text
stringlengths 1
2.56M
| id
stringlengths 40
40
| metadata
dict |
|---|---|---|
\section{Introduction}
An inverse atmospheric dispersion problem is stated as follows: given the
topography, the meteorological conditions and a set of detector readings
determine where and when the hazardous substance was released and in which
quantities. The problem is known as a source reconstruction problem, and it
is easy to state but harder to solve due to being, like most inverse
problems, ill-posed \cite{Tikhonov1963},\cite{Enting2002}. Spurred by
various applications including the locating of industrial plants \cite%
{Marchuk1986}, determining the amount of radioactive nuclides released from
Chernobyl \cite{GHL} and Fukushima \cite{StohlEtAl}, pin-pointing nuclear
tests \cite{RingbomEtAl}, estimation of material released from volcanoes
\cite{TheysEtAl},\cite{StohlEtAl2010},\cite{GrahnEtAl} a number of different
methods for addressing the inverse problem have been suggested. Even though
the main difference perhaps lies in the interpretation of the results these
methods are usually divided into two main categories: the probabilistic
approach with its Bayesian methods and the deterministic approach with its
optimisation methods. In the Bayesian setting a likelihood function is
calculated and weighted with any a priori information that one has at hand
to yield a posterior probability density function, which is then sampled to
yield an estimate of the sought source term (see e.g. \cite{Stuart2010} for
an introduction to general Bayesian inverse problems). In the deterministic
setting with optimisation methods a norm is devised under which the sensor
response of candidate sources is compared with the given sensor readings.
The candidate source that best fits the given sensor readings (minimizes the
distances under the chosen norm) is then deemed the solution to the inverse
problem.
For linear inverse atmospheric dispersion problems these methods come in
many different flavours and have often been devised with a given application
in mind: usually there are - a priori imposed - restrictions on the source
characteristics, e.g. the method may assume that the source is well
localised (located at a single point in space) and that the release was
instantaneous. For example Yee and coauthors have written a series of papers
adapting the Bayesian method to inverse dispersion problems of increasing
complexity \cite{KYL2007},\cite{Yee2007},\cite{YF2010},\cite{Yee2012} and
\cite{Yee2012B}.
To use an optimisation method the inverse dispersion problem has to be cast
in a manner where the distance (under a chosen norm) function between model
sensor data and the given sensor readings can be minimized. Usually a least
squares solution is sought, and in \cite{BP2015} conditions under which the
least square problem is well defined is presented. Much of the literature
focuses on the problem where it is a-priori assumed that there is only a
single source, see e.g. \cite{RL1998},\cite{THG2007},\cite{AYH2007} and \cite%
{ISS2012}. There are however exceptions, for example in \cite{SSI2012} the
renormalisation method (least square method under the renormalisation norm)
presented in \cite{ISS2012} is generalised to cover an unknown number of
point sources, and in \cite{Bocquet2005} the space-time has been discretised
and an optimal source term is constructed by forming a union of (space-time)
grid sized point sources.
In this paper we make a contribution to the literature on optimisation
methods by applying a bilevel optimisation method, see \cite{Bard1998}, to a
linear inverse dispersion problem. A bilevel optimisation method splits the
optimisation problem in two: into a leader (upper level) problem and a
follower (lower level) problem and they are solved concurrently rather than
simultaneously. We consider dispersion problems where the source is a single
point source emitting at a constant rate. This problem is well suited to a
bilevel optimisation method where the follower problem concerns solving for
the emission rate and the leader problem pinpointing the location of the
source. For the bilevel optimisation method to work properly the follower
problem is required to have minima, ideally a strict minimum. We therefore
study local strict convexity of the follower problem, see Theorem \ref%
{thm:local_convexity} for sufficient conditions. We then explore the concept
of local strict convexity and its connection to ill-possedness of inverse
problems through a few toy-model examples. Following this the paper is
rounded off with the bilevel optimisation method being applied to dispersion
data from a series of wind tunnel experiments of urban environments of
varying complexity. The wind tunnel data was collected as part of the
European Defence Agency category B project MODITIC. In all cases the
boundary layer is neutrally stable, and we only consider cases where the
released gas is neutrally buoyant making the dispersion problem linear.
\section{Bilevel optimization problems}
A bilevel optimization problem is a constrained optimization problem where
the constraints also includes an optimization problem. The problem is
divided into an upper level or leader problem, with decision variables $%
\boldsymbol{x}\in X\subseteq \mathbb{R}^{n}$, and a lower level or follower
problem, with decision variables $\boldsymbol{y}\in Y\subseteq \mathbb{R}%
^{m} $. Here $X$ and $Y$ may be restricted to integers or nonnegative
values. We follow the notation of \cite{Bard1998}, p. 6. The leader problem
has the form
\begin{eqnarray*}
V &=&\min_{\boldsymbol{x}\in X}F\left( \boldsymbol{x},\boldsymbol{y}\left(
\boldsymbol{x}\right) \right) \\
\boldsymbol{G}\left( \boldsymbol{x},\boldsymbol{y}\left( \boldsymbol{x}%
\right) \right) &\leq &\boldsymbol{0}
\end{eqnarray*}%
where $F$ and $\boldsymbol{G}$, respectively, are the leader objective
function and constraint function, and $\boldsymbol{y}\left( \boldsymbol{x}%
\right) $ is an optimal solution to the follower problem%
\begin{eqnarray*}
v\left( \boldsymbol{x}\right) &=&\min_{\boldsymbol{y}\in Y}f\left(
\boldsymbol{x},\boldsymbol{y}\right) \\
\boldsymbol{g}\left( \boldsymbol{x},\boldsymbol{y}\right) &\leq &\boldsymbol{%
0}\text{.}
\end{eqnarray*}%
An ambiguity occurs if the follower problem has several optimal solutions,
i.e., $\boldsymbol{y}\left( x\right) $ is set--valued. Then the follower is
indifferent towards these points, but the leader objective may be different
for different points in $\boldsymbol{y}\left( x\right) $, and there is no
way for the leader to direct the follower to the upper level optimal point.
Therefore, there may be no optimal solution to the bilevel program although
all functions are continuous and $X,Y$ are compact, cf. \cite{Bard1998}, p.
11. In our problem the sets $X,Y$ will be positive orthants, the follower
objective function $f\left( \boldsymbol{x},\boldsymbol{y}\right) $ will be a
Mahalanobis distance function measuring the discrepancy between model data
and measurements. The leader objective function $F$ will have the form%
\begin{equation*}
F\left( \boldsymbol{x},\boldsymbol{y}\right) =\exp \left( -\lambda \sum
y_{i}\right) +f\left( \boldsymbol{x},\boldsymbol{y}\right)
\end{equation*}%
hence minimizing the least square function, but penalizing large values of $%
\boldsymbol{y}$, which will act as a regularization of the problem ($\lambda
>0$ is a regularization parameter).
\section{The follower problem}
In the setting we are considering we have a priori assumed that the source
is a point source releasing a neutrally buoyant substance at a constant
rate. Under these assumptions the source location $\boldsymbol{x}\in \mathbb{%
R}^{d}$ is a\ nonlinear model parameter while the emission rate $\boldsymbol{%
y}\in \mathbb{R}^{1}$ is a linear model parameter. In general, our model
formulation allows a linear combination of basic sources, with a linear
positive weight vector $y\in \mathbb{R}^{n}$. To solve the inverse problem
we need a source-sensor relationship. Since the problem is linear the
source-sensor relationship is given a matrix relationship, which for
computational efficiency \cite{Marchuk1986} is expressed through the adjoint
formulation of the problem, thus $A:\mathbb{R}^{d}\rightarrow \mathbb{R}%
_{+}^{m\times n}$ is a matrix function with nonnegative elements (no sinks
are considered) and the adjoint model data $\boldsymbol{\mu }=\boldsymbol{%
\mu }\left( \boldsymbol{x},\boldsymbol{y}\right) =A\left( \boldsymbol{x}%
\right) \boldsymbol{y}\in \mathbb{R}^{m}$. The measured data $\boldsymbol{z}%
\in \mathbb{R}^{m}$ is the sensor response.\newline
We regard $\boldsymbol{z}$ as a random vector, and we assume that the
adjoint model data $\boldsymbol{\mu }$ represent the mean of $\boldsymbol{z}$%
. We also assume that the components $z_{i}$ of $\boldsymbol{z}$ are
statistically independent and that the variance of $z_{i}$ is%
\begin{equation*}
var\left( z_{i}\right) =\sigma ^{2}\left( \mu _{i}\right)
\end{equation*}%
where $\sigma :\mathbb{R\rightarrow R}_{+}$ is a given function. We let the
follower objective function be the \emph{Mahalanobis distance} between $%
\boldsymbol{z}$ and $\boldsymbol{\mu }$, viz.,
\begin{equation}
f\left( \boldsymbol{x},\boldsymbol{y}\right) =\sum_{i=1}^{m}\frac{\left(
z_{i}-\mu _{i}\right) ^{2}}{\sigma ^{2}\left( \mu _{i}\right) }\text{.}
\label{eqn:f}
\end{equation}%
In particular, we want to be able to choose a scale invariant distance
function, giving equal emphasis to all $\mu _{i}$, regardless of their size.
\subsection{Local convexity of the follower problem}
If the follower problem is strictly convex, the follower problem has a
unique optimal solution $\boldsymbol{y}\left( \boldsymbol{x}\right) $ for
each $\boldsymbol{x}\in X$, and by mild assumptions on $f\left( \boldsymbol{x%
},\boldsymbol{y}\right) $, an envelope theorem holds (e.g., \cite%
{MilgromSegal2002}, Theorem 2, p. 586), which implies that the optimal value
function%
\begin{equation*}
V\left( \boldsymbol{x}\right) =f\left( \boldsymbol{x},\boldsymbol{y}\left(
\boldsymbol{x}\right) \right) =\inf_{\boldsymbol{y\in Y}}\;f\left(
\boldsymbol{x},\boldsymbol{y}\right)
\end{equation*}%
is continuous. Therefore, the leader problem
\begin{equation*}
\inf_{\boldsymbol{x\in X}}F\left( \boldsymbol{x},\boldsymbol{y}\left(
\boldsymbol{x}\right) \right)
\end{equation*}%
has a solution since%
\begin{equation*}
F\left( \boldsymbol{x},\boldsymbol{y}\left( \boldsymbol{x}\right) \right)
=\exp \left( -\lambda \sum_{i}y_{i}\left( \boldsymbol{x}\right) \right)
V\left( \boldsymbol{x}\right) \text{. }
\end{equation*}%
However, in our setting the only situation when the follower problem is
guaranteed to be strictly convex is when $\sigma \left( \mu \right) $ is
constant, i.e., the classical least square method. However, we may derive
conditions for \emph{local convexity}, as the following theorem shows.
\begin{theorem}
\label{thm:local_convexity}Assume that
\begin{equation}
f\left( y_{1},...,y_{n}\right) =\sum_{i=1}^{m}\frac{r_{i}^{2}}{\sigma
^{2}\left( \mu _{i}\right) } \label{def:f}
\end{equation}%
where%
\begin{eqnarray*}
\mu _{i} &=&\sum_{j}a_{ij}y_{j} \\
r_{i} &=&z_{i}-\mu _{i}\text{. }
\end{eqnarray*}%
and $A=\left( a_{ij}\right) \in \mathbb{R}^{m\times n}$. Consider a fixed $%
\boldsymbol{y}$ and suppose that $\sigma $ is positive and twice
continuously differentiable in a neigbourhood of $\mu $.Then%
\begin{eqnarray*}
\frac{\partial f}{\partial y_{k}} &=&-2\sum_{i}a_{ik}\rho \left( \mu
_{i},r_{i}\right) \frac{r_{i}^{2}}{\sigma ^{2}\left( \mu _{i}\right) } \\
\frac{\partial ^{2}f}{\partial y_{k}\partial y_{l}} &=&2\sum_{i}a_{ik}a_{il}%
\eta \left( \mu _{i},r_{i}\right) \frac{r_{i}^{2}}{\sigma ^{2}\left( \mu
_{i}\right) }
\end{eqnarray*}%
where%
\begin{eqnarray*}
\rho \left( \mu ,r\right) &=&\frac{1}{r}+\frac{\sigma ^{\prime }\left( \mu
\right) }{\sigma \left( \mu \right) } \\
\eta \left( \mu ,r\right) &=&\left( \frac{1}{r}+2\frac{\sigma ^{\prime
}\left( \mu \right) }{\sigma \left( \mu \right) }\right) ^{2}-\left( \left(
\frac{\sigma ^{\prime }\left( \mu \right) }{\sigma \left( \mu \right) }%
\right) ^{2}+\frac{\sigma ^{\prime \prime }\left( \mu \right) }{\sigma
\left( \mu \right) }\right) \text{.}
\end{eqnarray*}%
Moreover, if $A$ has full rank then
\begin{enumerate}
\item If $\eta \left( \mu _{i},r_{i}\right) >0$ for all $i$ then $f$ is
strictly convex in a neighborhood of $\boldsymbol{y}$.
\item If $\eta \left( \mu _{i},r_{i}\right) <0$ for all $i$ then $f$ is
strictly concave in a neigborhood of $\boldsymbol{y}$.
\end{enumerate}
\end{theorem}
\begin{proof}
The formulas for the first and second derivatives of $f$ are proved by
elementary but tedious calculations, using the chain rule and the quotient
rule for differentiation. If $A$ has full rank, and $\eta \left( \mu
_{i},r_{i}\right) >0$ for all $i$, then%
\begin{equation*}
H\left( \boldsymbol{u}\right) \equiv \sum_{k,l}u_{k}\frac{\partial ^{2}f}{%
\partial y_{k}\partial y_{l}}u_{l}=\sum_{i}\sum_{k,l}\psi
_{i}a_{ik}u_{k}\psi _{i}a_{il}u_{l}=\sum_{i}\left( \sum_{k}\psi
_{i}a_{ik}u_{k}\right) ^{2}
\end{equation*}%
for all $\boldsymbol{u}\in \mathbb{R}^{n}$,where%
\begin{equation*}
\psi _{i}=\sqrt{\eta \left( \mu _{i},r_{i}\right) \frac{r_{i}^{2}}{\sigma
^{2}\left( \mu _{i}\right) }}\text{.}
\end{equation*}%
Hence $H\left( \boldsymbol{u}\right) \geq 0$. Assume that $H\left(
\boldsymbol{u}\right) =0$. Then $\sum_{k}\psi _{i}a_{ik}u_{k}=0$ for all $i$%
. Note that $\psi _{i}>0$, since when $r\rightarrow 0$ we have
\begin{equation*}
\eta \left( \mu ,r\right) \frac{r^{2}}{\sigma ^{2}\left( \mu \right) }%
\rightarrow \frac{1}{\sigma ^{2}\left( \mu \right) }>0\text{. }
\end{equation*}%
Hence $\boldsymbol{u}=0$, since the vectors $\boldsymbol{v}_{i}=\left( \psi
_{i}a_{i1},\psi _{i}a_{i2},...,\psi _{i}a_{in}\right) $, $i=1,2,...,m$ span $%
\mathbb{R}^{n}$. This shows that $H$ is positive definite at $\boldsymbol{y}$%
. By continuity, $H$ is positive definite, and hence $f$ strictly convex, in
a neighborhood of $\boldsymbol{y}$. Similarly, if $A$ has full rank and $%
\eta \left( \mu _{i},r_{i}\right) <0$, then
\begin{equation*}
H\left( \boldsymbol{u}\right) \equiv \sum_{k,l}u_{k}\frac{\partial ^{2}f}{%
\partial y_{k}\partial y_{l}}u_{l}=-\sum_{i}\left( \sum_{k}\psi
_{i}a_{ik}u_{k}\right) ^{2}
\end{equation*}%
where
\begin{equation*}
\psi _{i}=\sqrt{-\eta \left( \mu _{i},r_{i}\right) \frac{r_{i}^{2}}{\sigma
^{2}\left( \mu _{i}\right) }}
\end{equation*}%
and it follows with a similar argument as above that $f$ is strictly concave
in a neighborhood of $\boldsymbol{y}$.
\end{proof}
Most inverse modelling methods works perfectly for synthetic data, i.e. when
the "observed" sensor response is calculated using a dispersion model (the
same dispersion model that is then used to solve the inverse problem).
Indeed showing that an inverse modelling method works well for synthetic
data and, in particular, slightly perturbed synthetic data is usually
included in the body of work motivating the method. In view of the theorem
we make the following observation.
\begin{remark}
Note that when $r\rightarrow 0$, then $\eta \left( \mu ,r\right) \rightarrow
\infty $, so for small enough $r$, $\eta $ is always positive, if $\mu $ is
restricted to a compact set. Hence for $\boldsymbol{z}$ sufficiently close
to $\mathbf{\mu }\left( \boldsymbol{y}\right) $, $f\left( \boldsymbol{y}%
\right) $ is convex. This explains why most methods work well on synthetic
model data with small perturbations.
\end{remark}
We will now restrict ourselves to functions $\sigma \left( \mu \right) $
which are continuous, increasing, convex, and positive for $\mu >0$, and
satisfying%
\begin{equation*}
\lim_{\mu \rightarrow \infty }\frac{\sigma \left( \mu \right) }{\mu }=1\text{%
. }
\end{equation*}%
Moreover, we assume that $\sigma $ is twice continuously differentiable,
except possibly at a finite number of points. Hence $\sigma ^{\prime \prime
}\left( \mu \right) \geq 0$ at all points where $\sigma ^{\prime \prime }$
exists. Note that if $\sigma \left( \mu \right) $ satisfies these
conditions, so does
\begin{equation*}
\sigma _{\delta }\left( \mu \right) \equiv \delta \sigma \left( \mu /\delta
\right)
\end{equation*}%
for any $\delta >0$, so the class of functions $\sigma $ we consider is
scale--invariant. We consider three examples:%
\begin{eqnarray}
\sigma \left( \mu \right) &=&erf\left( \mu \right) -2\left( 1-\frac{1%
}{\sqrt{\pi }}\right) \exp \left( -\mu ^{2}\right) \text{ (Smooth threshold)}
\notag \\
\sigma \left( \mu \right) &=&\max \left( 1,\mu \right) \text{ (Piecewise
linear)} \label{eqn:thesigmas} \\
\sigma \left( \mu \right) &=&\mu \text{ (Linear)} \notag
\end{eqnarray}%
\begin{figure}[H]
\centering
\includegraphics[scale=0.25]{fig1.jpg}
\caption{Graphs of the three considered $%
\protect\sigma $--functions.}
\label{fig_simplex}
\end{figure}
Recall that
\begin{equation*}
f\left( \boldsymbol{y}\right) =\sum_{i}\frac{\left( z_{i}-\mu _{i}\right)
^{2}}{\sigma ^{2}\left( \mu _{i}\right) }
\end{equation*}%
Since we assume that $\sigma ^{\prime \prime }\geq 0$ and $\sigma >0$ for $%
\mu >0$, we can write $\eta $ on the form%
\begin{equation*}
\eta =\left( \frac{1}{r}-\alpha \right) \left( \frac{1}{r}-\beta \right)
\end{equation*}%
where%
\begin{eqnarray*}
\alpha \left( \mu \right) &=&-2\frac{\sigma ^{\prime }\left( \mu \right) }{%
\sigma \left( \mu \right) }-\sqrt{\left( \frac{\sigma ^{\prime }\left( \mu
\right) }{\sigma \left( \mu \right) }\right) ^{2}+\frac{\sigma ^{\prime
\prime }\left( \mu \right) }{\sigma \left( \mu \right) }} \\
\beta \left( \mu \right) &=&-2\frac{\sigma ^{\prime }\left( \mu \right) }{%
\sigma \left( \mu \right) }+\sqrt{\left( \frac{\sigma ^{\prime }\left( \mu
\right) }{\sigma \left( \mu \right) }\right) ^{2}+\frac{\sigma ^{\prime
\prime }\left( \mu \right) }{\sigma \left( \mu \right) }}
\end{eqnarray*}%
Clearly, $\eta >0$ if and only if the two factors have the same sign, i.e., $%
1/r\in \left[ \alpha ,\beta \right] ^{c}$, and $\eta <0$ if and only if $%
1/r\in \left( \alpha ,\beta \right) $. To analyze these conditions further
we need to consider the cases when $\alpha ,\beta $ have the same sign and
when they have different signs. If $\alpha ,\beta $ have the same sign then $%
1/r\in \left( \alpha ,\beta \right) $ if and only if $r\in \left( 1/\beta
,1/\alpha \right) $ (this includes the limiting cases $\beta \nearrow 0$ or $%
\alpha \searrow 0$, with $1/\beta =-\infty $ and $1/\alpha =\infty $). If $%
\alpha ,\beta $ have different signs, i.e., $\alpha <0<\beta $, then $1/r\in
\left( \alpha ,\beta \right) $ if and only if $r\in \left[ 1/\alpha ,1/\beta %
\right] ^{c}$. By the assumptions on $\sigma $, $\alpha $ and $\beta $ are
defined everywhere except possibly at a finite number of points. Henceforth,
we only consider points where $\alpha $ and $\beta $ are defined. We can now
conclude that $\eta <0$ if and only if either i) $z-\mu \in \left( \beta
^{-1},\alpha ^{-1}\right) $ and $\alpha <\beta \leq 0$, or ii) $z-\mu \in %
\left[ \alpha ^{-1},\beta ^{-1}\right] ^{c}$ and $\alpha <0<\beta $, or iii)
$z-\mu \in \left( \beta ^{-1},\infty \right) $ and $0=\alpha <\beta $.
Likewise, we can conclude by complementarity that $\eta >0$ if and only if
either i) $z-\mu \in \left[ \beta ^{-1},\alpha ^{-1}\right] ^{c}$ and $%
\alpha <\beta \leq 0$, or ii) $z-\mu \in \left( \alpha ^{-1},\beta
^{-1}\right) $ and $\alpha <0<\beta $, or iii) $z-\mu \in \left( -\infty
,\beta ^{-1}\right) $ and $0=\alpha <\beta $ or iv) $\alpha =\beta =0$.
Relying on these inequalities we can find domains where a distance function
is guaranteed to be strictly convex or strictly concave.
\begin{example}
Consider $\sigma \left( \mu \right) =\mu $, the linear case. Then $\alpha
(\mu )=-3/\mu $ and $\beta (\mu )=-1/\mu $. Hence $\alpha ^{2}>\beta $ and $%
\eta \left( \mu ,r\right) >0$ if and only if $r\in \left( -\infty ,-\mu
\right) \cup \left( -\mu /3,\infty \right) $, i.e., $z=\mu +r\in \left(
-\infty ,0\right) \cup \left( 2\mu /3,\infty \right) $. Moreover, $\eta
\left( \mu ,r\right) <0$ if and only if $z=\mu +r\in \left( 0,2\mu /3\right)
$. Hence for fixed $\boldsymbol{z}$, $f$ is strictly convex on the set $\cap
_{i}\left\{ 2\mu _{i}\left( \boldsymbol{y}\right) /3<z_{i}\right\} $, and
strictly concave on the set $\cap _{i}\left\{ 2\mu _{i}\left( \boldsymbol{y}%
\right) /3>z_{i}\right\} $. Let for example%
\begin{equation*}
A=%
\begin{bmatrix}
1 & 0 \\
0 & 1 \\
1 & 1%
\end{bmatrix}%
\text{, }z=%
\begin{bmatrix}
1 \\
1 \\
5/3%
\end{bmatrix}%
\text{.}
\end{equation*}%
Then%
\begin{equation*}
f\left( y_{1},y_{2},y_{3}\right) =\frac{\left( 1-y_{1}\right) ^{2}}{y_{1}^{2}%
}+\frac{\left( 1-y_{2}\right) ^{2}}{y_{2}^{2}}+\frac{\left(
3-y_{1}-y_{2}\right) ^{2}}{\left( y_{1}+y_{2}\right) ^{2}}\text{,}
\end{equation*}%
and $f$ is strictly convex on $\left\{ \left( y_{1},y_{2},y_{3}\right) \in
\mathbb{R}_{+}^{3}\mid 2y_{1}/3<1\text{, }2y_{2}/3<1\text{,}2\left(
y_{1}+y_{2}\right) /3<5/3\right\} $, and strictly concave on $\left\{ \left(
y_{1},y_{2},y_{3}\right) \in \mathbb{R}_{+}^{3}\mid 2y_{1}/3>1\text{, }%
2y_{2}/3>1\text{,}2\left( y_{1}+y_{2}\right) /3>5/3\right\} $ (the last
condition is not active). Of course, $f$ may be strictly convex or concave
on points outside these sets also;\ these conditions are sufficient, but not
necessary. The domains of local convexity and local concavity are plotted in
the next figure.
\begin{figure}[H]
\centering
\includegraphics[scale=0.25]{fig2.jpg}
\caption{This
figure shows isocurves of $f\left( y_{1},y_{2}\right) $, the local convexity
region (blue) and the local concavity region (red), determined from Theorem
1. The minimum point is marked with a red circle.}
\label{fig2}
\end{figure}
\end{example}
\begin{remark}
When $\sigma $ is rescaled\ (i.e., replacing $\sigma \left( \mu \right) $ by
$\sigma _{\delta }\left( \mu \right) $), the derivatives are scaled
according to $\sigma _{\delta }^{\prime }\left( \mu \right) =\sigma ^{\prime
}\left( \mu /\delta \right) $ and $\sigma _{\delta }^{\prime \prime }=\delta
^{-1}\sigma ^{\prime \prime }\left( \mu /\delta \right) $, and hence $\alpha
\left( \mu \right) $ is replaced by $\alpha _{\delta }\left( \mu \right)
=\delta ^{-1}\alpha \left( \mu /\delta \right) $, and $\beta \left( \mu
\right) $ is replaced by $\beta _{\delta }\left( \mu \right) =\delta
^{-1}\beta \left( \mu /\delta \right) $.
\end{remark}
\subsection{Examples of follower problems with unique and non-unique
solutions}
In Theorem \ref{thm:local_convexity} we established sufficient conditions
for the objective function $f$ to be locally strictly convex with the view
of determining when the follower problem is well posed. Alas, strict
convexity of the objective function is not sufficient to make the follower
problem convex as the constraints have to be taken into account. We explore
the interplay between the objective function and the constraints and their
effect on minima and well posedness of the minimization problem through a
series of examples. It is instructive to begin with the linear case $\sigma
(\mu )=\mu $, and then proceed to the nonlinear cases (smooth threshold and
piecewise linear), compare (\ref{eqn:thesigmas}).
\subsubsection{The linear case}
In the linear case, the minimization problem for $f$ can be formulated as a
constrained minimization problem with a convex objective function. However,
the constraints are not convex, which can cause multiple minimum points for
certain values of $A$ and $\boldsymbol{z}$. For $\sigma \left( \mu \right)
=\mu $, $f$ is a strictly convex function of $\boldsymbol{\xi }\equiv
\boldsymbol{\mu }^{-1}=\left( \mu _{1}^{-1},...,\mu _{m}^{-1}\right) $:%
\begin{equation*}
f\left( \boldsymbol{y}\right) =F\left( \boldsymbol{\xi }\right) \equiv
\sum_{i}\left( z_{i}\xi _{i}-1\right) ^{2}\text{.}
\end{equation*}%
Writing $F$ on the form%
\begin{equation*}
F\left( \boldsymbol{\xi }\right) =\sum_{i:b_{i}\neq 0}\frac{\left( \xi
_{i}-z_{i}^{-1}\right) ^{2}}{\left( z_{i}^{-1}\right) ^{2}}+m-m^{\prime }%
\text{,}
\end{equation*}%
where $m^{\prime }=\sum_{i:b_{i}\neq 0}1$ is the number of $b_{i}\neq 0$, we
see that for $c>0$, the level surface $F\left( \boldsymbol{\xi }\right)
=c+m-m^{\prime }$ is an $m^{\prime }$--axial ellipsoidal cylinder with
center coordinates $\xi _{i}=z_{i}^{-1}$ and corresponding semiaxes $\sqrt{c}%
z_{i}^{-1}$ for all $i$ such that $z_{i}\neq 0$, extending linearly along
all coordinates $\xi _{i}$ for which $z_{i}=0$. Let us assume that all $%
z_{i}\neq 0$ for simplicity. Then for $c>0$ the level surface $F\left(
\boldsymbol{\xi }\right) =c$ is an $m$--axial ellipsoid centered at $\left(
z_{1}^{-1},...,z_{m}^{-1}\right) $ with semiaxes $\sqrt{c}z_{1}^{-1},...,%
\sqrt{c}z_{m}^{-1}$. The minimization problem for $f$ can be formulated as a
constrained minimization problem%
\begin{eqnarray*}
&&\min F\left( \boldsymbol{\xi }\right) \\
\boldsymbol{G}\left( \boldsymbol{\xi },\boldsymbol{y}\right) &\leq &%
\boldsymbol{0} \\
\boldsymbol{H}\left( \boldsymbol{\xi },\boldsymbol{y}\right) &=&\boldsymbol{0%
}
\end{eqnarray*}%
where $\boldsymbol{G}\left( \boldsymbol{\xi },\boldsymbol{y}\right) =-%
\boldsymbol{y}$, and $\boldsymbol{H}=\left( H_{1},...,H_{m}\right) $, where%
\begin{equation*}
H_{i}\left( \boldsymbol{\xi },\boldsymbol{y}\right) =\varphi _{i}\left(
\boldsymbol{y}\right) -\xi _{i}\text{, }i=1,...,m
\end{equation*}%
and
\begin{equation*}
\varphi _{i}\left( \boldsymbol{y}\right) =\frac{1}{\left( A\boldsymbol{y}%
\right) _{i}}\text{.}
\end{equation*}%
The objective function $F\left( \boldsymbol{\xi },\boldsymbol{y}\right)
=F\left( \boldsymbol{\xi }\right) $, is convex, and depends on $\boldsymbol{z%
}$, but is independent of $A$. The inequality constraints are linear and
hence convex. The equality constraints depend on $A$, but are independent of
$\boldsymbol{z}$, and are not convex, so the problem is not convex. We have
thus separated the dependencies of $A$ and $\boldsymbol{z}$ into the
objective and constraint functions, respectively.
The equality constraints $\boldsymbol{H}\left( \boldsymbol{\xi },\boldsymbol{%
y}\right) =\boldsymbol{0}$ define an $n$--dimensional parametrized surface $%
S $ in $\mathbb{R}^{m}$ by $\boldsymbol{\xi }=\boldsymbol{\varphi }\left(
\boldsymbol{y}\right) $. The $n$--dimensional tangent space to $S$ at $%
\boldsymbol{\varphi }\left( \boldsymbol{y}\right) $ is spanned by the
tangent vectors%
\begin{equation*}
\frac{\partial \boldsymbol{H}}{\partial y_{j}}\left( \boldsymbol{y}\right)
\text{, }j=1,...,n\text{. }
\end{equation*}%
The Karush-Kuhn-Tucker (KKT) conditions (see e.g. \cite{NocedalWright2006})
for the minimization problem are%
\begin{eqnarray*}
-\frac{\partial F}{\partial \xi _{i}} &=&\sum_{k=1}^{n}\mu _{k}\frac{%
\partial G_{k}}{\partial \xi _{i}}+\sum_{l=1}^{m}\lambda _{l}\frac{\partial
H_{l}}{\partial \xi _{i}}=\lambda _{i}\text{, }i=1,...,m \\
0 &=&-\frac{\partial F}{\partial y_{j}}=\sum_{k=1}^{n}\mu _{k}\frac{\partial
G_{k}}{\partial y_{j}}+\sum_{l=1}^{n}\lambda _{l}\frac{\partial H_{l}}{%
\partial y_{j}}\text{, }j=1,...,n \\
\mu _{j}G_{j} &=&0\text{, }j=1,...,n \\
\mu _{j} &\geq &0\text{, }j=1,...,n
\end{eqnarray*}%
which are necessary conditions for minimum. Assume for simplicity that we
have a minimum with $\boldsymbol{y}^{\ast }>0$. Then $G_{j}<0$ so $\mu
_{j}=0 $. Substituting the first equation in the second we get
\begin{equation*}
\sum_{l=1}^{m}\frac{\partial F}{\partial \xi _{l}}\frac{\partial H_{l}}{%
\partial y_{j}}=0\text{, }j=1,...,n\text{.}
\end{equation*}%
Geometrically, this conditions means that the gradient of $F$ is orthogonal
to the tangent space of $S$ at $y^{\ast }$. Hence, in the $\boldsymbol{\xi }$%
--space, minimal points $\boldsymbol{y}^{\ast }$ are characterized by points
$\boldsymbol{\xi }^{\ast }=\boldsymbol{\varphi }\left( \boldsymbol{y}^{\ast
}\right) $ where the $n$--dimensional surface $S:\boldsymbol{\xi }=%
\boldsymbol{\varphi }\left( \boldsymbol{y}\right) $ is tangent to the
ellipsoid $F\left( \boldsymbol{\xi }\right) =c^{\ast }$. The optimal value
is then $c^{\ast }$.
Suppose now that $A$ is fixed. For $\boldsymbol{z}$, the center of the
ellipsoidal isosurfaces of $F$, let $E\left( \boldsymbol{z}\right) $ denote
the smallest isosurface of $F$ that intersects $S$. For generic $\boldsymbol{%
z}$, the ellipsoid $E\left( \boldsymbol{z}\right) $ will contain only one
point $\boldsymbol{\varphi }\left( \boldsymbol{y}^{\ast }\right) $, defined
by the unique minimum point $\boldsymbol{y}^{\ast }$. However, for $%
\boldsymbol{z}$ in some exceptional lower--dimensional set, $E\left(
\boldsymbol{z}\right) $ will touch $S$ at several points, in which case we
have several minimum points $\boldsymbol{y}^{\ast }$. In that case, $A$ or $%
\boldsymbol{z}$ can be perturbed so that either of the multiple minima is
perturbed into a single global minimum. Hence the global minimum point
varies discontinuously with $A$ and $\boldsymbol{z}$, and the problem is ill
posed (unless $\boldsymbol{z}$ is restricted to be generic, e.g., if $%
\boldsymbol{z}^{-1}$ is sufficiently close to the surface $S$).
\begin{example}
\label{ex:linear}Let%
\begin{equation*}
A=%
\begin{bmatrix}
1 & \varepsilon \\
\varepsilon & 1 \\
1 & 1%
\end{bmatrix}%
\text{, }\boldsymbol{z}=%
\begin{bmatrix}
1 \\
1 \\
t%
\end{bmatrix}%
\end{equation*}%
where $\varepsilon =0.1$.
\begin{figure}[H]
\centering
\includegraphics[scale=0.15]{fig3.jpg}
\caption{Case $%
t=3$. The figure shows minimum point $\protect\xi ^{\ast }=\left(
0.7729,0.7729,0.4251\right) $ as a red dot, the surface $S$ parametrized for
$0.5\leq y_{1},y_{2}\leq 5$, and the ellipsoidal isosurface for $f$ with
optimal value $c^{\ast }=0.17893$. The center $\boldsymbol{z}$ of the
ellipsoid is marked by a blue cross.}
\label{fig3}
\end{figure}
\begin{figure}[H]
\centering
\includegraphics[scale=0.15]{fig4.jpg}
\caption{Case $t=3$. The figure
shows the minimum point $y^{\ast }=\left( 1.1762;1.1762\right) $ marked with
a red circle, with the minimal value $c^{\ast }=0.17893$. Also isocurves for
$f$ are shown.}
\label{fig4}
\end{figure}
\begin{figure}[H]
\centering
\includegraphics[scale=0.15]{fig5.jpg}
\caption{Case $t=5.$, two minimum points $\protect\xi _{1}^{\ast }=\left(
0.9303,0.2920,0.2444\right) $, $\protect\xi _{2}^{\ast }=\left(
0.2920,0.9303,0.2444\right) $ marked by red dots, parametric surface $S$
plotted for $0.5\leq y_{1},y_{2}\leq 5$, and ellipsoidal isosurface for $f$
\ and optimal value $c^{\ast }=0.55556$. The center $\boldsymbol{z}$ of the
ellipsoid is marked by a blue cross.}
\label{fig5}
\end{figure}
\begin{figure}[H]
\centering
\includegraphics[scale=0.15]{fig6.jpg}
\caption{ Case $t=5$. Two
minimal points $y_{1}^{\ast }=\left( 0.73987,3.351\right) $, $y_{2}^{\ast
}=\left( 3.351,0.73987\right) $, marked by red circles. Minimum value $%
c^{\ast }=0.55556$. Isosurfaces for $f$ also shown.}
\label{fig:linear_case_t5}
\end{figure}
\end{example}
\paragraph{The one--dimensional case}
Distinct minima cannot occur in the case $n=1$. We have
\begin{proposition}
If $a_{i}>0$, and some $b_{i}>0$ then
\begin{equation*}
f\left( y\right) =\sum_{i=1}^{m}\frac{\left( b_{i}-a_{i}y\right) ^{2}}{%
\left( a_{i}y\right) ^{2}}=m-\frac{\alpha ^{2}}{\beta }+\beta \left( y^{-1}-%
\frac{\alpha }{\beta }\right) ^{2}
\end{equation*}%
where
\begin{equation*}
\chi _{i}=b_{i}/a_{i}\text{, }\alpha =\sum_{i}\chi _{i}\text{ and }\beta
=\sum_{i}\chi _{i}^{2}\text{, }
\end{equation*}%
and $f$ has minimum value $m-\alpha ^{2}/\beta $, attained for%
\begin{equation*}
y=\frac{\beta }{\alpha }\text{,}
\end{equation*}%
which is a strict minimum. Moreover, the minimum point $y$ depends
continuously on $a_{i}$, $b_{i}$, so the minimization problem is well--posed.
\end{proposition}
\begin{proof}
The formula for $f$ is proved by elementary calculations. The final
conclusion follows from the observation that $f\left( y\right) $ is a second
order polynomial in $y^{-1}$.
\end{proof}
\begin{remark}
Note that $\partial \chi _{i}/\partial a_{i}=-a_{i}^{-1}\chi _{i}^{2}$ and $%
\partial \chi _{i}/\partial b_{i}=a_{i}^{-1}$, and the sensitivity of the
minimum point $y=y\left( a_{1},...,a_{m},b_{1},...,b_{m}\right) $ is given by%
\begin{equation*}
\frac{\partial y}{\partial a_{i}}=\frac{1}{a_{i}}\frac{\left( y-2\chi
_{i}\right) \chi _{i}^{2}}{\alpha }\text{, }\frac{\partial y}{\partial b_{i}}%
=\frac{1}{a_{i}}\frac{2\chi _{i}-y}{\alpha }
\end{equation*}%
so the minimum point $y$ is very sensitive for $a_{i}$ and $b_{i}$ when $%
a_{i}$ is small, provided that $y\neq 2\chi _{i}$.
\end{remark}
\subsubsection{The nonlinear case}
In contrast to the linear case, $f$ can have multiple local minima for $n=1$
in the nonlinear case. For certain values of $A$,$\boldsymbol{z}$, these
minimum values may be identical, hence distinct global minima. When this
occurs, the minimum problem becomes ill--posed, as the following example
shows.
\begin{example}
Let%
\begin{equation*}
A=%
\begin{bmatrix}
0.3 \\
0.8%
\end{bmatrix}%
\text{, }\boldsymbol{z}=%
\begin{bmatrix}
t \\
0%
\end{bmatrix}%
\text{, }\delta =1
\end{equation*}%
and
\begin{equation*}
f\left( y\right) =\sum_{i=1}^{m}\frac{\left( z_{i}-a_{i}y\right) ^{2}}{%
\sigma ^{2}\left( a_{i}y\right) }\text{, }i=1,2\text{. }
\end{equation*}%
If $t=t_{0}\approx 1.0732$, then $f$ has two distinct global minimum points
for the smooth threshold $\sigma (\mu )$, (the smooth threshold is defined
in equation (\ref{eqn:thesigmas})), see Figure \ref{fig:globalmin_t10732}.
Moreover, for all $t$ in a neighborhood $U$ of $t_{0}$, $f$ has two distinct
local minima . For $t<t_{0}$ in $U$, the left local minimum of $f$ is
global, and for $t>t_{0}$ in $U$, the right local minimum of $f$ is global.
Hence when $t$ increases from $t=1.05$ to $1.10$, the global minimum point $%
y=y\left( t\right) $ of $f\left( y\right) $ jumps from $y\approx 0.4919$ to $%
y\approx 3.5733$ at $t=t_{0}$, see figures \ref{fig:globalmin_t105} and \ref%
{fig:globalmin_t110}. The function $f$ shows a similar behaviour for the
piecewise linear $\sigma $, with a slightly different value of $t_{0}$.%
\begin{figure}[H]
\centering
\includegraphics[scale=0.15]{fig9.jpg}
\caption{ The figure shows $f\left(
y\right) $ for $t=t_{0}\approx 1.0732$. At this transition point, the smooth
threshold $\protect\sigma $ gives two distinct global minima, while the
piecewise linear $\protect\sigma $ gives a single global minimum to the
right. The transition point for the piecewise linear $\protect\sigma $
occurs for a sligthly different value of $t$.}
\label{fig:globalmin_t10732}
\end{figure}
\begin{figure}[H]
\centering
\includegraphics[scale=0.15]{fig7.jpg}
\caption{ The figure shows $f\left( y\right) $ for $t=1.05$. Both smooth
threshold and piecewise linear $\protect\sigma $ give two distinct local
minima, the left one is also global minimum.}
\label{fig:globalmin_t105}
\end{figure}
\begin{figure}[H]
\centering
\includegraphics[scale=0.15]{fig8.jpg}
\caption{ The figure shows $f\left( y\right) $ for $t=1.10$. Both smooth
threshold and piecewise linear $\protect\sigma $ give two distinct local
minima, the right one is also global minimum.}
\label{fig:globalmin_t110}
\end{figure}
\end{example}
In higher dimensions, nonlinear $\sigma $ give similar behaviour of $f$
regarding multiple global minima and ill--posedness of the minimization
problem.
In higher dimensions, the nonlinear $\sigma $ functions causes a similar
behaviour as the linear $\sigma $ regarding multiple local minimas of $f$.
To illustrate this, we show a slight perturbation of the bimodal example,
Example \ref{ex:linear}, above for the linear $\sigma $. The perturbed
problem still have two local minima, but only one of them is a global min.
Let
\begin{equation*}
A=%
\begin{bmatrix}
1 & 0.1 \\
0.1 & 1 \\
1 & 1%
\end{bmatrix}%
\text{, }\boldsymbol{z}=%
\begin{bmatrix}
1.1 \\
0.9 \\
5%
\end{bmatrix}%
\text{.}
\end{equation*}%
Then we get
\begin{figure}[H]
\centering
\includegraphics[scale=0.15]{fig10.jpg}
\includegraphics[scale=0.15]{fig11.jpg}
\caption{Both smooth threshold and piecewise linear $\sigma $ functions give two
local minima in this case, and quite similar isocurve pattern. Comparison with the linear case, Figure %
\ref{fig:linear_case_t5}, in the previous example reveals a similar pattern.}
\label{fig:10_11}
\end{figure}
In some cases there are huge differences between smooth threshold and
piecewise linear $\sigma $. For example, by randomly choosing $A$ and $%
\boldsymbol{z}$, we found%
\begin{equation*}
A=%
\begin{bmatrix}
0.0557 & 0.8528 \\
0.3231 & 0.6868 \\
0.9336 & 0.1672%
\end{bmatrix}%
\text{, }\boldsymbol{z}=%
\begin{bmatrix}
1.3196 \\
0 \\
1.3499%
\end{bmatrix}%
\end{equation*}%
which gives
\begin{figure}[H]
\centering
\includegraphics[scale=0.15]{fig12.jpg}
\includegraphics[scale=0.15]{fig13.jpg}
\caption{Here we see a dramatic
difference between the discontinouous gradient and double local minima in
the piecewise linear case, and the smooth, single global minimum in the
smooth threshold and linear cases.}
\label{fig:qualitative_difference}
\end{figure}
A motivation for using a scaled piecewise linear $\sigma $ rather than the
linear $\sigma $ is that model values $\mu $ (averages) below a detection
limit should give a variance $\sigma $ of the measurement value $z$ which is
independent of $\mu $ (corresponding to the background noise). We think it
is a good compromise to use the smooth threshold $\sigma $, giving a
variance with a strictly positive lower bound, but avoiding potential
numerical difficulties as pictured in Figure \ref{fig:qualitative_difference}
above.
\section{Application: neutral gas in an urban environment}
We will now apply the bilevel optimisation method to experimental data
reconstructing the source term. The different choices of distance function,
smooth threshold, piecewise linear, linear and constant will be compared and
contrasted. The experimental data stems from the European Defence Agency
category B project MODITIC where wind tunnel experiments were conducted
involving release of both neutrally buoyant gas as well as dense gas over
urban environments of varying complexity. In all cases the wind tunnel
boundary layer was neutrally stable. The gas was released from a single
point source either as a puff or continuously at a constant rate. For the
full details on the experimental set up we refer to \cite{RobinsEtAl2016a}
and \cite{RobinsEtAl2016b}.
\subsection{Sensor data}
As the bilevel optimisation method described above is designed for linear
inverse dispersion problems we only consider cases where the released gas is
neutrally buoyant, and to further reduce the scope we focus on two scenarios
referred to as the simple array, a symmetric case with four buildings, and
the complex array, a less symmetric case involving fourteen buildings.
\subsubsection{The simple array}
The dimensions of the simple array, with the positions of the synchronized
detectors we use for backtracking is shown in Figure \ref{fig:SimpleArray}.%
\begin{figure}[H]
\centering
\includegraphics[scale=0.15]{Array.jpg}
\caption{Sensors network (A, B, C
arrangement) for the simple array cases. In each scenario 4 synchronized
detectors are used to measure the concentration of the released gas. The
dectors are located at the positions stated under A, B and C respectively.}
\label{fig:SimpleArray}
\end{figure}
The wind direction in the wind tunnel experiments is aligned with the x-axis
in Figure \ref{fig:SimpleArray}, and as shown in the figure the simple array
may aligned at two different angles: we denote the alignment in the upper
pane as \textquotedblleft 0 degrees\textquotedblright\ and the alignment in
the lower pane as \textquotedblleft 45 degrees\textquotedblright . In each
scenario four synchronized detectors were used, hence the reference 4 FFID
in Figure \ref{fig:SimpleArray}, and we chose three different sets of
locations for these detectors: we refer to these as case A, B and C. In
addition to this two different source locations were available: these we
denote S1 and S2, both S1 and S2 (at the origin) are located at 8H upwind
(=-0.88m) in the x direction, but S1 is shifted off the x-axis by 1.5H
(=0.165m) in the +y direction. The location of S2 was chosen to be the
origin of the coordinate system. The diameter of the sources is 0.1m (to be
compared to building height and sides of H=0.11m). A constant release rate
of 50l/min=8.33e-4 m$^{3}$/s was used in all scenarios.
\subsubsection{The complex array}
In the complex array there more buildings present and there is less
symmetry. The configuration also opens up for a larger number of sensible
detector locations, however, still only 4 synchronized detectors were used
in these scenarios. For the complex array we chose five different sets of
detector configurations: these are denoted case A through to E, and these
are shown together with the geometry of the complex array in Figure \ref%
{fig:ComplexArray}. Note that some detectors are located on the roof tops
(cases A and C) and other are inside the street canyons (cases B, D and E).
\begin{figure}[H]
\centering
\includegraphics[scale=0.35]{complexarray.jpg}
\caption{Configuration of the complex
array. In each scenario, denoted case A through to E, 4 synchronized
detectors were used to measure the concentration of the released gas. Note
that some detectors are located at the roof of the buildings. Two different
source locations were used, denoted S1 and S2. (There is a third source S3
indicated in the figure, but it was not used for inverse modelling).}
\label{fig:ComplexArray}
\end{figure}
Two different source locations were used S1 and S2. The diameter of the
sources is 0.1m. The source S1 is defined to be at the origin (see Figure %
\ref{fig:ComplexArray}) while S2 is located upwind at x=-8H=-0.88m and y=0.
S1 and S2 are both located on the ground. As for the empty and complex array
the source strength is 50l/min=8.33e-4m$^{3}$/s.
\subsection{Source-sensor relationship: adjoint CFD-plumes}
To apply the bilevel optimisation method to the sensor data retrieved from
the wind tunnel we need a relationship between the source and the sensors,
i.e. a dispersion model. Since we are in a setting where a) we are
considering an urban environment and b) we are trying to gauge the fidelity
of the inverse solver, we opt to use a computational fluid dynamics solver
for the wind field. We are only considering the steady state of the
dispersion scenario (not the transient behaviour when the source is turned
on and subsequently off), hence a Reynolds-averaged Navier-Stokes (RANS)
solver is suitable. See \cite{BurkhartBurman} for RANS solutions of MODITIC
scenarios. The RANS solution in each case defines a source-sensor
relationship which could be used to solve the inverse problem, however, due
to computational efficiency it is preferable to solve the adjoint dispersion
problem \cite{Marchuk1986}. The dispersion problem is self-adjoint, hence
the adjoint solution is obtained by reversing the advection while leaving
the diffusive component untouched. The RANS solver in PHEONICS was adapted
accordingly to yield adjoint RANS solutions for the simple array and the
complex array \cite{WP7000}. One adjoint wind field was computed for each
sensor. These adjoint flow fields constitute a source-sensor relationship.
\subsection{Source reconstruction}
We will briefly outline the bilevel optimisation algorithm before using it
to reconstruct the MODITIC sources.
\subsubsection{Bilevel optimisation algorithm}
In each case four synchronous sensors where used, let us number them $%
i=1,2,3,4$. For each sensor $i$ compute the adjoint dispersion plume $\chi
_{i}(x,y,z)$ for each grid point $(x,y,z)$. Using the adjoint plume the
model sensor data is given by $c_{i}=q\chi _{i}(x,y,z)$ where $q$ is the
source strength of a point source located at grid point $(x,y,z)$. We denote
the sensor data obtained in the wind tunnel $d_{i}$.
\begin{enumerate}
\item Choose the weight $\sigma $ in the distance function. Choose detection
threshold $\delta $ (may be known from the sensor manufacturer). Choose
penalty coefficient $\lambda $.
\item Follower problem: Solve for the optimal emission rate $q^{\ast
}(x,y,z) $ in each grid point $(x,y,z)$ by%
\begin{equation*}
\min_{q}\sum_{i=1}^{4}\frac{\left( d_{i}-c_{i}^{{}}\right) ^{2}}{\sigma
^{2}\left( c_{i}^{{}}\right) }.
\end{equation*}
\item Compute the envelope grid function $c_{i}^{\ast }=q^{\ast }(x,y,z)\chi
_{i}(x,y,z)$ for each $(x,y,z)$ in the grid.
\item Leader problem: Minimize the grid function%
\begin{equation*}
V(x,y,z)=\min_{(x,y,z)}\left( \sum_{i=1}^{4}\frac{\left( d_{i}-c_{i}^{\ast
}\right) ^{2}}{\sigma ^{2}\left( c_{i}^{\ast }\right) }+\lambda q^{\ast
}(x,y,z)\right) ,
\end{equation*}%
,where $\lambda q^{\ast }$ is a penalty term, to find the optimal source
location $(x,y,z)=\left( x^{\ast },y^{\ast },z^{\ast }\right) .$
\item The optimal solution is given by $\left( q^{\ast },x^{\ast },y^{\ast
},z^{\ast }\right) $.
\end{enumerate}
\subsubsection{Results}
Below we present the source reconstruction results by showing the estimation
error in the $x,y$ and $z$ directions (source location) and in the emission
rate respectively. As the estimated emission rate varies over a large range
we present the error in the $\log _{10}$ value of the emission rate. We
consider four different choices of weight $\sigma $ in the distance
function: smooth threshold, piecewise linear, linear (as defined in equation
(\ref{eqn:thesigmas})) and in addition a constant weight $\sigma \equiv 1$.
The error for each parameter for each choice of $\sigma $ for each data set
in the simple and complex arrays are presented in scatter plots for
comparison. These results are presented for two different choices of high
emission rate penalty factor $\lambda $, $\lambda =10^{-2}$ and $\lambda =1$%
. In addition the results are presented for two different flavours of the
inverse problem: the "unconstrained" one where the source could be located
anywhere on or above the ground, and the "conditional" case where the source
is assumed (correctly!) to be on the ground.
To be specific, $\sigma \equiv 1$ in the constant case. In the other cases,
a scaled $\sigma $ is used: $\sigma _{\delta }\left( \mu \right) =\delta
\sigma \left( \mu /\delta \right) $. In the probabilistic interpretation, $%
\delta $ may be thought of as a detection limit. When the model value $\mu $
is below the detection limit, the standard deviation of the measured value $%
z $ is exactly $\delta $ in the piecewise linear case, and approximately $%
\delta $ in the smooth case. We use $\delta =10^{-6}$ in this study (a
choice of $\ \delta =10^{-3}$ gives no noticeable difference).
\subsection{Results for $\protect\lambda =10^{-2}$, $\protect\delta =10^{-6}$%
}
Setting $\lambda =10^{-2}$ and $\delta =10^{-6}$ yield the following results.
\subsubsection{3D estimates}
\begin{figure}[H]
\centering
\includegraphics[scale=0.15]{fig15.jpg}
\includegraphics[scale=0.15]{fig16.jpg}
\includegraphics[scale=0.15]{fig17.jpg}
\end{figure}
The average errors (averaged over all data sets) is presented in the
following table:
\begin{tabular}{lllll}
& Smooth & Piecewise linear & Linear & Constant \\
$x$ & $1.5$ & $1.5$ & $1.5$ & $1.42$ \\
$y$ & $0.117$ & $0.117$ & $0.117$ & $0.155$ \\
$z$ & $0.214$ & $0.214$ & $0.117$ & $0.251$ \\
$\log _{10}\left( q\right) $ & $1.46$ & $1.46$ & $1.63$ & $1.83$%
\end{tabular}
\subsubsection{2D estimates - assuming the source is on the ground}
\begin{figure}[H]
\centering
\includegraphics[scale=0.15]{fig18.jpg}
\includegraphics[scale=0.15]{fig19.jpg}
\end{figure}
The average errors (averaged over all data sets) is presented in the
following table:
\begin{tabular}{lllll}
& Smooth & Piecewise linear & Linear & Constant \\
$x$ & $1.5$ & $1.5$ & $1.5$ & $1.41$ \\
$y$ & $0.117$ & $0.117$ & $0.117$ & $0.168$ \\
$\log _{10}\left( q\right) $ & $1.09$ & $1.09$ & $1.09$ & $0.853$%
\end{tabular}
\subsection{Results for $\protect\lambda =1$, $\protect\delta =10^{-6}$}
Increasing the penalty for large emission rates to $\lambda =1$, keeping $%
\delta =10^{-6}$, reduces the errors.
\subsubsection{3D estimates}
\begin{figure}[H]
\centering
\includegraphics[scale=0.15]{fig25.jpg}
\includegraphics[scale=0.15]{fig26.jpg}
\includegraphics[scale=0.15]{fig27.jpg}
\end{figure}
The average errors (averaged over all data sets) is presented in the
following table:
\begin{tabular}{lllll}
& Smooth & Piecewise linear & Linear & Constant \\
$x$ & $1.29$ & $1.29$ & $1.29$ & $1.42$ \\
$y$ & $0.0849$ & $0.0849$ & $0.0849$ & $0.22$ \\
$z$ & $0.0825$ & $0.0825$ & $0.0849$ & $0.22$ \\
$\log _{10}\left( q\right) $ & $0.873$ & $0.873$ & $1.48$ & $1.82$%
\end{tabular}
\subsubsection{2D estimates - assuming the source is on the ground}
\begin{figure}[H]
\centering
\includegraphics[scale=0.15]{fig28.jpg}
\includegraphics[scale=0.15]{fig29.jpg}
\end{figure}
The average errors (averaged over all data sets) is presented in the
following table:
\begin{tabular}{lllll}
& Smooth & Piecewise linear & Linear & Constant \\
$x$ & $1.35$ & $1.35$ & $1.35$ & $1.41$ \\
$y$ & $0.101$ & $0.101$ & $0.101$ & $0.168$ \\
$\log _{10}\left( q\right) $ & $0.969$ & $0.969$ & $0.969$ & $0.853$%
\end{tabular}
\section{Discussion and Conclusions}
The bilevel optimisation method is well suited to linear inverse atmospheric
dispersion problems where a single source is emitting at a constant rate:
the problem decouples in a manner where the optimal source strength in each
grid point can be solved for first (follower problem) and then the optimal
location of the source can be determined (leader problem). When gauging how
good a candidate source is, namely how near its candidate sensor readings
are to the observed sensor readings - indeed how near its averaged candidate
sensor readings are to the averaged observed sensor readings, a distance
function is required. The distance function can be weighted, and since
averaged sensor readings are compared a natural choice of weight involves
the variance of the measurements. The simple choice is the let the weight
vary linearly with the model sensor reading. As the model sensor reading $%
\mu $ drops below the detection threshold however it is desired that the
variance of the real measurement $z$ becomes independent of the model sensor
reading $\mu $. The piecewise linear weight remedies this, but it comes at
the price of introducing a non-smoothness. We propose that this shortcoming
is addressed by introducing a smooth threshold weight keeping the benefit of
the piecewise linear weight (giving a variance with a strictly positive
lower bound). As shown in Figure \ref{fig:qualitative_difference} the
non-smoothness of the piecewise linear weight can yield a minimization
problem which is qualitatively different from smooth threshold distance
function and the linear one.
The method has been applied to experimental data from wind tunnel studies of
built up environments. Comparing average errors in the estimated parameters,
the smooth threshold distance function performs as well as the piecewise
linear one and usually better than the linear one. We also note that the
unweighted (constant) distance function is outperforming the others when it
comes to estimating the emission rate in the 2D case where it is a priori
assumed that the source is located on the ground. Increasing the penalty
coefficient $\lambda ,$penalising high emission rates, from $10^{-2}$ to $1$
decreases the average errors. It would be interesting to study how $\lambda $
should be chosen optimally.
\begin{acknowledgement}
This work was conducted within the European Defence Agency (EDA) project
B-1097-ESM4-GP \textquotedblleft Modelling the dispersion of toxic
industrial chemicals in urban environments\textquotedblright\ (MODITIC).
\end{acknowledgement}
| bd45e4caca15d5129a3a5609f2808920a62bb3b0 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Network operators continuously upgrade their networks due to increasing demands for ubiquitous communications. As popularities of cloud computing, Internet of Things (IoT), and 5G mobile communications increase, network traffic is becoming heavy and extremely bursty. Thus, just enlarging the network capacity is not an economical choice, while neglecting it will strongly affect the QoS. Also, emerging applications, such as online gaming, data backups and virtual machine migrations, result in heterogeneous demands in telecom networks. Today's network users request more customized services, such as Virtual Private Network (VPN) and Video on Demand (VoD), and more differentiated demands with different prices \cite{he2005}. For some of the traffic which is delay-insensitive and can accept some compromise in bandwidth or other aspects, it can be preempted by more ``important" requests when the network becomes congested. Thus, to maintain cost-effectiveness and customers' loyalty, network operators can provide different grades of service besides sufficient bandwidth, instead of trying to support all the traffic without distinction \cite{muhammad2013}.
To address these problems, \emph{degraded provisioning} is proposed to provide a degraded level of service when network congestion occurs instead of no service at all. Generally, degraded provisioning has two approaches: 1) keep total amount of transferred traffic constant by time prolongation or modulation-level adjustment with immediate service access (QoS-assured), or 2) degrade requested bandwidth without time or modulation compensation, or no guarantee for immediate access (QoS-affected). We focus on QoS-assured degraded provisioning in this study.
In a multi-layer network, QoS-assured degradation has different implementation methods in different layers. In electric layer, for a delay-insensitive and degradation-tolerant request, when network bandwidth is scarce, we degrade its transmission rate to enlarge the available bandwidth and extend its holding time based on the premise that total traffic amount is unchanged. Note that a traffic request cannot be degraded arbitrarily, and it is constrained by a given deadline \cite{fawaz2010}. In elastic optical layer, degradation refers to decreasing the number of occupied spectrum slots of a lightpath and raising the modulation level to guarantee the capacity. In OFDM-based elastic optical networks, modulation level can be dynamically reconfigured in DSP and DAC/ADC via software \cite{zhang2013}. But optical degradation has a constraint that higher-order modulation has shorter transmission reach \cite{bocoi2009}.
Due to the flexibility enabled by degraded provisioning, there exist many studies on this topic in different kinds of optical networks. In WDM networks, Roy \emph{et al.} \cite{roy2008} studied degraded service using multipath routing in a QoS-affected way. Zhang \emph{et al.} \cite{zhang2010} studied reliable multipath provisioning, exploiting flexibility in bandwidth and delay. Andrei \emph{et al.}\cite{Andrei2010} proposed a deadline-driven method to flexibly provision services without immediate access. Savas \emph{et al.} \cite{sedef2014} introduced a dynamic scheme to reduce blocking by exploiting degraded-service tolerance in a QoS-affected way, and they also applied this method to increase network survivability \cite{sedef}. In Mixed-Line-Rate (MLR) networks, Vadrevu \emph{et al.} \cite{vadrevu2014} proposed a QoS-affected degradation scheme using multipath routing considering minimum-cost network design. But the ITU-T grid limit puts constraints on optical layer flexibility.
As a major development of optical technology, elastic optical networking enables more flexibility in optical modulation and spectrum allocation. Distance-adaptive spectrum allocation \cite{jinno2010} is a similar approach as optical degradation, but its limitations are that the modulation format of a lightpath is configured at one time and cannot be adjusted based on the fluctuation of traffic. Gkamas \emph{et al.} \cite{Gkamas2015} proposed a dynamic algorithm for joint multi-layer planning in IP-over-flexible-optical networks without considering dynamic adjustment of lightpath modulation. Recent progress in modulation format conversion \cite{huang2012} enables all-optical OOK to 16QAM adjustment, and its advantages in elastic optical networking were demonstrated in Yin \emph{et al.} \cite{Yin2014} with modulation-formats-convertible nodes. Thus, degraded provisioning can be extended to the elastic optical layer, and this important issue has not been fully understood, with no previous studies.
We summarize our contributions as follows: 1) to the best of our knowledge, this is the first investigation on QoS-assured degraded provisioning problem in multi-layer networks with optical elasticity; 2) we propose an enhanced multi-layer network architecture and solve the complex dynamic degraded routing problem; and 3) we further propose novel dynamic heuristic algorithms in both layers to achieve degraded resource allocation. Results show that a significant increase of service acceptance can be achieved by multi-layer degradation.
\thispagestyle{fancy}
\chead{\small Accepted, IEEE Globecom 2016, Optical Networks and Systems (ONS) Symposium, for ArXiv only.}
\section{Dynamic Degraded Provisioning Scheme}
We decompose the online degraded provisioning problem into two subroutines: 1) degraded routing, and 2) degraded resource allocation.
\subsection{Degraded Routing}
Degraded routing solves the subproblem of degraded-route computation when conventional routing cannot be performed due to resource shortage. Optical degraded routing acts similarly as electric degraded routing. The term \emph{request} here refers to the \emph{lightpath request} in optical layer, and the \emph{service request} in electric layer. There are two concerns: route hops (RH) and potential degraded requests (PDR). RH denotes the amount of resources occupied by the new request, while PDR denotes how many existing requests may be affected.
We define a link in any layer of the multi-layer network as a tuple: $V_{ij,k}=(\mathbf{\Theta}, C)$, which is the $k^{th}$ link from $i$ to $j$. $\mathbf{\Theta}$ is a set that contains existing requests routed on this link, and $C$ is the available capacity of this link. $r_{n}$ is a request, and $\mathcal{P}_{r_n}$ is a degraded route for $r_{n}$ in electric or optical layer.
\vspace{-0.3em}
\begin{equation}
\small
\mathcal{P}_{r_n}=\{V_{ij,k}|r_n \in V_{ij,k}.\mathbf{\Theta}\}
\vspace{-0.4em}
\end{equation}
We introduce two metrics to evaluate the route $\mathcal{P}_{r_n}$. Note that the $|\cdot|$ operation returns the number of elements in a set.
\vspace{-0.5em}
\begin{equation}
\small
\mathcal{N}_{RH}=|\mathcal{P}_{r_{n}}|
\vspace{-0.3em}
\end{equation}
\begin{equation}
\small
\mathcal{N}_{PDR}=|\bigcup^{|\mathcal{P}_{r_{n}}|}_{c=1}\mathcal{P}_{r_{n}}[c].\mathbf{\Theta}|
\vspace{-0.5em}
\end{equation}
To calculate a route for minimizing RH, Dijkstra algorithm is applied. However, minimizing PDR is not that easy, because the minimizing-PDR problem aims to obtain a route that has the smallest PDR among all possible routes between a given source-destination $(s,d)$ pair. A straight-forward idea is to list all possible routes between a given $(s,d)$ pair and compare their PDR. But the complexity of this process is $O((N-2)!)$ ($N$ denotes the number of nodes). Here, we propose the enhanced multi-layer network architecture by introducing the auxiliary \emph{service layer}, which lies directly above the electric layer, and we solve the problem in polynomial time.
In the enhanced multi-layer architecture (Fig. 1), all nodes in the optical layer are mapped to upper two layers. There are two kinds of directional weighed links, i.e., request link and resource link. Request link weight is equal to a given large number times a binary that indicates whether there are existing requests between the node pair. Resource link weight is binary, which means if there is sufficient resource for this request. Note that, if the upper layer has isolated nodes (no connected edges), and the isolated node happens to be the source or destination of the request, then we replace the isolated node with the originating or terminating nodes of lightpaths (or requests) that bypass the isolated node. For example, in Fig. 1, $e^*$ is bypassed by request \#2 ($a^*$-$e^*$-$c^*$) on service layer. Thus, if we want to compute a degraded route on electric layer with minimized PDR for a new request ($a^*,e^*$), we should replace $e^*$ with $c^*$, and execute a shortest-path algorithm on service layer for the auxiliary request ($a^*,c^*$). Therefore, the \emph{minimizing-PDR problem} on one layer is transformed into a \emph{weighed shortest-path problem} on the upper layer. Finally, an actual route will be acquired based on the computed shortest path on upper layer.
\begin{figure}[t]
\centering
\vspace {-1em}
\includegraphics[width=2.5in]{fig1-multilayer.pdf}
\vspace {-0.5em}
\caption{Illustration of enhanced multi-layer network architecture.}
\label{Algorithm graph}
\vspace {-1.7em}
\end{figure}
\begin{figure*}[t]
\centering
\vspace {-1.8em}
\subfigure[Dynamic electric degradation.]{
\label{fig:subfig:a}
\hspace{-3em}
\includegraphics[width=2in]{fig2-e_degrade.pdf}}
\hspace{0.5em}
\subfigure[Dynamic optical degradation: on a fiber.]{
\label{fig:subfig:a}
\includegraphics[width=2in]{fig2-o_degrade}}
\hspace{0.5em}
\subfigure[Dynamic optical degradation: along a route.]{
\label{fig:subfig:a}
\includegraphics[width=2.4in]{fig2-o_degrade_route}}
\hspace{-2.5em}
\vspace{-0.4em}
\caption{Illustration of multi-layer degradation principle in degraded resource allocation stage.}
\label{fig:subfig}
\vspace {-1.4em}
\end{figure*}
\vspace {-1em}
\algsetup{ linenosize=\normalsize, linenodelimiter=. }
\noindent %
\rule [-0.3em] {\linewidth} {0.75pt} \\%
\noindent \textbf{Algorithm I}: Minimizing-PDR Algorithm
\\ \rule [0.8em]{\linewidth} {0.4pt} \\%
\vspace {-2.4em}
\begin{algorithmic}[1]
\footnotesize
\STATE Upper-layer topology$\{t_{i,j}\}=\{0\}$; current layer topology $\{t'_{i,j}\}$; $M$ is a large number; request ($s,d$);
\FOR {all requests $r$ \textbf{in} current layer}
\STATE $t_{r.source,r.destination}=M$;
\ENDFOR
\FOR {all $i,j$}
\STATE $t_{i,j}=t'_{i,j}$;
\ENDFOR
\IF {$\sum_{j \in N} t_{s,j}+\sum_{i\in N} t_{i,s}=0\ || \sum_{j \in N} t_{d,j}+\sum_{i\in N} t_{i,d}=0$}
\STATE replace $s$ (or $d$) with the originating (or terminating) nodes of lightpaths (or request) that running bypass $s$ (or $d$);
\ENDIF
\STATE run shortest-path algorithm on upper-layer topology $\{t_{i,j}\}$, acquire $\mathcal{P}$;
\FOR {all links $V_{m,n}$ \textbf{in} $\mathcal{P}$}
\STATE find a $V_{m,n}$ with the shortest hops (if $m$ or $n$ is the replaced node, use the isolated node to count hops) in lower layer; acquire its route;
\ENDFOR
\STATE combine all the acquired routes together, cancel loops and return it;
\end{algorithmic}
\vspace {-1em} %
\rule {\linewidth} {0.75pt}
Now, we introduce two policies of degraded routing:
\subsubsection{Minimize Route Hops (MinRH)}
We manage to minimize RH as a primary goal, and then minimize PDR.
\subsubsection{Minimize Potential Degraded Requests (MinPDR)}
We try to minimize PDR first, then we minimize RH.
\vspace{-0.3em}
\subsection{Degraded Resource Allocation}
When a degraded route $\mathcal{P}^e$ (electric) or $\mathcal{P}^o$ (optical) is acquired, we need to decide which request or requests to degrade, and how much to degrade them. In the multi-layer network, degraded resource allocation refers to different operations in different layers, which should be further studied separately.
\subsubsection{Electric Degraded Bandwidth Allocation (ED-BA)}
We propose the ED-BA algorithm based on a degraded route $\mathcal{P}^e$. A traffic service request on electric layer is defined as: $r_{n}=(s, d, bw, t, \tau, \eta, \rho)$, which mean source, destination, bandwidth, arrival time point, holding time, prolongation deadline, and priority, respectively. We define a function $AS(S, k)$, which sorts elements in set $S$ in ascending order of $k$.
\vspace {-1em}
\algsetup{ linenosize=\normalsize, linenodelimiter=. }
\noindent %
\rule [-0.3em] {\linewidth} {0.75pt} \\%
\noindent \textbf{Algorithm II}: ED-BA Algorithm
\\ \rule [0.8em]{\linewidth} {0.4pt} \\%
\vspace {-2.4em}
\begin{algorithmic}[1]
\footnotesize
\STATE Current time $t_c$, arriving request $r_0$, $flag$ = 1;
\FOR {all links $V_n$ \textbf{in} $\mathcal{P}^e$}
\IF{$V_n.C < r_0.bw$}
\STATE $PDL(V_n)\leftarrow \{r_0\}$; \quad \emph{/*potential degraded links*/}
\FOR {all requests $r$ \textbf{in} $V_n.\mathbf{\Theta}$}
\IF {$r.\rho \leq r_0.\rho$}
\STATE $PDL(V_n).pushback(r)$;
\ELSE
\STATE continue;
\ENDIF
\ENDFOR
\IF {$\sum_{u \in PDR(V_n)} u.bw\times\frac{u.\eta}{(u.\tau+u.\eta)} \geq r_0.bw$}
\STATE $ac\_bw\leftarrow$ 0; \ \emph{/*accumulate available bandwidth*/}
\FOR {all requests $x$ \textbf{in} $AS(PDL(V_n), priority)$}
\STATE degrade $x.bw$ to its maximum extent $x.bw'$, s.t. $\frac{(x.bw\times x.\tau-(t_c-x.t)\times x.bw)}{x.bw'}+t_c-x.t \leq x.\eta+x.\tau$;
\STATE $ac\_bw= ac\_bw+x.bw-x.bw'$;
\IF {$ac\_bw > r_0.bw$}
\STATE request $r_0$ routed successfully on $V_n$; break;
\ENDIF
\ENDFOR
\ELSE
\STATE request $r_0$ blocked; $flag$ = 0; break;
\ENDIF
\ELSE
\STATE continue;
\ENDIF
\ENDFOR
\IF {$flag$ == 1}
\STATE request $r_0$ is routed successfully;
\ENDIF
\end{algorithmic}
\vspace {-1em} %
\rule {\linewidth} {0.75pt}
Fig. 2(a) shows the basic principle of the ED-BA algorithm, that requests with higher priorities can ``preempt" those with no higher priorities. Here, the term ``preempt" means that some existing requests are degraded in transmission rate due to their relatively low priorities. Meanwhile, we manage to degrade the minimal number of requests to provide just-enough bandwidth.
\thispagestyle{fancy}
\chead{\small Accepted, IEEE Globecom 2016, Optical Networks and Systems (ONS) Symposium, for ArXiv only.}
\subsubsection{Optical Degraded Modulation and Spectrum Allocation (OD-MSA)}
In an elastic optical network, optical degradation refers to the reduction of occupied-spectrum-slot number of a lightpath, and raise lightpath's modulation level to ensure lightpath capacity under the modulation-distance constraint.
Figs. 2(b) and 2(c) show how optical degradation works. The number of spectrum slots in a fiber is $B$. $S_f$ is a binary bitmask that contains $B$ bits to record the availability of each spectrum slot in fiber $f$, e.g., $S_f[p]=1$ means the $p^{th}$ slot is utilized. A lightpath is defined as a tuple: $L=(f, \xi_l, \xi_r, \eta, \delta)$, which denotes the fiber that the lightpath is routed through, left and right indices of occupied spectrum slots, modulation level, and lightpath distance, respectively. Note that $(\xi_r-\xi_l+1)\cdot log_2\eta$ should be constant when performing optical degradation. A lightpath request is defined as a tuple: $l_n=\{i, j, \theta\}$, which denotes source, destination, and requested bandwidth in spectrum slots. We define a function $Q(a)$ to get the transmission reach of modulation level $a$.
We define Available Spectrum Slots Intersection (ASSI), which is a set of slots that are available all along the optical degraded route $\mathcal{P}^o$, to evaluate the available resources for the new lightpath before optical degradation. Note that the operator $\lor$ represents the logical OR operation, and $p \in [1,B]$.
\vspace {-0.5em}
\begin{equation}
\small
ASSI=\{S_f[p] | \sum_{f =1}^{|\mathcal{P}^o|-1}S_f[p] \lor S_{f+1}[p]=0\}
\vspace {-0.5em}
\end{equation}
We define Slot Border Through Lightpaths (SBTL) to evaluate whether a slot border locates inside the occupied spectrum of a lightpath. $w$ denotes index of spectrum slot borders, and there are $B+1$ borders, thus $w \in [1, B+1]$. We define a decision function $D(x)$ which is equal to 0 if $x$ is positive, or returns 1 if $x$ is negative.
\vspace {-0.5em}
\begin{equation}
\small
SBTL=\{w|\sum_{L:L.f\in\mathcal{P}^o}D((w-\frac{1}{2}-L.\xi_l)(w-\frac{1}{2}-L.\xi_r))=0\}
\vspace {-0.5em}
\end{equation}
When a degradation location (available slots or a slot border) is found in the spectrum by ASSI or SBTL, the potential degraded lightpaths are those on both sides of the location. We first try to degrade the left one ($L_1$), which is called \emph{single-side degradation}. And if only $L_1$ degradation cannot provide enough slots, we continue to degrade the right one ($L_2$), which is called \emph{double-side degradation}. Note that, when there are multiple possibilities (choosing an element in ASSI or SBTL), a First-Fit policy is applied to choose the one with smaller index to reduce spectrum fragmentation.
On new lightpath establishment, we use the threshold-based grooming approach, which has been demonstrated to achieve lower blocking probability than the fixed-grid IP-over-WDM and elastic-spectrum non-grooming approaches \cite{wan2012}.
\vspace {-1em}
\algsetup{ linenosize=\normalsize, linenodelimiter=. }
\noindent %
\rule [-0.3em] {\linewidth} {0.75pt} \\%
\noindent \textbf{Algorithm III}: OD-MSA Algorithm
\\ \rule [0.8em]{\linewidth} {0.4pt} \\%
\vspace {-2.4em}
\begin{algorithmic}[1]
\footnotesize
\STATE Arriving lightpath request $l_0$;
\FOR {all fibers $f_n$ \textbf{in} $\mathcal{P}^o$}
\STATE scan the spectrum to acquire $ASSI$;
\IF{$|ASSI|>0$}
\STATE choose consecutive available slots $[l,r]$ with the largest $r-l$ value in $ASSI$;
\STATE scan the spectrum to acquire $L_1$ and $L_2$ ($L_1.\xi_r==l-1\ \&\&\ L_2.\xi_l==r+1$);
\IF {$Q(log_2a+1)< L_1.\delta \leq Q(log_2a)$ and $Q(log_2b+1)< L_2.\delta \leq Q(log_2b)$}
\STATE $L_1.\eta =a$; \quad\quad \emph{/*degrade $L_1$ to modulation level $a$*/}
\IF {$(L_1.\xi_r-L_1.\xi_l+1)(1-log_2(\eta_0-a))+(l-r+1)\geq l_0.\theta$}
\STATE setup a new lightpath with $l_0.spt$ slots, starting from $L_1.\xi_l+(L_1.\xi_r-L_1.\xi_l+1)log_2(\eta_0-a)$;
\ELSIF{$(L_1.\xi_r-L_1.\xi_l+1)(1-log_2(\eta_0-a))+(l-r+1)+(L_2.\xi_r-L_2.\xi_l+1)(1-log_2(\eta_0-b))\geq l_0.\theta$}
\STATE $L_2.\eta =b$; \quad\emph{/*degrade $L_2$ to modulation level $b$*/}
\STATE setup a new lightpath with $l_0.\theta$ slots, starting from $L_1.lft+(L_1.\xi_r-L_1.\xi_l+1)log_2(\eta_0-a)$;
\ELSE
\STATE request $l_0$ blocked; break;
\ENDIF
\ENDIF
\ELSIF {$|SBTL|>0$}
\STATE choose smallest $w$ in $SBTL$, and perform sentence 7 to 17 (here, let $l=r+1=w$);
\ELSE
\STATE request $l_0$ blocked; break;
\ENDIF
\ENDFOR
\end{algorithmic}
\vspace {-1em} %
\rule {\linewidth} {0.75pt}
\vspace {-0.6em}
\subsection{Complexity Analysis}
In degraded routing stage, minimizing-RH problem can be solved with Dijkstra algorithm with $O(N^2)$ complexity in a $N$-node topology. But the minimizing-PDR problem may have a complexity of $O(N^3)$ if both source and destination nodes of the request are isolated nodes\footnote{Here, the isolated nodes in request should be replaced by other nodes of the upper layer in path computation, and the problem becomes an all-pairs shortest-path one, which should be solved via Floyd algorithm ($O(N^3)$).}.
In degraded resource allocation stage, the worst case is that the degraded route goes through every node of the topology, and the complexity is $O(N)$. In ED-BA algorithm, we suppose that the maximum number of existing requests on each link is $R$ (related to traffic load), and the time complexity is $O(NR)$. In OD-MSA algorithm, the complexity\footnote{Here, $B$ is the number of spectrum slots, which is a constant parameter.} is $O(NB^2)$.
Hence, the complexity of the proposed dynamic degraded provisioning scheme is $O(N^3+NR)$, and is suitable for online decision making in dynamic traffic accommodation.
\thispagestyle{fancy}
\chead{\small Accepted, IEEE Globecom 2016, Optical Networks and Systems (ONS) Symposium, for ArXiv only.}
\section{Illustrative Numerical Evaluations}
\begin{figure}[!t]
\centering
\vspace {-1em}
\includegraphics[scale=0.22]{fig3-USNet}
\vspace {-1em}
\caption{USNet topology with fiber length in kilometers marked on links.}
\label{Algorithm graph}
\vspace {-1.5em}
\end{figure}
\subsection{Experimental Setup}
Table I summarizes the parameters of different modulation formats based on \cite{bocoi2009} \cite{zhu2013}. We consider the USNet topology (Fig. 3) for dynamic performance simulation. All fibers are unidirectional with 300 spectrum slots, and spectrum width of each slot is 12.5 GHz. Traffic requests are generated between all node pairs, and characterized by Poisson arrivals with negative exponential holding times. Granularities of requests are distributed independently and uniformly from 5 Gbps to 150 Gbps. The maximum acceptable ratio for transmission-rate degradation is uniformly distributed in [25\%, 100\%] \cite{sedef}. There are 5 priorities with equal amount each. The lightpath establishment threshold for grooming is chosen as 150 Gbps, which is equal to the largest request bandwidth and performs the best \cite{wan2012}. An event-driven dynamic simulator has been developed to verify the effectiveness of the heuristic algorithms. Six degradation policies, i.e. OE-MinPDR, O-MinPDR, E-MinPDR, OE-MinRH, O-MinRH, E-MinRH (OE: both-layer degradation, O: optical degradation only, E: electric degradation only) are studied.
\begin{figure*}[!t]
\centering
\vspace{-1em}
\hspace{-2em}
\subfigure[Requests with all priorities.]{
\label{fig:subfig:a}
\includegraphics[width=2.2in]{fig4-overall}}
\hspace{0.2em}
\subfigure[Requests with highest priority (\#5).]{
\label{fig:subfig:a}
\includegraphics[width=2.2in]{fig4-high-priority}}
\hspace{0.2em}
\subfigure[Requests with lowest priority (\#1).]{
\label{fig:subfig:a}
\includegraphics[width=2.2in]{fig4-low-priority}}
\hspace{-2em}
\vspace{-0.5em}
\caption{Bandwidth blocking probability vs. traffic load: different degradation policies.}
\label{fig:subfig}
\vspace {-1em}
\end{figure*}
\begin{figure*}[!t]
\centering
\hspace{-2.5em}
\subfigure[Instantaneous network throughput (MinPDR).]{
\label{fig:subfig:a}
\includegraphics[width=2.3in]{fig5-trans-PDR}}
\hspace{0.15em}
\subfigure[Instantaneous network throughput (MinRH).]{
\label{fig:subfig:a}
\includegraphics[width=2.3in]{fig5-trans-RH}}
\hspace{0.15em}
\subfigure[Instantaneous BBP.]{
\label{fig:subfig:a}
\includegraphics[width=2.2in]{fig5-trans-bbp}}
\vspace{-0.5em}
\hspace{-2em}
\caption{Transient analysis within 1.5 $hour$ (30 $Erlang$ per node, $\lambda=300\ hour^{-1}, \mu =10\ hour^{-1}$): different degradation policies.}
\label{fig:subfig}
\vspace {-1.5em}
\end{figure*}
\begin{table}[htbp]
\vspace {-0.6em}
\renewcommand{\arraystretch}{1}
\vspace {-1em}
\caption{Modulation format vs. Data rate vs. Transmission reach}
\label{table_example}
\centering
\vspace {-1em}
\begin{tabular}{m{71pt}<{\centering} m{50pt}<{\centering} m{23pt}<{\centering} m{23pt}<{\centering} m{23pt}<{\centering}}
\hline
Modulation format & BPSK~(default) & QPSK & 8QAM & 16QAM\\
\hline
Modulation~level & 2 & 4 & 8 & 16 \\
Bits~per~symbol & 1 & 2 & 3 & 4\\
Slot~bandwidth~(GHz)& 12.5 & 12.5 & 12.5 & 12.5\\
Data~rate~(Gbps)& 12.5 & 25 & 37.5 & 50\\
Transmission~reach~(km) & 9600 & 4800 & 2400 & 1200\\
\hline
\end{tabular}
\vspace {-1.5em}
\end{table}
\subsection{Dynamic Analysis}
Fig. 4 shows the bandwidth blocking probability (BBP) advantages of our proposed scheme over conventional scheme (threshold-based grooming \cite{wan2012}, no degradation). Fig. 4(a) shows the overall performance of all requests, and we can find there is a crossing point between optical degradation and both-layer degradation. In low-load area (26-34 $Erlang$), both-layer degradation (OE-MinPDR, OE-MinRH) performs the best, up to two orders of magnitude, while in high-load area (36-44 $Erlang$), optical-layer degradation (O-MinPDR, O-MinRH) performs the best. The reason is that, in high-load conditions, electric degradation (E-MinPDR, E-MinRH) achieves worse BBP than no degradation, which affects the blocking reduction by optical degradation in both-layer degradation. Figs. 4(b) and 4(c) show the BBP performance of requests in the highest and lowest priority. And we can conclude that all degradation policies can achieve significant blocking reduction for highest priority, while, for lowest priority, the blocking performance acts similar as requests with all priorities do.
We also observe some common patterns in these three graphs. First, optical degradation performs almost the same regardless of priorities, because optical degradation does not involve service priorities as electric degradation does. Second, MinPDR performs better in optical-related degradations (both-layer degradation and optical degradation), while MinRH performs better only in electric degradation. This is because the route MinPDR returns tends to have a smaller number of existing requests, which increases the elements in ASSI or SBTL, thus increasing possibility of successful optical degradation. Also, the route that MinRH returns tends to have more existing requests, which increases the possibility of successful electric degradation. \emph{Actually, different mechanisms of optical degradation and electric degradation determine that MinPDR performs better for optical degradation, while MinRH suits electric degradation better}. The result that both-layer degradation and optical degradation performs similarly reveals that optical degradation has stronger influence on blocking reduction because it can enlarge the network capacity by high-order modulation, while electric degradation just deals with the bandwidth-time exchange to trade time for bandwidth.
\vspace {-0.5em}
\subsection{Transient Analysis}
We conduct transient analysis on instantaneous network throughput and BBP. From Figs. 5(a) and 5(b), we obtain similar conclusions as the dynamic evaluations, that optical-related degradation achieves better compliance with the offered load in MinPDR, while electric degradation accomplishes better improvements in MinRH. Fig. 5(c) shows the instantaneous BBP variance over time, and we observe that different levels of blocking reduction can be achieved by different degradation policies. Both-layer degradation policies have the largest blocking reduction, and OE-MinPDR performs even better (almost zero blocking).
\thispagestyle{fancy}
\chead{\small Accepted, IEEE Globecom 2016, Optical Networks and Systems (ONS) Symposium, for ArXiv only.}
\vspace {-0.2em}
\section{Conclusion}
In this work, we investigated dynamic QoS-assured degraded provisioning in service-differentiated multi-layer networks with optical elasticity. We proposed and leveraged the enhanced multi-layer architecture to design effective algorithms for network performance improvements. Numerical evaluations showed that we can achieve significant blocking reduction, up to two orders of magnitude. We also conclude that optical-related degradation achieves better with MinPDR, while electric degradation has lower blocking with MinRH due to different mechanisms of multi-layer degradations.
\IEEEtriggeratref{20}
\bibliographystyle{IEEEtran}
\vspace {-0.3em}
| dc6744abb01cd666eaa00cb77e470db20d5accf3 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
\subsection{The main theorem}
Let $F$ be a totally real field, $p$ be an odd prime, and ${\mathcal O}$ the ring of integers of a finite extension of $\Q_p$. In proving automorphy of geometric representations $\rho:G_F \rightarrow GL_2({\mathcal O})$ which are residually automorphic, there is an assumption made (the Taylor-Wiles hypothesis) that the residual representation $ {\overline{\rho}} $ is irreducible when restricted to $F(\zeta_p)$. In \cite{Thorne}, Thorne has recently weakened this assumption in many cases to simply asking that $ {\overline{\rho}} $ itself is irreducible but making the auxiliary assumption that the quadratic subfield of $F(\zeta_p)/F$ is totally real. \\ \\
Under other assumptions on $ {\overline{\rho}} $, but still allowing that $ {\overline{\rho}} |_{G_{F(\zeta_p)}}$ is reducible, we prove automorphy lifting results wherein we do not assume that the quadratic subfield $K$ of $F(\zeta_p)/F$ is totally real. The assumption on $ {\overline{\rho}} $ is that there is a (``level raising'') place $v$ of $F$ that splits in $K$, such that the ratio of the eigenvalues of $ {\overline{\rho}} ({\rm Frob}_v)$ is $q_v$ with $q_v$ not 1 modulo $p$. This assumption is automatic when $K/F$ is totally real, as exploiting the oddness of $ {\overline{\rho}} $ we can take any $v$ such that $ {\overline{\rho}} ({\rm Frob}_v)$ is conjugate to the image of complex conjugation. \\
The main theorem we prove is the following:
\begin{theorem} Let $F$ be a totally real number field, let $p$ be an odd prime, and let $\rho: G_F \to \GL_2(\overline{\Q}_p)$ be a continuous representation satisfying the following:
\begin{itemize}
\item[(1)] The representation $\rho$ is almost everywhere unramified.
\item[(2)] For each $v|p$ of $F$, the local representation $\rho|_{G_{F_v}}$ is de Rham. For each embedding $\tau: F \hookrightarrow \overline{\Q}_p$, we have $\text{HT}_{\tau}(\rho) = \{0,1\}$.
\item[(3)] For each complex conjugation $c \in G_F$, we have $\det \rho(c) = -1$.
\item[(4)] The residual representation $\bar{\rho}$ is absolutely irreducible, but $\bar{\rho}|_{G_{F(\zeta_p)}}$ is a direct sum of two distinct characters. Further suppose that if $K$ is the unique quadratic subfield of $F(\zeta_p)/F$ and $\bar{\gamma}: G_K \to k^{\times}$ is the ratio of the two characters, then we have $F(\zeta_p) \not \subset K(\bar{\gamma}\epsilon^{-1}) \cap K(\bar{\gamma}\epsilon)$.
\end{itemize}
Then $\rho$ is automorphic: there exists a cuspidal automorphic representation $\pi$ of $\GL_2(\mathbb{A}_F)$ of weight $2$, an isomorphism $\iota: \overline{\Q}_p \to \mathbb{C}$, and an isomorphism $\rho \cong r_{\iota}(\pi)$.
\end{theorem}
The hypothesis that $F(\zeta_p) \not \subset K(\bar{\gamma}\epsilon^{-1}) \cap K(\bar{\gamma}\epsilon)$ is equivalent to the existence of a place $v$ of $F$ such that $q_v$ is not 1 modulo $p$, $v$ splits in $K$, and $ {\overline{\rho}} ({\rm Frob}_v)$ has eigenvalues with ratio $q_v$. Furthermore, as remarked above, this hypothesis is automatic when $K$ is totally real as one may, using the oddness of $ {\overline{\rho}} $, take any $v$ such that $ {\overline{\rho}} ({\rm Frob}_v)$ is conjugate to $ {\overline{\rho}} (c)$ and $q_v$ is -1 modulo $p$. Thus the theorem generalizes the main theorem of \cite{Thorne}, and in fact its proof is essentially the same as \cite{Thorne}. In loc. cit., the author synthesizes two methods:
-- the patching method of Taylor-Wiles, which relies on using places $v$ of $F$ such that $q_v$ is $1$ modulo $p^N$ and $ {\overline{\rho}} ({\rm Frob}_v)$ has distinct eigenvalues to kill certain elements of the mod $p$ dual Selmer, and
-- a method due to Khare, which proves automorphy of $\rho$ by ``$p$-adic approximation'' using Ramakrishna places $v$, such that $q_v$ is not $1$ modulo $p$ and $\rho_N({\rm Frob}_v)$ has eigenvalues with ratio $q_v$, to perform the task of killing certain elements of mod $p$ dual Selmer which cannot be killed with Taylor-Wiles primes. Here $\rho_N$ denotes the mod $p^N$ reduction of $\rho$. \\
The Ramakrishna places $v$ that Thorne used were such that $\rho_N({\rm Frob}_v)$ is conjugate to the image of complex conjugation. To ensure that these places could effectively kill the troublesome part of mod $p$ dual Selmer, he had to impose that the quadratic subfield of $F(\zeta_p)/F$ be totally real, which implies $p \equiv 1 \Mod 4$. \\
The central new remark of this paper is that if we assume there is a place $v$ of $F$ that splits in $K$, such that the ratio of the eigenvalues of $ {\overline{\rho}} ({\rm Frob}_v)$ is $q_v$ with $q_v$ not $1$ modulo $p$, then by choosing ``Teichmuller liftings'' one may find, for each $N$, places $v$ such that $q_v$ is not $1$ modulo $p$, $\rho_N({\rm Frob}_v)$ has eigenvalues with ratio $q_v$, and the $v$'s perform the task of killing the troublesome elements of mod $p$ dual Selmer which cannot be killed with Taylor-Wiles primes. We present the proof of this remark in Proposition 2.11, and it serves as a replacement for Proposition 5.20 in \cite{Thorne} which needed that $K$ be totally real. Note that this assumption we make is automatic when $K$ is totally real, which explains why our result generalizes the main theorem of loc. cit. \\
After this remark the proof of our theorem is identical to that of loc. cit. where patching arguments followed by level raising and lowing arguments are used to conclude the automorphy of $\rho$. \\
The case when there is no place $v$ of $F$ that splits in $K$ such that the ratio of the eigenvalues of $ {\overline{\rho}} ({\rm Frob}_v)$ is $q_v$ with $q_v$ not 1 modulo $p$ cannot be addressed by the methods of \cite{Thorne} nor by the modifications that we carry out in this paper. In a forthcoming work \cite{KT} of Khare and Thorne, this case is addressed. Instead of patching and $p$-adic approximation, the authors use patching together with Wiles' numerical isomorphism criterion and calculation of $\eta$-invariants using monodromy arguments.
\subsection{Automorphy of elliptic curves}
We get a modest application to automorphy of elliptic curves over totally real fields.
Current automorphy lifting theorems prove the modularity of elliptic curves $E$ over $F$ which satisfy the ``Taylor-Wiles'' hypothesis at some prime $p \in \{3,5,7\}$, i.e., $\bar{\rho}_{E,p}(G_{F(\zeta_p)})$ is absolutely irreducible for at least one $p \in \{3,5,7\}$. Recently, in \cite{FLHS}, the authors prove the modularity of all elliptic curves over real quadratic fields. Moreover, they show that over any totally real field $F$, there are only finitely many potentially non-modular elliptic curves over $F$, which necessarily do not satisfy the Taylor-Wiles hypothesis at any prime $p \in \{3,5,7\}$. \\ \\
In \cite{Thorne}, Thorne was able to cut down on this list of potentially non-modular elliptic curves by proving a new automorphy lifting theorem and applying it to the above situation of elliptic curves with $p=5$. However, the results in that paper do not deal with the case $p=7$ because the requirement was that the quadratic subfield of $F(\zeta_p)/F$ be totally real (which implies that $p$ is 1 modulo 4). In this note, we try to deal with the cases when this quadratic subfield is not totally real by examining different sorts of auxiliary primes. We use this theorem to prove:
\begin{theorem} Let $F$ be a totally real field, and let $E$ be an elliptic curve over $F$. Suppose:
\begin{itemize}
\item[(1)] $F \cap \Q(\zeta_7) = \Q$.
\item[(2)] $E$ has no $F$-rational $7$-isogeny.
\item[(3)] Either $\bar{\rho}_{E,7}(G_{F(\zeta_7)})$ is absolutely irreducible, or it is reducible and is conjugate to a subgroup of a split Cartan subgroup of $\GL_2(\F_7)$.
\end{itemize}
Then $E$ is modular.
\end{theorem}
We further note that one may also deduce by similar arguments the automorphy of elliptic curves $E$ defined over a totally real field $F$, with a prime $p$ such that $E$ has no $F$-rational $p$-isogeny, $\rho_{E,p}(G_F)$ normalizes a split Cartan subgroup of $\GL_2(\F_p)$, and $E$ has bad semistable reduction at a place $v$ such that
$q_v$ is not 1 mod $p$ and $v$ splits in the quadratic subfield of $F(\zeta_p)/F$.
\subsection{Acknowledgements}
I want to thank Chandrashekhar Khare for many useful discussions and his invaluable guidance regarding the methods used in this paper.
\section{Galois Theory}
\subsection{Global Deformation Problem}
Let $F$ be a totally real number field. Let $p$ be an odd prime, and let $E$ be a coefficient field. Fix a continuous, absolutely irreducible representation $\bar{\rho} \colon G_F \to \GL_2(k)$ and a continuous character $\mu: G_F \to \Sc{O}^{\times}$ lifting $\det \bar{\rho}$. We assume $k$ contains the eigenvalues of all elements in the image of $\bar{\rho}$.\\ \\ Let $S$ be a finite set of finite places of $F$, containing the places dividing $p$ and the places at which $\bar{\rho}$ and $\mu$ are ramified. For $v \in S$, fix a ring $\Lambda_v \in \CNL_{\Sc{O}}$. Define $\Lambda = \hat{\otimes}_{v \in S} \Lambda_v$, so that $\Lambda \in \CNL_{\Sc{O}}$. \\ \\ If $v \in S$, we write $\Sc{D}_v^{\Box} \colon \CNL_{\Lambda_v} \to \text{Sets}$ for the functor which takes $R \in \CNL_{\Lambda_v}$ to the set of continuous homomorphisms $r \colon G_{F_v} \to \GL_2(R)$ such that $r \Mod \I{m}_R = \bar{\rho}|_{G_{F_v}}$ and $\det r$ agrees with the composite
$$G_{F_v} \stackrel{\mu|_{G_{F_v}}}{\longrightarrow} \Sc{O}^{\times} \longrightarrow R^{\times}.$$ This functor $\Sc{D}_v^{\Box}$ is represented by an object $R_v^{\Box} \in \CNL_{\Lambda_v}$.
\begin{definition} Let $v \in S$. A local deformation problem for $\bar{\rho}|_{G_{F_v}}$ is a subfunctor $\Sc{D}_v \subset \Sc{D}_v^{\Box}$ satisfying:
\begin{itemize}
\item[(i)] The subfunctor $\Sc{D}_v$ is represented by a quotient $R_v$ of $R_v^{\Box}$
\item[(ii)] For all $R \in \CNL_{\Lambda_v}$, $a \in \ker(\GL_2(R) \to \GL_2(k))$ and $r \in \Sc{D}_v(R)$, we have $ara^{-1} \in \Sc{D}_v(R)$.
\end{itemize}
\end{definition}
\begin{definition} A global deformation problem is a tuple
$$\Sc{S} = (\bar{\rho},\mu,S,\{\Lambda_v\}_{v \in S},\{\Sc{D}_v\}_{v \in S}),$$ where all the notation is as above, and $\Sc{D}_v$ is a local deformation problem for $\bar{\rho}|_{G_{F_v}}$ for $v \in S$.
\end{definition}
\subsection{Galois Cohomology}
Given a finite place $v$ of $F$ and a local deformation problem $\Sc{D}_v$, we denote by $\Sc{L}_v$ the tangent space of the local deformation problem as a subspace $\Sc{L}_v \subset H^1(F_v,\ad^0 \bar{\rho})$. We also let $\Sc{L}_v^{\perp} \subset H^1(F_v,\ad^0 \bar{\rho}(1))$ be the annihilator of $\Sc{L}_v$ induced by the perfect pairing of Galois modules:
$$\ad^0 \bar{\rho} \times \ad^0 \bar{\rho}(1) \to k(1), \quad (X,Y) \mapsto \tr(XY).$$
\begin{definition} Given a global deformation problem $\Sc{S} = (\bar{\rho},\mu,S,\{\Lambda_v\}_{v \in S},\{\Sc{D}_v\}_{v \in S})$, we define the dual Selmer group
$$H^1_{\Sc{S},T}(\ad^0 \bar{\rho}(1)) = \ker \left( H^1(F_S/F,\ad^0 \bar{\rho}(1)) \to \prod_{v \in S-T} H^1(F_v,\ad^0 \bar{\rho}(1))/\Sc{L}_v^{\perp} \right).$$
\end{definition}
Let us now take a look at the specific deformation problem we want to consider.
\subsection{Special Deformations, case $q_v \not \equiv 1 \Mod p$}
Let $v \in S$ be a prime not dividing $p$, and suppose that $q_v \not \equiv 1 \Mod p$. Suppose further that $\bar{\rho}|_{G_{F_v}}$ is unramified, and that $\bar{\rho}(\Frob_v)$ has two distinct eigenvalues $\alpha_v ,\beta_v \in k$ such that $\alpha_v/\beta_v = q_v$. Let $\Lambda_v = \Sc{O}$. We define a subfunctor $\Sc{D}_v^{\St(\alpha_v)} \subset \Sc{D}_v^{\Box}$ directly. Let $R \in \CNL_{\Sc{O}}$ and let $r: G_{F_v} \to \GL_2(R)$ be an element of $\Sc{D}_v^{\Box}$. If $\phi_v \in G_{F_{v}}$ is a choice of geometric Frobenius, then by Hensel's lemma the characteristic polynomial of $r(\phi_v)$ factors as $(X-A_v)(X-B_v)$, where $A_v,B_v \in R^{\times}$ with $A_v$ lifting $\alpha_v$ and $B_v$ lifting $\beta_v$. We will say $r \in \Sc{D}_v^{\St(\alpha_v)}$ if $A_v = q_vB_v$ and $I_{F_v}$ acts trivially on $(r(\phi_v)-A_v)R^2$, which is a direct summand $R$-submodule of $R^2$. One checks that this condition is independent of the choice of $\phi_v$.
\begin{proposition} The functor $\Sc{D}_v^{\St(\alpha_v)}$ is a local deformation problem. The representing object $R_v^{\St(\alpha_v)}$ is formally smooth over $\Sc{O}$ of dimension $4$.
\end{proposition}
\subsection{Existence of Auxiliary Primes}
Continue with the notation from the previous section, and assume further that $\bar{\rho}$ is totally odd, i.e. that $\mu(c) = -1$ for all choices of complex conjugation $c \in G_F$. Write $\zeta_p \in \overline{F}$ for a primitive $p$-th root of unity, and now fix a choice of complex conjugation $c \in G_F$. \\ \\We will assume that $\bar{\rho}|_{G_{F(\zeta_p)}}$ is the direct sum of two distinct characters. By Clifford theory, we know that $\bar{\rho}$ is induced from a continuous character $\bar{\chi}: G_K \to k^{\times}$, where $K$ is the unique quadratic subfield of $F(\zeta_p)/F$. That is,
$\bar{\rho} \cong \Ind_{G_K}^{G_F} \bar{\chi}$. Write $w \in G_F$ for a fixed choice of element with nontrivial image in $\Gal(K/F)$. We can assume that, possibly after conjugation, that $\bar{\rho}$ has the form:
$$\bar{\rho}(\sigma) = \MatTwo{\bar{\chi}(\sigma)}{0}{0}{\bar{\chi}^w (\sigma)}, \quad \text{ for } \sigma \in G_K,$$
$$\bar{\rho}(w) = \MatTwo{0}{\bar{\chi}(w^2)}{1}{0}.$$
Now let $\bar{\gamma} = \bar{\chi}/\bar{\chi}^w$. By assumption, $\bar{\gamma}$ is nontrivial, even after restriction to $G_{F(\zeta_p)}$. We have the following:
\begin{lemma} We have that $\ad^0 \bar{\rho}$ decomposes as $\ad^0 \bar{\rho} \cong k(\delta_{K/F}) \oplus \Ind_{G_K}^{G_F} \bar{\gamma}$ as a $G_F$-module, where $\delta_{K/F}: \Gal(K/F) \to k^{\times}$ is the unique nontrivial character.
\end{lemma}
From now on, we will let $M_0 = k(\delta_{K/F})$ and $M_1 = \Ind_{G_K}^{G_F} \bar{\gamma}$. Fix the standard basis for $\ad^0 \bar{\rho}$:
$$E = \MatTwo{0}{1}{0}{0},\quad H = \MatTwo{1}{0}{0}{-1},\quad F = \MatTwo{0}{0}{1}{0}.$$ If $M \in \ad^0 \bar{\rho}$, we write $k_M \subset \ad^0 \bar{\rho}$ for the line that it spans. \\ \\ We have the following easy lemma:
\begin{lemma} Let $v \nmid p$ be a finite place of $F$ which splits in $K$, and suppose that the local deformation problem $\Sc{D}_v = \Sc{D}_v^{\St(\alpha_v)}$ is defined.
\begin{itemize}
\item[(1)] The subspace $\Sc{L}_v \subset H^1(F_v,\ad^0 \bar{\rho})$ respects the decomposition $\ad^0 \bar{\rho} = M_0 \oplus M_1$. That is,
$$\Sc{L}_v = (\Sc{L}_v \cap H^1(F_v,M_0)) \oplus (\Sc{L}_v \cap H^1(F_v,M_1)).$$
\item[(2)] The subspace $\Sc{L}_v^{\perp} \subset H^1(F_v,\ad^0 \bar{\rho}(1))$ respects the decomposition $\ad^0 \bar{\rho}(1) = M_0(1) \oplus M_1(1)$.
\end{itemize}
\end{lemma}
\begin{proof} The second part is dual to the first, so we only prove the first part. The fact that $\Sc{D}_v^{\St(\alpha_v)}$ is defined means $q_v \not \equiv 1 \Mod p$, that $\bar{\rho}|_{G_{F_v}}$ is unramified, and that $\bar{\rho}(\Frob_v)$ takes two distinct eigenvalues $\alpha_v,\beta_v \in k$ with $\alpha_v/\beta_v = q_v$. The fact that $v$ splits in $K$ means $M_0 = k_H$ and $M_1 = k_E(\epsilon) \oplus k_F(\epsilon^{-1})$ as $k[G_{F_v}]$-modules. The case $q_v \equiv -1 \Mod p$ was proved in \cite{Thorne}. If $q_v \not \equiv \pm 1 \Mod p$, then $\Sc{L}_v$ is $1$-dimensional, and is contained in $H^1(F_v,M_1)$, being spanned by $H^1(F_v,k_E(\epsilon))$.
\end{proof}
\begin{remark} The difference between this and the corresponding Lemma 5.18 from \cite{Thorne} is that we do not make the assumption that the inducing field $K$ is totally real. However, we do need to make sure that we choose primes of $F$ which split in $K$ for the rest of the method to work.
\end{remark}
Let $\Sc{S} = (\bar{\rho},\mu,S,\{\Lambda_v\}_{v \in S},\{\Sc{D}_v\}_{v \in S})$ be a global deformation problem, and let $T \subset S$ be a subset containing $S_p$ (the set of places above $p$). Suppose that for $v \in S-T$, the local deformation problem $\Sc{D}_v = \Sc{D}_v^{\St(\alpha_v)}$. The lemma implies we can decompose
$$H^1_{\Sc{S},T}(\ad^0 \bar{\rho}(1)) = H^1_{\Sc{S},T}(M_0(1)) \oplus H^1_{\Sc{S},T}(M_1(1)).$$
We now show that we can kill the $M_1(1)$ portion of dual Selmer using the special deformation problem we defined in the previous section, and then kill the $M_0(1)$ portion using traditional Taylor-Wiles primes.
\subsection{Killing the $M_1(1)$ portion}
First, we show that lemmas 5.21, 5.22, and 5.23 of \cite{Thorne} continue to hold even if $K$ is not totally real. Indeed, the only one which requires proof is the second, since this is the only place where Thorne used this assumption. However, we will need to impose an additional restriction. We state the other two lemmas here for convenience.
\begin{lemma} Let $\Gamma$ be a group, and $\alpha: \Gamma \to k^{\times}$ a character. Let $k' \subset k$ be the subfield generated by the values of $\alpha$. Then $k'(\alpha)$ is a simple $\F_p[\Gamma]$-module. If $\beta: \Gamma \to k^{\times}$ is another character, then $k'(\alpha)$ is isomorphic to a $\F_p[\Gamma]$-submodule of $k(\beta)$ if and only if there is an automorphism $\tau$ of $k$ such that $\beta = \tau \circ \alpha$.
\end{lemma}
\begin{proof} See the proof of Lemma 5.21 in \cite{Thorne}.
\end{proof}
\begin{lemma} Let $K(\overline{\gamma})$ be the fixed field of $\ker \bar{\gamma}$, let $L = F(\zeta_p) \cap K(\overline{\gamma})$ and assume that $\# \epsilon(G_L )> 1$. Then the $\F_p[G_K]$-module $k(\epsilon \overline{\gamma})$ has no Jordan-Holder factors in common with $k$, $k(\overline{\gamma})$, or $k(\overline{\gamma}^{-1})$. The characters $\epsilon \overline{\gamma}$ and $\overline{\gamma}$ are nontrivial.
\end{lemma}
\begin{proof} The second claim follows from the fact that $\overline{\gamma}|_{G_{F(\zeta_p)}}$ is nontrivial. For the first claim, we show there are no $\F_p[G_K]$-module homomorphisms from $k(\epsilon \overline{\gamma})$ to $k(\overline{\gamma})$ or $k(\overline{\gamma}^{-1})$. Let $f: k(\epsilon \overline{\gamma}) \to k(\overline{\gamma})$ be such a homomorphism, choose $a \in k(\epsilon \overline{\gamma})$, and assume $f(a) = b$ . By the hypothesis of the lemma, there is an element $\tau \in G_L \subset G_K$ such that $\epsilon(\tau) \neq 1$ and $\gamma(\tau) = 1$. Since $f$ is a $\F_p[G_K]$-module homomorphism and $\epsilon(\tau) \in \F_p^{\times}$, we know
$$f(\epsilon(\tau)\overline{\gamma}(\tau)a) = \epsilon(\tau)f(a).$$ On the other hand,
$$f(\epsilon(\tau)\overline{\gamma}(\tau)a) = \overline{\gamma}(\tau)b = b.$$ Thus, $\epsilon(\tau)b = b$, which implies $b = 0$. Since $a \in k(\epsilon \overline{\gamma})$ was arbitrary, this implies $f = 0$. Thus, there are no nontrivial homomorphisms between $k(\epsilon \overline{\gamma})$ and $k(\overline{\gamma})$. \\ \\ The same proof shows there are no nontrivial homomorphisms between $k(\epsilon \overline{\gamma})$ and $k(\overline{\gamma}^{-1})$ or $k$.
\end{proof}
\begin{lemma} Let $N \geq 1$ and let $K_N = F(\zeta_{p^N},\overline{\rho}_N)$, i.e. $K_N$ is the splitting field of $\overline{\rho}_N|_{F(\zeta_{p^N})}$. Then $H^1(K_N/F,M_1(1)) = 0$.
\end{lemma}
\begin{proof} When $K$ is totally real, this is proved in Lemma 5.23 of \cite{Thorne}. The same proof proves the lemma in the case $K$ is CM using the preceding lemma.
\end{proof}
The following proposition is the analog of Proposition 5.20 of \cite{Thorne} and is the only place where we argue differently from Thorne because of not having (in the case that $K$ is not totally real) the luxury to choose places $v$ such that $\rho_N({\rm Frob}_v)$ is the image of complex conjugation under $\rho_N$. The proof relies on the simple observation that given an element $g$ in $\GL_2({\mathcal O}/p^M)$, then for $N>>0$, the element $g^{q^N}$ has (ratio of) eigenvalues that are the Teichm\"uller lift of the (ratio of) the eigenvalues of the reduction of $g$.
\begin{proposition} Let $\Sc{S} = (\bar{\rho},\mu,S,\{\Lambda_v\}_{v \in S},\{\Sc{D}_v\}_{v \in S})$ be a global deformation problem, and let $T = S$. Let $N_0 \geq 1$ be an integer. Let $\rho: G_F \to \GL_2(\Sc{O})$ be a lifting of type $\Sc{S}$. Let $K(\overline{\gamma}\epsilon)$ (resp. $K(\overline{\gamma}\epsilon^{-1})$) be the fixed field of $\ker \bar{\gamma}\epsilon$ (resp. $\ker \bar{\gamma} \epsilon^{-1}$), and assume that $F(\zeta_p) \not \subset K(\overline{\gamma}\epsilon) \cap K(\overline{\gamma}\epsilon^{-1})$. Then for any $m \geq h^1_{\Sc{S},T}(M_1(1))$, there exists a set $Q_0$ of primes, disjoint from $S$, and elements $\alpha_v \in k^{\times}$, satisfying the following:
\begin{itemize}
\item[(1)] $\# Q_0 = m$
\item[(2)] For each $v \in Q_0$, the local deformation problem $\Sc{D}_v^{\St(\alpha_v)}$ is defined. We define the augmented deformation problem
$$\Sc{S}_{Q_0} = (\bar{\rho},\mu,S \cup Q_0,\{\Lambda_v\}_{v \in S} \cup \{\Sc{O}\}_{v \in Q_0}, \{\Sc{D}_v\}_{v \in S} \cup \{\Sc{D}_v^{\St(\alpha_v)}\}_{v \in Q_0}).$$
\item[(3)] Let $\rho_{N_0} = \rho \Mod \lambda^{N_0}: G_F \to \GL_2(\Sc{O}/\lambda^{N_0}\Sc{O})$. Then $\rho_{N_0}(\Frob_v)$ has distinct eigenvalues whose ratio is $q_v$ for each $v \in Q_0$.
\item[(4)] $H^1_{\Sc{S}_{Q_0},T}(M_1(1)) = 0$.
\end{itemize}
\end{proposition}
\begin{proof} We wish to find a set $Q_0$ of primes such that $h^1_{\Sc{S}_{Q_0},T}(M_1(1)) = 0$. Suppose $r = h^1_{\Sc{S},T}(M_1(1)) \geq 0$. Using induction, it suffices to find a single prime $v$ satisfying the conditions of the theorem such that $h^1_{\Sc{S}_{\{v\}},T}(M_1(1)) = \max(r-1,0)$. The case $r=0$ is easy, so assume $r \geq 1$. \\
Let $0 \neq [\varphi] \in H^1_{\Sc{S},T}(M_1(1))$ be a nonzero class. We wish to find a place $v \notin S$ such that:
\begin{itemize}
\item[(i)] $v$ splits in $K$
\item[(ii)] $\rho_{N_0}(\Frob_v)$ has distinct eigenvalues with ratio $q_v \Mod \lambda^{N_0}$.
\item[(iii)] $q_v \not \equiv \pm 1 \Mod \lambda^{N_0}$
\item[(iv)] $\varphi(\Frob_v) \neq 0$ ($\in M_1(1)$).
\end{itemize}
Indeed, the first three conditions imply that $\Sc{D}_v^{\St(\alpha_v)}$ is defined for the appropriate choice of $\alpha_v$. We also have an exact sequence
$$0 \to H^1_{\Sc{S}_{\{v\}},T}(M_1(1)) \to H^1_{\Sc{S},T}(M_1(1)) \to H^1(G_{F_v},M_1(1)) \cong k.$$ Condition (iv) implies the final map is surjective, which gives $h^1_{\Sc{S}_v,T}(M_1(1)) < h^1_{\Sc{S},T}(M_1(1))$, as desired. \\ \\ By the Cebotarev density theorem, it suffices to find an element $\sigma \in G_K$ such that:
\begin{itemize}
\item[(a)] $\rho_{N_0}(\sigma)$ has distinct eigenvalues with ratio $\epsilon(\sigma) \Mod \lambda^{N_0}$.
\item[(b)] $\epsilon(\sigma) \not \equiv \pm 1 \Mod \lambda^{N_0}$
\item[(c)] $\varphi(\sigma) \neq 0$.
\end{itemize}
If $N_0 = 1$, then the assumption in the theorem ensures we can find $\sigma_1$ in $G_K$ such that $\overline{\gamma}(\sigma_1) = \epsilon(\sigma_1) \Mod \lambda$. Indeed, the assumption $F(\zeta_p) \not \subset K(\overline{\gamma}\epsilon) \cap K(\overline{\gamma}\epsilon^{-1})$ ensures that either $G_{K(\overline{\gamma}\epsilon)}$ or $G_{K(\overline{\gamma}\epsilon^{-1})}$ is not contained in $G_{F(\zeta_p)}$. This means there exists $\sigma_1$ in either $G_{K(\overline{\gamma}\epsilon)}$ or $G_{K(\overline{\gamma}\epsilon^{-1})}$ such that $\epsilon(\sigma_1) \not \equiv 1 \Mod p$. In the latter case, we find our desired $\sigma_1$. In the former case, by exchanging the roles of the eigenvalues, we get our desired $\sigma_1$.\\ \\ If $\varphi(\sigma_1) \neq 0$, then take $\sigma = \sigma_1$, so suppose $\varphi(\sigma_1) = 0$. We have the inflation-restriction sequence:
$$0 \to H^1(K_1/F, M_1(1)^{G_{K_1}}) \to H^1(F,M_1(1)) \to H^1(K_1,M_1(1))^{\Gal(K_1/F)}.$$ By the previous lemma, the first group is zero, so the image of $\varphi$ in $H^1(K_1,M_1(1))$ is nonzero. This restriction is a nonzero homomorphism $\varphi|_{G_{K_1}}: G_{K_1} \to M_1(1)$. Thus, we can find $\tau \in G_{K_1}$ such that $\varphi(\tau) \neq 0$. Then take $\sigma = \tau \sigma_1$. Then $$\overline{\rho}(\sigma) = \overline{\rho}(\tau) \overline{\rho}(\sigma_1) = \overline{\rho}(\sigma_1)$$ as $\tau \in \ker(\overline{\rho})$. We also find
$$\epsilon(\sigma) = \epsilon(\tau) \epsilon(\sigma_1) \equiv \epsilon(\sigma_1) \Mod \lambda$$ as $\epsilon(\tau) \equiv 1 \Mod \lambda$. Thus, $\overline{\gamma}(\sigma) \equiv \epsilon(\sigma) \Mod \lambda$. Moreover,
$$\varphi(\sigma) = \varphi(\tau) + \varphi(\sigma_1),$$ meaning $\varphi(\sigma) \neq 0$, as required. \\ \\ If $N_0 > 1$, then consider the element $\sigma_1$ defined above. Then $\rho_{N_0}(\sigma_1)$ has distinct eigenvalues by Hensel's lemma, and we know this ratio modulo $\lambda$ is $\overline{\gamma}(\sigma_1) \equiv \epsilon(\sigma_1) \Mod \lambda$. Consider $\sigma_{N_0} = \sigma_1^{q^M}$ for some $M$ to be determined and $q = \# k$. For some sufficiently high power of $M$, $\epsilon(\sigma_{N_0}) \Mod \lambda^{N_0}$ will be the Teichmuller lifting of $\epsilon(\sigma_1) \Mod \lambda$ to the $\Mod \lambda^{N_0}$ ring (indeed, $M = q^{N_0-1}$ should do). But since $\overline{\gamma}(\sigma_1) \equiv \epsilon(\sigma_1) \Mod \lambda$, we deduce that the ratio of the eigenvalues of $\rho_{N_0}(\sigma_{N_0})$ will be equivalent to $\epsilon(\sigma_{N_0}) \Mod \lambda^{N_0}$. \\ \\ We still need to make sure $\varphi(\sigma) \neq 0$. If $\varphi(\sigma_{N_0}) \neq 0$, then we can take $\sigma = \sigma_{N_0}$. If $\varphi(\sigma_{N_0}) = 0$, consider $\tau \in G_{K_N}$ with $\varphi(\tau) \neq 0$ as before. Let $\sigma = \tau \sigma_{N_0}$. By the same reasoning as in the $N_0=1$ case, the ratio of the eigenvalue of $\rho_N(\sigma)$ will still be equivalent to $\epsilon(\sigma) \Mod \lambda^{N_0}$, and moreover $\varphi(\sigma) = \varphi(\tau) + \varphi(\sigma_{N_0}) \neq 0$ by construction. This concludes the proof.
\end{proof}
\begin{remark} Note that the assumption $F(\zeta_p) \not \subset K(\overline{\gamma}\epsilon) \cap K(\overline{\gamma}\epsilon^{-1})$ is implied by the more checkable condition that $(\# \epsilon(G_L ), \# \overline{\gamma}(G_L)) > 1$, where $L = F(\zeta_p) \cap K(\overline{\gamma})$.
\end{remark}
\subsection{Killing the $M_0(1)$ portion}
Having killed the $M_1(1)$ portion of dual Selmer, we can try and get auxiliary primes that take care of the remaining part of the group. Indeed, we have the following proposition:
\begin{proposition} Let $\Sc{S} = (\bar{\rho},\mu,S,\{\Lambda_v\}_{v \in S}, \{\Sc{D}_v\}_{v \in S})$ be a global deformation problem. Let $T \subset S$, and suppose for $v \in S-T$ we have $\Sc{D}_v = \Sc{D}_v^{\St(\alpha_v)}$. Suppose further that $h^1_{\Sc{S},T}(M_1(1)) = 0$, and let $N_1 \geq 1$ be an integer. Then there exists a finite set $Q_1$ of finite places of $F$, disjoint from $S$, satisfying:
\begin{itemize}
\item[(1)] We have $\# Q_1 = h^1_{\Sc{S},T}(M_0(1))$, and for each $v \in Q_1$, the norm $q_v \equiv 1 \Mod p^{N_1}$ and $\bar{\rho}(\Frob_v)$ has distinct eigenvalues.
\item[(2)] Define the augmented deformation problem
$$\Sc{S}_{Q_1} = (\bar{\rho},\mu,S \cup Q_1,\{\Lambda_v\}_{v \in S} \cup \{\Sc{O}\}_{v \in Q_1},\{\Sc{D}_v\}_{v \in S} \cup \{\Sc{D}_v^{\Box}\}_{v \in Q_1}).$$ Then $h^1_{\Sc{S}_{Q_1},T}(\ad^0 \bar{\rho}(1)) = 0$.
\end{itemize}
\end{proposition}
\begin{proof} See the proof of Proposition 5.24 in \cite{Thorne}.
\end{proof}
\section{The Main Theorem}
From this point, everything from sections 6 of \cite{Thorne} carries over. We can now work towards proving the main theorem. We proceed as in the aforementioned paper, making the necessary modifications. First, some preliminary results.
\begin{lemma}\label{BaseChange} Let $F$ be a totally real number field, and let $F'/F$ be a totally real, soluble extension. Let $p$ be a prime and let $\iota: \overline{\Q}_p \stackrel{\sim}{\to} \mathbb{C}$ be a fixed isomorphism.
\begin{itemize}
\item[(1)] Let $\pi$ be a cuspidal automorphic representaion of $\GL_2(\mathbb{A}_F)$ of weight $2$, and suppose that $r_{\iota}(\pi)|_{G_{F'}}$ is irreducible. Then there exists a cuspidal automorphic representation $\pi_{F'}$ of $\GL_2(\mathbb{A}_{F'})$ of weight $2$, called the \emph{base change} of $\pi$, such that $r_{\iota}(\pi_{F'}) \cong r_{\iota}(\pi)|_{G_{F'}}$.
\item[(2)] Let $\rho: G_F \to \GL_2(\overline{\Q}_p)$ be a continuous representation such that $\rho|_{G_{F'}}$ is irreducible. Let $\pi'$ be a cuspidal automorphic representation of $\GL_2(\mathbb{A}_{F'})$ of weight $2$ with $\rho|_{G_{F'}} \cong r_{\iota}(\pi')$. Then there exists a cuspidal automorphic representation $\pi$ of $\GL_2(\mathbb{A}_F)$ of weight $2$ such that $\rho \cong r_{\iota}(\pi)$.
\end{itemize}
\end{lemma}
\begin{proof} This is stated in \cite{Thorne}, Lemma 5.1. The proof follows from results of \cite{Lan}, using arguments of \cite{BLGHT}, Lemma 1.3.
\end{proof}
\begin{theorem}\label{Long} Let $F$ be a totally real field, and let $p$ be an odd prime. Let $\rho: G_F \to \GL_2(\overline{\Q}_p)$ be a continuous representation. Suppose that:
\begin{itemize}
\item[(1)] $[F:\Q]$ is even.
\item[(2)] Letting $K$ be the quadratic subfield of $F(\zeta_p)/F$, there exists a continuous character $\bar{\chi}: G_K \to \overline{\F}_p^{\times}$ such that $\bar{\rho} \cong \Ind_{G_K}^{G_F} \overline{\chi}$.
\item[(3)] Letting $w \in \Gal(K/F)$ be the nontrivial element, the character $\bar{\gamma} = \bar{\chi}/\bar{\chi}^w$ remains nontrivial even after restriction to $G_{F(\zeta_p)}$ (in particular, $\bar{\rho}$ is irreducible).
\item[(4)] We have $F(\zeta_p) \not \subset K(\bar{\gamma}\epsilon^{-1}) \cap K(\bar{\gamma}\epsilon)$.
\item[(5)] The character $\psi = \epsilon \det \rho$ is everywhere unramified.
\item[(6)] The representation $\rho$ is almost everywhere unramified.
\item[(7)] For each place $v|p$, $\rho|_{G_{F_v}}$ is semi-stable, and $\bar{\rho}|_{G_{F_v}}$ is trivial. For each embedding $\tau: F \hookrightarrow \overline{\Q}_p$, we have $\text{HT}_{\tau}(\rho) = \{0,1\}$.
\item[(8)] If $v \nmid p$ is a finite place of $F$ at which $\rho$ is ramified, then $q_v \equiv 1\Mod p$, $\text{WD}(\rho|_{G_{F_v}})^{F-ss} \cong \text{rec}_{F_v}^T(\text{St}_2(\chi_v))$ for some unramified character $\chi_v: F_v^{\times} \to \overline{\Q}_p^{\times}$, and $\bar{\rho}|_{G_{F_v}}$ is trivial. The number of such places is even.
\item[(9)] There exists a cuspidal automorphic representation $\pi$ of $\GL_2(\mathbb{A}_F)$ of weight $2$ and an isomorphism $\iota: \overline{\Q}_p \to \mathbb{C}$ satisfying:
\begin{itemize}
\item[(a)] There is an isomorphism $\overline{r_{\iota}(\pi)} \cong \overline{\rho}$.
\item[(b)] If $v|p$ and $\rho$ is ordinary, then $\pi_v$ is $\iota$-ordinary and $\pi_v^{U_0(v)} \neq 0$. If $v|p$ and $\rho$ is non-ordinary, then $\pi_v$ is not $\iota$-ordinary and $\pi_v$ is unramified.
\item[(c)] If $v \nmid p\infty$ and $\rho|_{G_{F_v}}$ is unramified, then $\pi_v$ is unramified. If $v \nmid p\infty$ and $\rho|_{G_{F_v}}$ is ramified, then $\pi_v$ is an unramified twist of the Steinberg representation.
\end{itemize}
\end{itemize}
Then $\rho$ is automorphic: there exists a cuspidal automorphic representation $\pi'$ of $\GL_2(\mathbb{A}_F)$ of weight $2$ and an isomorphism $\rho \cong r_{\iota}(\pi')$.
\end{theorem}
\begin{remark} Here, $U_0(v)$ is the set of matrices in $\GL_2(\Sc{O}_{F_v})$ whose reduction modulo a fixed uniformizer of $\Sc{O}_{F_v}$ is upper triangular.
\end{remark}
\begin{proof} This is Theorem 7.2 \cite{Thorne} with the necessary modifications (namely, the addition of condition (4) instead of the condition that $K$ be totally real). One can just repeat the proof the author gives in that paper, replacing Proposition 5.20 of loc. cit. with Proposition 2.11 from the previous section.
\end{proof}
Using this theorem, we arrive at the main theorem.
\begin{theorem}\label{Aut} Let $F$ be a totally real number field, let $p$ be an odd prime, and let $\rho: G_F \to \GL_2(\overline{\Q}_p)$ be a continuous representation satisfying the following:
\begin{itemize}
\item[(1)] The representation $\rho$ is almost everywhere unramified.
\item[(2)] For each $v|p$ of $F$, the local representation $\rho|_{G_{F_v}}$ is de Rham. For each embedding $\tau: F \hookrightarrow \overline{\Q}_p$, we have $\text{HT}_{\tau}(\rho) = \{0,1\}$.
\item[(3)] For each complex conjugation $c \in G_F$, we have $\det \rho(c) = -1$.
\item[(4)] The residual representation $\bar{\rho}$ is absolutely irreducible, but $\bar{\rho}|_{G_{F(\zeta_p)}}$ is a direct sum of two distinct characters. Further suppose that if $K$ is the unique quadratic subfield of $F(\zeta_p)/F$ and $\bar{\gamma}: G_K \to k^{\times}$ is the ratio of the two characters, then we have $F(\zeta_p) \not \subset K(\bar{\gamma}\epsilon^{-1}) \cap K(\bar{\gamma}\epsilon)$.
\end{itemize}
Then $\rho$ is automorphic: there exists a cuspidal automorphic representation $\pi$ of $\GL_2(\mathbb{A}_F)$ of weight $2$, an isomorphism $\iota: \overline{\Q}_p \to \mathbb{C}$, and an isomorphism $\rho \cong r_{\iota}(\pi)$.
\end{theorem}
\begin{proof} The proof is exactly the same as Theorem 7.4 of \cite{Thorne}, replacing Theorem 7.2 of loc. cit. with Theorem \ref{Long} above. The idea is to construct a soluble extension $F'/F$ such that $\bar{\rho}|_{G_{F'}}$ satisfies the conditions of Theorem \ref{Long} above. We then apply Lemma \ref{BaseChange} to deduce the automorphy of $\rho$. We should note that in \cite{Thorne}, the author makes use of Corollary 7.4 in loc. cit., but that goes unchanged for us because that corollary made no assumptions on the quadratic subfield $K$.
\end{proof}
\section{Application to Elliptic Curves}
We can apply this theorem to elliptic curves. In a paper of Freitas, Le Hung, and Siksek (see \cite{FLHS}) the authors prove there are only finitely many non-automorphic elliptic curves over any given totally real field. We can use the theorem above to prove automorphy of elliptic curves $E$ defined over totally real fields $F$ in some cases not covered in the existing literature. This comes about by applying our theorem to the 7-adic representations arising from $E$. Note that the assumptions of \cite{Thorne} imply that the theorem in that paper can be applied only to $p$-adic representations of elliptic curves with $p$ congruent to 1 modulo 4. \\ \\ The following theorem is \cite{FLHS}, Theorems 3 and 4.
\begin{theorem}\label{AbsIrr} Let $p \in \{3,5,7\}$. Let $E$ be an elliptic field over a totally real field $F$ and let $\bar{\rho}_{E,p}: G_F \to \GL_2(\F_p)$ be the representation given by the action on the $p$-torsion of $E$. If $\bar{\rho}_{E,p}(G_{F(\zeta_p)})$ is absolutely irreducible, then $E$ is modular.
\end{theorem}
We will call an elliptic curve $p$-bad if $E[p]$ is an absolutely reducible $\F_p[G_{F(\zeta_p)}]$-module. Otherwise $E$ is $p$-good. The theorem tells us the only elliptic curves $E$ which are potentially non-modular are those which are $p$-bad for $p = 3,5$, and $7$. In \cite{Thorne}, the author deals with some of these remaining cases:
\begin{theorem} Let $E$ be an elliptic curve over a totally real field $F$. Suppose:
\begin{itemize}
\item[(1)] $5$ is not a square in $F$.
\item[(2)] $E$ has no $F$-rational $5$-isogeny.
\end{itemize}
Then $E$ is modular.
\end{theorem}
We can prove a similar theorem, but before doing so we recall \cite{FLHS}, Proposition 9.1.
\begin{proposition} Let $F$ be a totally real number field and let $E$ be an elliptic curve over $F$. Suppose $F \cap \Q(\zeta_7) = \Q$ and write $\bar{\rho} = \bar{\rho}_{E,7}$. Suppose $\bar{\rho}$ is irreducible but $\bar{\rho}(G_{F(\zeta_7)})$ is absolutely reducible. Then $\bar{\rho}(G_F)$ is conjugate in $\GL_2(\F_7)$ to one of the groups
$$H_1 = \left\langle \MatTwo{3}{0}{0}{5}, \MatTwo{0}{2}{2}{0} \right\rangle, \quad H_2 = \left\langle \MatTwo{0}{5}{3}{0},\MatTwo{5}{0}{3}{2} \right\rangle.$$ The group $H_1$ has order $36$ and is contained as a subgroup of index $2$ in the normalizer of a split Cartan subgroup. The group $H_2$ has order $48$ and is contained as a subgroup of index $2$ in the normalizer of a non-split Cartan subgroup. The images of $H_1$ and $H_2$ in $\PGL_2(\F_7)$ are isomorphic to $D_3 \cong S_3$ and $D_4$, respectively.
\end{proposition}
\begin{theorem}\label{Main} Let $F$ be a totally real field, and let $E$ be an elliptic curve over $F$. Suppose:
\begin{itemize}
\item[(1)] $F \cap \Q(\zeta_7) = \Q$.
\item[(2)] $E$ has no $F$-rational $7$-isogeny.
\item[(3)] Either $\bar{\rho}_{E,7}(G_{F(\zeta_7)})$ is absolutely irreducible, or it is reducible and $\bar{\rho}_{E,7}(G_F)$ is conjugate to the group $H_1$ from the previous proposition.
\end{itemize}
Then $E$ is modular.
\end{theorem}
\begin{proof} Let $\rho: G_F \to \GL_2(\Q_7)$ be the representation given by the action of $G_F$ on the \'{e}tale cohomology $H^1(E_{\overline{F}},\mathbb{Z}_7)$, after a choice of basis. The goal is to show $\rho$ is automorphic. Hypothesis (2) is equivalent to $\bar{\rho}$ being irreducible, hence absolutely irreducible because of complex conjugation. If $\bar{\rho}|_{G_{F(\zeta_7)}}$ is irreducible, then $\rho$ is automorphic by Theorem \ref{AbsIrr} above. We now deal with the case when this restriction is not irreducible. \\ \\ If $\bar{\rho}|_{G_{F(\zeta_7)}}$ is absolutely reducible, the third hypothesis combined with the previous proposition gives that the projective image of $\bar{\rho}$ in $\PGL_2(\F_7)$ is isomorphic to $D_3$. This implies that $\bar{\rho}|_{G_{F(\zeta_7)}}$ cannot be scalar since $\Gal(F(\zeta_7)/F)$ is cyclic, and therefore cannot surject onto $D_3$. \\ \\ Let $K$ be the quadratic subfield of $F(\zeta_7)/F$, so that $[F(\zeta_7):K] = 3$ by hypothesis (1). Let $\bar{\gamma}: G_K \to \F_7^{\times}$ be the character which gives the ratio of the eigenvalues. We need to examine the subgroup
$$H_1 = \left \langle \MatTwo{3}{0}{0}{5}, \MatTwo{0}{2}{2}{0} \right \rangle \subset \GL_2(\F_7).$$ It is easy to check that these two matrices generate the projective image as well. By simply checking the ratio of eigenvalues of each of the matrices, one can check that the possible values for the image of $\bar{\gamma}$ are elements of $\{1,2,4\}$. Therefore, $[K(\bar{\gamma}):K]=3$ or $1$. However, it cannot be the latter as $\overline{\gamma}$ is nontrivial as a character on $G_K$ by assumption. Therefore $[K(\bar{\gamma}):K] = 3$, so $K(\bar{\gamma}) \cap F(\zeta_7) = K$ or $F(\zeta_7)$. But we know it cannot be $F(\zeta_7)$ since the image of $\bar{\rho}|_{G_{F(\zeta_7)}}$ is non-scalar. Thus, $K(\bar{\gamma})$ is disjoint over $K$ from $F(\zeta_7)$ and $[K(\bar{\gamma}):K] = [F(\zeta_7):K] = 3$. Thus, hypothesis (4) of the main theorem above is satisfied, and the theorem implies $E$ is modular.
\end{proof}
We end by remarking that we can extend Theorem \ref{Main} to primes other than $p=7$.
\begin{theorem} Let $F$ be a totally real field, and let $E$ be an elliptic curve over $F$. Let $p \geq 7$ be a prime such that $(p-1)/2 = q^n$ for some odd prime $q$ and $n \geq 1$. Suppose:
\begin{itemize}
\item[(1)] $F \cap \Q(\zeta_p) = \Q$.
\item[(2)] $E$ has no $F$-rational $p$-isogeny.
\item[(3)] $\bar{\rho}_{E,p}(G_{F})$ normalizes a split Cartan subgroup of $\GL_2(\F_p)$.
\end{itemize}
Then $E$ is modular.
\end{theorem}
\begin{proof} Let $\rho: G_F \to \GL_2(\Q_p)$ be the representation given by the action of $G_F$ on the \'{e}tale cohomology $H^1(E_{\overline{F}},\mathbb{Z}_p)$, after a choice of basis. The goal is to show $\rho$ is automorphic. Hypothesis (2) is equivalent to $\bar{\rho}$ being irreducible, hence absolutely irreducible because of complex conjugation. Hypothesis (3) says $\bar{\rho}_{E,p}(G_F)$ is contained in the normalizer of a split Cartan subgroup. Thus, $\bar{\rho}(G_{F(\zeta_p)})$ cannot be scalar, since $\Gal(F(\zeta_p)/F)$ is cyclic, and hence cannot surject onto a non-cylic group. \\ \\ Let $K$ be the quadratic subfield of $F(\zeta_p)/F$, so that $[F(\zeta_p):K] = q^n = (p-1)/2$ by hypothesis (1). Let $\bar{\gamma}: G_K \to \F_p^{\times}$ be the character which gives the ratio of eigenvalues of $\bar{\rho}|_{G_K}$. We want to examine $[K(\bar{\gamma}):K]$, where $K(\bar{\gamma}) = \bar{F}^{\ker(\bar{\gamma})}$ as always. In particular, we will show that $K(\bar{\gamma}) \cap F(\zeta_p)$ is a field $L$ which satisfies $(\# \epsilon(G_L),\#\bar{\gamma}(G_L)) > 1$, which implies hypothesis (4) of the main theorem. Note that hypothesis (1) implies that, as a character of $G_K$, that $\epsilon$ takes values in $(\F_p^{\times})^2$. \\ \\ Using the fact that $\det \bar{\rho}$ is the mod $p$ cyclotomic character, we find that $\bar{\chi} \bar{\chi}^w = \epsilon$, so that $\bar{\gamma} = \bar{\chi}/\bar{\chi}^w = \bar{\chi}^2 \epsilon^{-1}$, which is a character $G_K \to (\F_p^{\times})^2$. Thus, the order of $\bar{\gamma}$ divides $(p-1)/2 = q^n$, and moreover cannot equal $1$ as $\bar{\gamma}$ is a nontrivial character of $G_K$. Thus, $1<[K(\bar{\gamma}):K] \big| q^n$. Moreover, $[F(\zeta_p):K] = q^n$ by hypothesis (1) of the theorem. Lastly, we know $K(\bar{\gamma}) \not \subseteq F(\zeta_p)$ since $\bar{\gamma}$ is nontrivial upon restriction to $G_{F(\zeta_p)}$, and thus $K(\bar{\gamma}) \cap F(\zeta_p)$ is neither $K(\bar{\gamma})$ nor $F(\zeta_p)$. This intersection is therefore a field $L$ which satisfies $(\#\epsilon(G_L),\# \bar{\gamma}(G_L))>1$ as $q$ divides both quantities. Thus, hypothesis (4) of the main theorem is satisfied, and therefore $E$ is modular.
\end{proof}
\bibliographystyle{alpha}{}
| b2fc369a7ad3df1eb1dbeb42774e15e6cabb9a69 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}\label{sec1}
An important tool of modern harmonic analysis is the extrapolation theorem due to Rubio de Francia.
For a given operator $T$ we suppose that for some $p_0$, $1 \le p_0 < \infty$, and
for every weight belonging to the Muckenhoupt class $A_{p_0}$ the inequality
\begin{equation} \label{i1}
\norm{Tf}{\ensuremath{L_{p_0,w} (\ensuremath{{\mathbb R}^n})}}= \left(\int_{\ensuremath{{\mathbb R}^n}} |Tf(x)|^{p_0} w(x) dx \right)^\frac{1}{p_0} \le c \norm{f}{\ensuremath{L_{p_0,w} (\ensuremath{{\mathbb R}^n})}}
\end{equation}
holds, where the constant $c$ is independent of $f$ and depends on $w$ only in terms of its $A_{p_0}$-constant, $[w]_{A_{p_0}}$ (see Definition \ref{defap}).
Then for every $p$, $1 < p < \infty$, and every $w \in A_{p}$, there exists a constant depending
on $[w]_{A_{p}}$ such that
\begin{equation} \label{i2}
\norm{Tf}{\ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}}\le c \norm{f}{\ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}}
\end{equation}
(cf.\ \cite{Ru1, Ru2} for the original results, or \cite{CMP11, D11} for modern presentations).
One fact which makes extrapolation theory so powerful is that one does not need any assumption on $T$ besides having $T$ well-defined on $\bigcup_{w\in A_{p_0}} L_{{p_0},w}(\ensuremath{{\mathbb R}^n})$. (Actually one can extrapolate just inequalities between pairs of functions, not even an operator is needed.)
Even if one is not interested in any weighted results this is very interesting by the fact that for example the weighted $L_2$-boudednesses implies in particular unweighted boundednesses in the complete $L_p$ scale.
Appropriate assertions hold also for subclasses of $A_{p_0}$, that is, one assumes \eqref{i1} for a subclass (e.g. $A_1$) and deduces \eqref{i2} for appropriate subclasses (this applies to different Fourier multipliers, for instance).
We generalize results of this theory to Morrey spaces and will get even in the classical Morrey spaces new results on the boundednesses of operators in the sense that we do not need any condition on $T$ except that $T$ satisfies \eqref{i1} (for weights of $A_{p_0}$ or distinguished subclasses) and is well-defined on $\bigcup_{w\in A_{p_0}} L_{{p_0},w}(\ensuremath{{\mathbb R}^n})$.
There are lots of papers dealing with classes of operators in various Morrey type spaces which admit size estimates as
\begin{equation} \label{i5}
|(Tf)(y)|\leq c \int_{\mathbb{R}^n} \frac{|f(x)|}{|y-x|^n} dx
\end{equation}
for all $f\in \mathcal{D}(\ensuremath{{\mathbb R}^n})$ and $y\notin \ensuremath{\mathrm{supp \,}}(f)$, where $\mathcal{D}(\ensuremath{{\mathbb R}^n})$ are the compactly supported smooth functions (cf. \cite{Pee66,ChFr87,N94, Alv96, Y98, Sam09,KS09, G11, Gul12, Mu12, SFZ13, RS14, KGS14, PT15,IST15, Nak16,Wan16} and the references given there).
To obtain the boundedness results they are explicitly using and requiring different representation formulas as \eqref{i5} of the considered operators.
We are completely avoiding these assumptions and encapsulating them through the consideration of weighted spaces. Hence, we get even new results on the classical Morrey spaces regarding Mihlin-H\"ormander type operators, rough operators, pseudodifferential operators, square functions, commutators, Fourier integral operators\ldots
\par
Moreover, literature about (the nonseparable) Morrey spaces (starting from Peetre \cite{Pee66} and many following scholars) does not care about how to extend the considered operators (satisfying a required representation formula only on nondense subsets of the considered Morrey spaces).
Some forerunners dealing with the extension of the operators are \cite{Alv96, AX12, RT13, Ros13, RoT14_2, T-HS, RS14, Ad15, RS16}.
Mainly, they apply duality (with some exceptions as \cite{T-HS, RS14} justifying \eqref{i5} also for Morrey functions), which restricts the boundedness results to linear operators and moreover the range of the integration parameter $p$ (in particular, the extension problem was not covered for $p=1$ and the corresponding weak-type inequalities).
We show different (continuous) embeddings of (weighted) Morrey spaces into different Muckenhoupt $A_1$ weighted $L_p$ spaces, which enables us to define the operators on the considered Morrey spaces by restriction.
Another possibility to overcome the extension problem is using the Fatou property of the Morrey space. Nevertheless, this method requires at least a bigger space where the operator is also bounded to extend it by restriction (cf.~Triebel \cite{TrFat}), which is exactly the type of our embeddings.
However, by the fact that we do not require representation formulas of operators it is even sufficient to show a not necessarily continuous embedding of a Morrey type space into a set of the type $\bigcup_{w\in A_{p}} L_{{p},w}(\ensuremath{{\mathbb R}^n})$ (where the considered operator is required to be bounded). This can be seen as a powerful model case to extend operators which are a priori only defined on $\mathcal{D}(\ensuremath{{\mathbb R}^n})$ ~---as multipliers and (maximally truncated) singular integrals--- to the various generalizations of Morrey spaces given in the literature and to overcome the extension problem (by the fact that we avoid the requirement of justifying \eqref{i5} also for Morrey functions).
If one wants to deal with other spaces where $\mathcal{D}(\ensuremath{{\mathbb R}^n})$ is not dense and which are contained in the set $\bigcup_{w\in A_{p}} L_{{p},w}(\ensuremath{{\mathbb R}^n})$ and where the maximal operator is bounded (or in their appropriate associated spaces), then there are even very good chances that the presented results could be carried over partially.
A extrapolation result for Morrey spaces appears also in \cite{RS16}. Its proof uses the boundedness of the Hardy-Littlewood maximal operator on predual Morrey spaces, which are also used to define the operator by duality. We avoid this in our approach and obtain more general results, valid also for end-points and for limited-range type assumptions.
We also extrapolate from inequalities with $L_{p,w}$ (quasi)-norms for $0<p<\infty$ and $w\in A_\infty$ to Morrey (quasi)-norms. In this way we get new Morrey versions of several interesting inequalities of this type appearing in the literature for pairs of operators.
We consider only Muckenhoupt weights because the extrapolation techniques are adapted to them and because there are many operators for which the boundedness on Lebesgue spaces with (subclasses of) Muckenhoupt weights are known, so that we can immediately infer Morrey estimates from our general results. Nevertheless, a description of the admissible weighted Morrey estimates is not known even for the Hardy-Littlewood maximal operator and, in fact, in the case of the Morrey spaces $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ (see Definition \ref{d1:def}) the class of admissible weights goes beyond the Muckenhoupt $A_p$ class as shown in \cite{Ta15}. A similar situation holds for the Hilbert transform (\cite{Sam09}) and more generally for the Riesz transforms and other singular integrals (\cite{NkSa17}). On the other hand, in these cases the problem of defining the operators in the corresponding weighted spaces, which in our case is solved by the embeddings, should be considered.
After introducing the notation and some preliminaries in Section \ref{sec2}, Section \ref{sec3} is concerned with continuous embeddings of Morrey type spaces into Muckenhoupt weighted $A_1$ Lebesgue spaces.
In Section \ref{sec4} we present the extrapolation theory generalized to Morrey type spaces which will be applied to far-reaching boundedness results given in Section \ref{sec5}.
\section{Notation, Morrey spaces and preliminaries}\label{sec2}
A weight is a locally integrable nonnegative function. For $0< p <\infty$, $L_{p,w} (\ensuremath{{\mathbb R}^n})$ is the complex quasi-Banach space of functions whose $p$-th power is integrable with respect to the weight $w$, with the quasinorm given by
\[
\| f \, | L_{p,w} (\ensuremath{{\mathbb R}^n}) \| = \Big( \int_{\ensuremath{{\mathbb R}^n}} |f(x)|^p \, w(x) dx \Big)^{1/p}= \Big( \int_{\ensuremath{{\mathbb R}^n}} |f|^p \, w \Big)^{1/p}.
\]
Moreover, we write $w(M)=\int_{M} w(x) dx$ for the measure of the subset $M$ of $\ensuremath{{\mathbb R}^n}$. We similarly define $L_{p,w} (M)$.
If $w(x)\equiv 1$, we simply write $L_{p}(M)$, $\| \cdot \, | L_{p} (M) \|$ and $|M|$.
Furthermore,
$\chi_M$ denotes the characteristic function of $M$.
For any $p\in (1,\infty)$ we denote by $p'$ the conjugate index, namely, $1/p+1/{p'}= 1$. If $p=1$, then $p'=\infty$.
Moreover, $\mathcal{L}(\ensuremath{{\mathbb R}^n})$ collects all equivalence classes of almost everywhere coinciding measurable complex functions. A subset of $\mathcal{L}(\ensuremath{{\mathbb R}^n})$ is $L_1^{\mathrm{loc}}(\ensuremath{{\mathbb R}^n})$, which collects all locally integrable functions, that is, functions in $L_1 (M)$ for any bounded measurable set $M$ of $\ensuremath{{\mathbb R}^n}$.
We consider cubes whose sides are parallel to the coordinate axes. For such a cube $Q$, $\delta Q$ stands for the concentric cube with side-length $\delta$ times of the side-length of $Q$.
The concrete value of the constants may vary from one
formula to the next, but remains the same within one chain of (in)equalities.
\begin{Def} \label{defap}
Let
$w\in L_1^\textrm{loc}(\ensuremath{{\mathbb R}^n})$ with $w>0$ almost everywhere. We say that $w$ is a \textit{Muckenhoupt weight} belonging to $A_p$ for $1<p<\infty$ if
\begin{equation*}
[w]_{A_p}\equiv
\sup_Q \frac{w(Q)}{|Q|} \left( \frac{w^{1-p'}(Q)}{|Q|} \right)^{p-1}<\infty,
\end{equation*}
where the supremum is taken over all cubes $Q$ in $\mathbb{R}^n$. The quantity $[w]_{A_p}$ is the $A_p$ constant of $w$.
We say that $w$ belongs to $A_1$ if, for any cube $Q$,
\begin{equation*}
\frac{w(Q)}{|Q|}\le c w(x) \text{ for almost all } x\in Q.
\end{equation*}
The $A_1$ constant of $w$, denoted by $[w]_{A_1}$, is the smallest constant $c$ for which the inequality holds.
We say that $w$ is in $A_\infty$ if $w\in A_p$ for some $p$.
\end{Def}
\begin{Rem}
The $A_p$ weights are doubling. This means that if $w\in A_p$, there exists a constant $c$ such that
\begin{equation} \label{DC}
w(2Q)\le cw(Q),
\end{equation}
for every cube $Q$ in $\ensuremath{{\mathbb R}^n}$ (cf.~\cite[(7.3)]{D01}).
\end{Rem}
We recall that the Hardy-Littlewood maximal operator, which we denote by $M$, is bounded on $L_{p,w} (\ensuremath{{\mathbb R}^n})$ for $1< p <\infty$ if and only if $w\in A_p$ and is of weak-type $(1,1)$ with respect to the measure $w(x)dx$ if and only if $w\in A_1$. On the other hand, we will use the following construction of $A_1$ weights: if $Mf(x)$ is finite a.e.~and $0\le\delta<1$, then $Mf(x)^\delta$ is an $A_1$ weight whose $A_1$-constant depends only on $\delta$. Moreover, the factorization theorem says that $w\in A_p$ if and only if there exist $w_0, w_1\in A_1$ such that $w=w_0w_1^{1-p}$ (cf. \cite{Ru2} or \cite[Thm.~9.5.1]{G09}).
We define some Muckenhoupt weighted Morrey spaces.
\begin{Def}{} \label{d1:def}
\upshape
For $0< p< \infty$, $-\frac{n}{p}\leq r<0$ and a weight $w$ we define
\begin{equation*}
\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}\equiv\{f \in {\mathcal{L}(\ensuremath{{\mathbb R}^n})} \mbox{ : } \left\|f|\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}\right\|<\infty\}
\end{equation*}
with the quasinorm
\begin{align*}
\left\|f|\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}\right\|\equiv &
\sup_{Q} w(Q)^{-\left(\frac{1}{p}+\frac{r}{n} \right)} \left\|f|L_{p, w}(Q)\right\|,
\end{align*}
where the supremum is taken over all cubes $Q$ in $\mathbb{R}^n$.
Moreover,
\begin{equation*}
\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}\equiv\{f \in {\mathcal{L}(\ensuremath{{\mathbb R}^n})} \mbox{ : } \left\|f|\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}\right\|<\infty\}
\end{equation*}
with the quasinorm
\begin{align*}
\left\|f|\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}\right\|\equiv &
\sup_{Q} |Q|^{-\left(\frac{1}{p}+\frac{r}{n} \right)} \left\|f|L_{p, w}(Q)\right\|.
\end{align*}
We also define their weak versions given by the quasinorms
\begin{align*}
&\left\|f|W\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}\right\|\equiv
\sup_{Q} \sup_{t>0} w(Q)^{-\left(\frac{1}{p}+\frac{r}{n} \right)} t w(\{x\in Q:\ |f(x)|>t\})^{\frac{1}{p}} ,
\\ &\left\|f|W\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}\right\|\equiv
\sup_{Q} \sup_{t>0} |Q|^{-\left(\frac{1}{p}+\frac{r}{n} \right)} t w(\{x\in Q:\ |f(x)|>t\})^{\frac{1}{p}}.
\end{align*}
\end{Def}
\begin{Rem}
We observe that ${L}^{-{n}/{p}}_p(w,\ensuremath{{\mathbb R}^n})={L}^{-{n}/{p}}_p(\lambda,w,\ensuremath{{\mathbb R}^n})=L_{p, w}(\ensuremath{{\mathbb R}^n})$. If $w\equiv 1$, $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ and $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ coincide with the unweighted Morrey space $\ensuremath{{ L}^{r}_{p}(\ensuremath{{\mathbb R}^n})}$. The weak Morrey spaces coincide also with the weak $\ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}$ spaces for $r=-n/p$.
\end{Rem}
\begin{Def}\label{defRH}
\upshape
Let a nonnegative locally integrable function $w$ on $\mathbb{R}^n$ belong to the \textit{reverse H\"older class} $RH_\sigma$ for $1<\sigma<\infty$ if it satisfies the \textit{reverse H\"older inequality} with exponent $\sigma$, i.e. \begin{equation*}
\left(\frac{1}{|Q|}\int_{Q} w(x)^\sigma dx \right)^\frac{1}{\sigma}\le \frac{c}{|Q|} \int_{Q} w(x) dx,
\end{equation*}
where the constant $c$ is universal for all cubes $Q\subset \ensuremath{{\mathbb R}^n}$.
\end{Def}
The $RH_\sigma$ classes are decreasing, that is, $RH_\sigma \subset RH_\tau$ for $1<\tau<\sigma<\infty$. On the other hand, Gehring's lemma (\cite{gehring}) says that if $w\in RH_\sigma$, there exists $\epsilon>0$ such that $w\in RH_{\sigma+\epsilon}$ (openness of the reverse H\"older classes).
\begin{Rem} Let $w\in RH_\sigma$. For any cube $Q$ and any measurable $E\subset Q$ it holds that
\begin{equation}\label{ineqRH}
\frac{w(E)}{w(Q)}\le c \left(\frac{|E|}{|Q|}\right)^{1/\sigma'}.
\end{equation}
(See \cite[Cor.~7.6]{D01}). Since $w\in A_p$ implies that $w\in RH_\sigma$ for some $\sigma$ (cf. \cite[Thm.~7.4]{D01}), the inequality holds for each $A_p$ weight for the appropriate $\sigma$. Consequences of the reverse H\"older inequalities for weights are that if $w\in A_p$, then there exist $\epsilon>0$ and $s>1$ such that $w\in A_{p-\epsilon}$ (openness of $A_p$ classes) and $w^s\in A_p$.
\end{Rem}
In the paper we use sometimes classes of the form $A_p\cap RH_\sigma$. There is a characterization of R.~Johnson and C.~Neugebauer for them (\cite{JN91}):
\begin{equation}\label{apandrh}
A_p\cap RH_\sigma =\{w: w^\sigma\in A_{\sigma(p-1)+1}\}.
\end{equation}
This is also valid for $p=1$, that is, $A_1\cap RH_\sigma =\{w: w^\sigma\in A_{1}\}$, which can be easily proved from the definition of $A_1$.
\begin{Rem} \label{Ma3}
Let $w(x)=|x|^\alpha$ or $w(x)=(1+|x|)^\alpha$. Then we have the following:
\begin{enumerate}
\item It holds that $w \in A_p$ if and only if $-n<\alpha<n(p-1)$ whenever $p\in (1,\infty)$, and $-n<\alpha\le 0$ whenever $p=1$.
\item For $\alpha \in (-n,0)$ it holds $w \in RH_\sigma$ if and only if $1<\sigma<-n/\alpha$. \label{Ma1}
\item For $\alpha \ge 0$ it holds $w \in RH_\sigma$ for all $1<\sigma<\infty$. \label{Ma2}
\end{enumerate}
The first one is well known. The second one is easily obtained using \eqref{apandrh} (for $p=1$). The third one can be checked directly or derived from \eqref{apandrh}.
\end{Rem}
\section{Embeddings of Muckenhoupt weighted Morrey spaces into Muckenhoupt weighted Lebesgue spaces}\label{sec3}
In this section we establish different continuous embeddings between Morrey spaces of the types $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ and $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ into some Muckenhoupt $A_1$ weighted Lebesgue spaces which will give us the possibility to define operators on these Morrey spaces by restriction. Extension by continuity is usually not working in Morrey type spaces by their nonseparability and also by the fact that the smooth functions are not dense in these spaces (see \cite[Prop.~3.7]{RoT14_2} for $\ensuremath{{ L}^{r}_{p}(\ensuremath{{\mathbb R}^n})}$ and \cite[Prop.~3.2]{RS16} for $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ for the nonseparability, and see \cite[p.~22]{Pic69} for the nondensity). In recent literature this difficulty has been treated using duality for linear operators, whenever the integration parameter $p$ satisfies $1<p<\infty$. Dealing with the extension problem through embeddings one is neither restricted in the parameter range of $p$ (in particular, $p=1$ is admissible) nor the operator has to be linear.
\begin{Prop} \label{LEmb}
Let $1\le\gamma < p<\infty$, $-n/p \le r<0$ and $w\in A_{p/\gamma}$.
Then there exists $q_0\in(\gamma, p)$ such that for each $q\in [\gamma,q_0]$, $1<s\le s_0(q)$, each cube $Q$, and $h\in L_{(p/q)',w}(Q)$ with norm $1$, the continuous embedding
\begin{equation} \label{emb1new}
\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})} \ensuremath{\hookrightarrow} L_{q,M(h^sw^s\chi_Q)^{1/s}}(\ensuremath{{\mathbb R}^n})
\end{equation}
holds with constant independent of $h$ and depending on $Q$ as $w(Q)^{1/p+r/n}$.
Furthermore, with the additional assumptions $w\in RH_{\sigma}$ and $r\le-n/(p\sigma)$, it also holds that
\begin{equation} \label{emb2new}
\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})} \ensuremath{\hookrightarrow} L_{q,M(h^sw^s\chi_Q)^{1/s}}(\ensuremath{{\mathbb R}^n})
\end{equation}
with constant independent of $h$ and depending on $Q$ as $|Q|^{1/p+r/n}$.
As a consequence, under the same conditions on $p$, $r$ and $w$, for $q\in [\gamma,q_0]$ there is $0<\alpha<n$ such that
\begin{equation} \label{emb1}
\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}, \ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})} \ensuremath{\hookrightarrow} L_{q,(1+|x|)^{-\alpha}}(\ensuremath{{\mathbb R}^n}).
\end{equation}
\end{Prop}
\begin{proof} We prove the result for $\gamma=1$. The case $\gamma>1$ will follow as a consequence.
Choose $q_0\in (1,p)$ such that $w\in A_{p/q_0}$. For $q\in [1,q_0]$, $w\in A_{p/q}$. Fix one such $q$ and set $\tilde p=p/q$. Let $1<s<\tilde p'$ such that $w^{1-\tilde p'}\in A_{\tilde p'/s}$.
For a function $h$ with $\Vert h| L_{\tilde p',w}(Q)\Vert=1$, the weight $M(h^s w^s \chi_{Q})^{1/s}$ is well defined because $h^s w^s \chi_{Q} \in L_1(\ensuremath{{\mathbb R}^n})$, which implies $M(h^s w^s \chi_{Q}) < \infty$ almost everywhere (by the fact that $M$ maps $L_1(\ensuremath{{\mathbb R}^n})$ to the weak $L_1(\ensuremath{{\mathbb R}^n})$ space). We check that $h^s w^s \chi_{Q} \in L_1(\ensuremath{{\mathbb R}^n})$ and obtain an estimate that will be needed below. Indeed,
\begin{align}
\begin{split} \label{p3}
\left(\int_{Q} h^s w^{s-1} w\right)^\frac{1}{s}
&\le\norm{h}{L_{\tilde p',w}(\ensuremath{{\mathbb R}^n})}
\left(\int_{Q} w^{(s-1) \frac{\tilde p'}{\tilde p'-s} +1} \right)^{\frac1{s}-\frac{1}{\tilde p'}}\\
& \le c\ |Q|^{\frac{1}{s}}w^{1- \tilde p'}(Q)^{-\frac{1}{\tilde p'}}
\le c\ w(Q)^{\frac{1}{\tilde p}} |Q|^{-\frac{1}{s'}},
\end{split}
\end{align}
where the second inequality holds because $w^{1-\tilde p'}\in A_{\tilde p'/s}$ (the exponent of $w$ in the integral is the same as $(1-\tilde p')(1-(\tilde p'/s)')$) and in the last one we use
\begin{equation*}
|Q|\le w(Q)^{\frac{1}{\tilde p}} w^{1- \tilde p'}(Q)^{\frac{1}{\tilde p'}}.
\end{equation*}
Let $f\in \ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ and assume that $f$ is nonnegative. To estimate the norm of $f$ in $L_{q,M(h^sw^s\chi_Q)^{1/s}}(\ensuremath{{\mathbb R}^n})$ we decompose the integral into two parts: first we integrate over $2Q$ and next over $\ensuremath{{\mathbb R}^n}\setminus 2Q$. Over $2Q$ we use H\"older's inequality,
\begin{equation*}
\left(\int_{2Q} f^{q} M(h^s w^s \chi_{Q})^\frac{1}{s}\right)^\frac 1q \le
\left(\int_{2Q} f^{p}w\right)^{\frac1{p}}\left(\int_{2Q} M(h^s w^s \chi_{Q})^{\frac{\tilde p'}s}w^{1- \tilde p'}\right)^{\frac 1{q\tilde p'}}.
\end{equation*}
The first term of the right-hand side is bounded by $w(2Q)^{\frac{1}{p}+\frac{r}{n}} \norm{f}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}$ and using \eqref{DC} we can replace $w(2Q)$ by $w(Q)$. In the second one we use that $w^{1-\tilde p'}\in A_{\tilde p'/s}$ and the boundedness of $M$ on $L_{\tilde p'/s,w^{1-\tilde p'}}(\ensuremath{{\mathbb R}^n})$ to get a constant times the norm of $h$ in $L_{\tilde p',w}(\ensuremath{{\mathbb R}^n})$, which is $1$.
To handle the integral over $\ensuremath{{\mathbb R}^n}\setminus 2Q$, we use that $M(h^s w^s \chi_{Q})(x)$ for $x\in 2^{i+1} Q\setminus 2^i Q$ ($i\ge 1$) is comparable to the average of $h^s w^s \chi_{Q}$ over $ 2^{i+1} Q$.
Then we obtain
\begin{align}
\begin{split} \label{p4}
& \int_{\ensuremath{{\mathbb R}^n}\setminus 2Q} f^{q}M(h^s w^s \chi_{Q})^\frac{1}{s}
\le c_1 \sum_{i=1}^\infty \int_{2^{i+1} Q\setminus 2^i Q} f^{q}\left( \frac{\int_{Q} h^s w^s}{|2^{i}Q|} \right)^\frac{1}{s}
\\ &\le c_2 \sum_{i=1}^\infty \int_{2^{i+1} Q\setminus 2^i Q} f^{q} w^{\frac 1{\tilde p}} w^{-\frac 1{\tilde p}}\frac{w(Q)^{\frac 1{\tilde p}}}{|2^{i}Q|^\frac{1}{s} |Q|^{\frac{1}{s'}}}
\\ &\le c_2 \sum_{i=1}^\infty \left(\int_{2^{i+1} Q\setminus 2^i Q} f^{p} w \right)^\frac{q}{p} w^{1-\tilde p'} \left(2^{i+1}Q\right)^{\frac 1{\tilde p'}} \frac{w(Q)^{\frac 1{\tilde p}}}{2^{in/s}|Q|}
\\ &\le c_3 \norm{f}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}^q
\sum_{i=1}^\infty w\left(2^{i+1} Q \right)^{\left(\frac{1}{p}+\frac{r}{n}\right)q} w^{1-\tilde p'} \left(2^{i+1}Q\right)^{\frac 1{\tilde p'}} \frac{w(Q)^{\frac 1{\tilde p}}}{2^{in/s}|Q|}.
\end{split}
\end{align}
Since $w\in A_{\tilde p}$, then
\begin{equation}\label{p6b}
w\left(2^{i+1} Q \right)^{\frac 1{\tilde p}} w^{1-\tilde p'} \left(2^{i+1}Q\right)^{\frac 1{\tilde p'}}\le c |2^{i+1} Q|,
\end{equation}
and it is sufficient to show
\begin{align} \label{p5}
\sum_{i=1}^\infty w\left(2^{i+1} Q \right)^{\frac{r}{n}q} w(Q)^\frac{q}{p} 2^{\frac{in}{s'}}
\le c \, w\left( Q \right)^{\left(\frac{1}{p}+\frac{r}{n}\right)q}.
\end{align}
Taking into account \eqref{ineqRH}, there exists $\delta>0$ such that
\begin{align*
\begin{split}
w(2^{i+1}Q)^{\frac{r}{n}q} \le c\, 2^{i\delta r q} w(Q)^{\frac{r}{n}q}.
\end{split}
\end{align*}
Therefore, \eqref{emb1new} is satisfied for $s$ sufficiently close to $1$.
In the case of the spaces $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$, the part corresponding to the integral over $2Q$ is obtained in a similar way using
\[
\left(\int_{2Q} f^{p}w\right)^{\frac1{p}}
\le c |Q|^{\frac{1}{p}+\frac{r}{n}} \norm{f}{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}.
\]
For the integral over $\ensuremath{{\mathbb R}^n}\setminus 2Q$, we first modify the last line of \eqref{p4} and then use \eqref{p6b} to get
\begin{align*}
\begin{split}
& \int_{\ensuremath{{\mathbb R}^n}\setminus 2Q} f^{q}M(h^s w^s \chi_{Q})^\frac{1}{s}
\\ & \le c_1 \norm{f}{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}^q
\sum_{i=1}^\infty \left|2^{i+1} Q \right|^{\left(\frac{1}{p}+\frac{r}{n}\right)q} w^{1-\tilde p'} \left(2^{i+1}Q\right)^{\frac 1{\tilde p'}} \frac{w(Q)^{\frac 1{\tilde p}}}{2^{in/s}|Q|}
\\ & \le c_2 \norm{f}{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}^q
\sum_{i=1}^\infty \left|2^{i+1} Q \right|^{\left(\frac{1}{p}+\frac{r}{n}\right)q}
2^{\frac{in}{s'}} \left[w(2^{i+1}Q)^{-1}\ w(Q)\right]^\frac{q}{p}.
\end{split}
\end{align*}
By the latter we obtain \eqref{emb2new} showing that
\begin{align}
\begin{split} \label{p10}
\sum_{i=1}^\infty 2^{inq[\frac{1}{p}+\frac{r}{n}+\frac{1}{s'q}]}
\left[w(2^{i+1}Q)^{-1}\ w(Q)\right]^\frac{q}{p} \le c.
\end{split}
\end{align}
By the fact that $w\in RH_{\sigma}$, from \eqref{ineqRH} we deduce
\begin{align}
\begin{split} \label{p12}
\frac{w(Q)}{w(2^{i}Q)}
\le c \left(\frac{|Q|}{|2^{i}Q|}\right)^{1/\sigma'}
= c 2^{-i\frac n{\sigma'}}.
\end{split}
\end{align}
Inserting \eqref{p12} into the left hand-side of \eqref{p10} and taking into account that $s$ is as close to $1$ as desired, the geometric series is convergent for $r/n+1/(p\sigma)<0$. The endpoint $r=-n/(p\sigma)$ of the statement is attained from the openness property of the reverse H\"older classes.
Let now $1<\gamma<p$ and $f\in \ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ with $w\in A_{p/\gamma}$. Since
\begin{equation} \label{TFE}
\norm{f}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}
= \norm{|f|^\gamma}{L^{r\gamma}_{p/\gamma}(w,\ensuremath{{\mathbb R}^n})}^\frac{1}{\gamma},
\end{equation}
$|f|^\gamma\in L^{r\gamma}_{p/\gamma}(w,\ensuremath{{\mathbb R}^n})$ (with $w\in A_{p/\gamma}$) and we can apply the first part of the proof. It gives the existence of $q_0^*\in (1,p/\gamma)$ and $s>1$ such that for $q^*\in (1,q_0^*]$,
\begin{equation*}
\norm{|f|^\gamma}{L^{r\gamma}_{p/\gamma}(w,\ensuremath{{\mathbb R}^n})}\le c \norm{|f|^\gamma}{L_{q^*,M(h^sw^s\chi_Q)^{1/s}}(\ensuremath{{\mathbb R}^n})},
\end{equation*}
for any cube $Q$ and $h\in L_{(p/(\gamma q*))',w}(Q)$ with norm $1$. Define $q_0=q_0^*\gamma$ and $q=q^*\gamma$ and \eqref{emb1new} is proved. The case $f\in\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ with $w\in A_{p/\gamma}$ is treated similarly.
To obtain \eqref{emb1}, let $Q$ be the unit cube and $h=cw^{-1}$, with $c$ chosen such that $h$ has unit norm. Use now that $M(\chi_Q)(x)$ is bounded below by a positive constant if $x\in 2Q$ and by $c|x|^{-n}$ if $x\notin 2Q$. Take $\alpha=n/s$ to conclude.
\end{proof}
\begin{Rem} \label{3.2T} Since $w\in A_p$ implies that $w\in RH_\sigma$ for some $\sigma>1$, the range of values of $r$ satisfying the embedding \eqref{emb1} for $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ is not empty.
In the case of the weights $|x|^\beta$ and $(1+|x|)^\beta$ such range is
\[
\begin{cases}
-\frac{n}{p} \le r<0, & \text{ if }\ 0\le \beta<n(\frac p{\gamma}-1); \\
-\frac{n}{p} \le r<\frac{\beta}{p}, &\text{ if }\ -n<\beta<0.
\end{cases}
\]
\end{Rem}
The following proposition shows that the union of the $L_{q,w}(\ensuremath{{\mathbb R}^n})$ spaces for fixed $q$ and all $w\in A_q$ is invariant for $1<q<\infty$. This result appears in \cite{KMM16} in a more general context. We give here a direct proof.
\begin{Prop} \label{DfOp}
Let be $1\le \gamma< q<\infty$.
Then it holds that
\begin{equation} \label{DefOp}
\bigcup_{w\in A_{q/\gamma}} L_{q,w}(\ensuremath{{\mathbb R}^n}) = \bigcup_{\gamma<p<\infty} \bigcup_{w\in A_{p/\gamma}} \ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}
\subsetneq \bigcup_{w\in A_1} L_{\gamma,w}(\ensuremath{{\mathbb R}^n}) .
\end{equation}
Hence, the left-hand side is independent of $q$.
\end{Prop}
\begin{proof} Again it is enough to prove the case $\gamma=1$. Once this is proved, the general case is deduced from the observation that $f\in \ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}$ with $w\in A_{p/\gamma}$ is equivalent to $|f|^\gamma\in L_{p/\gamma, w}(\ensuremath{{\mathbb R}^n})$ (with $w\in A_{p/\gamma}$).
Let $\gamma=1$. We start showing the equality in \eqref{DefOp}.
We first prove that if $f$ is in $\ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}$ for $1<p<q$ and $w\in A_p$, then $f\in \bigcup_{v\in A_q} L_{q,v}(\ensuremath{{\mathbb R}^n})$.
For $s>1$ sufficient small $w\in A_{p/s}$ holds. Moreover, we have $f^s \in L_{p/s,w}$ and therefore $M(f^s)\in L_{p/s,w}$. Hence $(M(f^s))^{1/s}\in A_1$. Thus, $(M(f^s))^{(p-q)/s}w\in A_{q}$ (see \cite[Lemma 2.1]{D11}).
By reason of $p<q$ finally it holds that
\[
\int_{\ensuremath{{\mathbb R}^n}} |f|^q (M(f^s))^{\frac{1}{s} (p-q)}w \le \int_{\ensuremath{{\mathbb R}^n}} |f|^q |f|^{p-q}w
\]
and $f\in \bigcup_{v\in A_q} L_{q,v}(\ensuremath{{\mathbb R}^n})$.
That means that we have up to now shown
\begin{equation}\label{half}
\bigcup_{w\in A_q} L_{q,w}(\ensuremath{{\mathbb R}^n}) = \bigcup_{1< p\le q} \bigcup_{w\in A_p} \ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}.
\end{equation}
We assume now $f\in \ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}$ for some $p>q$ and $w\in A_p$. By \eqref{emb1} (applied to $r=-n/p$, since ) we obtain $f\in L_{\tilde q,(1+|x|)^{-\alpha}}(\ensuremath{{\mathbb R}^n})$ for some $\tilde q<q$ and so $f$ is in the right-hand side of \eqref{half}. Here, $\tilde q=1$ is also admitted.
It remains to prove that the inclusion of \eqref{DefOp} is strict. For this it is enough to notice that there exists $f\in L^1(\ensuremath{{\mathbb R}^n})$ for which $Mf$ is not locally integrable. Indeed, such an $f$ cannot be in $L_{p,w}(\ensuremath{{\mathbb R}^n})$ for any $p\in (1,\infty)$ because this would imply $Mf \in L_{p,w}(\ensuremath{{\mathbb R}^n})\subset L_1^{\textrm{loc}}(\ensuremath{{\mathbb R}^n})$. A typical example of a function as the one required here is
$f(x)= |x|^{-n} (\ln{|x|})^{-2}\ \chi_{\{|x|\le 1/2\}}$.
\end{proof}
The following embedding result on the considered Morrey spaces is a generalization of \cite[Prop.~2.10]{T-HS} to weighted situations.
\begin{Prop} \label{LEmbE}
Let $1\le p<\infty$ and $\nu \ge 1$. Let $w\in A_1\cap RH_\sigma$ (that is, $w^\sigma\in A_{1}$) for some $\sigma>\nu$. Then for $r\in [-n/p, -n\sigma'/(p\nu')]$ (if $\nu=1$, the endpoint $r=0$ is excluded), there exists $s_0>\nu$ such that for $\nu<s\le s_0$ and for any cube $Q$, the continuous embedding
\begin{equation} \label{emb1Enew}
\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})} \ensuremath{\hookrightarrow} L_{p,M(w^s\chi_Q)^{1/s}}(\ensuremath{{\mathbb R}^n})
\end{equation}
holds with a constant depending on $Q$ as $w(Q)^{1/p+r/n}$.
Moreover, whenever
$r\in [-n/p,-n(1/\sigma + 1/\nu')/p]$, it also holds that
\begin{equation} \label{emb2Enew}
\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})} \ensuremath{\hookrightarrow} L_{p,M(w^s\chi_Q)^{1/s}}(\ensuremath{{\mathbb R}^n}),
\end{equation}
with a constant depending on $Q$ as $|Q|^{1/p+r/n}$.
As a consequence, under the same conditions on $w$ and $r$, there exists some $0<\alpha<n/\nu$ such that
\begin{equation} \label{emb1E}
\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}, \ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})} \ensuremath{\hookrightarrow} L_{p,(1+|x|)^{-\alpha}}(\ensuremath{{\mathbb R}^n}).
\end{equation}
\end{Prop}
\begin{proof}
Let $s$ be such that $\nu<s<\sigma$. The weight $M(w^s\chi_Q)^{1/s}$ is in $A_1$ and satisfies $M(w^s\chi_Q)^{1/s}\le M(w^s)^{1/s}\le C w$ a.e.\ because $w^s\in A_1$. Moreover, for $x\in 2^{i+1} Q\setminus 2^i Q$, $M(w^s \chi_{Q}) (x)$ is comparable to the average of $w^s \chi_{Q}$ over $2^{i+1} Q$.
Let $f\in \ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ with $r$ as stated and assume that $f$ is nonnegative. Then
\begin{equation*}
\int_{\ensuremath{{\mathbb R}^n}} f^p M(w^s\chi_Q)^{\frac 1s}
\le c_1 \int_{2Q} f^p w + c_2\sum_{i=1}^\infty \int_{2^{i+1} Q\setminus 2^i Q} f^p\left( \frac{w^s(Q)}{|2^{i}Q|} \right)^\frac{1}{s}.
\end{equation*}%
The first term gives immediately the desired estimate. As for the second one, we notice that
\begin{equation*
\left(\frac {w^s(Q)}{|Q|}\right)^{1/s}\le C\frac {w(Q)}{|Q|},
\end{equation*}
because $w\in RH_\sigma\subset RH_s$ and also that $w\in A_1$ implies
\begin{equation*}
\frac{w(2^{i+1}Q)} {|2^{i+1}Q|}\le Cw(x)\quad\text{a.e. }x\in 2^{i+1}Q.
\end{equation*}
Inserting these in the above expression we obtain
\begin{align}
\begin{split} \label{p4E}
\sum_{i=1}^\infty & \int_{2^{i+1} Q\setminus 2^i Q} f^p w \ \frac{|2^{i+1}Q|}{w(2^{i+1}Q)} \frac{w(Q)}{2^{in/s} |Q|}
\\ & \le c_1 \norm{f}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}^p \sum_{i=1}^\infty {w(2^{i+1}Q)}^{\frac{rp}{n}} w(Q) 2^{i\frac{n}{s'}}
\\ & \le c_2 \norm{f}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}^p {w(Q)}^{1+\frac{rp}{n}}\sum_{i=1}^\infty {2}^{i\frac{rp}{\sigma'}}2^{i\frac{n}{s'}},
\end{split}
\end{align}
using that $w\in RH_{\sigma}$ and \eqref{p12}. If $rp<-n\sigma'/\nu'$, we can choose $s>\nu$ sufficiently small such that the geometric series converges. To reach the endpoint $rp=-n\sigma'/\nu'$, use the openness of the reverse H\"older classes.
In the case of the spaces $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$, the left-hand side of \eqref{p4E} is bounded by
\begin{equation*
C \norm{f}{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}^p \sum_{i=1}^\infty (2^{in} |Q|)^{1+\frac {rp}n} 2^{\frac{in}{s'}} \frac{w(Q)}{w(2^{i+1}Q)}.
\end{equation*}
Using \eqref{p12} the geometric series converges for $rp+n(1/s'+1/\sigma)<0$. This gives \eqref{emb2Enew} in the stated range of $r$ (using again the openness of the reverse H\"older classes to get the endpoint).
For the particular embeddings \eqref{emb1E} take as $Q$ the unit cube centered at the origin and $\alpha=n/s$.
\end{proof}
\begin{Rem}
Since we assume $w^\nu\in A_{1}$, we automatically have $w\in RH_{\nu}$. But each such $w$ will be in a reverse H\"older class $w\in RH_{\sigma}$ for some $\sigma>\nu$ and the range of values of $r$ obtained in the proposition is not empty.
In the case $w(x)=|x|^\beta$ or $w(x)=(1+|x|)^\beta$, we need $-n<\beta\nu\le 0$ to fulfill the condition $w^\nu\in A_{1}$ and in such case the range of values of $r$ satisfying the two embeddings of \eqref{emb1E} are respectively
\[
-\frac{n}{p} \le r<-\frac{n}{p} \frac{n}{n+\beta} \frac{1}{\nu'} \quad\text{and}\quad -\frac{n}{p} \le r<\frac{\beta}{p} -\frac{n}{p}\frac{1}{\nu'}.
\]
\end{Rem}
\begin{Corol}\label{newcor}
For every $1\le \gamma\le p_0<\infty$, every $\gamma<p<\infty$ (and also $p=p_0$ if $\gamma=p_0$), $-n/p\le r<0$ and $w\in A_{p/\gamma}$, it holds that
\begin{equation}\label{arb}
\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}\subset \bigcup_{u\in A_{p_0/\gamma}}L_{p_0,u}(\ensuremath{{\mathbb R}^n}).
\end{equation}
If moreover $w\in A_{p/\gamma}\cap RH_{\sigma}$, for every $-n/p\le r< -n/(p \sigma) $ it holds that
\begin{equation}\label{arb2}
\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}\subset \bigcup_{u\in A_{p_0/\gamma}}L_{p_0,u}(\ensuremath{{\mathbb R}^n}).
\end{equation}
\end{Corol}
\begin{proof}
If $1\le \gamma< p_0$ we apply the embedding for $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ given by \eqref{emb1new} of Proposition \ref{LEmb} together with Proposition \ref{DfOp}. If $\gamma=p_0$ and $p=p_0=\gamma$, \eqref{arb} follows from Proposition \ref{LEmbE} with $\nu=1$. Finally, if $\gamma=p_0$ and $p_0=\gamma<p<\infty$ we use the embedding for $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ of Proposition \ref{LEmb} with $q=\gamma$. Analogously, we obtain \eqref{arb2}.
\end{proof}
\begin{Rem}
Note that as a consequence of this corollary, the left-hand side of \eqref{DefOp} in Proposition \ref{DfOp} also coincides with the (formally larger) union
\[\bigcup_{\gamma<p<\infty} \bigcup_{-n/p\le r<0} \bigcup_{w\in A_{p/\gamma}} \ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}.\]
(We recall that ${L}^{-{n}/{p}}_p(w,\ensuremath{{\mathbb R}^n})=L_{p, w}(\ensuremath{{\mathbb R}^n})$.)
\end{Rem}
\begin{Prop} \label{LEmbP2}
Let $1\le \gamma< b<\infty$ and $\sigma(p)=(b/p)'$ for $\gamma<p<b$. Let $w\in A_{p/\gamma}\cap RH_{\sigma}$ for $\sigma>\sigma(p)$. Let $r\in \left[-n/p , -n\sigma'/b\right]$. Then there exist $q_0\in(\gamma, p)$ and for each $q\in [\gamma,q_0]$, some $s_0(q)>\sigma (q)$, such that for $\sigma(q)<s\le s_0(q)$, each cube $Q$ and $h\in L_{(p/q)',w}(Q)$ with norm $1$, the continuous embedding
\begin{equation} \label{emb11new}
\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})} \ensuremath{\hookrightarrow} L_{q,M(h^sw^s\chi_Q)^{1/s}}(\ensuremath{{\mathbb R}^n})
\end{equation}
holds with constant independent of $h$ and depending on $Q$ as $w(Q)^{1/p+r/n}$.
Furthermore, with the additional assumption $r\in \left[-n/p ,-n/(p\sigma) - n/b \right]$, it also holds that
\begin{equation} \label{emb22new}
\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})} \ensuremath{\hookrightarrow} L_{q,M(h^sw^s\chi_Q)^{1/s}}(\ensuremath{{\mathbb R}^n})
\end{equation}
with constant independent of $h$ and depending on $Q$ as $|Q|^{1/p+r/n}$.
As a consequence, under the same conditions on $p$, $r$ and $w$, there exists $0<\alpha<n/\sigma (q)$ such that
\begin{equation} \label{emb1P2}
\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}, \ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})} \ensuremath{\hookrightarrow} L_{q,(1+|x|)^{-\alpha}}(\ensuremath{{\mathbb R}^n}).
\end{equation}
\end{Prop}
\begin{proof}
We can consider $\gamma=1$. Once this is proved, the case $\gamma>1$ is obtained from it by considering $|f|^\gamma$ as in Proposition \ref{LEmb} (see \eqref{TFE}).
Let $f\in \ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ nonnegative with $w\in A_p\cap RH_{\sigma}$.
Choose $q\in(1,p)$ such that $w\in A_{p/q}$ and set $\tilde p=p/q$. By \eqref{apandrh}, $w\in A_{\tilde p}\cap RH_{\sigma}$ implies $w^\sigma \in A_{\sigma(\tilde p-1)+1}$, which in turn implies $w^{1-\tilde p'}\in A_{1+(\tilde p'-1)/\sigma}$ by the duality property of the weights. If $1+(\tilde p'-1)/\sigma\le \tilde p'/s$, then $w^{1-\tilde p'}\in A_{\tilde p'/s}$. This is possible with $s>\sigma (q)$ if
\begin{equation*}
1+\frac{\tilde p'-1}{\sigma}<\frac{\tilde p'}{\sigma(q)}=\frac{\tilde p'}{(b/q)'},
\end{equation*}
which is the same as
\begin{equation*}
\frac 1{\tilde p'}+\frac1{\tilde p \sigma}<\frac{1}{(b/q)'}= 1-\frac qb,
\end{equation*}
that is, $\sigma>\sigma(p)$.
The proof continues as for Proposition \ref{LEmb}. In particular, the estimate \eqref{p3} remains true because $w^{1-\tilde p'}\in A_{\tilde p'/s}$, so that we will get \eqref{emb11new} if \eqref{p5} holds. Since $w\in RH_{\sigma}$, we use \eqref{p12} and the left-hand side of \eqref{p5} is bounded by
\begin{equation*}
c \, w\left( Q \right)^{\left(\frac{1}{p}+\frac{r}{n}\right)q} \sum_{i=1}^\infty 2^{i\left(\frac{rq}{\sigma'}+\frac n{s'}\right)},\end{equation*}
The geometric series converges for $r < -n\sigma'/(qs')$. It is possible to have $s>\sigma(q)$, if $r<-n\sigma'/b$. For the endpoint $r=-n\sigma'/b$, use the openness property of the reverse H\"older classes.
In the case of the spaces $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$, to get \eqref{emb22new} we follow again the proof of Proposition \ref{LEmb} and we are left with \eqref{p10}. We insert \eqref{p12} and the geometric series converges if
\begin{equation*}
\frac{nq}p+r+\frac{n}{s'}-\frac{nq}{p\sigma'}=\frac{nq}{p\sigma}+r+\frac{n}{s'}<0.
\end{equation*}
We can get $s>\sigma(q)$ if
\begin{equation*}
r<-\frac{nq}{p\sigma}-\frac{n}{\sigma(q)'}=-\frac{nq}{p\sigma}-\frac{nq}{b}.
\end{equation*}
This holds for $q$ close to $1$ if $r<-n/(p\sigma) - n/b$.
Take as $Q$ the unit cube and as $h=cw^{-1}$ to get the embeddings of \eqref{emb1P2} with $\alpha=n/s$.
\end{proof}
\begin{Prop} \label{LEmbP2B}
Let $1\le \gamma< b<\infty$ and $\sigma(p)=(b/p)'$ for $\gamma<p<b$. Let $\gamma<q<b$.
Then
\begin{equation*}
\bigcup_{w\in A_{q/\gamma}\cap RH_{\sigma(q)}} L_{q,w}(\ensuremath{{\mathbb R}^n}) = \bigcup_{\gamma< p<b} \bigcup_{w\in A_{p/\gamma}\cap RH_{\sigma(p)}} \ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}.
\end{equation*}
\end{Prop}
\begin{proof}
We assume $\gamma=1$ as in the previous proposition.
We notice first that using the factorization theorem of $A_p$ weights we have
\begin{equation*}
A_p\cap RH_{\sigma(p)}=\{w: w^{\sigma(p)} \in A_{\sigma(p)(p-1)+1} \}
=\{u_0^{1-\frac pb}u_1^{1-p}: u_0, u_1\in A_1\}.
\end{equation*}
Let $f\in L_{p,w}(\ensuremath{{\mathbb R}^n})$ with $w=u_0^{1-p/b}u_1^{1-p}$ for some $u_0, u_1\in A_1$, and let $1<p<q<b$. We have $fu_0^{-1/b}\in L^p (u_0 u_1^{1-p})$. Since $u_0 u_1^{1-p}\in A_p$, there exists $s>1$ such that $u_0 u_1^{1-p}\in A_{p/s}$. Then $(fu_0^{-1/b})^s\in L^{p/s} (u_0 u_1^{1-p})$. Hence $M((fu_0^{-1/b})^s)^{1/s}\in A_1$ and
\begin{equation*}
|f(x)|\le M((fu_0^{-\frac 1b})^s)^{\frac 1s}(x) u_0^{\frac 1b}(x)\text{ a.e.}
\end{equation*}
We deduce that
\begin{equation*}
\int_{\ensuremath{{\mathbb R}^n}} |f|^q [M((fu_0^{-\frac 1b})^s)^{\frac 1s} u_0^{\frac 1b}]^{p-q} u_0^{1-\frac pb}u_1^{1-p} \le
\int_{\ensuremath{{\mathbb R}^n}} |f|^p u_0^{1-\frac pb}u_1^{1-p}.
\end{equation*}
The weight of the left side is
\begin{equation*}
u_0^{1-\frac qb} [M((fu_0^{-\frac 1b})^s)^{\frac 1s}]^{p-q} u_1^{1-p}=u_0^{1-\frac qb} [v^\theta u_1^{1-\theta}]^{1-q},
\end{equation*}
where $v=M((fu_0^{-1/b})^s)^{1/s}$ and $\theta=(q-p)/(q-1) \in (0,1)$. Thus, it is in $A_q\cap RH_{\sigma(q)}$, since $v^\theta u_1^{1-\theta}\in A_1$. This proves that
\begin{equation*}
\bigcup_{1< p\le q} \bigcup_{w\in A_p\cap RH_{\sigma(p)}} \ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})} = \bigcup_{w\in A_q\cap RH_{\sigma(q)}} L_{q,w}(\ensuremath{{\mathbb R}^n}).
\end{equation*}
This is enough to prove the statement because for $p>q$ we know from Proposition \ref{LEmbP2} (taking $r=-n/p$) that a function in $L_{p,w}(\ensuremath{{\mathbb R}^n})$ is in some $L_{\tilde q,(1+|x|)^{-\alpha}}(\ensuremath{{\mathbb R}^n})$ with $\tilde q$ near $1$ and the weight $(1+|x|)^{-\alpha}$ in $A_1 \cap RH_{\sigma(\tilde q)}\subset A_{\tilde q}\cap RH_{\sigma(\tilde q)}$.
\end{proof}
Our embeddings directly apply to similar results with respect to $A_p$-weighted weak $L_p(\ensuremath{{\mathbb R}^n})$ spaces ($W\ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}$ is the space $W\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ of Definition \ref{d1:def} for $r=-n/p$), which are nonseparable. These kind of embeddings allow one to extend operators to a domain which is a weak $L_p$ space via restriction. The connection with the Morrey spaces $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ lies in the use of a different norm for $W\ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}$.
For $1<p<\infty$ the space $W\ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}$ can be normed with the norm
\begin{equation} \label{2:GG}
\left\|f|W\ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}\right\|_*\equiv
\sup_{M} w(M)^{\left(\frac{1}{p}-\frac{1}{\tilde p} \right)} \left\|f|L_{\tilde p, w}(M)\right\|,
\end{equation}
where $0<\tilde p<p$ and the supremum is taken over all measurable subsets of $\mathbb{R}^n$ with $0<w(M)<\infty$. This norm is equivalent to the usual quasi-norm given in Definition \ref{d1:def} (\cite[Exercise 1.1.12]{G08}). It is obvious from \eqref{2:GG} that
\begin{equation*}
W\ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})} \ensuremath{\hookrightarrow} L_{\tilde p}^{r}(w,\ensuremath{{\mathbb R}^n}),
\end{equation*}
with $r=-n/p>-n/\tilde p$.
Moreover, if $w\in A_{p/\gamma}$ for some $\gamma\in [1,p)$, $\tilde p$ can be chosen such that $w\in A_{\tilde p/\gamma}$, and Proposition \ref{LEmb} can be used to deduce that there exists $q_0>\gamma$ such that for each $q\in [\gamma,q_0]$ there is $0<\alpha<n$ for which
\begin{equation*}
W\ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})} \ensuremath{\hookrightarrow} L_{q,(1+|x|)^{-\alpha}}(\ensuremath{{\mathbb R}^n}).
\end{equation*}
The same type of embeddings can be obtained from Propositions \ref{LEmbE} and \ref{LEmbP2}.
\section{Boundedness on Morrey spaces: general results}\label{sec4}
In this section we obtain the general theorems giving the boundedness on the corresponding Morrey spaces from the assumption of weighted estimates on $\ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}$ spaces. Instead of using operators, we can establish the general theorems in terms of pairs of functions, as it is usual in the recent presentations of the extrapolation theorems. This has the advantage of providing immediately several different versions.
\begin{Thm} \label{TE1}
Let $1\le p_0<\infty$ and let $\mathcal F$ be a collection of nonnegative measurable pairs of functions. Assume that for every $(f,g)\in \mathcal F$ and every $w\in A_{p_0}$ we have
\begin{equation} \label{1exa1}
\norm{g}{L_{p_0,w}(\ensuremath{{\mathbb R}^n})}\le c_1 \norm{f}{L_{p_0,w}(\ensuremath{{\mathbb R}^n})},
\end{equation}
where $c_1$ does not depend on the pair $(f,g)$ and it depends on $w$ only in terms of $[w]_{A_{p_0}}$.
Then for every
$1< p<\infty$ (and also for $p=1$ if $p_0=1$), every $-n/p \le r<0$ and every $w\in A_{p}$ we have
\begin{equation} \label{2exa2}
\norm{g}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}\le c_2 \norm{f}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}.
\end{equation}
Furthermore, for every
$1< p<\infty$ (and also for $p=1$ if $p_0=1$) and $w\in A_{p} \cap RH_{\sigma}$, if $-n/p \le r\le-n/(p\sigma)$ we have
\begin{equation} \label{3exa3}
\norm{g}{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}\le c_3 \norm{f}{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}.
\end{equation}
The constants $c_2$ and $c_3$ in \eqref{2exa2} and \eqref{3exa3}
do not depend on the pair $(f,g)$ but may depend on $w$ and the involved parameters.
If \eqref{1exa1} is assumed to hold only when the left-hand side is finite, \eqref{2exa2} and \eqref{3exa3} still hold assuming that their left-hand side is finite.
\end{Thm}
\begin{Rem}
In \eqref{1exa1} we do not need to require that $f$ and $g$ are in $L_{p_0,w}(\ensuremath{{\mathbb R}^n})$ for all $w\in A_{p_0}$. In the first part of the statement we assume that \eqref{1exa1} holds whenever the right-hand side is finite (and in such case, this forces the left-hand side to be finite). Then the conclusion of the theorem is that for every $f$ in $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ or in $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$, $g$ is in the same space and the norm inequality is satisfied. When $g=Tf$ for some operator $T$ this means that $Tf$ is in the same Morrey space as $f$ is, for all the functions in the space.
In the last part of the statement we only assume that \eqref{1exa1} holds when the left-hand side is finite. That is, it could happen that for a particular weight and a particular pair of functions the left-hand side is infinity and the right-hand side is finite. Under this less restrictive hypothesis, the fact that $f$ is in $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ or in $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ does not imply the same thing for $g$, but if $g$ also is assumed to be in the corresponding Morrey space, then the norm inequality holds.
In either case the extrapolation theorem in the usual weighted Lebesgue spaces theorem says that if \eqref{1exa1} holds for a particular $p_0$, it holds for $1<p<\infty$ and $A_p$ weights with the same interpretation of the inequalities. The proof given in \cite{D11}, for instance, is only valid in the first situation, although this is not explicitely indicated in the paper. The proofs in \cite{CMP11} cover the second case (left-hand side finite), and are valid also for the first case. In the applications to operators, the advantage of using the first version is that one proves directly the result for the whole weighted Lebesgue space, without making the extension from a dense subclass.
\end{Rem}
The part of the theorem corresponding to \eqref{2exa2} in the case $p>1$ was proved in \cite{RS16} using a different method, which involves the predual Morrey spaces.
\begin{proof} (a) \textit{The case $p>1$}. From the extrapolation theorem we can assume \eqref{1exa1} for any $p\in (1,\infty)$ and the weight class $A_p$.
Let $w\in A_{p}$. Choose $q\in (1,p)$ and $s>1$ such that the embedding \eqref{emb1new} holds. Set $\tilde p=p/q$.
Let $Q$ be a fixed cube. Since
\begin{equation*}
\left(\int_{Q} g^{p}w\right)^\frac{1}{p} = \left(\int_{Q} g^{\tilde p q}w\right)^\frac{1}{\tilde p q}
=\sup_{h\,:\,\Vert h| L_{\tilde p',w}(Q)\Vert=1}\left(\int_{Q} g^{q}h w\right)^\frac{1}{q},
\end{equation*}
we fix such a function $h$ and we have
\begin{align} \label{p2}
\begin{split}
\left(\int_{\ensuremath{{\mathbb R}^n}} g^{q}h w \chi_{Q}\right)^\frac{1}{q} & \le \left(\int_{\ensuremath{{\mathbb R}^n}} g^{q}M(h^s w^s \chi_{Q})^\frac{1}{s}\right)^\frac{1}{q}\\
& \le c \left(\int_{\ensuremath{{\mathbb R}^n}} f^{q}M(h^s w^s \chi_{Q})^\frac{1}{s}\right)^\frac{1}{q},
\end{split}
\end{align}
where the second inequality holds because $M(h^s w^s \chi_{Q})^{1/s}\in A_1\subset A_q$. We checked in the proof of Proposition \ref{LEmb} that the weight is well defined. Note that the constant in \eqref{p2} is independent of $Q$ and $h$ because the $A_1$-constant of $M(h^s w^s \chi_{Q})^{1/s}$, which is an upper bound of the $A_q$-constant, depends only on $s$ (actually, it behaves as $(s-1)^{-1}$; see \cite[Theorem 9.2.7]{G09}).
We are now in the situation of the embedding \eqref{emb1new} and the proof of \eqref{2exa2} for $p>1$ is complete.
In the case of the spaces $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$, we use the embedding \eqref{emb2new}.
(b) \textit{The case $p=1$}. We assume now that \eqref{1exa1} holds for $p_0=1$ and $w\in A_1$ and prove the theorem for $p=1$ and $w\in A_1$.
We want to apply the embeddings of Proposition \ref{LEmbE} with $\nu=1$. Choose $s>1$ such that \eqref{emb1Enew} and \eqref{emb2Enew} hold. Fix a cube $Q\subset\ensuremath{{\mathbb R}^n}$. The weight $M(w^s\chi_Q)^{1/s}$ is in $A_1$. Then we have
\begin{equation*}
\int_{Q} gw \le \int_{Q} g M(w^s\chi_Q)^{\frac 1s} \le c_1 \int_{\ensuremath{{\mathbb R}^n}} f M(w^s\chi_Q)^{\frac 1s}.
\end{equation*}%
Thus, \eqref{1exa1} and \eqref{2exa2} follow from \eqref{emb1Enew} and \eqref{emb2Enew}.
(c) If the assumption \eqref{1exa1} holds only when the left-hand side is finite, the proof is still valid assuming that the left-hand side of \eqref{2exa2} or \eqref{3exa3} is finite. Indeed, since $g$ is in $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ or in $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$, the embeddings \eqref{emb1new} and \eqref{emb2new} say that the second term in \eqref{p2} is finite and the proof of part (a) goes on. The same thing is true for part (b) of the proof using the embeddings of Proposition \ref{LEmbE}.
\end{proof}
There are weighted inequalities between pairs of operators with $L_{p,w}$-quasinorms for $0<p<\infty$ and $w\in A_\infty$. Such inequalities hold when the left-hand side is finite. We can get similar type of estimates with Morrey quasinorms.
\begin{Corol} \label{TEAinf}
Let $0< p_0<\infty$ and let $\mathcal F$ be a collection of nonnegative measurable pairs of functions. Assume that for every $(f,g)\in \mathcal F$ and every $w\in A_{\infty}$ we have
\begin{equation} \label{1exa1inf}
\norm{g}{L_{p_0,w}(\ensuremath{{\mathbb R}^n})}\le c_1 \norm{f}{L_{p_0,w}(\ensuremath{{\mathbb R}^n})},
\end{equation}
whenever the left-hand side is finite.
Then for every
$0< p<\infty$, every $-n/p \le r<0$ and every $w\in A_{\infty}$ we have
\begin{equation} \label{2exa2inf}
\norm{g}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}\le c_2 \norm{f}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}},
\end{equation}
whenever the left-hand side is finite.
Furthermore, if $w\in RH_\sigma$ and $-n/p \le r\le-n/(p\sigma)$ we have
\begin{equation} \label{3exa3inf}
\norm{g}{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}\le c_3 \norm{f}{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}},
\end{equation}
whenever the left-hand side is finite.
\end{Corol}
\begin{proof}
By extrapolation in $L_{p,w}$ spaces we can assume that \eqref{1exa1inf} holds for all $p_0\in (0,\infty)$ (see \cite{CMP04}).
Let $0<p<\infty$. Given $w\in A_{\infty}$, take $q\in [1,\infty)$ such that $w\in A_q$. If $p\ge q$, since $A_q\subset A_p$, the result follows directly from Theorem \ref{TE1}. If $p<q$, we have
\begin{equation*}
\norm{g^{p/q}}{L_{1,u}(\ensuremath{{\mathbb R}^n})}\le c \norm{f^{p/q}}{L_{1,u}(\ensuremath{{\mathbb R}^n})},
\end{equation*}
for $u\in A_1$, whenever the left-hand side is finite, and we apply Theorem \ref{TE1} to obtain
\begin{equation*}
\norm{g^{p/q}}{{ L}^{rp/q}_{q}(w,\ensuremath{{\mathbb R}^n})}\le c_2 \norm{f^{p/q}}{{ L}^{rp/q}_{q}(w,\ensuremath{{\mathbb R}^n})},
\end{equation*}
which is the desired result using the scaling \eqref{TFE} with $\gamma=p/q$.
The proof of \eqref{3exa3inf} is similar.
\end{proof}
\begin{Corol} \label{TE2}
Let $1\le \gamma \le p_0<\infty$.
Assume that $T$ is an operator acting from $\bigcup_{w\in A_{p_0/\gamma}} L_{p_0,w}(\ensuremath{{\mathbb R}^n})$ into the space of measurable functions satisfying
\begin{equation} \label{ex1}
\norm{Tf}{L_{p_0,w}(\ensuremath{{\mathbb R}^n})}\le c_1 \norm{f}{L_{p_0,w}(\ensuremath{{\mathbb R}^n})}
\end{equation}
for all $f\in L_{p_0,w}(\ensuremath{{\mathbb R}^n})$ and $w\in A_{p_0/\gamma}$, with a constant depending on $[w]_{A_{p_0}/\gamma}$.
Then for every
$\gamma< p<\infty$ (and also $p=\gamma$ if $p_0=\gamma$), every ${-n/p \le r<0}$ and every $w\in A_{p/\gamma}$, we have
that $T$ is well-defined on $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ by restriction and, moreover,
\begin{equation} \label{ex2}
\norm{Tf}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}\le c_2 \norm{f}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}},
\end{equation}
for all $f\in \ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$.
Furthermore, for $p$ as before and every $w\in A_{p/\gamma} \cap RH_{\sigma}$, if $-n/p \le r\le-n/(p\sigma)$ we have
that $T$ is well-defined on $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ by restriction and, moreover,
\begin{equation} \label{3exa3A}
\norm{Tf}{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}\le c_3 \norm{f}{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}
\end{equation}
for all $f\in \ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$.
The constants $c_2$ and $c_3$ in \eqref{ex2} and \eqref{3exa3A}
do not depend on $f$ but may depend on the weight $w$ and the involved parameters.
If $T$ satisfies instead of \eqref{ex1} the weak-type assumption
\begin{equation*
\norm{Tf}{WL_{p_0,w}(\ensuremath{{\mathbb R}^n})}\le c_1 \norm{f}{L_{p_0,w}(\ensuremath{{\mathbb R}^n})}
\end{equation*}
the estimates \eqref{ex2} and \eqref{3exa3A} are replaced by their weak-type counterparts, that is, $\norm{Tf}{W\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}$ and $\norm{Tf}{W\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}$ appear at the left-hand side of the inequalities.
\end{Corol}
\begin{proof}
The operator is well defined by restriction as a consequence of the embeddings of Corollary \ref{newcor}. To obtain the estimates \eqref{ex2} and \eqref{3exa3A} it is enough to apply Theorem \ref{TE1} to the pairs $(|f|^\gamma, |Tf|^\gamma)$ with \break $f\in\bigcup_{w\in A_{p_0/\gamma}} L_{p_0,w}(\ensuremath{{\mathbb R}^n})$ and to use the scaling property of the norms in
$\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ and $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ (see \eqref{TFE}).
To deal with the weak-type inequalities we apply Theorem \ref{TE1} to the pairs $(|f|^\gamma, t^\gamma\chi_{\{|Tf|>t\}})$.
\end{proof}
\begin{Rem}
That the operator is well-defined, for instance, on the set $\bigcup_{w\in A_{p}} \ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}$ for some $1<p<\infty$ is for practical reasons not an additional condition. Usually one applies the theorem to continuous operators which are defined on an common dense subset of $\bigcap_{w\in A_{p}} \ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}$, for example singular integrals defined on $\mathcal{D}(\ensuremath{{\mathbb R}^n})$. The singular integrals can be extended to each $\ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}$ by continuity. Such extensions coincide for functions in the intersection of two weighted Lebesgue spaces, because for each such function one can find a sequence in $\mathcal{D}(\ensuremath{{\mathbb R}^n})$ converging to it in both spaces. This implies the well-definedness of the singular integrals on $\bigcup_{1<p<\infty, w\in A_{p}} \ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}$.
\par
Concerning the spaces $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ it arises the question about the necessity of the additional weight condition, that is, whether for fixed $r$ being in some reverse H\"older class is necessary. It is known that our result could not hold for all $A_p$ weights (cf.~\cite{Sam09}).
\end{Rem}
\begin{Rem}
Chiarenza and Frasca in \cite{ChFr87} noticed that weighted $A_1$ inequalities for singular integrals yields boundedness on unweighted Morrey spaces. With a different approach Adams and Xiao \cite{AX12} revisited this method using weighted inequalities to obtain boundedness in the unweighted Morrey spaces.
\end{Rem}
\begin{Corol}\label{vvcorol}
Assume that we have a sequence of operators $\{T_j\}$ satisfying the assumptions of
Corollary \ref{TE2}, and that the constant $c_1$ in \eqref{ex1} is independent of $j$. Then for $\gamma<q, p<\infty$ and the same conditions on $w$ and $r$, we have the vector-valued estimates
\begin{equation*}
\bigg\Vert{\big(\sum_j |T_jf_j|^q\big)^{1/q}}\bigg|{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}\bigg\Vert\le c_4 \bigg\Vert{\big(\sum_j |f_j|^q\big)^{1/q}}\bigg|{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}\bigg\Vert,
\end{equation*}
and
\begin{equation*}
\bigg\Vert{\big(\sum_j |T_jf_j|^q\big)^{1/q}}\bigg|{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}\bigg\Vert
\le c_5 \bigg\Vert{\big(\sum_j |f_j|^q\big)^{1/q}}\bigg|{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}\bigg\Vert.
\end{equation*}
\end{Corol}
To prove these vector-valued inequalities in Morrey spaces it is enough to apply Theorem \ref{TE1} to pairs $\left(\big(\sum_j |f_j|^q\big)^{1/q}, \big(\sum_j |T_jf_j|^q\big)^{1/q}\right)$.
Some operators satisfy weighted inequalities for classes of weights of the type $A_q\cap RH_\sigma$. This is the case of adjoints of linear operators bounded on $\ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}$ for $p>\gamma$ and $w\in A_{p/\gamma}$. By duality the adjoint operator is bounded on $\ensuremath{L_{p,w} (\ensuremath{{\mathbb R}^n})}$ for $1<p\le \gamma$ and $\{w^{1-p}: w\in A_{p'/\gamma}\}$, which is the same as $A_{p}\cap RH_{{\gamma'}/{(\gamma'-p)}}$. Also operators bounded in a limited range of values of $p$ satisfy weighted inequalities for weights in classes of such type (see \cite{AM07} and \cite{CDL12}).
Similarly, we have sometimes the weak or strong boundedness with weights in a class of the type $\{w: w^{\nu}\in A_1\}$.
\begin{Thm} \label{TE1P}
Let $1\le p_-\le p_0\le p_+<\infty$ and let $\mathcal F$ be a collection of nonnegative measurable pairs of functions. Let $\mathcal F$ be a collection of nonnegative pairs of function in $L_{p_0,w}(\ensuremath{{\mathbb R}^n})$. Assume that for every $(f,g)\in \mathcal F$ and every $w\in A_{p_0/p_-}\cap RH_{(p_+/p_0)'}$ we have
\begin{equation} \label{1exa1PP}
\norm{g}{L_{p_0,w}(\ensuremath{{\mathbb R}^n})}\le c_1 \norm{f}{L_{p_0,w}(\ensuremath{{\mathbb R}^n})},
\end{equation}
where $c_1$ depends on the $A_{p_0}$-constant of the weight $w$.
Let $p_-< p<p_+$ (and $p=p_-$ if $p_0=p_-$) and $w\in A_{p/p_-}\cap RH_\sigma$ with $\sigma\ge (p_+/p)'$. Then for every $-n/p \le r\le -n\sigma'/p_+$ we have
\begin{equation} \label{2exa2P}
\norm{g}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}\le c_2 \norm{f}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}.
\end{equation}
Furthermore, if $-n/p \le r\le-n/(p\sigma)-n/p_+$ we have
\begin{equation} \label{3exa3P}
\norm{g}{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}\le c_3 \norm{f}{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}.
\end{equation}
The constants $c_2$ and $c_3$ in \eqref{2exa2P} and \eqref{3exa3P}
do not depend on $f$ but may depend on the weights and the involved parameters.
\end{Thm}
\begin{proof}
It is enough to prove the result for $p_-=1$, which we assume in what follows. To treat the case $p_->1$ write \eqref{1exa1PP} for the pair $(f^{p_-},g^{p_-})$ on $L_{p_0/p_-,w}(\ensuremath{{\mathbb R}^n})$, and apply the case $p_-=1$.
(a) \textit{The case $p>p_-$} (that is, $p>1$). First we observe that by limited range extrapolation (cf.~\cite[Thm.~4.9]{AM07} or \cite[Thm.~3.31]{CMP11}) \eqref{1exa1PP} holds in $L_{p,v}(\ensuremath{{\mathbb R}^n})$ with $v\in A_{p}\cap RH_{(p_+/p)'}$ for $1<p<p_+$.
Let $w\in A_{p}\cap RH_{\sigma}$. We will apply Proposition \ref{LEmbP2} with $b=p_+$ and $\gamma=1$. We choose $q\in (1,p)$ and $s>(p_+/q)'$ such that the embedding \eqref{emb11new} holds. Let $Q$ be a fixed cube. As in the proof of Theorem \ref{TE1}, we use duality, and for $h$ in $L_{\tilde p',w}(Q)$ of norm $1$ we get \eqref{p2}, that is,
\begin{equation*}
\left(\int_{\ensuremath{{\mathbb R}^n}} g^{q}h w \chi_{Q}\right)^\frac{1}{q}
\le c \left(\int_{\ensuremath{{\mathbb R}^n}} f^{q}M(h^s w^s \chi_{Q})^\frac{1}{s}\right)^\frac{1}{q},
\end{equation*}
with constant independent of $h$ and $Q$, because $M(h^s w^s \chi_{Q})^{1/s}\in A_{1}\cap RH_{(p_+/q)'}$, due to the choice of $s>(p_+/q)'$. Under the conditions required to $r$ we can apply Proposition \ref{LEmbP2} to obtain \eqref{2exa2P} and \eqref{3exa3P}.
(b) \textit{The case $p=p_-$} (that is, $p=1$).
Now we are assuming that the inequality holds for $p=1$ and $w\in A_{1}\cap RH_{p_+'}$ (that is, $w^{p_+'}\in A_1$). The proof is as in part (b) of the proof of Theorem \ref{TE1} except for the fact that we need to choose $s>p_+'$ to guarantee that $M(w^s \chi_{Q})^{p_+'/s}$ is in $A_1$.
Under the assumptions on $r$, we are in the conditions of Proposition \ref{LEmbE} with $\nu=p_+'$. Then \eqref{2exa2P} and \eqref{3exa3P} are deduced from \eqref{emb1Enew} and \eqref{emb2Enew} with $p=1$.
\end{proof}
Using this theorem and the embeddings given in Propositions \ref{LEmbE} and \ref{LEmbP2} we can obtain results analogous to those of Corollary \ref{TE2}, for the corresponding ranges of $p$ and classes of weights.
\begin{Corol} \label{TE2P}
Let $1\le p_-\le p_0\le p_+<\infty$.
Assume that $T$ is an operator acting from $\bigcup_{w\in A_{p_0/p_-}\cap RH_{(p_+/p_0)'}} L_{p_0,w}(\ensuremath{{\mathbb R}^n})$ into the space of measurable functions that satisfies
\begin{equation*}
\norm{Tf}{L_{p_0,w}(\ensuremath{{\mathbb R}^n})}\le c_1 \norm{f}{L_{p_0,w}(\ensuremath{{\mathbb R}^n})}
\end{equation*}
for all $f\in L_{p_0,w}(\ensuremath{{\mathbb R}^n})$ and $w\in A_{p_0/p_-}\cap RH_{(p_+/p_0)'}$, with a bound depending on the constant of the weight.
Then for every $p_-< p<p_+$ (and also $p=p_-$ if $p_0=p_-$) and $w\in A_{p/p_-}\cap RH_\sigma$ with $\sigma\ge (p_+/p)'$, and for every $-n/p \le r\le -n\sigma'/p_+$ we have
\begin{equation*
\norm{Tf}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}\le c_2 \norm{f}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}
\end{equation*}
for all $f\in \ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$.
Furthermore, if $-n/p \le r\le-n/(p\sigma)-n/p_+$ we have
\begin{equation*}
\norm{Tf}{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}\le c_3 \norm{f}{\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}}
\end{equation*}
for all $f\in \ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$.
\end{Corol}
Weak-type inequalities like in Corollary \ref{TE2} and vector-valued inequalities like in Corollary \ref{vvcorol} can be written also in this setting.
\begin{Rem}
In \cite{CMP11} the authors get very general results with respect to extrapolation in Banach function spaces. But in this abstract setting they do not get results either at the endpoint ($A_1$ weights) or with respect to limited range extrapolation (cf.~Theorems \ref{TE1P}). They use a kind of duality (associated spaces) which is available for Morrey spaces of type $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ (cf.~\cite{RS16}), but we completely avoid it. However, their results deliver weighted inequalities which means that for applications to operators one still has to deal with the well-definedness of the operators (which we have done via embeddings).
\end{Rem}
\section{Applications to mapping properties of operators}\label{sec5}
There is a plentiful list of operators fulfilling the requirements of one or more of the theorems of the previous section. As we mentioned in the introduction several of those operators have been proved to be bounded in Morrey spaces (unweighted or weighted) using in general the particular form of the operator or some particular pointwise bound. In most cases only the size estimate with the Morrey norm has been proved, without any discussion about the possible definition of the operator in the corresponding Morrey space. All of this is avoided with our approach. In what follows, we present a collection of applications of our general results. Vector-valued Morrey inequalities also hold in all cases, but we do not mention them. In the last subsection we give Morrey estimates with $A_\infty$ weights and $0<p<\infty$.
\subsection{Calder\'{o}n-Zygmund operators and their maximal truncations}
It is by now a classical result that Calder\'on-Zygmund operators are bounded on $L_{p,w}(\ensuremath{{\mathbb R}^n})$ for $w\in A_p$ ($1<p<\infty$), and that they are of weak-type $(1,1)$ with respect to $A_1$ weights. Here we can understand the term Calder\'on-Zygmund operator in the general sense, that is, when the operator is represented by a two-variable kernel $K(x,y)$ with appropriate size and regularity estimates. The maximal operator associated to the Calder\'on-Zygmund operators taking the supremum of the truncated integrals also satisfies similar weighted estimates. For a proof the reader can consult \cite[Chapter 9]{G09}. Then we can apply Corollary \ref{TE2}.
The regularity assumptions on $K(x,y)$ can be weakened and stated in some integral form called $L^r$-H\"ormander condition. In the convolution case it appears implicitely in \cite{KuW79}, and in a more general setting in \cite{RRT86}, for instance. In that case the weighted inequalities hold for $L_{p,w}(\ensuremath{{\mathbb R}^n})$ for $p\ge\gamma$ and $w\in A_{p/\gamma}$, where $\gamma=r'$. Then Corollary \ref{TE2} applies. If the estimates are symmetric in the variables $x$ and $y$ of the kernel, duality can be used for the weighted estimates and the results of Corollary \ref{TE2P} apply.
\subsection{Multipliers}
Kurtz and Wheeden studied in \cite{KuW79} weighted inequalities for classes of multipliers defined as follows (see also \cite{StTo89}). Let $\widehat{Tf}(\xi)=m(\xi)\hat f(\xi)$. We say that $m\in M(s,l)$ if
\begin{equation*}
\sup_{R>0} \left(R^{s|\alpha|-n}\int_{R<|\xi|<2R}|D^\alpha m(\xi)|^s\right)^{1/s}<+\infty \text{ for all }|\alpha|\le l.
\end{equation*}
The result of Kurtz and Wheeden says that for $1<s\le 2$ and $n/s<l\le n$, $T_m$ is bounded on $L_{p,w}(\ensuremath{{\mathbb R}^n})$ if $n/l<p<\infty$ and $w\in A_{pl/n}$; its dual result for $1<p<(n/l)'$; and of weak-type $(1,1)$ with respect to weights $w$ such that $w^{n/l}\in A_1$. Although they do not make it explicit, interpolation gives that $T_m$ is bounded on $L_{p,w}(\ensuremath{{\mathbb R}^n})$ for $w^{lp/n}\in A_1$, when $1<p\le n/l$.
We can apply Corollaries \ref{TE2} and \ref{TE2P}, depending on the values of $p$.
A particularly interesting case is $M(2,l)$, which corresponds to the Mihlin-H\"ormander multipliers. For $l=n$ the full range of results holds applying Corollary \ref{TE2} with $\gamma=1$. Let us particularize some of the results valid for any $l>n/2$:
\begin{itemize}
\item for $p\ge 2$ and $w\in A_{p/2}$, we have boundedness on $ \ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ without restriction on $r$; for $w\in A_{p/2}\cap RH_\sigma$ we have boundedness on $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ for $-n/p\le r\le -n/(p\sigma)$ (Corollary \ref{TE2P} with $p_-=2$);
\item for $1<p<2$ and $w\in A_{p}\cap RH_\sigma$ with $\sigma'<2/p$, we have boundedness on $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ when $r\in [-n/p,-n\sigma'/2]$ and on $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ for $--n/p\le r\le -n/(p\sigma)-n/2$ (Corollary \ref{TE2P});
\item for $p\ge 2$ and $w(x)=|x|^\beta$ (or $(1+|x|)^\beta$) with $-n<\beta<n(p/2-1)$, we have boundedness on $ \ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ without restriction on $r$ (Corollary \ref{TE2} with $\gamma=2$);
\item for $p\ge 2$ and $w(x)=|x|^\beta$ (or $(1+|x|)^\beta$) with $-n<\beta<0$ we have boundedness on $ \ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ for $-n/p\le r\le \beta/{p}$ and if $0\le\beta< n(p/2-1)$, we have boundedness on $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ for $-n/p \le r<0$ (Corollary \ref{TE2P} with $p_-=2$ together with Remark \ref{Ma3});
\item for $1<p\le 2$ and $w(x)=|x|^\beta$ (or $(1+|x|)^\beta$) with $-np/2<\beta\le 0$, the boundedness on $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ holds for $-\frac np\le r< -\frac {n^2}{n+\beta}\left(\frac 1p-\frac 12\right)$. This is valid for $p=1$ using the weak Morrey estimate. In the case of $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ the range is $-\frac np\le r<-n\left(\frac 1p-\frac 12\right)+\frac \beta p$. We apply Corollary \ref{TE2P} with $p_-=p$ and $(p_+/p)'=2/p$, and Remark \ref{Ma3}.
\item for $1<p\le 2$ and $w(x)=|x|^\beta$ (or $(1+|x|)^\beta$) with $0\le \beta<n(p-1)$, we have boundedness on $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ and on $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ when $r\in [-n/p,-n/2)$. Here we use Corollary \ref{TE2P} with $p_+=2$ and the fact that these weights are in $RH_\sigma$ for all $\sigma>1$.
\end{itemize}
The origin of Morrey spaces is in the study of the smoothness properties of solutions of PDEs. For this reason results on the multipliers $M(s,l)$ related to their smoothness are interesting (cf. \cite{Tr12, T-HS, R}).
Another interesting case of multipliers is that of Marcinkiewicz multipliers in one dimension. They are bounded on $L_{p,w}(\mathbb{R})$ for $w\in A_p$ (see \cite{Ku80}) and Corollary \ref{TE2} gives all the Morrey boundedness results for $1<p<\infty$.
\subsection{Rough operators}
The rough singular integrals are defined in general by
\begin{equation*}
T_{\Omega, h}f(x)=\text{p.v.}\int_{\ensuremath{{\mathbb R}^n}} \frac{\Omega(y')h(|y|)}{|y|^n}f(x-y)dy
\end{equation*}
with $\Omega\in L^1(S^{n-1})$ ($y'=y|y|^{-1}$) and integral zero, and $h$ defined on $(0,\infty)$. The (weighted) boundedness of the operator is proved using additional assumptions on $\Omega$ and $h$. For instance, if both $\Omega$ and $h$ are in $L^\infty$, it was proved in \cite{DuRu86} that $T_{\Omega, h}$ is bounded on $L_{p,w}(\ensuremath{{\mathbb R}^n})$ for $w\in A_p$ ($1<p<\infty$), so that the full range is obtained for Morrey spaces from Theorem \ref{TE1}.
If for fixed $q>1$ we know that $\Omega\in L^q(S^{n-1})$ and $h\in L^q((0,\infty);dt/t)$ then the boundedness on $L_{p,w}(\ensuremath{{\mathbb R}^n})$ holds for $w\in A_{p/q'}$ ($q'\le p<\infty$) and also for some classes of weights obtained by duality for the range $1<p\le q$. These classes can further be extended by interpolation. Results of this type are in \cite{Wat90} and \cite{D93}. The weighted inequalities are very much like those for the multipliers of the previous section, hence similar Morrey estimates are obtained. Due to the openness properties of $A_p$ weights, D.~Watson noticed that for $\Omega\in \bigcap_{q<\infty} L^q(S^{n-1})$ and $h\in \bigcap_{q<\infty}L^q((0,\infty);dt/t)$ the full range $L_{p,w}(\ensuremath{{\mathbb R}^n})$ for $w\in A_p$ and $1<p<\infty$ is also obtained. Hence the full range of Morrey estimates is deduced. Notice that in this case the kernel of the operator $T_{\Omega, h}$ need not satisfy the size estimate \eqref{i5}.
Similar weighted estimates are known for the maximal operator defined from the truncated integrals of the rough operators.
\subsection{Square functions}
There exist several types of square functions for which our theorems apply.
The classical Littlewood-Paley operators in one dimension associated to the dyadic intervals are bounded on $L_{p,w}(\mathbb{R})$ for $w\in A_p$ (see \cite{Ku80}). A similar result holds in all dimensions in we consider the square functions of discrete or continuous type built using dilations of a Schwartz function with zero integral (see \cite[Prop.~1.9]{Ry01} and the comments following it).
The Lusin area integral, the $g_\lambda$ functions, Stein's $n$-dimensional extension of the Marcinkiewicz integral and the intrinsic square function of M. Wilson are other examples of square functions for which the weighted inequalities on $L_{p,w}(\ensuremath{{\mathbb R}^n})$ for $w\in A_p$ ($1<p<\infty$) are known to hold (see \cite{ToW90} and \cite[Thm.~7.2 and notes of Chapter 7]{Wil08}). Of course, all of them are covered by our Corollary \ref{TE2}.
Rough variants have been considered also for square functions with the introduction of some $\Omega\in L^q(S^{n-1})$ as for the singular integrals mentioned above. The class of weights for which boundedness holds depends on $q$ (see \cite{DFP99} and \cite{DuS02}).
On the other hand, we can mention Rubio de Francia's results of Littlewood-Paley type. The square function built using comparable cubes in $\ensuremath{{\mathbb R}^n}$ is bounded on $L_{p,w}(\ensuremath{{\mathbb R}^n})$ for $w\in A_{p/2}$ and $2\le p<\infty$ (see \cite{Ru83}) and in this case we can apply Corollary \ref{TE2} with $\gamma=2$ to obtain the corresponding Morrey estimates. For the square function built using arbitrary disjoint intervals in $\mathbb{R}$ we can apply Corollary \ref{TE2} to the result of \cite{Ru85} in which the boundedness of the operator is obtained for $L_{p,w}(\ensuremath{{\mathbb R}^n})$ with $w\in A_{p/2}$ and $p>2$ (the endpoint $p=2$ remains open).
\subsection{Commutators}
There are several results on weighted inequalities for commutators to which we can apply our theorems to obtain the corresponding Morrey estimates. Let us mention in particular the remarkable result of \'Alvarez et al.~in \cite{ABKP93}, which allows to obtain many weighted inequalities for commutators with BMO functions. A good list of applications can be seen in their paper. For each one of them we obtain the corresponding result in the Morrey spaces.
\subsection{Other operators}
There are many more results in the literature concerning weighted inequalities with $A_p$ type weights. We briefly mention here some other remarkable operators for which our theorems apply directly. Although in some cases particular proofs of their Morrey estimates do exist, not all of them seem to have been considered before.
Oscillatory singular integrals with kernels of the form $e^{iP(x,y)}K(x,y)$ where $P(x,y)$ is a polynomial and $K(x,y)$ is a Calder\'on-Zygmund standard kernel or a kernel of rough type have been studied and weighted results appear for instance in \cite{LDY07}. A further variant are the so-called multilinear oscillatory singular integrals in which the kernel is modified by inserting a factor $|x-y|^{-m}R_{m+1}(A;x,y)$ where the second term is the $(m+1)$-th order remainder of the Taylor series of a function $A$ expanded in $x$ about $y$. Some weighted inequalities are in \cite{Wu04}. Other types of oscillating kernels are of the form $K_{a,b+iy}(t)=\exp(i|t|^a)(1+|t|)^{-(b+iy)}$ for which weighted inequalities are obtained in \cite{ChKS83}, and strongly singular convolution operators corresponding to multipliers of the form $\theta(\xi)e^{i|\xi|^b}|\xi|^{-a}$, where $\theta$ is a smooth radial cut-off function (vanishing in a neighborhood of the origin and which is identically $1$ outside a compact set), studied in \cite{Ch84}.
Bochner-Riesz operators at the critical index satisfy $A_p$-weighted estimates for $1<p<\infty$ and weak-type $(1,1)$ estimates with $A_1$ weights (\cite{ShSu92} and \cite{Var96}). The boundedness on the Morrey spaces are obtained from Corollary \ref{TE2}. Below the critical index only partial results can be obtained, but still there are some subclasses of $A_p$ weights (see \cite{CDL12}, for instance) for which the corresponding theorems in the previous section can be applied. Some weighted inequalities also hold for the maximal Bochner-Riesz operator.
Weighted inequalities for pseudodifferential operators are for instance in \cite{Ya85} and \cite{Sat05}, while for Fourier integral operators they can be found in \cite{DSS14}. In both cases, Morrey space estimates can be deduced from our results in Section \ref{sec4}.
\subsection{Morrey norm inequalities with $A_\infty$ weights}
Let $0<p<\infty$ and $w\in A_\infty$. Then the following inequalities hold whenever the left-hand side is finite:
\begin{align}
\label{five1}
\norm{Tf}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}&\le c_1 \norm{Mf}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}},\\
\label{five2}
\norm{Mf}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}&\le c_2 \norm{M^\sharp f}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}},\\
\label{five3}
\norm{I_\alpha f}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}}&\le c_3 \norm{M_\alpha f}{\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}},
\end{align}
where $T$ is a Calder\'on-Zygmund operator in \eqref{five1}, $M$ is the Hardy-Littlewood maximal operator in \eqref{five1} and \eqref{five2}, $M^\sharp$ is the sharp maximal function (see \cite[p. 117]{D01}) in \eqref{five2}, and $I_\alpha$ and $M_\alpha$ are the fractional integral and the fractional maximal operator of order $\alpha\in (0,n)$ (see \cite[p. 88-89]{D01}) in \eqref{five3}. Similar inequalities hold with $\ensuremath{{ L}^{r}_{p}(\lambda,w,\ensuremath{{\mathbb R}^n})}$ instead of $\ensuremath{{ L}^{r}_{p}(w,\ensuremath{{\mathbb R}^n})}$ if $w$ is assumed to be in $RH_\sigma$ and $-n/p \le r\le-n/(p\sigma)$.
The inequalities are derived from their $L_{p,w}$ counterparts using Corollary \ref{TEAinf}. A proof of the needed $L_{p,w}$ estimates using extrapolation can be found in \cite{CMP04}, where references concerning the original inequalities (obtained by means of good-$\lambda$ inequalities) are given. In a similar way, other $L_{p,w}$ inequalities appearing in \cite{CMP04} can also be written with Morrey norms. In that paper weak-type inequalities and vector-valued inequalities are also given and they can also be transferred to the Morrey setting, but we do not go into this in detail. Weighted inequalities with Morrey norms involving $M^\sharp$ with $A_\infty$ weights may be found also in \cite[Corollary 2]{NkSa17}.
| a0cc6337234a16dbd70dfe5ee472027beeb1158c | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
The phenomenon of Brillouin scattering, whereby an acoustic field mediates
transitions between optical modes in an optical
waveguide~\cite{Brillouin1922,Boyd2003,Agrawal2012}, has been the focus of
intense research interest in recent years, driven by a suite of key
applications in nanophotonics ranging from microwave photonic filters to
novel light sources~\cite{Eggleton2013}.
Brillouin scattering can be categorised according to the direction of light
scattered by the phonon field: \emph{Backward} Brillouin scattering, by broad
consensus called Stimulated Brillouin Scattering (SBS) because of the
self-reinforcing nature of the phenomenon, arises commonly in optical fibers
and results from the interaction of optical guided waves with a longitudinal
acoustic wave with a wave length of only few hundred nanometers.
In contrast, \emph{forward} Brillouin scattering (FBS) arises due to the
interaction of light with long-wavelength acoustic modes.
Depending on whether the interaction occurs within the same optical mode
(intra-mode FBS) or between different optical modes (inter-mode FBS), the
relevant acoustic field either forms a transverse wave with wave length on the
order of centimeters or complex a torsional or flexural mode with intermediate
wave length (see Fig.~\ref{fig:types_of_sbs}).
\begin{figure}
\includegraphics[width=\columnwidth]{types-of-sbs.eps}
\caption{
Sketch of the three main variants of guided Brillouin scattering:
backward (stimulated) scattering (left), inter-mode forward scattering
(center) and intra-mode forward scattering (right).
For each case the figure schematically shows the acoustically mediated
transistions within the optical dispersion relation as well as typical
examples for the corresponding acoustic modes.
Note that only for intra-mode forward scattering (right panel) all Stokes
order are connected by the same acoustic mode (\mbox{i.e.\ } same symmetry, wave
number and frequency).
This motivates why intra-mode FBS is particularly prone to cascading.
}
\label{fig:types_of_sbs}
\end{figure}
Although observed experimentally as early as 1985~\cite{Shelby1985} in
conventional step index fibres, FBS appears most naturally in waveguides that
possess complex transverse structure;
FBS has been observed in photonic crystal fibres (PCFs) and
tapers~\cite{Elser2006,Dainese2006,Beugnot2007,Stiller2011,Zhong2015,Renninger2016,Kang2008},
while the parallel development of on-chip integrated optical
circuits has led to experimental observation of FBS in semiconductor
nanowires~\cite{Rakich2013,VanLaer2015}, and theoretical consideration
of slot waveguides~\cite{VanLaer2014} and hybrid photonic-phononic
waveguides~\cite{Chen2014}.
The observation that FBS is much easier to generate in integrated silicon
circuits than its backward counterpart~\cite{VanLaer2015a,VanLaer2015,Rakich2016} means
that FBS is critically important for applications that seek to harness the
interaction of sound and light on the nanoscale.
The established theory for Brillouin processes, in which a single pair of
optical fields are related by an acoustic mode, has developed from the study of
backward SBS in homogeneous materials and optical fibers~\cite{Boyd2003,Bloembergen1965,Tang1966}.
With some important modifications~\cite{Rakich2012,Wolff2015,Rakich2016}, for backward
SBS this formalism can be directly applied to the nanostructures that are of
recent interest.
However the existing theory is not well-suited to the study of FBS.
The main reason for this is the phenomenon of cascading, whereby the
generation of a single, or first-order, Stokes line results in multiple
follow-on orders.
Under certain conditions cascading can occur in backwards SBS, however the
process usually only arises in the presence of reflections or
resonances~\cite{Buettner2014a,Buettner2014b}.
By contrast, cascading arises in FBS because of the
less-stringent phase-matching requirements.
In principle the existence of large numbers of Stokes lines in FBS could be
addressed by an \emph{ad-hoc} expansion of the existing formalism, however
this approach restricts the optical spectrum to a comb of Stokes lines,
introduces artifacts if the expansion of Stokes orders is truncated, and becomes
rather cumbersome in the presence of optical dispersion or in combination with
other optical nonlinearities such as Raman scattering or the instantaneous
Kerr effect.
This approach also requires severe restrictions on the optical fields, which
must vary slowly on the time and length scales of the {\em acoustic} field,
and on the opto-acoustic response, which is assumed to be perfectly Lorentzian.
The range of experimental situations that can be treated with existing
approaches is therefore highly limited.
With the recent intense interest in the generation of FBS in nanostructures, a
general theory that fully describes forward Brillouin scattering is not only
essential to gain physical insight and to guide experiments, but will form
part of the ongoing program aiming to adapt Brillouin processes for use in
nanophotonics and to unify the theory with that of traditional
optomechanics~\cite{VanLaer2016}.
\begin{figure}
\includegraphics[width=\columnwidth]{Figure_schematic.ps}
\caption{
Schematic of a prototypical setup for cascaded forward Brillouin scattering:
In a waveguide (\mbox{e.g.\ } a nanowire, rib waveguide or an optical fibre) a
mechanical vibration with near-zero wave number and group velocity is
excited by an incident optical pump and a weak Stokes seed (\mbox{e.g.\ } due to
intensity noise in the pump laser).
This results in the exiting light being predominantly phase-modulated with
pronounced higher-order (anti-)Stokes side bands.
}
\label{fig:schematic}
\end{figure}
Here we present a new and rigorous formulation of FBS that includes cascading
to all orders.
To this end, we allow optical envelopes to cover any number of
\mbox{(anti-)Stokes}
lines, which reduces fully cascaded FBS to an interaction between one optical
and one mechanical envelope.
Figure~\ref{fig:schematic} shows a schematic of a prototypical system:
a modulated pump is injected into a waveguide that supports nearly transversal
mechanical eigenmodes.
The pump modulation can be either a dedicated Stokes seed as commonly used in
silicon photonics experiments or the inevitable intensity noise of the pump
laser.
For vanishing dispersion, our model has a full general analytical solution in
the sense that the optical field at the waveguide's output is given as an
explicit integral of the arbitrary input envelope.
A key finding is that without dispersion the optically driven vibrations only
cause an additional optical phase modulation.
This is in agreement with the recent theoretical observation that FBS in the
absence of dispersion has negative single-pass gain~\cite{Rakich2016}.
We then show that weak dispersion leads to two different regimes: in one
regime dispersion washes out intensity fluctuations of an incident pump laser
while generating additional phase noise.
In the other regime, it leads to self-amplifying dynamics along the waveguide.
We also demonstrate the great similarity of FBS with Raman scattering as it is
encountered, for example, in the context of soliton dynamics.
Finally, we find an unexpected striking similarity between dispersionless FBS
and discrete diffraction in waveguide arrays.
This makes optically driven FBS an interesting candidate for few-photon
experiments.
Our single-envelope formulation of cascaded FBS is ideally suited for this
type of problem, as it conveniently describes superpositions of different
Stokes orders with a single operator.
Our formalism allows arbitrary optical spectra spanning several Stokes shifts,
including soliton-like pulses and it facilitates the inclusion of
optical dispersion and other optical nonlinearities, which we demonstrate for
Raman-scattering and the Kerr effect.
Furthermore, our formalism illuminates the placement of FBS within the family
of optical nonlinearities.
Forward Brillouin scattering bears resemblance with several other scattering
phenomena, and exhibits different characteristics that depend on the
parameters of the experiment.
This chimeric quality is reflected in the different names FBS has been given,
each of which highlights specific aspects of the phenomenon.
The term ``guided acoustic wave Brillouin scattering''
(GAWBS)~\cite{Shelby1985,Elser2006,Beugnot2007,Stiller2011,Zhong2015}
emphasizes the confined nature of the acoustic wave and the directionality of
FBS in silica fibres.
In contrast, ``Raman-like light scattering''~\cite{Dainese2006,Kang2009}
highlights that (unlike in backward SBS) the Stokes shift is nearly
independent of the optical wavelength and that FBS easily leads, as in the
Raman process, to cascading.
Finally, ``forward stimulated Brillouin scattering''
(FSBS)~\cite{Rakich2012,Chen2014} draws the strong formal connection with
backward (stimulated) Brillouin scattering.
A result of the formalism presented here is that these aspects are unified:
specific observations can be explained with the fact that fibres and
integrated waveguides can be quite different in optical loss, dispersion,
power handing, length, and opto-mechanical interaction strength.
For this reason we have adopted the very general term ``forward Brillouin
scattering'' (FBS) and devote Sec.\ref{sec:measurement} in our discussion to
the different regimes referred to by the above mentioned terms and the
corresponding measurement techniques.
This paper is structured as follows:
in Sec.~\ref{sec:approx}, we state the approximations and assumptions that are
required for describing an optical pump and all (anti-)Stokes orders in an
optical band with a single envelope function.
We also comment on the differences between a classical and a quantum-mechanical
description. In Sec.~\ref{sec:hamiltonian}, we then go on to first derive the
quantum-mechanical Hamiltonian for Brillouin scattering in a single-envelope
picture and state both quantum-mechanical and classical equations of motion,
which we then specialize for intra-mode forward scattering.
We demonstrate how the established multi-envelope theory is derived from this.
Next, we solve the dispersionless case analytically for any optical input in
Sec.~\ref{sec:dispersionless}.
Based on this result, the case of weak optical dispersion and shallow optical
intensity variations is solved perturbatively in Sec.~\ref{sec:dispersive}.
Sec.~\ref{sec:discussion} finally concludes our paper with an in-depth
discussion and a summary of the main findings.
\section{Approximations, assumptions and theoretical approach for cascaded
Brillouin scattering}
\label{sec:approx}
The established coupled mode
theory~\cite{Boyd2003,Bloembergen1965,Tang1966,Rakich2012,Wolff2015,Sipe2016}
of SBS is centered around two approximations:
a restrictive slowly varying envelope approximation (SVEA) for the wave
propagation and a restrictive rotating wave approximation (RWA) for the
opto-acoustic interaction.
The approximations are restrictive in the sense that all but the most slowly-varying
interaction terms are neglected and that the acoustic evelopes as well
as the envelopes describing individual Stokes orders (usually only the pump
and the first Stokes band) only vary on a time scale larger than the acoustic
period.
In intra-mode FBS settings the acoustic field is assumed to be the same
for the interaction between each pair of adjacent optical lines.
This approach has two main disadvantages for the study of problems where
cascading effects are relevant.
Firstly, each Stokes order requires an additional differential equation, which
can become cumbersome if other nonlinear effects such as Kerr-induced
four-wave mixing and optical dispersion are to be included.
This is mostly a technical problem.
Secondly, the decomposition into individual Stokes orders requires a very
restrictive implementation of the SVEA and the RWA in that the optical
envelopes have to vary slowly on the order of the \emph{acoustic} time and
length scales.
In the same vein, the opto-acoustic interaction term is stripped of all
contributions that deviate from a perfect Lorentzian response.
This restriction of the established theory precludes for example the
description of opto-acoustic dynamics of light pulses that vary on the
acoustic time scale.
We therefore seek a less restrictive description of Brillouin scattering.
An optical envelope can cover several THz and still be considered slowly
varying, thus one single optical envelope is sufficient to describe all Stokes
orders at once.
However, this is in conflict with the RWA as it is commonly applied to the
opto-acoustic interaction term: in the established literature, the
interaction term is averaged over a time window much longer than the acoustic
period.
Instead, we have to relax the RWA to allow for variations on the same time scale
as the optical envelope, \mbox{i.e.\ } several multiples of the acoustic period.
As a result, such a model has to allow for far off-resonant excitations of the
mechanical system and for example include static deformations.
Consequently, the acoustic amplitude cannot be described as a slowly varying
envelope, but instead has to explicitly contain the time dependence of the
mechanical oscillations.
We note that the single-envelope picture outlined above is quite common in
short-pulse Raman physics~\cite{Boyd2003,Agrawal2012,Gordon1986}.
Brillouin scattering can be formulated as a problem of classical
physics \mbox{e.g.\ } based on a Hamiltonian, a Lagrangian~\cite{Beugnot2015} or a
coupled mode~\cite{Wolff2015} picture, where only the latter two allow for an
elegant incorporation of loss.
Alternatively, it can be described via a quantum-mechanical
Hamiltonian~\cite{Sipe2016}.
As it turns out, the differences between the classical and quantum-mechanical
Hamiltonians are marginal and we choose to derive the loss-less
quantum-mechanical Hamiltonian, which will be the basis for future studies on
quantum-optics using FBS.
However, in this work we ultimately aim for a classical discussion of cascaded
FSB and therefore derive classical equations of motion from their
quantum-mechanical counterparts, where we also simply add the loss terms that
are well known from the literature.
\section{Hamiltonian and equations of motion}
\label{sec:hamiltonian}
We follow the example of Ref.~\cite{Sipe2016} and start our investigation with
the opto-acoustic Hamiltonian.
From this, we derive equations of motion for the quantum mechanical amplitudes
of the optical and acoustic fields.
Originating from a Hamiltonian representation without external baths, this model
is lossless.
We then move on to the classical realm, where we also introduce acoustic loss.
This is adequate to describe most experiments.
The Hamiltonian for opto-acoustic interactions in a longitudinally-invariant
waveguide (oriented along $z$) can be expressed in a modal expansion of the
form~\cite{Sipe2016}
\begin{align}
\mathcal{H} = &
\sum_\alpha \int_{-\infty}^\infty \mathrm{d} k \ \hbar \omega_{\alpha k}
\hat a^\dagger_{\alpha k} \hat a_{\alpha k}
\label{eqn:starting_point}
\\
\nonumber
& \quad +
\sum_\beta \int_{-\infty}^\infty \mathrm{d} q \ \hbar \Omega_{\beta q}
\hat b^\dagger_{\beta q} \hat b_{\beta q}
+
\mathcal{V},
\end{align}
where $\mathcal{V}$ is the opto-acoustic interaction term,
$\hat a_{\alpha k}$ is the annihilation operator for a photon in the
$\alpha$-th optical mode with wave number $k$ and energy
$\hbar \omega_{\alpha k}$, and $\hat b_{\beta q}$
likewise for a phonon in the $\beta$-th acoustic mode with wave number $q$ and
energy $\hbar \Omega_{\beta q}$.
The ladder operators $\hat a_{\alpha k}$ and $\hat b_{\beta q}$ are not to be
confused with the classical envelope functions $a_\alpha(z,t)$ and $b_\beta(z,t)$
that will be introduced towards the end of this section.
We furthermore introduce the electric induction field operator and the
mechanical strain operator
\begin{align}
\nonumber
\hat{\vec D} = & \sum_\alpha \frac{1}{2\sqrt{\pi}} \int_{-\infty}^\infty \mathrm{d} k \
\sqrt{\hbar \omega_{\alpha k}} \hat a_{\alpha k} \bar{\vec d}_{\alpha k}
\exp(\mathrm{i} k z) + \text{h.c.},
\\
\nonumber
\hat S^{ij} = & \sum_\beta \frac{1}{2\sqrt{\pi}} \int_{-\infty}^\infty \mathrm{d} q \
\sqrt{\hbar \Omega_{\beta q}} \hat b_{\beta q} \bar s^{ij}_{\beta q}
\exp(\mathrm{i} q z) + \text{h.c.},
\end{align}
with ``$\text{h.c.}$'' representing the Hermitian conjugate.
Here, $\bar{\vec d}_{\alpha k}$ is the electric induction field of the
$\alpha$-th classical optical eigenmode with wave number $k$ and has been
normalized according to
\begin{align}
\bar{\vec d}_{\alpha k} = & \left[ \int \mathrm{d}^2 r \
\frac{\vec d_{\alpha k}^\ast \cdot \vec d_{\alpha k}}{\varepsilon_\text{r}(x,y) \varepsilon_0}
\right]^{-\frac{1}{2}} \vec d_{\alpha k}
= \sqrt{\frac{2 }{\mathcal{E}_{\alpha k}}} \vec d_{\alpha k},
\end{align}
where $\vec d_{\alpha k}$ is an arbitrary solution to the classical optical
wave equation and $\mathcal{E}_{\alpha k}$ its classical energy per unit length of
the waveguide~\cite{Wolff2015}.
Analogously, the modal strain field $\bar s^{ij}_{\beta q}$ associated with
wave number $q$ and mode index $\beta$ is normalized to its classical energy
$\mathcal{E}_{\beta q}$ per unit length of waveguide~\cite{Wolff2015}:
\begin{align}
\bar{s}^{ij}_{\beta q} = & \sqrt{\frac{2}{\mathcal{E}_{\beta q}}}
s^{ij}_{\beta q}.
\end{align}
Note that in the original Hamiltonian formulation~\cite{Sipe2016}, the optical
and acoustic modal fields were each normalized to carry a single quantum per
unit waveguide length, as opposed to the arbitrary normalization we allow here.
The two pictures can be connected by choosing
\mbox{$\mathcal{E}_{\alpha k} =\hbar \omega_{\alpha k}$} and
\mbox{$\mathcal{E}_{\beta q} =\hbar \Omega_{\beta k}$}.
To obtain a representation suitable for describing pulse propagation,
we transform to the spatial domain by introducing envelope operators
$\hat \psi_{\alpha}(z,t)$, $\hat \phi_{\beta}(z,t)$
\begin{align}
\hat \psi_{\alpha}(z,t) = & \frac{1}{\sqrt{2 \pi}} \int_{-\infty}^\infty
\mathrm{d} k \ \hat a_{\alpha k}(t) \exp(\mathrm{i} k z),
\label{eqn:psi_op_def}
\\
\hat \phi_{\beta}(z,t) = & \frac{1}{\sqrt{2 \pi}} \int_{-\infty}^\infty
\mathrm{d} q \ \hat b_{\beta q}(t) \exp(\mathrm{i} q z),
\label{eqn:phi_op_def}
\end{align}
and propagation operators $\mathcal{L}_\alpha$ and $\mathcal{L}_\beta$:
\begin{align}
\mathcal{L}_\alpha = & \hbar \omega_0 +
\hbar \sum_{n=1}^\infty \frac{(-\mathrm{i})^n}{n!} v_\alpha^{(n)} (\partial_z - \mathrm{i} k_\alpha)^n,
\label{eqn:Lalpha_def}
\\
\mathcal{L}_\beta = & \hbar \Omega_{\beta q_0} +
\hbar \sum_{n=1}^\infty \frac{(-\mathrm{i})^n}{n!} v_\beta^{(n)} (\partial_z - \mathrm{i} q_\beta)^n,
\label{eqn:Lbeta_def}
\end{align}
where we explicitly assume that all optical carriers have the same frequency
$\omega_0$, because in basically every experiment the optical frequency is
defined by the vacuum wave length of the used laser; the different optical wave
numbers and the wave numbers and frequencies of the excited acoustic modes
follow from this frequency via the optical and acoustic dispersion relations.
The first ($n=1$) elements in each set of coefficients
\begin{align}
v_\alpha^{(n)} = & \frac{\partial^n \omega_{\alpha k_\alpha}}{\partial k^n} \Big|_{k_\alpha},
&
v_\beta^{(n)} = & \frac{\partial^n \Omega_{\beta q_\beta}}{\partial q^n} \Big|_{q_\beta}
\label{eqn:group_velocity_def}
\end{align}
are the optical and acoustic acoustic group velocities, respectively.
Their products with the modal energies $\mathcal{E}$ per unit length of waveguide
are the modal powers through the transversal waveguide plane:
\begin{align}
\nonumber
\mathcal{P}_\alpha = & v^{(1)}_\alpha \mathcal{E}_\alpha,
&
\mathcal{P}_\beta = & v^{(1)}_\beta \mathcal{E}_\beta.
\end{align}
The operator $\mathcal{L}_\alpha$ emerges from Eq.\eqref{eqn:starting_point}
by expressing the optical dispersion relation $\omega_{\alpha k}$ as a Taylor
series in $k$ and by substituting
\begin{align*}
k \rightarrow -\mathrm{i} \partial_z,
\end{align*}
in the course of the Fourier transformation to the real-space representation.
The operator $\mathcal{L}_\beta$ is defined likewise for the acoustic dispersion
relation.
This description is based on the convergence of the $k$-space Taylor series and
is therefore valid inside a disk in complex $k$-space where the radius is
determined by the nearest complex point of degeneracy (found at band crossings
and near anti-crossing).
The operators $\hat \phi_\alpha$ and $\hat \psi_\beta$ are fast operators in the
sense that they oscillate according to $\exp(\mathrm{i} k_\alpha z)$ and
$\exp(\mathrm{i} q_\beta z)$, respectively.
They have dimensions of \mbox{$1/\sqrt{\text{length}}$} and fulfil the
commutation relations
\begin{align}
[\hat \psi_\alpha(z), \hat \psi^\dagger_{\alpha'}(z')] = &
\delta_{\alpha \alpha'} \delta(z-z'),
\label{eqn:commutator_psi}
\\
[\hat \phi_\beta(z), \hat \phi^\dagger_{\beta'}(z')] = &
\delta_{\beta \beta'} \delta(z-z').
\label{eqn:commutator_phi}
\end{align}
This differs from the formally very similar operators introduced
for each individual Stokes order by Sipe and Steel~\cite{Sipe2016}.
The difference is that the integrals in
Eqs.~(\ref{eqn:psi_op_def},\ref{eqn:phi_op_def}) formally cover the complete
$k$-range instead of small $k$-space intervals of width
$\Delta k \approx \Omega / v_\alpha$ as in Ref.~\cite{Sipe2016}.
Our first approximation is an SVEA: the field distributions are approximated
as modulations of the carrier eigenmodes
\begin{align}
\hat{\vec D} \approx & \sum_\alpha
\sqrt{\frac{\hbar \omega_0}{2}}
\bar{\vec d}_{\alpha k_\alpha} \hat\psi_{\alpha} + \text{h.c.},
\label{eqn:psi_svea}
\\
\hat S^{ij} \approx & \sum_\beta \sqrt{\frac{\hbar \Omega_{\beta q_\beta}}{2}}
{\bar s}^{ij}_{\beta q_\beta} \hat\phi_{\beta} + \text{h.c.},
\label{eqn:phi_svea}
\end{align}
where we explicitly allow for envelopes that vary on a time scale shorter than
the acoustic period (what we call a ``relaxed'' SVEA).
Eqs.~(\ref{eqn:psi_svea},\ref{eqn:phi_svea}) are justified if neither the
optical dispersion relation nor the optical mode patterns vary appreciably
over the wave number range of multiple Stokes shifts.
This is usually the case away from special points in the dispersion relation
such as band edges.
\subsection{Hamiltonian}
We now derive the interaction term that is appropriate for our choice of
envelopes.
Here, we focus on the electrostrictive interaction term in order to keep
our equations short.
Discussions of how to derive radiation pressure within a coupled-mode picture
of SBS can be found in the literature~\cite{Rakich2012,Wolff2015,Sipe2016}.
The electrostrictive interaction term is~\cite{Sipe2016}:
\begin{align}
\mathcal{V} = & \frac{1}{2 \varepsilon_0} \int \mathrm{d}^3 r \
\sum_{ijkl} \hat D_i \hat D_j p_{ijkl} \hat S^{kl}
\\
\nonumber
= & \sum_{\beta, \alpha, \alpha'}
\frac{\hbar \omega_0 \sqrt{\hbar \Omega_{\beta q_\beta}}}{2 \varepsilon_0 \sqrt{8} }
\int \mathrm{d}^3 r \ \sum_{ijkl} \Big\{
\\
\nonumber
& \
[\bar d^i_{\alpha k_\alpha} \hat \psi_\alpha +
(\bar d^i_{\alpha k_\alpha})^\ast \hat \psi^\dagger_\alpha]
[\bar d^j_{\alpha' k_{\alpha'}} \hat \psi_{\alpha'} +
(\bar d^j_{\alpha' k_{\alpha'}})^\ast \hat \psi^\dagger_{\alpha'}]
\\
& \ \times p_{ijkl}
[\bar s^{kl}_{\beta q_\beta} \hat \phi_{\beta} +
(\bar s^{kl}_{\beta q_\beta})^\ast \hat \phi^\dagger_{\beta}] \Big\},
\end{align}
where $p_{ijkl}(\vec r)$ is the photoelastic tensor distribution.
We now employ an optical rotating wave approximation, \mbox{i.e.\ } we neglect the terms
proportional to $\hat \psi_\alpha^2$ and $(\hat \psi^\dagger_\alpha)^2$,
because these terms oscillate at twice the optical frequency and cannot excite
the acoustic system.
We do, however, retain terms describing off-resonant excitation at harmonics of
the acoustic frequency.
Within this approximation, we find:
\begin{align}
\nonumber
\mathcal{V} = & \frac{1}{2} \sum_{\beta, \alpha, \alpha'}
\int_{-\infty}^\infty \mathrm{d} z \
\Big(
\bar \Gamma_{\alpha \alpha' \beta}
\hat \psi_\alpha \hat \psi^\dagger_{\alpha'}
\hat \phi_{\beta}
\\
& \quad +
\bar \Gamma_{\alpha' \alpha \beta}
\hat \psi^\dagger_{\alpha} \hat \psi_{\alpha'}
\hat \phi_{\beta} \ + \ \text{h.c.} \Big),
\label{eqn:coupling_deriv}
\end{align}
with the coupling coefficient
\begin{align}
\bar \Gamma_{\alpha \alpha' \beta} =
\sqrt{\frac{\hbar^3 \omega_0^2 \Omega_{\beta q_{\beta}}}{
\mathcal{E}_{\alpha} \mathcal{E}_{\alpha'} \mathcal{E}_{\beta}}}
Q^\text{pe}_{\alpha \alpha' \beta},
\end{align}
where $Q^\text{pe}_{\alpha \alpha' \beta}$ is the transverse
electrostrictive overlap integral involving conventional eigenmodes:
\begin{align}
Q^\text{pe}_{\alpha \alpha' \beta} = \frac{1}{\varepsilon_0}
\int \mathrm{d}^2 r \sum_{ijkl}
d^i_{\alpha k_\alpha} (d^j_{\alpha' k_{\alpha'}})^\ast
p_{ijkl} s^{kl}_{\beta q_\beta}.
\end{align}
This integral is only defined up to a phase factor that is given by the
arbitrary global phase of the acoustic eigenmode.
$Q^\text{pe}_{\alpha \alpha' \beta}$ and as a result
$\bar \Gamma_{\alpha \alpha' \beta}$ can be chosen to be real-valued for one
pair of optical modes $\alpha$ and $\alpha'$ (but not necessarily for all
pairs of optical modes at once).
The addition of radiation pressure leads to the same result except for the
addition of a surface integral $Q^\text{mb}_{\alpha \alpha' \beta}$
to the total opto-acoustic overlap term $Q_{\alpha \alpha' \beta}$.
This is well documented in the literature~\cite{Rakich2012,Wolff2015,Sipe2016}
and therefore need not be repeated here.
The total Hamiltonian for the FBS problem is thus
\begin{align}
\mathcal{H} = &
\nonumber
\sum_\alpha \int_{-\infty}^\infty \mathrm{d} z \ \hat \psi^\dagger_{\alpha}
\mathcal{L}_\alpha \hat \psi_{\alpha}
+
\sum_\beta \int_{-\infty}^\infty \mathrm{d} z \ \hat \phi^\dagger_{\beta}
\mathcal{L}_\beta \hat \phi_{\beta}
\\
\nonumber
& \quad + \frac{1}{2} \sum_{\beta, \alpha, \alpha'} \int_{-\infty}^\infty \mathrm{d} z \
\Big(
\bar \Gamma_{\alpha \alpha' \beta}
\hat \psi_\alpha \hat \psi^\dagger_{\alpha'}
\hat \phi_{\beta}
\\
& \quad \quad +
\bar \Gamma_{\alpha' \alpha \beta}
\hat \psi^\dagger_{\alpha} \hat \psi_{\alpha'}
\hat \phi_{\beta} \ + \ \text{h.c.} \Big).
%
\label{eqn:hamiltonian}
\end{align}
If desired, the optical operators in the interaction term can be normal-ordered:
\begin{align}
\hat \psi_\alpha \hat \psi_{\alpha'}^\dagger
=
\hat \psi_{\alpha'}^\dagger \hat \psi_\alpha
+ [ \hat \psi_{\alpha}, \hat \psi^\dagger_{\alpha'} ],
\label{eqn:opt_commutator}
\end{align}
where the commutator diverges according to Eq.~\eqref{eqn:commutator_psi},
because it involves operators $\hat \psi_{\alpha}(z, t)$ and
$\hat \psi^\dagger_{\alpha'}(z, t)$ evaluated at the same coordinate $z$.
This only leads to a static
deformation of the waveguide and is irrelevant for the dynamics of FBS.
It can be interpreted as the contribution of the guided optical modes to the
total Casimir force on the waveguide.
We note that this force is not only restricted to boundary forces, but also
has an electrostrictive contribution.
\subsection{Equations of motion}
The Heisenberg equations for the envelope operators are:
\begin{align}
\partial_t \hat \psi_\alpha &=
\frac{1}{\mathrm{i} \hbar} [\hat \psi_\alpha, \mathcal{H}]
\\
\nonumber
&
=
\frac{1}{\mathrm{i} \hbar} \mathcal{L}_\alpha \hat \psi_\alpha + \frac{1}{2 \mathrm{i} \hbar}
\sum_{\alpha' \beta} \Big[
(\bar \Gamma_{\alpha \alpha' \beta} + \bar \Gamma_{\alpha' \alpha \beta} )
\hat \phi_\beta
\\
\nonumber
& \quad
+
(\bar \Gamma_{\alpha' \alpha \beta}^\ast
+ \bar \Gamma_{\alpha \alpha' \beta}^\ast) \hat \phi_\beta^\dagger
\Big] \hat \psi_{\alpha'} ,
\\
\partial_t \hat \phi_\beta &= \frac{1}{\mathrm{i} \hbar}
[\hat \phi_\beta, \mathcal{H}]
\\
\nonumber
&
=
\frac{1}{\mathrm{i} \hbar} \mathcal{L}_\beta \hat \phi_\beta + \frac{1}{2 \mathrm{i} \hbar}
\sum_{\alpha \alpha'}
(\bar \Gamma_{\alpha' \alpha \beta}^\ast
\hat \psi_\alpha \hat \psi_{\alpha'}^\dagger
+ \bar \Gamma_{\alpha \alpha' \beta}^\ast
\hat \psi_\alpha^\dagger \hat \psi_{\alpha'}).
\end{align}
As we have already discussed, the commutator
\mbox{$[\hat \psi_{\alpha'}, \hat \psi_\alpha^\dagger]$} is irrelevant for the
dynamics of FBS and therefore it can be safely ignored:
\begin{align}
\partial_t \hat \phi_\beta = &
\frac{1}{\mathrm{i} \hbar} \mathcal{L}_\beta \hat \phi_\beta
\\
\nonumber
& \quad
+
\frac{1}{2 \mathrm{i} \hbar} \sum_{\alpha \alpha'}
(
\bar \Gamma_{\alpha' \alpha \beta}^\ast
\hat \psi_{\alpha'}^\dagger \hat \psi_{\alpha}
+
\bar \Gamma_{\alpha \alpha' \beta}^\ast
\hat \psi_{\alpha}^\dagger \hat \psi_{\alpha'}
).
\end{align}
It is convenient to transform the optical operators to a rotating frame that
eliminates both the temporal and the spatial carrier.
It should be stressed that we do not apply such a transformation to the
acoustic envelopes as this would lead to an explicitly time-dependent
interaction term.
Furthermore, the (nearly) vanishing acoustic group velocity suggests a
rescaling to normalize with respect to energies rather than powers:
\begin{align}
\hat \Psi_\alpha(z, t)
= & \hat \psi_{\alpha}(z, t) \sqrt{\hbar \omega_0}
\exp(\mathrm{i} \omega_0 t - \mathrm{i} k_\alpha z),
\\
\hat \Phi_\beta(z, t) = & \hat \phi_{\beta}(z, t)
\sqrt{\hbar \Omega_{\beta q_\beta}},
\\
\Gamma_{\alpha \alpha' \beta}
= & \frac{1}{ \hbar \omega_0 \sqrt{\hbar \Omega_{\beta q_\beta}} }
\bar \Gamma_{\alpha \alpha' \beta}
=
\frac{1}{\sqrt{\mathcal{E}_{\alpha} \mathcal{E}_{\alpha'} \mathcal{E}_{\beta}}} Q_{\alpha \alpha' \beta}
.
\end{align}
The product operators $\hat \Psi^\dagger \hat \Psi$ and
$\hat \Phi^\dagger \hat \Phi$ thus express the local optical and acoustic
energies rather than the local number of photons and phonons, respectively.
\begin{widetext}
\noindent
Using Eqs.~(\ref{eqn:Lalpha_def},\ref{eqn:Lbeta_def}), the resulting quantum
equations of motion are:
\begin{align}
\nonumber
\left[ \partial_t + v_\alpha^{(1)} \partial_z -
\mathrm{i} \frac{v_\alpha^{(2)}}{2} \partial_z^2 + \ldots \right]
\hat \Psi_\alpha
= & - \frac{\mathrm{i} \omega_0}{2} \sum_{\alpha' \beta}
\mathrm{e}^{\mathrm{i}(k_{\alpha'} - k_\alpha)z}
\Big[
(\Gamma_{\alpha \alpha' \beta} + \Gamma_{\alpha' \alpha \beta} )
\hat \Phi_\beta
\\
& \quad +
(\Gamma_{\alpha \alpha' \beta}^\ast + \Gamma_{\alpha' \alpha \beta}^\ast )
\hat \Phi_\beta^\dagger
\Big] \hat \Psi_{\alpha'}
,
\label{eqn:eom_psi_qm}
\\
\nonumber
\left[ (\partial_t + \mathrm{i} \Omega_{\beta q_\beta})
+ v_\beta^{(1)} (\partial_z - \mathrm{i} q_\beta) -
\mathrm{i} \frac{v_\beta^{(2)}}{2} (\partial_z - \mathrm{i} q_\beta)^2 +
\ldots
\right]
\hat \Phi_\beta
= & - \frac{\mathrm{i} \Omega_{\beta q_\beta}}{2} \sum_{\alpha \alpha'}
\Big[
\mathrm{e}^{\mathrm{i}(k_{\alpha'} - k_{\alpha})z}
\Gamma_{\alpha' \alpha \beta}^\ast
\hat \Psi_\alpha^\dagger \hat \Psi_{\alpha'}
\\
& \quad +
\mathrm{e}^{\mathrm{i}(k_{\alpha} - k_{\alpha'})z}
\Gamma_{\alpha \alpha' \beta}^\ast
\hat \Psi_{\alpha'}^\dagger \hat \Psi_\alpha)
\Big] .
\label{eqn:eom_phi_qm}
\end{align}
In most situations a classical description is sufficient and allows for the
simple inclusion of loss via damping coefficients $\gamma_\alpha$ and
$\gamma_\beta$.
To this end, we identify the expectation value of the QM energy density
operator with the classical energy density of modulated modes:
\begin{align}
\langle \hat \Psi_\alpha^\dagger \hat \Psi_\alpha \rangle
= & \mathcal{E}_\alpha |a_\alpha|^2,
&
\langle \hat \Phi_\beta^\dagger \hat \Phi_\beta \rangle
= & \mathcal{E}_\beta |\tilde b_\beta|^2,
\end{align}
where $a_\alpha(z,t)$ and $\tilde b_{\beta}(z,t)$ are classical
dimensionless envelope functions.
The tilde indicates that the acoustic envelopes are subject to a different
phase convention than the optical envelopes.
The classical equations of motion are therefore obtained from
Eqs.~(\ref{eqn:eom_psi_qm},\ref{eqn:eom_phi_qm}) by identifying:
\begin{align}
\hat \Psi_\alpha \rightarrow & \sqrt{\mathcal{E}_\alpha} a_\alpha,
&
\hat \Phi_\beta \rightarrow & \sqrt{\mathcal{E}_\beta} \, {\tilde b}_\beta;
\end{align}
\begin{align}
\nonumber
\left[ \gamma_\alpha + \partial_t + v_\alpha^{(1)} \partial_z -
\mathrm{i} \frac{v_\alpha^{(2)}}{2} \partial_z^2 + \ldots \right]
a_\alpha
= & - \frac{\mathrm{i} \omega_{\alpha k_\alpha}}{\mathcal{E}_\alpha}
\sum_{\alpha' \beta}
\mathrm{e}^{\mathrm{i}(k_{\alpha'} - k_\alpha)z}
\\
& \quad \times \Re\left\{
\big( Q_{\alpha \alpha' \beta} + Q_{\alpha' \alpha \beta} \big) \tilde b_\beta
\right\} a_{\alpha'},
\label{eqn:eom_psi_cl}
\\
\nonumber
\left[ \gamma_\beta + (\partial_t + \mathrm{i} \Omega_{\beta q_\beta})
+ v_\beta^{(1)} (\partial_z - \mathrm{i} q_\beta) -
\mathrm{i} \frac{v_\beta^{(2)}}{2} (\partial_z - \mathrm{i} q_\beta)^2 +
\ldots
\right]
\tilde b_\beta
= & - \frac{\mathrm{i} \Omega_{\beta q_\beta} }{2\mathcal{E}_\beta}
\sum_{\alpha \alpha'} \big(
\mathrm{e}^{\mathrm{i}(k_{\alpha'} - k_\alpha)z}
Q_{\alpha' \alpha \beta}^\ast a^\ast_\alpha a_{\alpha'}
\\
& \quad +
\mathrm{e}^{\mathrm{i}(k_{\alpha} - k_{\alpha'})z}
Q_{\alpha \alpha' \beta}^\ast a^\ast_{\alpha'} a_\alpha
\big) .
\label{eqn:eom_phi_cl}
\end{align}
Note that the optical damping coefficient is defined with respect to amplitudes
and not powers.
Thus, it satisfies $\gamma_\alpha=v\alpha_\text{loss}/2$, where
$\alpha_\text{loss}$ is the conventional power attenuation coefficient.
\end{widetext}
The framework presented thus far is directly applicable intra-mode FBS -- which
is the main focus of this paper -- and for non-cascaded inter-mode backward SBS.
The treatment of cascaded inter-mode FBS and intra-mode backward SBS requires
\emph{two} carriers either in an acoustic or in an optical mode in order to
describe forward and backward propagating acoustic waves (in the case of
inter-mode FBS) or forward and backward propagating optical waves (in the case
of intra-mode backward SBS).
operators defined via Fourier transformations over half the reciprocal space.
\subsection{Intra-mode FBS}
Often, only one optical mode $\alpha=\alpha'$ and one acoustic mode $\beta$
are relevant.
In this case, the coefficients simplify and phase factors disappear:
\begin{align}
Q_{\alpha \alpha' \beta} = & Q_{\alpha' \alpha \beta} = Q,
\\
\mathrm{e}^{\mathrm{i}(k_{\alpha'} - k_\alpha)z} = &
\mathrm{e}^{\mathrm{i}(k_\alpha - k_{\alpha'})z} = 1,
\end{align}
the mode-index summations in the interaction terms disappear
and we drop the subscripts $\alpha$, $\beta$, $k_\alpha$ and $q_\beta$
where appropriate and use $k_0$, $\Omega_0$, and $q_0$ to denote the optical
and acoustic carrier wave parameters.
This leads to simplified quantum equations of motion
\begin{align}
\left[ \partial_t + v_\alpha^{(1)} \partial_z
+ \ldots \right] \hat \Psi
= - \mathrm{i} \omega_0
(\Gamma \hat \Phi +
\Gamma^\ast \hat \Phi^\dagger)
\hat \Psi,
\\
\left[ (\partial_t + \mathrm{i} \Omega_0)
+ v_\beta^{(1)} (\partial_z - \mathrm{i} q_0)
+ \ldots \right] \hat \Phi
= - \mathrm{i} \Omega_0 \Gamma^\ast \hat \Psi^\dagger \hat \Psi,
\end{align}
and their classical counterparts:
\begin{align}
\left[ \gamma_\alpha + \partial_t + v_\alpha^{(1)} \partial_z
+ \ldots \right] a
= - \frac{2 \mathrm{i} \omega_0}{\mathcal{E}_\alpha} \Re\{Q \tilde b\} a,
\\
\left[ \gamma_\beta + (\partial_t + \mathrm{i} \Omega_0)
+ v_\beta^{(1)} (\partial_z - \mathrm{i} q_0)
+ \ldots \right] \tilde b
= - \frac{\mathrm{i} \Omega_0}{\mathcal{E}_\beta} Q^\ast |a|^2.
\end{align}
The remainder of this paper is entirely based on the classical equations of
motion with only one relevant optical and one relevant acoustic mode.
We now demonstrate how the Eqs.~(\ref{eqn:eom_psi_cl},\ref{eqn:eom_phi_cl})
can be reduced to the standard treatment with individual optical amplitudes
for each Stokes order.
First, we assume that the optical amplitude can be represented as a
superposition of a pump and $N$ Stokes and $M$ anti-Stokes orders:
\begin{align}
a(z,t) = \sum_{n=-M}^N a_n(z,t) \exp[-\mathrm{i} n (\Omega_0 - q_0 z t)],
\end{align}
where each individual amplitude $a_n$ is assumed to vary slowly on the time
and length scales of the \emph{acoustic} problem.
Furthermore, it is now convenient to transform the acoustic envelope to a
rotating frame:
\begin{align}
b(z,t) = \tilde b(z, t) \exp(\mathrm{i} \Omega_0 t - \mathrm{i} q_0 z).
\end{align}
We insert this into Eqs.~(\ref{eqn:eom_psi_cl}) to obtain:
\begin{align}
0 = & \sum_{n=-M}^N a_n
\Bigg\{
\mathrm{e}^{\mathrm{i} n (\mathrm{i} \Omega_0 t - q_0 z)}
\Bigg[ \gamma_\alpha + (\partial_t - \mathrm{i} n \Omega_0)
\\
\nonumber
&
+ v_\alpha^{(1)} (\partial_z + \mathrm{i} n q_0)
- \mathrm{i} \frac{v_\alpha^{(2)}}{2} (\partial_z^2 + 2 n \mathrm{i} q_0 - q_0^2)
+ \ldots \Bigg]
\\
\nonumber
&
+ \frac{\mathrm{i} \omega_0}{\mathcal{E}_\alpha} \Bigg[
\mathrm{e}^{\mathrm{i} (n + 1) (\mathrm{i} \Omega_0 t - q_0 z)}
Q b
+
\mathrm{e}^{\mathrm{i} (n - 1) (\mathrm{i} \Omega_0 t - q_0 z)}
Q^\ast b^\ast
\Bigg] \Bigg\}.
\end{align}
Given the slowly varying constraint,
this equation is equivalent to a family of equations for the individual Stokes
orders:
\begin{align}
\nonumber
&
\left[ \gamma_\alpha + (\partial_t - \mathrm{i} n \Omega_0) +
v_\alpha^{(1)} (\partial_z + \mathrm{i} n q_0) +
\ldots \right] a_n
\\
= & - \frac{\mathrm{i} \omega_0}{\mathcal{E}_\alpha} (Q b a_{n-1} + Q^\ast b^\ast a_{n+1}),
\label{eqn:expanded_psi}
\end{align}
with the truncation condition
\begin{align}
a_{-M-1} = & a_{N+1} = 0.
\label{eqn:expanded_truncation_condition}
\end{align}
Analogously, we now insert the optical decomposition into the acoustic equation
of motion~\eqref{eqn:eom_psi_cl}:
\begin{align}
& \left[ \gamma_\beta + \partial_t + v_\beta^{(1)} \partial_z -
\mathrm{i} \frac{v_\beta^{(2)}}{2} \partial_z^2 + \ldots \right]
b
\label{eqn:expanded_phi}
\\
\nonumber
= & - \frac{\mathrm{i} \Omega_0 Q^\ast}{\mathcal{E}_\beta} \sum_{n=-M}^{N-1}
a_n^\ast a_{n+1} + \text{non-resonant terms},
\end{align}
where ``non-resonant terms'' contains all products of optical envelopes where
the beating is not proportional to $\exp(\mathrm{i} q_0 z - \mathrm{i} \Omega_0 t)$.
Neglecting these terms is equivalent to tightening the optical RWA to an
acousto-optical RWA that ignores all contributions that disappear when averaged
over a time scale much longer than the \emph{acoustic} period.
Eqs.~(\ref{eqn:expanded_psi}--\ref{eqn:expanded_phi}) are the conventional
coupled-mode equations for a large number of (anti-) Stokes orders.
The common case of a pump and a single Stokes side-band are obtained for $M=0$, $N=1$:
\begin{align}
\left( \gamma_\alpha + \partial_t + v_\alpha^{(1)} \partial_z
+ \ldots \right) a_0
= & - \frac{\mathrm{i} \omega_0 Q}{\mathcal{E}_\alpha} b a_1,
\label{eqn:dual_mode_1}
\\
\left( \gamma_\alpha + \partial_t + v_\alpha^{(1)} \partial_z
+ \ldots \right) a_1
= & - \frac{\mathrm{i} \omega_0 Q^\ast}{\mathcal{E}_\alpha} b^\ast a_0,
\label{eqn:dual_mode_2}
\\
\left( \gamma_\beta + \partial_t + v_\beta^{(1)} \partial_z
+ \ldots \right) b
= & - \frac{\mathrm{i} \Omega_0 Q^\ast}{\mathcal{E}_\beta} a_1^\ast a_0,
\label{eqn:dual_mode_3}
\end{align}
where we have used the fact that the terms containing $\Omega_0$ and $q_0$
cancel each other in Eq.~\eqref{eqn:dual_mode_2} if the acoustic wave satisfies
the phase-matching condition.
The terms with $\Omega_0$ and $q_0$ never appear in Eq.~\eqref{eqn:dual_mode_1},
because $n=0$.
Equations~(\ref{eqn:dual_mode_1}--\ref{eqn:dual_mode_3}) are in agreement with
the literature on conventional classical coupled mode theory for
SBS~\cite{Rakich2012,Wolff2015}.
With this, we have demonstrated that the commonly employed multi-envelope theory
to FBS emerges from our treatment by restricting the assumptions about the
optical spectrum and neglecting terms oscillating at multiples of the acoustic
frequency in the interaction.
\section{Complete analytical solution in dispersionless intra-mode FBS}
\label{sec:dispersionless}
Optical and acoustic dispersion are typically very small over the narrow
frequency ranges (usually MHz to GHz) of Brillouin phenomena.
Furthermore the propagation length of sound in a waveguide can be far smaller
than the variation of the acoustic and optical envelopes, and so the
propagation of acoustic waves is often neglected in Brillouin calculations.
Having established the governing
equations~(\ref{eqn:eom_psi_cl},\ref{eqn:eom_phi_cl})
we now examine the important limit where both optical and acoustic dispersion
can be entirely neglected, and where the propagation length of acoustic waves
is assumed to be equal to zero.
We will see that this canonical situation provides insight, in the form of
analytic solutions to the governing equations, into the overall behavior of
FBS processes.
In this section, we will solve the dispersionless initial value problem
corresponding to Eqs.~(\ref{eqn:eom_psi_cl},\ref{eqn:eom_phi_cl}) in this
no-dispersion, no-acoustic-propagation limit:
\begin{align}
\left( \partial_t + v \partial_z \right) a
= & - \left(\frac{2 \mathrm{i} \omega_0}{\mathcal{E}_\alpha}
\Re\{Q \tilde b\} + \gamma_\alpha
\right) a,
\label{eqn:anasol_eom_opt}
\\
\left( \partial_t + \gamma_\beta + \mathrm{i} \Omega_0 \right) \tilde b
= & - \frac{\mathrm{i} \Omega_0 Q^\ast}{\mathcal{E}_\beta} |a|^2,
\label{eqn:anasol_eom_ac}
\\
a(z=0, t) = & A(t),
\label{eqn:anasol_eom_bc}
\\
a(z, t) = & \tilde b(z, t) = A(t) = 0 \quad \text{for} \quad t < 0,
\label{eqn:anasol_eom_ic}
\end{align}
where $A(t)$ is the incident optical envelope at the start of the waveguide.
The general approach in solving this system is to first integrate the optical
equation~\eqref{eqn:anasol_eom_opt} to find the optical amplitude as a function
of $z$ and $t$.
This solution will depend on the incident optical field $A(t)$ and the yet
unknown acoustic envelope $\tilde b(z, t)$.
We then insert it into Eq.~\eqref{eqn:anasol_eom_ac} and find that the
resulting equation happens to be solvable.
By back-substitution into the solution to Eq.~\eqref{eqn:anasol_eom_opt} we
find the analytical solution to cascaded FBS as an explicit integral and
evaluate it for a typical experimental setup.
\subsection{General solution}
Eq.~\eqref{eqn:anasol_eom_opt} is separable when integrating along the
characteristic of the one-way wave equation.
In other words, we obtain the phase of the optical amplitude at the point
\mbox{$(z,t)$} by integrating along a line \mbox{$(z',t')$} in space time
defined by
\begin{align}
z' = & z - vs, & t' = & t - s,
\end{align}
where $s$ is the running parameter.
This leads to the solution (expressed in the original coordinates):
\begin{align}
a(z, t) = & E\Big(t - \frac{z}{v}\Big)
\exp\Big(-\frac{\gamma_\alpha z}{ v}\Big)
\label{eqn:anasol_step1}
\\
\nonumber & \ \times
\exp\left[\frac{2 \omega_0}{\mathrm{i} \mathcal{P}_\alpha}\int_0^z \mathrm{d} z' \
\Re\left\{Q \tilde b\left(z', t - \frac{z - z'}{v}\right)\right\}
\right].
\end{align}
It should be pointed out that the integrand is purely real-valued and thus
the acoustic envelope only introduces a phase.
In the absence of optical dispersion, the effect of the acoustic deformation on
the optical wave is a pure phase modulation and the optical intensity
does not depend on $\tilde b$.
However, the only thing relevant for Eq.~\eqref{eqn:anasol_eom_ac} is the
optical intensity distribution:
\begin{align}
|a(z, t)|^2 = & \left|E\left(t - \frac{z}{v}\right)\right|^2
\exp\Big(-\frac{2 \gamma_\alpha z }{ v} \Big).
\end{align}
With this, Eq.~\eqref{eqn:anasol_eom_ac} becomes an ordinary differential
equation that can be explicitly solved via its Green function:
\begin{align}
\tilde b(z, t) = & \frac{\Omega_0 Q^\ast}{\mathrm{i} \mathcal{E}_\beta}
\int_0^\infty \mathrm{d} t' \
\text{e}^{-(\gamma_\beta + \mathrm{i} \Omega_0)t' -
\frac{2\gamma_\alpha z}{v}}
\left| E\left(t - t' - \frac{z}{v}\right) \right|^2,
\label{eqn:anasol_step2}
\end{align}
This expression only depends on $A(t)$, but not on $a(z, t)$.
Therefore, Eq.~\eqref{eqn:anasol_step1} is a simple integral rather than an
integral equation.
Furthermore, the $z$-integral in Eq.~\eqref{eqn:anasol_step1} is trivial.
We find for the $\tilde b$-related term appearing under the $z$-integral:
\begin{align}
& - \Re\left\{ Q \tilde b\left(z', t - \frac{z - z'}{v}\right)\right\}
\\
\nonumber
= & \frac{\Omega_0 |Q|^2}{\mathcal{E}_\beta} \int_0^\infty \mathrm{d} t' \
\text{e}^{-\gamma_\beta t' - \frac{2\gamma_\alpha z'}{v}}
\left| E\left(t - t' - \frac{z}{v}\right) \right|^2
\sin(\Omega_0 t').
\end{align}
The stationary solution for the optical envelope of cascaded FBS in the
dispersionless case and for any incident optical envelope $A(t)$ is therefore:
\begin{align}
\nonumber
& a(z, t) = E\bigg(t - \frac{z}{v}\bigg)
\exp\bigg\{-\frac{\gamma_\alpha z}{v}
+ \frac{\mathrm{i} \Omega_0 \omega_0 |Q|^2
\big(1 - \text{e}^{-\frac{2 \gamma_\alpha z}{v}}\big)}{
\gamma_\alpha \mathcal{E}_\alpha \mathcal{E}_\beta
}
\\
\nonumber
& \quad \times
\int_0^\infty \mathrm{d} t' \ \text{e}^{-\gamma_\beta t'}
\left| E\left(t - t' - \frac{z}{v}\right) \right|^2
\sin(\Omega_0 t')
\bigg\},
\\
\nonumber
& = E\bigg(t - \frac{z}{v}\bigg)
\exp\bigg\{-\frac{\gamma_\alpha z}{v}
+ \frac{\mathrm{i} \Omega_0 \omega_0 |Q|^2 }{ \mathcal{E}_\alpha \mathcal{E}_\beta }
\frac{2 L_\text{eff}}{v}
\\
& \quad \times
\int_0^\infty \mathrm{d} t' \ \text{e}^{-\gamma_\beta t'}
\left| E\left(t - t' - \frac{z}{v}\right) \right|^2
\sin(\Omega_0 t')
\bigg\},
\label{eqn:dispersionless_solution}
\end{align}
where the effective propagation length of the optical field at point $z$ is
\begin{align}
L_\text{eff}=\frac{
v \big(1 - \text{e}^{-\frac{2 \gamma_\alpha z}{v}}\big)}{ 2 \gamma_\alpha }.
\end{align}
The resulting optical field can then be understood as the input field $A(t)$
attenuated according to the damping $\gamma_a/v$ while being phase modulated by
its own intensity over the effective length.
\subsection{Example: two-tone excitation and zero optical loss}
\label{sec:two-tone}
The explicit formula~\eqref{eqn:dispersionless_solution} is the complete
solution to the problem of cascaded FBS in the absence of optical dispersion.
To gain more insight into its nature,
we next evaluate it for an example that is typical both for theoretical and
experimental studies of SBS: the excitation with a strong optical pump and a
weak Stokes side band:
\begin{align}
A(t) = A_0 [1 - \mathrm{i} M \exp(\mathrm{i} \Omega_0 t)],
\end{align}
where $A_0$ is the amplitude of the pump wave and $M A_0$ is that of the
Stokes seed.
The corresponding pump and Stokes seed powers are
\begin{align}
P_\text{P} = & \mathcal{P}_\alpha |A_0|^2
&
P_\text{S} = & \mathcal{P}_\alpha |M A_0|^2
\end{align}
We assume that $M$ is real-valued, which is always possible by an
appropriately chosen time origin.
We furthermore neglect the optical loss $\gamma_\alpha$ in this specific
example; the inclusion is straight-forward but tends to obscure the
equations.
On thus letting $L_\text{eff}\to z$ and $\gamma_\alpha \to 0$,
the solution becomes
\begin{align}
a(z, t) = &
A\Big(t - \frac{z}{v}\Big)
\exp\Bigg\{\frac{2 \mathrm{i} \Omega_0 \omega_0 |Q|^2 z}{\mathcal{P}_\alpha \mathcal{E}_\beta}
\\
\nonumber
& \quad \times
\int_0^\infty \mathrm{d} t' \ \text{e}^{-\gamma_\beta t'}
\left| A\Big(t - t' - \frac{z}{v}\Big) \right|^2
\sin(\Omega_0 t')
\Bigg\}
\\
= &
A\Big(t - \frac{z}{v}\Big)
\exp\Bigg\{
\underbrace{
\frac{2 \mathrm{i} \Omega_0^2 \omega_0 |Q|^2 (P_\text{P} + P_\text{S}) z}{
\mathcal{P}_\alpha^2 \mathcal{E}_\beta (\gamma_\beta^2 + \Omega_0^2) }
}_{\text{static term}}
\\ & \quad +
- \frac{2 \mathrm{i} \Omega_0 \omega_0 |Q|^2 \sqrt{P_\text{P} P_\text{S}} z}{\gamma_\beta \mathcal{P}_\alpha^2 \mathcal{E}_\beta}
\Bigg[
\nonumber
\underbrace{
\cos\Big(\Omega_0 t - \frac{\Omega_0 z}{v} \Big)
}_{\text{dominant term}}
\\
\nonumber
& \quad +
\underbrace{
\frac{\gamma_\beta \mathrm{e}^{-\mathrm{i} \Omega_0 (t - z/v)} }{2 (\gamma_\beta - 2 \mathrm{i} \Omega_0)}
+ \frac{\gamma_\beta \mathrm{e}^{\mathrm{i} \Omega_0 (t - z/v)} }{2 (\gamma_\beta + 2 \mathrm{i} \Omega_0)}
}_{\text{counter-rotating term}}
\Bigg] \Bigg\}.
\end{align}
For high mechanical quality factors (\mbox{i.e.\ } $\Omega_0 \gg \gamma_\beta$), the
additional phase factor due to static waveguide deformations and the
out-of-phase contributions due to the counter-rotating term are negligible
compared to the resonant effect.
We introduce a natural unit of length that depends on the optical input powers
and the intrinsic SBS parameters and we shift the origin of the time coordinate
to a more convenient point:
\begin{align}
\zeta = & z / L_\text{nat},
\\
\tau = & t - \frac{z}{\Omega_0 v},
\\
\text{with} \quad L_\text{nat} = &
\frac{\mathcal{P}_\alpha^2 \mathcal{E}_\beta \gamma_\beta}{2 \Omega_0 \omega_0 |Q|^2 \sqrt{P_\text{P} P_\text{S}} }.
\label{eqn:units}
\end{align}
In these units, the cascaded FBS effect [excluding static deformations that
only introduce a very slow phase factor $\exp(\mathrm{i} A \zeta)$ where $A$ is
approximately the inverse mechanical quality factor]
takes a generic and simple form:
\begin{align}
a(\zeta, \tau) = & A(\tau) \exp[\mathrm{i} \zeta \cos(\Omega_0 \tau)].
\label{eqn:dispersionless_solution_natunits}
\end{align}
We will now investigate three properties of this solution that are of particular
interest for the characterization of SBS-based frequency combs: the frequency
spectrum, the intensity evolution and the auto-correlation function.
\begin{figure}
\includegraphics[width=\columnwidth]{dispersionless_spectrum.eps}
\caption{
Evolution of the absolute amplitude (gray scale) of various \mbox{(anti-)}Stokes orders
(frequency on the abscissa) as a function of position $\zeta$ along the
wave guide (ordinate) assuming vanishing dispersion and an optical
excitation with a strong pump and a weak Stokes side-band.
}
\label{fig:stokes_evolution}
\end{figure}
The frequency spectrum of the total
signal Eq.~\eqref{eqn:dispersionless_solution_natunits} is the convolution of
the spectrum of the incident signal $A(\tau)$ and the SBS-modulation function
\begin{align}
\mathcal{R}(\zeta, \tau) = \exp[\mathrm{i} \zeta \cos(\Omega_0 \tau)],
\end{align}
which reflects the effect of cascaded FBS.
Due to the periodicity of the problem, its spectrum only contains narrow lines
at integer multiples of the Stokes shift $\Omega_0$.
Therefore, the spectral component of the $n$-th (anti-)Stokes order is given by
\begin{align}
\mathcal R_n(\zeta) = & \frac{1}{2\pi} \int_0^{2 \pi / \Omega_0} \mathrm{d} \tau \
\exp[\mathrm{i} n \Omega_0 \tau + \mathrm{i} \zeta \cos(\Omega_0 \tau)]
\\
= & \frac{1}{\Omega_0} J_n(\zeta),
\end{align}
where $J_n(\zeta)$ is the Bessel function of order $n$.
We therefore can directly plot the generation of the individual Stokes orders
along the waveguide (see Fig.~\ref{fig:stokes_evolution}).
Cascaded FBS clearly generates a frequency comb in the sense that it generates
many equally spaced frequency lines.
However, very often frequency combs are associated with the generation of
light pulses that are equally spaced in time domain.
The solution of cascaded FBS in the absence of optical dispersion in a
straight waveguide does not predict the formation of pulses.
Even more, the SBS response function $\mathcal{R}(\zeta, \tau)$ is a pure
phase factor, \mbox{i.e.\ } the FBS process does not affect the intensity distribution
at all.
A related feature of this is that (in the absence of optical loss and
dispersion) the acoustic amplitude is constant along the waveguide: the sound
wave is not amplified.
This is in agreement with an earlier work on the optical single-pass gain of
FBS structures~\cite{Rakich2016}.
As a result, the effect of cascaded FBS in the non-dispersive regime cannot
be observed in the RF-domain by detecting the interference within the optical
output with a photodetector.
\begin{figure}
\includegraphics[width=\columnwidth]{dispersionless_auto_correlation.eps}
\caption{
Absolute square $|\mathcal{L}(\tau)|^2$ of the auto-correlation function
[see Eq.~\eqref{eqn:auto_correlation_def}]
of the SBS-generated signal modulation $\mathcal{R}(\zeta, \tau)$ for
a range of waveguide lengths $\zeta$ expressed in natural units of length
$L_\text{nat}$ [see Eq.~\eqref{eqn:units}].
The delay parameter $\tau$ is normalized to the acoustic period
$T = 2\pi/\Omega_0$.
}
\label{fig:auto_correlation}
\end{figure}
Finally, auto-correlation measurements of optical combs are often used instead
of measuring the time-dependence of the optical intensity in order to grade the
quality of a frequency comb.
It turns out that the cascaded FBS process in a dispersionless straight
waveguide causes a pronounced auto-correlation peak despite the complete lack
of intensity modulation.
The amplitude auto-correlation for this problem with time period
$T=2\pi/\Omega_0$ can be defined as:
\begin{align}
\mathcal{L}(\tau) = & \frac{1}{T} \int_0^T \mathrm{d} \tau' \
\mathcal{R}^\ast(\zeta, \tau') \mathcal{R}(\zeta, \tau' + \tau)
\label{eqn:auto_correlation_def}
\\
\nonumber
= & \frac{1}{T} \int_0^T \mathrm{d} \tau' \
\exp\big\{
\mathrm{i} \zeta [\cos \Omega_0 \tau' - \cos\Omega_0(\tau'+\tau)]
\big\}
\\
= & J_0\left(2 \zeta \sin\frac{\Omega_0 \tau}{2}\right),
\label{eqn:auto_correlation}
\end{align}
where $J_0(x)$ is the Bessel function of order zero.
In Fig.~\ref{fig:auto_correlation} we show the absolute value squared of this
result for a range of normalized waveguide lengths $\zeta$.
\section{Intra-mode FBS in dispersive waveguides}
\label{sec:dispersive}
In the previous section, we have shown that in the absence of optical dispersion
our cascaded model of FBS does not exhibit variations in the acoustic amplitude
nor the intensity modulation along the waveguide.
This is in fact not due to some hidden assumptions within our model but the
result of the conservation of momentum and energy, \mbox{i.e.\ } phase matching.
To illustrate this, we assume a waveguide that is acoustically modulated
with the angular frequency $\Omega_0$ and ask how the forward-scattering of
light by this modulation modifies the acoustic intensity.
\begin{figure}
\includegraphics[width=0.7\columnwidth]{dispersion_sketch.eps}
\caption{
Sketch of a dispersive optical band.
Three points are highlighted: a carrier wave at $\omega_0$ and $k_0$,
an anti-Stokes wave at $\omega_0 + \Omega_0$ and $k_0 + k_+$ and a
Stokes wave at $\omega_0 - \Omega_0$ and $k_0 + k_-$ (note that $k_-$ is
chosen to be negative).
}
\label{fig:dispersion_sketch}
\end{figure}
Fig.~\ref{fig:dispersion_sketch} shows schematically a small section of an
optical dispersion relation: a pump field at angular frequency $\omega_0$ and
wave number $k_0$ is scattered into the Stokes and anti-Stokes side bands at
angular frequencies $\omega_0 \mp \Omega_0$ and wave numbers $k_0 + k_\mp$.
The wave number differences can be expanded in terms of the frequency:
\begin{align}
k_\pm = & \pm s^{(1)} \Omega_0 + \frac{s^{(2)}}{2} \Omega_0^2 + \ldots ,
\end{align}
where the group slowness and slowness dispersion parameters are connected to
the more familiar group velocity and dispersion parameters
[Eq.~\eqref{eqn:group_velocity_def}] via:
\begin{align}
s^{(1)} = & [v^{(1)}]^{-1}, & s^{(2)} = -v^{(2)}[v^{(1)}]^{-3}.
\end{align}
When part of the optical pump at $\omega_0$ is scattered, its energy and
momentum are distributed between the acoustic system and the two optical side
bands expressed by amplitudes $a_-$ and $a_+$ at the Stokes and anti-Stokes
frequencies, respectively.
First, we will find a condition for case that the acoustic amplitude is not
modified, \mbox{i.e.\ } that the energy and momentum are distributed only among the
optical side bands:
\begin{align}
\hbar k_+ |a_+|^2 + \hbar k_- |a_-|^2 = & 0,
\\
\hbar \Omega_0 |a_+|^2 - \hbar \Omega_0 |a_-|^2 = & 0.
\end{align}
This is only possible if $k_+ = -k_-$, \mbox{i.e.\ } for vanishing optical dispersion
(note that the acoustic wave number $q=k_+$ is not zero).
In addition, we find that $|a_+|^2 = |a_-|^2$, \mbox{i.e.\ } that both side bands are
equally excited in agreement with Sec.~\ref{sec:dispersionless}.
Finally, we may conclude that in the presence of optical dispersion, a two-wave
process is not sufficient to simultaneously conserve energy and momentum, \mbox{i.e.\ }
the acoustic system is required as a third participating wave.
Next, we we introduce $B$ as the (positive or negative) change in the
acoustic intensity $|b|^2$.
The conservation of energy requires:
\begin{align}
\hbar \Omega_0 |a_+|^2 - \hbar \Omega_0 |a_-|^2 + \hbar \Omega_0 B = & 0,
\\
\Rightarrow \quad B = & |a_-|^2 - |a_+|^2.
\label{eqn:qualitative_disp_energy_cons}
\end{align}
Self-amplification occurs if energy is dumped into the acoustic field, \mbox{i.e.\ }
for $B>0$.
For the conservation of momentum, we find:
\begin{align}
\hbar k_+ |a_+|^2 + \hbar k_- |a_-|^2 + \hbar q B = & 0,
\\
\Rightarrow \quad
\frac{s^{(2)}\Omega_0^2}{2} (|a_-|^2 + |a_+|^2) + (q - s^{(1)} \Omega_0) B = & 0.
\label{eqn:qualitative_disp_momentum_cons}
\end{align}
From this we can see that a change in the acoustic intensity is necessary in
the presence of optical dispersion.
Furthermore, we can see that (unlike in the dispersionless case) the acoustic
phase velocity must slightly differ form the optical group velocity.
This very simple argument based on the conservation of energy and momentum
alone demonstrates that dispersion is necessarily required to modify the
acoustic intensity via FBS.
However, it cannot predict in which regime self-amplification occurs, because
no assumptions are made about the opto-acoustic interaction.
This requires a more detailed approach, as we will show in the following.
\subsection{Amplification \& suppression in the weak dispersion limit}
We now study the full problems of FBS with optical dispersion,
where we neglect optical loss for the sake of simplicity:
\begin{align}
\bigg( \partial_t + v^{(1)}_\alpha \partial_z -
\frac{\mathrm{i} v^{(2)}_\alpha}{2} \partial_z^2 \bigg) a
= & \frac{2 \mathrm{i} \omega_0}{\mathcal{E}_\alpha}
\Re\{Q \tilde b\} a,
\label{eqn:dispersive_eom_opt_orig}
\\
\left( \partial_t + \gamma_\beta + \mathrm{i} \Omega_0 \right) \tilde b
= & - \frac{\mathrm{i} \Omega_0 Q^\ast}{\mathcal{E}_\beta} |a|^2,
\label{eqn:dispersive_eom_ac_orig}
\\
a(z=0, t) = & A(t),
\label{eqn:dispersive_eom_bc_orig}
\\
a(z, t) = b(z, t) = A(t) = & 0 \quad \text{for} \quad t < 0.
\label{eqn:dispersive_eom_ic_orig}
\end{align}
In this problem, the optical dispersion gradually shifts the phase between
adjacent Stokes lines with propagation along $z$ and therefore slowly
converts the pure phase-modulation
of the opto-acoustic interaction into a partial amplitude-modulation.
Depending on the sign of this process, the optical interference pattern is
amplified or attenuated along the waveguide.
Unfortunately, the parabolic nature of the optical equation precludes
a closed form solution similar to Eq.~\eqref{eqn:dispersionless_solution}.
We therefore restrict ourselves to a few relevant aspects of the problem.
First, we eliminate constants that only clutter the notation by substituting
\begin{align}
f(z, t) = & \frac{|Q|}{\sqrt{\mathcal{E}_\alpha \mathcal{E}_\beta}}
a(z, t),
\\
g(z, t) = & - \frac{Q}{\mathcal{E}_\alpha} \tilde b(z, t),
\end{align}
and by introducing the symbols $v=v_\alpha^{(1)}$ and $w = v_\alpha^{(2)}/2$.
This leads to the equivalent equations:
\begin{align}
[\mathrm{i} \partial_t + \mathrm{i} v \partial_z + w \partial_z^2 + \omega_0 (g + g^\ast)] f = & 0,
\label{eqn:dispersive_eom_opt_rescaled}
\\
(\partial_t + \gamma_\beta + \mathrm{i} \Omega_0) g = & \mathrm{i} \Omega_0 |f|^2.
\label{eqn:dispersive_eom_ac_rescaled}
\end{align}
In the case of weak dispersion ($w \approx 0$), an approximate solution to
Eqs.~(\ref{eqn:dispersive_eom_opt_rescaled},\ref{eqn:dispersive_eom_ac_rescaled})
over a short length of waveguide ($z\approx0$) can be found via first order
perturbation theory.
To this end, we first make the ansatz
\begin{align}
f(z, t) = \exp[h(z, t) + \mathrm{i} j(z, t)],
\end{align}
with real-valued functions $h(z, t)$ and $j(z, t)$.
Inserting this into Eq.~\eqref{eqn:dispersive_eom_opt_rescaled} and
separating real and imaginary parts of this equation, we reach the system
\begin{align}
(\partial_t + \gamma_\beta + \mathrm{i} \Omega_0) g = & \mathrm{i} \Omega_0 \text{e}^{2 h},
\\
\partial_t h + [v + 2 w (\partial_z j)] \partial_z h = & - w \partial^2_z j,
\\
(\partial_t + v \partial_z) j + w [ (\partial_z j)^2 - (\partial_z h)^2
- \partial_z^2 h] = & \omega_0 (g + g^\ast).
\end{align}
We then write the exact function $h$ ($j$ and $g$ likewise) as an expansion:
\begin{align}
h(z, t) = h_0(z, t) + h_1(z, t) + h_2(z, t) + \ldots\ ,
\end{align}
where $h_0$ is the solution to the dispersionless problem, $h_n$ are the
$n^\text{th}$-order corrections corresponding to the powers $w^n$, and we
will truncate after the first order term.
As in the dispersionless solution, we take our previous result from
Sec.~\ref{sec:two-tone}:
\begin{align}
g_0(z, t) = & \mathrm{i} G_0 \exp\Big(- \mathrm{i} \Omega_0 \frac{vt - z}{v} \Big),
\\
h_0(z, t) = & \log|F_0| + \log\Big|1 - \mathrm{i} M
\text{e}^{\frac{\mathrm{i} \Omega_0 (v t - z)}{v}} \Big|,
\\
j_0(z, t) = & \frac{2 G_0 \omega_0 z}{v} \sin\Big(\Omega_0 \frac{vt - z}{v} \Big),
\end{align}
where
\begin{align}
F_0 = \frac{|Q| A_0 }{\sqrt{\mathcal{E}_\alpha \mathcal{E}_\beta}}
\end{align}
is the transformed optical mean amplitude and
we have omitted overall meaningless phase factors due to static deformation
in order to keep the equations concise.
The intensity modulation depth $M$ is assumed to be small and connected to
the acoustic amplitude $G_0 = \mathrm{i}|F_0|^2 M \Omega_0 / \gamma_\beta$.
Next, we calculate the first-order correction to $h$ and its effect on $g$;
the resulting correction to $j$ is not relevant within this context and
therefore not shown.
The correction to $h$ satisfies:
\begin{align}
\partial_t h_1 + [v + 2 w (\partial_z j_0)] \partial_z h_1 = & - w \partial_z^2 j_0,
\end{align}
with the boundary condition at $z=0$
\begin{align}
h_1(0, t) = 0.
\end{align}
\begin{figure}
\includegraphics[width=\columnwidth]{neg_dispersion_spectrum.eps}
\caption{
Evolution of the \mbox{(anti-)}Stokes orders for a normalized dispersion
of $\bar w = -0.15$ assuming that the waveguide is long compared to the
acoustic wave length and an optical excitation with a strong pump and a
weak Stokes side-band (see also Fig.~\ref{fig:stokes_evolution}).
}
\label{fig:neg_dispersion_evolution}
\end{figure}
This leads to the first order correction expressions (the intermediate steps
leading up to this result can be found in Appendix~\ref{appx:perturbation}):
\begin{align} \label{eq:h1corr}
h_1(z, t) = & \frac{w \omega_0 \Omega_0 G_0}{v^4} \Big[
4 vz \cos\Big(\Omega_0 \frac{vt-z}{v}\Big)
\\
\nonumber \quad &
+ \Omega_0 z^2 \sin\Big(\Omega_0 \frac{vt-z}{v}\Big)
\Big],
\\ \label{eq:g1corr}
g_1(z, t) \approx &
\frac{2 \mathrm{i} w \omega_0 \Omega_0^2 G_0 |F_0|^2}{v^4 \gamma_\beta}
\Big[ 4 v z + \mathrm{i} \Omega_0 z^2 \Big] \text{e}^{\mathrm{i} \Omega_0 \frac{z - vt}{v}},
\end{align}
where we have neglected any non-resonant excitation of the mechanical system.
This is justified for high mechanical quality factors.
This means that to first approximation and over a small distance $z$ the
acoustic amplitude is modified by a relative factor of
\begin{align}
\frac{g_1(z, t)}{g_0(z, t)} =
\frac{2 w \omega_0 \Omega_0^2 |F_0|^2 ( 4 v z + \mathrm{i} \Omega_0 z^2 )}{v^4 \gamma_\beta} .
\end{align}
This suggests to make the ansatz
\begin{align}
g(z, t) = & G(z) g_0(z, t),
\\
\text{with} \quad G(\Delta z) \approx &
\frac{2 w \omega_0 \Omega_0^2 |F_0|^2 ( 4 v \Delta z + \mathrm{i} \Omega_0 \Delta z^2 )}{v^4 \gamma_\beta}
\\
= & (a \Delta z + b \Delta z^2).
\end{align}
for infinitesimally small $\Delta z$.
We can then approximate a waveguide of finite length $z$ as a concatenation of
pieces of decreasing length $\Delta z$:
\begin{align}
G(n \Delta z) \approx & (1 + a \Delta z + b \Delta z^2) G[(n-1) \Delta z],
\\
\Rightarrow G(z) = & \lim_{z \rightarrow 0}
[1 + a\Delta z + b \Delta z^2]^{z/\Delta z}
\\
= & \exp\Big( \frac{8 w \omega_0 \Omega_0^2 |F_0|^2}{v^3 \gamma_\beta} z \Big).
\\
= & \exp\Big(
\frac{8 w \omega_0 \Omega_0^2 |Q|^2 |A_0|^2}{\mathcal{P}_\alpha \mathcal{E}_\beta v^2 \gamma_\beta} z \Big).
\label{eqn:dispersive_correction}
\end{align}
\begin{figure}
\includegraphics[width=\columnwidth]{pos_dispersion_spectrum.eps}
\caption{
Evolution of the \mbox{(anti-)}Stokes orders for a normalised dispersion
of $\bar w = 0.15$ assuming that the waveguide is long compared to the
acoustic wave length and an optical excitation with a strong pump and a
weak Stokes side-band (see also Fig.~\ref{fig:stokes_evolution}).
}
\label{fig:pos_dispersion_evolution}
\end{figure}
This can be expressed in terms of the normalised length variable $\zeta$ by
introducing a normalised dispersion parameter $\bar w$:
\begin{align}
G(\zeta) = & \exp(\bar w \zeta),
\\
\text{with} \quad \bar w = & \frac{4 \Omega_0 w}{v^2}
\sqrt{\frac{P_\text{P}}{P_\text{S}}} \ .
\end{align}
Fig.~\ref{fig:neg_dispersion_evolution} and
Fig.~\ref{fig:pos_dispersion_evolution}, show the evolution of the optical
spectrum along a waveguide for $\bar w = -0.15$ and $\bar w = +0.15$,
respectively.
This nicely illustrates that -- within the bounds of this first-order
perturbation theory, \mbox{i.e.\ } weak dispersion, weak optical intensity contrast --
the optical modulation pattern is compressed along $z$ for $w>0$ and
stagnates for $w<0$.
The underlying reason for this is that anomalous dispersion translates the
opto-acoustic phase modulation into an amplitude modulation with the same
sign as the optical fluctuation that drives the acoustic oscillation.
As a result, both the intensity modulation and the acoustic amplitude grow
along the waveguide.
In contrast, normal dispersion leads to an amplitude modulation that
diminishes optical intensity fluctuations along the waveguide.
\subsection{
Connection of intra-mode FBS to Raman scattering and soliton dynamics
}
\label{sec:raman-like}
The formalism presented so far is particularly well suited for the combination
of FBS with other optical nonlinearities, especially Raman-scattering and the
Kerr effect.
This is because the multi-Stokes optical envelope readily covers nonlinear
interference terms between various Stokes orders, which have to be explicitly
included in more conventional coupled mode theory~\cite{Wolff2015c} and would
clutter the equations for cascaded Brillouin scattering~\cite{Winful2016}.
The dynamics of solitons in optical fibres can be expressed by a retarded
nonlinear Schroedinger equation,~\cite{Gordon1986} where the instantaneous Kerr
effect and Raman scattering enter through the convolution of the optical
intensity with a (time-dependent) response function $r(t)$ as the convolution
kernel.
Our FBS equations of motion including this response function and for the
complete set of mechanical modes read:
\begin{widetext}
\begin{align}
\left[ \gamma_\alpha + \partial_t + v_\alpha^{(1)} \partial_z
+ \mathrm{i} \frac{v_\alpha^{(2)}}{2} \partial_z^2 \right] a(z, t)
= &
\Bigg[ \underbrace{\frac{2 \omega_0}{\mathrm{i} \mathcal{E}_\alpha}
\sum_\beta \Re\{Q_\beta \tilde b_\beta(z, t)\}}_{\text{FBS}}
+ \underbrace{\mathrm{i} \int_0^\infty \mathrm{d} t' \ r(t') |a(z, t-t')|^2}_{\text{Raman}} \Bigg]
a(z, t),
\\
\left[ \gamma_\beta + (\partial_t + \mathrm{i} \Omega_\beta)
+ v_\beta^{(1)} (\partial_z - \mathrm{i} q_0) \right] \tilde b_\beta(z, t)
= & - \frac{\mathrm{i} \Omega_\beta}{\mathcal{E}_\beta} Q_\beta^\ast |a(z, t)|^2.
\end{align}
The acoustic equations of motion can be formally solved using Green functions:
\begin{align}
\tilde b_\beta(z, t) = & - \frac{\mathrm{i} Q_\beta^\ast}{\mathcal{E}_\beta}
\int_0^\infty \mathrm{d} t' \
|a(z-v_\beta t', t-t')|^2
\mathrm{e}^{-(\gamma_\beta + \mathrm{i} \Omega_\beta - \mathrm{i} v^{(1)}_\beta q_0) t'}.
\end{align}
This allows us to combine the Raman kernel with the Brillouin response in a single
function $r_\text{eff}(z, t)$:
\begin{align}
\left[ \gamma_\alpha + \partial_t + v_\alpha^{(1)} \partial_z
+ \mathrm{i} \frac{v_\alpha^{(2)}}{2} \partial_z^2 \right] a(z, t)
= &
\mathrm{i} a(z, t) \int_0^\infty \mathrm{d} t' \ \int_0^\infty \mathrm{d} z' \
r_\text{eff}(z', t') |a(z-z', t-t')|^2,
\end{align}
\end{widetext}
where the effective response function is spatially nonlocal due to the
travelling nature of the sound waves in addition to the temporal non-locality
that is known from Raman scattering:
\begin{align}
r_\text{eff}(z, t) = & \delta(z) r(t) -
\sum_\beta
\frac{2 \omega_0 \Omega_\beta |Q_\beta|^2 }{\mathcal{E}_\alpha \mathcal{E}_\beta}
\mathrm{e}^{-\gamma_\beta t}
\\
\nonumber
& \quad \times
\sin\Big[\Big(\Omega_\beta - v^{(1)}_\beta q_0\Big) t\Big] \
\delta\Big(z - v^{(1)}_\beta t\Big) .
\end{align}
This spatial nonlocality is the main difference between FBS and Raman
scattering.
This nonlocality would be relevant for FBS involving a Dirac cone in the
acoustic dispersion relation at $q=0$.
However, in practice FBS occurs at acoustic band edges, where the
the acoustic group velocity $v^{(1)}_\beta$ so small that the phonon propagation
length (\mbox{i.e.\ } the spatial nonlocality) is but a few nanometers.
In this case, spatial nonlocality is negligible and intra-mode forward
Brillouin scattering can be regarded as a simple contribution to the
low-frequency tail of the Raman spectrum:
\begin{align}
r_\text{eff}(t) = & r(t) - \sum_\beta
\frac{2 \omega_0 \Omega_\beta |Q_\beta|^2 }{\mathcal{E}_\alpha \mathcal{E}_\beta}
\mathrm{e}^{-\gamma_\beta t}
\sin(\Omega_\beta t).
\end{align}
FBS can therefore be expected to contribute to the formation and the
self-frequency shift of nanosecond-scale solitons in
waveguides~\cite{Gordon1986}.
\comment{
\subsection{Approximate solitary (pulse train) solutions}
Solitary solutions propagate along the waveguide without changing their shape.
They can be therefore described by amplitudes that have a common wave-like
dependence on space and time:
\begin{align}
f(z, t) = & f(x),
\\
g(z,t) = & g(x),
\\
\text{with} \quad x(z, t) = & \frac{\Omega_0 t - \mu z}{2}.
\end{align}
Here, we assumed that the acoustic system is resonantly excited; the parameter
$\mu = \Omega_0 / \tilde v$, where $\tilde v$ is the unknown velocity of the
pulse train and can be expected close to the optical group velocity $v$.
This substitution leads to the coupled equations
\begin{align}
\frac{w \mu^2}{4} f'' + \frac{\mathrm{i} (v \mu - \Omega_0)}{2} f' - (g + g^\ast) f = & 0,
\\
g' + 2 \mathrm{i} g + \frac{2 \gamma}{\Omega_0} g = & \mathrm{i} |f|^2.
\end{align}
Exact solutions to this system of equations are not easy to find.
Instead, we assume that the acoustic response can be approximated as
a harmonic oscillation at the acoustic frequency:
\begin{align}
g(x) \approx g_0 \exp(2 \mathrm{i} x),
\end{align}
with $g_0 \in \mathbb{R}$.
This is a valid assumption given the exceedingly high quality factors (several
100 to several 1000) of the mechanical system.
We eliminate the first derivative by the ansatz:
\begin{align}
f(x) = \exp\Big[\frac{\mathrm{i} x (\Omega_0 - v\mu)}{w \mu^2} \Big] \tilde f(x).
\end{align}
This reduces our problem to the Mathieu equation
\begin{align}
& \tilde f'' + [A - 2 Q \cos(2 x)] \tilde f \ = \ 0,
\\
& Q \ = \ - \frac{4 g_0}{w\mu^2}, \quad \quad
A \ = \ \frac{2 (v\mu - \Omega_0)^2}{w^2 \mu^4}.
\end{align}
The obvious candidates for stable pulse train solutions are the Mathieu
functions that occur for a discrete set of specific real values of $A$ for
any given real $Q$~\cite{AbramowitzStegun}.
However, the Mathieu functions are periodic and real-valued, but either even
or odd, which means that the resulting $|f(x)|^2$ is an even function.
This means that the optical forces are $\pi/2$ out of phase with the mechanical
oscillation they are supposed to drive.
Therefore, the Mathieu functions do not provide solution to the problem.
Likewise, the associated Mathieu functions are unsuited for symmetry reasons.
We have to look for candidates within the zoo of non-periodic solutions to the
Mathieu equation.
Stable solutions to the Mathieu equation exist for certain $A$-dependent
intervals of $Q$.
The general solution for given stable $Q$ can be represented in terms of a
function $P(x)$ that is $\pi$-periodic, \mbox{i.e.\ } has the same periodic as the
acoustic amplitude~\cite{AbramowitzStegun}:
\begin{align}
\tilde f(x) = \alpha \text{e}^{\mathrm{i} \lambda x} P(x) + \beta \text{e}^{-\mathrm{i} \lambda x} P(-x).
\end{align}
Here, $\alpha$ and $\beta$ are complex constants.
The $Q$-dependent real $\lambda$ and the periodic function $P(x)$ are defined
through the eigenvalue problem:
\begin{align}
P''(x) - 2 \mathrm{i} \lambda P'(x) + 2 Q \cos(2 x) P(x) = (A + \lambda^2) P(x),
\label{eqn:Mathieu_periodic_def}
\end{align}
where integer multiples of $pi$ must be excluded for $\lambda$ (they correspond
to the already discussed Mathieu functions).
The substitution $x \rightarrow -x$ leads to the equation
\begin{align}
P''(x) + 2 \mathrm{i} \lambda P'(x) + 2 Q \cos(2 x) P(x) = (A + \lambda^2) P(x),
\end{align}
which is the complex conjugate of Eq.~\eqref{eqn:Mathieu_periodic_def}.
Thus, the periodic parts have the property
\begin{align}
P(-x) = P^\ast(x).
\end{align}
As a consequence, the optical intensity
\begin{align}
\nonumber
|f(x)|^2 = & |\alpha|^2 |P(x)|^2 + |\beta|^2 |P(-x)|^2
\\
& \quad + 2 \Re\{\alpha \beta^\ast P(x) P(-x)\}
\\
\nonumber
= & |\alpha|^2 |P^\ast(-x)|^2 + |\beta|^2 |P^\ast(x)|^2
\\
& \quad + 2 \Re\{\alpha \beta^\ast P(x) P(-x)\}
\end{align}
is an even function, which means that again the optical forces are $\pi/2$
out of phase with the mechanical motion they are supposed to drive.
Hence, stable pulse train solutions with harmonic acoustic oscillations
at the mechanical resonance are not possible.
In reality, $g(x)$ is never purely harmonic, but involves a static
displacement and other off-resonant excitations at harmonics of the acoustic
resonance.
The static displacement only adds to $A$ and is therefore irrelevant.
}
\section{Discussion}
\label{sec:discussion}
After the analysis of the previous sections we now combine the results
obtained in the different regimes and discuss their implications.
\subsection{Phase modulation versus intensity modulation}
We found in Sec.~\ref{sec:dispersionless}, that intra-mode FBS without optical
dispersion has a closed analytical solution in the form of a simple integral.
One key observation within this result is that the acoustic field does not
affect the optical intensity, \mbox{i.e.\ } neither the optical beat pattern nor the
acoustic amplitude vary along the waveguide.
Consequently, dispersionless intra-mode FBS results only in a phase modulation
of the optical signal.
Thus, the acoustic amplitude is constant throughout the waveguide and the
phase modulation grows linearly.
The situation changes fundamentally if the waveguide is dispersive.
The optical dispersion slowly shifts the various (anti-)Stokes orders out of
phase and thereby creates an FBS-related beating pattern, \mbox{i.e.\ } an optical
intensity modulation.
Depending on the sign of the dispersion constant, this new beating pattern
can have the opposite or the same sign as the intensity fluctuation that
drives the acoustic wave in the first place.
In the former case, the total optical intensity variations (and thus the
acoustic amplitude) diminish along the waveguide.
However, the entropy of the fluctuating pump cannot disappear, so the reduction
in the intensity modulation has to be countered by a phase modulation.
In the other case (additive beat patterns), optical intensity fluctuations (as well
as the acoustic amplitude) are amplified along the waveguide and both the
intensity and the phase modulation grow exponentially along the waveguide.
We note that, having used a perturbative approach to solve this problem,
these results are strictly only valid for weak dispersion and weak intensity
contrasts --- this approximation will therefore break down as the Stokes grows
to be comparable with the pump.
Nevertheless these results can be expected to remain qualitatively valid even
in the cases of strong dispersion and large intensity contrasts.
\subsection{Measurement of FBS}
\label{sec:measurement}
FBS occurs in various fields such as photonic crystal fibres or integrated
silicon waveguides.
These systems are typically studied with different experimental techniques
because of the very dissimilar loss regimes, power handling and waveguide
lengths.
Most fiber experiments capture the modulation of the optical wave with high
fidelity either by using an interferometric setup or by measuring in another
polarization.
They usually do not inject a dedicated Stokes seed and instead observe
scattering from phonons that are either thermally excited or generated by
the intensity noise of the pump laser.
The long interaction lengths and power handling in fiber optics allows for
complex interplay of FBS with other nonlinearities and often result in cascaded
dynamics.
Extreme examples of this are lasers that are passively mode-locked at the
acoustic resonance frequency~\cite{Kang2013,Stiller2013}.
It should be stressed that in the case of a straight dispersionless waveguide
and for strict intra-mode scattering the intensity variations at the end of a
waveguide do not differ from those at the front, because the beat of the pump
and Stokes lines and among the Stokes lines vanishes exactly.
This is the reason why many measurements
are either done in orthogonal polarizations or in interferometric setups that
are designed to eliminate the main pump
line~\cite{Elser2006,Dainese2006,Beugnot2007,Kang2008,Stiller2011,Zhong2015}.
Through the amplitude and phase modifications of the individual Stokes order
this leads to a beat signal, which can then be easily detected by a
photodiode and analyzed in the RF domain.
It should be noted that the relative amplitudes of the lines in the RF spectrum
can differ considerably from relative amplitudes in the actual optical
spectrum.
In contrast, experiments on FBS in integrated photonics often require a
pump-probe configuration with fairly strong Stokes seeds in order to lift the
spectral peaks above the detection threshold.
As a consequence, the measurements are usually not sensitive enough to
distinguish between ``stimulated'' (in the sense of ``self-amplifying'')
dynamics and an ``optically driven'' phase modulation.
Furthermore, these experiments may not require special ``phase-mixing''
elements (\mbox{e.g.\ } interferometric loops) to convert the pure phase modulation into
a partial intensity modulatoin, because (narrow-band) grating couplers
readily introduce phases and the waveguides are dispersive in themselves.
Quite often, the interaction lengths are too short for cascading to set in.
Without the need to capture cascading, the theoretical description is
often restricted to basic coupled-mode theory borrowed from backward SBS.
Only recently, an extension of the coupled-mode formulation to include the
first anti-Stokes line has led to the realization~\cite{Rakich2016} that
self-amplifying FBS is not generally expected in integrated waveguides.
Care should be taken when studying FBS in weakly dispersive integrated
waveguides with broad-band couplers (or even butt-coupling):
without the phase modification to the individual Stokes orders, this will
only result in a pure phase modulation that will go unnoticed if directly
analyzed with a photo diode.
\subsection{Stimulated versus Raman-like scattering}
\begin{figure}
\includegraphics[width=0.43\columnwidth]{ac_bs_conventional.eps}
\hspace{0.08\columnwidth}
\includegraphics[width=0.43\columnwidth]{ac_bs_dirac.eps}
\caption{
Schematic examples for acoustic dispersion relations with low acoustic
group velocity of the higher-order acoustic mode (red dot in left panel)
and with a Dirac point at $q=0$ leading to high acoustic group velocities
of some of the higher-order acoustic modes (red dot in right panel).
FBS in the conventional case with low acoustic group velocity can be
described as a low-frequency contribution to the local Raman response.
In contrast, FBS in the right-hand case manifests as a non-local effect
along the waveguide.
It should be noted that the question whether self-amplification can occur
is not related to this, but determined by the \emph{optical} dispersion
relation.
}
\label{fig:dirac_point}
\end{figure}
As we have mentioned in the introduction, FBS is known under different names,
especially as ``Raman-like'' or as ``forward stimulated'' scattering.
Each of them is justified, because they indicate features that may or may
not be dominant depending on the specific type of waveguide and the measurement
technique employed.
For example, in Sec.~\ref{sec:raman-like}, we have clearly shown that intra-mode
FBS is phenomenologically indistinguishable from a low-frequency tail of the
Raman spectrum if and only if the acoustic group velocity (\mbox{i.e.\ } the velocity at
which an acoustic pulse travels along the waveguide) is negligibly small.
This is usually the case in intra-mode FBS and inter-mode FBS between degenerate
optical modes (see Fig.~\ref{fig:dirac_point}).
In these systems the term ``Raman-like'' is well justified.
Large acoustic group velocities can occur for inter-mode FBS or if the acoustic
dispersion relation has a Dirac point at $q=0$.
The formation of Dirac points through the combination of symmetry-related and
accidental degeneracies at the $\Gamma$-point is well understood in the context
of photonic crystals~\cite{Huang2011} and similar effects are expected in
mechanical systems, as well.
In these two cases, FBS differs from Raman scattering in that it becomes
non-local over the length scale of the acoustic decay length
(approx. $50-100\,\mu$m), which can be very relevant in integrated
photonics~\cite{Wolff2015b}.
Both Brillouin scattering processes, forward and backward, are referred to as
\emph{Stimulated} Brillouin Scattering (SBS) in parts of the literature, while
some authors restrict this terminology to the backward scattering process, only.
The observation that FBS in the absence of dispersion does not result in any
amplification of the Stokes signal prompts the question:
should FBS be classified as a stimulated process?
The answer to this depends on one's definition of what a stimulated process
entails.
\comment{
A stimulated process may be defined as one in which the probability of a
transition between states --- in this case a phonon generated by the transition
between a high-energy photon state to a lower energy photon state --- is
increased in proportion to the population of the emitted quantum.
However there are also significant differences between Brillouin transitions,
which involve bosons, and the transitions in a conventional laser, which
involve fermions.
In particular, Brillouin scattering can be described in a purely classical
form, and Planck's constant does not appear in the rate equations.
In the end, the classification of such classical systems, which include not
only (backward) SBS, but also Raman and X-ray lasers, as ``stimulated'' has
become a matter of convention and is widely accepted.
}
In the context of backward Brillouin scattering and Raman scattering, the term
``stimulated'' scattering has been used since the early
papers~\cite{Bloembergen1965,Tang1966} to denote the positive feedback that results in
self-amplification of both the optical and acoustic fields.
Using this definition a clear distinction is drawn between the spontaneous
processes, in which the pump signal is too weak to cause a runaway effect so
strong that thermal excitations are amplified to levels comparable to the pump
power, and the stimulated process, in which significant self-amplification
occurs and the backscattered Stokes signal grows exponentially along the
length of the waveguide.
This definition applies to FBS provided that the waveguide exhibits an
appropriate optical dispersion, as we have shown in Sec.~\ref{sec:dispersive},
but does not if this condition is not met.
Adopting this pragmatic approach, it seems fair to call FBS a stimulated
process, but to keep in mind that whether or not the effect is ``stimulated''
will always depend on the details of the waveguide and experiment.
While the presence of dispersion certainly allows forward amplification, our
analysis shows that for the zero dispersion case, or more physically, for
distances shorter than the dispersion length, there is no significant
self-amplification at powers where amplification would certainly be seen in
the backwards case.
This suggests that while FSBS is a defensible terminology, perhaps FBS is to
be preferred in generality.
\subsection{Similarities to cavity opto-mechanics and discrete diffraction }
The problem of FBS bears similarities to other fields of optics apart from
backward SBS and Raman scattering.
One example is the close similarity between FBS in dispersive waveguides and
side-band heating and cooling in cavity opto-acoustics~\cite{Bahl2012}.
In both cases the opto-acoustic interaction can either suppress acoustic
vibrations or lead to an exponential growth.
The particular outcome is determined by the imbalance of the optical
density of states on the red and the blue side of the optical pump beam.
Such a connection is somewhat expected given the intimate connection between
both fields~\cite{VanLaer2016}.
However, there is one fundamental difference:
in cavity opto-acoustics, the acoustic oscillations are amplified or suppressed
over time, whereas in FBS they are modified along the waveguide.
While cavity opto-acoustics and FBS are clearly based on the same physics, the
intimate connection between FBS and discrete diffraction in waveguide arrays
is on a formal rather than a physical level.
The various Stokes orders constitute channels of light propagation similar to
an array of waveguides and a transversal acoustic mode provides a coupling
between these channels, whose magnitude is proportional to the acoustic
amplitude.
It is a fortunate coincidence that in the absence of optical dispersion the
acoustic amplitude is constant along the waveguide, \mbox{i.e.\ } that the process of FBS
is formally equivalent to an array of parallel waveguides.
As a result, the evolution of the optical intensity through the Stokes orders
along the waveguide is governed by the same equations and the solution of the
two-tone problem (Sec.~\ref{sec:two-tone}) is identical to the diffraction of
light in an array of parallel waveguides.
One of the intriguing features of this analogy is that the inter-channel
coupling coefficient (or alternatively the effective length $\zeta$ of the
waveguide array) can be tuned with extreme flexibility via the optical pump
power.
This similarity raises the possibility for FBS-based quantum-walk experiments.
\subsection{Summary and conclusions}
We have presented here a new and rigorous formulation of the forward Brillouin
scattering process, which implicitly includes all coupling amongst the comb of
cascaded orders.
Because the differences between the quantum and classical formulations are
minimal, we have chosen the more general case for our derivation of the
equations of motion, beginning with the full optomechanical Hamiltonian of the
system.
This formalism can therefore be applied to problems of interest in the limit of
low photon/phonon numbers, and would be especially useful for the prediction and
interpretation of single phonon scattering experiments.
The established description of FSB is derived from the theory of the backward
SBS process in fibres.
In this, the pump and Stokes fields are modelled as independent, spectrally
extremely narrow fields.
This is partially motivated by the strong dependence of the acoustic frequency
on the optical wave length and the narrow acoustic resonances.
Cascading effects can occur in backward SBS~\cite{Buettner2014a}, but are not a
very common feature.
Instead, exponential growth of the optical and acoustic amplitudes along the
waveguide is the most common manifestation.
In contrast, FBS is mostly independent of the optical wavelength, it has a
strong tendency towards cascading and requires the help of optical dispersion
to have any impact on the optical and acoustic intensities.
Thus, the multi-mode ansatz that is so successful in BSBS is not ideal for FBS.
We find that a description coupling only one broad-band optical field to an
acoustic field is much better suited.
In particular for the dispersionless case we find closed-form solutions, in
which the amplitude of the cascaded orders is given simply by a series of
Bessel functions.
The simplicity of this result follows from one of the main insights arising
from this formalism: that the acoustic field exerts, in the absence of
dispersion, a pure phase modulation on the optical field.
This implies that for dispersionless waveguides FBS is not capable of
amplifying the intensity beat between a pump and a Stokes signal.
\section{Acknowledgements}
We acknowledge financial support of the Australian Research
Council via the ARC Center of Excellence CUDOS (CE110001018) and
its Laureate Fellowship (Prof. Eggleton, FL120100029) program.
Furthermore, we are deeply indebted to Dr Mark Craddock for insightful
discussions.
\begin{appendix}
\comment{
\section{Effect of optical commutator on mechanical system}
\label{appx:casimir}
The non-trivial commutator $[\hat \psi_{\alpha}^\dagger, \hat \psi_\alpha ]$
in Eq.~\eqref{eqn:opt_commutator} is a constant and can be incorporated into
the mechanical part of the Hamiltonian.
For each acoustic mode $\beta$, this can be expressed in reciprocal space:
\begin{align}
\mathcal{H}_\beta = &
\int_{-\infty}^\infty \mathrm{d} q \ \hbar \Omega_{\beta, q}
b^\dagger_{\beta, q} b_{\beta, q}
+ A_{\beta, q} (b^\dagger_{\beta, q} + b_{\beta, q}),
\intertext{with}
A_{\beta, q} = & \sum_{\alpha} \Gamma_{\alpha \alpha \beta}
[\hat \psi_\alpha^\dagger, \hat \psi_\alpha] \delta(q) = A_\beta \delta(q),
\label{eqn:casimir_coeff}
\end{align}
where we assumed that $A_\beta$ is real-valued.
This can be achieved by an appropriate phase choice of the acoustic eigenmode,
even though some of the coupling constants $\Gamma_{\alpha \alpha \beta}$
might have non-trivial phases.
\begin{align}
\mathcal{H}_\beta(q=0) = &
\frac{\hbar \Omega_{\beta, 0}}{4} \Big[
( b_{\beta, 0} +b^\dagger_{\beta, 0} + x_0)^2
\\
\nonumber
& \quad -
( b_{\beta, 0} - b^\dagger_{\beta, 0})^2
- 2 - x_0^2 \Big].
\end{align}
The change to the vacuum energy is of no consequence and the real-valued
parameter
\begin{align}
x_0 = & \frac{A_\beta}{2 \hbar \Omega_{\beta, 0}}
\end{align}
can be interpreted as a static offsets of the generalized position of the
harmonic oscillator, \mbox{i.e.\ } as a static deformation of the waveguide according to
the acoustic mode $\beta$ at $q=0$.
This can be absorbed into the phononic ladder operators by substituting:
\begin{align}
b_{\beta, 0} \longrightarrow b_{\beta, 0} + \frac{x_0}{2}.
\end{align}
}
\section{Derivation of first-order perturbative dispersion terms}
\label{appx:perturbation}
Here we derive the corrections to the optical and acoustic fields
for the dispersive dynamics shown in Eqs.~\eqref{eq:h1corr}
and~\eqref{eq:g1corr}.
We start with the dispersion-free stationary two-tone solution:
\begin{align}
h_0(z, t) \approx & \log | F_0 |,
\\
g_0(z, t) = & \mathrm{i} G_0 \exp\Big(- \mathrm{i} \Omega_0 \frac{vt - z}{v} \Big),
\\
j_0(z, t) = & \frac{2 G_0 \omega_0 z}{v}
\sin\Big(\Omega_0 \frac{vt - z}{v} \Big),
\end{align}
with real positive $G_0$.
The correction to $p$ satisfies:
\begin{align}
\partial_t h_1 + [v + 2 w (\partial_z j_0)] \partial_z p = &
- w \partial_z^2 j_0,
\end{align}
with the boundary condition at $z=0$
\begin{align}
h_1(0, t) = 0.
\end{align}
The derivatives of the unperturbed phase function are:
\begin{align}
\partial_z j_0 = & \frac{G_0\omega_0}{v} \sin\Big(\Omega_0 \frac{vt - z}{v}\Big)
- \frac{G_0 \omega_0 \Omega_0 z}{v^2} \cos\Big(\Omega_0 \frac{vt - z}{v}\Big),
\\
\partial_z^2 j_0 = & -\frac{4G_0\omega_0\Omega_0}{v^2} \cos\Big( \Omega_0 \frac{vt - z}{v} \Big)
\\
\nonumber
& \quad - \frac{2G_0 \omega_0 \Omega_0^2 z}{v^3} \sin\Big(\Omega_0\frac{vt-z}{v}\Big).
\end{align}
The dispersive correction to the effective group velocity is of order $w$ and
therefore introduces corrections of order $w$.
We may therefore approximate:
\begin{align}
\nonumber
(\partial_t + v \partial_z) h_1 = &
\frac{2 w \omega_0 \Omega_0 G_0}{v^3}
\Big[ 2 v \cos\Big( \Omega_0 \frac{vt - z}{v} \Big)
\\
& \quad
+ \Omega_0 z \sin\Big(\Omega_0\frac{vt-z}{v}\Big)\Big].
\end{align}
Through the substitution
\begin{align}
y = & vt + z, & s = & t - \frac{z}{v};
\\
z = & \frac{y - vs}{2}, & t = & \frac{y + vs}{2v},
\end{align}
this can be reduced to the ordinary differential equation
\begin{align}
v \partial_y = &
\frac{2 w \omega_0 \Omega_0 G_0}{v^3}
\Big[ 2 v \cos \Omega_0 s
+ \frac{\Omega_0 (y - vs)}{2} \sin\Omega_0 s\Big].
\end{align}
with the explicit
solution:
\begin{widetext}
\begin{align}
h_1(y, s) = & \frac{w \omega_0 \Omega_0 G_0}{v^4} \int_{0}^{y}
\mathrm{d} y' \
\Big[ 2 v \cos (\Omega_0 s)
+ \Omega_0 \frac{y' - vs}{2} \sin(\Omega_0 s) \Big] + C(s),
\\
= & \frac{w \omega_0 \Omega_0 G_0}{v^4} \Big[ 2 v y \cos (\Omega_0 s)
+ \frac{\Omega_0}{4} (y^2 - 2 y vs) \sin(\Omega_0 s) \Big] + C(s),
\end{align}
where $C(s)$ is defined through the boundary condition expressed in the
transformed coordinates \mbox{$h_1(vs, s) = 0$}.
Therefore, the first order correction to $p$ is:
\begin{align}
h_1(y, s) = & \frac{w \omega_0 \Omega_0 G_0}{v^4} \Big[
2 v(y - vs) \cos(\Omega_0 s)
+ \frac{\Omega_0}{4} (y - vs)^2 \sin(\Omega_0 s)
\Big];
\\
h_1(z, t) = & \frac{w \omega_0 \Omega_0 G_0}{v^4} \Big[
4 vz \cos\Big(\Omega_0 \frac{vt-z}{v}\Big)
+ \Omega_0 z^2 \sin\Big(\Omega_0 \frac{vt-z}{v}\Big)
\Big].
\end{align}
This recovers Eq.~\eqref{eq:h1corr} in the main text.
The remaining step is to find the correction to the acoustic problem:
\begin{align}
g_0(z, t) + g_1(z, t) = & \mathrm{i} \Omega_0 \int_0^\infty \mathrm{d} t' \
\text{e}^{-\gamma t' - \mathrm{i} \Omega_0 t'} \big|A(t-t'-z/v)\big|^2 \exp[2 h_1(z, t-t')]
\\
\approx & \mathrm{i} \Omega_0 \int_0^\infty \mathrm{d} t' \
\text{e}^{-\gamma t' - \mathrm{i} \Omega_0 t'} \big|A(t-t'-z/v)\big|^2 [1 + 2 h_1(z, t-t')]
\\
= & g_0(z, t) + \underbrace{2 \Omega_0 \mathrm{i} \int_0^\infty \mathrm{d} t' \
\text{e}^{-\gamma t' - \mathrm{i} \Omega_0 t'} \big|A(t-t'-z/v)\big|^2 h_1(z,t-t')}_{=g_1(z, t)}.
\end{align}
Assuming that the intensity of the incident optical signal varies only weakly
(weak incident Stokes seed, \mbox{e.g.\ } due to thermal fluctuations), we can replace
\mbox{$|F(t)|^2 \approx |F_0|^2$} in the second term:
\begin{align}
g_1(z, t) \approx & \frac{2 \mathrm{i} w \omega_0 \Omega_0^2 G_0 |F_0|^2}{v^4}
\int_0^\infty \mathrm{d} t' \ \text{e}^{-\gamma t' - \mathrm{i} \Omega_0 t'}
\Big[ 4 vz \cos\Big(\Omega_0 \frac{vt-vt'-z}{v}\Big)
+ \Omega_0 z^2 \sin\Big(\Omega_0 \frac{vt-vt'-z}{v}\Big)
\Big]
\\
\approx &
\frac{2 \mathrm{i} w \omega_0 \Omega_0^2 G_0 |F_0|^2}{v^4 \gamma}
\Big[ 4 v z + \mathrm{i} \Omega_0 z^2 \Big] \exp\Big(\mathrm{i} \Omega_0 \frac{z - vt}{v}\Big)
\end{align}
where only retained the resonant contribution to the convolution.
This is the second expression required in the main text.
\end{widetext}
\end{appendix}
| 63ce2a033043234b2d56d7eeb74ad948f22f1478 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction \label{sec:intro}}
\IEEEPARstart{A}{utomatic} speech recognition, music classification, audio
indexing, and to a less visible extent biometric voice authentication, have now
achieved some degree of commercial success in mass markets. A vast proportion
of these applications are supported by PC platforms, cloud computing and
powerful smartphones. Nevertheless, a new market is quickly emerging in the
domain of the {\em Internet of Things} (IoT) \cite{gubbi2013internet}. In this market, Automatic
Environmental Sound Recognition (AESR) \cite{chachada2014environmental}
has a significant potential to create value, e.g., for security
or home safety applications \cite{istrate2006information,vacher2010challenges,sitte2007non}.
However, in this context, industrial reality imposes hard constraints on the
available computing power. Indeed, IoT devices can be broadly divided into
two classes: (a) devices which are perceived as ``doing only one thing'', thus
requiring the use of low cost processors to hit a price point that users are
willing to pay for what the device does; (b) embedded devices where sound
recognition comes as a bolt-on to add value to an existing product, for
example adding audio analytic capabilities to a consumer grade camera or to a
set-top box, thus requiring the algorithm to fit into the device's existing
design and price points. These two cases rule out the use of higher end processors.
Generally speaking, the following features jointly define the financial cost of
a processor and the level of constraint imposed on embedded computing: a) the {\em clock
speed} is related to energy consumption; b) the {\em instruction set} is related to
chip size and manufacturing costs, but in some cases includes special
instruction sets to parallelise more operations into a single clock cycle;
c) the {\em architecture} defines, e.g., the number of registers, number of cores,
presence/absence of a Floating Point Unit (FPU), a Graphical Processing Unit
(GPU) and/or a Digital Signal Processing (DSP) unit. These features affect applications
in the obvious way of defining an upper limit on the number and the type of
operations which can be executed in a given amount of time, thus ruling
the possibility to achieve real time audio processing at the proportion of
processor load allocated to the AESR application. Finally, {\em onboard memory size} is
an important factor related to processor cost, as it affects
both the computational performance, where repetitive operations can be cached
to trade speed against memory, and the scalability of an algorithm, by imposing
upper limits on the number of model parameters that can be stored and manipulated.
If trying to work within these limitations, and given that most IoT embedded
devices allow Internet connectivity, it could be argued that cloud computing can
solve the computational power constraints by abstracting the computing platform
and making it virtually as powerful as needed. However, a number of additional
design considerations may rule out the use of cloud computing for
AESR applications: the {\em latency} introduced by cloud communications can be
a problem for, e.g., time critical security applications \cite{bonomi2014fog}; regarding {\em
Quality of Service (QoS)}, network interruptions introduce an extra point of
failure into the system; regarding {\em privacy} and {\em bandwidth
consumption}, respectively, sending alerts rather than streaming audio or
acoustic features out of the sound recognition device both rules out any
possibility of eavesdropping~\cite{medaglia2010overview} and requires less
bandwidth.
Thus, the reality of embedded industrial applications is that at the price
points acceptable in the marketplace, IoT devices will most likely be devoid of
an FPU, will operate in the hundreds of Megahertz clock speed range, won't offer
onboard DSP or specialised instruction sets, and may prefer to run AESR on board
rather than in the cloud. At the same time, most AESR performance evaluations
are still produced with limited interest for the computational cost involved:
many experimental results are obtained with floating point arithmetics on
powerful computing platforms, and under the assumption that sufficient
computing power will sooner or later become available for the algorithm to be
practicable in the context of commercial applications.
This article aims at crossing this divide, by comparing the performance
of three AESR algorithms as a function of their computational cost. In other
terms, it seeks which AESR algorithm is able to make the most of a
limited amount of computational power. The rest of the paper is organised as follows:
Section~\ref{sub:prior} reviews prior art in AESR. Section~\ref{sec:cost}
defines the measure of computational cost used in this study, and
applies it to a range of popular machine learning algorithms.
Section~\ref{sec:perf} defines the other axis of this evaluation, which is the
metric used to measure classification performance. The data sets and methodology
are detailed in Section~\ref{sec:methodology}, while results are discussed in
Section~\ref{sec:results}.
\section{Background \label{sub:prior}}
Audio or sound recognition research has primarily been focused on
speech and music signals. However in the past decade, the problem of classifying
and identifying environmental sounds has received more attention \cite{chachada2014environmental}.
AESR has applications in content-based search \cite{virtanen2007probabilistic}, robotics
\cite{yamakawa2011environmental}, security \cite{istrate2006information} and a host of other tasks.
Recognition of environmental sounds is a part of the much broader field of computational
auditory scene analysis (CASA) \cite{wang2006computational}. We are interested in the problem
of identifying the presence of non-speech, non-music sounds or \emph{environmental sounds}
in a given audio recording. Environmental sound recognition involves the study of two closely related
problems. Acoustic Scene Classification (ASC) involves selecting an appropriate set of labels
that encompasses all the sound events which occur in a recording \cite{stowell2015detection}, e.g., ``crowded train station''. With regards to machine
learning, ASC can be seen as a sequence classification task where a set
of labels is assigned to a variable length input sequence. A related problem is the problem
of Acoustic Event Detection (AED). In addition to identifying the correct label, AED aims to
segment the audio recording into regions where the label occurs \cite{stowell2015detection}, e.g., ``stationmaster blowing whistle'' or ``doors closing''. Again with regards to machine
learning, AED involves identifying the correct labels and the correct \emph{alignment}
of labels with the recording. In this study, we compare Gaussian mixture models (GMMs),
support vector machines (SVMs) and deep neural networks (DNNs) in terms of their
performance on an AED task as a function of their computational cost.
GMMs are generative probabilistic models
and have been extensively applied to many problems in speech
recognition~\cite{rabiner1993fundamentals}. In a number of cases, GMMs have
therefore been considered a sensible starting point for research on the
classification of environmental sounds. For example, they have been used to
detect gunshot sounds~\cite{clavel2005events} and to discriminate between
gunshots and screams~\cite{valenzise2007scream}. Port\^{e}lo et
al.~\cite{portelo2009non} have used Gaussian Mixtures as part of a hidden
Markov model (HMM) approach to detecting sirens in audio. Elsewhere,
Radhakrishnan et al.~\cite{radhakrishnan2005audio} have used GMMs when
attempting to develop a model for recognising new `outlier' sounds against a
GMM trained to model background sounds. Principi et al.~\cite{Principi2015}
proposed another novelty detection system, with an embedded implementation on a
Beagle Board~\cite{Bonfigli2014}. Atrey et al.~\cite{atrey2006audio} have used
GMMs to classify audio frames into foreground or background sounds, then into
vocal or nonvocal sounds and finally into `excited' events (e.g. shouting,
running) or `normal' events (e.g. talking, walking).
SVMs are non-parametric classifiers
that can generalise well given a limited training set and have well defined
generalisation bounds~\cite{burges1998tutorial}. Port\^{e}lo et al.
\cite{portelo2009non} have experimented with an SVM when attempting to detect
events in film audio. They found it to perform well on some sounds
(aeroplanes, helicopters) but poorly on others (sirens, cats). By adding new
features and applying PCA they found better performance on siren sounds.
Moncrieff et al. \cite{moncrieff2001detecting} have also experimented with SVMs
when attempting to classify sound events in film audio.
Chen et al. \cite{chen2006mixed} have used SVMs to classify audio data into
music, speech, environmental sounds, speech mixed with music and environmental
sounds mixed with music. Rouas et al.
\cite{rouas2006audio} have presented a comparison of SVMs and GMMs when
attempting to detect shouting in public transport vehicles. Testing on audio
recorded by actors on a train, they find that the SVM achieves a lower
false alarm rate whilst the GMM seems better at making positive identifications.
More recently, Deep Learning, i.e., artificial neural networks with more
than one intermediate hidden layer, have gained much popularity and have
achieved impressive results on several machine learning
tasks~\cite{lecun2015deep}. However, they do not seem to have been extensively
applied to AESR so far. Toyoda et al.~\cite{toyoda2004environmental} have used a
neural network to classify up to 45~different environmental sounds. They
achieved high recognition accuracies, though it should be mentioned that both
the training and test data sets used in their study may appear quite small in
comparison to the data sets used nowadays as benchmarks for speech recognition
and computer vision research. Baritelli et al.~\cite{beritelli2008pattern} used
a DNN to classify 10~different environmental sounds. The authors found that the
proposed system, which classified MFCCs directly, performed similarly to other
AESR systems which were requiring more complex feature extraction.
In~\cite{vidalsound}, the authors used neural networks to estimate the presence
or absence of people in a given recording, as well as the number of people and
their proximity to the recording device. Gencoglu et al.
\cite{gencoglu2014recognition} applied a deep neural network to an
acoustic event detection task for classifying $61$ different classes of sounds.
Their results showed this classifier to outperform a
conventional GMM+HMM classification system.
Recently, the problem of AESR has been receiving more attention. Challenges like
CLEAR \cite{temko2006clear} and DCASE \cite{stowell2015detection} have been proposed
in order to benchmark the performance of different approaches and to stimulate
research. But, if drawing a parallel between the evolution of AESR and
the evolution of speech recognition, the CLEAR and DCASE data sets
could be thought of as being at the stage of the TIMIT era \cite{zue1990speech}:
although the TIMIT data set was widely used for many years, it was limited in size
to about 5.4 hours of speech data. The subsequent growth in the volume of
available speech evaluation data sets has significantly contributed to
bringing speech recognition to the levels of performance known today,
where modern speech recognition systems are being trained over
1700~hours of data \cite{senior2015context}.
AED research is following a similar trend where the domain is now facing a need to collect
larger data sets \cite{foster2015chime,Salamon:UrbanSound:ACMMM:14} in order to make larger scale evaluations
possible and to foster further technical progress.
Many studies explore
feature learning to derive useful acoustic features that can be used to classify
acoustic events \cite{salamon2015feature,salamon2015unsupervised}. A limited number of publications
apply k-Nearest Neighbours (k-NNs) \cite{sawhney1997situational,lukowicz2003soundbutton}
or Random Forests \cite{stowell2014automatic} to AESR. However, these models have not been
included in our comparative study, which is restricted to GMM, SVM and DNN
models: these three types of models can be considered mainstream in terms of
having been applied more extensively to a larger range of audio classification
tasks which include speech, speaker and music classification and audio event
detection. In preliminary experiments,
we observed that k-NN classifiers were clearly outperformed by the classifiers
included in this study.
\section{Computational cost \label{sec:cost}}
\subsection{Measure of computational cost used in this article}
The fundamental question studied in this paper is that of comparing various
machine learning algorithms {\em as a function of their computational cost}. A
corollary to this question is to compare the rate of performance degradation
that occurs when down-scaling various AESR algorithms by a comparable amount of
computing operations: a given class of algorithms might be more resilient to
down-scaling than another. While computational cost is related to computational
complexity, both notions are distinct: computational cost simply counts the
number of operations at a given model dimension, whereas computational
complexity expresses the mathematical law according to which the computational
cost scales up with the dimension of the input feature space.
Computational cost is studied here at the sound recognition stage,
known as acoustic decoding. Unless specifically required by the
application, it is indeed fairly rare to see the training stage of machine
learning being realised directly on-board an embedded device. In most cases, it is
possible to design the training as an offline process realised on powerful
scientific computing platforms equipped with FPUs and GPUs. Offline training
delivers a floating point model which is then quantised according to the
needs of fixed point decoding in the embedded system \cite{smith1997scientist}.
It is generally accepted that the quantisation error introduced by this
operation is unlikely to affect the acoustic modelling performance \cite{gupta2015deep}.
The computational cost estimate used in this study simply counts four~types of
basic operations: addition, comparison, multiplication, and lookup table
retrieval (LUT).
{\em Multiply-add} operations are commonly found in dot products, matrix
multiplications and FFTs. Correlation operations, linear filtering or the
Mahalanobis distance, at the heart of many recognition algorithms, rely solely
on multiply-adds. Fixed point precision is fairly straightforward to manage
over this type of operation. {\em Divisions}, on the other hand, and in
particular matrix inversions, might be more difficult to manage. Regarding
matrix inversion, it can happen that quantisation errors add up and make the
matrix inversion algorithm unstable \cite{parhami2009computer}. In many cases, it is possible to
pre-compute the inversion of key parameters (e.g., variances), possibly offline
and in floating point before applying fixed-point quantisation. {\em Non-linear
operations} such as logarithms, exponentials, cosines, $N^{th}$~roots etc.
are required by some algorithms. Two approaches can commonly be taken to
transform non-linear functions into a series of multiply-adds:
either Taylor series expansion, or look-up tables (LUTs). LUTs are more memory
consuming than Taylor series expansions, but may require a lower number of
instructions to achieve the desired precision. Generally speaking, it is
important to be aware of the presence of a non-linearity in a particular
algorithm, before assessing its cost: algorithms which rely on a majority of
multiply-adds are more desirable than algorithms relying more heavily on
non-linearities.
For all the estimates presented below, it is assumed that these four~types
of operations have an equal cost. While this assumption would be true for a
majority of processors on the market, it may underestimate the cost of
non-linear functions if interpolated LUTs or Taylor series are used instead of
simpler direct LUT lookups. In addition to the 4 operations considered here,
there are additional costs incurred with other operations (like data handling)
in the processor. However, we use the simplifying assumption that the
considered operations are running on the same core, thus minimizing data handling overhead.
As a case study, let us assume an attempt at running the K-nearest neighbours
algorithm on a Cortex-M4 processor running at a clock speed of 80MHz and
with 256kB of onboard memory. This is a very common hardware configuration for
consumer electronic devices such as video cameras. Assuming that 20\% of the processor
activity is needed for general system tasks means that
a maximum of 64 million of multiply or adds per second can be used for sound
recognition ($80\mbox{MHz} \times 80\% = 64\mbox{MHz}$).
Assuming sound sampled at 16kHz and audio feature extracted at, e.g., a window
shift of 256 samples, equivalent to 62.5 frames per second, means that the AESR
algorithm should use no more than 1,024,000 multiply or adds per analysis
frame (64MHz / 62.5Hz = 1\,024K instructions).
From there, assuming a K-Nearest neighbour algorithm with a 40 dimensional
observation vector $x$, mean vector $\mu$ and a Mahalanobis distance $( x - \mu ) \sigma^{-1} (x -
\mu)$ with a diagonal covariance matrix $\sigma$, there would be one~subtraction
and two~multiplications per dimension for the
Mahalanobis distance calculation, plus one~operation for the accumulation across
each dimension: four~operations in total, thus entailing a maximum of 1,024,000
instructions / 4~operations / 40~dimensions = 6\,400 nearest neighbours on this
platform to achieve real-time recognition.
However, assuming that each nearest neighbour is a 40-dimensional vector of
32\,bits/2\,Bytes values, the memory requirement to store the model would be
512kB, which is double the 256kB available on the considered platform. Again
assuming that the code size occupies 20\% of the memory, leaving 80\% of the
memory for model storage, a maximum of $256\mbox{kB} \times 80\% \,/\,
40\,\mbox{features} \,/\, 2\mbox{Bytes} = 2\,560$ nearest neighbours only
which could be used as the acoustic model.
Depending on the choice of algorithm, memory is not always the limiting factor.
In the case of highly non-linear algorithms, the computational cost of achieving
real-time audio recognition might outweigh the model storage requirements.
While only the final integration can tell if the desired computational load and
precision are met, estimates can be obtained as illustrated in this
case study in order to forecast if an algorithm will fit on a particular
platform.
\begin{table*}[t!]
\newcommand\T{\rule{0pt}{2.6ex}}
\newcommand\B{\rule[-1.2ex]{0pt}{0pt}}
\renewcommand{\arraystretch}{1.3}
\begin{center}
\footnotesize
\begin{tabular}{| l | c | c | c | c |}
\hline
\textbf{Feature} & \textbf{Addition} & \textbf{Multiplication} & \textbf{
nonlinearity LUT lookup} \\
\hline
\hline
GMM & $2(M \cdot (D+1) + M)$ & $2M \cdot 2D$ & $M$ \\
SVM - Linear & $\lambda D + \lambda + 1$ & $\lambda \cdot D$ & 0 \\
SVM - Polynomial & $\lambda D +\lambda + 1$ & $\lambda(D+d)$ & 0 \\
SVM - Radial Basis Function & $2\lambda D + \lambda + 1$ & $\lambda(D+2)$ & $\lambda$ \\
SVM - Sigmoid & $\lambda D + \lambda +1$ & $\lambda(D+1)$ & $\lambda$ \\
DNN - Sigmoid & $H \cdot (1+D+L+(L-1)H)+1$ & $H \cdot (1+D+(L-1)H)$ &
$L \cdot H + 1$\\
DNN - ReLU & $H \cdot (1+D+L+(L-1)H)+1$ & $H \cdot (1+D+(L-1)H)$ &
$L \cdot H + 1$\\
RNN - Tanh & $H \cdot
(2+D+H+2(L-1)(H+1))+1$ & $H \cdot
(1+D+H+2(L-1)H)$ & $L \cdot H + 1$\\
\hline
\end{tabular}
\small
\caption{Computational cost per frame of each compared AESR algorithm. $D$ is
the dimensionality of the feature vector, $M$ is the number of Gaussian
mixtures for a GMM. $\lambda$ is the number of support vectors for a SVM, $d$
is the degree of a polynomial kernel. For the neural networks: $H$ is the
number of hidden units in each layer and $L$ is the number of layers.
\label{num_ops}
}
\end{center}
\end{table*}
\subsection{Computational cost estimates for the compared algorithms}
\label{sub:cost}
Most AED systems use a common 3-step pipeline \cite{stowell2015detection}. The audio
recording is first converted into a time-frequency representation, usually
by applying the short-time Fourier transform (STFT)
to overlapping windows. This is followed by feature extraction from
the time-frequency representation. Typically, Mel Frequency Cepstral
Coefficients (MFCCs) have been used as standard acoustic features since the seventies, due to favourable
properties when it comes to computing distances between sounds \cite{mermelstein1976}.
Then the acoustic features are input into an acoustic model to obtain a
classification score, which in some cases is analogous to a posterior probability.
The classification score is then thresholded in order to obtain a classification decision.
Below, we provide an estimate of the computational cost for each of these operations.
\textbf{Feature extraction -}
Engineering the right feature space for AESR is essential,
since the definition of the feature space affects the separability of the
classes of acoustic data. Feature extraction from sound recordings thus forms
an additional step in the classification pipeline, which contributes to the
overall computational cost. In this study, the GMMs, SVMs and DNNs are trained
on a set of input features that are
typically used for audio and speech processing (Section \ref{sub:feature_extraction}). Although
recent studies demonstrate that neural networks can be trained to jointly learn
the features and the classifier \cite{lecun2015deep}, we have found that this method is impractical for most
AESR problems where the amount of labelled data for training and testing is limited. Additionally,
by training the different algorithms on the same set of features, we study the performance
of the various classifiers as a function of computation cost, without having to account
for the cost of extracting each type of feature from the raw audio waveform\footnote{For example in preliminary experiments, we observed that DNNs were able to achieve the same performance with log-spectrogram inputs as speech features (Section \ref{sub:feature_extraction}). However, in this case the GMMs and SVMs performed badly making it difficult to clearly compare performance.}. It also allows
a fair comparison of the discriminative properties of different classification algorithms on
the same input feature space. Therefore, we factor out the computational cost of
feature extraction.
\textbf{Gaussian Mixture Models -}
Given a $D$ dimensional feature vector $x$, a GMM~\cite{reynolds2009gaussian} is
a weighted sum of Gaussian component densities which provides an estimate of the
likelihood of $x$ being generated by the probability distribution defined by
a set of parameters $\theta$:
\begin{equation}
p(x|\theta) = \sum\limits_{i=0}^{M-1} w_{i} \cdot g(x | \mu_{i}, \Sigma_{i} )
\label{eq:gauss1}
\end{equation}
where:
\begin{equation}
g(x | \mu_{i}, \Sigma_{i} ) = \frac{1}{(2\pi)^{\frac{D}{2}}
|\Sigma_{i}|^{\frac{1}{2}}} \cdot e^{ -\frac{1}{2} (x - \mu_i)'
\Sigma_{i}^{-1} (x - \mu_i)}
\label{eq:gauss2}
\end{equation}
where $\theta = (\mu_{i}, \Sigma_{i}, w_{i})_{0 \leq i < M}$ is the parameter
set, $M$ is the number of Gaussian components, $\mu_{i}$ are the means,
$\Sigma_{i}$ the covariance matrices and $w_{i}$ the weights of each Gaussian
component. In practice, computing the likelihood in the log domain avoids
numerical underflow. Furthermore, the covariance matrices are constrained to be
diagonal, thus transforming equations~(\ref{eq:gauss1}) and~(\ref{eq:gauss2}) into:
\begin{equation}
\begin{array}{l}
\log p(x|\theta) = \\
\;\; \textrm{logsum}_{i=0}^{M-1} \left\{
\sum_{d=0}^{D-1}
\left(
x_d - \mu_{i,d}
\right)^2
\cdot
\sigma_{i,d}^{-1}
+ w_{i}
\right\}
+ K
\end{array}
\label{eq:gauss3}
\end{equation}
where $d$ is the dimension index, constant $K$ can be neglected for
classification purposes, and $\textrm{logsum}$ symbolises a recursive version of
function $\log(a+b) = \log a + \log(1 + e^{\left( \log b - \log a\right)})$ \cite{logsumexp},
which can be computed using a LUT on non-linearity $\log(1 + e^x)$.
Calculating the log likelihood of a $D$-dimensional vector given a GMM with
$M$~components therefore requires the following operations:
\begin{itemize}
\item $D$ subtractions (i.e., additions of negated means) and $2D$
multiplications per Gaussian to calculate the argument of the exponent.
\item $1$ extra addition per Gaussian to apply the weight in the log domain.
\item $M$ LUT lookups and $M$ additions for the non-linear $\textrm{logsum}$
cumulation across the Gaussian components.
\end{itemize}
This leads to the computational cost per audio frame expressed in the first
row of table~\ref{num_ops}.
\textbf{Support Vector Machines -}
\label{SVMs}
Support Vector Machines (SVMs) \cite{burges1998tutorial} are discriminative
classifiers. Given a set of data points belonging to two different
classes, the SVM determines the optimal separating hyperplane between the two
classes of data. In the linearly separable case, this is achieved by maximising
the margin between two hyperplanes that pass through a number of \emph{support
vectors}. The optimal separating hyperplane is defined by all points
$\textrm{x}$ that satisfy:
\begin{equation}
\textrm{x} \cdot \textrm{w} + b = 0
\end{equation}
where $\textrm{w}$ is a normal vector to the hyperplane and $\frac{|b|}{\|w\|}$
is the perpendicular distance from the hyperplane to the origin. Given that all
data points $\textrm{x}_{i}$ satisfy:
\begin{equation}
\left\{
\begin{array}{ll}
\textrm{x}_{i} \cdot \textrm{w} + b \geq 1 & \mbox{for labels $y_{i} = +1$} \\
\textrm{x}_{i} \cdot \textrm{w} + b \leq -1 & \mbox{for labels $y_{i} = -1$}
\end{array}
\right.
\end{equation}
It can be shown that the maximum margin is defined by
minimising $\frac{\|\textrm{w}\|^2}{2}$. This can be solved using a Lagrangian
formulation of the problem, thus producing the multipliers $\alpha_{i}$ and the
decision function:
\begin{equation}
f(\textrm{x}) = \textrm{sgn} \Big( \sum\limits_{i=0}^{N-1} y_{i} \alpha_{i} \textrm{x} \cdot \textrm{x}_i + b \Big)
\end{equation}
where $N$ is the number of training examples and $\textrm{x}$ is a feature vector
we wish to classify. In practice, most of the $\alpha_{i}$ will be zero, and
the $\textrm{x}_i$ for non-zero $\alpha_{i}$ are called the \emph{support
vectors} of the algorithm.
In the case where the data is not linearly separable, a non-linear kernel function
$K(\textrm{x}_{i},\textrm{x}_{j})$ can be used to replace the dot products $\textrm{x} \cdot \textrm{x}_i$,
with the effect of projecting the data into a higher dimensional space where it
could potentially become linearly separable. The decision function then becomes:
\begin{equation}
\label{SVM_classification}
f(\textrm{x}) = \textrm{sgn} \Big( \sum\limits_{i=0}^{N-1} y_{i} \alpha_{i} K(\textrm{x},\textrm{x}_{i}) + b \Big)
\end{equation}
Commonly used kernel functions include:
\medskip \noindent
\begin{tabular}{ll}
\hspace{-1.5ex} {$\vcenter{\hbox{\tiny$\bullet$}}$} Linear: &
\hspace{-2ex} $K(\textrm{x},\textrm{x}_{i}) = \textrm{x} \cdot \textrm{x}_{i}$
\\
\hspace{-1.5ex} {$\vcenter{\hbox{\tiny$\bullet$}}$} Polynomial: &
\hspace{-2ex} $K(\textrm{x},\textrm{x}_{i}) = (\textrm{x} \cdot
\textrm{x}_{i})^d$ \\
\hspace{-1.5ex} {$\vcenter{\hbox{\tiny$\bullet$}}$} Radial basis function (RBF):
& \hspace{-2ex} $K(\textrm{x},\textrm{x}_{i}) = \exp(-\gamma |
\textrm{x}-\textrm{x}_{i}|^2)$ \\
\hspace{-1.5ex} {$\vcenter{\hbox{\tiny$\bullet$}}$} Sigmoid: &
\hspace{-2ex} $K(\textrm{x},\textrm{x}_{i}) = \tanh(\textrm{x} \cdot
\textrm{x}_{i})$
\end{tabular}
\medskip \noindent
A further refinement to the SVM algorithm makes use of a \emph{soft margin}
whereby a hyperplane can still be found even if the data is non-separable
(perhaps due to mislabeled examples) \cite{burges1998tutorial}. The modified objective function is defined
as follows:
\begin{equation}
\label{soft_margin}
\arg\min_{\textrm{w},\mathbf{\xi}, b } \left\{\frac{1}{2} \|\textrm{w}\|^2 + C
\sum_{i=1}^n \xi_i \right\}
\end{equation}
\vspace{-0.25em} \hspace{1cm} subject to: $ y_i(\mathbf{w}\cdot\mathbf{x_i} - b)
\ge 1 - \xi_i, \;\;\xi_i \ge 0$
\vspace{0.5em}
\noindent $\xi_i$ are non-negative \emph{slack variables}. The modified algorithm
finds the best hyperplane that fits
the data while minimising the error of misclassified data points. The importance
of these error terms is determined by parameter~$C$,
which can control the tendency of the algorithm to over fit or under fit the data.
Given a support vector machine with $\lambda$ support vectors, a new
$D$-dimensional example can be classified using $\lambda$ additions to sum the
dot product results for each support vector, plus a further addition for the
bias term. For each dot product, a support vector requires the
computation of the kernel function, then $\lambda$ multiplications to apply
the $\alpha_i$ multipliers.
Some kernels require a number of elementary operations, whereas others require a
LUT lookup. Thus, depending on the kernel, the final costs are:
\begin{itemize}
\item SVM Linear -- $(\lambda \cdot D) + \lambda + 1$ additions and $(\lambda \cdot D) + \lambda$ multiplications.
\item Polynomial -- $\lambda(D + d + 2) + 1$ additions and $\lambda(D+2)$
multiplications, where $d$ is the degree of the polynomial.
\item Radial Basis Function -- $2\lambda D + \lambda + 1$ additions, $\lambda(D+2)$ multiplications and $\lambda$ exponential functions.
\item Sigmoid -- $\lambda(D+2) + 1$ additions and $\lambda(D+2)$ multiplications and $\lambda$ hyperbolic tangent functions.
\end{itemize}
The computational cost of SVMs per audio frame for these kernels is
summarized in Table~\ref{num_ops}.
\textbf{Deep Neural Networks -}
Two types of neural network architectures are compared in this study:
Feed-Forward Deep Neural Networks (DNNs) and Recurrent Neural Networks (RNNs).
A {\em Feed-Forward Deep Neural Network} processes the input by a series
of non-linear transformations given by:
\begin{equation}
\label{ffNN}
{h}^i = f \left( {W}^i {h}^{i-1}+{b}^i \right) \; \mbox{for} \; 0 \leq i \leq L
\end{equation}
where ${h}^0$ corresponds to the input $x$, ${W}^i,{b}^i$
are the weight matrix and bias vector for the $i^{th}$ layer,
and the output of the final layer ${h}^L$ is the desired output. The
non-linearity $f$ in Equation~\ref{ffNN} is usually a sigmoid $f(x) =
1/\left( 1 + e^{-x} \right)$ or a hyperbolic tangent function $\tanh(x)$.
However recent research in various domains has showed that the Rectified Linear Unit
(ReLU)~\cite{glorot2011deep} non-linearity can lead to
faster convergence of models during training. The ReLU function is defined as
$f(x) = max(0,x)$, which is simpler than the sigmoid or $\tanh$ activation
because it only requires a piecewise linear operator instead of a fixed point sigmoid or
$\tanh$ LUT.
While feed forward networks are designed for stationary data, {\em Recurrent
Neural Networks} (RNNs) are a class of neural networks designed for modelling
temporal data. RNNs predict the output $y_t$ given an input $x_t$, at some time $t$ in
the following way:
\begin{equation}
\label{RNN}
{h}^i_t = f \left( {W}^i_f {h}^{i-1}_t+{W}^i_r {h}^{i}_{t-1}+{b}^i \right) \;
\mbox{for} \; 0 \leq i \leq L
\end{equation}
where ${W}^i_f,{W}^i_r$ are the forward and recurrent weight matrices for the
$i^{th}$ recurrent layer, $b_i$ is bias for layer $i$, and
$\theta = \left\{{W}^i_f,{W}^i_r,{b}^i \right\}_0^L$ are the model parameters.
The input layer $h^{0}_{t}$ corresponds to the input $x_t$ and the
output layer, $h^{L}_{t}$ corresponds to the output $y_t$. In Equation
\ref{RNN}, $f$ is the hyperbolic tangent function. The choice of $f$ for the
output layer depends on the particular modelling problem.
RNNs are characterised by the recurrent connections ${W}^i_r$ between the
hidden units. Similar to feed forward neural networks, RNNs can have several
recurrent layers stacked on top of each other to learn more complex
functions~\cite{graves2013speech}. Due to the recursive structure of the hidden
layer, at any time $t$, the RNN makes a prediction conditioned on the entire
sequence seen until time $t$: $y_t = f(x_0^t)$.
For simplicity, every neural network in this study was constrained to have the
same number $H$ of hidden units in each of its $L$ layers. The input dimensionality
is denoted by $D$. The forward pass through a feed-forward neural network therefore
involves the following computations:
\begin{itemize}
\item Multiplying a column vector of size $n$ with a matrix
of size $n \times k$ involves $n \cdot k$ additions and an equal number of
multiplication operations.
\item Therefore the first layer involves $\left( D \cdot H \right)$
multiplication operations and $H \cdot \left( D +1 \right)$ additions, where
the additional $H$ additions are due to the bias term.
\item The remaining $L-1$ layers involve the following computations: $\left( L-1
\right) \cdot \left( H \cdot H\right)$ multiplications and $\left( L-1
\right) \cdot \left( H \cdot (H+1) \right)$ additions.
\item The output layer involves $H$ multiplications and $H+1$ additions.
\item A non-linearity is applied to the outputs of each layer, leading to a
total of $L \cdot H+1$ non-linearities.
\item The RNN forward pass includes an additional matrix multiplication due to the
recurrent connections. This involves $H \cdot H$ multiplications and an equal
number of addition operations.
\end{itemize}
The computational cost estimates for DNNs and RNNs are summarised in Table~\ref{num_ops}.
\section{Performance metrics for sound recognition \label{sec:perf}}
Classification systems produce two types of errors: False Negatives (FN), a.k.a. Missed Detections (MD),
i.e., target sounds which have not been detected, and False
Positives (FP), a.k.a. False Alarms (FA), i.e., non-target sounds which have been
mistakenly classified as the target sound~\cite{murphy2012machine}.
Their correlates are the True Positives (TP), or
correctly detected target sounds, and True Negatives (TN), or correctly
rejected non-target sounds. In our experiments, ``sounds'' refer to feature
frames associated with a particular sound class label.
The above metrics are usually expressed as percentages of the available
testing samples. They can be computed \emph{after an operating point
or threshold} has been selected \cite{japkowicz2011evaluating} for a given classifier.
Indeed, most classifiers output a real valued \emph{score}, which can be, e.g., a
log-likelihood ratio, a posterior probability or a distance. The binary
target/non-target classification is then made by comparing the score to a
threshold. Thus, a system can be made either more permissive, by choosing a
threshold which lets through more True Positives at the expense of
generating more False Positives, or the system can be made more conservative,
if the threshold is set to filter out more False Positives at the expense of
rejecting more sounds in general and thus generating more Missed Detections.
For example, the operation point of a GMM classifier is defined by choosing a
log-likelihood threshold.
Ideally, AESR systems should be compared independently of the choice of threshold.
This can be achieved through the use of
Detection Error Tradeoff (DET) curves~\cite{martin1997det}, which depict the
FA/MD compromises achieved across a sweep of operation points. DET curves are
essentially equivalent to Receiver Operating Curves
(ROC)~\cite{martin1997det}, but plotted into normal deviate scale to
linearise the curve, under the assumption that FA and MD scores are normally
distributed. Given that the goal is to minimise the FA/MD compromises globally
across all possible operation points, a DET curve closer to the origin means a
better classifier. DET curve performance can thus be summarised for convenience
into a single figure called the Equal Error Rate (EER), which is the point
where the DET curve crosses the \%FA = \%MD diagonal line. In other terms, the
EER represents the operation point where the system generates an equal rate of
FA and MD errors.
For GMM based classification systems, the output score is the ratio between the likelihood
of the target GMM and the likelihood of the universal background model (UBM): $P(x|\theta_{\mbox{\tiny
target sound}})/P(x|\theta_{\mbox{\tiny UBM}})$ \cite{reynolds2009gaussian}.
For neural networks, the output layer consists in a single unit followed by a
sigmoid function, thus yielding a real-valued class membership
probability $P(C=1|x)$ taken as the score value. For SVMs, the score can be
defined as the argument of the $\textrm{sgn}()$ function in
Equation~\ref{SVM_classification}: whereas the $\textrm{sgn}()$ function always
amounts to classifying against a threshold fixed to zero, using its argument as
a score suggests a more gradual notion of deviation from the margin defined by
the support vectors, or in other terms it suggests a more gradual measure of
belonging to one side or the other of the margin.
\section{Data sets and methodology \label{sec:methodology}}
\subsection{Data sets}
For the problem of AED, the training data set consists of audio recordings along with corresponding
annotations or ground truth labels. The ground truth labels are usually provided in
the form of alignments, i.e. onset times, offset times and the audio event
labels. This is in contrast to ASC problems where the ground truth does not include alignments.
ASC is similar to the problem of speech recognition where each recording has an
associated word-level transcription. The alignments are then obtained as a by-product of the
speech recognition model. For the problem of AED, annotating large quantities of data with
onset and offset times requires significant effort from human annotators.
Historically, relatively small data sets of labelled audio have been used for evaluating AED systems.
As a matter of fact, the two most popular benchmarks for AED algorithms known to date
are the DCASE challenge \cite{stowell2015detection} and the CLEAR challenge \cite{temko2006clear}.
The DCASE challenge data provides 20 examples for each of the 16 event classes. While
the data set for the CLEAR challenge contains approximately 60 training instances for
each of the 13 classes. The typical approach to AED so far has been to extract features from the audio
and then use a classifier or an \emph{acoustic model} to classify the frames into event labels.
Additionally, the temporal relationship between acoustic frames and outputs is modelled
using HMMs \cite{stowell2015detection}.
In this study, we investigate whether a data-driven approach to AED can yield improved performance.
Inspired by the recent success of Deep Learning \cite{lecun2015deep}, we investigate whether neural network
acoustic models can outperform existing models when trained on sufficiently large data sets.
Deep Neural Networks (DNNs) have recently contributed to significant progress in the fields
of computer vision, speech recognition, natural
language processing and other domains of machine learning. The superior performance of neural
networks can be largely attributed to more computational power and the availability of large quantities
of labelled data. Neural network models with millions of parameters can now be trained on
distributed platforms to achieve good generalisation performance \cite{szegedy2015going}. However,
despite the strong motivation to use neural networks for various tasks, their application
is limited by the availability of labelled data. For example, we trained many DNN architectures
on the DCASE data, but were unable to beat the performance of a baseline GMM system.
In order to train large neural network models, we perform experiments using
three private data sets made available
by Audio Analytic Ltd.\footnote{ {\tt \url{http://www.AudioAnalytic.com}} \\
The data can be made available to selected research partners
upon setting suitable contractual agreements.}
to support this study. These three data sets are subsets of much larger data
collection campaigns led by Audio Analytic Ltd. to support the
development of smoke alarm and baby cry detection products. Ground truth
annotations with onset and offset times obtained from human annotators were
also provided. The motivation for using
these data sets is twofold. Firstly, to provide a large number of training and test
examples for training models that recognize baby cries and smoke alarm sounds.
Secondly, a very large \emph{world} data set of ambient sounds is used as a source of
impostor data for training and testing, in a way which emulates the quasi infinite
prior probability of the system being exposed to non-target sounds instead of target ones.
In other words, there are potentially thousands of impostor sounds which
the system must reject in addition to identifying the target sounds correctly. While
the DCASE challenge evaluates how an AED system performs at detecting one sound in the presence
of 15 impostors or non-target sounds, the world data set allows to evaluate the
performance of detecting smoke alarms or baby cries against the presence of over 1000 impostor
sounds, in order to reflect a more realistic use case.
The \emph{Baby Cry} data subset comprises a
proportion of baby cries obtained from two different recording conditions:
one through camcorder in a hospital environment, one through uncontrolled
hand-held recorders in uncontrolled conditions (both indoors and outdoors).
Each recording was from a different baby. The train/test split was done evenly,
and set to achieve the same balance of both conditions in both the train and
test set (about 3/4 hospital condition and 1/4 uncontrolled condition) without
any overlap between sources for training and testing. The recordings
in this data set sum up to $4\,822$ seconds of training data and $3\,669$
seconds of test data, from which $224\,076$ feature frames in the training set
and $90\,958$ feature frames in the test set correspond to target baby cry
sounds.
The \emph{Smoke Alarm} data subset comprises recordings of 10 smoke
alarm models, recorded through 13 different channels across 3 British homes.
While it may be thought that smoke alarms are simple sounds, the variability
associated with different tones, different audio patterns, different room
responses and different channel responses does require some generalisation
capability which either may not be available in simpler fingerprinting or
template matching methods, or would require a very large database of templates.
The training set comprises recordings
from two of the homes, while the test set is composed of recordings from the
remaining home. Thus, while the same device can appear in
both the training and evaluation sets, the recording conditions are always different.
So the results presented over the smoke alarm set focus mainly on the generalisation
across room responses and recording conditions.
The data set provided for
this study consists of $15\,271$ seconds of training data and $5\,043$~seconds
of testing data, of which $194\,142$ feature frames in the training set and
$114\,753$ feature frames in the test set correspond to target smoke alarm
sounds.
The \emph{World} data set contains about 900 files covering 10 second samples
of a wide variety of complex acoustic scenes recorded
from numerous indoor and outdoor locations around the UK, across a wide and
uncontrolled range of devices, and potentially covering more than a thousand
sound classes. For example, a 10 seconds sample of “train station” scene from
the world data set covers, e.g., train noise, speech, departure whistle and more,
while a 10 second sample of “supermarket” covers babble noise, till beeps and
children’s voices. While it does not seem feasible to enumerate all the specific
classes from the World set, assuming that there are more than two classes
per file across the 410 files reserved for testing leads to
an estimate of more than a thousand classes in the test set.
The 500 other files of the world set are used to train a non-target model,
i.e., a Universal Background Model (UBM)~\cite{reynolds2009gaussian}
in the case of GMMs, or a negative input for neural network and SVM training.
It was ensured that the audio files in the both the training and testing world
sets were coming from completely disjoint recording sessions. The World data set provided
for this study consists of $5\,000$~seconds of training data and $4\,089$ seconds of test
data, thus amounting to $312\,500$ feature frames for training and $255\,562$
feature frames for testing.
Where necessary, the recordings were converted to 16kHz, 16~bits using a high
quality resampling algorithm. No other pre-processing was applied to the
waveform before feature extraction.
\subsection{Feature Extraction \label{sub:feature_extraction}}
For this experiment, Mel-frequency cepstral coefficient (MFCC) features were
extracted from the audio. MFCC features are extensively used in speech
recognition and in environmental sound recognition \cite{clavel2005events,valenzise2007scream,
portelo2009non,radhakrishnan2005audio}. In addition to MFCCs, the
spectral centroid \cite{clavel2005events}, spectral flatness
\cite{portelo2009non}, spectral rolloff \cite{valenzise2007scream}, spectral
kurtosis \cite{valenzise2007scream} and zero crossing rate
\cite{valenzise2007scream,atrey2006audio} were also computed. With audio
sampled at 16kHz, all the features were calculated with a window size of
$512$~samples and a hop size of $256$~samples. The first $13$~MFCCs
were used for our experiments. Concatenating all the features thus lead
to an $18$ dimensional feature vector of real values. Temporal
information was included in the input representation by appending the first and
second order differences of each feature. This led to an input representation with
$18 \times 3 = 54$~dimensions. It should be noted that the temporal difference features were
used as inputs to all models except the RNNs, where only the $18$~dimensional
features were used as inputs. This is due to the fact that the RNN is designed to
make predictions conditioned on the entire sequence history (Equation \ref{RNN}).
Therefore, explicitly adding temporal information to the inputs is unnecessary.
The reduction in input dimensionality and related reduction in computational
cost is offset by the extra weight matrix in the RNN (Table~\ref{num_ops}).
For all the experiments, the data was normalised to have zero mean and
unit standard deviation, with the mean and standard deviation calculated
over the entire training data independently for each feature dimension.
\subsection{Training Methodology \label{training}}
\textbf{Gaussian Mixture Models -} GMMs were trained using the expectation
maximisation algorithm (EM) \cite{bishop2006pattern}. The covariance matrices were
constrained to be diagonal\footnote{In preliminary experiments, we observed that
a full covariance matrix did not yield any improvement in performance.}.
A grid search was performed to find the optimal
number of Gaussians in the mixture, on the training set. The best model with number
of components $M \in \{ 1,2,4,8,16,32,64,128,256,512,1024 \}$ was used for evaluation. A validation set was
created by randomly selecting 20\% of the training data, while the remaining 80\% of the data
was used for parameter estimation. The model that performed the best on the validation set
was later used for testing. All the Gaussians were initialised
using the k-means++ algorithm \cite{arthur2007k}.
A single UBM was trained on the World training data set. One GMM per target class was then trained
independently of the UBM. We also performed experiments where GMMs for each
class were estimated by adapting the UBM by Maximum A Posteriori (MAP) adaptation \cite{reynolds2000speaker}.
We observed that this method yielded inferior results. We also observed worse results when no UBM
was used and class membership was determined by thresholding the outputs of individual
class GMMs. In the following, results are therefore only reported for the
best performing method, which is the likelihood ratio between independently
trained UBM and target GMMs.
\textbf{Support Vector Machines -} The linear, polynomial, radial basis function
(RBF) and sigmoid kernels were compared in our experiments. In addition to
optimising the kernel parameters ($d,\gamma$), a grid search was performed over the penalty
parameter $C$ (Equation \ref{soft_margin}). In practice, real-world data are often
non-separable, thus a soft margin has been used as necessary.
Model performance was estimated by evaluating
performance on a validation set. The validation set was formed by randomly selecting 20\%
of the training data, while the remaining data was used for parameter estimation. Again, the
best performing model on the validation set was used for evaluation.
SVM training is known not to scale well to very large
data sets~\cite{burges1998tutorial}: it has a time-complexity of $O(T^3)$, where
$T$ is the number of training examples. As a matter of fact, our preliminary
attempts at using the full training data sets led to prohibitive training times.
Standard practice in that case is to down-sample the training set by randomly
drawing $T$ frames from the target set, and the same number $T$ of frames from
the world set as negative examples for training. Results presented in
section~\ref{sec:results} with $T \in \left\{ 500,2000 \right\}$ show that
contrary to the intuitive thought that down-sampling might have disadvantaged
the SVMs by reducing the amount of training data compared to the other
machines, such data reduction actually improves SVM performance. As such,
down-sampling can be thought of as an integral part of the training process,
rather than as a reduction of the training data.
\textbf{Deep Neural Networks -} All neural network architectures
were trained to output a distribution over the presence or absence of a class,
using the backpropagation algorithm and stochastic gradient descent (SGD)
\cite{lecun2012efficient}. The training data was divided into a $80/20$
training/validation split to find the optimum training hyper-parameters. All the
target frames were used for each class, and an equal number of non-target frames
was sampled randomly from the World data set, for both training and validation.
For the DNNs, a grid search was performed over the following parameters: number
of hidden layers $L \in \left\{ 1,2,3,4 \right\}$, number of hidden units per layer $H \in
\left\{ 10,25,50,100,150 \right\}$, hidden activations $act \in \left\{
sigmoid, ReLU \right\}$. In order to minimise parameter tuning, we used
ADADELTA \cite{zeiler2012adadelta} to adapt the learning rate over iterations.
The networks were trained using mini-batches of size $100$ and training was
accelerated using an NVIDIA Tesla K40c GPU. A constant dropout rate
\cite{srivastava2014dropout} of $0.2$ was used for all layers. The training
was stopped if the cost on the validation set did not decrease after $20$
epochs.
For the RNNs, a grid search was performed over the following parameters:
number of hidden layers $L \in \left\{ 1,2,3 \right\}$ and number of hidden units
per layer $H \in \left\{10,25,50,100,150 \right\}$. An initial learning rate of $0.001$ was used and
linearly decreased to $0$ over $1000$ iterations. A constant momentum
rate of $0.9$ was used for all the updates. The training was stopped if the
error on the validation set did not decrease after $20$ epochs. The training
data was further divided into sub-sequences of length $100$ and the networks
were trained on these sub-sequences without any mini-batching. Gradient clipping
\cite{bengio2013advances} was also used to avoid the exploding gradient problem
in the early stages of RNN training. Clipping was triggered if the norm of the
gradient update exceeded $10$.
\section{Experimental results \label{sec:results}}
\begin{figure*}[ht]
\begin{center}
\begin{tabular}{ccc}
\subfloat[Baby Cry data set]{\includegraphics[width=0.42\textwidth,height=0.42\textwidth]{figure_1.png}\label{DET_baby_cry}}
&
\hspace{1cm}
&
\subfloat[Smoke Alarm data set]{\includegraphics[width=0.42\textwidth,height=0.42\textwidth]{figure_2.png}\label{DET_smoke_alarms}}\\
\end{tabular}
\end{center}
\caption{DET curves comparing frame classification performance of the acoustic classifiers.}
\label{DET_curves}
\end{figure*}
\begin{table*}
\begin{center}
\begin{tabular}{ccc}
\scalebox{1.}{
\subfloat[]{
\begin{tabular}{| l | c | c |}
\hline
\textbf{Best Baby Cry classifiers} & \textbf{EER} & \textbf{\# Ops.} \\
\hline
GMM, $M=32$ & 14.0 & 10\,560 \\
\parbox[t]{28ex}{Linear SVM, $T=2000$, \\ $C=1.0$, $\lambda=655$} & 12.9 &
101\,985
\\
\parbox[t]{28ex}{Feed-forward DNN, sigmoid, \\ $L=2$, $H=50$} & 10.8 &
10\,702 \\
\hline
\end{tabular}
\label{performance_baby_cry}
}
}
&
\hspace{1cm}
&
\scalebox{1.}{
\subfloat[]{
\begin{tabular}{| l | c | c |}
\hline
\textbf{Best Smoke Alarm classifiers} & \textbf{EER} & \textbf{\# Ops.} \\
\hline
GMM, $M=16$ & 2.9 & 5\,280 \\
\parbox[t]{28ex}{Linear SVM, $T=2000$, \\ $C=0.1$, $\lambda=152$} & 3.0 &
46\,655
\\
\parbox[t]{28ex}{Feed-forward DNN, sigmoid, \\ $L=2$, $H=25$} & 1.7 & 4\,102 \\
\hline
\end{tabular}
\label{performance_smoke_alarms}
}
}
\end{tabular}
\caption{Performance of the best classifiers on the Baby Cry and Smoke Alarm data sets.}
\label{classifier_performance}
\end{center}
\end{table*}
In this section, classification performance is analysed for the task of recognising
smoke alarms and baby cries against the large number of impostor sounds from the world set.
Table \ref{classifier_performance} presents the EER for the best performing classifiers of each
type and their associated computational cost. Figure \ref{DET_curves} shows the corresponding DET
curves. In Figure \ref{scatter_plots} we present a more detailed
comparison between the performance of various classifier types against the computational
cost involved in classifying a frame of input at test time.
\textbf{Baby Cry data set -} From Table \ref{performance_baby_cry} we observe that the
best performing GMM has 32 components and achieves an EER of 14.0\% for frame-wise
classification. From Figure \ref{baby_cry_scatter} (+ markers) we observe that there is no performance
improvement when the number of Gaussian components is increased beyond 32.
It should be noted that the best performing GMM (with $32$ components) has lower computational cost compared to
any SVM classifier, and a cost comparable on average with DNN classifiers.
From Table \ref{performance_baby_cry}, we observe that a SVM with a linear kernel
is the best performing SVM classifier, with an EER of $12.9\%$. The SVM was trained on
2000 examples, resulting in 655 support vectors. From Figure \ref{baby_cry_scatter}, we
note that the performance of the linear SVM (blue triangles) increases as the computational complexity
(number of support vectors) is increased. The number of support vectors can also
be controlled by varying the parameter $C$. For the best performing linear SVM, $C=1.0$ yields the best
results. We observe that SVMs with a sigmoid kernel (light blue triangles) yield similar results to the linear SVM
(Figure \ref{baby_cry_scatter}). The best SVM with a sigmoid kernel yields an EER of 13.5\%, while
the second best sigmoid SVM has an EER of 13.9\%. As in the linear case, we observe
an improvement in test performance as the computational cost or the number
of support vectors is increased.
\begin{figure*}[t!]
\subfloat[Baby Cry Data set]{\includegraphics[width=0.5\textwidth,height=0.45\textwidth]{figure_3.png}\label{baby_cry_scatter}}
\hfill
\subfloat[Smoke Alarm Data set]{\includegraphics[width=0.5\textwidth,height=0.45\textwidth]{figure_4.png}\label{smoke_alarm_scatter}}
\caption{Acoustic frame classification performance (EER percentage) as a
function of the number of operations per frame, for each of the tested models,
across the Baby Cry and Smoke Alarms data sets. }
\label{scatter_plots}
\end{figure*}
From Figure \ref{baby_cry_scatter}, we note that SVMs with RBF
and polynomial kernels (green and red triangles) are outperformed by linear and sigmoid SVMs, both
in terms of \%EER and computational cost. There is also no observable trend between
test performance and computational cost. For the polynomial SVM (red triangles), we found a
kernel with $d = 3$ yielded the best performance, while low values of gamma
$\gamma \in (0.005,0.01)$ provided the best results for the RBF kernel (Section \ref{SVMs}).
The number of training examples $T = 2000$ was empirically determined to be the
optimal value: adding more training examples did not yield any improvement in
performance. Conversely, we tried training the same classifiers with $T
= 500$ training examples, since the computational cost of SVMs is determined
both by the type of kernel and the number of support vectors, the latter being
controlled by varying the number of training examples and the penalty parameter
$C$. With $T = 500$ and $C=0.1$, we were able to
achieve a minimum EER rate of $13.7\%$ with the linear SVM, while halving the number of operations.
From Table \ref{performance_baby_cry}, the best performing neural network
architecture, which achieves an EER of $10.8\%$, is a feed-forward DNN with
sigmoid activations for the hidden units. Figure \ref{DET_baby_cry} shows that
the neural network clearly outperforms all the other models. From Figure \ref{baby_cry_scatter} we observe that the
feed forward DNNs (circle and square markers) achieve similar test performance over a wide range of computational
costs. This demonstrates that the network performance is not particularly sensitive to the
specific number of hidden units in each layer. However, we did observe that
networks which were deeper ($>1$ hidden layer) yielded better performance. From Figure \ref{baby_cry_scatter},
we observe that the RNN architectures (hexagonal markers) yield slightly worse performance and are computationally
more expensive. An interesting observation from Figure \ref{baby_cry_scatter} is that
feed-forward DNNs with both sigmoid and ReLU activations yield similar results. This is
a very important factor when deploying these models on embedded hardware, since a ReLU net
can be implemented with only linear operations (multiplications and additions), without
the need for costly Taylor series expansions or LUTs.
\textbf{Smoke Alarm data set -} From Table \ref{performance_smoke_alarms}, we
observe that the best performing GMM yields an EER of 2.9\% and uses 16 mixture
components. From Figure~\ref{smoke_alarm_scatter} (+ markers), we observe that the GMM
performance improves as the computation cost (number of Gaussians) increases till $M=32$.
Beyond this, increasing the number of Gaussian components does not improve results.
Again, we observe that the best performing GMM with $M=32$ has much lower computational
cost compared to SVMs, and a cost comparable on average with DNNs.
Similar to the results on the Baby Cry data set, the linear and sigmoid
kernel SVMs show the best performance, out of all four SVM kernel types.
The best linear SVM has an EER of 3.0\% -- a small
improvement over the best GMM approach, but with a small increase in the number
of operations. The model used 2000 training examples and $C=1.0$. From Figure
\ref{smoke_alarm_scatter} we again observe an improvement in performance, with
an increase in computation cost for both the linear and sigmoid kernels (blue triangles).
The best sigmoid kernel SVM scored 3.5\% with 2000 training examples, while
another configuration scored 3.6\% with $T = 500$ and $C = 1$, at half the
number of operations. Again, the polynomial and RBF kernels (red and green triangles) yield lower performance,
with no observable trend in terms of performance versus computational cost
(Figure \ref{smoke_alarm_scatter}).
From Table \ref{performance_smoke_alarms}, we observe that a feed-forward sigmoid DNN
yields the best performance, with an EER of $1.6\%$.
From the DET curves (Figure \ref{DET_smoke_alarms}), we see that the neural network
clearly outperforms the other models. From figure \ref{smoke_alarm_scatter} we note
that the neural networks consistently perform better than the other models, over a
wide range of computational costs, which correspond to different network configurations
(number of layers, number of units in each layer). The ReLU networks perform similarly
to the sigmoid networks, while the RNNs perform worse and are computationally more
costly.
It is worth noting that the performance of all classifiers is significantly better for the smoke alarm
sounds, since the smoke alarms are composed of simple tones. On the other
hand, baby cries have a large variability and are therefore more difficult to classify.
\section{Conclusion \label{sec:conclusion}}
In this study, we compare the performance of neural network acoustic models with
GMMs and SVMs on an environmental audio event detection task. Unlike other machine learning systems,
AESR systems are usually deployed on embedded hardware, which imposes many computational constraints.
Keeping this in mind, we compare the performance of the models as a function of their computational
cost. We evaluate the models on two tasks, detecting baby cries and detecting smoke alarms against
a large number of impostor sounds.
These data sets are much larger than the popular data sets found in AESR literature, which
enables us to train neural network acoustic models. Additionally, the large
number of impostor sounds allows to investigate the performance of the proposed
models in a testing scenario that is closer to practical use cases than previously
available data sets.
Results suggest that GMMs provide a low cost baseline for classification, across
both data sets. The GMM acoustic models are able to perform reasonably well at
a modest computational cost. SVMs with linear and sigmoid kernels yield similar
EER performance compared to GMMs, but their computational cost is overall higher.
The computational cost of the SVM is determined by the number of support vectors.
Unlike GMMs, SVMs are non-parametric models which do not allow the direct specification
of model parameters, although the number of support vectors can be indirectly controlled
with regularisation. Finally, our results suggest that deep neural
networks consistently outperform both the GMMs and the SVMs on both
data sets. The computational cost of DNNs can be controlled by
limiting the number of hidden units and the number of layers. While changes in
the number of units in the hidden layers did not appear to have a large impact
on performance, deeper networks appeared to perform better in all cases.
Additionally, neural networks with ReLU activations achieved good performance,
while being an attractive choice for deployment on embedded devices because
they do not require expensive LUT lookup operations.
In the future, we would like to expand the evaluations presented here to include more
event classes. However, the lack of large data sets for AESR problems is a major limitation.
We hope that studies like this one will encourage collaboration with industry partners to collect
large data sets for more rigorous evaluations. We would also like to investigate the performance
of acoustic models as a function of the memory efficiency, since memory is an important
consideration when designing models for embedded hardware.
\section{Acknowledgements}
The authors would like to thank the three anonymous reviewers whose comments helped to improve the article.
This work was supported by funding from Innovate UK, EPSRC grants EP/M507088/1 \& EP/N014111/1 from the UK Engineering and Physical Sciences Research Council, as well as private funding from Audio Analytic Ltd.
\ifCLASSOPTIONcaptionsoff
\newpage
\fi
\bibliographystyle{IEEEtran}
| 400a206513a253f92278ccc4218edf1aa83c3cbf | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Let $\Omega$ be a domain of $\mathbb{R}^n$ and $W_0^{1,q}(\Omega)$ be the closure of $C_0^\infty(\Omega)$ with respect to the norm
$$\|u\|_{W_0^{1,q}(\Omega)}=\le(\int_\Omega\le(|\nabla u|^q+|u|^q\ri)dx\ri)^{1/q},$$
where $q\geq 1$. Then the Sobolev embedding theorem reads
$$W_0^{1,q}(\Omega)\hookrightarrow\le\{\begin{array}{lll}
L^p(\Omega)\,\,\,{\rm for}\,\,\, q\leq p\leq q^*=\f{nq}{n-q},&{\rm when}\,\,\, n>q\\[1.5ex]
L^p(\Omega)\,\,\,{\rm for}\,\,\, q\leq p<+\infty, &{\rm when}\,\,\, q=n\\[1.5ex]
C^{1-n/q}(\overline{\Omega}), &{\rm when}\,\,\, q>n.
\end{array}\ri.$$
The model problem
\be\label{P}\le\{\begin{array}{lll}
-\Delta u+\lambda u=|u|^{p-2}u,\\[1.5ex]
u\in W_0^{1,2}(\Omega)
\end{array}\ri.\ee
or its variants has been extensively studied since 1960's. Let $J:W_0^{1,2}(\Omega)\ra \mathbb{R}$
be a functional defined by
$$J(u)=\f{1}{2}\int_\Omega(|\nabla u|^2+\lambda u^2)dx-\f{1}{p}\int_\Omega|u|^pdx.$$
Clearly the critical points of $J$ are weak solutions to the problem (\ref{P}). In the case $2<p<+\infty$ when $n=1,\,2$, or
$2<p\leq 2^*=2n/(n-2)$ when $n\geq 3$, one can check that $\sup_{W_0^{1,2}(\Omega)}J=+\infty$ and $\inf_{W_0^{1,2}(\Omega)}J
=-\infty$. In \cite{Nehari-1}, Nehari obtained a nontrivial solution of (\ref{P}) when $\lambda\geq 0$ and $\Omega=(a,b)$, by minimizing
$J$ in the manifold $$\mathcal{N}=\{u\in W_0^{1,2}(\Omega):\langle J^\prime(u),u\rangle=0,\,u\not= 0\}.$$ In \cite{Nehari-2},
he proved the existence of infinitely many solutions and in \cite{Nehari-3}, he solved the case
where $\Omega=\mathbb{R}^3$, $\lambda>0$ and $2<p<6$ after reduction to an ordinary differential equation. When
$\Omega$ is unbounded or when $p=2^*$, there is a lack of compactness in Sobolev spaces
because of invariance by translation or
by dilation. Some nonexistence results follow from the Pohozaev identity. General existence theorems were first
obtained by Strauss \cite{Strauss} when $\Omega=\mathbb{R}^n$ and by Brezis-Nirenberg \cite{Brezis-Nirenberg} when $p=2^*$. When $p=2^*$, Bahri and Coron \cite{Bahri-Coron} proved that if there exist a positive integer $d$ such that the $d$-dimentional homology group of domain $\Omega$ is nontrivial, then (\ref{P}) has a solution. Pohozaev
\cite{Pohozaev1} proved that if $\Omega$ is starshaped, then (\ref{P}) has no solution.
The Brezis-Lieb lemma and Lions' concentration compactness principle are important tools in solving those problems. For other existence results
for variants of (\ref{P}), we refer the reader to \cite{Willem}.
Analogous to (\ref{P}), one can consider the problem
\be\label{PT}\le\{\begin{array}{lll}
-\Delta_n u=f(x,u) \quad {\rm in}\quad \Omega\\[1.5ex]
u\in W_0^{1,n}(\Omega),
\end{array}\ri.\ee
where $\Delta_n$ is the $n$-Laplace operator and $f(x,s)$ has exponential growth as $s\ra+\infty$. Instead of the Sobolev embedding
theorem, the key tool in solving the problem (\ref{PT}) is the Trudinger-Moser embedding contributed by Yudovich \cite{Yudovich},
Pohozaev \cite{Pohozaev}, Peetre \cite{Peetre}, Trudinger \cite{Trudinger} and Moser \cite{Moser}. In \cite{Adimurthi}, Adimurthi
proved an existence
of positive solution to (\ref{PT}) by using a method of Nehari manifold.
In \cite{FMR}, de Figueiredo, Miyagaki and Ruf considered (\ref{PT}) in the case that $\Omega$ is a bounded domain in $\mathbb{R}^2$,
by using the critical point theory. In \cite{doo1}, by using the mountain-pass theorem
without the Palais-Smale condition, do \'O improve the results of \cite{Adimurthi,FMR}.
In \cite{doo2}, using the same method, he extended these results to the case that $\Omega$ is the whole Euclidean space $\mathbb{R}^n$.
For related works, we refer the reader to \cite{doo3,Adi-Yang,Yang1,Yang2} and the references there in.
On Riemannian manifolds, an analog of the model problem (\ref{P}) arises from the Yamabe problem: Let $(M,g)$ be a
compact $n$ $(\geq 3)$ dimensional Riemannian manifold without boundary. Does there exist a good metric $\tilde{g}$ in the conformal class of $g$ such that
the scalar curvature $R_{\tilde{g}}$ is a constant? This problem was studied by Yamabe \cite{Yamabe}, Trudinger \cite{Tru},
Aubin \cite{Aubin}, and completely solved by Schoen \cite{Schoen}. Though there is no background of geometry or physics,
there are still some works concerning the problem (\ref{PT})
on Riemannian manifolds, see for examples \cite{Yang-Zhao,do-Yang,Zhao,Yang-jfa}.\\
Our goal is to consider problems (\ref{P}) and (\ref{PT}) when
an Euclidean domain $\Omega$ is replaced by a graph. Such problems can be viewed as discrete versions of (\ref{P}) and (\ref{PT}).
In this paper, we concern bounded domain on locally finite graphs or finite graphs.
The key point is an observation of pre-compactness of the Sobolev
space in our setting.
Using the mountain pass theorem due to Ambrosseti-Rabinowich \cite{Ambrosetti-Rabinowitz}, we
prove the existence of nontrivial solutions to Yamabe type equations on graphs. Our results are quite different from
that of \cite{Bahri-Coron} and \cite{Pohozaev1} when $p=2^*$, namely, the existence results are not related to the homology group of the graphs.\\
This paper is organized as follows: In Section 2, we give some notations on graph and state main results.
In Section 3, we establish Sobolev embedding such that the mountain pass theorem can be applied to our problems.
Local existence results (Theorems \ref{Theorem 1}-\ref{Theorem 4}) are proved in Section 4, and
global existence results (Theorems \ref{Theorem 5}-\ref{Theorem 8})
are proved in Section 5.
\section{Settings and main results}
Let $G=(V,E)$ be a finite or locally finite graph, where $V$ denotes the vertex set and $E$ denotes the edge set.
For any edge $xy\in E$, we assume that its weight $w_{xy}>0$ and that $w_{xy}=w_{yx}$. The degree of $x\in V$ is defined
as ${\rm deg}(x)=\sum_{y\sim x}w_{xy}$, where we write $y\sim x$ if
$xy\in E$. Let $\mu:V\ra \mathbb{R}^+$ be a finite measure. For any function $u:V\ra \mathbb{R}$, the $\mu$-Laplacian (or Laplacian for short)
of $u$ is defined as
\be\label{lap}\Delta u(x)=\f{1}{\mu(x)}\sum_{y\sim x}w_{xy}(u(y)-u(x)).\ee
The associated gradient form reads
\be\label{grad-form}\Gamma(u,v)(x)=\f{1}{2\mu(x)}\sum_{y\sim x}w_{xy}(u(y)-u(x))(v(y)-v(x)).\ee
Write $\Gamma(u)=\Gamma(u,u)$. We denote the length of its gradient by
\be\label{grd}|\nabla u|(x)=\sqrt{\Gamma(u)(x)}=\le(\f{1}{2\mu(x)}\sum_{y\sim x}w_{xy}(u(y)-u(x))^2\ri)^{1/2}.\ee
Similar to the Euclidean case, we define the length of $m$-order gradient of $u$ by
$$\label{high-deriv}|\nabla^mu|=\le\{\begin{array}{lll}
|\nabla\Delta^{\f{m-1}{2}}u|,\,\,{\rm when}\,\,\, m
\,\,{\rm is\,\,odd}\\[1.5ex]
|\Delta^{\f{m}{2}}u|,\,\,{\rm when}\,\,\, m
\,\,{\rm is\,\,even},
\end{array}\ri.
$$
where $|\nabla\Delta^{\f{m-1}{2}}u|$ is defined as in (\ref{grd}) with $u$ is replaced by $\Delta^{\f{m-1}{2}}u$, and
$|\Delta^{\f{m}{2}}u|$ denotes the usual absolute of the function $\Delta^{\f{m}{2}}u$.
Let $\Omega$ be a domain (or a connected set) in $V$.
To compare with the Euclidean setting, we denote, for any function $u:V\ra\mathbb{R}$,
\be\label{integration}\int_\Omega ud\mu=\sum_{x\in \Omega}\mu(x)u(x).\ee
The first eigenvalue of the Laplacian with respect to Dirichlet boundary condition reads
\be\label{eigen}\lambda_1(\Omega)=\inf_{u\not\equiv 0,\,u|_{\p\Omega}=0}\f{\int_\Omega|\nabla u|^2d\mu}{\int_\Omega u^2d\mu},\ee
where $\p\Omega$ is the boundary of $\Omega$, namely
$\p\Omega=\{x\in\Omega: \exists y\not\in \Omega\,\,{\rm such\,\,that}\,\, xy\in E\}$.
Moreover, we denote the interior of $\Omega$ by
$\Omega^\circ=\Omega\setminus\p\Omega$. Our first result is the following:
\begin{theorem}\label{Theorem 1}
Let $G=(V,E)$ be a locally finite graph, $\Omega\subset V$ be a bounded domain with $\Omega^\circ\not=\varnothing$, and
$\lambda_1(\Omega)$ be defined as in (\ref{eigen}). Then for any $p>2$ and any $\alpha<\lambda_1(\Omega)$, there exists a
solution to the equation
$$\label{1}\left\{
\begin{array}{lll}
-\Delta u-\alpha u=|u|^{p-2}u\quad{\rm in}\quad \Omega^\circ\\[1.5ex]
u>0\,\,\,{\rm in}\,\,\, \Omega^\circ,\,\,\, u=0\,\,\,{\rm on}\,\,\, \p\Omega.
\end{array}
\right.
$$
\end{theorem}
The $p$-Laplacian of $u:V\ra\mathbb{R}$, namely $\Delta_pu$, is defined in the distributional sense by
$$\int_V(\Delta_pu)\phi d\mu=-\int_V |\nabla u|^{p-2}\Gamma(u,\phi)d\mu,\quad\forall
\phi\in \mathcal{C}_{\rm c}(V),$$
where $\Gamma(u,\phi)$ is defined as in (\ref{grad-form}), $\mathcal{C}_{\rm c}(V)$ denotes
the set of all functions with compact support, and the integration is defined by (\ref{integration}). Point-wisely, $\Delta_pu$ can be written as
$$\Delta_pu(x)=\f{1}{2\mu(x)}\sum_{y\sim x}\le(|\nabla u|^{p-2}(y)+|\nabla u|^{p-2}(x)\ri)w_{xy}(u(y)-u(x)).$$
When $p=2$, $\Delta_p$ is the standard graph Laplacian $\Delta$ defined by (\ref{lap}).
The first eigenvalue of the $p$-Laplacian with respect to Dirichlet boundary
condition reads
\be\label{p-eig}\lambda_p(\Omega)=\inf_{u\not\equiv 0,\,u|_{\p\Omega}=0}\f{\int_\Omega|\nabla u|^pd\mu}
{\int_\Omega |u|^pd\mu}.\ee
Our second result can be stated as follows:
\begin{theorem}\label{Theorem 2}
Let $G=(V,E)$ be a locally finite graph, $\Omega\subset V$ be a bounded domain with $\Omega^\circ\not=\varnothing$.
Let
$\lambda_p(\Omega)$ be defined as in (\ref{p-eig}) for some $p>1$. Suppose that $f:\Omega\times\mathbb{R}\ra \mathbb{R}$ satisfies the following
hypothesis:\\
$(H_1)$ For any $x\in \Omega$, $f(x,t)$ is continuous in $t\in\mathbb{R}$;\\
$(H_2)$ For all $(x,t)\in\Omega\times[0,+\infty)$, $f(x,t)\geq 0$ , and $f(x,0)=0$ for all $x\in \Omega$;\\
$(H_3)$ There exists some $q>p$ and $s_0>0$ such that if $s\geq s_0$, then there holds
$$F(x,s)=\int_0^sf(x,t)dt\leq \f{1}{q}sf(x,s),\quad \forall x\in \Omega;$$
$(H_4)$ For any $x\in \Omega$, there holds
$$\limsup_{t\ra 0+}\f{f(x,t)}{t^{p-1}}<\lambda_p(\Omega).$$
Then there exists a nontrivial solution to the equation
$$\label{2}\left\{
\begin{array}{lll}
-\Delta_p u=f(x,u)\quad{\rm in}\quad \Omega^\circ\\[1.5ex]
u\geq 0\,\,\,{\rm in}\,\,\, \Omega^\circ,\,\,\, u=0\,\,\,{\rm on}\,\,\, \p\Omega.
\end{array}
\right.
$$
\end{theorem}
In Theorem \ref{Theorem 2}, if $p=2$, then
$|s|^{q-2}s$ $(q> 2)$ satisfies $(H_1)-(H_4)$. Moreover, the nonlinearities in Theorem \ref{Theorem 2}
include the case of exponential growth as in the problem (\ref{PT}). For further extension, we
define an analog of $\lambda_p(\Omega)$ by
\be\label{l-mp}\lambda_{mp}(\Omega)=\inf_{u\in\mathcal{H}}\f{\int_\Omega|\nabla^mu|^pd\mu}{\int_\Omega|u|^pd\mu},\ee
where $m$ is any positive integer and $\mathcal{H}$ denotes the set of all functions $u\not\equiv 0$ with $u=|\nabla u|=\cdots=|\nabla^{m-1}u|=0$
on $\p\Omega$. Then we have the following:
\begin{theorem}\label{Theorem 4}
Let $G=(V,E)$ be a locally finite graph and $\Omega\subset V$ be a bounded domain with $\Omega^\circ\not=\varnothing$.
Let $m\geq 2$ be an integer, $p>1$, and $\lambda_{mp}(\Omega)$ be defined by (\ref{l-mp}). Suppose that $f:\Omega\times\mathbb{R}\ra \mathbb{R}$ satisfies the following
assumptions:\\
$(A_1)$ $f(x,0)=0$, $f(x,t)$ is continuous with respect to $t\in\mathbb{R}$;\\
$(A_2)$ $\limsup_{t\ra 0}\f{|f(x,t)|}{|t|^{p-1}}<\lambda_{mp}(\Omega)$;\\
$(A_3)$ there exists some $q>p$ and $M>0$ such that if $|s|\geq M$, then
$$0<qF(x,s)\leq sf(x,s),\quad\forall x\in \Omega.$$
Then there exists a nontrivial
solution to the equation
$$\label{e-3}\left\{
\begin{array}{lll}
\mathcal{L}_{m,p}\, u=f(x,u)\quad{\rm in}\quad \Omega^\circ\\[1.5ex]
|\nabla^j u|=0\,\,{\rm on}\,\, \p\Omega,\,0\leq j\leq m-1,
\end{array}
\right.
$$
where $\mathcal{L}_{m,p}\, u$ is defined as follows: for any $\phi$ with $\phi=|\nabla\phi|=\cdots=
|\nabla^{m-1}\phi|=0$ on $\p\Omega$, there holds
$$\int_\Omega(\mathcal{L}_{m,p}\, u)\phi d\mu=\le\{\begin{array}{lll}
\int_\Omega|\nabla^mu|^{p-2}\Gamma(\Delta^{\f{m-1}{2}}u,\Delta^{\f{m-1}{2}}\phi)d\mu,\,\,\,{\rm when} \,\, m\,\,{\rm is\,\,odd},\\
[1.5ex] \int_\Omega |\nabla^mu|^{p-2}\Delta^{\f{m}{2}}u\Delta^{\f{m}{2}}\phi d\mu,\,\,\,{\rm when} \,\, m\,\,{\rm is\,\,even}.
\end{array}\ri.$$
In particular, if $p=2$, then $\mathcal{L}_{m,p}\, u=(-\Delta)^mu$, the poly-Laplacian of $u$.
\end{theorem}
If $G=(V,E)$ is a finite graph, we also have existence results similar to the above theorems. Analogous to Theorem \ref{Theorem 1},
we state the following:
\begin{theorem}\label{Theorem 5}
Let $G=(V,E)$ be a finite graph. Suppose that $p>2$ and $h(x)>0$ for all $x\in V$. Then there exists a
solution to the equation
$$\label{e-4}\left\{
\begin{array}{lll}
-\Delta u+h u=|u|^{p-2}u\quad{\rm in}\quad V\\[1.5ex]
u>0\,\,\,{\rm in}\,\,\, V.
\end{array}
\right.
$$
\end{theorem}
Similar to Theorem \ref{Theorem 2}, we have
\begin{theorem}\label{Theorem 6}
Let $G=(V,E)$ be a finite graph. Suppose that $h(x)>0$ for all $x\in V$.
Suppose that $f:V\times\mathbb{R}\ra \mathbb{R}$ satisfies the following
hypothesis:\\
$(H_V^1)$ For any $x\in V$, $f(x,t)$ is continuous in $t\in\mathbb{R}$;\\
$(H_V^2)$ For all $(x,t)\in V\times[0,+\infty)$, $f(x,t)\geq 0$ , and $f(x,0)=0$ for all $x\in V$;\\
$(H_V^3)$ There exists some $q>p>1$ and $s_0>0$ such that if $s\geq s_0$, then there holds
$$F(x,s)=\int_0^sf(x,t)dt\leq \f{1}{q}sf(x,s),\quad \forall x\in V;$$
$(H_V^4)$ For any $x\in V$, there holds
$$\limsup_{t\ra 0+}\f{f(x,t)}{t^{p-1}}<\lambda_p(V)=\inf_{u\not\equiv 0}\f{\int_V(|\nabla u|^p+h|u|^p)d\mu}
{\int_V|u|^pd\mu}.$$
Then there exists a nontrivial solution to the equation
$$\label{e-5}\left\{
\begin{array}{lll}
-\Delta_p u+h|u|^{p-2}u=f(x,u)\quad{\rm in}\quad V\\[1.5ex]
u\geq 0\,\,\,{\rm in}\,\,\, V,
\end{array}
\right.
$$
where $\Delta_pu$ denotes the $p$-Laplacian of $u$.
\end{theorem}
Finally we have an analog of Theorem \ref{Theorem 4}, namely
\begin{theorem}\label{Theorem 8}
Let $G=(V,E)$ be a finite graph. Let $m\geq 2$ be an integer and $p>1$. Suppose that
$h(x)>0$ for all $x\in V$. Assume $f(x,u)$ satisfies the following assumptions:\\
$(A_V^1)$ For any $x\in V$, $f(x,0)=0$, $f(x,t)$ is continuous with respect to $t\in\mathbb{R}$;\\
$(A_V^2)$ $\limsup_{t\ra 0}\f{|f(x,t)|}{|t|^{p-1}}<\lambda_{mp}(V)=\inf_{u\not\equiv 0}
\f{\int_V(|\nabla^m u|^p+h|u|^p)d\mu}{\int_V|u|^pd\mu}$;\\
$(A_V^3)$ there exists some $q>p$ and $M>0$ such that if $|s|\geq M$, then
$$0<qF(x,s)\leq sf(x,s),\quad\forall x\in V.$$
Then there exists a nontrivial solution to
$$\label{e-6}
\mathcal{L}_{m,p}\, u+h|u|^{p-2}u=f(x,u)\quad{\rm in}\quad V,
$$
where $\mathcal{L}_{m,p}\, u$ is defined in the distributional sense: for any function $\phi$, there holds
$$\int_V(\mathcal{L}_{m,p}\, u)\phi d\mu=\le\{\begin{array}{lll}
\int_V|\nabla^mu|^{p-2}\Gamma(\Delta^{\f{m-1}{2}}u,\Delta^{\f{m-1}{2}}\phi)d\mu,\,\,\,{\rm when} \,\, m\,\,{\rm is\,\,odd},\\
[1.5ex] \int_V |\nabla^mu|^{p-2}\Delta^{\f{m}{2}}u\Delta^{\f{m}{2}}\phi d\mu,\,\,\,{\rm when} \,\, m\,\,{\rm is\,\,even}.
\end{array}\ri.$$
\end{theorem}
\section{Preliminary analysis}
Let $G=(V,E)$ be a locally finite graph, $\Omega\subset V$ be an domain, $\p\Omega$ be its boundary and $\Omega^\circ$ be its
interior. For any $p>1$,
$W^{m,p}(\Omega)$ is defined as a space of all functions $u:V\ra\mathbb{R}$ satisfying
\be\label{norm}\|u\|_{W^{m,p}(\Omega)}=\le(\sum_{k=0}^m\int_\Omega|\nabla^ku|^pd\mu\ri)^{1/p}<\infty.\ee
Denote $\mathcal{C}_0^m(\Omega)$ be a set of all functions $u:\Omega\ra\mathbb{R}$ with
$u=|\nabla u|=\cdots=|\nabla^{m-1}u|=0$ on $\p\Omega$.
We denote $W_0^{m,p}(\Omega)$ be the completion of $\mathcal{C}_0^m(\Omega)$ under the norm (\ref{norm}).
If we further assume that $\Omega$ is a bounded, then $\Omega$ is a finite set.
Observing that the dimension of $W_0^{m,p}(\Omega)$ is finite when $\Omega$ is bounded,
we have the following Sobolev embedding:
\begin{theorem} \label{compactlemma} Let $G=(V,E)$ be a locally finite graph,
$\Omega$ be a bounded domain of $V$ such that $\Omega^\circ\not=\varnothing$.
Let $m$ be any positive integer and $p>1$.
Then $W_0^{m,p}(\Omega)$ is embedded in $L^q(\Omega)$ for all $1\leq q\leq+\infty$. In particular,
there exists a constant $C$ depending only on $m$, $p$ and $\Omega$ such that
\be\label{Poincare}\le(\int_\Omega |u|^qd\mu\ri)^{1/q}\leq C\le(\int_\Omega|\nabla^mu|^pd\mu\ri)^{1/p}\ee
for all $1\leq q\leq +\infty$ and for all $u\in W_0^{m,p}(\Omega)$.
Moreover, $W_0^{m,p}(\Omega)$ is pre-compact, namely, if $u_k$ is bounded in $W_0^{m,p}(\Omega)$, then up to a subsequence,
there exists some $u\in W_0^{m,p}(\Omega)$ such that $u_k\ra u$ in $W_0^{m,p}(\Omega)$.
\end{theorem}
{\it Proof.} Since $\Omega$ is a finite set, $W_0^{m,p}(\Omega)$ is a finite dimensional space. Hence $W_0^{m,p}(\Omega)$
is pre-compact. We are left to show (\ref{Poincare}). It is not difficult to see that
\be\label{w0-norm}\|u\|_{W_0^{m,p}(\Omega)}=\le(\int_\Omega|\nabla^mu|^pd\mu\ri)^{1/p}\ee
is a norm equivalent to (\ref{norm}) on $W_0^{m,p}(\Omega)$. Hence for any $u\in W_0^{m,p}(\Omega)$, there exists some
constant $C$ depending only on $m$, $p$ and $\Omega$ such that
\be\label{qp}\le(\int_\Omega |u|^pd\mu\ri)^{1/p}=\le(\sum_{x\in\Omega}\mu(x)|u(x)|^p\ri)^{1/p}\leq
C\le(\int_\Omega|\nabla^mu|^pd\mu\ri)^{1/p},\ee
since $W_0^{m,p}(\Omega)$ is a finite dimensional space. Denote $\mu_{\min}=\min_{x\in \Omega}\mu(x)$. Then (\ref{qp}) leads to
$$\label{l-infty}\|u\|_{L^\infty(\Omega)}\leq \f{C}{\mu_{\min}}\|u\|_{W_0^{m,p}(\Omega)},$$
and thus for any $1\leq q<+\infty$,
$$\le(\int_\Omega |u|^qd\mu\ri)^{1/q}\leq \f{C}{\mu_{\min}}|\Omega|^{1/q}\|u\|_{W_0^{m,p}(\Omega)}\leq
\f{C}{\mu_{\min}}(1+|\Omega|)\|u\|_{W_0^{m,p}(\Omega)},$$
where $|\Omega|=\sum_{x\in\Omega}\mu(x)$ denotes the volume of $\Omega$. Therefore (\ref{Poincare}) holds. $\hfill\Box$\\
If $V$ is a finite graph, then $W^{m,p}(V)$ can be defined as a set of all functions $u:V\ra\mathbb{R}$ under the norm
\be\label{Wh}\|u\|_{W^{m,p}(V)}=\le(\int_V(|\nabla^m u|^p+h|u|^p)d\mu\ri)^{1/p},\ee
where $h(x)>0$ for all $x\in V$. Then we have an obvious analog of Theorem \ref{compactlemma} as follows:
\begin{theorem}\label{compact-V}
Let $V$ be a finite graph. Then $W^{m,p}(V)$ is embedded in $L^q(V)$ for all $1\leq q\leq+\infty$.
Moreover, $W^{m,p}(V)$ is pre-compact.
\end{theorem}
Obviously, both $W_0^{k,p}(\Omega)$ and
$W^{k,p}(V)$ with norms (\ref{w0-norm}) and (\ref{Wh}) respectively are Banach spaces.
Let $(X,\|\cdot\|)$ be a Banach space, $J: X\ra\mathbb{R}$ be a functional. We say that $J$ satisfies
the $(PS)_c$ condition for some real number $c$, if for any sequence of functions $u_k: X\ra\mathbb{R}$
such that $J(u_k)\ra c$ and $J^\prime(u_k)\ra 0$ as $k\ra +\infty$, there holds up to a subsequence,
$u_k\ra u$ in $X$. To prove Theorems \ref{Theorem 1}-\ref{Theorem 8}, we need the following mountain pass theorem.
\begin{theorem}(Ambrosetti-Rabinowitz \cite{Ambrosetti-Rabinowitz}).\label{Mountain-pass-Theorem}
Let
$(X,\|\cdot\|)$ be a Banach space, $J\in C^1(X,\mathbb{R})$, $e\in X$ and
$r>0$ be such that $\|e\|>r$ and
$$b:=\inf_{\|u\|=r}J(u)>J(0)\geq J(e).$$
If $J$ satisfies the $(PS)_c$ condition with $c:=\inf_{\gamma\in\Gamma}\max_{t\in[0,1]}J(\gamma(t))$,
where $$\Gamma:=\{\gamma\in C([0,1],X): \gamma(0)=0,\gamma(1)=e\},$$
then $c$ is a critical value of $J$.
\end{theorem}
\section{Local existence }
In this section, we prove Theorems \ref{Theorem 1}-\ref{Theorem 4} by applying Theorem \ref{Mountain-pass-Theorem}.
\\
{\it Proof of Theorem \ref{Theorem 1}}. Let $p>2$ and $\alpha<\lambda_1(\Omega)$ be fixed.
For any $u\in W_0^{1,2}(\Omega)$, we let
$$J(u)=\f{1}{2}\int_\Omega(|\nabla u|^2-\alpha u^2) d\mu-\f{1}{p}
\int_\Omega (u^+)^pd\mu,$$
where $u^+(x)=\max\{u(x),0\}$.
It is clear that $J\in \mathcal{C}^1(W_0^{1,2}(\Omega),\mathbb{R})$. We claim that
$J$ satisfies the $(PS)_c$ condition for any $c\in\mathbb{R}$.
To see this, we take a sequence of functions $u_k\in W_0^{1,2}(\Omega)$ such that
$J(u_k)\ra c$, $J^\prime(u_k)\ra 0$ as $k\ra+\infty$.
This leads to
\bea
\label{p1}\f{1}{2}\int_\Omega(|\nabla u_k|^2-\alpha u_k^2)d\mu-\f{1}{p}\int_\Omega(u_k^+)^pd\mu=c+o_k(1),\\[1.5ex]
\label{p2}\le|\int_\Omega(|\nabla u_k|^2-\alpha u_k^2)d\mu-\int_\Omega (u_k^+)^pd\mu\ri|\leq o_k(1)\|u_k\|_{W_0^{1,2}(\Omega)}.
\eea
Noting that $p>2$, we conclude from (\ref{p1}) and (\ref{p2}) that $u_k$ is bounded in $W_0^{1,2}(\Omega)$. Then the Sobolev
embedding (Theorem
\ref{compactlemma})
implies that up to a subsequence, $u_k$ converges to some function $u$ in $W_0^{1,2}(\Omega)$. Hence the $(PS)_c$ condition holds.
To proceed, we need to check that $J$ satisfies all conditions in the mountain pass theorem (Theorem
\ref{Mountain-pass-Theorem}). Note that
\be\label{j0}J(0)=0.\ee
By Theorem \ref{compactlemma}, there exists some constant $C$ depending only on $p$ and $\Omega$ such that
$$\le(\int_\Omega (u^+)^pd\mu\ri)^{1/p}\leq C\le(\int_\Omega|\nabla u|^2d\mu\ri)^{1/2}.$$
Hence there holds for all $u\in W_0^{1,2}(\Omega)$
$$J(u)\geq \f{1}{2}\|u\|_{W_0^{1,2}(\Omega)}^2-\f{C^p}{p}\|u\|_{W_0^{1,2}(\Omega)}^p.$$
Since $p>2$, one can find some sufficiently small $r>0$ such that
\be\label{c-2}\inf_{\|u\|_{W_0^{1,2}(\Omega)}=r}J(u)>0.\ee
Take a function $u^*\in W_0^{1,2}(\Omega)$ satisfying $u^*>0$ in $\Omega^\circ$.
Passing to the limit $t\ra+\infty$, we have
$$J(tu^*)=\f{t^2}{2}\int_\Omega(|\nabla u^*|^2-\alpha (u^*)^2)d\mu-\f{t^p}{p}\int_\Omega ({u^*}^+)^pd\mu\ra-\infty.$$
Hence there exists some $u_0\in W_0^{1,2}(\Omega)$ such that
\be\label{c-3}J(u_0)<0,\quad\|u_0\|_{W_0^{1,2}(\Omega)}>r.\ee
Combining (\ref{j0}), (\ref{c-2}) and (\ref{c-3}), we conclude by Theorem \ref{Mountain-pass-Theorem} that
$c=\inf_{\gamma\in\Gamma}\max_{t\in[0,1]}J(\gamma(t))$ is a critical value of $J$, where
$\Gamma=\{\gamma\in C([0,1],W_0^{1,2}(\Omega)): \gamma(0)=0,\gamma(1)=u_0\}$. In particular, there exists some function
$u\in W_0^{1,2}(\Omega)$ such that
\be\label{eqn}-\Delta u-\alpha u=(u^+)^{p-1}\quad{\rm in}\quad \Omega^\circ.\ee
Testing (\ref{eqn}) by $u^-=\min\{u,0\}$ and noting that $u^-u^+=u^-(u^+)^{p-1}=0$, we have
$$-\int_\Omega u^-\Delta ud\mu-\alpha\int_\Omega (u^-)^2d\mu=0.$$
Since $u=u^++u^-$, the above equation leads to
\bea\nonumber\f{\alpha}{\lambda_1(\Omega)}\int_\Omega|\nabla u^-|^2d\mu&\geq& \alpha\int_\Omega (u^-)^2d\mu\\
&=&\nonumber
-\int_\Omega u^-(\Delta u^++\Delta u^-)d\mu\\&=&
\int_\Omega|\nabla u^-|^2d\mu-\int_\Omega u^-\Delta u^+d\mu.\label{g1}\eea
Note that
\bea\nonumber
-\int_\Omega u^-\Delta u^+d\mu&=&-\sum_{x\in\Omega^\circ}u^-(x)\sum_{y\sim x}w_{xy}(u^+(y)-u^+(x))\\
&=&-\sum_{x\in\Omega^\circ}\sum_{y\sim x}w_{xy}u^-(x)u^+(y)\geq 0.\label{g2}
\eea
Inserting (\ref{g2}) into (\ref{g1}) and recalling that $\alpha<\lambda_1(\Omega)$, we obtain
$\int_\Omega|\nabla u^-|^2d\mu= 0$, which implies that $u^-\equiv 0$ in $\Omega$. Whence $u\geq 0$ and (\ref{eqn})
becomes
\be\label{eqn-1}\left\{
\begin{array}{lll}
-\Delta u-\alpha u=u^{p-1}\quad{\rm in}\quad \Omega^\circ\\[1.5ex]
u\geq0\,\,\,{\rm in}\,\,\, \Omega^\circ,\,\,\, u=0\,\,\,{\rm on}\,\,\, \p\Omega.
\end{array}
\right.\ee
Suppose $u(x)=0=\min_{x\in\Omega}u(x)$ for some $x\in\Omega^\circ$. If $y$ is adjacent to $x$, then
we know from (\ref{eqn-1}) that $\Delta u(x)=0$, and thus by the definition of $\Delta$ (see (\ref{lap}) above),
$u(y)=0$. Therefore we conclude that $u\equiv 0$ in $\Omega$, which contradicts (\ref{c-2}). Hence $u>0$ in
$\Omega^\circ$ and this completes the proof of the theorem. $\hfill\Box$\\
{\it Proof of Theorem \ref{Theorem 2}}. Let $p>1$ be fixed. For any $u\in W_0^{1,p}(\Omega)$, we let
$$J_p(u)=\f{1}{p}\int_\Omega|\nabla u|^pd\mu-
\int_\Omega F(x,u^+)d\mu,$$
where $u^+(x)=\max\{u(x),0\}$.
In view of $(H_4)$, there exist two constants $\lambda$ and $\delta>0$ such that $\lambda<\lambda_p(\Omega)$ and
$$F(x,u^+)\leq \f{\lambda}{p} (u^+)^p+\f{(u^+)^{p+1}}{\delta^{p+1}}F(x,u^+).$$
For any $u\in W_0^{1,p}(\Omega)$ with $\|u\|_{W_0^{1,p}(\Omega)}\leq 1$, we have by the Sobolev embedding
(Theorem \ref{compactlemma}) that $\|u\|_\infty\leq C$ for some constant $C$ depending only on $p$ and $\Omega$,
and that $$F(x,u^+)\leq \f{\lambda}{p} (u^+)^p+C{(u^+)^{p+1}}.$$
This together with (\ref{p-eig}), the definition of $\lambda_p(\Omega)$,
and Theorem \ref{compactlemma} leads to
$$\int_\Omega F(x,u^+)d\mu\leq \f{\lambda}{p\lambda_p(\Omega)}\int_\Omega|\nabla u^+|^pd\mu+
C\le(\int_\Omega|\nabla u^+|^pd\mu\ri)^{1+1/p}.$$
Here and throughout this paper, we often denote various constants by the same $C$. Noting that
\bna
|\nabla u|^2(x)&=&\f{1}{2\mu(x)}\sum_{y\sim x}w_{xy}(u(y)-u(x))^2\\
&=&\f{1}{2\mu(x)}\sum_{y\sim x}w_{xy}\le\{(u^+(y)-u^+(x))^2+(u^-(y)-u^-(x))^2\ri.\\
&&\quad\quad\quad\le.-2u^+(y)u^-(x)-2u^+(x)u^-(y)\ri\}\\
&\geq&|\nabla u^+|^2(x),
\ena
we have $|\nabla u^+|^p\leq |\nabla u|^p$. Hence if $\|u\|_{W_0^{1,p}(\Omega)}\leq 1$,
then we obtain
$$J_p(u)\geq \f{\lambda_p(\Omega)-\lambda}{p\lambda_p(\Omega)}\int_\Omega|\nabla u|^pd\mu
-C\le(\int_\Omega|\nabla u|^pd\mu\ri)^{1+1/p}.$$
Therefore
\be\label{p-1-0}\inf_{\|u\|_{W_0^{1,p}(\Omega)}=r}J_p(u)>0\ee
provided that $r>0$ is sufficiently small.
By $(H_3)$, there exist two positive constants $c_1$ and $c_2$ such that
$$F(x,u^+)\geq c_1(u^+)^{q}-c_2.$$
Take $u_0\in W_0^{1,p}(\Omega)$ such that $u_0\geq 0$ and $u_0\not\equiv 0$. For any $t>0$, we have
$$J_p(tu_0)\leq\f{t^p}{p}\int_\Omega|\nabla u_0|^pd\mu-c_1t^{q}\int_\Omega u_0^{q}d\mu
-c_2|\Omega|.$$
Since $q>p$, we conclude $J_p(tu_0)\ra -\infty$ as $t\ra+\infty$. Hence there exists some $u_1\in W_0^{1,p}(\Omega)$ satisfying
\be\label{p-2-0}J_p(u_1)<0,\quad \|u_1\|_{W_0^{1,p}(\Omega)}>r.\ee
We claim that $J_p$ satisfies the $(PS)_c$ condition for any $c\in\mathbb{R}$.
To see this, we assume $J_p(u_k)\ra c$ and $J_p^\prime(u_k)\ra 0$ as $k\ra+\infty$. It follows that
\bna
&&\f{1}{p}\int_\Omega|\nabla u_k|^pd\mu-\int_\Omega F(x,u_k^+)d\mu=c+o_k(1)\\[1.5ex]
&&\int_\Omega|\nabla u_k|^pd\mu-\int_\Omega u_kf(x,u_k^+)d\mu=o_k(1)\|u_k\|_{W_0^{1,p}(\Omega)}.
\ena
In view of $(H_3)$, we obtain from the above two equations that $u_k$ is bounded in $W^{1,p}_0(\Omega)$.
Then the $(PS)_c$ condition follows from Theorem \ref{compactlemma}.
Combining (\ref{p-1-0}), (\ref{p-2-0}) and the obvious fact $J_p(0)=0$, we conclude from Theorem \ref{Mountain-pass-Theorem} that
there exists a function $u\in W_0^{1,p}(\Omega)$ such that $J_p(u)=\inf_{\gamma\in\Gamma}\max_{t\in[0,1]}J_p(\gamma(t))>0$
and $J_p^\prime(u)=0$, where $\Gamma=\{\gamma\in C([0,1],W_0^{1,p}(\Omega)): \gamma(0)=0,\gamma(1)=u_1\}$. Hence
there exists a nontrivial solution $u\in W_0^{1,p}(\Omega)$ to the equation
$$-\Delta_p u=f(x,u^+)\quad{\rm in}\quad \Omega^\circ.$$
Noting that $\Gamma(u^-,u)=\Gamma(u^-)+\Gamma(u^-,u^+)\geq |\nabla u^-|^2$, we obtain
$$\int_\Omega|\nabla u^-|^pd\mu\leq \int_\Omega|\nabla u^-|^{p-2}\Gamma(u^-,u)d\mu=
-\int_\Omega u^-\Delta_p ud\mu=\int_\Omega u^-f(x,u^+)dx=0.$$
This implies that $u^-\equiv 0$ and thus $u\geq 0$. $\hfill\Box$\\
{\it Proof of Theorem \ref{Theorem 4}}. Let $m\geq 2$ and $p>1$ be fixed. For any $u\in W_0^{m,p}(\Omega)$, we write
$$J_{mp}(u)=\f{1}{p}\int_\Omega|\nabla^mu|^pd\mu-\int_\Omega F(x,u)d\mu.$$
By $(A_2)$, there exists some $\lambda<\lambda_{mp}(\Omega)$ and $\delta>0$ such that for all $s\in\mathbb{R}$,
$$F(x,s)\leq \f{\lambda}{p}|s|^p+\f{|s|^{p+1}}{\delta^{p+1}}|F(x,s)|.$$
By Theorem \ref{compactlemma}, we have
\bea\label{p+1}
&&\int_\Omega|u|^{p+1}d\mu\leq C\le(\int_\Omega|\nabla^mu|^pd\mu\ri)^{1+1/p},\\
&&\label{infty}\|u\|_{L^\infty(\Omega)}\leq C\le(\int_\Omega|\nabla^mu|^pd\mu\ri)^{1/p}.
\eea
For any $u\in W_0^{m,p}(\Omega)$ with $\|u\|_{W_0^{m,p}(\Omega)}\leq 1$,
in view of (\ref{infty}), there exists some constant $C$, depending only on $m$, $p$ and $\Omega$, such that
$\|u\|_{L^\infty(\Omega)}\leq C$ and thus $\|F(x,u)\|_{L^\infty(\Omega)}\leq C$. This together with (\ref{p+1}) gives
\bna
\int_\Omega F(x,u)d\mu&\leq&\f{\lambda}{p}\int_\Omega |u|^pd\mu+\f{C}{\delta^{p+1}}\int_\Omega|u|^{p+1}|F(x,u)|d\mu\\
&\leq&\f{\lambda}{\lambda_{mp}(\Omega)p}\int_\Omega|\nabla^mu|^pd\mu+C\le(\int_\Omega|\nabla^mu|^pd\mu\ri)^{1+1/p}.
\ena
Hence
$$J_{mp}(u)\geq \f{\lambda_{mp}(\Omega)-\lambda}{\lambda_{mp}(\Omega)p}\int_\Omega|\nabla^mu|^pd\mu
-C\le(\int_\Omega|\nabla^mu|^pd\mu\ri)^{1+1/p},$$
and thus
\be\label{lowerb}\inf_{\|u\|_{W_0^{m,p}(\Omega)}=r}J_{mp}(u)>0\ee
for sufficiently small $r>0$.
By $(A_3)$, there exist two positive constants $c_1$ and $c_2$ such that
$$F(x,s)\geq c_1|s|^{q}-c_2,\quad\forall s\in \mathbb{R}.$$
For any fixed $u\in W_0^{m,p}(\Omega)$ with $u\not\equiv 0$, we have
$$\label{e-0}J_{mp}(tu)\leq\f{|t|^p}{p}\int_\Omega|\nabla^mu|^pd\mu-c_1|t|^{q}\int_\Omega|u|^{q}d\mu+c_2|\Omega|\ra-\infty$$
as $t\ra\infty$. Hence there exists some $u_2\in W_0^{m,p}(\Omega)$ such that
\be\label{less}J_{mp}(u_2)<0,\quad \|u_2\|_{W_0^{m,p}(\Omega)}>r.\ee
Now we claim that $J_{mp}$ satisfies the $(PS)_c$ condition for any $c\in\mathbb{R}$. For this purpose, we take
$u_k\in W_0^{m,p}(\Omega)$ such that $J_{mp}(u_k)\ra c$ and $J^\prime_{mp}(u_k)\ra 0$ as $k\ra+\infty$. In particular,
\bea
\f{1}{p}\int_\Omega|\nabla^mu_k|^pd\mu-\int_\Omega F(x,u_k)d\mu=c+o_k(1),\label{p-s-1}\\[1.5ex]
\int_\Omega|\nabla^mu_k|^pd\mu-\int_\Omega u_kf(x,u_k)d\mu=o_k(1)\|u_k\|_{W_0^{m,p}}.\label{p-s-2}
\eea
By $(A_3)$, we have
\bea
q\int_\Omega F(x,u_k)d\mu&\leq&q\int_{|u_k|\leq M}|F(x,u_k)|d\mu+q\int_{|u_k|>M}F(x,u_k)d\mu\nonumber\\
&\leq&\int_{|u_k|> M}u_kf(x,u_k)d\mu+C\nonumber\\
&\leq&\int_\Omega u_kf(x,u_k)d\mu+C.\label{pp1}
\eea
Combining (\ref{p-s-1}), (\ref{p-s-2}) and (\ref{pp1}), we conclude that $u_k$ is bounded in $W_0^{m,p}(\Omega)$.
Then the $(PS)_c$ condition follows from Theorem \ref{compactlemma}.
In view of (\ref{lowerb}) and (\ref{less}), applying the mountain-pass theorem as before, we finish the proof of the theorem. $\hfill\Box$
\section{Global existence }
In this section, using Theorem \ref{Mountain-pass-Theorem}, we prove global existence results (Theorems \ref{Theorem 5}-\ref{Theorem 8}).
These procedures are very similar to that of Theorems \ref{Theorem 1}-\ref{Theorem 4}. We only give the outline of the proof.\\
{\it Proof of Theorem \ref{Theorem 5}}. Let $p>2$ be fixed. For $u\in W^{1,2}(V)$, we let
$$J_V(u)=\f{1}{2}\int_V(|\nabla u|^2+hu^2)d\mu-\f{1}{p}\int_V(u^+)^pd\mu.$$
We first prove that $J_V$ satisfies the $(PS)_c$ condition for any $c\in\mathbb{R}$.
To see this, we assume $J_V(u_k)\ra c$ and $J_V^\prime(u_k)\ra 0$ as $k\ra \infty$, namely
\bea\label{psc1}&&\f{1}{2}\int_V(|\nabla u_k|^2+hu_k^2)d\mu-\f{1}{p}\int_V(u^+)^pd\mu=c+o_k(1),\\
\label{psc2}&&\le|\int_V(|\nabla u_k|^2+hu_k^2)d\mu-\int_V(u_k^+)^pd\mu\ri|\leq o_k(1)\|u_k\|_{W^{1,2}(V)}.
\eea
It follows from (\ref{psc1}) and (\ref{psc2}) that $u_k$ is bounded in $W^{1,2}(V)$. By Theorem \ref{compact-V},
$W^{1,2}(V)$ is pre-compact,
we conclude up to a subsequence, $u_k\ra u$ in $W^{1,2}(V)$. Thus $J_V$ satisfies the $(PS)_c$ condition.
To apply the mountain pass theorem, we check the conditions of $J_V$. Obviously
\be\label{zer}J_V(0)=0.\ee
Note that $h(x)>0$ for all $x\in V$. By (\ref{Wh}),
$$\|u\|_{W^{1,2}(V)}^2=\int_V(|\nabla u|^2+hu^2)d\mu,\quad \forall u\in W^{1,2}(V).$$
By the Sobolev embedding (Theorem \ref{compact-V}), we have for any $p>2$,
$$\int_V (u^+)^pd\mu\leq C\|u\|_{W^{1,2}(V)}^p.$$
Hence
\be\label{inf}\inf_{\|u\|_{W^{1,2}(V)}=r}J_V(u)>0\ee
for sufficiently small $r>0$. For any fixed $\tilde{u}\geq 0$ with $\tilde{u}\not\equiv 0$, we have
$J_V(t\tilde{u})\ra-\infty$ as
$t\ra-\infty$. Hence there exists some $e\in W^{1,2}(V)$ such that
\be\label{con3}J_V(e)<0,\quad \|e\|_{W^{1,2}(V)}>r.\ee
Combining (\ref{zer}), (\ref{inf}) and (\ref{con3}) and applying the mountain-pass theorem (Theorem \ref{Mountain-pass-Theorem}),
we find some $u\in W^{1,2}(V)\setminus\{0\}$ satisfying
$$-\Delta u+hu=(u^+)^{p-1}\quad{\rm in}\quad V.$$
Testing this equation by $u^-$, we have
$$\int_V(|\nabla u^-|^2+h(u^-)^2)d\mu\leq\int_Vu^-(-\Delta u+hu)d\mu=0,$$
which leads to $u^-\equiv 0$. Therefore $u\geq 0$ and
$$-\Delta u+hu=u^{p-1}\quad{\rm in}\quad V.$$
If $u(x)=0$ for some $x\in V$, then $\Delta u(x)=0$, which together with $u\geq 0$ implies that $u\equiv 0$ in $V$.
This contradicts $u\not\equiv 0$. Therefore $u>0$ in $V$ and this completes the proof of the theorem. $\hfill\Box$\\
{\it Proof of Theorem \ref{Theorem 6}.} Let $p>1$ be fixed. For $u\in W^{1,p}(V)$, we define
$$J_{V,p}(u)=\f{1}{p}\int_V(|\nabla u|^p+h|u|^p)d\mu-\int_VF(x,u^+)d\mu.$$
Write
$$\|u\|_{W^{1,p}(V)}=\le(\int_V(|\nabla u|^p+h|u|^p)d\mu\ri)^{1/p}.$$
By $(H_V^4)$, there exist two constants $\lambda$ and $\delta>0$ such that $\lambda<\lambda_p^h(V)$ and
$$F(x,u^+)\leq \f{\lambda}{p} (u^+)^p+\f{(u^+)^{p+1}}{\delta^{p+1}}F(x,u^+).$$
For any $u\in W^{1,p}(V)$ with $\|u\|_{W^{1,p}(V)}\leq 1$, we have by Theorem \ref{compact-V} that $\|u\|_\infty\leq C$,
and that $$F(x,u^+)\leq \f{\lambda}{p} (u^+)^p+C{(u^+)^{p+1}}.$$
This together with the definition of $\lambda_p^h(V)$ and the Sobolev embedding (Theorem \ref{compact-V}) leads to
$$\int_V F(x,u^+)d\mu\leq \f{\lambda}{p\lambda_p^h(V)}\|u^+\|_{W^{1,p}(V)}^p
+ C\|u^+\|_{W^{1,p}(V)}^{p+1}.$$
Then we obtain for all $u$ with $\|u\|_{W^{1,p}(V)}\leq 1$,
$$J_{V,p}(u)\geq \f{\lambda_p^h(V)-\lambda}{p\lambda_p^h(V)}\|u\|_{W^{1,p}(V)}^p
-C\|u\|_{W^{1,p}(V)}^{p+1}.$$
Therefore
\be\label{p-1}\inf_{\|u\|_{W^{1,p}(V)}=r}J_{V,p}(u)>0\ee
provided that $r>0$ is sufficiently small.
By the hypothesis $(H_V^3)$, there exist two positive constants $c_1$ and $c_2$ such that
$$F(x,u^+)\geq c_1(u^+)^{q}-c_2.$$
For some $u^*\in W^{1,p}(V)$ with $u^*\geq 0$ and $u^*\not\equiv 0$, we have for any $t>0$,
$$J_{V,p}(tu^*)\leq\f{t^p}{p}\|u^*\|_{W^{1,p}(V)}^p-c_1t^{q}\int_V (u^*)^{q}d\mu
-c_2|V|.$$
Hence $J_{V,p}(tu^*)\ra -\infty$ as $t\ra+\infty$, from which one can find some $e\in W^{1,p}(V)$ satisfying
\be\label{p-2}J_{V,p}(e)<0,\quad \|e\|_{W^{1,p}(V)}>r.\ee
We claim that $J_{V,p}$ satisfies the $(PS)_c$ condition for any $c\in\mathbb{R}$.
To see this, we assume $J_{V,p}(u_k)\ra c$ and $J_{V,p}^\prime(u_k)\ra 0$ as $k\ra+\infty$. It follows that
\bna
&&\f{1}{p}\|u_k\|_{W^{1,p}(V)}^p-\int_V F(x,u_k^+)d\mu=c+o_k(1)\\[1.5ex]
&&\|u_k\|_{W^{1,p}(V)}^p-\int_V u_k^+f(x,u_k^+)d\mu=o_k(1)\|u_k\|_{W^{1,p}(V)}.
\ena
In view of $(H_V^3)$, we conclude from the above two equations that $u_k$ is bounded in $W^{1,p}(V)$.
Then the $(PS)_c$ condition follows from Theorem \ref{compact-V}.
Combining (\ref{p-1}), (\ref{p-2}) and $J_{V,p}(0)=0$, we conclude from Theorem \ref{Mountain-pass-Theorem} that
there exists a nontrivial solution $u\in W^{1,p}(V)$ to the equation
$$-\Delta_p u+h|u|^{p-2}u=f(x,u^+)\quad{\rm in}\quad V.$$
Noting that $\Gamma(u^-,u)=\Gamma(u^-)+\Gamma(u^-,u^+)\geq |\nabla u^-|^2$, we obtain
$$\int_V(|\nabla u^-|^p+h|u^-|^p)d\mu\leq -\int_V u^-\Delta_p ud\mu+\int_Vhu^-|u|^{p-2}ud\mu
=\int_V u^-f(x,u^+)dx=0.$$
This implies that $u^-\equiv 0$ and thus $u\geq 0$ in $V$. $\hfill\Box$\\
{\it Proof of Theorem \ref{Theorem 8}.} Let $m\geq 2$ and $p>1$ be fixed. We define
$$J_{mp}^V(u)=\f{1}{p}\int_V(|\nabla^mu|^p+h|u|^p)d\mu-\int_V F(x,u)d\mu.$$
By $(A_V^2)$, there exists some $\lambda<\lambda_{mp}(V)$ and $\delta>0$ such that for all $x\in V$ and $s\in\mathbb{R}$,
$$F(x,s)\leq \f{\lambda}{p}|s|^p+\f{|s|^{p+1}}{\delta^{p+1}}|F(x,s)|.$$
For any $u\in W^{m,p}(V)$ with $\|u\|_{W^{m,p}(V)}\leq 1$, we have $\|u\|_\infty\leq C$ and thus $\|F(x,u)\|_\infty\leq C$. Hence
\bna
\int_V F(x,u)d\mu&\leq&\f{\lambda}{p}\int_V |u|^pd\mu+\f{C}{\delta^{p+1}}\int_V|u|^{p+1}|F(x,u)|d\mu\\
&\leq&\f{\lambda}{\lambda_{mp}(V)p}\|u\|_{W^{m,p}(V)}^p+C\|u\|_{W^{m,p}(V)}^{p+1}.
\ena
It follows that
$$J_{mp}^V(u)\geq \f{\lambda_{mp}(V)-\lambda}{\lambda_{mp}(V)p}\|u\|_{W^{m,p}(V)}^p
-C\|u\|_{W^{m,p}(V)}^{p+1},$$
and thus
\be\label{lowerb10}\inf_{\|u\|_{W^{m,p}(V)}=r}J_{mp}^V(u)>0\ee
for sufficiently small $r>0$.
By $(A_V^3)$, there exist two positive constants $c_1$ and $c_2$ such that
$$F(x,s)\geq c_1|s|^{q}-c_2,\quad\forall s\in \mathbb{R}.$$
For any fixed $u\in W^{m,p}(V)$ with $u\not\equiv 0$, we have
$$\label{e-0}J_{mp}^V(tu)\leq\f{|t|^p}{p}\int_V(|\nabla^mu|^p+h|u|^p)d\mu-
c_1|t|^{q}\int_V|u|^{q}d\mu+c_2|V|\ra-\infty$$
as $t\ra\infty$. Hence there exists some $e\in W^{m,p}(V)$ such that
\be\label{less10}J_{mp}^V(e)<0,\quad \|e\|_{W^{m,p}(V)}>r.\ee
Now we claim that $J_{mp}^V$ satisfies the $(PS)_c$ condition for any $c\in\mathbb{R}$. For this purpose, we take
$u_k\in W^{m,p}(V)$ such that $J_{mp}^V(u_k)\ra c$ and $({J_{mp}^V})^\prime(u_k)\ra 0$ as $k\ra+\infty$. In particular,
\bea
\f{1}{p}\int_V(|\nabla^mu_k|^p+h|u_k|^p)d\mu-\int_V F(x,u_k)d\mu=c+o_k(1),\label{p-s-11}\\[1.5ex]
\int_V(|\nabla^mu_k|^p+h|u_k|^p)d\mu-\int_V u_kf(x,u_k)d\mu=o_k(1)\|u_k\|_{W^{m,p}(V)}.\label{p-s-22}
\eea
By $(A_V^3)$, we have
\bea
q\int_V F(x,u_k)d\mu&\leq&q\int_{|u_k|\leq M}|F(x,u_k)|d\mu+q\int_{|u_k|>M}F(x,u_k)d\mu\nonumber\\
&\leq&\int_{|u_k|> M}u_kf(x,u_k)d\mu+C\nonumber\\
&\leq&\int_V u_kf(x,u_k)d\mu+C.\label{pp11}
\eea
Combining (\ref{p-s-11}), (\ref{p-s-22}) and (\ref{pp11}), we conclude that $u_k$ is bounded in $W^{m,p}(V)$.
Then the $(PS)_c$ condition follows from Theorem \ref{compact-V}.
In view of (\ref{lowerb10}), (\ref{less10}) and the fact $J_{mp}^V(0)=0$, applying the mountain-pass theorem, we finish the proof of the theorem. $\hfill\Box$\\
{\bf Acknowledgements.} A. Grigor'yan is partly supported by SFB 701 of the German Research Council. Y. Lin is supported by the National Science Foundation of China (Grant No.11271011). Y. Yang is supported by the National Science Foundation of China (Grant No.11171347).
\bigskip
| 2f026c2df66c134600ee27cb21d7c1f7adeab6ed | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Suppose that $F$ is a free group of finite rank, $\rr(F)$ denotes the rank of $F$, and $\brr(F) := \max
(\rr(F)-1,0)$ is the reduced rank of $F$. Let $H, K$ be finitely
generated subgroups of $F$ and let $\langle H, K \rangle$ denote the the subgroup generated by $H, K$, called the
{\em join} of $H, K$. Hanna Neumann \cite{N1} proved
that $ \brr (H \cap K) \le 2 \bar \rr(H) \bar \rr(K)$ and
conjectured that $\brr (H \cap K) \le \bar \rr(H) \bar \rr(K)$. This problem, known as the Hanna Neumann conjecture
on subgroups of free groups, was solved in the affirmative by Friedman \cite{F} and Mineyev \cite{M},
see also Dicks's proof \cite{D1}. Relevant results and generalizations of this conjecture can be found in
\cite{D}, \cite{DIv}, \cite{DIv2}, \cite{I99}, \cite{I01}, \cite{I08}, \cite{WN}, \cite{St}.
Imrich and M\"uller \cite{IM} introduced the reduced rank $\brr (\langle H, K \rangle)$ of the join $\langle H, K \rangle$ in this context and attempted to prove the following inequality
\begin{equation}\label{inq11}
\bar \rr(H \cap K) \le 2 \bar \rr(H) \bar \rr(K) - \brr (\langle H, K \rangle) \min(\bar \rr(H), \bar \rr(K))
\end{equation}
under the assumption that if $H^*, K^*$ are free factors of $H, K$, resp., then the equality
$H^*\cap K^* = H \cap K$ implies that $H= H^*$ and $K = K^*$. We remark that this inequality provides a stronger
than Hanna Neumann conjecture's bound for $\bar \rr(H \cap K)$ whenever
$\brr (\langle H, K \rangle) > \max(\bar \rr(H), \bar \rr(K))$ and this looks quite remarkable. We also note that
\eqref{inq11} was an improvement of an earlier result of Burns \cite{B}, see also \cite{N2}, \cite{S}, stating
that
\begin{equation*}
\bar \rr(H \cap K) \le 2 \bar \rr(H) \bar \rr(K) - \min(\bar \rr(H), \bar \rr(K)) .
\end{equation*}
Later Kent \cite{K} discovered a serious gap in the proof of a key lemma of Imrich--M\"uller \cite[p.~195]{IM} and gave his own proof to the inequality \eqref{inq11} under the weakened assumption that $H \cap K \ne \{ 1 \}$. Kent
\cite[p.~312]{K} remarks that the lemma in \cite{IM} ``would be quite useful, and though its proof is incorrect, we do
not know if the lemma actually fails". In this note we give an example that shows that the lemma of \cite{IM} is indeed false.
On positive side, we suggest a repair for this lemma so that the arguments of Imrich--M\"uller \cite{IM} could be saved. Our approach seems to be of independent interest and can be outlined as follows. Given subgroups $H, K$ of a free group $F$ with $H \cap K \ne \{ 1 \}$, we modify $H, K, F$ by certain deformations of Stallings graphs of $H, K, F$ so that the modified subgroups
$\wtl H, \wtl K$ would satisfy $\brr(\wtl H) = \brr( H)$, $\brr(\wtl K) = \brr( K)$, $\bar \rr(\wtl H \cap \wtl K) = \bar \rr(H \cap K)$, and
$\brr (\langle \wtl H, \wtl K \rangle) \ge \brr (\langle \wtl H, \wtl K \rangle)$. Furthermore, our modification
is done so that Stallings graphs of subgroups $\langle \wtl H, \wtl K \rangle$, $\wtl H, \wtl K$ have vertices of degree
2 or 3 only and every vertex of the graph of $\langle \wtl H, \wtl K \rangle$ of degree 3 has a preimage of degree 3
in the graph of $\wtl H$ or $ \wtl K$. These properties mean that the lemma of Imrich--M\"uller \cite[p.~195]{IM} holds in a modified setting and the arguments of Imrich--M\"uller are retained otherwise. As an application of this strategy developed in Sect.~3, we will prove Theorem~\ref{th1} in Sect.~4.
Theorem~\ref{AB} in Sect.~5 deals with a natural question related to our main construction
used in proofs of key Lemmas~\ref{ED}--\ref{mn}.
\section{A Counterexample to Imrich--M\"uller Lemma}
Similarly to Stallings \cite{St}, see also \cite{D}, \cite{K}, \cite{KM}, we consider
finite graphs associated with finitely generated subgroups of a free group $F$.
We consider the ambient free group $F$ as the fundamental group of a finite connected graph $U$, $F = \pi_1(U)$, without vertices of degree 1. Let $H, K$ be finitely generated subgroups of $F$ and let $X, Y$ denote
finite graphs associated with $H, K$, resp. Recall that there are locally injective graph maps $\ph_X : X \to U$,
$\ph_Y : Y \to U$. Conjugating $H, K$ if necessary, we may assume that the graphs $X, Y$ have no vertices of degree 1, i.e., $\core (X) = X$ and $\core (Y) = Y$, where $\core(\Gamma )$ is the subgraph of a graph $\Gamma $ consisting of all edges
that can be included into circuits of $\Gamma $.
Let $S(H,K)$ denote a set of representatives of those double cosets $H tK \subseteq F$, $t \in F$,
that have the property $H t Kt^{-1} \ne \{ 1\}$. Recall that the connected components of the core $W := \core (X \times_U Y)$ of the pullback $X \times_U Y$ of graph maps $\ph_X : X \to U$, $\ph_Y : Y \to U$
are in bijective correspondence with elements of the set $S(H,K)$, see \cite{D}, \cite{KM}, \cite{WN}. Hence, we can write
\begin{equation*}
W = \bigvee_{t \in S(H,K)} W_t ,
\end{equation*}
where $\bigvee$ denotes a disjoint union. In addition, if $W_t$ is a \cncm\ of $W$ that corresponds to $t \in S(H,K)$, then
\begin{equation*}
\brr ( H \cap tKt^{-1} ) = | E W_t | - | V W_t | ,
\end{equation*}
where $E \Gamma $ is the set of (nonoriented) edges of a graph $\Gamma $, $V \Gamma $ is the set of vertices of $\Gamma $ and $|A|$ is the cardinality of a set $A$. Using notation $\brr (\Gamma) := | E \Gamma | - | V \Gamma |$, we have
\begin{equation*}
\brr (W) = \sum_{t \in S(H,K) } \brr( W_t ) = \sum_{t \in S(H,K) } \brr ( H \cap tKt^{-1} ) .
\end{equation*}
For $S_1 \subseteq S(H,K)$, $S_1 \ne \varnothing$, denoting $W(S_1) := \bigvee_{s \in S_1} W_s$, we obtain
\begin{equation}\label{inq2}
\brr (W(S_1) ) = \sum_{s \in S_1 } \brr( W_s ) = \sum_{t \in S_1 } \brr ( H \cap sKs^{-1} ) = \brr(H, K, S_1) .
\end{equation}
Let $\al_X : W(S_1) \to X$, $ \al_Y : W(S_1) \to Y$ denote the restrictions on $W(S_1) \subseteq X \x_U Y$ of the pullback projection maps
$$
\bar \al_X : X \x_U Y \to X , \quad \bar \al_Y : X \x_U Y \to Y ,
$$
resp. Consider the pushout $P$ of these maps $\al_X : W(S_1) \to X$, $ \al_Y : W(S_1) \to Y$. The corresponding
pushout maps are denoted $\beta_X : X \to P$, $\beta_Y : Y \to P$. We also consider the Stallings graph $Z$ corresponding to the subgroup $\langle H, K, S_1 \rangle$ of $F$. It is clear from the definitions that there are
graph maps $\gamma : P \to Z$ and $\delta : Z \to U$ such that the diagram depicted in Fig.~1 is commutative.
\begin{center}
\begin{tikzpicture}[scale=.57]
\draw [-latex](.5,.5) -- (1.5, 1.5);
\draw [-latex](.5,-.5) -- (1.5, -1.5);
\draw [-latex](2.5,1.5) -- (3.5, 0.5);
\draw [-latex](2.5,-1.5) -- (3.5, -0.5);
\draw [-latex](4.5,0) -- (5.5, 0);
\draw [-latex](6.5,0) -- (7.5, 0);
\node at (-.2,0) {$W(S_1)$};
\node at (2,2) {$X$};
\node at (2,-2) {$Y$};
\node at (.6,1.4) {$\alpha_X$};
\node at (.6,-1.4) {$\alpha_Y$};
\node at (3.5,1.4) {$\beta_X$};
\node at (3.5,-1.4) {$\beta_Y$};
\node at (5.,0.5) {$\gamma$};
\node at (7.,0.5) {$\delta$};
\node at (4,0) {$P$};
\node at (6,0) {$Z$};
\node at (8,0) {$U$};
\node at (4.5,-3) {Figure 1};
\end{tikzpicture}
\end{center}
As was found out by Kent \cite[Fig.~1]{K}, one has to distinguish between $P$ and $Z$ as $P \ne Z$ in general. Since the map $\delta : Z \to U$ is locally injective and $P \ne Z$, we see that
the map $\gamma : P \to Z$ need not be locally injective, hence, $\gamma$ factors out through a sequence of edge
foldings and $\brr(P) \ge \brr(Z)$. According to Kent \cite{K}, the erroneous identification $P = Z$ is the source of a mistake in a key lemma of Imrich--M\"uller \cite[p.~195]{IM}. Recall that this lemma claims, in our terminology, that if
$Z$ is an {\em almost trivalent} graph, i.e., every vertex of $Z$ has degree 3 or 2, then every vertex of $Z$ of degree 3, has a preimage of degree 3 in $X$ or $Y$. Kent \cite{K} explains in detail a mistake in the proof of this lemma and comments that, while
the proof of lemma of \cite[p.~195]{IM} can be somewhat corrected if one replaces $Z$ by $P$, the subsequent arguments of
Imrich--M\"uller rely on the property that $Z=U$ is trivalent, the property that the graph $P$ does not possess. Furthermore, Kent \cite[p.~312]{K} points out that the lemma in \cite{IM} ``would be quite useful, and though its proof is incorrect, we do not know if the lemma actually fails".
We now present an example that shows that the lemma of \cite{IM} does fail. Consider subgroups
\begin{align}
H & = \langle b_1b_2 a_2 b_2^{-1} b_1^{-1}, \ a_1^2, \ a_1b_1b_3 a_3 (a_1b_1b_3)^{-1} \rangle , \\
K & = \langle b_1b_2 a_2 b_2^{-1} b_1^{-1}, \ a_1^3, \ a_1 b_1b_3 a_3 (a_1b_1b_3)^{-1} \rangle
\end{align}
of a free group $F= \pi_1(U)$ whose graphs $ X, Y, Z$, resp., are depicted in Fig.~2, where the base vertices of
$ X, Y, Z$ are dashed circled.
\begin{center}
\begin{tikzpicture}[scale=.57]
\draw (-2,0)[fill = black]circle (0.06);
\draw (-1,0)[fill = black]circle (0.06);
\draw (0,0)[fill = black]circle (0.06);
\draw (2,0)[fill = black]circle (0.06);
\draw (3,0)[fill = black]circle (0.06);
\draw (4,0)[fill = black]circle (0.06);
\draw (0,0)[dashed]circle (0.2);
\draw[-latex](-3.1,1) -- (-2.95,1);
\draw[-latex](0.95,1) -- (1.1,1);
\draw[-latex](4.95,1) -- (5.1,1);
\draw[-latex](1.1,-1) -- (0.95,-1);
\draw[-latex](-1.5,0) -- (-1.6,0);
\draw[-latex](-0.5,0) -- (-0.6,0);
\draw[-latex](2.5,0) -- (2.6,0);
\draw[-latex](3.5,0) -- (3.6,0);
\draw(-2,0) -- (0,0);
\draw (-3,0) ellipse (1 and 1);
\draw (1,0) ellipse (1 and 1);
\draw(2,0) -- (4,0);
\draw (5,0) ellipse (1 and 1);
\node at (-3,1.5) {$a_2$};
\node at (1,1.5) {$a_1$};
\node at (1,-1.5) {$a_1$};
\node at (5,1.5) {$a_3$};
\node at (-1.5,0.5) {$b_2$};
\node at (-0.5,0.5) {$b_1$};
\node at (2.5,0.5) {$b_1$};
\node at (3.5,0.5) {$b_3$};
\node at (-1.5,-1.5) {$X$};
\begin{scope}[shift={(12,0)}]
\draw (-2,0)[fill = black]circle (0.06);
\draw (-1,0)[fill = black]circle (0.06);
\draw (0,0)[fill = black]circle (0.06);
\draw (2,0)[fill = black]circle (0.06);
\draw (3,0)[fill = black]circle (0.06);
\draw (4,0)[fill = black]circle (0.06);
\draw (0,0)[dashed]circle (0.2);
\draw[-latex](-3.1,1) -- (-2.95,1);
\draw[-latex](0.95,1) -- (1.1,1);
\draw[-latex](4.95,1) -- (5.1,1);
\draw[-latex](-1.5,0) -- (-1.6,0);
\draw[-latex](-0.5,0) -- (-0.6,0);
\draw[-latex](2.5,0) -- (2.6,0);
\draw[-latex](3.5,0) -- (3.6,0);
\draw (1,-1)[fill = black]circle (0.06);
\draw[-latex](1.73,-.7) -- (1.6,-.8);
\draw[-latex](.3,-.7) -- (.2,-.6);
\node at (2.1,-1.1) {$a_1$};
\node at (-.1,-1.1) {$a_1$};
\node at (-1.5,-1.5) {$Y$};
\draw(-2,0) -- (0,0);
\draw (-3,0) ellipse (1 and 1);
\draw (1,0) ellipse (1 and 1);
\draw(2,0) -- (4,0);
\draw (5,0) ellipse (1 and 1);
\node at (-3,1.5) {$a_2$};
\node at (1,1.5) {$a_1$};
\node at (5,1.5) {$a_3$};
\node at (-1.5,0.5) {$b_2$};
\node at (-0.5,0.5) {$b_1$};
\node at (2.5,0.5) {$b_1$};
\node at (3.5,0.5) {$b_3$};
\end{scope}
\draw (5,-4) ellipse (1 and 1);
\draw (9,-4) ellipse (1 and 1);
\draw (7,-8) ellipse (1 and 1);
\draw (1,0) ellipse (1 and 1);
\draw(7.,-7) -- (7,-5.5);
\draw (5.7,-4.7) -- (7,-5.5);
\draw (8.3,-4.7) -- (7,-5.5);
\draw[-latex](7,-6.3) -- (7,-6.1);
\draw[-latex](6.38,-5.12)--(6.18,-5.0);
\draw[-latex](7.65,-5.12)--(7.8,-5.04);
\draw (5,0) ellipse (1 and 1);
\draw (7,-5.5)[fill = black]circle (0.06);
\draw (5.7,-4.7)[fill = black]circle (0.06);
\draw (8.3,-4.7)[fill = black]circle (0.06);
\draw (7,-7)[fill = black]circle (0.06);
\draw (7,-7)[fill = black]circle (0.06);
\draw (7,-7)[dashed]circle (0.2);
\draw(-2,0) -- (0,0);
\draw[-latex](4.95,-3) -- (5.1,-3);
\draw[-latex](8.95,-3) -- (9.1,-3);
\draw[-latex](6,-8.1) -- (6,-7.9);
\node at (5,-2.5) {$a_2$};
\node at (9,-2.5) {$a_3$};
\node at (5.5,-8) {$a_1$};
\node at (6,-5.5) {$b_2$};
\node at (8,-5.5) {$b_3$};
\node at (7.5,-6.2) {$b_1$};
\node at (4.5,-6.5) {$Z$};
\node at (7,-10.25) {Figure 2};
\end{tikzpicture}
\end{center}
\noindent
It is easy to check that $S(H,K)$ has a single element, say $S(H,K) = \{ 1\}$, and
$$
H \cap K = \langle b_1b_2 a_2 b_2^{-1} b_1^{-1}, a_1^6, a_1b_1b_3 a_3 (a_1b_1b_3)^{-1} \rangle .
$$
Furthermore, the graphs $W, P$ look like those in Fig.~3. Hence, we can see that $Z$ is a trivalent graph that has a vertex of degree 3, which is the center of $Z$, without a preimage of degree 3 in $X \vee Y$, as desired.
\begin{center}
\begin{tikzpicture}[scale=.52]
\draw (-2,0)[fill = black]circle (0.06);
\draw (-1,0)[fill = black]circle (0.06);
\draw (0,0)[fill = black]circle (0.06);
\draw (2,0)[fill = black]circle (0.06);
\draw (3,0)[fill = black]circle (0.06);
\draw (4,0)[fill = black]circle (0.06);
\draw (0,0)[dashed]circle (0.2);
\draw[-latex](-3.1,1) -- (-2.95,1);
\draw[-latex](0.95,1) -- (1.1,1);
\draw[-latex](4.95,1) -- (5.1,1);
\draw[-latex](1.1,-1) -- (0.95,-1);
\draw[-latex](-1.5,0) -- (-1.6,0);
\draw[-latex](-0.5,0) -- (-0.6,0);
\draw[-latex](2.5,0) -- (2.6,0);
\draw[-latex](3.5,0) -- (3.6,0);
\draw(-2,0) -- (0,0);
\draw (-3,0) ellipse (1 and 1);
\draw (1,0) ellipse (1 and 1);
\draw(2,0) -- (4,0);
\draw (5,0) ellipse (1 and 1);
\node at (-3,1.5) {$a_2$};
\node at (1,1.5) {$a_1$};
\node at (1,-1.5) {$a_1^5$};
\node at (5,1.5) {$a_3$};
\node at (-1.5,0.5) {$b_2$};
\node at (-0.5,0.5) {$b_1$};
\node at (2.5,0.5) {$b_1$};
\node at (3.5,0.5) {$b_3$};
\node at (-1.5,-1.5) {$W$};
\draw (9.5,1) ellipse (1 and 1);
\draw (13.5,1) ellipse (1 and 1);
\draw (11.5,-3) ellipse (1 and 1);
\draw (1,0) ellipse (1 and 1);
\draw(11.5,-2) -- (10.2,0.3);
\draw(11.5,-2) -- (12.8,0.3);
\draw[-latex](11.2,-1.5) -- (11.1,-1.3);
\draw[-latex](11.8,-1.5) -- (11.9,-1.3);
\draw[-latex](10.57,-0.4)--(10.48,-.2);
\draw[-latex](12.4,-0.4)--(12.52,-0.2);
\draw (5,0) ellipse (1 and 1);
\draw (10.85,-0.88)[fill = black]circle (0.06);
\draw (12.15,-0.88)[fill = black]circle (0.06);
\draw (10.2,0.3)[fill = black]circle (0.06);
\draw (12.8,0.3)[fill = black]circle (0.06);
\draw (11.5,-2)[fill = black]circle (0.06);
\draw (11.5,-2)[fill = black]circle (0.06);
\draw (11.5,-2)[dashed]circle (0.2);
\draw(-2,0) -- (0,0);
\draw[-latex](9.45,2) -- (9.6,2);
\draw[-latex](13.45,2) -- (13.6,2);
\draw[-latex](10.5,-3.1) -- (10.5,-2.9);
\node at (9.5,2.5) {$a_2$};
\node at (13.5,2.5) {$a_3$};
\node at (10,-3) {$a_1$};
\node at (10.,-0.5) {$b_2$};
\node at (13,-0.5) {$b_3$};
\node at (12.5,-1.6) {$b_1$};
\node at (10.5,-1.6) {$b_1$};
\node at (9,-1.5) {$P$};
\node at (6.5,-4) {Figure 3};
\end{tikzpicture}
\end{center}
\section{Fixing the Lemma of Imrich and M\"uller}
We now discuss how to do certain deformations over the graphs $W(S), X$, $Y, P$, $Z, U$ to achieve the situation when
$P = Z =U$, $U$ is almost trivalent and lemma of \cite[p.~195]{IM} would hold for $Z$.
Our idea could be illustrated by the remark that, when studying graphs $W(S)$, $X$, $Y, P$, $Z, U$, or corresponding subgroups, we can replace the ambient group $F = \pi_1(U)$ by $\wht F = \pi_1(Z)$, i.e., we can replace the graph $U$ by $Z$.
We can go further and replace the group $F = \pi_1(U)$ with $\wtl F = \pi_1(P)$ by using the pushout graph $P$ in place of $U$.
Either of these replacements $U \to Z$, $U \to P$ results in obvious cosmetic changes to subgroups $H, K$, to sets $S(H,K)$, $S_1 \subseteq S(H,K)$, and to subgroups $ H \cap sKs^{-1}$, $s \in S_1$, $\langle H, K, S_1 \rangle$.
Either of these replacements $U \to Z$, \ $U \to P$ preserves the graphs $W(S)$, $X, Y$, $P$ and the maps
$\al_X, \al_Y, \beta_X, \beta_Y$. The replacement $U \to Z$ turns $\delta$ into the identity map and the replacement
$U \to P$ turns both $\delta, \gamma$ into the identity maps. Otherwise, the structure of the diagram depicted on Fig.~1 is retained. In particular, the ranks $\brr(H), \brr(K)$, $\brr(W(S))$ do not change but the rank $\brr(P)$ could increase
as $\brr(P) \ge \brr(Z)$ originally.
For future references, we record this replacement in the following.
\begin{lemma}\label{rpl} Replacing the graph $U$ in the diagram depicted in Fig.~1 by $P$ and
changing the maps $\gamma : P \to Z$, $\delta: Z \to U$ by $\gamma = \delta = \mbox{id}_{U}$, resp., and
changing the maps $\ph_X : X \to U$, $\ph_Y : Y \to U$ by $\beta_X : X \to P$, $\beta_Y : Y \to P$, resp.,
preserve the properties that $W(S_1) = \bigvee_{s \in S_1} W_{s}$ consists of connected components of
$\core(X \x_{U} Y)$ and that $P = X \vee_{W(S_1)} Y$. In particular, this replacement does not
change the ranks $\brr(X), \brr(Y), \brr(W(S_1))$ but could increase $\brr(Z)$.
\end{lemma}
\begin{proof} If $(e_1, e_2) $ is an edge of $W(S_1) \subseteq \core(X \x_{U} Y)$, where $\ph_X \al_X((e_1, e_2)) = e_1$ is an edge of $X$ and $\ph_Y \al_Y((e_1, e_2)) = e_2$ is an edge of $Y$, then the edges $e_1, e_2$ are identified in $P$, whence, $(e_1, e_2) $ is also an edge of $X \x_{P} Y$. Since $\core (W(S_1) ) = W(S_1)$, it follows that
$\core(X \x_{P} Y)$ contains all edges $(e_1, e_2) $ of $W(S_1)$, hence, $\core(X \x_{P} Y)$ will contain the graphs naturally isomorphic to \cncm s of the graph $W(S_1) = \bigvee_{s\in S_1} W_s$. Therefore, the original pushout
$P = X \vee_{W(S_1)} Y$ will also be preserved.
Since $\brr(P) \ge \brr(Z)$ for the original graphs $P, Z$ and since $P = Z = U$ after the replacement $U \to P$, we see that the rank $\brr(Z)$ could increase after the replacement.
\end{proof}
Suppose that $v$ is a vertex of $U$. Subdividing the edges incident to $v$ if necessary, we may assume that
every oriented edge $e$ of $U$ whose terminal vertex, denoted $e_+$, is $e_+ = v$ has the initial vertex, denoted $e_-$, of degree 2, $\deg e_- = 2$. Hence, we may assume that $U$ contains a subgraph $\St(v)$ isomorphic to a star
with $\deg v$ rays whose center is $v$. Let $T$ be a tree whose vertices of degree 1 are in bijective correspondence with
vertices of degree 1 of $\St(v)$ whose set we denote by $V_1 \St(v)$. Taking $\St(v)$ out of $U$ and putting the tree $T$ in place of $\St(v)$, using the bijective correspondence to identify vertices of degree 1 of $(U \setminus \St(v)) \cup V_1\St(v)$ and those of $T$, results in a transformation of $U$ which we call an {\em elementary deformation} of $U$ around $v$ by means of $T$.
It is clear that the obtained graph $U_T : = (U \setminus \St(v))\cup T$ is homotopically equivalent to $U$ and
$\pi_1(U_T)$ is isomorphic to $F = \pi_1(U)$. We lift this \ED\ of the star around $v$ in $U$ to all the graphs $X, Y, W, Z$ by replacement of stars around preimages of the vertex $v$ by trees isomorphic to suitable subtrees of $T$ and change, accordingly, the maps $\al_X, \al_Y$, $\beta_X, \beta_Y$,
$\ph_X, \ph_Y$. The new graphs and maps obtained this way we denote by $X_T, Y_T$, $W(S_1)_T$, $P_{T}, Z_T$,
$\al_{X_T}, \al_{Y_T}$, $\beta_{X_T}, \beta_{Y_T}$,
$\ph_{X_T}, \ph_{Y_T}$, $\gamma_T, \delta_T$.
Clearly, $\brr(Q_T) = \brr(Q)$, where $Q \in \{ X, Y, W, Z, U\}$.
As far as the new pushout graph $P_T$ is concerned, we can only claim that $\brr(P_{T}) \ge \brr(P)$
which would be analogous to the following.
\begin{lemma}\label{ZT} Let $P = Z = U$, let $\gamma$, $\delta$ be identity maps and let the graphs
$X_T, \ldots$, $U_T$ be obtained from $X, \ldots$, $U$ by an \ED\ around a vertex $v \in VU$ by means of a tree $T$.
Then the map $\delta_T : Z_T \to U_T$ is an isomorphism, i.e., the graphs
$Z_T, U_T$ are naturally isomorphic. Furthermore, the restriction of the map
$\delta_T\gamma_T : P_{T} \to U_T$ is bijective on the set $ \gamma_T^{-1}\delta_T^{-1} (U_T \setminus T)$ and the subgraph $ \widehat T := \gamma_T^{-1}\delta_T^{-1} (T)$ of $P_{T}$ is connected. In particular, $\brr(P_{T}) \ge \brr(P)$ and the equality holds if and only if $ \widehat T$ is a tree.
\end{lemma}
\begin{proof} It follows from the definitions that if we collapse edges of $T \subset U_T$ into a point then we obtain the graph $U$ back. Similarly, collapsing lifts of the edges of $T$ in $ Q_T$, $Q \in \{ X, Y, W(S_1), P \}$, into points, we obtain the original graph $Q$. This observation, together with the local injectivity of the map $\delta_T$, implies that
$Z_T = U_T$ and that the graph $ \widehat T := \gamma_T^{-1}\delta_T^{-1} (T)$ is connected.
Furthermore, the graph $P_{T}$ consists of a subgraph isomorphic to $ (U_T \setminus T ) \cup V_1T $, where $V_1T $ is the set of vertices of $T$ of degree 1, along with the graph $\widehat T= \gamma_T^{-1}\delta_T^{-1} (T)$ which is mapped by $ \delta_T\gamma_T$ to $T$. Surjectivity of the restriction of $\delta_T\gamma_T$ on $\widehat T$ follows from connectedness of $\widehat T$.
\end{proof}
The graph $\widehat T$ of Lemma~\ref{ZT} can be regarded as a ``blow-up" of the vertex $v \in VU$. If $\widehat T$ turns out to be a tree, then our attempt to nontrivially ``blow up" the vertex $v$ is unsuccessful. On the other hand, if $\widehat T $ is not a tree, then we can increase $\brr (U)$ by invoking Lemma~\ref{rpl} and picking $P_{T}$ in place of $U$.
It is of interest to note that even when $P = Z = U$ and $\brr(P_{T}) = \brr(P)$, i.e., $ \widehat T$ is a tree,
the tree $ \widehat T$ might look different from $T$, in this connection, see Lemma~\ref{ED} and its proof.
The following is due to Kent \cite{K}. We provide a proof for completeness.
\begin{lemma}\label{knt} Every vertex of $P$ of degree at least $3$ has a preimage in $X \vee Y$ of degree at least $3$.
\end{lemma}
\begin{proof} Arguing on the contrary, assume that $u \in V P$ has degree $\deg u >2$ and every lift $v \in VX \vee VY$ of $u$ has degree 2. Let $v, v' \in VX \vee VY$ be two arbitrary lifts of $u$. It follows from the definitions of $ P, W(S_1)$ that there is a sequence of vertices $v_1 = v , \dots, v_k = v' $ in $VX \vee VY$ with the following properties:
(a) The vertices $v_1 = v , \dots, v_k = v' $ are mapped by
$\beta_X, \beta_Y$ to $u$ and have degree 2;
(b) For every $i =1, \dots, k-1$, the vertices $v_i$, $v_{i+1}$ belong to distinct connected components of $X \vee Y$;
(c) For every $i =1, \dots, k-1$, there is a vertex $w_i \in VW$ of degree 2 such that either $\alpha_X(w_i) = v_i$, $\alpha_Y(w_i) = v_{i+1}$ if $v_i \in VX$ or $\alpha_Y(w_i) = v_i$, $\alpha_X(w_i) = v_{i+1}$ if $v_i \in VY$.
It is clear from these properties and from the definitions of the graphs $ P, W(S_1)$ that the vertices $v_1 = v , \dots, v_k = v' \in VX \vee VY $ will be identified in the pushout $P$ so that the resulting vertex will have degree 2. Since $v, v' \in VX \vee VY$ were chosen arbitrarily, it follows that the degree of $u \in V P $ is also 2. This contradiction to $\deg u >2$ proves our claim.
\end{proof}
We say that $T$ is a {\em trivalent tree} if every vertex of $T$ has degree 1 or 3.
\begin{lemma}\label{ED} Suppose that $P = Z = U$ and $v_0$ is a vertex of $U$ of degree at least 4.
Then there is an \ED\ of $U$ around $v_0$ by means of a trivalent tree $T$ such that either $\brr(P_{T}) > \brr(P)$
or the following are true: $\brr(P_{T}) = \brr(P)$, the subgraph $ \widehat T = \gamma_T^{-1}\delta_T^{-1} (T)$ of $P_{T}$ is a tree
and
$$
\sum_{v \in V P_{T} } \max(\deg v -3, 0) < \sum_{v \in V P } \max(\deg v -3, 0) .
$$
\end{lemma}
\begin{proof} Denote $\deg v_0 = m_0 >3$. Let $e_1, \dots, e_{m_0}$ be all of the oriented edges of $U$ that end in $v_0$.
By Lemma~\ref{knt}, there is a vertex $v_1 \in VX \vee VY$, say $v_1 \in VX$, such that $\beta_X(v_1) = v_0$ and $\deg v_1 = m_1$, where $3 \le m_1 \le m_0$. Reindexing if necessary, we may assume that if $f_1, \dots , f_{m_1} $ are all of the edges of $X$ that end in $v_1$, then $\beta_X(f_i) = e_i$, $i =1, \dots, m_1$.
Let $T$ be a trivalent tree with $m_0$ vertices of degree 1.
Let $U_T$ denote the graph obtained from $U$ by an \ED\ around $v_0$ by means of $T$.
Let $u_i$ denote the vertex of $T$ which becomes $(e_i)_-$ in $U_T$, $i =1, \dots, m_0$.
Here we are assuming that $\deg(e_i)_- = 2$ for every $i$ as in the definition of an \ED\ around $v$.
It follows from Lemma~\ref{ZT} that if the subgraph $\widehat T = \gamma_T^{-1} \delta_T^{-1}(T) $ is not a tree for some $T$, that is, if $\brr(\widehat T) \ge 0$, then $\brr(\widehat Z_T) > \brr(\widehat Z)$ and our lemma is proven. Hence, we may assume that $\widehat T $ is a tree for every trivalent tree $T$.
Suppose that $\widehat T $ contains no vertex of degree $m_0$ for some $T$. Then, obviously,
$$
\sum_{v \in V U_T} \max( \deg v -3, 0) < \sum_{v \in VU} \max( \deg v -3, 0)
$$
and our lemma is true.
Thus we may suppose that, for every trivalent tree $T$, $\widehat T$ is a tree which contains a vertex $\widehat u$ of degree $m_0$, i.e., $\widehat T$ is homeomorphic to a star.
It follows from the definitions of the graphs $P_{T}, \widehat T$ that the minimal subtree $S$ of $T$ that contains the vertices $u_1, \dots, u_{m_1}$ (recall $u_i = (e_i)_-$ in $P_{T}$) is isomorphic to
a subtree $\widehat S$ of $\widehat T$.
Indeed, a copy of $S$ will show up in $X_T$ in place of a star around the vertex $v_1$ of $X$, hence,
a copy $\widehat S$ of $S$ will also show up in $P_T$ as a subgraph of $\widehat T$.
Since $T$ is trivalent and
$\widehat T$ is a tree that has $m_0$ vertices of degree 1, one vertex $\widehat u$ of degree $m_0$ and other vertices of degree 2, it follows that $\widehat S$ contains a single vertex of degree $\ge 3$ which implies $m_1 = 3$.
Since $T$ is an arbitrary trivalent tree, we may pick a tree $T = T_{12}$ that has adjacent vertices $w_{12}$, $w_3$ of degree 3 such that $w_{12}$ is adjacent to both $u_1, u_2$ and $w_3$ is adjacent to $u_3$,
see Fig. 4(a). Note that, in this case, $S$ is the subtree of $T = T_{12}$ that contains vertices $u_1, u_2$, $u_3$, $w_{12}, w_3$
and has 4 edges that connect these vertices, see Fig. 4(a).
\begin{center}
\begin{tikzpicture}[scale=.64]
\draw (0,0)[fill = black]circle (0.06);
\draw (-1,1)[fill = black]circle (0.06);
\draw (-1,-1)[fill = black]circle (0.06);
\draw (2,0)[fill = black]circle (0.06);
\draw (3,1)[fill = black]circle (0.06);
\draw(0,0) -- (-1,1);
\draw(0,0) -- (-1,-1);
\draw(0,0) -- (2,0);
\draw(2,0) -- (3,1);
\draw[dashed](2,0) -- (3,-.5);
\draw (2.5,-1.5)[fill = black]circle (0.06);
\draw (4.4,-0.5)[fill = black]circle (0.06);
\node at (3.5,-1) {$\ldots$};
\node at (-1,1.5) {$u_1$};
\node at (-1,-1.5) {$u_2$};
\node at (0.25,.5) {$w_{12}$};
\node at (1.75,.5) {$w_{3}$};
\node at (3,1.5) {$u_3$};
\node at (2,-1.5) {$u_4$};
\node at (5,-0.5) {$u_{m_0}$};
\node at (1,1.5) {$T_{12}$};
\node at (1.25,-2.5) {Figure 4(a)};
\begin{scope}[shift={(9,0)}]
\draw (0,0)[fill = black]circle (0.06);
\draw (-1,1)[fill = black]circle (0.06);
\draw (-1,-1)[fill = black]circle (0.06);
\draw (2,0)[fill = black]circle (0.06);
\draw (3,1)[fill = black]circle (0.06);
\draw(0,0) -- (-1,1);
\draw(0,0) -- (-1,-1);
\draw(0,0) -- (2,0);
\draw(2,0) -- (3,1);
\draw[dashed](2,0) -- (3,-.5);
\draw (2.5,-1.5)[fill = black]circle (0.06);
\draw (4.4,-0.5)[fill = black]circle (0.06);
\node at (3.5,-1) {$\ldots$};
\node at (-1,1.5) {$u_1$};
\node at (-1,-1.5) {$u_3$};
\node at (0.25,.5) {$w_{13}$};
\node at (1.75,.5) {$w_{2}$};
\node at (3,1.5) {$u_2$};
\node at (2,-1.5) {$u_4$};
\node at (5,-0.5) {$u_{m_0}$};
\node at (1,1.5) {$T_{13}$};
\node at (1.25,-2.5) {Figure 4(b)};
\end{scope}
\end{tikzpicture}
\end{center}
\noindent Since the image $\widehat w_{12}$ of $w_{12}$ in
$\widehat S$ has degree 3, we have $\deg \widehat w_{12} = m_0$ in $\widehat T_{12}$.
This, in particular, means that every preimage of $w_3$ in $\widehat T_{12}$ has degree 2 and adjacent to a preimage of $w_{12}$.
Consider a vertex $u \in V Q$, $Q \in \{ X, Y \}$, such that $\beta_Q(u) = v_0$.
Let $g_1, \dots, g_k$ be all of the oriented edges of $Q$ that end in $u$. We claim that the set $\{ \beta_Q(g_1), \dots, \beta_Q(g_k) \}$ may not contain both
$e_3$ and $e_j$, where $j >3$. Arguing on the contrary, assume that
\begin{equation}\label{inc}
\{ e_3, e_j \} \subseteq \{ \beta_Q(g_1), \dots, \beta_Q(g_k) \} .
\end{equation}
Then a star neighborhood of $u$ in $Q$ would turn in $Q_{T_{12}}$ into a tree isomorphic
to a subtree $S_u$ of $T_{12}$ which contains vertices $w_3, u_3, u_j$, see Fig.~4(a).
Clearly, either $\deg w_3 = 3$ in ${S_u}$ or $\deg w_3 = 2$ in ${S_u}$ and then $S_u$ contains no vertex $w_{12}$. In either case, a copy of $S_u$ is present in $\widehat T_{12}$ which is impossible because, as we saw above, every preimage of $w_3$ in $\widehat T_{12}$ has degree 2 and adjacent to a preimage of $w_{12}$. Thus the inclusion \eqref{inc} is impossible.
Recall that we can pick any trivalent tree $T$, in particular, we can pick a tree $T = T_{13}$ that has adjacent vertices $w_{13}$, $w_2$ of degree 3 such that $w_{13}$ is adjacent to both $u_1, u_3$ and $w_2$ is adjacent to $u_2$, see Fig.~4(b).
Repeating the above argument with indices 2 and 3 switched, we can show that
$$
\{ e_2, e_j \} \not\subseteq \{ \beta_Q(g_1), \dots, \beta_Q(g_k) \}
$$
for every $j > 3$.
Similarly, switching indices 1 and 3, we prove that
$$
\{ e_1, e_j \} \not\subseteq \{ \beta_Q(g_1), \dots, \beta_Q(g_k) \}
$$
for every $j > 3$.
Now we see that for every vertex $u \in VX \vee VY$ such that $\beta_Q(u) = v_0$, where
$Q \in \{ X, Y \}$, the edges $ g_1, \dots, g_k $ of $Q$ that end in $u$ have the following property: either
$$ \{ \beta_Q(g_1), \dots, \beta_Q(g_k) \} \subseteq
\{ e_1, e_2, e_3 \} \quad \mbox{or} \quad
\{ \beta_Q(g_1), \dots, \beta_Q(g_k) \} \subseteq
\{ e_4, \dots, e_{m_0} \} .
$$
Since the edges of $X \vee Y$ are identified in the pushout $P = X \vee_{W(S_1)} Y$ if and only if their $\beta_X$-, $\beta_Y$-images in $P$ are equal, it follows that the terminal vertex of $e_1, e_2, e_3$ must be different in $P$ from the terminal vertex of an edge $e_j$ where $j \ge 4$. This contradiction to the definition of the edges $e_1, \dots, e_{m_0}$ of $U$ completes the proof.
\end{proof}
We are now ready to prove our key lemma.
\begin{lemma}\label{mn} Suppose that the graphs $X, Y, W(S_1), P, Z, U$ and corresponding maps
$\al_X, \ldots, \delta$ are defined as in Fig.~1. Then there exists a finite sequence $\tau$ of alternating replacements
as in Lemma~\ref{rpl} and \ED s as in Lemma~\ref{ED} that result in graphs
$X^\tau, Y^\tau$, $W(S_1)^\tau$, $P^\tau$, $Z^\tau, U^\tau$ that have the following properties: $\brr(X^\tau) = \brr(X)$,
$\brr(Y^\tau) = \brr(Y)$, $\brr( W(S_1)^\tau) = \brr(W(S_1))$, \ $P^\tau = Z^\tau = U^\tau$, $\brr(P^\tau) \ge \brr(P)$, every vertex of $U^\tau$ has degree 3 or 2 and every vertex of degree 3 of $U^\tau$ has a preimage of degree 3 in $X^\tau$ or $Y^\tau$.
\end{lemma}
\begin{proof} Applying Lemma~\ref{rpl}, we may assume that $P = Z = U$. If every vertex of $U$ has degree 2 or 3, then our claim holds true as follows from Lemma~\ref{knt}. Hence, we may assume that $U$ contains a vertex $v_0$ of
degree at least 4. In view of Lemma~\ref{ED}, by checking all possible \ED s of $U$ around $v_0$, we can find an \ED\ by means of a trivalent tree $T$
such that either $\brr(P_{T}) > \brr(P)$ or $\brr(P_{T}) = \brr(P)$ and
$$
\sum_{v \in VP_{T} } \max(\deg v -3, 0) < \sum_{v \in V P } \max(\deg v -3, 0) .
$$
Invoking Lemma~\ref{rpl}, we replace $U_T$ with $P_{T}$ and, completing one cycle of changes in graphs $X, Y, W(S_1), P, Z, U$, we rename $Q := Q_T$, where $Q \in \{ X, Y, W(S_1), P, Z, U \}$, and start over.
Observe that the rank $\brr(Z) = \brr( \langle H, K, S_1 \rangle) $ is bounded above by the number of generators
$\rr(H) + \rr(K) + |S_1|$
and this bound does not change as we perform cycles of changes of graphs $X, Y, W(S_1), P, Z, U$,
because $|S_1|$ is equal to the number of connected components of $W(S_1)$ and $\brr(H) = \brr(X)$, $\brr(K) = \brr(Y)$. This implies that the number
$$
\sum_{v \in VZ } \max(\deg v -3, 0)
$$
is bounded above by $ 2\brr(Z) \le 2(\rr(H) + \rr(K) + |S_1|)$. Thus the total number of cycles will not exceed
$2(\rr(H) + \rr(K) + |S_1|)^2$ and our lemma is proved.
\end{proof}
Note that the arguments of the proof of Lemma~\ref{mn} provide an algorithm that deterministically constructs
the desired graphs $X^\tau, Y^\tau$, $W(S_1)^\tau$, $P^\tau$, $Z^\tau, U^\tau$ in polynomial space
(of size of input which are graphs $X, Y, W(S_1)$, $P, Z, U$ along with associated maps $\al_X, \ldots, \delta$).
\begin{lemma}\label{sum}
$\brr(Z) \le \brr(X) + \brr(Y)$.
\end{lemma}
\begin{proof} In view of Lemma~\ref{mn}, we may assume that Stallings graphs $X, Y, Z$ of subgroups
$H, K, \langle H, K, S_1 \rangle$ satisfy the conclusion of Lemma~\ref{mn}. Then all of the vertices of $X, Y, Z$ have degree 2 or 3.
Hence, $2\brr(U) = | V_3 U|$, where $ V_3 U$ is the set of vertices of degree 3 of $U$.
Similarly, we have $2\brr(X) = | V_3 X |$ and $2\brr(Y) = | V_3 Y |$. Now application of Lemma~\ref{knt} yields
$|V_3 Z | \le | V_3 X | + | V_3 Y | $ which implies
the desired inequality $\brr(Z) \le \brr(X) + \brr(Y)$.
\end{proof}
\section{Applications}
As an application of Lemma~\ref{mn}, we state and prove a couple of specific inequalities for
reduced ranks of the generalized intersections and joins of subgroups in free groups.
\begin{thm}\label{th1} Let $H, K$ be finitely generated subgroups of a free group $F$.
Let $S(H,K) \subseteq F$ denote a set of representatives of those double cosets $H tK \subseteq F$, $t \in F$,
that have the property $H t Kt^{-1} \ne \{ 1\}$, and let $S_1 \subseteq S(H,K)$ be nonempty.
Then $\brr (\langle H, K, S_1 \rangle) \le \brr( H) + \brr( K)$ and
\begin{align}\notag
\brr(H, K, S_1) & := \sum_{s \in S_1} \brr(H \cap s K s^{-1}) \\ \label{inq1t}
& \le
2 \brr(H) \brr(K) - \brr (\langle H, K, S_1 \rangle) \min(\brr(H), \brr(K)) .
\end{align}
Moreover,
\begin{align}\label{inq2t}
\! \brr(H, K, S_1) \! \le \tfrac 12 ( \brr(H) \! + \! \brr(K)\! - \! \brr(\langle H, K, S_1 \rangle) )
( \brr(H) \! + \! \brr(K)\! -\! \brr(\langle H, K, S_1 \rangle)\! + \! 1) .
\end{align}
\end{thm}
Note that the inequality \eqref{inq1t} is a strengthened version of \eqref{inq11} and
this strengthening is analogous to a strengthened version of
the Hanna Neumann conjecture introduced by Walter Neumann \cite{WN}.
We also remark that the inequality \eqref{inq1t} in the cases when $S_0 = \{ 1 \}$ and $S_0 = S(H,K)$ is due to Kent \cite{K}
and the inequality \eqref{inq2t} is shown by Dicks \cite{D2} who also obtained other inequalities.
It is worthwhile to mention that the natural question on the existence of a bound that would be
a stronger version of \eqref{inq1t}, in which the first term would have the coefficient 1 in place of 2
and the second negative term would contain the factor $\brr (\langle H, K, S_1 \rangle)$ with some coefficient, has a negative solution. Indeed, according to \cite{I12}, there are finitely generated subgroups $H, K$ of a free group $F$
such that $\brr(H, K, S(H,K)) = \brr(H) \brr(K) >0$ and $\brr (\langle H, K, S(H,K)) >C$ for any constant $C >0$.
\begin{proof}
In view of Lemma~\ref{mn}, we may assume that Stallings graphs $X, Y, Z$ of subgroups
$H, K, \langle H, K, S_1 \rangle$, resp., and graphs $W(S_1)^\tau$, $P^\tau$ satisfy the conclusion of Lemma~\ref{mn}.
Recall that $\brr(H) = \brr(X)$, $\brr(K) = \brr(Y)$ and $\brr(\langle H, K, S_1 \rangle ) \le \brr(Z) = \brr(U)$,
As before, if $Q$ is a graph whose vertices have degree 2 or 3, then $V_3 Q$ denotes the set of vertices of $Q$ of degree 3. Recall that $2 \brr(Q) = |V_3 Q |$.
For every $v \in VU$, denote
$$
k_v := |V_3X \cap \beta_X^{-1}(v)|, \quad \ell_v := |V_3Y \cap \beta_Y^{-1}(v)|, \quad m := |V_3 W(S_1) |, \quad
n := |V_3 U | .
$$
We also let $V_3U := V_B U \vee V_X U\vee V_Y U$, where $V_B U$ is the set of vertices of $U$ that have preimages
of degree 3 in both $X$ and $Y$, $V_X U$ is the set of vertices of $U$ that have preimages
of degree 3 in $X$ only, and $V_Y U$ is the set of vertices of $U$ that have preimages
of degree 3 in $Y$ only. Denote $n_B := |V_B U|$, $n_X := |V_X U|$, $n_Y := |V_Y U|$.
We let $V_3 X = V_{31} X \vee V_{32} X$, where $V_{31} X := \beta_X^{-1}(V_X U)$ and $V_{32} X := \beta_X^{-1}(V_B U)$. Similarly, we let $V_3 Y = V_{31} Y \vee V_{32} Y$, where $V_{31} Y := \beta_Y^{-1}(V_Y U)$ and $V_{32} Y := \beta_Y^{-1}(V_B U)$. Denote
$$
k := |V_3X|, \ k_1 := |V_{31} X|, \ k_2 := |V_{32} X|, \ \ell := |V_3Y|, \ \ell_1 := |V_{31} Y|, \ \ell_2 := |V_{32} Y| .
$$
Clearly,
$
k = k_1 +k_2 , \ \ell = \ell_1 +\ell_2 , \ n = n_B+n_X +n_Y, \ k_1 \ge n_X, \ \ell_1 \ge n_Y$, and
$$
m \le \sum_{v \in VU} k_v \ell_v = \sum_{v \in V_BU} k_v \ell_v .
$$
We now continue with arguments similar to those of Imrich-M\"uller \cite{IM} that follow the proof of their lemma.
Since $\sum_{v \in V_BU} \ell_v = \ell_2$ and $\ell = \ell_1 +\ell_2 $, we obtain
\begin{align}\notag
m & \le \sum_{v \in V_BU} k_v \ell_v = k\ell - \sum_{v \in V_BU} (k -k_v) \ell_v - k \ell_1 \\ \notag
& \le k \ell - \sum_{v \in V_BU} (k -k_v) - k \ell_1 = k \ell - kn_B + k_2 - k \ell_1 \\ \label{in3}
& \le k \ell - k(n_B + n_Y -1 ) .
\end{align}
Switching $X$ and $Y$, we analogously obtain
\begin{align}\label{in4}
m & \le k \ell - \ell (n_B + n_X -1 ) .
\end{align}
Assume that $n_B = 0$. Then $\brr(W(S_1)) = 0$ and the inequality \eqref{inq1t} is equivalent to
\begin{align*}
0 \le 2 \brr(X) \brr(Y) - \min( \brr(X), \brr(Y))\brr(Z)
\end{align*}
which is equivalent to $\brr(Z) \le 2 \max( \brr(X), \brr(Y))$. Since
$\brr(Z) \le \brr(X) + \brr(Y)$ by Lemma~\ref{sum},
it follows that \eqref{inq1t} holds true. A reference to Lemma~\ref{sum} also proves that
$$
\brr(\langle H, K, S_1 \rangle ) \le \brr(Z) \le \brr(X)+ \brr(Y) = \brr(H) +\brr(K) .
$$
Suppose that $n_B \ge 1$. Since $n_B +n_X+n_Y = 2\brr(U) $, it follows that
$$
n_X+n_Y \le 2\brr(U)-1 .
$$
Hence,
$\min(n_X, n_Y) \le \brr(U)-1$ and $\max(n_B + n_X -1, n_B + n_Y -1 ) \ge \brr(U)$. Therefore, the inequalities
\eqref{in3}--\eqref{in4} imply that
\begin{align*}
2 \brr(W(S_1)) = m & \le 4 \brr(X) \brr(Y) - 2 \min( \brr(X), \brr(Y))\brr(Z) .
\end{align*}
Dividing by 2, we obtain the required bound \eqref{inq1t}.
\medskip
Observe that Lemma~\ref{sum} together with the foregoing classification of vertices of degree 3 in graphs
$W(S), X, Y, P=Z=U$ make it possible to produce other inequalities for $\brr(H, K, S_1) = \brr(W(S_1)) $ that would involve $\brr(\langle H, K, S_1 \rangle ) = \brr(Z)$. For example, when maximizing the sum
$\sum_{v \in VU} k_v\ell_v$ which gives an upper bound for $m = 2 \brr(W(S_1))$, we may assume that there is a single vertex
$v$ such that both $k_v, \ell_v >0$, whence $n_B = 1$, $k_2 = k_v$ and $\ell_2 = \ell_v$.
We may further assume that $k_1 = n_X$, $\ell_1 = n_Y$. Indeed, if, say $k_1 > n_X$, then we could increase
$k_v$ by $k_1 - n_X$ and decrease $k_1$ by $k_1 - n_X$ making thereby $k_2\ell_2 = k_v\ell_v$ greater. Hence, to get an upper bound for $m$, we can maximize the product $k_2\ell_2 = (k - k_1) (\ell -\ell_1)$ subject to $k_1 + \ell_1+ 1 = n$.
Since $k, \ell, n$ are fixed positive even integers, equal to $2\brr(X)$, $2\brr(Y)$, $2\brr(Z)$, resp.,
it follows that the product $(k- k_1)(\ell -n+1 - k_1)$ has the maximum at $k_1 = \tfrac 12( k-\ell +n)$. Since $k_1$ is an integer, it follows that $(k- k_1)(\ell -n+1 - k_1)$ has a maximal value for integer $k_1$ when
$k_1 = \tfrac 12( k-\ell +n)$ or $k_1 = \tfrac 12( k-\ell +n)-1$. This means that, unconditionally, we have
$$
m \le (k- k_1)(\ell -n+1 - k_1) \le \tfrac 12( k+\ell -n+2 ) \cdot \tfrac 12 (k+\ell -n ) .
$$
Equivalently,
\begin{align*}
2 \brr(W(S_1)) = m & \le ( \brr(X) + \brr(Y) - \brr(Z) +1) ( \brr(X) + \brr(Y) - \brr(Z))
\end{align*}
which implies the bound \eqref{inq2t}. Theorem~\ref{th1} is proved.
\end{proof}
\section{One More Question}
One might wonder what would be the conditions that guarantee the existence of a nontrivial
``blow-up" of a vertex $v_0$ of the graph $P = Z =U$ by an \ED\ around $v_0$, as defined in Sect.~3.
Here we present a result that provides a criterion for the existence of such a ``blow-up" of a vertex $v_0$ of $U$, i.e.,
the existence of an \ED\ around a vertex $v_0$ by means of a tree $T$ such that $\brr(P_{T}) > \brr(P)$.
Assume that $P = Z =U$ and the maps $\gamma, \delta$ in Fig.~1 are identity maps on $U$. Let $v_0$ be a fixed vertex of $U$ and let $D$ denote the set of all oriented edges of $U$ that end in $v_0$.
A vertex $v$ of one of the graphs $W(S_1), X, Y$ is called a {\em $v_0$-vertex} if the image of $v$ under $\beta_X \al_X$, $\beta_X$, $\beta_Y$, resp., is $v_0$.
For every $v_0$-vertex $v \in V Q$, where $Q \in \{ X, Y, W(S_1) \}$, define the set $D(v) \subseteq D$ so that $e \in D(v)$ if and only if there is an edge $f \in \vec E Q$ so that the terminal vertex $f_+$ of $f$ is $v$ and the image of the edge $f$ in $U$ is $e$.
Consider a partition $D = A \vee B$ of the set $D $
into two nonempty subsets $A, B$.
A $v_0$-vertex $v \in V Q$, where $Q \in \{ X, Y, W(S_1) \}$, is said to be of {\em type $\{A, B\}$} if both intersections $D(v) \cap A$ and $D(v) \cap B$ are nonempty. The set of all vertices of type $\{A, B\}$ of a graph ${Q}$, where
$Q \in \{ X, Y, W(S_1) \}$, is denoted $V_{\{A,B\}} Q$.
Consider a bipartite graph $\Psi(\{A, B \} )$ whose set of vertices
$$
V \Psi(\{A, B \}) = V_X \Psi(\{A, B \}) \vee V_Y \Psi(\{A, B \} )
$$
consists of two disjoint parts
$$
V_X \Psi(\{A, B \}) := V_{ \{A, B \} }X \quad \mbox{and } \quad
V_Y \Psi(\{A, B \} ):= V_{\{ A, B \} }Y .
$$
Two vertices $v_X \in V_X \Psi(\{A, B \}) $ and
$v_Y \in V_Y \Psi(\{A, B \})$ are connected in $\Psi(\{A, B \})$ by an edge if and only if there is a vertex $v \in W(S_1) $ such that
$\alpha_{X}(v) = v_X$, $\alpha_{Y}(v) = v_Y$ and $v \in V_{\{A, B \} }W(S_1)$.
\begin{theorem}\label{AB} Suppose $P = Z =U$, $v_0$ is a vertex of $U$. The equality $\brr (P_{T}) = \brr (U_T) = \brr (U)$ holds for every \ED\ around $v_0$ by means of a tree $T$ if and only if the following conditions hold.
For every partition $D = A \vee B$ with nonempty $A, B$ if the graph $\Psi( \{ A, B \})$ contains $k$
connected components, then there exists a partition $D = \bigvee_{i=1}^{k+1} C_i$ such that, for every $i$, $C_i$ is nonempty and either $C_i \subseteq A$ or $C_i \subseteq
B$. Next, for every vertex $v \in V Q$, where $Q \in \{ X, Y \}$, such that $\beta_Q(v) = v_0$, the intersection
$D(v) \cap A$ is either empty or $D(v) \cap A \subseteq C_{i_{vA}}$ for
some $i_{vA}$, $i_{vA} = 1, \dots, k+1$, and the intersection $D(v) \cap B$ is either empty or $D(v) \cap B \subseteq C_{i_{vB}}$ for some $i_{vB}, i_{vB} = 1, \dots, k+1$.
\end{theorem}
\begin{proof} First we introduce the notation we will need below. As above, let $T$
be a tree whose set of vertices of degree 1 is $D'$. If $H' \subseteq D'$
is a nonempty subset of vertices of degree 1 in $T$, we let $M_T(H')$ denote the
minimal subtree of $T$ that contains $H'$. Clearly, the set of vertices of degree
$\le 1$ of $M_T(H')$ is $H'$. Note that $M_T(H') = H'$ if $H'$ consists of a single
vertex. If $E \subseteq D$, then $E' \subseteq D'$ denotes the image of $E$
under the map $d \to d'$ for every $d \in D$.
For every vertex $v \in V Q$, where $Q \in \{ X, Y, W(S_1) \}$, consider a tree
$M_T( D(v) )$ which is isomorphic to the subtree
$M_T( D(v)')$ of $T$ and let
\begin{equation}\label{zeta}
\zeta_v : M_T( D(v) ) \to T
\end{equation}
denote the natural monomorphism whose image is $M_T( D(v)' )$.
If $w$ is a vertex of $W(S_1)$, then the tree $M_T( D( \alpha_{X}(w) ))$ contains a subtree
isomorphic to $M_T( D( w) )$ which we denote $M_{T, X}( D( w) )$.
Similarly, the tree $M_T( D( \alpha_{Y}(w) ))$ contains a subtree isomorphic
to $M_T( D( w) )$, denoted $M_{T, Y}( D(w) )$.
It follows from the definitions that the graph $\widehat T = \gamma_T^{-1}
\delta_T^{-1} (T)$ can be described as follows.
$\widehat T $ is the union of graphs $M_T( D(v) )$ over all $v \in V Q$, $Q \in
\{ X, Y \}$, which are identified along their subgraphs of the form
$M_{T, X}( D( w) )= M_{T, Y}( D( w) )$ over all $w \in V W(S_1)$.
Now assume that the equality $\brr (P_{T})= \brr (U_T)$ holds for every tree $T$ or,
equivalently, see Lemma~\ref{ZT}, the graph $\widehat T$ is a tree for every $T$. Note that
$\widehat T$ being a tree need not imply that $\wht T$ is naturally isomorphic to $T$.
Let $D = A \vee B$ be a partition of $D$ with nonempty $A, B$. Our goal is to find
a partition $D = \bigvee_{i=1}^{k+1} C_i$ with the properties described in
Theorem~\ref{AB}.
Consider a trivalent tree $T = T(\{ A, B\})$ such that $T$ contains a nonoriented edge $e$ so
that the graph $T \setminus \{ e \}$ consists of two connected components $T_{eA}$, $T_{eB}$
such that $A' \subseteq V T_{eA}$ and $B' \subseteq V T_{eB}$.
Note that the tree $M_T( D(v) )$, where $v \in V Q$, $Q \in \{ X, Y, W(S_1) \}$,
contains an edge $f$ with $\zeta_v(f) = e$, where $\zeta_v$ is defined by \eqref{zeta}, if and only if $v$ has type $\{ A, B \}$.
Hence, the set $\gamma_T^{-1} \delta_T^{-1} (e) $ consists of images of such edges $f$ of
$M_T( D(v) )$ with $\zeta_v(f)=e$.
Furthermore, let $v, v' \in V_{ \{A, B \} } X \vee V_{ \{A, B \} } Y $ be two vertices that belong to the same connected component of $\Psi(\{A, B \} )$. Then there exists a
sequence $v = v_1, v_2, \dots, v_{\ell} = v'$ of vertices in the graph $\Psi(\{A, B \} )$ so that $v_i, v_{i+1}$ are connected by an edge. It follows from the definitions that the edges
$f_1, f_2, \dots, f_\ell $ such that $\zeta_{v_i}(f_i)=e$, $i =1, \dots, \ell$, are identified in $\widehat T$. Conversely, it follows from the definitions in a similar fashion that if
$v, v' \in V_{ \{A, B \} } X \vee V_{ \{A, B \} } Y $
and the edges $f, f'$ such that $\zeta_{v}(f)=e=\zeta_{v'}(f')$ are identified in
$\widehat T$, then $v, v'$ belong to the same connected component of $\Psi(\{A, B \} )$.
These remarks prove that there is a bijection between the set
$\gamma_T^{-1} \delta_T^{-1} (e) = \{ f_1,
\dots, f_k \} \subseteq \widehat T $ and the set of connected components of the graph
$\Psi(\{A, B \} )$. Since $\widehat T$ is a tree with $|D|$ vertices of degree 1 whose set we denote $D''$, it follows that the graph $\widehat T - (\gamma_T^{-1} \delta_T^{-1} (e ) )$ splits into $k+1$ connected components which induce a partition $D'' = \bigvee_{i=1}^{k+1} C''_i$, where for every $i$, $C''_i$ is nonempty and either $C''_i \subseteq A''$ or $C_i \subseteq B''$. Then $D = \bigvee_{i=1}^{k+1} C_i$,
where $C''_i $ is the image of $C_i$ under the map $d \to d''$ for every $d \in D$, provides a desired partition for $D$
because, for every $v_0$-vertex $v \in V Q$, where $Q \in \{ X, Y \}$, the vertices of
$A'' \cap D(v)'' \subseteq D''$ and those of $B'' \cap D(v)''$ are connected in $\widehat T - (\gamma_T^{-1} \delta_T^{-1} (e ) )$ by images of edges of $M_T( D(v) ) - \zeta_v^{-1}( \{ e \} ) $ of $\widehat T$. Therefore, $D(v) \cap A \subseteq C_{i_{vA}}$ for some $i_{vA}$ whenever the intersection $D(v) \cap A$ is not empty and
$D(v) \cap B \subseteq C_{i_{vB}}$ for some $i_{vB}$ whenever the intersection $D(v) \cap B$ is nonempty.
\smallskip
Now we will prove the converse. Assume that for every partition $D = A \vee B$
there exists a partition $D = \bigvee_{i=1}^{k+1} C_i$ with the properties of Theorem~\ref{AB}. Arguing on the contrary, suppose that there is a tree $T$ for which the graph $\widehat T$ is not a tree. Pick a shortest circuit $p$ in
$\widehat T$ of positive length. Let $e$ be a nonoriented edge of $T$ such that $e \in \delta_T\gamma_T(p)$. The graph $T - \{ e \}$ consists of two
connected components $T_{eA}, T_{eB}$ which define a partition $D' = A' \vee B'$, where $ A' \subseteq T_{eA}$, $ B' \subseteq T_{eB}$, and both $ A', B' $ are nonempty. Let $k$ denote the number of connected components of the graph
$\Psi(\{A, B \} )$. It is clear from Lemma~\ref{ZT} and the definitions that $k \ge 1$. According to our assumption, there exists a partition
$D = \bigvee_{i=1}^{k+1} C_i$, where, for every $i$, $C_i $ is nonempty, $C_i \subseteq A$ or $C_i \subseteq B$, and, for every $v_0$-vertex $v \in V Q$, where $Q \in \{ X, Y \}$, we have that $D(v) \cap A \subseteq C_{i_{vA}}$ for some $i_{vA}$ whenever $D(v) \cap A$ is nonempty and we have that $D(v) \cap B \subseteq C_{i_{vB}}$ for some $i_{vB}$ whenever $D(v) \cap B$ is nonempty.
As above, we observe that if $v, v' \in V_{ \{A, B \} } X \vee V_{ \{A, B \} } Y $ are two vertices, then the edges $\zeta^{-1}_{v}(e)$, $\zeta^{-1}_{v'}(e)$ are identified in $\widehat T$ if and only if $v, v'$ belong to the same connected component of the graph $\Psi(\{A, B \} )$. This implies that the edges $\gamma_T^{-1} \delta_T^{-1} (e ) \subseteq \widehat T$ are in bijective correspondence with connected components of $\Psi(\{A, B \} )$. Hence, there are $k$ edges in $\gamma_T^{-1} \delta_T^{-1} (e )$. Denote
$\gamma_T^{-1} \delta_T^{-1} (e ) = \{ f_1, \dots, f_k \}$.
Since identification of subgraphs
$$
M_{T, X}(D(w)) \subseteq M_{T}(D(\alpha_X(w) )) \quad \mbox{and} \quad M_{T, Y}(D(w)) \subseteq M_{T}(D(\alpha_Y(w) ))
$$
in process of construction of $\widehat T$
respects the partition $D' = \bigvee_{i=1}^{k+1} C'_i$, it follows that the graph
$\widehat T$ consists of pairwise disjoint subgraphs $S_1, \dots, S_{k+1}$, where
$$
\delta_T\gamma_T(S_j) \cap D'= C'_j , \quad j = 1, \dots, k+1 ,
$$
which are joined by $k$ nonoriented edges of $ \gamma_T^{-1} \delta_T^{-1} (e )$. Collapsing subgraphs $S_1, \dots, S_{k+1}$ into points, we see that $\widehat T$ turns into a tree with $k$ nonoriented edges and $k+1$ vertices. Consequently, a shortest circuit in $\widehat T$ of positive length may not contain any edge of
$ \gamma_T^{-1} \delta_T^{-1} ( e )$. This contradiction to the choice of the path $p$ and the edge $e$ in
$ \delta_T\gamma_T(p)$ completes the proof of Theorem~\ref{AB}.
\end{proof}
| 7a2eddd429c0d4359b90d98479c325ea2599527d | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
The Downwards L\"{o}wenheim-Skolem-Tarski Theorem states that every model $M$ for a language $\mathcal{L}$, $|M| = \kappa \geq \aleph_0$, and cardinal $|\mathcal{L}| + \aleph_0 \leq \mu \leq \kappa$ there is an elementary submodel $M^\prime \prec M$, of cardinality $\mu$. Informally speaking, this means that first order logic (with countable language) cannot distinguish between infinite cardinals.
Second order logic, in which we are allowed to quantify over subsets of the structure, is strong enough to distinguish between different infinite cardinals. For example, it is easy to express the statement ``There are exactly $\aleph_7$ elements in the structure'' in second order logic. By a theorem of Magidor \cite{Magidor1971}, a variant of the Downwards L\"{o}wenheim-Skolem Theorem for full second order logic can hold only above a supercompact cardinal. In fact, there is a specific $\Pi^1_1$-formula $\Phi$ such that if $\kappa$ is a cardinal, and for every model $M$ of cardinality at least $\kappa$ that models $\Phi$ there is an elementary submodel $M^\prime$, $|M^\prime| < \kappa$, $M^\prime \models \Phi$, then there is a supercompact cardinal $\kappa_0 \leq \kappa$.
Thus, it is natural to ask how strong can be a fraction of the second order logic such that it still does not distinguish between ``most'' pairs of infinite cardinals. One candidate is first order logic enriched with the Chang's Quantifier:
\begin{definition}
Let $M$ be a model. We write \[Q x,\, \varphi(x, \vec{p})\]
if \[|\{x \in M : M \models \varphi(x, \vec{p})\}| = |M|\]
\end{definition}
We let $L(Q)$ be first order logic enriched with the quantifier $Q$. We write $M^\prime \prec_{Q} M$ if $M^\prime$ is an elementary submodel of $M$ relative to all formulas in $L(Q)$.
\begin{lemma}
The following are equivalent for infinite cardinals $\mu < \kappa$:
\begin{enumerate}
\item For every model $M$ of cardinality $\kappa^{+}$ there is an $L(Q)$-elementary submodel of cardinality $\mu^+$.
\item For every model $M$ for a language $\mathcal{L}$ that contains a predicate $A$, if $|M| = \kappa^+$ and $|A| = \kappa$, then there is $M^\prime\prec M$ with $|M^\prime| = \mu^+, |M^\prime \cap A| = \mu$.
\item For every function $f\colon (\kappa^{+})^{<\omega} \to \kappa$ there is a set $X \subseteq \kappa^{+}$, $|X| = \mu^+$, such that $|f \image X^{<\omega}| \leq \mu$.
\end{enumerate}
\end{lemma}
The second assertion is called \emph{Chang's Conjecture,} and it is denoted by:
\[(\kappa^+, \kappa) \twoheadrightarrow (\mu^+, \mu).\]
For basic facts about Chang's Conjecture see, for example, \cite[section 7.3]{ChangKeisler1990}.
Note that if $M$ is a model of cardinality $\kappa$ of enough set theory, $M^\prime \prec_Q M$, and $|M^\prime| = \mu$, then $\kappa$ is a successor cardinal iff $\mu$ is a successor cardinal. Thus the restriction in the above lemma to successor cardinals is unavoidable.
This is not the only instance of Chang's Conjecture which provably fails. For example, assuming {\rm GCH\xspace}, for every singular cardinal $\kappa$, and every $\mu$ such that $\mu > \cf \kappa$ and $\cf \mu \neq \cf \kappa$, $(\kappa^+, \kappa) \not\twoheadrightarrow (\mu^+, \mu)$, \cite[Lemma 1]{LevinskiMagidorShelah1990}. Without assuming ${\rm GCH\xspace}$ one can prove weaker results using Shelah's ${\rm PCF\xspace}$ mechanism: for example $(\aleph_{\omega+1},\aleph_\omega)\twoheadrightarrow (\aleph_{n+1}, \aleph_n)$, implies that for every scale on $\aleph_\omega$ there are stationarily many bad points of cofinality $\omega_{n+1}$ (see \cite{ForemanMagidor1997}). It is provable that for every scale on $\aleph_{\omega}$ there is a club in which every ordinal of cofinality $\geq \omega_4$ is good (see, for example, \cite{AbrahamMagidor2010}), and therefore $(\aleph_{\omega+1},\aleph_\omega)\not\twoheadrightarrow (\aleph_{n+1}, \aleph_n)$ for every $n \geq 3$. The consistency of the cases $(\aleph_{\omega+1},\aleph_\omega)\twoheadrightarrow (\aleph_2, \aleph_1)$ and $(\aleph_{\omega+1},\aleph_\omega)\twoheadrightarrow (\aleph_3, \aleph_2)$ is completely open.
There are many open questions about Chang's Conjecture at successors of singular cardinals. In this paper we will concentrate on the questions about Chang's Conjecture at successors of regular cardinals.
In Section \ref{sec: local cc}, we will show how to force instances of Chang's Conjecture of the form $(\kappa^{+},\kappa)\twoheadrightarrow (\mu^+,\mu)$, where $\kappa > \mu^+$ for various values of $\kappa$ and $\mu$ from large cardinals weaker than a supercompact cardinal. This improves the known upper bounds for the consistency strength of those instances.
In Section \ref{sec: global cc} we will show that it is consistent relative to a huge cardinal that for every successor $\kappa$ and every $\mu < \kappa$, $(\kappa^+,\kappa)\twoheadrightarrow (\mu^+,\mu)$. This answers a question of Foreman from \cite[Section 12, Question 7]{Foreman2010ideals}.
In Section \ref{sec: segment cc} we will construct a model in which for all $m < n < \omega$, $(\aleph_{n+1}, \aleph_n)\twoheadrightarrow(\aleph_{m+1}, \aleph_m)$ and $(\aleph_{\omega+1},\aleph_{\omega})\twoheadrightarrow (\aleph_1, \aleph_0)$.
Our notations are mostly standard. We work in {\rm ZFC\xspace}, and specify any usage of large cardinals. For basic facts about forcing see \cite{Jech2003}. For standard facts and definitions about large cardinals, see \cite{Kanamori2009}.
\section{Preliminaries and preservation lemmas}\label{sec: preliminaries}
We begin with some standard notations and definitions.
\begin{definition}[Levy collapse]
Let $\kappa < \lambda$ be cardinals. $\Col(\kappa, \lambda)$ is the set of all partial functions, $f\colon \kappa \to \lambda$, $|\dom f| < \kappa$, ordered by reverse inclusion.
\end{definition}
$\Col(\kappa,\lambda)$ adds a surejction from $\kappa$ onto $\lambda$. If $\kappa$ is regular cardinal, this forcing is $\kappa$-closed. The forcing $\Col(\kappa, {<}\lambda)$ is the product with support ${<}\kappa$ of $\Col(\kappa, \alpha),\ \alpha < \lambda$. If $\lambda$ is inaccessible then this forcing is $\lambda$-c.c.
\begin{definition}[Easton Collapse, Shioya]\cite{Shioya2011}
Let $\kappa < \lambda$ be cardinals. $\E(\kappa,\lambda)$ is the Easton-support product of $\Col(\kappa, \alpha)$, over inaccessible $\alpha \in (\kappa,\lambda)$.
\end{definition}
\begin{definition}
A partial order $\mathbb P$ has \emph{precaliber} $\kappa$ if for every $A \subseteq \mathbb P$ of size $\kappa$, there is $B \subseteq A$ of size $\kappa$ that generates a filter.
\end{definition}
\begin{lemma}
Assume that $\kappa$ is regular and $\lambda$ is Mahlo. Then $\E(\kappa,\lambda)$ has precaliber $\lambda$ and is $\kappa$-closed.
\end{lemma}
The following absorption lemmas will be important in our use of huge cardinal embeddings.
\begin{lemma}[Folklore]
\label{folk}
If $\mathbb P$ is $\kappa$-closed and forces $|\mathbb P| = \kappa$, then $\mathbb P$ is forcing-equivalent to $\Col(\kappa,|\mathbb P|)$.
\end{lemma}
The following notion and lemma are due to Laver, which we will show in somewhat more generality in Lemma \ref{easton projection}.
\begin{definition}Suppose $\mathbb P$ is a partial order and $\dot{\mathbb Q}$ is a $\mathbb P$-name for a partial order. The termspace forcing for $\dot{\mathbb Q}$, denoted $T(\mathbb P,\dot{\mathbb Q})$, is the set of equivalence classes of $\mathbb P$-names of minimal rank for elements of $\dot{\mathbb Q}$. Two names $\tau,\sigma$ are equivalent when $1 \Vdash \tau = \sigma$. The ordering is $[\tau] \leq [\sigma]$ iff $1 \Vdash_{\mathbb P} \tau \leq \sigma$.
\end{definition}
\begin{lemma}If $\dot{\mathbb Q}$ is a $\mathbb P$-name for a partial order, then the identity map from $\mathbb P \times T(\mathbb P,\dot{\mathbb Q})$ to $\mathbb P * \dot{\mathbb Q}$ is a projection.
\end{lemma}
In this paper, we will use Erd\H{o}s-Rado theorem repeatedly. Since, at some points, we will need to refer to its proof we provide here a proof as well for the case that interests us.
\begin{theorem}[Erd\H{o}s-Rado]\label{thm: Erdos-Rado}\cite{ErdosRado1956}
Let $\kappa$ be a regular cardinal and let $\rho < \kappa$.
\[(2^{{<}\kappa})^+\to (\kappa + 1)^2_\rho\]
Namely, for every function $f$ from the unordered pairs of $(2^{{<}\kappa})^+$ to $\rho$ there is a set $H$ of order type $\kappa+1$ and an ordinal $\alpha$ such that for all $x, y\in H$, $f(x, y) = \alpha$.
\end{theorem}
\begin{proof}
Let $\lambda = (2^{{<}\kappa})^+$. Let $M$ be an elementary submodel of $H(\chi)$ (for large enough $\chi$), with $f\in M$, $^{{<}\kappa} M \subseteq M$, $M \cap \lambda$ is an ordinal and $|M| = 2^{{<}\kappa}$.
Let $\delta = M \cap \lambda$. By the closure assumptions on $M$, $\cf \delta \geq \kappa$.
Let us construct, by induction, an increasing sequence of ordinals $\alpha_i \in M$ such that:
\[\forall i < j < \kappa,\, f(\alpha_i, \alpha_j) = f(\alpha_i, \delta)\]
Assume $\eta < \kappa$ and we have constructed the first $\eta$ members of the sequence $\langle \alpha_i \colon i < \eta \rangle$. The element $r = \langle f(\alpha_i, \delta) \colon i < \eta\rangle\in \,^\eta \rho$ belongs to $M$, as $M$ is closed under ${<}\kappa$-sequences and thus contains $H(\kappa)$. Similarly, the function $g\colon \eta \to M \cap \lambda$, $g(i) = \alpha_i$, belongs to $M$.
\[H(\chi) \models \exists \zeta,\, \forall i < \eta,\, f(g(i), \zeta) = r(i),\, \zeta > \sup g\image \eta\]
as witnessed by $\zeta = \delta$. By elementarity, the same holds in $M$, so there is $\zeta\in M$ such that $f(\alpha_i, \zeta) = r(i) = f(\alpha_i, \delta)$ for all $i < \eta$. Take $\alpha_\eta = \zeta$.
Next, let us narrow down the sequence $\langle \alpha_i \colon i < \kappa\rangle$ to a homogeneous set. Let $\rho_i = f(\alpha_i, \delta) < \rho$. Since $\kappa$ is regular and $\rho < \kappa$, there is some $\rho_\star < \kappa$ and an unbounded set $I \subseteq \kappa$ such that for all $i\in I$, $\rho_i = \rho_\star$. Let $H = \{\alpha_i \colon i \in I\} \cup \{\delta\}$. For every $\alpha < \beta$ in $H$, $f(\alpha, \beta) = f(\alpha,\delta) = \rho_\star$, as wanted.
\end{proof}
\begin{remark}\label{remark: erdos-rado for 2k colors}
The same proof shows that whenever $\kappa$ is regular, $\lambda^{<\kappa} = \lambda$, and \[f\colon [\lambda^+]^2 \to \lambda,\] there is an increasing sequence of ordinals $\langle \alpha_i \colon i < \kappa + 1\rangle$, $\delta = \alpha_{\kappa}$ such that $f(\alpha_i, \alpha_j) = f(\alpha_i, \delta)$ for all $i < j < \kappa$. This observation will come in handy later.
\end{remark}
In this paper, we are interested in transfer properties between pairs of cardinals. However, consideration of transfer between larger collections of cardinals will aid in the investigation of pairs. Suppose $\langle \lambda_i \rangle_{i \in I}$, $\langle \kappa_i \rangle_{i \in I}$ are sequences of cardinals. The notation
\[ \langle \lambda_i \rangle_{i \in I} \twoheadrightarrow \langle \kappa_i \rangle_{i \in I} \]
signifies the assertion that for every $f : [\lambda]^{{<}\omega} \to \lambda$, where $\lambda = \sup \lambda_i$, there is $X \subseteq \lambda$ closed under $f$ such that $|X \cap \lambda_i| = \kappa_i$ for each $i \in I$.
Let us note a few easy facts about these principles:
\begin{itemize}
\item If $J \subseteq I$ and $\langle \lambda_i \rangle_{i \in I} \twoheadrightarrow \langle \kappa_i \rangle_{i \in I}$, then $\langle \lambda_i \rangle_{i \in J} \twoheadrightarrow \langle \kappa_i \rangle_{i \in J}$.
\item If $\langle \lambda_i \rangle_{i \in I} \twoheadrightarrow \langle \kappa_i \rangle_{i \in I}$ and $\langle \kappa_i \rangle_{i \in I} \twoheadrightarrow \langle \mu_i \rangle_{i \in I}$, then $\langle \lambda_i \rangle_{i \in I} \twoheadrightarrow \langle \mu_i \rangle_{i \in I}$.
\item If $\lambda_j,\kappa_j$ are the maximum elements of $\langle \lambda_i \rangle_{i \in I}$, $\langle \kappa_i \rangle_{i \in I}$ respectively, $\lambda' > \lambda_j$, and $\kappa_j > \kappa' \geq \sup_{i \not= j} \kappa_i$, then $\{ (j,\lambda') \} \cup \langle \lambda_i \rangle_{i \not= j} \twoheadrightarrow \{ (j,\kappa') \} \cup \langle \kappa_i \rangle_{i \not= j}$
\end{itemize}
During the construction of our models, we would like to use the fact that Chang's Conjecture is indestructible under a wide variety of forcing notions.
\begin{definition}
If $\lambda_1 \geq \lambda_0$ and $\kappa_1 \geq \kappa_0$ are cardinals and $\xi$ is an ordinal, let
\[ (\lambda_1,\lambda_0) \twoheadrightarrow_\xi (\kappa_1,\kappa_0) \]
stand for the statement that for all $f : \lambda_1^{{<}\omega} \to \lambda_1$, there is $X \subseteq \lambda_1$ of size $\kappa_1$ such that $f\image (X^{{<}\omega}) \subseteq X$, $|X \cap \lambda_0| = \kappa_0$, and $\xi \subseteq X$.
\end{definition}
\begin{lemma}[Folklore]
\label{preserve small}
The statement $(\lambda_1,\lambda_0) \twoheadrightarrow_{\kappa_0} (\kappa_1,\kappa_0)$ is preserved by $\kappa_0^+$-c.c.\ forcing.
\end{lemma}
\begin{proof}
Suppose $\mathbb P$ is a $\kappa_0^+$-c.c.\ forcing and $\dot f$ is a $\mathbb P$-name for a function from $\lambda_1^{{<}\omega}$ to $\lambda_1$. For every $x \in \lambda_1^{{<}\omega}$, let us look at the $\mathbb{P}$-name $\dot{f}(x)$. By the chain condition of $\mathbb{P}$, the set of possible values for $\dot f(x)$ has size $\leq \kappa_0$. If $g(\alpha,x)$ returns the $\alpha^{th}$ possible value for $\dot f(x)$, then a set closed under $g$ of the appropriate type will be closed under $\dot f$ in the extension by $\mathbb P$.
\end{proof}
\begin{lemma}
Suppose either $\lambda_0 = \kappa_0^{+\xi}$, or there is $\lambda \leq \lambda_0$ such that $\lambda_0 = \lambda^{+\xi}$ and $\lambda^{\kappa_0} \leq \lambda_0$. If $(\lambda_1,\lambda_0) \twoheadrightarrow_\xi (\kappa_1,\kappa_0)$, then $(\lambda_1,\lambda_0) \twoheadrightarrow_{\kappa_0} (\kappa_1,\kappa_0)$.
\end{lemma}
\begin{proof}
See \cite[Section 2.2.1]{Foreman2009}.
\end{proof}
Under {\rm GCH\xspace}, in many cases $(\lambda_1,\lambda_0) \twoheadrightarrow (\kappa_1,\kappa_0)$ implies $(\lambda_1,\lambda_0) \twoheadrightarrow_{\kappa_0} (\kappa_1,\kappa_0)$ (see \cite[Proposition 19]{Foreman2009}). For example, this is true for regular cardinals $\kappa_0, \kappa_1, \lambda_0, \lambda_1$.
Foreman \cite{Foreman1983} proved the next result for the case where $\mathbb{P}$ is trivial.
\begin{lemma}\label{preserve closed}
Let $\kappa_1$ be regular. Suppose $\mathbb P$ is $\kappa_1$-c.c., $\mathbb Q$ is $\kappa_1$-closed, and $\Vdash_{\mathbb P} \dot{\mathbb R} \lhd \check{\mathbb Q}$. If $\Vdash_{\mathbb P} (\kappa_1,\kappa_0) \twoheadrightarrow (\mu_1,\mu_0)$, then $\Vdash_{\mathbb P * \dot{\mathbb R}} (\kappa_1,\kappa_0) \twoheadrightarrow (\mu_1,\mu_0)$.
\end{lemma}
\begin{proof}
First, let us claim that if $\mathbb P \times \mathbb Q$ forces that $(\kappa_1,\kappa_0) \twoheadrightarrow (\mu_1,\mu_0)$ holds then also $\Vdash_{\mathbb P \ast \dot{\mathbb R}} (\kappa_1,\kappa_0) \twoheadrightarrow (\mu_1,\mu_0)$.
By Easton's lemma, $\Vdash_{\mathbb P}$ ``$\mathbb Q$ is $\kappa_1$-distributive,'' thus $\Vdash_{\mathbb P * \dot{\mathbb R}}$ ``$\mathbb Q / \mathbb R$ is $\kappa_1$-distributive.'' Let $G * \dot H$ be $\mathbb P * \dot{\mathbb R}$-generic, and let $f : \kappa_1^{<\omega} \to \kappa_1$ be in $V[G][H]$. If $H^\prime$ is $\mathbb Q / H$-generic, then in $V[G][H][H^\prime]$, there is an $X \subseteq \kappa_1$ closed under $f$ such that $|X| = \mu_1$ and $|X \cap \kappa_0| = \mu_0$. By the distributivity of $\mathbb Q / \mathbb R$, $X \in V[G][H]$.
Now suppose $\Vdash_{\mathbb P \times \mathbb Q}$ ``$\dot f : \kappa_1^{<\omega} \to \kappa_1$.'' Let $\langle s_\alpha : \alpha < \kappa_1 \rangle$ enumerate $\kappa_1^{<\omega}$, and let $(p,q) \in \mathbb P \times \mathbb Q$ be arbitrary. Let $(p^0_0,q^0_0)$ decide $\dot f(s_0)$, with $(p^0_0,q^0_0) \leq (p,q)$. If possible, choose $p^1_0 \perp p^0_0$ also below $p$, and let $q^1_0 \leq q^0_0$ be such that $(p^1_0,q^1_0)$ decides $\dot f(s_0)$. For as long as possible, keep choosing a sequence of pairs $(p_0^\alpha,q_0^\alpha)$, such that the $p_0^\alpha$'s form an antichain below $p$ and the $q_0^\alpha$'s form a descending sequence. If we have chosen less than $\kappa_1$ $q^0_\alpha$'s, we can choose a condition below all of them by $\kappa_1$-closure. By the $\kappa_1$-c.c., this must terminate at some $\eta_0 < \kappa_1$, at which point $\langle p_0^\alpha : \alpha < \eta_0 \rangle$ is a maximal antichain below $p$. Let $q_0^* \leq q_0^\alpha$ for all $\alpha < \eta_0$. Next, do the same with respect to $\dot f(s_1)$, but starting with $q^0_1 \leq q^*_0$. Continuing in this way, we get maximal antichains $\langle p^\beta_\alpha : \beta < \eta_\beta \rangle$ in $\mathbb P \restriction p$ for each $\alpha < \kappa_1$, and a descending sequence $\langle q^*_\alpha : \alpha < \kappa_1 \rangle$, with the property that whenever $\alpha < \kappa_1$, $\beta < \eta_\alpha$ and $\alpha \leq \gamma < \kappa_1$, $(p^\beta_\alpha,q^*_\gamma)$ decides $\dot f(s_\alpha)$.
Let $G \subseteq \mathbb P$ be generic over $V$ with $p \in G$. For each $\alpha < \kappa_1$, there is a unique $\beta < \eta_\alpha$ such that $p^\beta_\alpha =_{\mathrm{def}} p^*_\alpha \in G$. In $V[G]$, define a function $f^\prime : \kappa_1^{<\omega} \to \kappa_1$ by $f^\prime(s_\alpha) = \beta$ iff $(p^*_\alpha,q^*_\alpha) \Vdash^V_{\mathbb P \times \mathbb Q} \dot f(s_\alpha) = \beta$. By the hypothesis about $\mathbb P$, let $X \subseteq \kappa_1$ be such that $X$ is closed under $f^\prime$, $|X| = \mu_1$ and $|X \cap \kappa_0| = \mu_0$. Let $\gamma < \kappa_1$ be such that $X^{<\omega} \subseteq \{ s_\alpha : \alpha < \gamma \}$. Next, take $H \subseteq \mathbb Q$ generic over $V[G]$ with $q^*_\gamma \in H$. Then for all $\alpha < \gamma$, $\dot f^{G \times H}(s_\alpha) = f^\prime(s_\alpha)$, since $\{ (p^*_\alpha,q^*_\alpha) : \alpha < \gamma \} \subseteq G \times H$. As $(p,q)$ was arbitrary, $\mathbb P \times \mathbb Q$ forces that there is $X \subseteq \kappa_1$ closed under $\dot f$ such that $|X \cap \kappa_0 | = \mu_0$.
\end{proof}
\section{Local Chang's Conjecture from subcompact cardinals}\label{sec: local cc}
In this section we will prove the consistency of certain instances of Chang's Conjecture relative to the existence of large cardinals at the level of supercompact cardinals.
We start with the concept of \emph{subcompactness}. This large cardinal notion was isolated by Jensen. We will use a generalization due to Brooke-Taylor and Friedman.
\begin{definition}\cite{BrookeTaylorFriedman2013}
Let $\kappa \leq \lambda$ be cardinals. $\kappa$ is $\lambda$-subcompact if for every $A\subseteq H(\lambda)$ there are $\bar\kappa, \bar\lambda < \kappa$, $\bar{A}\subseteq H(\bar\lambda)$ and an elementary embedding \[j\colon \langle H(\bar\lambda), \bar{A}, \in\rangle \to \langle H(\lambda), A, \in \rangle\]
with critical point $\bar\kappa$ and $j(\bar\kappa) = \kappa$.
\end{definition}
Following Neeman and Steel, we say that $\kappa$ is $(+\alpha)$-subcompact if it is $\kappa^{+\alpha}$-subcompact.
It follows immediately from a theorem of Magidor \cite[Lemma 1]{Magidor1971}, that $\kappa$ is $\lambda$-subcompact for all $\lambda$ iff $\kappa$ is supercompact. Nevertheless, subcompactness is level-by-level weaker that supercompactness.
Assuming {\rm GCH\xspace}, if $\kappa$ is $\kappa^{+\alpha + 1}$-supercompact cardinal and $\alpha < \kappa$, then $\kappa$ is $(+\alpha + 1)$-subcompact and the normal measure derived from the supercompact embedding concentrates on $(+\alpha + 1)$-subcompact cardinals.
On the other hand, if $\kappa$ is $(+\alpha + 2)$-subcompact then there are unboundedly many cardinals $\rho < \kappa$ such that $\rho$ is $\rho^{+\alpha + 1}$-supercompact.
\begin{theorem}\label{thm: cc at single point}
Assume {\rm GCH\xspace}. Let $\kappa$ be $(+2)$-subcompact. Then for every regular $\mu < \kappa$ there is $\rho < \kappa$ such that forcing with $\Col(\mu, \rho^{+})\times\Col(\rho^{++}, \kappa)$ forces $(\mu^{+3},\mu^{++})\twoheadrightarrow(\mu^{+},\mu)$.
\end{theorem}
\begin{proof}
As a warm-up, let us show first an easier fact:
\begin{claim} There is $\rho < \kappa$ such that $(\kappa^{++}, \kappa^{+})\twoheadrightarrow (\rho^{++}, \rho^{+})$
\end{claim}
\begin{proof}
Let $\kappa$ be a $(+2)$-subcompact cardinal. Assume, toward a contradiction, that for every $\rho < \kappa$, $(\kappa^{++}, \kappa^{+})\not\twoheadrightarrow (\rho^{++}, \rho^+)$. In particular, for every $\rho < \kappa$ we can pick a function $f_\rho\colon (\kappa^{++})^{<\omega}\to\kappa^{+}$ such that for every $A\subseteq \kappa^{++}$, $|A|=\rho^{++}$, $|f_\rho\image A^{<\omega}| = \rho^{++}$.
Let us code the sequence $\langle f_\eta \colon \eta < \kappa\rangle$ as a subset of $H(\kappa^{++})$. By the $(+2)$-subcompactness of $\kappa$, there $\rho < \kappa$ and elementary embedding:
\[j\colon \langle H(\rho^{++}), \in, \langle g_\eta \colon \eta < \rho\rangle \rangle \to \langle H(\kappa^{++}),\in,\langle f_\eta \colon \eta < \kappa\rangle\rangle\]
Let us look at $A = j\image \rho^{++}$. By the definition of $f_\rho$, $f_\rho \image A^{<\omega}$ has cardinality $\rho^{++}$. In particular, there is a sequence of finite subsets of $A$, $\langle a_\xi \colon \xi < \rho^{++}\rangle$, such that $f_\rho (a_\xi) < f_\rho(a_\zeta) < \kappa^{+}$ for all $\xi < \zeta < \rho^{++}$. Since $a_\xi$ is a finite subset of $j\image \rho^{++}$, $a_\xi = j(b_\xi)$, where $b_\xi$ is a finite subset of $\rho^{++}$. For every pair $\xi < \zeta$, by elementarity, there is some $\eta < \rho$ such that $g_\eta (b_\xi) < g_\eta (b_\zeta) < \rho^{+}$.
Let us define for $\xi < \zeta < \rho^{++}$, $c(\xi, \zeta) = \min \{\eta < \rho \colon g_\eta (b_\xi) < g_\eta(b_\zeta)
\}$. $c$ is a coloring of the pairs of ordinals below $\rho^{++}$. By {\rm GCH\xspace}, $2^\rho = \rho^+$. By the Erd\H{o}s-Rado Theorem, there is a homogeneous subset of $\rho^{++}$ with order type $\rho^{+} + 1$, $H$. Let $\eta$ be its color.
Let us look at the sequence $\langle g_\eta (b_\xi) \colon \xi \in H\rangle$. This is an increasing sequence of length $\rho^+ + 1$ of ordinals below $\rho^+$ - a contradiction.
\end{proof}
Let us now return to the proof of the theorem, which is very similar to the proof of the claim.
Let $\mathbb{L}_\rho = \Col(\mu, \rho^+)\times \Col(\rho^{++}, \kappa)$.
Assume, towards a contradiction, that there is no such $\rho$, i.e. for every $\rho < \kappa$ there is a $\mathbb{L}_\rho$-name, $\dot{f}_\rho$ of a function from $(\kappa^{++})^{<\omega}$ to $\kappa^{+}$ (in the sense of $V$) such that for every subset of cardinality $(\rho^{++})^{V}$, $A$, we have $\Vdash |\dot{f}_\rho\image (A^{{<}\omega})| = (\rho^{++})^{V}$. The sequence $\langle \dot{f}_\rho \colon \rho < \kappa\rangle$ can be coded as a subset of $H(\kappa^{++})$.
Using the $(+2)$-subcompactness, there is $j$ and $\rho$ such that:
\[j\colon \langle H(\rho^{++}), \in, \langle \dot{g}_\eta \colon \eta < \rho\rangle \rangle \to \langle H(\kappa^{++}),\in,\langle \dot{f}_\rho \colon \rho < \kappa\rangle\rangle\]
is elementary. As before, let us look at $A = j\image \rho^{++}$.
By the assumption, $\Vdash_{\mathbb{L}_\rho} |\dot{f}_\rho\image A^{<\omega}| = (\rho^{++})^{V}$.
Let $\langle \dot{x}_\alpha \colon \alpha < \rho^{++}\rangle$ be a sequence of $\mathbb{L}_\rho$-names, such that for every $\alpha < \beta$, $\Vdash \dot{f}_\rho(\dot{x}_\alpha) < \dot{f}_\rho(\dot{x}_\beta)$. Let us pick, by induction, conditions \[p_\alpha = \langle q_\alpha, r_\alpha\rangle \in \mathbb{L}_\rho = \Col(\mu, \rho^+)\times \Col(\rho^{++}, \kappa).\] For every $\alpha$, we find $a_\alpha \in A^{{<}\omega}$ and pick $p_\alpha$ such that $p_\alpha \Vdash \dot{x}_\alpha = \check{a}_\alpha$. Moreover, using the $\rho^{++}$-closure of $\Col(\rho^{++}, \kappa)$, we may pick the sequence $\langle r_\alpha \colon \alpha < \rho^{++}\rangle$ to be decreasing. $|\Col(\mu, \rho^{+})| = \rho^{+}$ and therefore, there is an unbounded subset $J \subseteq \rho^{++}$ such that for every $\alpha \in J$, $q_\alpha = q_\star$, for some fixed $q_\star \in \Col(\mu, \rho^{+})$. By re-arranging the sequence and omitting all elements outside of $J$, we may assume that $J = \rho^{++}$.
To conclude, we can find a sequence of conditions $\langle p_\alpha \colon \alpha < \rho^{++}\rangle$ and a sequence of finite subsets of $A$, $\langle a_\alpha\colon \alpha < \rho^{++}\rangle$, such that:
\begin{enumerate}
\item For all $\alpha$, $p_\alpha = \langle q_\star, r_\alpha\rangle$, where $q_\star\in\Col(\mu, \rho^{+})$ is fixed and $r_\alpha$ is decreasing.
\item For all $\alpha < \beta < \rho^{++}$, $p_\beta \Vdash \dot{f}_\rho (\check{a}_\alpha) < \dot{f}_\rho (\check{a}_\beta)$.
\end{enumerate}
Reflecting downward for every pair $\xi < \zeta$ separately and using the fact that $a_\xi = j(b_\xi)$ for some $b_\xi \in (\rho^{++})^{<\omega}$, we get that there is some $\eta < \rho$ and some condition $s$ in $\Col(\mu,\eta^+)\times\Col(\eta^{++}, \rho)$ such that $s\Vdash \dot{g}_\eta(\check{b}_\xi) < \dot{g}_\eta(\check{b}_\zeta) < (\check{\rho}^{+})^{V}$.
Let us define a coloring, $c\colon [\rho^{++}]^2\to \rho \times V_\rho$ that assigns for each pair $\xi < \zeta < \rho^{++}$ such pair $(\eta, s) \in \rho \times V_\rho$ as above. Since $\rho$ is measurable, and in particular inaccessible, this coloring obtains only $\rho$ many colors. Therefore, by Erd\H{o}s-Rado, we have a homogeneous set $H$ of order type $\rho^{+} + 1$. Let $(\eta, s)$ be its color.
For every $\xi < \zeta$ in $H$, $s\Vdash \dot{g}_\eta(\check{b}_\xi) < \dot{g}_\eta(\check{b}_\zeta)$. We conclude that in the generic extension relative to the forcing $\Col(\mu,\eta^{+})\times\Col(\eta^{++}, \rho)$, there is a subset of $\rho^{+}$ of order type $\rho^{+} + 1$, which is impossible.
\end{proof}
A similar method can be used in order to get instances of Chang's Conjecture with larger gaps between the cardinals, starting from stronger large cardinal assumptions:
\begin{theorem}
Assume {\rm GCH\xspace}, and let $\alpha \geq 2$ be an ordinal. If there is $\kappa > \alpha$ such that $\kappa$ is $(+\alpha)$-subcompact cardinal then there is $\rho < \kappa$ such that $\langle \kappa^{+i} \rangle_{1 \leq i \leq \alpha} \twoheadrightarrow \langle \rho^{+i} \rangle_{1 \leq i \leq \alpha}$. Moreover, if $\alpha$ is a successor ordinal, we may find $\rho < \kappa$ such that
\[\Vdash_{\Col(\rho^{+\alpha},\kappa)} \langle \rho^{+\alpha+i} \rangle_{1 \leq i \leq \alpha} \twoheadrightarrow \langle \rho^{+i} \rangle_{1 \leq i \leq \alpha}.\]
\end{theorem}
\begin{proof}
We will only sketch the case when we collapse cardinals. Towards a contradiction, suppose $\kappa$ is $(+\alpha)$-subcompact and there is no such $\rho < \kappa$. Let $\langle \dot f_\eta : \eta < \kappa \rangle$ be a sequence such that $\dot f_\eta$ is a $\Col(\eta^{+\alpha},\kappa)$-name for a counterexample. Let $\rho < \kappa$ be such that there is an elementary embedding
\[j\colon \langle H(\rho^{+\alpha}), \in, \langle \dot{g}_\eta \colon \eta < \rho\rangle \rangle \to \langle H(\kappa^{+\alpha}),\in,\langle \dot{f}_\eta \colon \eta < \kappa\rangle\rangle.\]
Let $A = j \image \rho^{+\alpha}$, and consider a $\Col(\rho^{+\alpha},\kappa)$-name for $\dot f_\rho \image A$. We may assume that $\Vdash A \subseteq \dot f_\rho \image A$, so that $\Vdash |\dot f_\rho \image A \cap \rho^{+\alpha+i}| \geq \rho^{+i}$ for $1 \leq i \leq \alpha$. It suffices to prove that for all successor $\beta < \alpha$, $\Vdash |\dot f_\rho \image A \cap \rho^{+\alpha+\beta}| = \rho^{+\beta}$, since the limit cases follow by continuity. For such $\beta$, the argument proceeds exactly as before.
\end{proof}
\begin{remark}
Suppose that in the above, $\alpha$ is the successor of a limit ordinal $\lambda$. Using the lemmas of the previous section, we can then force with $\Col(\mu,\rho^{+\lambda})$, where $\mu \leq \cf(\lambda)$, to obtain $(\mu^{+\lambda+ 1}, \mu^{+\lambda}) \twoheadrightarrow (\mu^+,\mu)$. This gives an alternate proof of consistency results from \cite{HayutMagidorMalitz}.
\end{remark}
In the next theorem, we will get the consistency of instances of Chang's Conjecture where the source is double successor of singular cardinal, e.g.\ $(\aleph_{\omega+2}, \aleph_{\omega+1})\twoheadrightarrow(\aleph_{n+1}, \aleph_n)$. In order to achieve this, we need to start with slightly stronger assumption.
\begin{lemma}\label{lem: cc between successor points}
Assume ${\rm GCH\xspace}$. Let $\kappa$ be $\kappa^{++}$-supercomapct cardinal and let $\mathcal{U}$ be a normal measure on $\kappa$, derived from a supercompact embedding.
\begin{enumerate}
\item There is a set of measure one $A$ such that for all $\rho\in A$, $(\kappa^{++}, \kappa^{+})\twoheadrightarrow(\rho^{++},\rho^{+})$. Moreover, there is a set of measure one $A^\prime$ such that for every $\rho\in A^\prime$ and every forcing notion of cardinality $\leq \rho^+$, $\mathbb{Q}$, $\mathbb{Q}\times\Col(\rho^{++}, \kappa)$ forces $(\kappa^{++}, \kappa^{+})\twoheadrightarrow(\rho^{++},|\rho^{+}|)$.
\item For every $\rho \in A$, there is a set $B_\rho\in\mathcal{U}$ such that for every $\eta\in B_\rho$, $(\eta^{++}, \eta^{+})\twoheadrightarrow(\rho^{++},\rho^{+})$. Moreover, for every $\rho \in A^\prime$, there $B_\rho^\prime\in\mathcal{U}$ such that for every $\eta\in B^\prime_\rho$ and every forcing notion $\mathbb{Q}$, $|\mathbb{Q}| \leq \rho^{+}$, $\mathbb{Q} \times \Col(\rho^{++}, \eta)$ forces $(\eta^{++}, \eta^{+})\twoheadrightarrow(\rho^{++},|\rho^{+}|)$.
\item There is $C\in\mathcal{U}$ such that for every $\zeta < \xi$ in $C$, $(\xi^{++}, \xi^{+})\twoheadrightarrow(\zeta^{++},\zeta^{+})$ and moreover for every forcing notion $\mathbb{Q}$ or cardinality $\leq \zeta^+$, $\mathbb{Q} \times \Col(\zeta^{++}, \xi)$ forces $(\xi^{++}, \xi^{+})\twoheadrightarrow(\zeta^{++},|\zeta^{+}|)$.
\end{enumerate}
\end{lemma}
\begin{proof}
Let us show the first assertion. Assume otherwise, and let us pick $\dot{f}_\rho, \mathbb{Q}_\rho$ witnessing it. Namely:
\begin{enumerate}
\item $\mathbb{Q}_\rho$ is a forcing notion of cardinality $\leq \rho^+$.
\item $\dot{f}_\rho$ is a name for a function from $(\kappa^{++})^{<\omega}$ into $\kappa^+$ in the forcing $\mathbb{Q}_\rho\times\Col(\rho^{++},\kappa)$
\item For every set of cardinality $\rho^{++}$, $X\subseteq \kappa^{++}$
\[\Vdash_{\mathbb{Q}_\rho\times\Col(\rho^{++},\kappa)} |\dot{f}_\rho\image \check{X}^{<\omega}| = \rho^{++}\]
\end{enumerate}
Let us assume, without loss of generality, that $\mathbb{Q}_\rho$ is a partial order on $\rho^+$.
Let $j\colon V\to M$ be a $\kappa^{++}$-supercompact embedding such that \[\mathcal{U} = \{Y\subseteq \kappa \colon \kappa\in j(Y)\}.\] Let us work in $M$. There, by elementarity, the forcing $j(\mathbb{Q})_\kappa\times \Col(\kappa^{++}, j(\kappa))$ adds a function $j(\dot{f})_\kappa\colon j(\kappa^{++})^{<\omega}\to j(\kappa^{+})$ such that for every $X\subseteq j(\kappa^{++})$ of cardinality $(\kappa^{++})^M = \kappa^{++}$
\[\Vdash_{j(\mathbb{Q})_\kappa\times\Col(\kappa^{++},j(\kappa))} |j(\dot{f})_\kappa\image \check{X}^{<\omega}| = \check{\kappa}^{++}\]
Take $X = j\image\kappa^{++}$. Let us denote $\mathbb{L} = j(\mathbb{Q})_\kappa\times\Col(\kappa^{++},j(\kappa))$. By the same arguments of Theorem~\ref{thm: cc at single point}, we can find a sequence of conditions $p_\alpha\in \mathbb{L}$ and $a_\alpha \subseteq \kappa^{++}$, finite, such that for every $\alpha < \beta$, $p_\beta \Vdash j(\dot{f})_\kappa(\widecheck{j(a_\alpha)}) < j(\dot{f})_\kappa(\widecheck{j(a_\beta)})$. Reflecting this fact back to $V$, we obtain that for every $\alpha < \beta < \kappa^{++}$ there is an ordinal $\rho < \kappa$ and a condition $r\in\mathbb{Q}_\rho \times \Col(\rho^{++}, \kappa)$ such that \[r\Vdash_{\mathbb{Q}_\rho \times \Col(\rho^{++},\kappa)} \dot f_\rho(\check a_\alpha) < \dot f_\rho(\check a_\beta) < \check \kappa^+.\]
This defines a coloring of pairs of elements in $\kappa^{++}$ with $\kappa$ many colors. By Erd\H{o}s-Rado, there is a homogeneous set of order type $\kappa^{+}+1$, which is impossible.
The second statement follows from the reflection properties of the supercompact cardinal. For the third statement take $C = A \cap \triangle_{\rho\in A} B_\rho$.
\end{proof}
\begin{theorem}
Assume {\rm GCH\xspace}. Let $\kappa$ be $\kappa^{++}$-supercompact and let $\mu < \kappa$ be regular. There is a generic extension in which $\kappa = \mu^{+\omega}$ and \[(\kappa^{++}, \kappa^+) \twoheadrightarrow (\mu^+, \mu).\]
\end{theorem}
\begin{proof}
Let us consider the Prikry type forcing with collapses relative to a normal measure $\mathcal{U}$ on $\kappa$ which is a projection of $\kappa^{++}$-supercompact measure.
Let us describe explicitly the conditions in the forcing notion.
Let $j_\mathcal{U}\colon V\to M$ be the ultrapower embedding. Let us consider the forcing notion $\Col^M((\kappa^{++})^M, j(\kappa))$. This forcing is $\kappa^{+}$-closed (in $V$) and by standard counting arguments, it has only $|\mathcal{P}^M(j(\kappa))|^V = \kappa^{+}$ dense subsets in $M$. Therefore, there is an $M$-generic filter for $\Col^M((\kappa^{++})^M, j(\kappa))$ in $V$, $\mathcal{K}$.
Let us define the conditions for the forcing, $\mathbb{P}$.
\[p = \langle \alpha_0, f_0, \dots, \alpha_{n-1}, f_{n-1}, A, F \rangle\]
is a condition in $\mathbb{P}$ iff:
\begin{enumerate}
\item $\mu < \alpha_0 < \dots < \alpha_{n-1} < \kappa$.
\item $f_i\in\Col(\alpha_{i}^{++}, \alpha_{i+1})$ for $i < n - 1$, and $f_{n- 1} \in \Col(\alpha_{n - 1}^{++}, \kappa)$.
\item $A\in \mathcal{U}$, and $\min A > \alpha_{n - 1} + \rank f_{n - 1}$.
\item $F\colon A\to V$, for every $\alpha\in A$, $F(\alpha)\in\Col(\alpha^{++},\kappa)$ and $[F]_{\mathcal{U}}\in \mathcal{K}$.
\end{enumerate}
Let $p, q\in \mathbb{P}$.
\[p = \langle \alpha_0, f_0, \dots, \alpha_{n-1}, f_{n-1}, A, F\rangle\]
\[q = \langle \beta_0, g_0, \dots, \beta_{m-1}, g_{m-1}, B, G\rangle\]
$p\leq q$ iff:
\begin{enumerate}
\item $n \geq m$
\item For every $i < m$, $\alpha_i = \beta_i$.
\item For every $i < m$, $f_i \supseteq g_i$.
\item If $m \leq i <n$, then $\alpha_i\in B$ and $f_i \supseteq G(\alpha_i)$.
\item $A \subseteq B$ and for all $\alpha \in A$, $F(\alpha) \supseteq G(\alpha)$.
\end{enumerate}
We say that $p\leq^\star q$ if $p\leq q$ and they have the same length.
\begin{claim}
$\mathbb{P}$ satisfies the Prikry property. Namely, for every statement $\Phi$ in the forcing language and condition $p\in\mathbb{P}$, there is $q\leq^\star p$ that decides the truth value of $\Phi$.
\end{claim}
This is folklore, and a complete proof can be found in \cite{Faubion2012}. We remark that there are some minor differences between the presentation there and our presentation. As is standard, if $p = \langle \alpha_0, f_0, \dots, \alpha_{n}, f_n, A, F\rangle$, then $\mathbb P \restriction p \cong \Col(\alpha_{0}^{++},\alpha_1) \times \dots \times \Col(\alpha_{n-1}^{++},\alpha_{n}) \times \mathbb Q$, where $\mathbb Q$ adds no subsets of $\alpha_n^+$.
Since conditions with the same stem are compatible, $\mathbb P$ is $\kappa^+$-c.c. Moreover, every ${<}\kappa$-sized set of conditions sharing a common stem has a lower bound with the same stem.
From the Prikry property and the chain condition, we conclude that every cardinal greater than or equal to $\kappa$ is preserved and that $\kappa$ becomes $\alpha_0^{+\omega}$.
Let $$X_0 = \{ \alpha < \kappa : \forall g \exists Y \exists F \langle \alpha, g ,Y,F \rangle \Vdash (\kappa^{++},\kappa^+) \twoheadrightarrow (\alpha^{++},\alpha^+) \},$$
$$X_1 = \{ \alpha < \kappa : \exists g \exists Y \exists F \langle \alpha, g,Y,F \rangle \Vdash (\kappa^{++},\kappa^+) \not\twoheadrightarrow (\alpha^{++},\alpha^+) \}.$$
By the Prikry property, either $X_0$ or $X_1$ is in $\mathcal U$. Towards a contradiction, suppose $X_1 \in \mathcal U$. Let $\dot{f}\colon(\kappa^{++})^{<\omega}\to\kappa^{+}$ be a name for a function such that if $p_\alpha$ witnesses $\alpha \in X_1$, then $p_\alpha$ forces that for all $A \subseteq \kappa^{++}$ of size $\alpha^{++}$, $| \dot f \image A^{<\omega} | = \alpha^{++}$.
Let $j\colon V\to M$ be a $\kappa^{++}$-supercompact embedding such that the normal measure on $\kappa$ that it defines is the one that we use in the Prikry forcing. Let $A = j \image \kappa^{++}$.
By hypothesis, there is a condition $p = j(p)_\kappa = \langle \kappa, g, Y,F \rangle \in j(\mathbb P)$ such that $p \Vdash^M_{j(\mathbb P)} | j(\dot f) \image A^{<\omega} | = \kappa^{++}$. There is a sequence of names $\langle \dot b_\alpha : \alpha < \kappa^{++} \rangle \subseteq A^{<\omega}$ such that $p$ forces $\langle \dot f(b_\alpha) : \alpha < \kappa^{++} \rangle$ is an increasing sequence of ordinals below $j(\kappa^+)$. The sequence of $\dot b_\alpha$'s is forced to be $j \image \vec a$, where $\vec a$ is a sequence contained in $(\kappa^{++})^{<\omega}$. Let $q = \langle \kappa,g_0', \gamma,g_1',Y',F' \rangle \leq p$. By the Prikry property, $\vec a$ is added by the factor $\Col(\kappa^{++},\gamma)$, and there is an extension $q'$ of $q$ of the same length that decides on a $\Col(\kappa^{++},\gamma)$-name $\dot{\tau}$ for $\vec a$.
Let $\langle r_\alpha : \alpha < \kappa^{++} \rangle$ be a descending sequence of conditions in $\Col(\kappa^{++},\gamma)$ such that $r_0 \leq g_0'$, and $r_\alpha$ decides a value $a_\alpha$ the $\alpha$-th value in the vector $\dot\tau$, $\dot{\tau}(\alpha)$. Thus we have that for each $\alpha < \beta < \kappa^{++}$,
$$\langle \kappa,r_\beta, \gamma, g_1', Y',F' \rangle \Vdash j(\dot f)(\widecheck{j( a_\alpha)}) < j(\dot f)(\widecheck{j(a_\beta)}) < j(\kappa^+).$$
Reflecting this downwards to $V$, we have the following coloring problem: For every pair of ordinals, $\alpha < \beta < \kappa^{++}$, there is a condition $p_{\alpha,\beta}$ that forces $\dot f(a_\alpha) < \dot f(a_\beta) < \kappa^+$. We want to show that this situation is impossible.
There are $2^\kappa$ many conditions in $\mathbb{P}$ and therefore we can think of the above coloring as a coloring from $[\kappa^{++}]^2$ to $2^\kappa$. By Remark~\ref{remark: erdos-rado for 2k colors} we obtain an ordinal $\delta <\kappa^{++}$ and a sequence of ordinals $\{\alpha_i \colon i < \kappa^{+}\}$, cofinal in $\delta$ such that $p_{\alpha_\xi, \alpha_\zeta} = p_{\alpha_\xi, \delta}$ for all $\xi < \kappa^{+}$.
Let us denote by $q_i = p_{\alpha_i, \delta}$. Let us claim that there is a condition $r$ that forces that the set $\{i < \kappa^{+} \colon q_i \in G\}$ (where $G$ is the generic filter) is unbounded. Otherwise, using the chain condition of the forcing, we can find an ordinal $\beta<\kappa^+$ such that $\Vdash \{i < \kappa^{+} \colon q_i \in G\} \subseteq \check\beta$. This is absurd, since $q_{\beta}$ forces the opposite statement.
Let $G$ be a generic filter containing $r$. Let $I = \{i < \kappa^{+} \colon q_i \in G\}$. Since $\kappa^{+}$ is regular in $V[G]$, $\otp I = \kappa^{+}$. For every $\xi, \zeta \in I$, $\xi < \zeta$, $p_{\alpha_\xi, \alpha_\zeta} = p_{\alpha_\xi, \delta} = q_\xi \in G$. Therefore, $V[G]\models f(a_\xi) < f(a_\zeta) < f(a_\delta) < \kappa^+$.
So, in $V[G]$, there is a sequence of ordertype $\kappa^{+} + 1$ of ordinals below $\kappa^{+}$ -- a contradiction.
Therefore, $X_0 \in \mathcal U$, and any condition of the form $\langle X_0,F \rangle$ forces $(\kappa^{++},\kappa^+) \twoheadrightarrow (\alpha_0^{++},\alpha_0^+)$, where $\alpha_0$ is the first member of the Prikry sequence. If we then force with $\Col(\mu,\alpha_0^+)$, we get the desired conclusion.
\end{proof}
\begin{corollary}
It is consistent relative to a $(+2)$-supercompact cardinal that for all even ordinals $\beta < \alpha \leq \omega$, $(\aleph_{\alpha+2},\aleph_{\alpha+1}) \twoheadrightarrow (\aleph_{\beta+2},\aleph_{\beta+1})$.
\end{corollary}
\begin{proof}
If $p = \langle \alpha_0, f_0, \dots, \alpha_{n}, f_n, A, F\rangle \in \mathbb P$ as above, then
$$\mathbb P \restriction p \cong \Col(\alpha_0^{++},\alpha_1) \times \dots \times \Col(\alpha_{n-1}^{++},\alpha_{n}) \times \mathbb P \restriction \langle \alpha_n,f_n,A,F \rangle.$$
If $G \subseteq \mathbb P$ is generic, then for some $n_0 < \omega$, we have $(\kappa^{++},\kappa^+) \twoheadrightarrow (\alpha_m^{++},\alpha_m^+)$ for all $m \geq n_0$. By Lemma \ref{lem: cc between successor points}, there is some $n_1 \geq n_0$ such that $(\alpha_{m+1}^{++},\alpha_{m+1}^{+}) \twoheadrightarrow (\alpha_m^{++},\alpha_m^{+})$ for all $m \geq n_1$. These relations continue to hold after forcing with $\Col(\omega,\alpha_{n_1})$. By the facts mentioned after Remark \ref{remark: erdos-rado for 2k colors}, the desired statement holds in this extension.
\end{proof}
The learned reader may perceive how to use Radin forcing to extend the above result to obtain a model of ZFC in which for \emph{all} even ordinals $\beta < \alpha$, $(\aleph_{\alpha+2},\aleph_{\alpha+1}) \twoheadrightarrow (\aleph_{\beta+2},\aleph_{\beta+1})$. Instead of pursuing this line, we will now work towards showing the consistency of a ``denser'' global Chang's Conjecture using much stronger large cardinal assumptions, answering a question of Foreman.
\section{Global Chang's Conjecture}\label{sec: global cc}
In this section we obtain a model in which $(\kappa^{+}, \kappa) \twoheadrightarrow (\mu^{+}, \mu)$
holds for every cardinal $\mu$ and every successor cardinal $\kappa$, starting from a huge cardinal.
\subsection{Getting Chang's Conjecture between many pairs of regular cardinals}
Towards the goal of the section, we start with a simpler task:
\[(\nu^{+}, \nu) \twoheadrightarrow (\mu^+, \mu)\]
for all regular cardinals $\mu < \nu$. This answers Question 7 of Foreman in \cite{Foreman2010ideals}. Foreman asked whether a weaker statement is consistent, where we assume the larger $\nu$ is a successor cardinal. In our model, we retain many large cardinals. We will then force further to obtain a model in which the smaller cardinal $\mu$ can be singular.
\begin{lemma}
\label{easton projection}
Suppose $\mu<\kappa \leq \lambda < \delta$ are regular, $\kappa$ and $\delta$ are Mahlo, and
\begin{enumerate}
\item $\Vdash_{\E(\mu,\kappa)}$ ``$\dot{\mathbb Q}$ is $\mu$-closed, of size $\leq \lambda$, and preserves the regularity of $\lambda$.''
\item $\Vdash_{\E(\mu,\kappa) * \dot{\mathbb Q}}$ ``$\dot{\mathbb R}$ is $\mu$-closed and of size ${<}\delta$.''
\end{enumerate}
Then there is a projection $\pi : \E(\mu,\delta) \to (\E(\mu,\kappa) * \dot{\mathbb Q}) * (\dot{\mathbb R} \times \dot{\E}(\lambda,\delta))$ such that for all $p$, the first coordinate of $\pi(p)$ is $p{\restriction}\kappa$.
\end{lemma}
\begin{proof}
For brevity, let $\mathbb P = \E(\mu,\kappa) * \dot{\mathbb Q}$. For every inaccessible cardinal $\alpha \in (\lambda,\delta)$, $T(\mathbb P, \dot{\Col}(\lambda,\alpha))$ is $\lambda$-closed and has size $\alpha$. By Lemma \ref{folk}, since $\mathbb P * \dot{\Col}(\lambda,\alpha)$ collapses $\alpha$ to have size $\lambda$, and since $|\mathbb{P}| < \alpha$, $T(\mathbb P, \dot{\Col}(\lambda,\alpha))$ collapses $\alpha$ to $\lambda$ as well and therefore $T(\mathbb P, \dot{\Col}(\lambda,\alpha))$ is forcing-equivalent to $\Col(\lambda,\alpha)$. Thus,
\[ \mathbb P \times \E(\lambda,\delta) \cong \mathbb P \times \prod_{\alpha \in (\lambda,\delta) \cap I}^E T(\mathbb P, \dot{\Col}(\lambda,\alpha)), \]
where $I$ is the class of inaccessible cardinals, and the superscript $E$ indicates that Easton supports are used. We define a projection
\[ \pi_0 : \mathbb P \times \prod_{\alpha \in (\lambda,\delta) \cap I}^E T(\mathbb P, \dot{\Col}(\lambda,\alpha)) \to \mathbb P * \dot{\E}(\lambda,\delta) \]
as follows. $\pi_0(p,q) = \langle p,\tau_q \rangle$, where $\tau_q$ is the canonical $\mathbb P$-name for a function with domain $\dom q$, and $\Vdash \forall \alpha,\, \dot \tau_q(\alpha) = \check q(\alpha)$. To show $\pi_0$ is a projection, suppose $\langle p_1,\dot q_1 \rangle \leq \langle p_0,\tau_{q_0} \rangle$. Since $|\mathbb P| \leq \lambda$, there is an Easton set $X$ such that $\Vdash \dom(\dot q_1) \subseteq \check X$. For each $\alpha \in X$, let $\sigma_\alpha$ be a $\mathbb P$-name such that $p_1 \Vdash \sigma_\alpha = \dot q_1(\check \alpha)$ and if $p \perp p_1$, $p \Vdash \sigma_\alpha= \check q_0(\check \alpha)$. If $q_2 = \{ \langle \alpha,\sigma_\alpha \rangle : \alpha \in X \}$, then $\langle p_1,q_2 \rangle \leq \langle p_0,q_0 \rangle$, and $\langle p_1,\tau_{q_2} \rangle \leq \langle p_1,\dot q_1 \rangle$ because $p_1 \Vdash \tau_{q_2} = \dot q_1$.
Let $\rho < \delta$ be regular and such that $\Vdash_{\mathbb P} |\dot{\mathbb R}| \leq \rho$. Applying Lemma \ref{folk} coordinate-wise, we have the following sequence of projections:
\begin{align*}
\E(\mu,\delta) &\cong \E(\mu,\kappa) \times \E(\mu,\delta) \restriction [\kappa,\delta) \\
&\cong \E(\mu,\kappa) \times \Col(\mu,\lambda) \times \Col(\mu,\rho) \times \E(\mu,\delta) \restriction (\lambda,\delta) \\
&\to (\E(\mu,\kappa) * \dot{\mathbb Q} * \dot{\mathbb R}) \times \E(\lambda,\delta) \\
&\cong (\mathbb P * \dot{\mathbb R}) \times \prod_{\alpha \in (\lambda,\delta) \cap I}^E T(\mathbb P, \dot{\Col}(\lambda,\alpha)) \\
&\to \mathbb P * (\dot{\mathbb R} \times \dot{\E}(\lambda,\delta)),
\end{align*}
as desired.
\end{proof}
The next lemma answers a question of Shioya \cite{Shioya2011}, who asked if $\kappa$ is a huge cardinal with target $\delta$, and $\mu< \kappa$ is regular, does $\E(\mu,\kappa) * \dot \E(\kappa,\delta)$ force $(\mu^{++},\mu^{+})\twoheadrightarrow(\mu^+,\mu)$? He noted in the same paper that if we allow more distance between the cardinals, then the answer is yes: for example $\E(\mu,\kappa) * \dot \E(\kappa^+,\delta)$ forces $(\mu^{+3},\mu^{++})\twoheadrightarrow(\mu^+,\mu)$. The next lemma covers these cases and many others. The main issue is to understand the behavior of a potential master condition. The argument shows that the use of posets like the Silver collapse, which is designed to get master conditions under control (see \cite{Foreman2010ideals}), is not actually needed in this context.
\begin{lemma}
\label{lemma: easton collapse give cc}
Suppose $\kappa$ is a huge cardinal, $j : V \to M$ is a huge embedding with critical point $\kappa$, $j(\kappa) = \delta$, and $\mu,\lambda$ are regular cardinals with $\mu<\kappa \leq \lambda < \delta$. Suppose also:
\begin{enumerate}
\item $\Vdash_{\E(\mu,\kappa)}$ ``$\dot{\mathbb Q}$ is $\check\kappa$-directed-closed, of size $\leq \check\lambda$, and preserves the regularity of $\check\lambda$.''
\item $\Vdash_{\E(\mu,\kappa) * \dot{\mathbb Q}}$ ``$\dot{\mathbb R}$ is $\check\kappa$-directed-closed, of size ${<}\check\delta$.''
\end{enumerate}
Then it is forced by $(\E(\mu,\kappa) * \dot{\mathbb Q}) * (\dot{\mathbb R} \times \dot{\E}(\lambda,\delta))$ that $(\lambda^+,|\lambda|) \twoheadrightarrow (\mu^+,\mu)$.
\end{lemma}
\begin{proof}
Let $G * g * (h \times H)$ be $(\E(\mu,\kappa) * \dot{\mathbb Q}) * (\dot{\mathbb R} \times \dot{\E}(\lambda,\delta))$-generic. By Lemma \ref{easton projection} there is a further forcing yielding an $\E(\mu,\delta)$-generic $\hat G$ with $G * g * (h \times H) \in V[\hat G]$. The embedding can be extended to $\bar{j} : V[G] \to M[\hat G]$.
Since $M[\hat G] \models |g*h| < \check\delta$, and $\bar{j}(\mathbb Q * \mathbb R)$ is $\delta$-directed-closed, there is a master condition $(q,r) \in \bar{j}(\mathbb Q * \mathbb R)$ below $\bar{j} \image (g * h)$. Forcing below this, we get an extended embedding $\hat{j} : V[G*g*h] \to M[\hat G * \hat g * \hat h]$.
In $M[\hat{G} * \hat g]$, for any $\alpha < \delta$, $\hat{j} \image H {\restriction} \alpha$ is a directed subset of $E(j(\lambda),j(\alpha))^{M[\hat{G} * \hat g]}$ of size $<\delta$. Hence we define $m_\alpha = \inf \hat{j} \image H {\restriction} \alpha$.
Note that the restriction maps are continuous in the sense that for any ordinals $\alpha < \beta < \gamma$ and any $X \subseteq \E(\alpha,\gamma)$ with a lower bound, $(\inf X) {\restriction} \beta = \inf \{ p {\restriction} \beta : p \in X \}$. Since $H {\restriction} \alpha = \{ p {\restriction} \alpha : p \in H {\restriction} \beta \}$ for any $\alpha < \beta < \delta$, we have:
\begin{align*}
m_\beta {\restriction} j(\alpha) & =(\inf \{ \hat{j}(p) : p \in H {\restriction} \beta \}) \restriction j(\alpha) = \inf \{ \hat{j}(p) {\restriction} j(\alpha) : p \in H {\restriction} \beta \} \\
& = \inf \{ \hat{j}(p {\restriction} \alpha) : p \in H {\restriction} \beta \} = \inf \{ \hat{j}(p) : p \in H {\restriction} \alpha \} = m_\alpha.
\end{align*}
In $M[\hat{G} * \hat g]$, let $m = \bigcup_{\alpha < \delta} m_\alpha$. To show that $m \in E(j(\lambda),j(\delta))^{M[\hat G * \hat g]}$, let $\gamma = \sup j \image \delta < j(\delta)$. First note $M[\hat G * \hat g]$ thinks that $\dom m$ is an Easton subset of $\gamma$. This is because $\dom m = \bigcup_{p \in H} \dom j(p)$, $M[\hat G] \models |H| = \delta$, and for all $p \in H$, $M[\hat G] \models$ ``$\dom j(p)$ is an Easton subset of $\gamma \setminus \delta$.'' Second, for every $\beta < \gamma$, $\dom m(\beta)$ is a bounded subset of $j(\lambda)$. For $\beta < \gamma$, if $\alpha < \delta$ is such that $j(\alpha) > \beta$, then $m(\beta)$ is ``frozen'' by $m_\alpha$, as $m_\xi(\beta) = m_\alpha(\beta)$ for all $\xi \geq \alpha$.
Therefore if we take a generic $\hat{H} \subseteq E(j(\lambda),j(\delta))^{M[\hat{G} *\hat g]}$ over $V[\hat{G} * \hat g * \hat h]$ with $m \in \hat{H}$, then we get an extended elementary embedding $\tilde{j} : V[G * g * (h \times H)] \to M[\hat G * \hat g * (\hat h \times \hat H)]$. If $f : \delta^{{<}\omega} \to \delta$ is in $V[G * g * (h \times H)]$, then $j \image \delta = \tilde{j} \image \delta$ is closed under $j(f)$. In $M[\hat G * \hat g * (\hat h \times \hat H)]$, $| j \image \delta \cap j(\lambda) | = |\lambda| = \mu$, and $j \image \delta$ has size $\delta = j(\kappa)$. Thus by elementarity, there is some $X \subseteq \delta$ of size $\kappa$ closed under $f$ and such that $|X \cap \lambda | = \mu$ in $V[G * g * (h \times H)]$.
\end{proof}
\begin{lemma}
\label{below huge}
Let $\kappa$ be huge with witnessing embedding $j : V \to M$, and let $\mathcal{U}$ be the normal measure on $\kappa$ derived from $j$.
There is $A\in\mathcal{U}$ such that for all regular cardinals $\alpha < \beta \leq \gamma < \delta$ with $\beta, \delta \in A$, and for every notion of forcing of the form $(\E(\alpha,\beta) * \dot{\mathbb Q}) * (\dot{\mathbb R} \times \dot{\E}(\gamma,\delta))$, where
\begin{enumerate}
\item $\Vdash_{\E(\alpha,\beta)}$ ``$\dot{\mathbb Q}$ is $\beta$-directed-closed, of size $\leq \gamma$, and preserves the regularity of $\gamma$,''
\item $\Vdash_{\E(\alpha,\beta) * \dot{\mathbb Q}}$ ``$\dot{\mathbb R}$ is $\beta$-directed-closed and of size ${<}\delta$.''
\end{enumerate}
forces $(\gamma^+, |\gamma|)\twoheadrightarrow (\alpha^+, \alpha)$.
\end{lemma}
\begin{proof}
Let $\varphi(\beta,\delta)$ stand for the assertion that whenever $\alpha < \beta \leq \gamma < \delta$ are regular and
(1) and (2) hold of $\dot{\mathbb Q}$ and $\dot{\mathbb R}$ as above, then $(\E(\alpha,\beta) * \dot{\mathbb Q}) * (\dot{\mathbb R} \times \dot{\E}(\gamma,\delta))$ forces $(\gamma^+, |\gamma|)\twoheadrightarrow(\alpha^+, \alpha)$.
By Lemma \ref{lemma: easton collapse give cc}, $V \models \varphi(\kappa,j(\kappa)).$ Since $M^{j(\kappa)} \subseteq M$, this holds in $M$ as well. Reflecting this statement once, we find $A_0\in\mathcal{U}$ such that $\varphi(\beta,\kappa)$ holds for all $\beta \in A_0$. Next, for all $\beta \in A_0$, we can reflect again to find a set $A_\beta \in \mathcal U$ such that $\varphi(\beta,\delta)$ holds for all $\delta \in A_\beta$. Take $A = A_0 \cap \triangle_{\beta \in A_0} A_\beta$.
\end{proof}
Let $\kappa$ be a huge cardinal, with $A \subseteq \kappa$ as above, and without loss of generality assume $A$ contains only ${<}\kappa$-supercompact cardinals. Let $\langle \alpha_i : i < \kappa \rangle$ be the increasing enumeration of the closure of $A \cup \{\omega\}$. We define an Easton-support iteration $\langle \mathbb P_i, \dot{\mathbb Q}_j : i \leq \kappa, j < \kappa \rangle$ as follows:
If $\alpha_i$ is regular, $\Vdash_i \dot{\mathbb Q}_i = \dot{\E}(\alpha_i,\alpha_{i+1})$. If $\alpha_i$ is singular, $\Vdash_i \dot{\mathbb Q}_i = \dot{\E}(\alpha_i^+,\alpha_{i+1})$. Note that since each cardinal in $A$ is sufficiently supercompact, the cardinality assumptions of Lemma \ref{below huge} are satisfied at singular limits. If $i<j$, then $\mathbb P_{j+1}$ forces $(\alpha_{j+1},\beta)\twoheadrightarrow(\alpha_i^+,\alpha_i)$, where $\beta$ is the cardinal predecessor of $\alpha_{j+1}$ in $\mathbb P_{j+1}$. By Lemma \ref{preserve closed}, this is preserved by the tail $\mathbb P_\kappa / \mathbb P_{j+1}$, which is $\alpha_{j+1}$-closed.
After forcing with $\mathbb P_\kappa$, we have:
\begin{claim}\label{cc between all successors of regulars}
$V_\kappa^{\mathbb{P}_\kappa} \models (\alpha^+,\alpha) \twoheadrightarrow (\beta^+,\beta)$ for all pairs of regular cardinals $\beta < \alpha$. Furthermore, this is preserved by $\Col(\gamma_0,\gamma_1)$, whenever $\beta \leq \gamma_0 \leq \gamma_1 \leq \alpha$ are regular (where $\beta < \alpha$).
\end{claim}
We do not exclude the cases in which $\beta = \gamma_0$ or $\alpha = \gamma_1$. Note that if both equations hold then in the generic extension $\beta^{+} = \alpha^{+}$ and $|\alpha| = \beta$. In this case the assertion holds trivially.
The stronger claim holds because if $\beta$ is the next regular cardinal $\geq |\mathbb P_i|$ and $\alpha$ is such for $\mathbb P_j$, then it is forced by $\mathbb P_i$ that $(\mathbb P_{j+1} / \mathbb P_i)\times \Col(\gamma_0,\gamma_1)$ has a form satisfying the hypotheses of Lemma \ref{below huge}. As $\Col(\gamma_0,\gamma_1)$ is $\alpha^+$-c.c.\ and $\mathbb P_\kappa / \mathbb P_{j+1}$ is $\alpha^+$-closed, Lemma \ref{preserve closed} implies this instance of Chang's Conjecture continues to hold in $V^{\mathbb P_\kappa \ast \Col(\gamma_0,\gamma_1)}$.
\begin{claim}\label{claim:preserving almost hugeness}
Assume that $\kappa$ is almost huge cardinal with a witnessing elementary embedding $j$ such that $\delta=j(\kappa)$ is Mahlo and $\kappa \in j(A)$. Assume also that $\sup j\image \delta = j(\delta)$. Then in the generic extension by $j(\mathbb{P}_\kappa)$, $j$ extends to an elementary embedding witnessing that $\kappa$ is almost huge.
\end{claim}
\begin{proof}
Let $\hat{\mathbb{P}} = j(\mathbb{P}_\kappa)$. Let us analyse the forcing notion $j(\hat{\mathbb{P}})$.
$\hat{\mathbb{P}}$ is defined as an Easton support iteration of Easton collapses, between the elements in the closure of the set $j(A)$. Note that since $M \cap V_{\delta} = V_{\delta}$, $V$ and $M$ compute this iteration in the same way. $j(\hat{\mathbb P})$ is an Easton support iteration in the model $M$, of length $j(\delta)$, between the points in the closure of $j^2(A)$. Note that \[j^2(A) \cap \delta = j(j(A) \cap \kappa) = j(A).\] Therefore, $j(\hat{\mathbb{P}}) = \hat{\mathbb{P}} \ast \dot{\mathbb{Q}}$ where $\dot{\mathbb{Q}}$ is forced to be a $\delta$-closed forcing notion.
Let $G \subseteq \hat{\mathbb{P}}$ be a $V$-generic filter. In $V[G]$, we will define a filter $H \subseteq j(\hat{\mathbb{P}})$ generic over $M[G]$ such that for every $p\in G$, $j(p) \in G \ast H$.
We imitate the proof of Lemma~\ref{lemma: easton collapse give cc}. For every $\alpha < \delta$ inaccessible, let $m_\alpha = \bigcup_{p\in G \cap V_\alpha} j(p)$.
Since we apply $j$ on $\alpha$ many elements and $\alpha < \delta$, $m_\alpha \in M[G]$. Also, for every $\alpha < \beta$, $m_\beta \restriction j(\alpha) = m_\alpha$.
$\hat{\mathbb{P}}$ is $\delta$-c.c.\ and therefore $M$ models that $j(\hat{\mathbb{P}})$ is $j(\delta)$-c.c. As $\delta$ is inaccessible, $M[G]$ can compute an enumeration of the set of all maximal antichains of $j(\hat{\mathbb{P}})\restriction [\delta, j(\delta))$ in a sequence of length $j(\delta)$. $|j(\delta)|^V = \delta$, so $V[G]$ can enumerate those maximal antichains in a sequence of length $\delta$. Since all those antichains are bounded below $j(\delta)$, we may pick an enumeration $\langle \mathcal{A}_i \colon i < \delta\rangle \in V[G]$ in which $\mathcal{A}_\alpha$ is a maximal antichain in $j(\hat{\mathbb{P}} \restriction \alpha)$ (here we use the fact that $\sup j\image \delta = j(\delta)$).
Let us define a decreasing sequence of conditions $\langle q_i \colon i < \delta\rangle \subseteq \mathbb{Q}$. We require that $q_\alpha \leq m_\alpha$, $q_\alpha \in \mathcal{A}_\alpha$ and that the support of $q_\alpha$ is a subset of $j(\alpha)$. For every $\alpha < \delta$, the sequence $\langle q_\beta \colon \beta < \alpha\rangle$ is a member of $M$. By the $\delta$-closure of $\mathbb{Q}$, one can always pick a condition $q_\alpha$ stronger than all previous conditions. By the properties of $\mathbb{Q}$, it is clear that one can choose $q_\alpha$ to have support which is contained in the union of the supports of $q_\beta$, $\beta < \alpha$.
Let $H$ be the filter generated by $\{q_i \colon i < \delta\}$. By the construction of this sequence, $H$ meets every maximal antichain in $M$ of the forcing notion $j(\hat{\mathbb{P}})$. Therefore, it is $M[G]$-generic.
By Silver's criteria, $j\colon V\to M$ extends to an elementary embedding $\tilde{j}\colon V[G] \to M[G][H]$. Let us claim that $M[G][H]^{{<}\delta} \subseteq M[G][H]$. Indeed, let $\langle \dot{x}_i \colon i < \alpha\rangle$ be a sequence of names of elements of $M$ and assume that $\alpha < \delta$. Without loss of generality, we may assume that $\dot{x_i}$ is a name of an ordinal for every $i < \alpha$. By the chain condition of $\hat{\mathbb{P}}$, this sequence can be encoded as a set of ordinals of cardinality $<\delta$, and therefore belongs to $M$. Since $G\in M[G][H]$, we conclude that also its realization is in $M$, as needed.
\end{proof}
In fact, for our goals it is sufficient to note only that some of the supercompactness of $\kappa$ is preserved.
For the next section we need a stronger version of Lemma~\ref{lemma: easton collapse give cc} and Claim~\ref{cc between all successors of regulars}. We will need to know that a stronger type of reflection holds between pair of elements from $A$.
\begin{definition}[Magidor-Malitz Quantifiers]
Let $M$ be a model over the language $\mathcal{L}$. We enrich $\mathcal{L}$ with the quantifiers $Q^n$ with the following interpretation:
\[ M \models Q^n x_0, \dots x_{n-1} \varphi(x_0, \dots, x_{n-1}, p)\]
iff
\[\exists I \subseteq M,\, |I| = |M|,\, \forall a_0,\dots, a_{n-1}\in I,\, M \models \varphi(a_0, \dots, a_{n-1}, p)\]
\end{definition}
A set $I\subseteq M$ satisfying
\[\forall a_0,\dots, a_{n-1}\in I,\, M \models \varphi(a_0, \dots, a_{n-1}, p)\]
is called a $\varphi$-block.
The Magidor-Malitz quantifiers were defined by Menachem Magidor and Jerome Malitz in \cite{MagidorMalitz1977}. In this paper, they showed that under $\diamondsuit(\aleph_1)$ a certain compactness theorem holds for the language $\mathcal{L}(Q^{<\omega})$ - the first order logic extended by adding the Magidor-Malitz quantifiers.
We say that $A\prec_{Q^n} B$ if $A$ is $\mathcal{L}(Q^n)$-elementary substructure of $B$. We write $\mu\twoheadrightarrow_{Q^n} \nu$ if for every model $B$ of cardinality $\mu$, there is a $Q^n$-elementary submodel $A$ of cardinality $\nu$.
\begin{lemma}
Suppose $\kappa$ is a huge cardinal, $j : V \to M$ is a huge embedding with critical point $\kappa$, $j(\kappa) = \delta$, and $\mu,\lambda$ are regular cardinals with $\mu<\kappa \leq \lambda < \delta$. Suppose also:
\begin{enumerate}
\item $\Vdash_{\E(\mu,\kappa)}$ ``$\dot{\mathbb Q}$ is $\kappa$-directed-closed, of size $\leq \lambda$, and preserves the regularity of $\lambda$.''
\item $\Vdash_{\E(\mu,\kappa) * \dot{\mathbb Q}}$ ``$\dot{\mathbb R}$ is $\kappa$-directed-closed, of size ${<}\delta$.''
\end{enumerate}
Then it is forced by $(\E(\mu,\kappa) * \dot{\mathbb Q}) * (\dot{\mathbb R} \times \dot{\E}(\lambda,\delta))$ that $\lambda^+ \twoheadrightarrow_{Q^{<\omega}} \mu^+$.
\end{lemma}
\begin{proof}
First, let us note that it is enough to deal with models $\mathcal{A}$ which are transitive elementary submodels of $H(\delta^{+})$. Indeed let $\mathcal{A}^\prime$ be an algebra on $\delta$. Clearly, $\mathcal{A}^\prime \in H(\delta^+)$. Let $\mathcal{A}$ be a transitive elementary submodel of $H(\delta^{+})$ of cardinality $\delta$ that contains $\mathcal{A}^\prime$ as an element. Assume that $\mathcal{B} \prec_{Q^n} \mathcal{A}$. Let us claim that $\mathcal{B}^\prime = \mathcal{B} \cap \mathcal{A}^\prime \prec_{Q^n} \mathcal{A}^\prime$. This is true, as any $Q^n$ statement in $\mathcal{A}^\prime$ is equivalent to a $Q^n$ statement in $\mathcal{A}$.
Let $\mathcal A$ be a transitive elementary structure of $H(\delta^+)$ of size $\delta$. We may assume that for every formula $\Phi$ of the form $Q^n x_0, \dots, x_{n-1} \varphi(x_0, \dots, x_{n-1}, p)$ there is function in the language of $\mathcal{A}$, $f_\Phi$ such that $f_\Phi\colon \mathcal{A}\to \mathcal{A}$ is either constant (if $\neg \Phi$) or one-to-one (if $\Phi$ holds), and if it is one-to-one, then \[\mathcal{A} \models \forall y_0, \dots, y_{n-1} \varphi(f_\Phi(y_0), \dots, f_\Phi(y_{n-1}), p).\]
Let $j : V[G * g * (h \times H)] \to M[\hat G * \hat g * (\hat h \times \hat H)]$ be as in Lemma \ref{lemma: easton collapse give cc}, and for brevity denote the domain and codomain by $V'$ and $M'$ respectively. We want to show that $j \image \mathcal A$ is a $Q^{{<}\omega}$-elementary substructure of $j(\mathcal A)$.
Let $\Phi$ be a formula of the form:
\[Q^n x_0, \dots, x_{n-1} \varphi(x_0, \dots, x_{n-1}, j(p)),\]
Let us assume, by induction, that every proper subformula of $\Phi$ is satisfied by $j(\mathcal{A})$ in $M'$ if and only if it is satisfied by $j\image \mathcal{A}$ in $M'$.
First, let us assume $M' \models ``j(\mathcal{A}) \models \Phi."$ By elementarity, \[V' \models `` \mathcal{A} \models Q^n x_0, \dots, x_{n-1} \varphi(x_0, \dots, x_{n-1}, p)."\] By the observation above, there is a function $f_\Phi$ witnessing this fact in $V'$ and clearly, $j(f_\Phi)$ is a one-to-one function on $j \image \mathcal{A}$ witnessing $j \image \mathcal{A} \models \Phi$.
On the other hand, assume that
\[M' \models `` j\image \mathcal{A} \models Q^n x_0, \dots, x_{n-1} \varphi(x_0, \dots, x_{n-1}, j(p))."\]
Let $I\subseteq j\image \mathcal{A}$ be a $\varphi$-block. We want to show that there is a corresponding $\varphi$-block (for the parameter $p$) also in $V'$. Note that the forcing to obtain $\hat G * \hat g * (\hat h \times \hat H)$ from $G * g * (h \times H)$ is of the form $\mathbb{Q}_0\ast \mathbb{Q}_1$ where $\mathbb{Q}_0$ is a precaliber-$\delta$ forcing, and $\mathbb{Q}_1$ is a $\delta$-closed forcing. Let us denote the generic filter for $\mathbb{Q}_0$ by $K_0$ and the generic filter for $\mathbb{Q}_1$ by $K_1$.
In order to find the $\varphi$-block in $V'$, we will show that the existence of such $\varphi$-block in $V'[K_0][K_1]$ implies the existence of a corresponding $\varphi$-block in $V[K_0]$, and that the existence of the later $\varphi$-block in $V'[K_0]$ implies the existence of a similar $\varphi$-block in $V'$.
In $M'$, there is a $\varphi$-block $I \subseteq j\image \mathcal{A}$ of size $\delta$. Note that all its elements are of the form $j(a)$ for some $a\in V'$. Since $M' \subseteq V'[K_0][K_1]$, there is a $\varphi$-block in $V[K_0][K_1]$. In $V[K_0]$, let $\langle \dot{x}_i \colon i < \delta\rangle$ be a sequence of $\mathbb{Q}_1$-names such that $V[K_0][K_1]\models\text{``} \{ j(\dot{x}_i^{K_1}) \colon i < \delta\}$ is a $\varphi$-block''. In $V[K_0]$ let us construct a decreasing sequence of conditions $\langle q_i \colon i < \delta\rangle\subseteq \mathbb{Q}_1$ such that $q_0 \Vdash$``$ \{j(\dot{x}_i) \colon i < \delta\}$ is a $\varphi$-block in $j\image \mathcal{A}$'' and $q_i \Vdash \dot{x}_i = \check{a}_i$, for some $a_i \in V'$. We claim that $\{a_i \colon i < \delta\}$ is a $\varphi$-block for $\mathcal{A}$ in $V'$ (with parameter $p$). For every $\alpha_0 < \alpha_1 < \cdots < \alpha_{n-1} < \delta$, the condition $q = q_{\alpha_{n-1}+1}$ forces $j(a_{\alpha_0}), j(a_{\alpha_1}), \dots, j(a_{\alpha_{n-1}})\in I$. Thus, $q \Vdash j\image \mathcal{A} \models \varphi(j(a_{\alpha_0}), \dots, j(a_{\alpha_{n-1}}), j(p))$. This is a proper subformula of $\Phi$ and thus, by the induction hypothesis, \[q \Vdash M ' \models j(\mathcal{A}) \models \varphi(j(a_{\alpha_0}), \dots, j(a_{\alpha_{n-1}}), j(p)).\] By elementarity, \[q \Vdash V' \models \mathcal{A} \models \varphi(a_{\alpha_0}, \dots,a_{\alpha_{n-1}}, p).\]
This statement is about the ground model $V'$, so it does not depend on the condition $q$.
We conclude that in $V[K_0]$ there is a $\varphi$-block, $I'\subseteq \mathcal{A}$ of cardinality $\delta$.
Work in $V'$. Let $\langle \dot{y}_i \colon i < \delta\rangle$ be a $\mathbb{Q}_0$-name such that $\langle a_i \colon i < \delta\rangle$ is its $K_0$ realization. In $V'$, let us pick conditions $\langle r_i \colon i < \delta\rangle \subseteq \mathbb{Q}_0$ such that $r_i \Vdash \dot{y}_i = \check{b}_i$ for some $b_i\in V'$, and each $r_i$ forces that $\{ \dot{y}_i \colon i < \delta\}$ is a $\varphi$-block.
By the $\delta$-precaliber of $\mathbb{Q}_0$, there is $X \in [\delta]^\delta$ such that $\{ r_\alpha : \alpha \in X \}$ generates a filter. The set $I'' = \{ b_\alpha : \alpha \in X \}$ is a $\varphi$-block for $\mathcal A$ in $V'$:
indeed, if $\alpha_0 < \alpha_1 < \cdots < \alpha_{n-1} \in X$, then there is a condition $r\in \mathbb{Q}_0$ stronger than all the conditions $r_{\alpha_1}, \dots, r_{\alpha_{n-1}}$.
$r \Vdash V' \models \mathcal{A} \models \varphi(b_{\alpha_1}, \dots, b_{\alpha_{n-1}}, p)$.
But this is a statement about the ground model, so it does not depend on the condition $r$. We conclude that:
\[V' \models ``\mathcal A \models Q^n x_0, \dots, x_{n-1} \varphi(x_0, \dots, x_{n-1}, p),"\]
so by elementarity, $M' \models ``j(\mathcal{A}) \models \Phi$.''
\end{proof}
Applying the reflection argument of Lemma \ref{below huge}, we conclude that the measure $\mathcal{U}$ generated from the huge embedding contains a set $A$ such that every pair of elements $\alpha < \beta$ in $A$ satisfies the conclusion of Lemma \ref{below huge}, when replacing the Chang's relation $\twoheadrightarrow$ with the stronger relation $\twoheadrightarrow_{Q^{{<}\omega}}$.
Let us look at the model after the iteration of the Easton collapses. We can't conclude that the stronger version of Claim \ref{cc between all successors of regulars} holds, since we do not have a preservation lemma similar to Lemma \ref{preserve closed} for Magidor-Malitz reflection. Thus we can only conclude the following version:
\begin{claim}\label{claim: mm reflection in successor of regulars}
Let $\mathbb{P}_\kappa$ be the iteration defined in \ref{cc between all successors of regulars}.
If $\alpha > \beta$ are regular in the generic extension by $\mathbb{P}_\kappa$ then there is an $\alpha^+$-c.c.\ complete subforcing $\mathbb{P}_i$ such that $\Vdash_{\mathbb{P}_i} \alpha^{+}\twoheadrightarrow_{Q^{{<}\omega}} \beta^+$ and $\mathbb{P}_\kappa / \mathbb{P}_i$ is $\alpha^{+}$-closed. The same holds when replacing $\mathbb{P}_\kappa$ by $j(\mathbb{P}_\kappa)$.
\end{claim}
\subsection{Radin forcing}
Work in the model of Claim~\ref{claim:preserving almost hugeness}. In this model, ${\rm GCH\xspace}$ holds high above $\kappa$, and $\kappa$ is almost-huge. In particular, $\kappa$ is measurable and $o(\kappa) = \kappa^{++}$ so we can force with the Radin forcing, while collapsing the cardinals between points in the Radin club. We will show that in the generic extension, for every $\mu < \nu < \kappa$, where $\nu$ is a successor, $(\nu^+,\nu)\twoheadrightarrow (\mu^+,\mu)$ holds.
We start with a pair of preservation lemmas:
\begin{lemma}\label{lemma: small forcing force cc after q2}
Let $\alpha < \beta$ be regular cardinals such that $\beta^{<\beta} = \beta$, and assume that $\beta^+\twoheadrightarrow_{Q^2} \alpha^+$. Assume that $\mathbb{Q}$ is a $\beta$-c.c.\ forcing notion of size $\leq \beta$, and $\mathbb{Q}$ preserves $\alpha^+$. Then $\mathbb{Q}$ forces $(\beta^+, \beta)\twoheadrightarrow (\alpha^+, |\alpha|)$.
\end{lemma}
\begin{proof}
Let $\dot{f}$ be a name of a function from $(\beta^+)^{<\omega}$ to $\beta$, such that it is forced that there is no set $A$ of cardinality $\alpha^+$ such that $|\dot{f}\image A^{<\omega}| \leq \alpha$.
Let $\mathcal{A}$ be an elementary substructure of $H(\chi)$ of cardinality $\beta^+$, $\chi$ large enough, $\dot{f}, \mathbb{Q}\in \mathcal{A}$ and let $\mathcal{B} \prec_{Q^2} \mathcal{A}$, $|\mathcal{B}| = \alpha^+$, and $\dot{f}, \mathbb{Q}\in \mathcal{B}$. Let us look at the elements $\dot{f}(a)$, $a\in (\mathcal{B}\cap \beta^+)^{<\omega} = \mathcal{B} \cap (\beta^{+})^{{<}\omega}$. It is forced that there are $\alpha^+$ many elements $a\in \mathcal{B}$ with different realizations for $\dot{f}(a)$.
Thus, in the generic extension, there is a set $\dot{I}\subseteq \mathcal{B}$ such that every pair of elements of it obtain different values under $\dot{f}$. We claim that in the ground model, one can find a set of full cardinality $I \subseteq \mathcal{B}$ such that for every $a, b\in I$ there is a condition $q\in \mathbb{Q}$ that forces $f(a) \neq f(b)$.
Let us look at the collection of all subsets of $I^\prime \subseteq \mathcal{B} \cap (\beta^{+})^{{<}\omega}$ such that for every pair of distinct elements $a, b\in I^\prime$ there is a condition $q\in \mathbb{Q}$, such that $q\Vdash \dot{f}(\check a) \neq \dot{f}(\check b)$. Let $I_m$ be maximal with respect to this condition. If $|I_m| \leq \alpha$, then in the generic extension for every $a\in \mathcal{B}\cap (\beta^{+})^{{<}\omega}$, $\dot{f}(a) \in \dot{f}\image I_m$. If $a\in I_m$, this is clear, and otherwise, there is no condition $q$ that forces it to be different from every element in this set, so every condition forces it to be equal to one of them. In particular, in the generic extension $\alpha^{+} = |\dot{f} \image (\mathcal{B} \cap (\beta^{+})^{{<}\omega})| \leq |I_m| \leq |\alpha|$ - a contradiction to the assumption that $\alpha^{+}$ is not collapsed in the generic extension.
Let $I\subseteq \mathcal{B} \cap (\beta^{+})^{{<}\omega}$ be a set of cardinality $\alpha^{+}$ such that for every $a, b\in I$, there is $q\in \mathbb{Q}$ that forces $\dot{f}(a) \neq \dot{f}(b)$.
By elementarity, for every $a, b\in I$, there is $q\in \mathbb{Q} \cap \mathcal{B}$ forcing the same statement. Therefore $\mathcal{B}$ satisfies the following $Q^2$-sentence:
\[Q^2 a, b\in \mathcal{B}\, \exists q\in \mathbb{Q},\, q\Vdash \dot f(a) \neq \dot f(b)\]
Thus, $\mathcal{A}$ satisfies the same formula: There is a set $I\subseteq \mathcal{A}$, $|I| = \beta^+$, and for every $a, b\in I$, there is $q\in \mathbb{Q}$ such that $q \Vdash \dot{f}(a) < \dot{f}(b)$ or $q \Vdash \dot{f}(a) > \dot{f}(b)$. This defines a coloring $h\colon [\beta^+]^2 \to 2 \times \mathbb{Q}$.
By Remark \ref{remark: erdos-rado for 2k colors}, there are sequences $\langle a_i : i < \beta + 1 \rangle \subseteq (\beta^+)^{<\omega}$ and $\langle q_i : i < \beta \rangle \subseteq \mathbb Q$ such that for all $i < \beta$:
\[ (\forall j >i)\, q_i \Vdash \dot f(a_i) < \dot f(a_j) < \beta, \text{ or }
(\forall j >i) q_i \Vdash \beta > \dot f(a_i) > \dot f(a_j). \]
For each $i$, the second option is impossible by well-foundedness.
By the $\beta$-c.c., there is some $q$ forcing $\beta$-many of $q_i$ to be in the generic filter. Then $q$ forces that there is an increasing sequence of order type $\beta +1$ below $\beta$, which is impossible.
\end{proof}
\begin{corollary}\label{cor: small forcing force cc after half q2}
Work in the generic extension by $\mathbb{P}_\kappa$. Let $\alpha < \gamma_0 \leq \gamma_1 \leq \beta < \kappa$ be regular cardinals. Assume that
$\mathbb{Q}$ is a $\beta$-c.c.\ forcing notion of size $\leq \beta$, and $\mathbb{Q} \times \Col(\gamma_0,\gamma_1)$ preserves $\alpha^+$.
Then $\mathbb{Q} \times \Col(\gamma_0,\gamma_1)$ forces $(\beta^+, \beta)\twoheadrightarrow (\alpha^+, |\alpha|)$.
\end{corollary}
\begin{proof}
Let $\mathbb{P}_i$ be a $\beta^+$-c.c.\ regular subforcing of $\mathbb{P}_\kappa$ such that $\Vdash_{\mathbb{P}_i} \beta^+\twoheadrightarrow_{Q^{{<}{\omega}}} \alpha^+$ and $\mathbb{P}_\kappa / \mathbb{P}_i$ is $\beta^+$-closed.
By Claim \ref{claim: mm reflection in successor of regulars}, $\beta^+ \twoheadrightarrow_{Q^{<\omega}} \alpha^+$ holds in $V^{\mathbb P_i * \dot \Col(\gamma_0,\gamma_1)}$.
Lemma \ref{lemma: small forcing force cc after q2} implies that $\mathbb Q$ forces $(\beta^+, \beta)\twoheadrightarrow (\alpha^+, |\alpha|)$ over this model.
In $V^{\mathbb P_i}$, $\mathbb{P}_\kappa / \mathbb{P}_i$ is $\beta^+$-closed, and $| \mathbb Q \times \Col(\gamma_0,\gamma_1) | \leq \beta$. Therefore, Lemma \ref{preserve closed} implies that in the generic extension Chang's Conjecture, $(\beta^+, \beta)\twoheadrightarrow (\alpha^+, |\alpha|)$, holds.
\end{proof}
We are now ready for the main theorem. We start by defining a notion of Radin forcing with interleaved collapses. For simplicity, we assume ${\rm GCH\xspace}$.
Recall the following definition of a measure sequence, due to Radin.
\begin{definition}\cite{Radin1982}
Let $j\colon V\to M$ be an elementary embedding, $\crit j = \alpha$. Let us define, by induction, a sequence of normal measures on $V_\alpha$, $u$. Let $u(0) = \alpha$.
For every $i < j(\alpha)$, if $u\restriction i \in M$, let $u(i) = \{X \subseteq V_\alpha \colon u\restriction i \in j(X)\}$. Otherwise, we halt.
$u$ is called the \emph{measure sequence} derived from $j$.
\end{definition}
Let $j\colon V\to M$ be an elementary embedding with critical point $\kappa$. Let $\mathcal{U}$ be the measure sequence derived from $j$. Let $MS$ be the class of all measure sequences.
We say that a measure $u$ on $V_\alpha$ is \emph{normal} if it is closed under diagonal intersections in the following sense: if $\langle A_v \mid v \in MS \cap V_\alpha\rangle$ is a list of sets from $u$ then also:
\[\triangle_{v} A_v = \{x \in V_\alpha \cap MS \mid \forall v\in V_{x(0)} \cap MS,\, x \in A_v\}\]
Let us start with the following fact:
\begin{lemma}\label{lemma: guiding generics}
There is a sequence $\langle \mathcal{G}_\alpha \colon \alpha \leq \kappa\rangle$ such that:
\begin{enumerate}
\item For every measurable $\alpha$, and every $f\in \mathcal{G}_\alpha$, $f$ is a function with domain $V_{\alpha}$ and for every measure sequence $u \in \dom f$, $f(u) \in \Col(u(0)^+, \alpha)$.
\item Let $u$ be a normal measure on $V_\alpha$, $\alpha \leq \kappa$, and let $j_u\colon V\to M_u$ be the ultrapower embedding. Then the set $\{[f]_u \colon f\in \mathcal{G}_\alpha\}$ is an $M_u$-generic filter for $\Col(\alpha^{+}, j_u(\alpha))$. Moreover, for every function $D\colon V_\alpha \to V$, such that for every measure sequence $v\in V_\alpha$, $D(v)$ is a dense open subset of $\Col(v(0)^+, \alpha)$, there is $f\in \mathcal{G}_\alpha$ such that $\{v \in MS \cap V_\alpha \colon f(v)\in D(v)\}$ belongs to every normal ultrafilter on $V_\alpha$.
\end{enumerate}
\end{lemma}
\begin{proof}
Let $\delta < \kappa$. Using {\rm GCH\xspace}, enumerate all the functions $D : V_\delta \to V_{\delta+1}$, such that $D(u)$ is a dense open subset of $\Col(u(0)^+, \delta)$ for every measure sequence $u\in V_\delta$, as $\langle D_\alpha : \alpha < \delta^+ \rangle$. Let $p_0 : V_\delta \to V_\delta$ be such that $p_0(u) \in D_0(u)$ for every measure sequence $u \in V_\delta$. Given $\langle p_i : i \leq \alpha \rangle$, $\alpha < \delta^+$, let $p_{\alpha+1}$ be such that $p_{\alpha+1}(u) \leq p_\alpha(u)$ and $p_{\alpha+1}(u) \in D_{\alpha+1}(u)$ for all measure sequences $u$. At limit stages $\lambda$ in the construction, we use the following inductive assumption: For every $\alpha < \beta < \lambda$, there is a club $C_{\alpha,\beta} \subseteq \delta$ such that whenever $u(0) \in C_{\alpha,\beta}$ for a measure sequence $u$, $p_\alpha(u) \geq p_\beta(u)$. Let $\langle \lambda_\alpha : \alpha < \cf(\lambda) \rangle$ be increasing and cofinal in $\lambda$. The diagonal intersection, $C = \{ \alpha < \delta :$ for all $\beta < \gamma < \alpha$, $\alpha \in C_{\lambda_\beta,\lambda_\gamma} \}$, is club. For all $u$ such that $u(0) \in C$, $\langle p_{\lambda_\alpha}(u) : \alpha < u(0) \rangle$ is a decreasing sequence in $\Col(u(0)^+,\delta)$. Let $p_\lambda$ be such that $p_\lambda(u) \in D_\lambda(u)$ is a lower bound to this sequence for all such $u$. To continue the induction, we define $C_{\lambda_\alpha,\lambda} = C \cap \{ \beta \mid \alpha < \beta \}$ for $\alpha < \cf(\lambda)$. For $\beta < \lambda$ not among the $\lambda_\alpha$, let $C_{\beta,\lambda} = C_{\lambda_\alpha,\lambda} \cap C_{\beta,\lambda_\alpha}$, where $\alpha$ is the least ordinal such that $\lambda_\alpha > \beta$.
For every normal measure $u$ on $V_\delta$, $\{[p_\alpha]_u \colon \alpha < \delta^+ \}$ is a descending sequence in $\Col(\delta^+,j_u(\delta))$, and $[p_\alpha]_u \in [D_\alpha]_u$ for every $\alpha < \delta^+$. We let $\mathcal{G}_\delta = \{ p_\alpha : \alpha < \delta^+ \}$. Finally, we let $\mathcal{G}_\kappa = j(\langle \mathcal{G}_\alpha : \alpha < \kappa \rangle)(\kappa)$.
\end{proof}
\begin{lemma}\label{lemma: guiding generics reflection}
Under the same assumptions:
\begin{enumerate}
\item Assume that $\langle f_\alpha \colon \alpha < \kappa\rangle$ is a sequence of functions, $f_\alpha \in \mathcal{G}_\alpha$. Then there is a function $f \in \mathcal{G}_\kappa$ such that the collection: $\{v \in V_\kappa \colon f \restriction V_{v(0)} = f_{v(0)}\}$ is in $\bigcap_{0 < \beta < \len \mathcal{U}} \mathcal{U}(\beta)$.
\item Let $B$ be the set of all measure sequences $u \in V_\kappa$ such that for every $\langle f_\gamma \colon \gamma < u(0)\rangle$, with $f_\gamma \in \mathcal{G}_\gamma$ there is $f$ such that $\{v \in V_{u(0)} \colon f \restriction V_{v(0)} = f_{v(0)}\}$ is in $\bigcap_{0 < \beta < \len u} u(\beta)$. Then $B \in \bigcap_{0 < \beta < \len \mathcal{U}} \mathcal{U}(\beta)$.
\end{enumerate}
\end{lemma}
\begin{proof}
For (1), let $f = j(\langle f_\alpha \colon \alpha < \kappa\rangle) \in \mathcal{G}_\kappa$. Note that $j(f) \restriction V_\kappa = f$. Let $A = \{ \alpha : f \restriction V_\alpha = f_\alpha \}$. The set $A'$ of measure sequences $u \in V_\kappa$ such that $u(0) \in A$ is in $\mathcal U(\beta)$ for every $\beta < \len \mathcal U$, since $\kappa \in j(A)$.
Now let $B$ be as in (2). For all $\beta < \len \mathcal U$, $M$ can see that for all sequences $\langle f_\alpha \colon \alpha < \kappa\rangle$ as in (1), there is $f \in \mathcal{G}_\kappa$ such that $\{ v \in V_\kappa : f \restriction V_{v(0)} = f_{v(0)} \} \in \bigcap_{\alpha < \beta} \mathcal U(\alpha)$. Thus $\mathcal U \restriction \beta \in j(B)$, and $B \in \mathcal U(\beta)$.
\end{proof}
Let us define the forcing notion $\mathbb{P}$. $p\in \mathbb{P}$ iff
\[p = \langle f_{-1}, q_0, f_0, \dots, q_n\rangle\]
Where
\[q_i = \langle u_i, A_i, F_i\rangle\]
and:
\begin{enumerate}
\item $u_i$ is a measure sequence. We denote $u_i(0)$ by $\alpha_i$.
\item If $\len u_i > 0$, $A_i \in \bigcap_{0 < \beta < \len(u_i)} u_i(\beta)$. Otherwise $A_i = \emptyset$.
\item $F_i \colon A_i \to V$, and for all $\beta \in A_i$, $F_i(\beta) \in \Col(\beta^+, \alpha_i)$ and $F_i \in \mathcal{G}_{u_i(0)}$.
\item $u_n = \mathcal{U}$.
\item For every $i \geq 0$, $f_i \in \Col(\alpha_i^{+}, \alpha_{i+1})$.
\item $f_{-1} \in \Col(\omega, \alpha_0)$.
\item If $v\in A_i$ then $v(0) > \sup \range f_{i-1}$.
\end{enumerate}
Let
\[p = \langle f_{-1}, u_0, A_0, F_0, f_0, \dots, u_n, A_n, F_n\rangle \in \mathbb{P}\]
\[p^\prime = \langle f^\prime_{-1}, u^\prime_0, A^\prime_0, F^\prime_0, f^\prime_0, \dots, u^\prime_m, A^\prime_m, F^\prime_m\rangle \in \mathbb{P}\]
$p \leq p^\prime$ iff:
\begin{enumerate}
\item $m \leq n$ and there is a strictly increasing sequence of indices $i_0, \dots, i_m$ such that $u_{i_j} = u^\prime_j$. Let us set $i_{-1} = -1$.
\item For all $-1 \leq j \leq m$, $f_{i_j} \supseteq f^\prime_j$.
\item $A_{i_j} \subseteq A^\prime_j$.
\item For every $-1 \leq j < m$ and every $i_j < k < i_{j+1}$, $u_k \in A_{i_{j+1}}$, $f_k \supseteq F^\prime_{i_{j+1}}(u_k)$ and $A_k \subseteq A^\prime_{j+1}$.
\item For all $j$ and $i_j < k \leq i_{j+1}$, for all $\beta \in A_k$, $F_k(\beta) \supseteq F^\prime_{i_{j+1}} (\beta)$.
\end{enumerate}
We say that $p \leq^\star q$ if $p\leq q$ and $\len p = \len q$.
For a condition $p = \langle f_{-1}, u_0, A_0, F_0, f_0, \dots, u_n, A_n, F_n\rangle \in \mathbb{P}$, let us denote:
\[\stem p = \langle f_{-1}, u_0, A_0, F_0, f_0, \dots, u_n\rangle\]
\begin{lemma}\label{lemma: radin centered}
$\mathbb{P}$ is $\kappa$-centered.
\end{lemma}
For any measure sequence $u$ which is derived from some elementary embedding, let us denote by $\mathbb{P}^{u}$ the forcing notion which is defined as $\mathbb{P}$ when replacing $\mathcal{U}$ by $u$. Lemma \ref{lemma: radin centered} holds for $\mathbb{P}^u$. We have the standard decomposition:
\begin{claim}
For every condition $p\in \mathbb{P}$ of length $n$, and every measure sequence $u$ in the stem of $p$, the forcing $\mathbb{P} \restriction p$ of all conditions below $p$ splits into the product:
$\mathbb{P} \restriction p = \mathbb{P}^{>u} \restriction p_{\uparrow} \times \mathbb{P}^{u} \restriction p_{\downarrow}$, where $\mathbb{P}^{>u}$ is the forcing $\mathbb{P}$ when we modify the definition of $\alpha_{-1}$ to be $u(0)^+$. $p_{\uparrow}, p_{\downarrow}$ is the decomposition of the condition $p$ to the parts above and below $u$, respectively.
\end{claim}
\begin{lemma}
$\mathbb{P}$ satisfies the Prikry property. Moreover, this is true for $\mathbb{P}^{>u}$ for every measure sequence $u$.
\end{lemma}
\begin{proof}
We sketch a proof for the lemma. The proof is similar to the one in \cite[Section 3, Section 4]{Gitik2010}, with minor changes. We will prove it only for the case of $\mathbb{P}$. The other cases are similar.
Let $\Phi$ be a statement in the forcing language. Let $p\in \mathbb{P}$ be a condition.
In order to show that $\mathbb{P}$ satisfies the Prikry property, we will show that this is true for $\mathbb{P}^u$, for every measure sequence $u$.
Let us assume, by induction, that this is true for every measure sequence $u$ such that $u(0) < \kappa$.
Let us start with the case $\len p = 0$:
\begin{claim}
Assume that $p = \langle f_{-1}, \mathcal U, A, F\rangle$. There is a direct extension of $p$ that decides the truth value of $\Phi$.
\end{claim}
\begin{proof}
First, let us consider, for every stem $s$, the following sets:
\[D^0_{s, v} = \{g \leq F(v) \colon \exists A_v, F_v, A^\prime, F^\prime, s^\smallfrown \langle A_v, F_v, g, A^\prime, F^\prime\rangle \parallel \Phi\}\]
\[D^1_{s, v} = \{g \leq F(v) \colon \forall A_v, F_v, A^\prime, F^\prime, \forall g' \leq g,\, s^\smallfrown \langle A_v, F_v, g', A^\prime, F^\prime\rangle \not\parallel \Phi\}\]
Clearly, $D^0_{s, v} \cup D^1_{s,v}$ is a dense subset of $\Col(v(0)^{+},\kappa)$.
By the distributivity of $\Col(v(0)^{+},\kappa)$, the intersection $D_v = \bigcap_{s\in V_{v(0)}} (D^0_{s, v} \cup D^1_{s,v})$ is dense subset of $\Col(v(0)^{+},\kappa)$. By Lemma \ref{lemma: guiding generics}, there is $F'\in \mathcal{G}_\kappa$ such that the set of all $v\in V_\kappa$ with $F'(v)\in D_v$ belongs to $\bigcap_{0 < \alpha < \len \mathcal{U}} \mathcal{U}(\alpha)$. Let $A'$ be the intersection of the above set with $A$. Let $F^\star$, be a condition in $\mathcal{G}_\kappa$ stronger than $F, F'$.
Let us define for every possible stem of a condition stronger than $\langle f_{-1}, \mathcal{U}, A', F^\star\rangle$,
\[s = \langle f^s_{-1}, u_0^s, A_0^s, F_0^s, f_0^s, \dots, u_{k-1}^s, A_{k-1}^s, F_{k-1}^s, f_{k-1}^s \rangle,\]
and for every $\alpha < \len \mathcal{U}$, a set $A(s, \alpha)\in \mathcal{U}(\alpha)$. This is a measure one set, relative to $\mathcal{U}(\alpha)$, such that one of the three possibilities holds for it:
\begin{enumerate}
\item For every $v\in A(s,\alpha)$ there is a choice of $B^s_v, F^s_v, f^s_v$ such that an extension of $p$ with the stem $s^\smallfrown \langle v, B^s_v, F^s_v, f^s_v\rangle$ forces $\Phi$.
\item For every $v\in A(s,\alpha)$ there is a choice of $B^s_v, F^s_v,f^s_v$ such that an extension of $p$ with the stem $s^\smallfrown \langle v, B^s_v, F^s_v,f^s_v\rangle$ forces $\neg\Phi$.
\item For every $v\in A(s,\alpha)$, there is no extension of $p$ with stem $s ^\smallfrown \langle v, B_v, F_v,f_v\rangle$, that forces either $\Phi$ or $\neg\Phi$.
\end{enumerate}
Using the closure of the generic filter $\mathcal{G}_{v(0)}$, we may assume that $F^s_v, f^s_v$ depend only on $v$ (by taking the lower bound of all the $F^s_v, f^s_v$ with $s \in V_{v(0)}$ - there are only $v(0)$ many such stems).
Let $A(\alpha)$ be the diagonal intersection of all the $A(s, \alpha)$, and let $A^\star = A' \cap \bigcup_{\alpha < \len \mathcal{U}} A(\alpha)$. Let $p^\star = \langle f_{-1}, \mathcal U, A^\star, F^\star\rangle$.
Let us observe first that for every $v\in A(s, \alpha) \cap A'$, if one of the first two options holds, then we may take $f_v = F^\star(v)$. Recall that $F^\star(v) \in D^0_{s, v} \cup D^1_{s,v}$. $f_v \leq F^\star(v)$, and it decides the truth value of $\Phi$. Thus, $F^\star(v) \in D^0_{s, v}$. In particular, there are $B_v^\prime, F_v^\prime$ that together with $F^\star(v)$ decide the truth value of $\Phi$. By compatibility, a condition with stem $s^\smallfrown \langle v, B_v \cap B_v^\prime, F_v \wedge F^\prime_v, F^\star(v)\rangle$ must force the same truth value for $\Phi$ as a condition with stem $s^\smallfrown \langle v, B_v, F_v, f_v\rangle$.
Let us take an extension of $p^\star$ which decides $\Phi$ and has a minimal length. If it is a direct extension, we are done. Let us assume, towards a contradiction, that this extension has length $n + 1$:
\[r = \langle f_{-1}^r, v^r_0, A^r_0, F^r_0,f^r_0 \dots, \mathcal U, A^r_{n+1}, F^r_{n+1}\rangle\]
Let $s$ be the lower stem (up to length $n$). By our assumption, there is $\alpha$ such that $A(s, \alpha)\in \mathcal{U}(\alpha)$ contains only measure sequences $v$, which when appended to $s$ together with $A_v, F_v$, form a condition that decides the statement in the same direction as $r$. Without loss of generality, they all force $\Phi$.
For every $v \in A(s, \alpha)$, $F_v$ is stronger than the restriction of $F^\star$ to $A_v$ (pointwise) and belongs to $\mathcal{G}_{v(0)}$.
Let us consider the function $g\colon A(s, \alpha) \to V_\kappa$, $g(v) = \langle A_v, F_v\rangle$. By the definition of $\mathcal U(\alpha)$, $\mathcal{U}\restriction \alpha \in \dom j(g)$. Let $\langle A^{<\alpha}, F^{<\alpha}\rangle = j(g)(\mathcal{U}\restriction \alpha)$.
By elementarity, $A^{<\alpha} \in \bigcap_{\beta < \alpha} \mathcal{U}(\beta)$ and for $\mathcal{U}(\alpha)$-almost all $v \in V_\kappa$, $A^{<\alpha} \cap V_{v(0)} = A_v$ and $F^{<\alpha} \restriction V_{v(0)} = F_v$. Let $A^{\alpha}$ be the collection of all $v \in A(s,\alpha)$ that satisfy the above assertion. By Lemma \ref{lemma: guiding generics reflection}, $F^{<\alpha}\in \mathcal{G}_\kappa$, so let $F^{\star\star} = F^{<\alpha} \wedge F^\star$.
Let $A^{>\alpha}$ to be all the sets that reflect $A^{\alpha}$, namely $A^{>\alpha} = \{u\in A^\star \colon \exists \beta,\, A^{\alpha} \cap V_{u(0)} \in u(\beta)\}$. Now let $A^{\star\star} = A^{<\alpha} \cup A^\alpha \cup A^{>\alpha}$, and let us restrict the domain of $F^{\star\star}$ to $A^{\star\star}$.
Let us show that any extension of the condition $q_s = s ^\smallfrown \langle\mathcal{U}, A^{\star\star}, F^{\star\star}\rangle$ is compatible with a choice of an element from $A(s, \alpha)$. Therefore, any extension of the current condition is compatible with an extension that forces $\Phi$.
This is true by our choice of $A^{\star\star}$. If we extend $q_s$ by only adding elements below $v_n$ and strengthening the collapses, then this condition is compatible with any condition in which we extend $q_s$ by adding a single element from $A(s, \alpha)$ above $v_{n-1}$. Otherwise, let $q \leq q_s$ be any extension of $q_s$ and assume that the Radin club of $q$ contains elements above $v_{n-1}$. Let $m = \len q$,
\[q = \langle f_{-1}^q, u_0^q, A_0^q,F_0^q,f_0^q, \dots, u_m^q, A^q_m, F^q_m \rangle\]
and assume that $k < m$ is the first index of an element in the Radin sequence which
is a measure sequence of length $>0$ such that $(A^{\alpha} \cup A^{> \alpha}) \cap V_{u_k(0)} \in \bigcup_{\beta < \len u_k} u_k(\beta)$. If there is no such element, let us pick any nontrivial measure sequence $v \in A^\alpha \cap A^q_m$. Then $A^{<\alpha} \cap A^q_i \in \bigcap_{\beta < \len u_i} u_i(\beta)$ for all $i < m$, and $A_v = A^{<\alpha} \cap V_{v(0)}$. Thus adding $\langle v,A_v,F^{\star\star} \restriction A_v \rangle$ to $q_s$ results in a condition that forces $\Phi$ and is compatible with $q$.
So, let us assume that there is such element $u_k^q$. If $A^{\alpha} \in u_k^q(\beta)$ for some $\beta < \len u_k^q$, then there is $v \in A^q_k$ such that $A_v \cap A^q_k \in \bigcap_{\beta < \len v} v(\beta)$, so as above, $q_s$ may be extended by $\langle v,A_v,F^{\star\star} \restriction A_v \rangle$ to get a condition compatible with $q$ that forces $\Phi$.
If $A^{>\alpha} \in u_k^q(\beta)$ for some $\beta < \len u_k^q$, then by our choice of $A^{>\alpha}$ there is some $v \in A^q_k$ that can be added to $q$ to put us into the previous case.
We conclude that in any case, any extension of $q_s$ has an extension that forces $\Phi$ and thus, $q_s \Vdash \Phi$. But this is a contradiction to the minimality of $n$.
\end{proof}
Let us continue to the general case.
Let $p\in \mathbb{P}$ be a general condition. Let us assume, by induction, that Prikry property holds for every shorter condition. By the claim above, we may assume that $\len p > 0$.
We want to find a direct extension of $p$ that decides the truth value of statement $\Phi$.
We can decompose the forcing notion $\mathbb{P} \restriction p$ into a product $\mathbb{P}^{>u}\restriction p_\uparrow \times \mathbb{P}^u \restriction p_\downarrow$ for some measure sequence $u$ that appears in $p$.
Recall that $\mathbb{P}^u$ is $\alpha$-centered, where $\alpha = u(0)$. Let $\langle r_i \colon i < \alpha\rangle$ enumerate all possible stems of conditions in $\mathbb{P}^u$.
Let us define, by induction, a sequence of conditions in $\mathbb{P}^{>u}$, $\langle p_i \colon i \leq \alpha\rangle$ in the following way. Let $p_0 = p_\uparrow$. Given $p_i$, let $p_{i+1} \leq^\star p_i$ decide whether there is a condition in $\mathbb P^u$ with stem $r_i$ deciding $\Phi$, and if so, whether it forces $\Phi$ or $\neg \Phi$. At limit ordinals $i \leq \alpha$, we use the closure of the order $\leq^\star$ and take $p_i$ to be a lower bound of $p_{i'}$ for every $i' < i$.
If $G \times H \subseteq \mathbb P^u \times \mathbb{P}^{>u}$ is generic with $\langle p_\downarrow,p_\alpha \rangle$, then there is a condition $\langle r,q \rangle \in G \times H$ deciding $\Phi$. The stem of $r$ is $r_i$ for some $i < \alpha$, and $p_\alpha$ must already decide which way $r$ decides $\Phi$. Thus it is forced by $\mathbb P^u$ that $p_\alpha$ decides $\Phi$. By induction, we may take a direct extension $r' \leq^\star p_\downarrow$ which decides which way $p_\alpha$ decides $\Phi$. $r^{\prime \smallfrown} p_\alpha$ is the desired direct extension of $p$.
\end{proof}
Recall that $\mathcal{U}$ was derived from an elementary embedding $j\colon V\to M$.
\begin{lemma}
Assume that $V_{\kappa + 2} \subseteq M$. Then $\mathbb{P}$ preserves the inaccessibility of $\kappa$.
\end{lemma}
\begin{proof}
By ${\rm GCH\xspace}$ and the strength of the elementary embedding, $j$, we have $\len \, \mathcal{U} \geq \kappa^{++}$, while $|V_{\kappa + 1}| = \kappa^{+}$. Therefore, there is a \emph{repeat point}, namely an ordinal $\alpha < \kappa^{++}$ such that \[\bigcap_{0 < \beta < \alpha} \mathcal{U}(\beta) = \bigcap_{0 < \beta < \kappa^{++}} \mathcal{U}(\beta).\] Let $\alpha$ be the first repeat point in the measure sequence derived from $j$.
Clearly, replacing $\mathcal{U}$ with $\mathcal{U}\restriction \alpha$ does not change the forcing. Let $k\colon V \to M$ be an elementary embedding generated by $\mathcal{U}(\alpha)$. Let us look at $k(\mathbb{P})$. Let:
\[p = \langle f_{-1}, u_0, A_0, F_0, f_0, \dots, u_n, A_n, F_n\rangle \in \mathbb{P}.\]
Let us extend the condition $k(p)$ by adding $\langle \mathcal{U}\restriction \alpha, A_n, F_n\rangle$ at the $n$-th coordinate, and let $q$ be the obtained condition. It is clear that forcing below $q$ introduces a generic filter for $\mathbb{P}$, $K$. The Radin forcing below the condition $q$ is equivalent to a product $\mathbb{P} \times \mathbb{Q}$ (where $\mathbb{Q}$ consists of all the upper parts of the conditions in $\mathbb{P}$). Recall that $\mathbb{Q}$ is $\kappa^{+}$-weakly closed. Applying $k$ to the upper part of the conditions in $\mathbb{P}$ generates an $M$-generic filter for $\mathbb{Q}^\prime$, since any dense open set in $M$ is represented by a function from $\kappa$ to dense open sets of the closed part of $\mathbb{P}$. Therefore, there is a condition in the intersection of all of them. Let $H$ be a $\mathbb{Q}$-generic filter, extending the $\mathbb{Q}^\prime$-generic filter which is generated by the $k$ images of the elements of $\mathbb{P}$.
Let $K \times H$ be the $M$-generic filter for $k(\mathbb{P})$. Silver's criteria holds, and therefore one can extend $k$ to an elementary embedding $\tilde{k}\colon V[K] \to M[K][H]$. In particular, since $\kappa$ is the critical point of $\tilde{k}$, it is regular in $V[K]$.
\end{proof}
In fact, the forcing $\mathbb{P}$ preserves also the measurability of $\kappa$, but for our purposes it is enough to know that $V_\kappa$ is a model of ${\rm ZFC\xspace}$.
There is a natural projection from a measure on measure sequences in $V_\kappa$ to a measure on $\kappa$ by taking each measure sequence $u$ to its first element $u(0)$. When saying that a subset of $\kappa$ is large relative to a measure on the measure sequences of $V_\kappa$ we mean that it is large relative to the corresponding projection.
Let us return now to the model that was obtained in the previous section.
\begin{theorem}
Let $\mathbb{P}$ be the Radin forcing for adding a club through $\kappa$, with interleaved collapses, collapsing $\rho_{i+1}$ to be of cardinality $\rho_i^+$ for any two successive Radin points. Let $A$ be the set obtained in Lemma \ref{below huge}. Assume that $A$ is $\mathcal{U}$-large relative to all relevant measures.
Then $\mathbb{P}$ forces $(\beta^{++}, \beta^+)\twoheadrightarrow (\alpha^+, \alpha)$ for all $\alpha \leq \beta < \kappa$.
\end{theorem}
\begin{proof}
Let $p \in \mathbb P$ force that $\beta$ be a cardinal in the extension. Assume $p$ is strong enough to decide three successive points $\zeta < \xi < \rho$ in the Radin club such that $p \Vdash \beta^+ = (\zeta^+)^V,\, \beta^{++} = (\xi^+)^V$, and $\beta^{+3} = (\rho^+)^V$.
The forcing $\mathbb{P}\restriction p$ splits into a product $\mathbb{Q}_0 \times \Col(\xi^+,\rho) \times \mathbb Q_1$, where $\mathbb{Q}_0$ is the Radin forcing below $\xi$, and $\mathbb Q_1$ adds no subsets of $\xi^+$.
Note that $\mathbb{Q}_0 = \mathbb{P}_0 \times \Col(\zeta^+, \xi)$, where $\mathbb P_0$ is $\zeta^+$-c.c., has size $\leq \zeta^+$, and preserves $\alpha^+$, which is the successor cardinal in $V$ to some member of the Radin club. Corollary \ref{cor: small forcing force cc after half q2} implies that $(\beta^{++},\beta^+) \twoheadrightarrow (\alpha^+,\alpha)$ holds after forcing with $\mathbb{P}_0 \times \Col(\zeta^+, \xi)$.
Since $\mathbb Q_0$ is $\xi^+$-c.c.\ and $\Col(\xi^+,\rho)$ is $\xi^+$-closed, Lemma \ref{preserve closed} implies that the instance of Chang's Conjecture holds after forcing with $\mathbb{Q}_0 \times \Col(\xi^+, \rho)$. It continues to hold after forcing with $\mathbb Q_1$, since no new algebras on $\xi^+$ are added.
\end{proof}
In the above model, the instances of Chang's Conjecture of the form $(\mu^+, \mu)\twoheadrightarrow (\nu^+, \nu)$ where $\mu$ is singular, always fail. Since every singular cardinal in the generic extension is inaccessible in the ground model, ${\Square}_\mu^\star$ holds there. Since we preserve its successor, it still holds in the generic extension. Any instance of Chang's Conjecture of the form $(\mu^+, \mu)\twoheadrightarrow (\nu^+, \nu)$, where $\mu$ is singular, implies the failure of the weak square, $\Square_\mu^\star$ \cite{ForemanMagidor1997}.
\section{Segments of Chang's Conjecture}\label{sec: segment cc}
In the previous section we dealt with obtaining Chang's Conjecture between all pairs of the form $(\mu^+, \mu)$ and $(\nu^+, \nu)$ where $\mu$ is a successor cardinal. The cases of $\mu$ singular cardinal, which were not covered in the previous section, are much harder.
Recall that any instance of Chang's Conjecture of the form $(\mu^+,\mu)\twoheadrightarrow(\nu^+,\nu)$ where $\mu$ is singular, $\nu < \mu$ (in which we assume that the elementary submodel of cardinality $\nu^+$ contains $\nu$) implies the failure of the weak square $\square_{\mu}^\star$. Indeed, it implies that there are no good scales. Thus, the problem of getting, for example $(\mu^+,\mu)\twoheadrightarrow((\cf \mu)^+, \cf \mu)$ globally requires us to get a failure of weak square at all singular cardinals. See \cite{DiagonalSCPrikry} for the best known consistency result towards this goal.
We want to attack a more modest problem. We will get all the instances of Chang's Conjecture which are compatible with {\rm GCH\xspace}, in a small segment of cardinals not covered by the previous section.
Let $\kappa$ be a huge cardinal, and let $j\colon V\to M$ be an elementary embedding witnessing it. Let $\delta = j(\kappa)$.
\begin{lemma}\label{lem: universal laver function}
There is a function $\ell\colon \delta \to V_\delta$ with the following properties:
\begin{enumerate}
\item $j(\ell\restriction \kappa) = \ell$.
\item For every cardinal $\mu$, and $\gamma < \delta$, if $x\in V_\gamma$ and $\mu$ is $2^\gamma$-supercompact then there is an elementary embedding $k\colon V\to N$, $\crit k = \mu$, $k\image \gamma \in N$, $k(\ell)(\mu) = x$.
\item For every $\mu < \kappa$ non-measurable, $\ell(\mu) = \emptyset$.
\item For every $\mu < \kappa$, if $\mu$ is ${<}\delta$-supercomapct, $\ell(\mu) = \emptyset$.
\end{enumerate}
\end{lemma}
\begin{proof}
We pick a universal Laver function $\ell\restriction \kappa$ on $V_\kappa$, using minimal counter examples (if there are counter examples). I.e.\ we pick $\ell(\rho) = x$ if $x\in H(\theta)$ and $\theta < \delta$ is the minimal cardinal for which there is no supercompact measure on $P_\rho \theta$, $\mathcal{U}$ such that $j_{\mathcal{U}}(\ell\restriction \rho) = x$. We pick $x$ to be the minimal such witness of failure, relative to some fixed well order.
Apply $j$ on this function to obtain $\ell$. Since $V_\delta \subseteq M$, the first item holds. The second item holds below $\kappa$ by the usual proof for the Laver diamond: the minimal $x$ for which property (2) fails would be exactly the value of $j(\ell)$ at the point $\mu$ in any closed enough ultrapower. The third item holds by definition. The fourth item holds as for any fully supercompact cardinal (from the point of view of $V_\delta$), $\mu$, $\ell\restriction \mu$ is already a Laver diamond and therefore there is no counter example to pick.
\end{proof}
Let $\ell$ be a Laver function for $V_\delta$ as in the conclusion of Lemma \ref{lem: universal laver function}.
We wish to make every $\alpha< \kappa$ which is ${<}\kappa$-supercompact indestructible under any $\alpha$-directed-closed forcing. We do the usual Laver iteration $\mathbb P$ with respect to $\ell$. We claim that if $G \subseteq j(\mathbb P)$ is generic, then $\kappa$ is still huge in $V[G]$. Since $\kappa$ is ${<}\delta$-supercompact in $M$, $j(\mathbb P)/(G \cap \mathbb P)$ is $(2^\kappa)^+$-directed-closed in $V[G\cap \mathbb P]$. Thus we may take a master condition and build an $M[G]$-generic filter for $j^2(\mathbb P)/G$.
Now let $A$ be the set obtained from Lemma \ref{below huge}. By reflection arguments, we may assume every cardinal in $A$ is ${<}\kappa$-supercompact. Let us pick such cardinals $\langle \mu_n \mid n < \omega\rangle$.
\begin{theorem}
There is $\rho < \mu_0$ such that:
\[\Col(\omega, \rho^{+\omega}) \ast \E(\rho^{+\omega+1}, \mu_0) \ast \E(\mu_0, \mu_1) \ast \cdots \ast \E(\mu_n, \mu_{n+1})\ast \cdots\]
forces:
\begin{enumerate}
\item For every $m < n < \omega$, $(\aleph_{n+1}, \aleph_n) \twoheadrightarrow (\aleph_{m+1}, \aleph_m)$.
\item $(\aleph_{\omega+1}, \aleph_{\omega}) \twoheadrightarrow (\aleph_1, \aleph_0)$.
\end{enumerate}
\end{theorem}
In the generic extension $\aleph_1 = \rho^{+\omega + 1}$ and $\aleph_{n + 2} = \mu_n$ for all $n <\omega$.
We split the proof of this theorem into three parts:
\begin{lemma}
For every choice of $\rho$, $(\aleph_{2}, \aleph_1)\twoheadrightarrow (\aleph_1, \aleph_0)$.
\end{lemma}
\begin{proof}
After forcing with $\Col(\omega, \rho^{+\omega})$, $\mu_0$ is still measurable. Since the forcing $\E(\omega_1, \mu_0)$ is $\mu_0$-c.c.\ and $\sigma$-closed, after forcing with it there is an $\omega_2$-complete ideal on $\omega_2$, in which the positive sets have a dense subset which is $\sigma$-closed. In particular, the Strong Chang conjecture holds (see, for example, \cite{Sakai2005}[Theorem 1.1]).
\end{proof}
\begin{lemma}
For every choice of $\rho$, for every $0 < m < n < \omega$, $(\aleph_{n + 1}, \aleph_n)\twoheadrightarrow (\aleph_{m + 1}, \aleph_m)$.
\end{lemma}
\begin{proof}
This is a special case of Lemma \ref{below huge}.
\end{proof}
\begin{lemma}
There is a choice of $\rho$ for which $(\aleph_{\omega+1}, \aleph_{\omega}) \twoheadrightarrow (\aleph_1, \aleph_0)$.
\end{lemma}
\begin{proof}
Let us show first that there is $\rho$ such that after forcing with:
\[(\Col(\omega, \rho^{+\omega}) \ast \E(\rho^{+\omega+1}, \mu_0)) \times (\E(\mu_0, \mu_1) \times \cdots \times \E(\mu_n, \mu_{n+1})\times \cdots) \]
Chang's conjecture $(\aleph_{\omega+1}, \aleph_{\omega}) \twoheadrightarrow (\aleph_1, \aleph_0)$, holds.
Replace the order of the product and force with the second component first. By the indestructibility of $\mu_0$, $\mu_0$ remains ${<}\kappa$-supercompact after this forcing.
Assume that for every $\rho < \mu_0$ there is a name for a function $\dot{f}_\rho \colon (\mu_0^{+\omega+1})^{<\omega} \to \mu_0^{+\omega}$ witnessing the failure of Chang's conjecture in the generic extension.
Let $k\colon V \to M$ be an elementary embedding with critical point $\mu_0$ and $A = k\image \mu_0^{+\omega+1} \in M$. Let $\dot f = k(\langle \dot f_\rho : \rho < \mu_0 \rangle)(\mu_0)$. By our assumption, $\Vdash |\dot{f} \image A^{<\omega}| = \aleph_1$. Since the image of $\dot{f}$ is contained in $k(\mu_0^{+\omega})$, there is an integer $n >0$ and a condition $p \in \Col(\omega, \mu_0^{+\omega}) \ast \E(\mu_0^{+\omega+1}, k(\mu_0))$ such that $p \Vdash |\dot{f} \image A^{<\omega} \cap k(\mu_0^{+n})| = \aleph_1$.
Let us find a sequence of decreasing conditions $p_\alpha$ below $p$ and a sequence of sets $a_\alpha \in (\mu_0^{+\omega+1})^{{<}\omega}$ such that $p_\beta \Vdash \dot f(k(a_\alpha)) < \dot f(k(a_\beta)) < k(\mu_0^{+n})$ for every $\alpha < \beta < \mu_0^{+\omega+1}$. We find this sequence in the same way as we did in Theorem \ref{thm: cc at single point}. Namely, let $\langle \dot{x}_i \mid i < \mu_0^{+\omega+1}\rangle$ be a sequence of names for elements in $A^{<\omega}$ such that $p \Vdash \dot{f}(\dot{x}_\alpha) < \dot{f}(\dot{x}_\beta) < k(\mu_0^{+n})$ for every $\alpha < \beta$. Let us pick for every $\alpha < \mu_0^{+\omega + 1}$ a condition $p_\alpha = \langle r_\alpha, \dot{q}_\alpha\rangle$ below $p$ such that:
\begin{enumerate}
\item For some $a_i \in (\mu_0^{+\omega+1})^{{<}\omega}$, $p_\alpha \Vdash \dot{x}_\alpha = k(\check a_\alpha)$.
\item For every $\beta < \alpha$, $r_0 \Vdash \dot{q}_\alpha \leq \dot{q}_\beta$.
\end{enumerate}
Given the partial sequence $\langle r_\alpha, \dot q_\alpha, a_\alpha : \alpha<\beta \rangle$ satisfying the above conditions, we let $\dot q$ be a name for a lower bound to $\langle \dot q_\alpha : \alpha < \beta \rangle$. Then we pick $\langle r_\beta, \dot q' \rangle \leq \langle r_0, \dot q \rangle$ and $a_\beta$ such that $\langle r_\beta, \dot q' \rangle \Vdash \dot x_\beta = k(\check a_\beta)$. Then let $\dot q_\beta$ be such that $r_\beta \Vdash \dot q_\beta = \dot q'$ and $r' \Vdash \dot q_\beta = \dot q$ for all $r' \perp r_\beta$.
By the regularity of $\mu_0^{+\omega + 1}$, there is a fixed condition $r_\star$ and a cofinal set of ordinals $\alpha < \mu_0^{+\omega+1}$ such that $r_\alpha = r_\star$. Without loss of generality, for every $\alpha$, $r_\alpha = r_\star$.
By elementarity, for every $\alpha < \beta < \mu_0^{+\omega + 1}$ there is $\rho < \mu_0$ and a condition $q \in \Col(\omega, \rho^{+\omega}) * \mathbb E(\rho^{+\omega+1}, \mu_0)$ that forces $\dot f_\rho(a_\alpha) < \dot f_\rho(a_\beta) < \mu_0^{+n}$. Applying the Erd\H{o}s-Rado theorem on the first $\mu_0^{+n + 1}$ elements in this sequence, we obtain a sequence of ordinals $I$ of order type $\mu_0^{+n} + 1$ and a single $\rho_\star < \mu_0$, $q_\star \in \Col(\omega, \rho_\star^{+\omega}) * \mathbb E(\rho_\star^{+\omega+1}, \mu_0)$ such that for every $\alpha < \beta$ in $I$, $q_\star \Vdash \dot f_{\rho_\star}(a_\alpha) < \dot f_{\rho_\star}(a_\beta) < \mu_0^{+n}$. This is a contradiction, since it is impossible to get an increasing sequence of ordinals of length $\mu_0^{+n}+1$ below $\mu_0^{+n}$.
Thus, there is $\rho < \mu_0$ such that the product forces $(\aleph_{\omega+1}, \aleph_{\omega}) \twoheadrightarrow (\aleph_1, \aleph_0)$. Let us show that in this case the iteration does the same.
\begin{claim}
There is a projection
\begin{align*}
\pi & : (\Col(\omega, \rho^{+\omega}) \ast \E(\rho^{+\omega+1}, \mu_0)) \times \E(\mu_0, \mu_1) \times \cdots \times \E(\mu_n, \mu_{n+1})\times \cdots \\
& \to \Col(\omega, \rho^{+\omega}) \ast \E(\rho^{+\omega+1}, \mu_0) \ast \E(\mu_0, \mu_1) \ast \cdots \ast \E(\mu_n, \mu_{n+1})\ast \cdots
\end{align*}
\end{claim}
\begin{proof}
Let $\mathbb P = \Col(\omega, \rho^{+\omega}) \ast \E(\rho^{+\omega+1}, \mu_0)$. The argument for Lemma \ref{easton projection} shows the following: For each $n < \omega$, there is a map $$\sigma_n : \E(\mu_{n},\mu_{n+1}) \to T(\mathbb P \ast \E(\mu_0, \mu_1) \ast \cdots \ast \E(\mu_{n-1}, \mu_{n}), \E(\mu_{n},\mu_{n+1}))$$ such that $\langle p,q \rangle \mapsto \langle p, \sigma_n(q) \rangle$ is a projection from $(\mathbb P \ast \E(\mu_0, \mu_1) \ast \cdots \ast \E(\mu_{n-1}, \mu_{n})) \times \E(\mu_n,\mu_{n+1})$ to $(\mathbb P \ast \E(\mu_0, \mu_1) \ast \cdots \ast \E(\mu_{n-1}, \mu_{n}) \ast \E(\mu_{n}, \mu_{n+1})$. Furthermore, if $p \Vdash \dot q_1 \leq \sigma_n(\check q_0)$, then there is $q_2 \leq q_0$ such that
$p \Vdash \sigma_n(\check q_2) = \dot q_1$.
For a condition $r = \langle p,q_0,q_1,\dots \rangle$ in the infinite product we define $\pi(r) = \langle p, \sigma_0(q_0),\sigma_1(q_1),\dots \rangle$. To verify that $\pi$ is a projection, suppose $\langle p',q'_0,q'_1,\dots \rangle \leq \langle p, \sigma_0(q_0),\sigma_1(q_1),\dots \rangle.$ For each $n$, there is $q''_n \leq q_n$ such that $\langle p',q'_0,\dots,q'_{n-1} \rangle \Vdash \sigma_n(q''_n) = q'_n$. An easy induction argument shows that $\langle p', \sigma_0(q''_0),\sigma_1(q''_1),\dots \rangle \leq \langle p',q'_0,q'_1,\dots \rangle$.
\end{proof}
As the product is $\mu_0$-closed in the ground model, it is $\mu_0$-distributive in the generic extension by the $\mu_0$-c.c.\ forcing $\Col(\omega, \rho^{+\omega}) \ast \E(\rho^{+\omega+1}, \mu_0)$. Therefore, if $f : \aleph_{\omega+1}^{{<}\omega} \to \aleph_\omega$ is in the extension by the iteration, the $\aleph_1$-sized witness for Chang's conjecture with respect to $f$ already exists in the extension by the iteration.
\end{proof}
\section{Open Questions}
We conclude this paper with a list of open questions:
\begin{question} Is it consistent, relative to large cardinals, that for every pair of cardinals $\kappa < \lambda$ such that $\kappa < \cf \lambda$ or $\cf \kappa = \cf \lambda$, \[(\lambda^+, \lambda)\twoheadrightarrow (\kappa^+,\kappa)?\]
\end{question}
In the model of Section~\ref{sec: global cc} we gave a positive answer to this question when restricting $\lambda$ to be a successor cardinal.
\begin{question}
What is the consistency strength of $(\aleph_4, \aleph_3) \twoheadrightarrow (\aleph_2, \aleph_1)$?
\end{question}
In Section~\ref{sec: local cc} we gave an upper bound of $(+2)$-subcompact cardinal. The known lower bound, due to Levinski is $0^\dagger$, \cite{Levinski1984}.
\providecommand{\bysame}{\leavevmode\hbox to3em{\hrulefill}\thinspace}
\providecommand{\MR}{\relax\ifhmode\unskip\space\fi MR }
\providecommand{\MRhref}[2]{%
\href{http://www.ams.org/mathscinet-getitem?mr=#1}{#2}
}
\providecommand{\href}[2]{#2}
| d0052cf9ffad899326dd92b26d4d96a2dbc12445 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Recent years have witnessed a remarkable convergence of two broad trends. The first of these concerns information i.e.\ data -- rapid technological advances coupled with an increased presence of computing in nearly every aspect of daily life, have for the first time made it possible to acquire and store massive amounts of highly diverse types of information. Concurrently and in no small part propelled by the environment just described, research in artificial intelligence -- in machine learning \cite{Aran2012g,Aran2012h,Aran2015,Aran2015d}, data mining~\cite{BeykAranPhunVenk+2014}, and pattern recognition, in particular -- has reached a sufficient level of methodological sophistication and maturity to process and analyse the collected data, with the aim of extracting novel and useful knowledge~\cite{Aran2015c,BeykAranPhunVenk+2014}. Though it is undeniably wise to refrain from overly ambitious predictions regarding the type of knowledge which may be discovered in this manner, at the very least it is true that few domains of application of the aforesaid techniques hold as much promise and potential as that of medicine and health in general.
Large amounts of highly heterogeneous data types are pervasive in medicine. Usually the concept of so-called ``big data'' in medicine is associated with the analysis of Electronic Health Records \cite{ChriElli2016,Aran2015g,Aran2016,VasiAran2016,VasiAran2016a}, large scale sociodemographic surveys of death causes \cite{RGI2009}, social media mining for health related data~\cite{BeykAranPhunVenk+2015} etc. Much less discussed and yet arguably no less important realm where the amount of information presents a challenge to the medical field is the medical literature corpus itself. Namely, considering the overarching and global importance of health (to say nothing of practical considerations such as the availability of funding), it is not surprising to observe that the amount of published medical research is immense and its growth is only continuing to accelerate. This presents a clear challenge to a researcher. Even restricted to a specified field of research, the amount of published data and findings makes it impossible for a human to survey the entirety of relevant publications exhaustively which inherently leads to the question as to what kind of important information or insight may go unnoticed or insufficiently appreciated. The premise of the present work is that advanced machine learning techniques can be used to assist a human in the analysis of this data. Specifically, we introduce a novel methodology based on Bayesian non-parametric inference that achieves this, as well as free software which researchers can use in the analysis of their corpora of interest.
\subsubsection{Previous work}
A limitation of most models described in the existing literature lies in their assumption that the data corpus is static. Here the term `static' is used to describe the lack of any associated temporal information associated with the documents in a corpus -- the documents are said to be exchangeable~\cite{BleiLaff2006a}. However, research articles are added to the literature corpus in a temporal manner and their ordering has significance. Consequently the topic structure of the corpus changes over time~\cite{Dyso2012,BeykPhunAranVenk2015,BeykAranPhunVenk2015a}: new ideas emerge, old ideas are refined, novel discoveries result in multiple ideas being related to one another thereby forming more complex concepts or a single idea multifurcating into different `sub-ideas' etc. The premise in the present work is that documents are not exchangeable at large temporal scales but can be considered to be at short time scales, thus allowing the corpus to be treated as \emph{temporally locally static}.
\section{Proposed approach\label{s:proposed}}
In this section we introduce our main technical contributions. We begin by reviewing the relevant theory underlying Bayesian mixture models, and then explain how the proposed framework employs these for the extraction of information from temporally varying document corpora.
\subsection{Bayesian mixture models}\label{ss:mixModels}
Mixture models are appropriate choices for the modelling of so-called heterogeneous data whereby heterogeneity is taken to mean that observable data is generated by more than one process (source). The key challenges lie in the lack of observability of the correspondence between specific data points and their sources, and the lack of \emph{a priori} information on the number of sources~\cite{RichGree1997}.
Bayesian non-parametric methods place priors on the infinite-dimensional space of probability distributions and provide an elegant solution to the aforementioned modelling problems. Dirichlet Process~(DP) in particular allows for the model to accommodate a potentially infinite number of mixture components~\cite{Ferg1973}:
\begin{align}
p\left(x|\pi_{1:\infty},\phi_{1:\infty}\right)=\sum_{k=1}^{\infty}\pi_{k}f\left(x|\phi_{k}\right).
\end{align}
where $\text{DP}\left(\gamma,H\right)$ is defined as a distribution of
a random probability measure $G$ over a measurable space $\left(\Theta,\mathcal{B}\right)$,
such that for any finite measurable partition $\left(A_{1},A_{2},\ldots,A_{r}\right)$
of $\Theta$ the random vector $\left(G\left(A_{1}\right),\ldots,G\left(A_{r}\right)\right)$
is a Dirichlet distribution with parameters $\left(\gamma H\left(A_{1}\right),\ldots,\gamma H\left(A_{r}\right)\right)$.
Owing to the discrete nature and infinite dimensionality of its draws, the DP is a useful prior for Bayesian mixture models. By associating different mixture components with atoms $\phi_{k}$, and assuming $x_{i}|\phi_{k}\overset{iid}{\sim}f\left(x_{i}|\phi_{k}\right)$ where $f\left(.\right)$ is the kernel of the mixing components, a Dirichlet process mixture model (DPM) is obtained~\cite{Radf2000}.
\subsubsection{Hierarchical DPMs}
While the DPM is suitable for the clustering of exchangeable data in a single group, many real-world problems are more appropriately modelled as comprising multiple groups of exchangeable data. In such cases it is desirable to model the observations of different groups jointly, allowing them to share their generative clusters. This so-called ``sharing of statistical strength'' emerges naturally when a hierarchical structure is implemented.
The DPM models each group of documents in a collection using an infinite number of topics. However, it is desired for multiple group-level DPMs to share their clusters. The hierarchical DP (HDP)~\cite{TehJordBealBlei2006} offers a solution whereby base measures of group-level DPs are drawn from a corpus-level DP. In this way the atoms of the corpus-level DP are shared across the documents; posterior inference is readily achieved using Gibbs sampling~\cite{TehJordBealBlei2006}.
\subsection{Modelling topic evolution over time\label{ss:contrib}}
We now show how the described HDP based model can be applied to the analysis of temporal topic changes in a \emph{longitudinal} data corpus.
Owing to the aforementioned assumption of a temporally locally static corpus we begin by discretizing time and dividing the corpus into epochs. Each epoch spans a certain contiguous time period and has associated with it all documents with timestamps within this period. Each epoch is then modelled separately using a HDP, with models corresponding to different epochs sharing their hyperparameters and the corpus-level base measure. Hence if $n$ is the number of epochs, we obtain $n$ sets of topics
$\boldsymbol{\phi}=\left\{ \boldsymbol{\phi}_{t_{1}},\ldots,\boldsymbol{\phi}_{t_{n}}\right\} $
where $\boldsymbol{\phi}_{t}=\left\{ \theta_{1,t},\ldots,\phi_{K_{t},t}\right\} $
is the set of topics that describe epoch $t$, and $K_{t}$ their number.
\subsubsection{Topic relatedness\label{ss:similarity}}
Our goal now is to track changes in the topical structure of a data corpus over time. The simplest changes of interest include the emergence of new topics, and the disappearance of others. More subtly, we are also interested in how a specific topic changes, that is, how it evolves over time in terms of the contributions of different words it comprises. Lastly, our aim is to be able to extract and model complex structural changes of the underlying topic content which result from the interaction of topics. Specifically, topics, which can be thought of as collections of memes, can merge to form new topics or indeed split into more nuanced memetic collections. This information can provide valuable insight into the refinement of ideas and findings in the scientific community, effected by new research and accumulating evidence.
The key idea behind our tracking of simple topic evolution stems from the observation that while topics may change significantly over time, changes between successive epochs are limited. Therefore we infer the continuity of a topic in one epoch by relating it to all topics in the immediately subsequent epoch which are sufficiently similar to it under a suitable similarity measure -- we adopt the well known Bhattacharyya distance (BHD). This can be seen to lead naturally to a similarity graph representation whose nodes correspond to topics and whose edges link those topics in two epochs which are related. Formally, the weight of the directed edge that links $\phi_{j,t}$, the $j$-th topic in epoch $t$, and $\phi_{k,t+1}$ is $\rho_\text{BHD}\left(\phi_{j,t},\phi_{k,t+1}\right)$ where $\rho_\text{BHD}$ denotes the BHD.
In constructing a similarity graph a threshold to used to eliminate automatically weak edges, retaining only the connections between sufficiently similar topics in adjacent epochs. Then the disappearance of a particular topic, the emergence of new topics, and gradual topic evolution can be determined from the structure of the graph. In particular if a node does not have any edges incident to it, the corresponding topic is taken as having emerged in the associated epoch. Similarly if no edges originate from a node, the corresponding topic is taken to vanish in the associated epoch. Lastly when exactly one edge originates from a node in one epoch and it is the only edge incident to a node in the following epoch, the topic is understood as having evolved in the sense that its memetic content may have changed.
\begin{figure*}[t]
\centering
\subfigure[Topic speciation]{\includegraphics[width=0.8\columnwidth]{speciation.pdf}\label{f:speciation}}~~~~~~~~~~~~~~~~~~~~~~~
\subfigure[Topic splitting]{\includegraphics[width=0.8\columnwidth]{splitting.pdf}\label{f:splitting}}
\caption{ This paper is the first work to describe the difference between two topic evolution phenomena: (a) topic speciation and (b) topic splitting. }
\end{figure*}
A major challenge to the existing methods in the literature concerns the detection of topic merging and splitting. Since the connectedness of topics across epochs is based on their similarity what previous work describes as `splitting' or indeed `merging' does not adequately capture these phenomena. Rather, adopting the terminology from biological evolution, a more accurate description would be `speciation' and `convergence' respectively. The former is illustrated in Fig~\ref{f:speciation} whereas the latter is entirely analogous with the time arrow reversed. What the conceptual diagram shown illustrates is a slow differentiation of two topics which originate from the same `parent'. Actual topic splitting, which does not have a biological equivalent in evolution, and which is conceptually illustrated in Fig~\ref{f:splitting} cannot be inferred by measuring topic similarity. Instead, in this work we propose to employ the Kullback-Leibler divergence (KLD) for this purpose. This divergence is asymmetric can be intuitively interpreted as measuring how well one probability distribution `envelops' another. KLD between two probability distributions $p(i)$ and $q(i)$ is defined as follows:
\begin{align}
\rho_\text{KLD} = \sum_i p(i) \log \frac{p(i)}{q(i)}
\end{align}
It can be seen that a high penalty is incurred when $p(i)$ is significant and $q(i)$ is low. Hence, we use the BHD to track gradual topic evolution, speciation, and convergence, while the KLD (computed both in forward and backward directions) is used to detect topic splitting and merging.
\subsubsection{Automatic temporal relatedness graph construction\label{ss:construction}}
Another novelty of the work first described in this paper concerns the building of the temporal relatedness graph. We achieve this almost entirely automatically, requiring only one free parameter to be set by the user. Moreover the meaning of the parameter is readily interpretable and understood by a non-expert, making our approach highly usable.
Our methodology comprises two stages. Firstly we consider all inter-topic connections present in the initial fully connected graph and extract the empirical estimate of the corresponding cumulative density function (CDF). Then we prune the graph based on the operating point on the relevant CDF. In other words if $F_\rho$ is the CDF corresponding to a specific initial, fully connected graph formed using a particular similarity measure (BHD or KLD), and $\zeta \in [0, 1]$ the CDF operating point, we prune the edge between topics $\phi_{j,t}$ and $\phi_{k,t+1}$ iff $\rho(\phi_{j,t},\phi_{k,t+1}) < F^{-1}_\rho (\zeta)$.
\section{Evaluation and discussion}
We now analyse the performance of the proposed framework empirically on a large real world data set.
\subsection{Evaluation data}\label{sss:rawData}
We used the PubMed interface to access the US National Library of Medicine and retrieve from it scholarly articles. We searched for publication on the metabolic syndrome (MetS) using the keyphrase``metabolic syndrome'' and collected papers written in English. The earliest publication found was that by Berardinelli~\textit{et al.}~\cite{BeraCorddeAlCouc1953}. We collected all matching publications up to the final one indexed by PubMed on 10th Jan 2016, yielding a corpus of 31,706 publications.
\subsubsection{Pre-processing\label{sss:preprocessing}}
The raw data collected from PubMed is in the form of free text. To prepare it for automatic analysis a series of `pre-processing' steps are required. The goal is to remove words which are largely uninformative, reduce dispersal of semantically equivalent terms, and thereafter select terms which are included in the vocabulary over which topics are learnt.
We firstly applied soft lemmatization using the WordNet$^\circledR$ lexicon~\cite{Mill1995} to normalize for word inflections. No stemming was performed to avoid semantic distortion often effected by heuristic rules used by stemming algorithms. After lemmatization and the removal of so-called stop-words, we obtained approximately 3.8 million terms in the entire corpus when repetitions are counted, and 46,114 unique terms. Constructing the vocabulary for our method by selecting the most frequent terms which explain 90\% of the energy in a specific corpus resulted in a vocabulary containing 2,839 terms.
\subsection{Results}
We stared evaluation by examining whether the two topic relatedness measures (BHD and KLD) are capturing different aspects of relatedness. To obtain a quantitative measure we looked at the number of inter-topic connections formed in respective graphs both when the BHD is used as well as when the KLD is applied instead. The results were normalized by the total number of connections formed between two epochs, to account for changes in the total number of topics across time. Our results are summarized in Fig~\ref{f:common}. A significant difference between the two graphs is readily evident -- across the entire timespan of the data corpus, the number of Bhattacharyya distance based connections also formed through the use of the KLD is less than 40\% and in most cases less than 30\%. An even greater difference is seen when the proportion of the KLD connections is examined -- it is always less than 25\% and most of the time less than 15\%.
\begin{figure}
\centering
\subfigure[BHD-KLD normalized overlap]{\includegraphics[width=0.99\columnwidth]{BH_KLD_overlap.pdf}}
\subfigure[KLD-BHD normalized overlap]{\includegraphics[width=0.99\columnwidth]{KLD_BH_overlap.pdf}}
\caption{ The proportion of topic connections shared between the BHD and the KLD temporal relatedness graphs, normalized by (a) the number of BHD connections, and (b) the number of KLD connections, in an epoch. }
\label{f:common}
\end{figure}
To get an even deeper insight into the contribution of the two relatedness measures, we examined the corresponding topic graphs before edge pruning. The plot in Fig~\ref{f:smooth} shows the variation in inter-topic edge strengths computed using the BHD and the KLD (in forward and backward directions) -- the former as the $x$ coordinate of a point corresponding to a pair of topics, and the latter as its $y$ coordinate. The scatter of data in the plot corroborates our previous observation that the two similarity measures indeed do capture different aspects of topic behaviour.
We performed extensive qualitative analysis which is necessitated by the nature of the problem at hand and the so-called `semantic gap' that underlies it. In all cases we found that our algorithm revealed meaningful and useful information, as confirmed by an expert in the area of metabolic MetS research.
Our final contribution comprises a web application which allows users to upload and analyse their data sets using the proposed framework. The application allows a range of powerful tasks to be performed quickly and in an intuitive manner. For example, the user can search for a given topic using keywords (and obtain a ranked list), trace the origin of a specific topic backwards in time, or follow its development in the forward direction, examine word clouds associated with topics, display a range of statistical analyses, or navigate the temporal relatedness graph.
\section{Summary and Conclusions}
In this work we presented a case for the importance of use of advanced machine learning techniques in the analysis and interpretation of medical literature. We described a novel framework based on non-parametric Bayesian techniques which is able to extract and track complex, semantically meaningful changes to the topic structure of a longitudinal document corpus. Moreover this work is the first to describe and present a method for differentiating between two types of topic structure changes, namely topic splitting and what we termed topic speciation. Experiments on a large corpus of medical literature concerned with the metabolic syndrome was used to illustrate the performance of our method. Lastly, we developed a web application which allows users such as medical researchers to upload their data sets and apply our method for their analysis; the application and its code will be made freely available following publication.
\begin{figure}
\centering
\includegraphics[width=0.99\columnwidth]{KLD_BH_smooth.pdf}
\caption{ Relationship between inter-topic edge strengths computed using the BHD and the KLD before the pruning of the respective graphs. }
\label{f:smooth}
\end{figure}
\balance
\bibliographystyle{ieee}
| e89a56f1867c07c34411a92851ee6cef5d631cb1 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}\label{sec:intro}
Many central problems in geometry, topology, and mathematical physics lead to questions concerning the long-time dynamics of solutions to ordinary and parabolic partial differential equations. Three basic scenarios are possible. \\
First, under suitable dissipativity conditions, longtime existence is guaranteed and a low-dimensional global attractor of uniformly bounded eternal solutions exists. The structure of such global attractors details longtime behavior. Some of those structures are encoded in meanders -- which open doors to a broad variety of algebraic structures far beyond the original PDE scope of the attractor question. \\
Second, longtime existence without dissipativity features trajectories which escape to infinity in infinite time (grow-up). In some cases it is possible to construct compactifications of an unbounded attractor containing, for example, equilibria at infinity. As an example we show how equivariant bifurcations give rise to black hole initial data that is not spherically symmetric. \\
Third, singularities may develop after finite time. In other words, some solutions escape to infinity within finite time (blow-up). Some of the ramifications are explored in complex time, here. Cosmologically, the most meaningful singularity is the big-bang. We address the Bianchi models: a system of ordinary differential equations which represent the homogenous, but anisotropic, Einstein equations. The Belinsky--Khalatnikov--Lifshitz (BKL) conjecture favors them as models for a spacelike singularity. We obtain stability results for a primordial universe tumbling between different Kasner states, each of which represents a self-similarly expanding universe. Furthermore we address the BKL question of particle horizons. To conclude, we embed general relativity in the broader context of Ho\v{r}ava--Lifschitz gravity, to study the influence on the tumbling behavior between the Kasner states and the resulting chaoticity.
\section{Global attractors of parabolic equations}\label{sec:globalattractors}
\numberwithin{equation}{section}
\numberwithin{figure}{section}
In this section we survey recent progress on the global dynamics, mostly of the scalar parabolic semilinear equation
\begin{equation}\label{C8:eq:2.1.1}
u_t = u_{xx} +f\,.
\end{equation}
The emerging global theory gives rise to surprising topological structures, as finite regular cell complexes, in Section~\ref{subsec:sturmglobalattractors}, and to combinatorial algebraic structures like Temperley--Lieb algebras, in Section~\ref{subsec:meanders}. An application to axisymmetric, but not spherically symmetric, initial data for black holes which satisfy the Einstein constraint is given in Section~\ref{subsec:blackholedata}. The question of stability of stationary solutions is addressed in Section~\ref{subsec:statsol}. \\
For simplicity we consider the PDE~\eqref{C8:eq:2.1.1} under Neumann $\mathcal{N}$ or periodic $\mathcal{P}$ boundary conditions, on the unit interval $0<x<1$. Other separated linear boundary conditions can be treated analogously to the Neumann case. The quasilinear uniformly parabolic case may easily be believed to yield identical results, but the long necessary groundwork is not complete at this stage.
\subsection{Sturm global attractors}\label{subsec:sturmglobalattractors}
We consider the PDE~\eqref{C8:eq:2.1.1} for several classes of nonlinearities $f$. In particular, we survey our results~\cite{FiedlerRochaWolfrum2012-SturmAttractors, Fiedler15, FiedlerRochaWolfrum12, FiedlerGrottaRocha14, RochaFiedler14, FiedlerRocha14}. We denote these classes by $\mathcal{N}(x,u,u_x)$, $\mathcal{N}(u,u_x)$, $\dots$ for $f=f(x,u,u_x)$, $f=f(u,u_x)$, $\dots$ under Neumann boundary conditions, and by $\mathcal{P}(x,u,u_x)$, $\mathcal{P}(u,u_x)$, $\dots$ in the $x$-periodic case. Dirichlet or other separated boundary conditions of mixed type can be treated analogously to the Neumann case. We always assume $f\in C^1$ to be dissipative, i.e., the set of bounded \emph{eternal solutions} $u(t,x),t\in\mathbb{R}$ is itself bounded in a suitable Sobolev type interpolation space $u(t,\cdot)\in X \hookrightarrow C^1$. Explicit assumptions on $f$ which guarantee dissipativity involve, for example, sign conditions $f(x,u,u_x) \cdot u<0$, for large $|u|$, and subquadratic growth in $u_x$. The bounded set $\mathcal{A}_f \subseteq X$ of spatial profiles $u(t,\cdot)$ on bounded eternal solutions is called the \emph{global attractor} of~\eqref{C8:eq:2.1.1}. The concept originated with Ladyzhenskaya's work on the two-dimensional Navier--Stokes equation. \\
For general surveys on global attractors $\mathcal{A}$ see for example~\cite{hale,vishikchepyzhov}. The general theory establishes properties like existence, global attractivity (hence the name), compactness, finite-dimensionality in the fractal box counting sense, and sometimes connectivity. From such a general, and hence necessarily superficial, point of view the global attractors $\mathcal{A}_f$ of our model class~\eqref{C8:eq:2.1.1} are just an easy exercise. \\
However, our objective is a much deeper global understanding of the detailed global structure of the global attractors $\mathcal{A}_f$. Let us address the general Neumann case $f\in\mathcal{N}(x,u,u_x)$ first. By results of Zelenyak and Matano, the PDE~\eqref{C8:eq:2.1.1} then possesses a gradient-like Morse structure due to a decreasing Lyapunov, or energy, or Morse, functional $V$. Throughout we assume all Neumann equilibrium solutions
\begin{equation}\label{C8:eq:2.1.2}
0=v_{xx}+f
\end{equation}
to be hyperbolic, i.e., to be quadratically nondegenerate critical points of $V$. Then
\begin{equation}\label{C8:eq:2.1.3}
\mathcal{A}_f =\mathcal{E}_f \cup\mathcal{H}_f
\end{equation}
consists of the finitely many equilibria $v\in\mathcal{E}_f$ and \emph{heteroclinic orbits} $u(t,\cdot)\in\mathcal{H}_f$ \emph{connecting} them:
\begin{equation}\label{C8:eq:2.1.4}
u(t,\cdot)\rightarrow v_{\pm}\in\mathcal{E}_f \quad \text{for} \quad t \rightarrow \pm \infty\,,
\end{equation}
with equilibria $v_+\neq v_-$. We call $\mathcal{A}_f$ a \emph{Sturm global attractor}, to emphasize the particular origin from the one-dimensional parabolic PDE class~\eqref{C8:eq:2.1.1}. Information on the vertex pairs $(v_-,v_+)$ which do possess a heteroclinic edge~\eqref{C8:eq:2.1.4}, and on the pairs which don't, can be encoded in the (directed, acyclic, loop-free) \emph{connection graph} $\mathcal{C}_f$; see~\cite{FiedlerCastaneda2012-Meander} and Fig.~\ref{C8:fig:2.1.1}(a). \\
\begin{figure}
\centering \includegraphics[angle=0,width=\textwidth]
{figs/REPORT_SFB647_Fig1_Meander.pdf}
\caption{\emph{(a) Sturm global attractor $\mathcal{A}_f$ and connection graph $\mathcal{C}_f$ of a PDE~\eqref{C8:eq:2.1.1} with 9 hyperbolic equilibria $v$, labeled by $1,\dots, 9=N$.
One-dimensional heteroclinic orbits between equilibria of adjacent Morse index are indicated by arrows.
(b) Open meander, at $x=1$, which leads to the 9 equilibrium solutions $v$ by shooting from the Neumann boundary condition at $x=0$.
(c) Closed meander which arises by replacing the dashed upper arch $(1,2)$, in (b), with the new dashed upper arch $(9,2)$; see~\cite{FiedlerCastaneda2012-Meander}.}}
\label{C8:fig:2.1.1}
\end{figure}
As was discovered by Fusco and Rocha~\cite{FuscoRocha}, a permutation $\sigma$ given by the boundary values of the equilibria $\mathcal{E}_f= \{ v_1,\dots, v_N\}$ is the key to a global analysis of the Sturm global attractor $\mathcal{A}_f$ and its heteroclinic connection graph $\mathcal{C}_f$. The \emph{Sturm permutation} $\sigma=\sigma_f$ encodes the orderings of the equilibria,
\begin{equation}\label{C8:eq:2.1.5}
\begin{aligned}
v_1 \;&<\; v_2 \!\!\!\!&<\; \dots \;&< \; v_N &\quad \text{at}\quad x=0\,,\\
v_{\sigma (1)} \;&<\; v_{\sigma (2)} \!\!\!\!&< \;\dots \;&< \;v_{\sigma (N)} &\quad \text{at} \quad x=1\,,
\end{aligned}
\end{equation}
at the respective Neumann boundaries. More precisely, the Sturm permutations $\sigma_f$ determine all \emph{unstable dimensions}, alias \emph{Morse indices}, $i=i(v_k)$ of the equilibria $v_k \in \mathcal{E}$; see~\cite{FuscoRocha}. This is a linear ODE question of Sturm--Liouville type; hence the name Sturm permutation for $\sigma_f$. Moreover $\sigma_f$ determines the (PDE) heteroclinic connection graph $\mathcal{C}_f$, explicitly and constructively, in a combinatorial and algorithmic style~\cite{FiedlerRocha1996-SturmPDE}. Global Sturm attractors $\mathcal{A}_f$ and $\mathcal{A}_g$ with the same Sturm permutation $\sigma_f = \sigma_g$ are in fact $C^0$ orbit equivalent, i.e.\ homeomorphic in $X$ under a homeomorphism which maps eternal PDE orbits $\{ u(t,\cdot),t\in\mathbb{R}\}\subseteq\mathcal{A}_f$ to eternal orbits in $\mathcal{A}_g$; see~\cite{FiedlerRocha00}. \\
Not all permutations $\sigma$ are Sturm permutations $\sigma_f$ arising from the Neumann class $f \in \mathcal{N}(x,u,u_x)$ of~\eqref{C8:eq:2.1.1}. In fact $\sigma=\sigma_f$ for some $f$, if and only if the permutation $\sigma\in S_N$ is a meander permutation which, in addition, is dissipative and Morse~\cite{FiedlerRocha1999-SturmPermutations}. Following Arnold, here, we call $\sigma$ a \emph{meander permutation} if $\sigma$ describes the transverse crossings of a planar $C^1$ Jordan curve, or ``meandering river'', oriented from south-west to north-east, with the horizontal axis, or ``road''. Labeling the crossings along the ``river'', sequentially, the meander permutation $\sigma$ is defined by the labels $\sigma(1),\sigma(2),\dots,\sigma(N)$ read along the road, left to right. We call a meander $\sigma$ \emph{dissipative} if $\sigma(1)=1$ and $\sigma(N)=N$.
\begin{figure}
\centering\includegraphics[angle=0,width=0.80\textwidth]
{figs/REPORT_SFB647_Fig2_SturmPermutations}
\caption{\emph{All Sturm permutations in $S(9)$, up to trivial symmetries, and their meanders. We indicate all cases which cannot be realized by nonlinearities $f =f(u)$; see~\cite{FiedlerRochaWolfrum2012-SturmAttractors}.}}
\label{C8:fig:2.1.2}
\end{figure}
We call $\sigma$ \emph{Morse} if the integers
\begin{equation}\label{C8:eq:2.1.6}
i_k:= \sum_{j=1}^{k-1}(-1)^{j+1}\operatorname{sign}\big(\sigma^{-1}(j+1)-\sigma^{-1} (j)\big)
\end{equation}
are nonnegative, for all $k=1,\dots,N$. Geometrically, this amounts to a net right winding of the river, as counted from the first crossing; see Fig.~\ref{C8:fig:2.1.1}(b) and Fig.~\ref{C8:fig:2.1.2} for many examples. The PDE connection graphs $\mathcal{C}_f$ then follow, in each case, without further PDE analysis. \\
It is not difficult to see why Sturm permutations $\sigma=\sigma_f$ satisfy these three properties. Dissipativeness of $\sigma$ follows because $\mathcal{E}_f$ must contain extreme equilibria $v_1(x)<v_2(x),\dots,v_{N-1}(x)<v_N(x)$, for all $0\leq x\leq1$, by dissipativeness of~\eqref{C8:eq:2.1.1} and parabolic comparison. The Morse property $i_k\geq0$ follows because $i_k=i(v_k)\geq0$ are the nonnegative Morse indices; see~\cite{FuscoRocha}. The meander property follows by ODE shooting for the equilibrium boundary value problem~\eqref{C8:eq:2.1.2} in the $(v,v_x)$ plane. Indeed, the ``river'' is the image of the horizontal Neumann $v$-axis $\{ v_x =0\}$, at $x=0$, propagated to $x=1$ under the ODE-flow~\eqref{C8:eq:2.1.2}. The ``road'' is the horizontal Neumann $v$-axis $\{v_x=0\}$, at $x=1$. See Section~\ref{subsec:meanders} for a deeper discussion of meanders. The converse claim, how all dissipative Morse meanders $\sigma$ are actually Sturm permutations $\sigma=\sigma_f$ for some $f$, is beyond the limitations of this introductory survey. See \cite{FiedlerRocha1999-SturmPermutations}. The deceptively elementary description of Sturm permutations $\sigma_f$ should not conceal the formidable complications arising from the meander property. Already Gauss got intrigued by the related question of self-intersecting orders in planar curves; see~\cite{Gauss}. Rescaling $x=1$ to become large, the evident complexities of chaotic stroboscope maps for the periodically forced pendulum or the forced van~der~Pol oscillator are also lurking here, as special cases. \\
A key technical ingredient to the results indicated above are nodal properties, i.e.\ a refinement of the parabolic comparison principle in one space dimension, or in radially symmetric settings. Indeed, let $u_1(t,x)$ and $u_2(t,x)$ be any two non-identical solutions of~\eqref{C8:eq:2.1.1}, in $X \hookrightarrow C^1$, and let the \emph{zero number} $z(\varphi)\leq\infty$ count the strict sign changes of any continuous $x$-profile $\varphi: [0,1]\rightarrow\mathbb{R}$. Then the zero number
\begin{equation}\label{C8:eq:2.1.7}
t \mapsto z\big(u_1(t, \cdot) -u_2 (t,\cdot )\big)
\end{equation}
is finite, for any $t >0$, nonincreasing with $t$, and drops strictly at any multiple zeros of $x\mapsto u_1(t,x)-u_2(t,x)$; see~\cite{Angenent}. Early versions of this, in the linear case, in fact date back to Sturm and convinced us to name the nonlinear ramifications after him; see~\cite{Sturm}. \\
It is the zero number dropping property~\eqref{C8:eq:2.1.7} which generates surprising rigidity in the global Sturm attractors $\mathcal{A}_f$ and enables a detailed global analysis. For example, stable and unstable manifolds of hyperbolic equilibria $v_+$ and $v_-$ intersect transversely, automatically, along any heteroclinic PDE orbit $u(t, \cdot)\rightarrow v_\pm$, for $t \rightarrow\pm\infty$, in $\mathcal{H}_f$; see~\cite{Angenent, Henry85}. In particular, this implies local structural stability of Sturm attractors and, moreover, a strict dropping of Morse indices,
\begin{equation}\label{C8:eq:2.1.8}
i(v_-) > z\big(u(t,\cdot)-v_-\big) \geq z(v_+ -v_-) \geq z\big(v_+ -u(t,\cdot)\big) \geq i(v_+)\,.
\end{equation}
By a (standard) transitivity property of transverse heteroclinicity and a (Sturm specific) cascading principle it turns out, in fact, that the full heteroclinic connection graph $\mathcal{C}_f$ is generated by the edges $(v_-,v_+)$ with adjacent Morse index. Saddle-saddle heteroclinic orbits $i(v_-) = i(v_+)$ are also excluded by Morse dropping~\eqref{C8:eq:2.1.8}. This precludes global topological ambiguities which, for example, homotopy-invariant Conley--Morse theory is not allowed to discard. Indeed the positive integer polynomial terms $(1+t)\,q(t)$ in the Morse (in)equalities, alias connecting homomorphisms or Conley--Franzosa connection matrices, always need to accommodate global reshufflings of the connection graph due to nongeneric saddle-saddle connections. This is where nodal arguments involving the zero number $z(u_1-u_2)$ step in. \\
We summarize more recent results on Sturm global attractors $\mathcal{A}_f$ and their characterizing Sturm permutations $\sigma_f$ for the parabolic PDE~\eqref{C8:eq:2.1.1} under Neumann $(\mathcal{N})$ and periodic $(\mathcal{P})$ boundary conditions, next. We conclude with recent progress in our global geometric understanding of $\mathcal{A}_f$ as regular topological cell complexes. \\
Little information is provided, in our approach, on the global Sturm attractor $\mathcal{A}_f$ of particular nonlinearities $f\in\mathcal{N}(x,u,u_x)$. The reason is purely on the ODE pendulum side~\eqref{C8:eq:2.1.2}. It is not an easy task to determine the equilibrium boundary order of $v\in\mathcal{E}_f$, as encoded in the Sturm permutation $\sigma_f$, for particular nonlinearities $f$. Indeed we address the whole class of dissipative nonlinearities $f=f(x,u,u_x)$, rather than any particular instance. \\
As a partial remedy we also consider the subclass $\mathcal{N}(u)$ of nonlinearities $f=f(u)$, for which the equilibrium ODE~\eqref{C8:eq:2.1.2} becomes $x$-autonomous Hamiltonian. See~\cite{FiedlerRochaWolfrum2012-SturmAttractors} and the survey~\cite{RochaFiedler14} for an explicit combinatorial characterization of this subclass of Sturm permutations $\sigma_f$. For a complete listing with $N=9$ equilibria see again Fig.~\ref{C8:fig:2.1.2}. The case $f\in\mathcal{N}(u,u_x^2)$ of nonlinearities $f(u,p) =f(u,-p)$ which are even in $p=u_x$ is only superficially more general.\ The equilibrium ODE~\eqref{C8:eq:2.1.2} is then only reversible in $x$, but not necessarily Hamiltonian. Still, reversibility implies integrability here. The ODE~\eqref{C8:eq:2.1.2}, but not the PDE~\eqref{C8:eq:2.1.1}, can be smoothly transformed to become Hamiltonian. The original Sturm permutation $\sigma_f$ becomes an involution, $\sigma_f=\sigma_f^{-1}$. Moreover $\sigma_f$ can be represented by a Hamiltonian nonlinearity $g\in\mathcal{N}(u)$ such that $\sigma_f=\sigma_g$. In particular, the Sturm global attractors $\mathcal{A}_f$ in the larger $x$-reversible class $\mathcal{N}(u,u_x^2)$ coincide with those of the $x$-reversible Hamiltonian class $\mathcal{N}(u)$. \\
Next, we turn to the case $f\in\mathcal{P}(x,u,u_x)$ of periodic boundary conditions $x\in S^1$. The nodal dropping property~\eqref{C8:eq:2.1.7} of zero numbers prevails. It has long been known, however, that any planar autonomous ODE vector field embeds into this larger class of PDEs~\cite{SandstedeFiedler92}. In particular, saddle-saddle heteroclinic and homoclinic trajectories may arise, and automatic transversality of stable and unstable manifolds fails. In the $S^1$-equivariant case of $x$-independent nonlinearities $f\in\mathcal{P}(u,u_x)$, on the other hand, automatic transversality does follow from mere (normal) hyperbolicity of equilibria and nonconstant rotating waves $u(t,x) = U\, (x-ct)$. In fact the global attractor then decomposes into equilibria $\mathcal{E}$, rotating waves $\mathcal{R}$, and their heteroclinic orbits $\mathcal{H}$ as
\begin{equation}\label{C8:eq:2.1.9}
\mathcal{A}_f = \mathcal{E}_f \cup \mathcal{R}_f \cup \mathcal{H}_f\,.
\end{equation}
In this periodic class, the PDE~\eqref{C8:eq:2.1.1} therefore defines a Morse--Smale system. In the $x$-reversible, i.e.\ $O(2)$-equivariant, class $f \in \mathcal{P}(u,u_x^2)$ an explicit Lyapunov function has been constructed; see~\cite{FiedlerGrottaRocha14}. Then all ``rotating'' waves are frozen, i.e.\ of wave speed $c=0$, and time-periodic solutions other than equilibria cannot arise. \\
In~\cite{FiedlerRochaWolfrum12}, we have developed an explicit algorithm to determine the heteroclinic connection graph in the general $S^1$-equivariant case $f_0\in\mathcal{P}(u,u_x)$. The basic procedure consisted in a homotopy $f_\tau$ from $f_0$ to $f_1\in\mathcal{P}(u,u_x^2)$ which preserved (normal) hyperbolicity of all equilibria and rotating waves, freezing the latter to wave speed $c=0$ at $\tau =1$. We were then able to derive the heteroclinic orbits in $\mathcal{P}$ from those in the reversible Neumann class $f_1\in \mathcal{N }(u,u_x^2)$ or, equivalently, the Hamiltonian Neumann class $f_1\in\mathcal{N}(u)$. \\
As an explicit example we mention the \emph{Sturm spindle} attractor $\mathcal{A}$ of Fig.~\ref{C8:fig:2.1.3},
\begin{figure}
\centering \includegraphics[angle=0,scale=1]
{figs/REPORT_SFB647_Fig3_SturmSpindel.pdf}
\caption{\emph{A geometric representation of the Sturm spindle attractor with three equilibria and one rotating wave; see also~\cite{KrisztinWalther01}.}}
\label{C8:fig:2.1.3}
\end{figure}
which consists of three equilibria $e_k$ of Morse indices $i(e_1)=i(e_3)=0$, $i(e_2)=3$ and one rotating wave $w$ with (strong) unstable dimension one. Surprisingly, the same global attractor $\mathcal{A}$ has also been encountered by Krisztin and Walther in the quite different setting of delay differential equations
\begin{equation}\label{C8:eq:2.1.10}
\dot{u}(t) = g\big(u(t),u(t-1)\big)\,,
\end{equation}
for positive monotone feedback $g$ in the delayed argument; see~\cite{KrisztinWalther01}. There is a nodal property similar to the zero number $z$ of~\eqref{C8:eq:2.1.8} at work, here.
The delay class is not exhaustively classified; it might turn out to be a subclass of the parabolic class, together with the closely related class of negative monotone delayed feedback. This striking analogy remains unexplored at present. \\
At the end of this section we return to the general Neumann class $f\in\mathcal{N}(x,u,u_x)$. We have a combinatorial description of this PDE class in terms of ODE shooting meanders, the Sturm permutations $\sigma_f$, and the explicit derivation of the heteroclinic connection graph $\mathcal{C}_f$. We did not yet succeed, however, to a priori characterize the class of all connection graphs $\mathcal{C}_f$ or, more ambitiously, the topological properties of the global Sturm attractors $\mathcal{A}_f$ themselves -- other than by mindless successive enumeration of all $\sigma_f$. However, some promising steps have been achieved. \\
We recall the gradient-like structure which implies a general decomposition $\mathcal{A}=\mathcal{E}\cup\mathcal{H}$ of the global attractor into (hyperbolic) equilibria $\mathcal{E}$ and their heteroclinic orbits $\mathcal{H}$. In other words,
\begin{equation}\label{C8:eq:2.1.11}
\mathcal{A}=\bigcup_{v\in\mathcal{E}}W^u(v)
\end{equation}
decomposes into the disjoint union of the unstable manifolds $W^u(v)$ of all equilibria $v$. Here $W^u(v)$ consists of all ancient trajectories $u(t,\cdot)$ in $\mathcal{A}$ which limit onto $v$, for $t\rightarrow-\infty$. In particular, $v \in W^u(v)$. Note that $\dim W^u(v)=i(v)$ is the Morse index of $v$, and $W^u(v)=\{ v\}$ for $i(v) = 0$. We call~\eqref{C8:eq:2.1.11} the \emph{dynamic complex} of $\mathcal{A}$. In general, the boundaries $\partial W^u =(\operatorname{clos}W^u)\smallsetminus W^u$ need not be manifolds. In fact,~\eqref{C8:eq:2.1.11} need not even be a cell complex, even though the $W^u(v)$ are embedded balls of dimensions $i(v)$. In the class of Sturm global attractors $\mathcal{A}= \mathcal{A}_f$, $f \in \mathcal{N}(x,u,u_x)$, however, the decomposition~\eqref{C8:eq:2.1.11} is a \emph{regular cell complex}. This means that the embedded $i(v)$-cells $W^u(v)$ possess embedded sphere boundaries $\partial W^u(v) \cong S^{i(v)-1}$, so that the attaching maps onto the boundary complexes are homeomorphisms, rather than just continuous maps. The proof requires a theorem of Schoenflies type for the embeddings of the closures $\operatorname{clos}W^u(v)$; see~\cite{Fiedler15}. For the resulting regular dynamic cell complex~\eqref{C8:eq:2.1.11} see~\cite{FiedlerRocha14}. \\
For a more concrete example, the case of a single closed Sturm 3-ball
\begin{equation}\label{C8:eq:2.1.12}
\mathcal{A}_f = \operatorname{clos}W^u(0) = W^u(0) \cup S^2\,.
\end{equation}
with trivial equilibrium $i(0)=3$ has been addressed in~\cite{FiedlerRocha14}. It has been shown how any finite regular cell decomposition of the Schoenflies boundary $\partial W^u(0) = S^2$ can be realized, topologically, as the dynamic complex
\begin{equation}\label{C8:eq:2.1.13}
S^2 = \bigcup_{v \in \mathcal{E} \smallsetminus \{0\}} W^u(v)
\end{equation}
of the remaining equilibria $v$ with Morse index $i(v) \leq 2$, in the Neumann class $f \in \mathcal{N}(x,u,u_x)$. \\
A more careful analysis of this example, however, also keeps track of the two-dimensional \emph{fast unstable manifold} $W^{uu}(0) \subset W^u(0)$ of those solutions $u(t, \cdot)$ in $W^u$ which do not decay to $v$ at the slowest possible exponential rate in $W^u(0)$, for $t \rightarrow -\infty$. By our Schoenflies results $\partial W^{uu}(0)=S^1$ decomposes $S^2$, as an ``equator'', into two (closed) hemispheres $S_\pm^2$. Each closed hemisphere, separately, is realizable as a planar Sturm attractor $\mathcal{A}_\pm = \mathcal{A}_{f_\pm}$ for some nonlinearities $f_\pm \in \mathcal{N}(x,u,u_x)$. To prove our Sturm realization~\eqref{C8:eq:2.1.13} in~\cite{FiedlerRocha14} we took one of the hemispheres to be a single closed 2-cell. The general question of admissible, i.e.\ Sturm realizable, hemisphere decompositions is unexpectedly delicate; see our work in progress~\cite{FiedlerRocha16a,FiedlerRocha16b,FiedlerRocha16c}. As a caveat, we just mention that the stable extreme equilibria $v_1, v_{27}$ in any dissipative Sturm realization of the 3d solid octahedron complex $\mathbb{O}$ of 27 equilibria must be chosen as adjacent corners of $\mathbb{O}$, rather than antipodally. For adjacent corners $v_1, v_{27}$, the two hemispheres $S_\pm^2$ are allowed to decompose $S^2$ into 1+7 or 2+6 octahedral faces, but not into 3+5 or 4+4 faces, by the fast unstable manifold $W^{uu}(0)$.\\
To complete our ambitious goal of a complete description of the Sturm regular dynamic complexes~\eqref{C8:eq:2.1.11}, an inductive process comes to mind. Suppose all Sturm global attractors $\mathcal{A}_f$, $f \in \mathcal{N}(x,u,u_x)$, have been understood for dimensions up to $n-1$. Two steps remain. In a first step we have to understand the sphere boundary of single Sturm balls $\mathcal{A}=\operatorname{clos}W^u(0)$, for $i(0)=n$, as gluings of closed hemisphere complexes $\mathcal{A}_\pm$ of dimension $n-1$, which are assumed to be known by induction hypothesis. In a second step we have to understand how these new $n$-cells attach to each other, and to lower-dimensional cells, to form a general Sturm global attractor of dimension $n$. Again, the respective hemisphere decompositions should play a crucial role here. At present, we have understood step~2 only up to the planar case $n=2$; see~\cite{FiedlerRocha2008-SturmAttractors2,FiedlerRocha2009-SturmAttractors1,FiedlerRocha2010-SturmAttractors3}. Step~1 for $n=3$ is imminent~\cite{FiedlerRocha16a,FiedlerRocha16b,FiedlerRocha16c}.
\subsection{Meanders}\label{subsec:meanders}
In the previous section, we introduced meander curves in the plane as shooting curves of parabolic PDEs~\eqref{C8:eq:2.1.1} in one space dimension; see in particular~\eqref{C8:eq:2.1.2},~\eqref{C8:eq:2.1.5} and Figs.~\ref{C8:fig:2.1.1},~\ref{C8:fig:2.1.2}. Quite independently from this ODE and PDE context, meanders also arise as trajectories of Cartesian billiards~\cite{FiedlerCastaneda2012-Meander}, and as representations of monomials in Temperley--Lieb algebras~\cite{DiFrancescoGolinelliGuitter1997-Meander}.
\begin{figure}
\centering \includegraphics[angle=-90,width=\textwidth]
{figs/REPORT_SFB647_Fig4_RainbowMeander.pdf}
\caption{\emph{(a) A seaweed meander $\mathcal{M}(2,2\,|\,1,3)$ with $N=8$ intersections.
(b) The associated cleaved bi-rainbow meander $\mathcal{M}_*(2,2,3,1\,|\,8)$ with $2N=16$ intersections.
Note how upper arches $(i,j)$ of $\mathcal{M}$ have been preserved.
Lower arches $(i,j)$ of $\mathcal{M}$ have been converted to upper arches $(2N+1-i,\, 2N+1-j)$ of $\mathcal{M}_*$; see~\cite{FiedlerCastaneda2012-Meander}.}}
\label{C8:fig:2.2.1}
\end{figure}
For simplicity we now represent meander curves, or equivalently meander permutations, by $\left[N/2\right]$ standardized upper and lower half-circular arches joining at vertices $1,\dots,N$ along the horizontal axis. Also we replace the open meanders from south-west to north-east, of Section~\ref{subsec:sturmglobalattractors}, by their closed Jordan curve counterparts. See Fig.~\ref{C8:fig:2.1.1}(c) for one simple closing procedure. \\
More generally, we now regard closed meanders $\mathcal{M}$ as the pattern created by one or several disjoint closed smooth Jordan curves in the plane as they intersect a given horizontal line transversely. Let $c(\mathcal{M}) \geq 1$ count the connected components of $\mathcal{M}$. In view of Sturm global attractors, of course, the case $c(\mathcal{M}) =1$ of connected meanders generated by a single closed Jordan curve is of primary interest. \\
Several new results have been obtained which are based on rainbows. Here a \emph{proper upper rainbow} consists of a number $\alpha$ of concentric and adjacently nested upper arches in any meander. It is understood that $\alpha$ is chosen maximal, in all cases. An \emph{upper rainbow} consists of one or several neighboring proper upper rainbows. Lower rainbows are defined analogously. Note that upper and lower rainbows may themselves be nested, giving rise to a forest of their sizes $\alpha$. A closed meander without rainbow nesting is called a \emph{seaweed meander}. Symbolically a seaweed meander can be described as
\begin{equation}\label{C8:eq:2.2.1}
\mathcal{M}(\alpha_1,\dots,\alpha_{m_+}\,|\, \beta_1,\dots,\beta_{m_-})\,,
\end{equation}
where $\alpha_1,\dots, \alpha_{m_+}$ are the sizes of the neighboring proper upper rainbows, and $\beta_1,\dots, \beta_{m_-}$ enumerate the proper lower rainbows, left to right. See Fig.~\ref{C8:fig:2.2.1}(a) for an example. Trivially
\begin{equation}\label{C8:eq:2.2.2}
\alpha_1+\dots+\alpha_{m_+}=\beta_1+\dots+\beta_{m_-}=N/2
\end{equation}
for any closed meander of $N$ vertices. The special seaweed case $m_- =1$ of a single lower rainbow of size $\beta = N/2$ is called a \emph{bi-rainbow meander}.
\begin{figure}
\centering \includegraphics[angle=-90,width=0.8\textwidth]
{figs/REPORT_SFB647_Fig5_MeanderCartesianBilliard.pdf}
\caption{\emph{(a) A bi-rainbow meander with two connected components, one distinguished by dashed arches.
(b) A corresponding nontransitive plane Cartesian billiard. One closed flight path is distinguished by dashed lines; see~\cite{FiedlerCastaneda2012-Meander}.}}
\label{C8:fig:2.2.2}
\end{figure}
The case where the upper rainbows may be arbitrarily nested, but there is only one single lower rainbow $\beta = N/2$, is called \emph{lower rainbow meander}; see Fig.~\ref{C8:fig:2.2.1}(b). For example any closed meander with $N$ vertices can be ``opened'' to a closed lower rainbow meander with $2N$ vertices and the same number of connected components. Indeed we just duplicate the original vertices $j \mapsto j,j'$ and open up the horizontal axis to accommodate the vertices in the order $1,\dots, N,N',\dots, 1'$; see again Fig.~\ref{C8:fig:2.2.1}. Recall how only connected meanders relate to an interpretation as ODE shooting curves for PDE Sturm global attractors. Lower rainbow meanders of the above general type were related to Cartesian billiards by Fiedler and Casta\~{n}eda; see~\cite{FiedlerCastaneda2012-Meander} and Fig.~\ref{C8:fig:2.2.2}. \\
In fact, there exists a plane with only horizontal and vertical billiard boundaries, with vertices all on the Cartesian integer grid and billiard slopes of $\pm 45^\circ$ between half-integer boundary points, such that the connectivities of paths in the billiard and in the meander coincide. As an almost immediate consequence one obtains the following list for the number $c(\alpha_1,\dots,\alpha_m)$ of connected components of bi-rainbows $\mathcal{M}(\alpha_1,\dots, \alpha_n\,|\,\beta)$ in terms of greatest common divisors $\gcd$:
\begin{equation}\label{C8:eq:2.2.3}
\begin{aligned}
c(\alpha_1) &= \alpha_1\,,\\
c(\alpha_1,\alpha_2) &= \gcd (\alpha_1, \alpha_2)\,,\\
c(\alpha_1, \alpha_2, \alpha_3) &= \gcd (\alpha_1 +\alpha_2, \alpha_2 + \alpha_3)\,.
\end{aligned}
\end{equation}
The dissertation of Karnauhova~\cite{Karnauhova2016-Dissertation} establishes how such formulas cannot be extended to $m\geq4$. In fact $c$ cannot be written as the $\gcd$ of any two polynomials $f_1,f_2$ in $\alpha_1,\dots,\alpha_m$ with integer coefficients. In~\cite{KarnauhovaLiebscher2015-Meanders} this obstacle is overcome, replacing the algorithm ``$\gcd$'' by an explicit recursion of related logarithmic complexity in terms of the data $\alpha_1,\dots,\alpha_m$. \\
Encouraged by suggestions of Matthias Staudacher (projects C4 ``Structure of Quantum Field Theory: Hopf Algebras versus Integrability'' and C5 ``AdS/CFT Correspondence: Integrable Structures and Observables'') and earlier hints by Eberhard Zeidler, we also began to explore relations between meanders and Temperley-Lieb Hopf algebras $TL_N(\tau)$; see~\cite{FiedlerCastaneda2012-Meander}. Algebraically these are given by $N$ multiplicative generators $e_0=1,e_2,\dots,e_{N-1}$ with the three relations
\begin{equation}\label{C8:eq:2.2.4}
\begin{aligned}
e_i^2 &=\tau e_i\,,\\
e_i e_j &= e_j e_i\,,\\
e_i e_{i \pm 1} e_i &= e_i\,,
\end{aligned}
\end{equation}
for all appropriate $i,j >0$ with $|i-j|\, >1$.
\begin{figure}
\centering \includegraphics[angle=0,width=\textwidth]
{figs/REPORT_SFB647_Fig6_TLGenerators.pdf}
\caption{\emph{Diagram representation of Temperley--Lieb generators $e_0=1$ and $e_i$, for $i=1,\dots,N-1$; see~\cite{FiedlerCastaneda2012-Meander}.}}
\label{C8:fig:2.2.3}
\end{figure}
See Fig.~\ref{C8:fig:2.2.3} for a representation of the Temperley--Lieb generators $e_i \in TL_N(\tau)$ by $N$-strand diagrams. Multiplication of generators is realized by horizontal concatenation of the generator diagrams, up to strand isotopy. Isolas effect multiplication by $\tau$, each. As illustrated in Fig.~\ref{C8:fig:2.2.4} for $e$:$=e_2e_1e_3 \in TL_4(\tau)$, Temperley--Lieb monomials $e$ correspond to lower rainbow meanders $\mathcal{M}[e]$. The number $c$ of connected components of $\mathcal{M}[e]$ is then related to the \emph{Markov trace} $\operatorname{tr}_F$ of $e$:
\begin{equation}\label{C8:eq:2.2.5}
\operatorname{tr}_F(e) = \tau^{c(\mathcal{M}[e])}\,,
\end{equation}
see~\cite{DiFrancescoGolinelliGuitter1997-Meander}. For further comments and a relation to the classical Young-Baxter equations see also the dissertation by Karnauhova~\cite{Karnauhova2016-Dissertation} and the references there.
\begin{figure}
\centering \includegraphics[angle=0,width=\textwidth]
{figs/REPORT_SFB647_Fig7_TLMonomial.pdf}
\caption{\emph{(a) Diagram representation of the monomial $e=e_2e_1e_3 \in TL_4(\tau)$.
(b) Homotopy equivalent diagram of $e$ with dashed exterior strands matching right and left ends of the same strand index.
Note the number $k=1$ of exterior connected components of the closed diagram.
(c) Joining the lower ends ``$o$'' of the vertical strand boundaries of $e$, and opening the boundaries to become horizontal we obtain an equivalent lower rainbow meander with $8$ intersections and the same number of connected components as in (b); see~\cite{FiedlerCastaneda2012-Meander}.}}
\label{C8:fig:2.2.4}
\end{figure}
\\
The case~\eqref{C8:eq:2.2.1},~\eqref{C8:eq:2.2.2} of seaweed meanders $\mathcal{M}(\alpha_1,\dots, \alpha_{m_+}\,|\, \beta_1,\dots,\beta_{m_-})$ is related to the index of seaweed algebras, as studied by Dergachev and Kirillov~\cite{DK00}. The seaweed algebra of type $(\alpha_1,\dots,\alpha_{m_+}\,|\, \beta_1,\dots,\beta_{m_-})$ is the $\operatorname{gl}(N/2)$ subalgebra of matrices which preserve the vector spaces $V_j$ spanned by unit vectors $e_1,\dots,e_{\alpha_1+\ldots+\alpha_j}$ and $W_j$ spanned by $e_{\beta_1+\ldots+\beta_j+1},\dots,e_{N/2}$. In fact the seaweed meander corresponds to the Cartesian billiard on the domain of admissible nonzero entries of the seaweed matrices. The index of the algebra is the minimal kernel dimension of the map $A \mapsto f[A,\cdot]$, for any linear functional $f$ in the dual algebra. By~\cite{DK00} this index coincides with the number $c(\mathcal{M})$ of connected components of the meander $\mathcal{M}$. \\
In her dissertation, Karnauhova considers the maximally disconnected case $\alpha_j=\beta_j$ of seaweed algebras with maximal index $N/2$, generated by trivial $2$-cycles of $\sigma$, for the closed meander $\mathcal{M}$; see~\cite{Karnauhova2016-Dissertation}. She then shifts the lower $\beta$-rainbows one step to the right. Ratcheting into the upper $\alpha$-rainbows, she obtains a dissipative Morse meander $\sigma_f$, and hence a Sturm global attractor $\mathcal{A}_f$. The Chafee--Infante attractors \cite{Henry85}, for example, which arise for symmetric cubic nonlinearities $f$ and have been the first Sturm global attractors studied historically, correspond to $m_\pm=1$ and $\alpha=\beta$. They are the Sturm global attractors of maximal dimension $\alpha =N/2$ for any given (odd) number $N+1$ of equilibria. The same construction in fact works for general, but identical, upper and lower arch configurations. Although the resulting global attractors are all pitchforkable, and thus have been known ``in principle'' by Fusco and Rocha~\cite{FuscoRocha}, the new approach is elegant and geometric: the Sturm global attractors can be obtained explicitly by successive double cone suspensions and elementary gluing concatenations. \\
In~\cite{Fiedler14}, Fiedler embarked on a different approach to explore the relation between the chaotic dynamics of ($x$-periodically forced) equilibrium equations~\eqref{C8:eq:2.1.2} with nonlinearity $f$ and the Sturm permutations $\sigma_f$ of the resulting shooting meanders. Homoclinic tangles with their associated shift dynamics are taken as a paradigm. In this direction, meander permutations have been defined for standard hyperbolic linear $SL_2(\mathbb{Z})$ torus automorphisms. The meander permutations compare the orderings of homoclinic points along the one-dimensional stable and unstable manifolds of their limits, respectively. As such, they are new topological invariants -- with intriguing relations to continued fraction expansions and quadratic number fields.
\subsection{Stability, instability, and obstacles}\label{subsec:statsol}
Consider the higher-dimensional analogue of~\eqref{C8:eq:2.1.1}, that is, for semilinear equations
$$u_t=\Delta u+f(u)$$
on a domain $\Omega\subseteq{\mathbb R}^N$ with $N\ge1$. More generally consider systems $u\colon\Omega\to{\mathbb R}^n$
\begin{equation}\label{C8:eq:semilinear}
u_t=Lu+f(u)
\end{equation}
with an elliptic differential operator $L$ and associated Dirichlet, Neumann, or mixed boundary values. Then the naturally associated function spaces $W_\Gamma^{1,2}(\Omega)$ and $L^2(\Omega)$ are not continuously embedded into $C(\overline\Omega)$. Therefore, differentiability of the corresponding superposition operator $F(u)(x)=f(u(x))$ fails in $L^2(\Omega)$ (except in the linear case). This imposes an unrealistically restrictive growth condition for $f$ in case $F\colon W_\Gamma^{1,2}(\Omega)\to L^2(\Omega)$. Moreover, no analogue of the zero number~\eqref{C8:eq:2.1.7} with a property like~\eqref{C8:eq:2.1.8} is available. Nevertheless, we are interested in the long-term behavior of the equation in these two function spaces. \\
A particularly natural question is whether a ``formally'' linearized stability of some stationary solution $u_0$ implies asymptotic stability in these spaces. In other words, suppose the spectrum of the ``formal'' linearization $L+f'(u_0)$ lies in the complex left half-plane. Do all solutions starting close to $u_0$ in the topology of the space $W_\Gamma^{1,2}(\Omega)$ or $L^2(\Omega)$ indeed remain close to $u_0$ and eventually tend to $u_0$ in the corresponding topology? Let $A$ denote the operator naturally associated to $-L$ and the boundary conditions. Suppose $A$ solves Kato's square root problem, that is, if $A^{1/2}\colon W_\Gamma^{1,2}(\Omega)\to L^2(\Omega)$ is an isomorphism. Then $W_\Gamma^{1,2}(\Omega)=D(A^{1/2})$. The asymptotic stability in $W_\Gamma^{1,2}(\Omega)$ is known for $F\colon W_\Gamma^{1,2}(\Omega)\to L^2(\Omega)$; see e.g.~\cite{Henry}. Gurevich and V\"{a}th~\cite{GurevichVaeth} have relaxed this hypothesis to a subcritical growth condition on $f$. Moreover, they obtained an asymptotic stability result in $L^2(\Omega)$. The method of proof was to consider extrapolated spaces and extrapolated semigroups which were obtained by two different procedures. As a side result, they have shown that these extrapolation procedures coincide if and only if $A$ solves Kato's square root problem. \\
One particular motivation to consider the topology of the spaces $W_\Gamma^{1/2}(\Omega)$ and $L^2(\Omega)$ is that the following question so far can be addressed only in these topologies. Suppose that the problem~\eqref{C8:eq:semilinear} is endowed with some obstacle, i.e., the values of $u$ on some parts of $\Omega$ or of its boundary are forced to stay within a certain cone by some constraint. One of our surprising results is that the principal of linearize asymptotic stability may then fail in $W_\Gamma^{1,2}(\Omega)$. The proof is based on an instability criterion for variational inequalities and a rather general formula relating the fixed point index of the flow of the obstacle problem with the topological degree of a certain associated operator~\cite{KimVaeth}. For specific examples see~\cite{KimVaeth,VaethProc}.
\begin{figure}
\centering \includegraphics[angle=0,width=\textwidth]
{figs/REPORT_SFB647_Fig8_BlackHoleData.pdf}
\caption{\emph{A $2d$-caricature of black hole initial data for the Einstein equations. The vertical variable is the radius $r$. Each horizontal section $\{ r=\operatorname{constant}\}$ in this figure is a circle, which represents a $2$-sphere $S^2$. The region outside the apparent horizon (and above it in this picture) was treated in the literature. In~\cite{FiedlerHellSmith15} we construct the region $M$ between an initial radius $r_0>0$ and a blow-up radius $r_1 =1$ whose section $\{r_1\} \times S^2$ is a critical point of the area functional.\ The matching and center regions remain to be constructed; see~\cite{FiedlerCastaneda2012-Meander}.}}
\label{C8:fig:2.3.1}
\end{figure}
\begin{figure}
\centering \includegraphics[angle=-90,width=\textwidth]
{figs/REPORT_SFB647_Fig9_BifurcationPitchfork.pdf}
\caption{\emph{Local transcritical and pitchfork bifurcation branches $(\lambda_K(s), v_K(s))$ from $\lambda_K(0) = \lambda_\ell=\ell(\ell+1)$, $v_K(0)=0$ with maximal isotropies $K$ as indicated.
Numbers in brackets indicate co-dimensions of the strong unstable manifolds of their normally hyperbolic group orbits; see~\cite{FiedlerHellSmith15}.}}
\label{C8:fig:2.3.2}
\end{figure}
\begin{figure}
\centering \includegraphics[angle=0,width=\textwidth]
{figs/REPORT_SFB647_Fig10_BifurcatingSymmetry.pdf}
\caption{\emph{The graphs of the bifurcating symmetry breaking equilibria $v_K$ over the sphere $S^2$, with isotropies $K$ as indicated; see~\cite{FiedlerHellSmith15}.}}
\label{C8:fig:2.3.3}
\end{figure}
\subsection{Symmetric and asymmetric black hole data}\label{subsec:blackholedata}
The Cauchy initial-value problem for the relativistic Einstein equations requires initial data which satisfy the Einstein constraint.
For black holes these constraints take the form of an equation for the metric $g$ with prescribed scalar curvature $R$ on an asymptotically flat Riemannian $3$-manifold $M$; see Fig.~\ref{C8:fig:2.3.1} for an illustration and~\cite{Ringstroem2009-CauchyGeneralRelativity} for a general background. For simplicity we consider a collar region $M= [r_0, r_1] \times S^2$ with standard $2$-sphere cross section. In a suitable coordinate foliation, the metric can be expressed as
\begin{equation}\label{C8:eq:2.3.1}
g=u^2dr^2 + r^2\omega
\end{equation}
with the standard metric $\omega$ on $S^2$. For example consider prescribed scalar curvatures $R$ such that $\lambda$:= $r^2 R -2$ is a constant bifurcation parameter. Then $u=u(r,x)$ can be rescaled to
\begin{equation}\label{C8:eq:2.3.2}
u(r,x) = \big( \tfrac{\lambda}{2} (\tfrac{1}{r}-1) \big)^{-1/2}
\big( v \big(-\tfrac{1}{2} \log (1-r),x\big) +1 \big)\,,
\end{equation}
and the Einstein constraints on the radial metric coefficient $u$ take the quasilinear parabolic form
\begin{equation}\label{C8:eq:2.3.3}
v_t = (1+v)^2 \big( \Delta v + \lambda f(v)\big)\,.
\end{equation}
Here $v = v(t,x)$, $x \in S^2$, $\Delta$ is the Laplace-Beltrami operator on $S^2$, and $f(v) = v- \tfrac{1}{2}v^2/(1+v)$ defines a cubic nonlinearity. \\
In~\cite{FiedlerHellSmith15}, we obtain spherically anisotropic solutions $v$ for isotropically prescribed scalar curvature $R$. Equilibria $v(b,x)=v_*(x)\not\equiv\operatorname{const}$ provide self-similar anisotropic metrics $g$. Heteroclinic trajectories $u(t,x)\rightarrow v_\pm(x)$, for $t \rightarrow\pm\infty$, are asymptotically self-similar for $r \rightarrow 0$ and $r \rightarrow 1$. The metric $g$ remains smooth, in fact, in the latter limit. We obtain these solutions by an $O(3)$ symmetry breaking bifurcation analysis of the scaled quasilinear PDE in a center manifold at the trivial isotropic equilibrium $v_*\equiv0$. \\
Symmetry breaking bifurcations occur at $\lambda_\ell = \ell(\ell+1)$, for $l=0,1,2,\dots$, in the spherical harmonics kernels $V_\ell$ of dimension $2\ell+1$. See Fig.~\ref{C8:fig:2.3.2} for the bifurcating branches of nonconstant equilibria $v$, and Fig.~\ref{C8:fig:2.3.3} for their remaining reduced, nonspherical spatial isotropies. Heteroclinic trajectories follow, between $v_*$ and $0$, by standard bifurcation theory in the one-dimensional isotropy subspaces of each kernel $V_\ell$. For heteroclinic trajectories between equilibria of different isotropies; see also~\cite{ChossatLauterbach}. \\
The results above are of a local nature, near the spherically isotropic trivial equilibrium $v_* \equiv 0$. The fully isotropic case $\ell=0$ was treated in~\cite{smith2007blow} previously, by more direct methods. In practice specific computations, particularly of the unstable dimensions and the bifurcation directions, are limited to low $\ell$ and have been carried out for $\ell\leq4$, only. Axisymmetric solutions $v=v(t,\vartheta)$, however, which depend only on the azimuthal angle $\vartheta$ of $x \in S^2$, are known to bifurcate for all $\ell=0,1,2,\dots$. Clearly~\eqref{C8:eq:2.3.3} becomes ``spatially'' one-dimensional,
\begin{equation}\label{C8:eq:2.3.4}
v_t=(1+v)^2 \big(v_{\vartheta \vartheta} + \cot \vartheta\, v_\vartheta + \lambda f(v) \big)\,,
\end{equation}
with Neumann boundary on $0 < \vartheta < \pi$. The dissertation of Lappicy explores these versions of quasilinear Sturm parabolic equations with a singular spatial coefficients, by the global methods of Section~\ref{subsec:sturmglobalattractors}; see~\cite{Lappicy2016-Dissertation}. In particular, this yields a plethora of axisymmetric, but not fully spherically isotropic, equilibrium solutions and their heteroclinic trajectories as parts of black hole initial data which satisfy the Einstein constraint.
\section{Blow-up and Bianchi cosmology}
\numberwithin{equation}{section}
\numberwithin{figure}{section}
\subsection{Dynamics at infinity and blow-up in complex time}
In the previous Section~\ref{sec:globalattractors}, a class of dissipative equations~\eqref{C8:eq:2.1.1} was studied, where longtime existence and boundedness of the solutions was guaranteed. In the present section we focus on equations which develop singularities in finite or infinite time. In other words, some trajectories escape to infinity. In order to study the dynamics at infinity, one possible approach uses \emph{Poincar\'e ``compactification''}: the solution space is projected centrally onto a tangent upper hemisphere. More precisely the hemisphere at its north pole is tangent to the solution space at its origin. Arbitrarily far points in the solution space are thereby mapped to the equator of the hemisphere, also called the \emph{sphere at infinity}. For details see~\cite{helldiss,hell13,perko}. We keep the name Poincar\'e ``compactification'' which originates from ODE theory, where the solution space is $\mathbb{R}^n$ and the compactified space is a compact hemisphere. In the PDE context, the solution space is infinite dimensional and its ``compactification'' is a bounded hemisphere, which is not compact. \\
As an elementary but instructive example, consider the linear equation:
\begin{equation}\label{C8:linearreacdiff}
u_t=u_{xx}+bu\,,\\
\end{equation}
for $0<x<\pi$, with Neumann boundary conditions; see~\cite{helldiss}. The equation~\eqref{C8:linearreacdiff} is not dissipative for positive $b$: some trajectories escape to infinity, in infinite time. The class of equations at the boundary between dissipativity and finite time blow-up is called \emph{slowly nondissipative} in~\cite{BGdiss}: they admit at least one immortal trajectory escaping to infinity in forward time. The linear example~\eqref{C8:linearreacdiff} admits only one bounded equilibrium, located at the origin, and generates infinitely many equilibria at infinity. Those are located at the intersections of the one-dimensional eigenspaces, associated to the eigenfunctions of the Laplacian, with the sphere at infinity. Let us denote the equilibria at infinity by $\pm\Phi_m$, $m\in\mathbb{N}$. For all $b\neq n^2$ which are not square integers, all equilibria, both bounded and at infinity, are hyperbolic. Furthermore the dynamics within the sphere at infinity has a Chafee--Infante structure; see Section~\ref{subsec:sturmglobalattractors}. More precisely the two equilibria $\pm\Phi_0$ at infinity are stable. The equilibria $\pm\Phi_{m}$ at infinity each possess an $m$-dimensional unstable manifold that consists of heteroclinic orbits to the equilibria $\pm\Phi_{j}$, $j\in\{0,\dots,m-1\}$. The unstable manifold of the origin is of finite dimension $[\sqrt{b}]+1$, where $[\sqrt{b}]$ denotes the integer part of $\sqrt{b}$. This unstable manifold consists of heteroclinic orbits to the equilibria $\pm \Phi_i$, $i\in\{0,\dots,[\sqrt{b}]\}$. The stable manifold of the equilibrium at the origin is infinite dimensional and consists of heteroclinic orbits from the equilibria $\pm\Phi_m$ at infinity, $m>[\sqrt{b}]$, to the origin. This describes the global unbounded attractor for~\eqref{C8:linearreacdiff}. \\
When~\eqref{C8:linearreacdiff} is augmented by a smooth bounded nonlinearity $g=g(u)$,
\begin{equation}\label{C8:linreacdiff+bd}
u_t=u_{xx}+bu+g\,,\\
\end{equation}
the dynamics at infinity remains essentially unchanged. In her dissertation~\cite{BGdiss}, Ben-Gal adapts the techniques of Section~\ref{subsec:sturmglobalattractors} to study the global heteroclinic structure of~\eqref{C8:linreacdiff+bd}. In particular, she proves the existence of a finite-dimensional unbounded inertial manifold containing the so called \emph{unbounded} (or \emph{noncompact}) \emph{attractor}. This object consists of all bounded equilibria (generically hyperbolic), their finite dimensional unstable manifolds, the finitely many equilibria at infinity (also generically hyperbolic) to which those unstable manifolds connect, and the finite dimensional unstable manifolds of those equilibria at infinity. Theorems similar to the dissipative case reveal the heteroclinic structure of the unbounded global attractor of slowly nondissipative equations. Pimentel and Rocha generalized some of these results to bounded nonlinearities $g(x,u,u_x)$; see~\cite{pim14, pim15}. \\
We consider blow-up in finite time for~\eqref{C8:eq:2.1.1}, $f=f(u)$, next. For smooth right hand side with polynomial growth, time can be rescaled such that trajectories do not run into equilibria at infinity in finite time. The resulting semiflow on the closed Poincar\'e hemisphere is nonsingular. Conley index theory then provides heteroclinic orbits between equilibria, or general isolated invariant sets. Isolated invariant sets are contained in isolating neighborhoods. An isolating neighborhood is a closed neighborhood whose boundary points eventually leave the neighborhood in forward or backward time direction. For details see~\cite{helldiss} and the references there. \\
It turns out that even for some polynomial ordinary differential equations, equilibria at infinity might not be isolated invariant sets.
\begin{figure}
\begin{center}
\includegraphics[width=0.3\textwidth]{figs/REPORT_SFB647_Fig16_classhamilton.pdf}
\hspace{2cm}
\includegraphics[width=0.28\textwidth]{figs/REPORT_SFB647_Fig15_ersatzinfinity.pdf}
\end{center}
\caption{\emph{Left: the lower equilibrium at infinity is not an isolated invariant set, but is of isolated invariant complement. Right: the ersatz infinity for that equilibrium.}}
\label{homflower}
\end{figure}
Fig.~\ref{homflower}, left hand side, illustrates such an equilibrium at infinity. This equilibrium is isolated, invariant, and stable within the sphere at infinity. Towards the interior of the Poincar\'e hemisphere, however, it shows two petals of nested homoclinic loops. Therefore it does not possess small isolating neighborhoods. The complement of any small neighborhood of the degenerate equilibrium, however, is an isolating neighborhood. An invariant set with this property is called of \emph{ isolated invariant complement}. Although the complementary invariant set isolated by such a neighborhood is possibly very complex, its Conley index is well-defined. Poincar\'e duality then defines a meaningful Conley index for the degenerate equilibrium at infinity. An invariant set at infinity attracts some finite initial conditions, if and only if it is not a repeller. This can be characterized in terms of Conley index. Therefore this tool allows us to determine which part of the dynamics in the sphere at infinity is reachable by grow-up or blow-up solutions. \\
Via a standard construction illustrated in the right hand side of Fig.~\ref{homflower}, and described in detail in~\cite{helldiss,hell13}, it is possible to replace the invariant set of isolated invariant complement by a benign \emph{ersatz} with a well-defined Conley index. This ersatz can be used for the analysis of the global heteroclinic structure via classical Conley index theory. \\
Based on the idea of self-similar scaling and the Poincar\'e sphere at infinity, Fiedler has contributed to an unexpected application in a certain renormalization approach to the experimentally observed formation of spin liquids in carbon nanotubes; see~\cite{hsufied}. Renormalization has been suggested to overcome the limitations of mean-field descriptions. In its most simple form, homogeneously quadratic ODE systems
\begin{equation}
\frac{dx}{dt}=-\nabla_{x}V
\end{equation}
were addressed. The variational structure features a cubic functional
\begin{equation}
V(x)=-\frac13 \sum_{i=1}^{N} x_i (x^T A_i x)
\end{equation}
and real symmetric $N\times N$-matrices $A_i$. As a result we obtain self-similar and asymptotically self-similar blow-up solutions, as well as their blow-up rates. These results provide a rational foundation for the renormalization group approach to large, strongly correlated electron systems. \\
For fast nonlinear diffusion in several space dimensions the phenomenon of extinction of $u$, alias blow-up of $1/u$, has been addressed by Fila and Stuke~\cite{FilaStuke14,FilaStuke16}. As a very simple paradigm for blow-up itself, consider the one-dimensional semilinear heat equation
\begin{equation}\label{C8:eq:nonlinear_heat}
u_t=u_{xx}+u^2
\end{equation}
on a bounded or unbounded interval.\ For real $u,x$ and \emph{real time} $t$, this equation has already been well-studied in the literature~\cite{Quittner07}. The solution is analytic in time and admits a local analytic continuation into the \emph{complex time plane}. For Dirichlet boundary conditions there exists a unique positive equilibrium $u^*$ which is one-dimensionally unstable. Solutions in the one-dimensional analytic unstable manifold of $u^*$ possess a real solution with complete blow-up in finite real time. Motivated by analyticity, Stuke studied extensions to complex, rather than real, time $t$. Of course this entails complex $u$. Following an idea of Masuda~\cite{Masuda84} and Guo at al.~\cite{Guo12}, Stuke attempts to bypass real time blow-up at $t=T$ by an excursion into the complex plane; see Fig.~\ref{fig:complex_time_domain}. Stuke proves that solutions on the complex unstable manifold of $u^*$ stay bounded on time paths $\Gamma_p^\pm$ parallel to the real axis. The solutions converge to zero: they are heteroclinic from $u^*$ to zero, along $\Gamma_p^\pm$, and locally foliate the nonsingular part of the complex one-dimensional unstable manifold $W_*^u$ of $u^*$. After some real time $t=T+\delta>T $, at the latest, the solutions $\Gamma^{\pm}_c$ extend back to the real axis from, both, the upper and the lower half-plane. Complete real blow-up, however, prevents these conjugate complex analytic branches to coincide, at real time overlap $t>T+\delta$. \\
\begin{figure}
\centering \includegraphics[angle=0,scale=1.1]{figs/REPORT_SFB647_Fig13_RectangularTimePath.pdf}
\caption{\emph{The complex time domain in which we consider the continuation $\Gamma^\pm_p$, $\Gamma^\pm_c$ of blow-up solutions. The dot indicates the blow-up time $ T$. After $t=T$ we introduce a cut along the positive real axis, drawn as a bold line.}}\label{fig:complex_time_domain}
\end{figure}
Traveling pulse waves $u=U(x-ct)$ with wave speed $c$, and self-similar solutions, open a second approach to blow-up in complex time. Specifically, they correspond to ODE solutions which are homoclinic to zero. The standing wave $u(t,x)=U(x)=-6/x^2$ with wave speed $c=0$, for example, is an explicit solution of~\eqref{C8:eq:nonlinear_heat}. In general, homoclinic traveling waves $u=U(\psi)$, $\psi=x-ct $, possess local expansions in $\psi$ and $\log\psi$. \\
For the branched real blow-up on the complex unstable manifold $W^u_*$ of $u^*$, matters are not quite that simple. Indeed a "center manifold", rather than a fully hyperbolic structure, appears in rescaled variables. This heralds essential singularities beyond $\log t$, and sectors with exponentially flat correction terms. For further details see the dissertation~\cite{Stuke16}. \\
\subsection{The tumbling universe}
Arguably, the big-bang is the largest imaginable blow-up. For us it is the initial singularity of a purely gravitational cosmological model. We briefly describe the general setting, and the exact reduction to ODE models of Bianchi class. In the next section we present our own results. \\
Cosmological models of general relativity are solutions of the full Einstein equations
\begin{equation}\label{C8:eq:Einstein}
\mathrm{Ric}(g) - \tfrac{1}{2} \mathrm{Scal}(g) g \;\;=\;\; T\,,
\end{equation}
which relate the geometry of spacetime -- the curvature of a 4-dimensional Lorentzian metric $g$ -- to the matter content encoded in the stress-energy tensor $T$, with or without a cosmological constant. The Einstein PDEs are of hyperbolic type and are usually coupled to kinetic equations of matter. A complete understanding of the full PDE system~\eqref{C8:eq:Einstein} is currently out of reach. Therefore, additional symmetries are often assumed to discuss special solutions, and their perturbations. \\
The simplest model -- and core of the standard cosmological model still in use -- is the Friedmann model of spatially homogeneous and isotropic space-time. This assumption of a 6-dimensional symmetry group allows a reduction of~\eqref{C8:eq:Einstein} to a single scalar ODE that determines the expansion rate of the universe. This expansion rate can be compared with measurements of the Hubble constant and with the consequences of large-scale thermodynamics of the matter part. \\
The second simplest class are \emph{Bianchi models} of spatially homogeneous but anisotropic spacetime. In other words, the spacetime is assumed to be foliated into spatial hypersurfaces given by the orbits of a three-dimensional symmetry group. Belinskii, Khalatnikov, and Lifschitz (BKL) described the dynamics of cosmological models near the big-bang singularity as vacuum dominated, local, and oscillatory~\cite{BelinskiiKhalatnikovLifshitz1970-OscillatoryApproach}. This suggests to approximate the early universe by the dynamics of homogeneous, but not necessarily isotropic, vacuum spacetime. \\
The Einstein system~\eqref{C8:eq:Einstein} with a non-tilted, perfect-fluid matter model can be reduced to a 5-dimensional ODE system in expansion-reduced variables,
\begin{equation}\label{C8:eq:Bianchi}
\begin{array}{rcl}
N_1' &=& ( q - 4 \Sigma_+ ) N_1\,,
\\ N_2' &=& ( q + 2 \Sigma_+ + 2\sqrt{3}\Sigma_- ) N_2\,,
\\ N_3' &=& ( q + 2 \Sigma_+ - 2\sqrt{3}\Sigma_- ) N_3\,,
\\ \Sigma_\pm' &=& (q-2) \Sigma_\pm - 3 S_\pm\,,
\end{array}
\end{equation}
with the abbreviations
\begin{equation}\label{C8:eq:BianchiAbbrv}
\begin{array}{rcl}
S_+ &=& \tfrac{1}{2}\left(
\left( N_2 - N_3 \right)^2
- N_1 \left( 2 N_1 - N_2 - N_3 \right)
\right)\,,
\\ S_- &=& \tfrac{1}{2}\sqrt{3}
\left( N_3 - N_2 \right) \left( N_1 - N_2 - N_3 \right)\,,
\\ q &=& 2 \left( \Sigma_+^2 + \Sigma_-^2 \right)
+ \tfrac{1}{2}(3\gamma-2)\Omega\,,
\\ \Omega &=& 1 - \Sigma_+^2 - \Sigma_-^2 - K\,,
\\ K &=& \frac{3}{4} \Big( N_1^2 + N_2^2 + N_3^2
- 2\left( N_1 N_2 + N_2 N_3 + N_3 N_1 \right)
\Big)\,.
\end{array}
\end{equation}
This system is due to Wainwright and Hsu~\cite{WainwrightHsu1989-BianchiA}. The initial big-bang singularity is approached in the rescaled time limit $t\to-\infty$. The variables $N_k$ describe the curvature of the spatial hypersurfaces. Their signs determine the Lie-algebra type of the associated spatial symmetry imposed by the homogeneity assumption. Due to Bianchi's classification of three-dimensional Lie algebras -- the tangent spaces to the assumed symmetry group -- these models are called Bianchi models, although they have been introduced by G{\"o}del~\cite{Goedel1949-Cosmology} and Taub~\cite{Taub1951-SpaceTimes}. The shear variables $\Sigma_\pm$ relate to the rescaled eigenvalues of the second fundamental form of the spatial hypersurfaces. The matter density $\Omega$ is nonnegative, and the vacuum boundary $\Omega=0$ is invariant. The coefficient $\gamma<2$ determines the equation of state of the perfect fluid, e.g.\ $\gamma=4/3$ for radiation and $\gamma=1$ for dust. See also~\cite{WainwrightEllis1997-Cosmology} for further details on this dynamics approach to cosmology, and~\cite{HeinzleUggla2009-Mixmaster} for a survey on Bianchi models and open questions. \\
The most prominent features of system~\eqref{C8:eq:Bianchi} are the \emph{Kasner circle} $\mathcal{K}$ of equilibria,
\begin{equation}\label{C8:eq:Kasner}
\mathcal{K} \;=\; \{\; \Sigma_+^2+\Sigma_-^2=1, \; N_1=N_2=N_3=0 \;\}\,,
\end{equation}
and the invariant \emph{Kasner caps}
\begin{equation}\label{C8:eq:Caps}
\mathcal{H}_k^\pm \;=\; \{\; \Sigma_+^2+\Sigma_-^2=1-N_k^2, \;
\pm N_k > 0, \; N_{k+1}=N_{k-1}=0, \; k \;\mathrm{mod}\; 3 \;\}\,.
\end{equation}
\begin{figure}
\centering
\includegraphics[angle=0,scale=1.1]{figs/REPORT_SFB647_Fig11_KasnerCap.pdf}
\caption{\emph{Heteroclinic cap of vacuum Bianchi II solutions to the Kasner circle $\mathcal{K}$. Remaining caps are given by $2\pi/3$ rotations over $2\pi/3$ in $\Sigma_\pm$ and cyclic permutations of $\{N_1, N_2, N_3\}$.}}
\label{C8:fig:Caps}
\end{figure}
The Kasner caps are filled with heteroclinic orbits between equilibria on $\mathcal{K}$; see Fig.~\ref{C8:fig:Caps}. Equilibria of~\eqref{C8:eq:Bianchi} indicate self-similar blow-up in the big-bang limit which occurs at rescaled time $t\to -\infty$. The projections of the heteroclinic orbits to the $\Sigma$-plane lie on straight lines through the corners of a circumscribed equilateral triangle; see Fig.~\ref{C8:fig:Kasner}. In reversed physical time, these sequences define the \emph{Kasner map} as an expansive map from the Kasner circle onto its double cover; see Fig.~\ref{C8:fig:Kasner} and Section~\ref{subsec:kasnerdynamics}. \\
\begin{figure}
\centering
\includegraphics[angle=0,scale=1.0]{figs/REPORT_SFB647_Fig12_KasnerMap.pdf}
\caption{\emph{The Kasner map and the three Taub points $T_1, T_2, T_3$ given by $N_1=N_2=N_3=0$, $(\Sigma_+ + i \Sigma_-)^3=-1$.}}
\label{C8:fig:Kasner}
\end{figure}
Concatenation of the heteroclinic orbits in the Kasner caps, i.e.\ iteration of the Kasner map, defines a subshift of finite type. Misner~\cite{Misner1969-Mixmaster} introduced the name \emph{Mixmaster} for the heuristic picture of generic trajectories following these formal heteroclinic sequences close to the big-bang singularity: the universe tumbles between various close-to self-similar solutions given by the various Kasner equilibria. \\
A crucial question is the \emph{locality} of the BKL/Mixmaster picture. In the (non-vacuum) Friedmann model, the particle horizon (backward light cone) remains finite in expansion reduced variables, close to the big-bang singularity. This bounded particle horizon also bounds the domain of dependence by particle interaction, and therefore represents locality. In the Bianchi model~\eqref{C8:eq:Bianchi}, the particle horizon corresponds to the \emph{Mixmaster integral}
\begin{equation}\label{C8:eq:MixIntegral}
I \;\;=\;\; \int_{-\infty}^0(\sqrt{N_1N_2}+\sqrt{N_2N_3}+\sqrt{N_1N_3}) \mathrm{d}t\,,
\end{equation}
see~\cite{HeinzleRingstroem2009-FutureBianchiVI}. Locality holds true, if and only if this integral remains finite. Obvious counterexamples are the
solutions on the line of Taub equilibria
\begin{equation}
N_1=0\ne N_2=N_3\,,\qquad \Sigma_-=0\,,\quad\Sigma_+=-1\,,\qquad \Omega=0\,,
\end{equation}
where the Mixmaster integral is $+\infty$. It is currently not known, whether nontrivial counterexamples with $I=\infty$ exist. The first nontrivial positive examples with $I<\infty$ are quite recent~\cite{LiebscherHaerterichWebsterGeorgi2011-BianchiA}. More precisely, the longstanding BKL/Mixmaster conjecture therefore claims that $I<\infty$ for generic initial conditions $N_j(0)$, $\Sigma_{\pm}(0)$ in the respective Bianchi class~\cite{BelinskiiKhalatnikovLifshitz1970-OscillatoryApproach,HeinzleRingstroem2009-FutureBianchiVI}. Here \emph{genericity} can be either understood as \emph{residual}, in the topological sense of Baire category, or as \emph{almost everywhere} with respect to Lebesgue measure. In the Bianchi type IX of class A, where all $N_j>0$, Ringstr\"om~\cite{Ringstroem2001-BianchiIX} has shown Baire and Lebesgue generic convergence to the Kasner caps. Of course, the Mixmaster integral $I$ vanishes on the Kasner caps. The mere generic convergence to the Kasner caps, however, does not imply generic finiteness of $I$ if this convergence is too slow.
\subsection{Kasner dynamics}\label{subsec:kasnerdynamics}
In the vacuum model~\eqref{C8:eq:Bianchi}, $\Omega=0$, each non-Taub Kasner equilibrium possesses one trivial eigenvalue zero of the linearization in the direction of the Kasner circle and three nontrivial eigenvalues in the remaining directions of the three heteroclinic caps. At the Taub points, two nontrivial eigenvalues become zero and a \emph{bifurcation without parameters} occurs, in the sense of the habilitation thesis~\cite{Liebscher15-Bifurcation} of Stefan~Liebscher. \\
To clarify the relation of generic trajectories to formal heteroclinic sequences, we ask for the set of points converging to the heteroclinic sequence attached to a given Kasner equilibrium. This is a difficult question because each heteroclinic step, by itself, already consumes infinite rescaled time $t$. We might hope for an invariant codimension-one foliation transverse to the Kasner circle, which in particular contains the local strong unstable and strong stable invariant manifolds of each Kasner equilibrium. Each fiber would then become the stable set of the Kasner equilibrium, in backwards time $t \to -\infty$, as the big-bang singularity is approached. \\
However, there are topological obstructions: the sets of initial conditions converging to the Taub points, backwards, were classified in~\cite{Ringstroem2001-BianchiIX} and form manifolds of codimension two. Thus, if the heteroclinic sequence of a Kasner equilibrium terminates at a Taub point, this equilibrium cannot possess an attached stable fiber of codimension one. In particular, any transverse backwards stable foliation has to miss at least a dense subset of the Kasner circle. \\
Stable foliations along Kasner sequences away from Taub points have first been addressed by Liebscher et al.\ in~\cite{LiebscherHaerterichWebsterGeorgi2011-BianchiA}, as follows. Consider a Kasner equilibrium such that its attached (unique, backwards) heteroclinic sequence does not accumulate at a Taub point. Then this equilibrium possesses a unique, locally invariant, codimension-one, stable Lipschitz leaf. The leaves of equilibria in the same heteroclinic sequence form a backwards invariant continuous foliation. The distance in which the trajectories pass the equilibria in this foliation, tends to zero exponentially. \\
The proof basically consists of three parts. First, a suitable eigenvalue property guarantees an arbitrarily strong contraction, transversely to the Kasner circle, and an arbitrarily small drift along the Kasner circle, locally. Second, the global passage along a heteroclinic orbit in the Kasner caps is modelled by a diffeomorphism close to the Kasner map. This provides a bounded expansion along the Kasner circle, away from Taub points. Third, the combination of both maps yields a hyperbolic map between suitably chosen cross sections near the Kasner circle: contraction transverse to the Kasner circle is inherited from the local passage, expansion along the Kasner circle is inherited from the global passage. The stable leaf is then obtained as the fixed point of a graph transform, for the above construction. \\
In~\cite{LiebscherRendalTchapnda2012-BianchiMaxwell}, Liebscher et al.\ generalize this approach to weaker eigenvalue conditions. This applies to non-vacuum Bianchi systems with arbitrary ideal fluids, and to Einstein--Maxwell equations of Bianchi type VI${}_0$. In the Einstein--Maxwell system one of the heteroclinic caps is no longer induced by the geometry but rather by the dynamics of the magnetic field. \\
An alternative approach in~\cite{Beguin2010-BianchiAsymptotics}, and the dissertation~\cite{Buchner2013-Dissertation} by Buchner, utilize Takens linearization~\cite{Takes1971-Linearization} to study the local passage. Indeed, after linearization, contraction and absence of drift follow trivially. However, the linearization requires strong spectral non-resonance conditions. Non-resonance imposes severe limitations on this approach and on possible generalizations. In~\cite{Beguin2010-BianchiAsymptotics}, for example, all formal sequences which are periodic or accumulate to periodic sequences (or the Taub points) are excluded to avoid resonances. In~\cite{Buchner2013-Dissertation}, Buchner has studied the non-resonance conditions more carefully, and certain periodic sequences are now admissible. His results on local passages extend to Bianchi class-B systems. \\
The techniques described above require to avoid the Taub points. This is their main restriction. In fact, the Kasner equilibria which, under iteration of the Kasner map, do not accumulate at Taub points form an (albeit uncountable) meager set of measure zero. They are not generic, in any sense, but possess a finite Mixmaster integral~\eqref{C8:eq:MixIntegral} and thus a finite particle horizon. \\
Brehm, in his dissertation~\cite{Brehm16}, is the first to answer the BKL/Mixmaster question in the Lebesgue sense. This problem had remained open for more than four decades. Brehm obtains two new results in the study of the Wainwright--Hsu system~\eqref{C8:eq:Bianchi},~\eqref{C8:eq:BianchiAbbrv} for the vacuum case $\Omega=0$.The first result extends Ringstr\"om's attractor result to Bianchi type \textsc{VIII} vacuum solutions. The second result asserts finiteness $I<\infty$ of the Mixmaster integral~\eqref{C8:eq:MixIntegral} for Lebesgue almost all initial conditions. \\
Consider the Bianchi type \textsc{IX} first, where all $N_j>0$ initially (and then for all times). The \emph{Taub space} $\{N_2=N_3,\, \Sigma_-=0\}$ is time invariant, together with its two rotated ``cousins'' given by cyclic permutations of the indices $\{1,2,3\}$. Ringstr\"om~\cite{Ringstroem2001-BianchiIX} has shown convergence to the Kasner caps, for initial conditions outside the Taub spaces, in the time-rescaled big-bang limit $t\to-\infty$. The proof proceeded by averaging over rotations around the invariant Taub-spaces, which are known explicitly. \\
The Bianchi type \textsc{VIII} is defined by $N_2,N_3>0>N_1$ (up to index permutations). The above Taub-space remains invariant, but its two cousins with $N_1=N_2$ and $N_1=N_3$ are now incompatible with $N_2,N_3>0>N_1$ and have disappeared. Even worse: all solutions with $N_3=0$ and $N_2>0>N_1$ converge to the Kasner-cicle $\mathcal K$ in \emph{both} time-directions $t\to\pm\infty$ and fail to lie on the Kasner-caps. This poses a serious obstacle to the attractivity of the caps. For these reasons, attractor results have remained elusive in the Bianchi \textsc{VIII} setting. \\
For both Bianchi \textsc{VIII} ($N_2,N_3>0>N_1$) and Bianchi \textsc{IX} ($N_1,N_2,N_3>0$), the attractor problem for Bianchi class A vacuum ($\Omega=0$) has been settled by Brehm in his dissertation~\cite{Brehm16} as follows. The dynamics in the big-bang limit $t\to-\infty$ falls into one of the following three mutually exclusive classes:
\begin{itemize}
\item [$(a)$] The solution converges to the three Kasner caps and has at least one $\alpha$-limit point on the Kasner-circle, which is distinct from the three Taub-points $T_j$ of Fig.~\ref{C8:fig:Kasner} and their antipodal points $Q_j=-T_j$.
\item [$(b)$] The solution is contained in the invariant Taub subspace $\{N_2=N_3,\, \Sigma_-=0\}$, or, in the case of Bianchi \textsc{IX}, one of its two cousins.
\item[$(c_2)$] The solution has exactly two $\alpha$-limit points on the Kasner-circle, which are $T_2$ and $Q_2=-T_2$. It has $\limsup |N_1N_3|>0 = \liminf|N_1N_3|$, for $t\to-\infty$, and $\lim|N_1N_2|=\lim|N_2N_3|=0$.
\item[$(c_3)$] The analogon of $(c_2)$, with the indices $2$ and $3$ interchanged.
\end{itemize}
Case $(a)$ applies for an open set of initial conditions which are a neighborhood of the three caps minus the three Taub spaces. Furthermore, for any solution in case $(a)$, the following non-Mixmaster integral $J$ remains finite:
\begin{equation}\label{C8:eq:brehm:weak-integral}
J=\int_{-\infty}^0 \left(|N_1N_2|+|N_2N_3|+|N_3N_1|\right)\mathrm{d} t < \infty\,.
\end{equation}
The cases $(c_2)$ and $(c_3)$ are impossible in the case of Bianchi \textsc{IX}; whether they can occur at all is currently unknown. Taken together with suitable measure-theoretic estimates, one can see that case $(a)$ applies in a set of initial conditions, both of fat Baire category and of full Lebesgue measure. In other words $(b),(c_2),(c_3)$ occur for a meager set of Lebesgue-measure zero, only. The estimate~\eqref{C8:eq:brehm:weak-integral} for the non-Mixmaster integral $J$ is novel also in the Bianchi \textsc{IX} case. It is insufficient to bound the Mixmaster integral $I$ of~\eqref{C8:eq:MixIntegral}, but nevertheless promising: from $\lim_{t\to-\infty}|N_iN_j|= 0$, to $\int_{-\infty}^0|N_iN_j|\mathrm{d} t <\infty$, and finally towards the goal $\int^0_{-\infty}\sqrt{|N_iN_j|}\mathrm{d} t <\infty$.
The proof in Bianchi \textsc{IX}, where only the $J<\infty$ estimate~\eqref{C8:eq:brehm:weak-integral} is novel compared to~\cite{Ringstroem2001-BianchiIX}, proceeds along the lines of Ringstr\"om, via a refined quantitative averaging over rotations around the three invariant Taub subspaces combined with a more careful gluing of the different averaging regimes.
In the Bianchi \textsc{VIII} case, the two missing invariant Taub subspaces are replaced by $\{|N_1|=|N_3|,\, \Sigma_-=\sqrt{3}\Sigma_+\}$ and its index-exchanged cousin. These two replacement Taub subspaces are invariant only up to leading order; this suffices for the averaging arguments and yields the trichotomy $(a)$, $(b)$, $(c)$.
The second main result in~\cite{Brehm16} establishes boundedness $I<\infty$ of the Mixmaster integral~\eqref{C8:eq:MixIntegral}, for Lebesgue almost every initial condition in Bianchi \textsc{VIII} and \textsc{IX} vacuum. In other words, particle horizons form towards the singularity and the blow-up is local in the sense of BKL. \\
The above considerations reside within the general relativity (GR) framework of a purely gravitational Einstein universe. A larger framework is provided by the Ho\v{r}ava-Lifshitz models (HL). HL gravity has been proposed as a candidate for the theory of quantum gravity; see~\cite{Muko} and the references there. The motivation of Hell to study those models in~\cite{HellUggla}, with Uggla, is the following. HL gravity breaks relativistic first principles and introduces anisotropic scalings between space and time. How do such modifications of the first principles affect the initial singularity and the structure of the BKL conjecture? As an attempt to answer this question, Hell and Uggla address the question of the dynamics on its first level of complexity, i.e.\ the discrete dynamics contained in the heteroclinic chains arising in HL. Similarly to the Bianchi models for general relativity, it is possible to formulate HL models as a system of ODEs with constraints that admit a circle of equilibria, which we call the Kasner circle by analogy. It corresponds to the Bianchi type I model for HL. On the next level of the symmetry hierarchy, Bianchi type II, we find three invariant hemispheres of heteroclinic orbits that intersect only at their boundary, the Kasner circle, for HL and GR alike. On each hemisphere, the projections of the heteroclinic orbits to the plane of the Kasner circle emanate from a single point, as a straightforward calculation shows\cite{HellUggla}. Those three points of emanation form an equilateral triangle whose center coincide with the center of the circle of equilibria: for the HL parameter $v=1/2$ corresponding to GR, the Kasner circle of equilibria is inscribed in the triangle, while this is not the case any more when the parameter changes to $v\neq 1/2$; see Fig.~\ref{HLKasnerdyn}. Concatenation of heteroclinic orbits on different hemispheres form heteroclinic chains, just as they did in the GR case. However, the discrete dynamics on the Kasner circle induced by those heteroclinic chains change. \\
\begin{figure}
\begin{center}
\includegraphics[scale=0.8]{figs/REPORT_SFB647_Fig14_Overlapping+stablearcs.pdf}
\end{center}
\caption{\emph{The Kasner circle of equilibria, the three expanding arcs and the three points of emanation for the HL models. Left $0<v<1/2$; the dashed circle zooms into one of the three overlapping regions, where the overlapping sectors are indicated by dashed lines. Right $1/2<v<1$.}}
\label{HLKasnerdyn}
\end{figure}
The BKL conjecture for GR describes the asymptotic behavior towards the big-bang singularity. This conjecture can be seen as an extrapolation from the dynamics of the heteroclinic chains to the continuous dynamics near the invariant hemispheres. Studying the discrete dynamics for the HL Bianchi models reveals how the BKL conjecture should be modified in this context. \\
Let us focus on the discrete dynamics induced by the heteroclinic chains, which map each equilibrium of the Kasner circle to the end of the heteroclinic leaving from it. Each point of emanation expands an arc of the Kasner circle enclosed by the two tangents to the circle through this point, by this HL Kasner map. Varying $v$ amounts to changing the distance between the points of emanation of the heteroclinics and the circle of equilibria. This changes the size of the three expanding arcs and the factor of expansion; see Fig.~\ref{HLKasnerdyn}. Two very different behaviors appear, depending on whether $v$ is below or above $1/2$. \\
Let us first discuss the case $1/2<v<1$. As can be seen in Fig.~\ref{HLKasnerdyn}, right, the three expanding arcs are separated by three arcs of stable fix points at which heteroclinic chains end, due to the lack of heteroclinic orbits leaving from there. As proved in~\cite{HellUggla}, generic initial conditions (both in the sense of Baire and Lebesgue) on the Kasner circle will lead to finite heteroclinic chains that end in a stable arc. Hence an analogue to the BKL conjecture should state that the asymptotic behavior of the system is, in general, convergence to a single equilibrium in a stable arc. This behavior is very different from the GR case where oscillatory chaos is present in the asymptotic behavior towards the big-bang singularity. However, by~\cite{HellUggla}, not all initial conditions on the Kasner circle lead to finite heteroclinic chains. An invariant Cantor set of measure zero leads to heteroclinic chains that never enter the stable arcs, and the discrete dynamics on this Cantor set is chaotic. The proofs of these facts use symbolic dynamics and equivalence to a subshift of finite type, known to be chaotic. Hence chaos has not completely vanished from the system for $1/2<v<1$, but it is not generic. \\
Let us now discuss the opposite case $0<v<1/2$. We see from Fig.~\ref{HLKasnerdyn}, left, how the three expanding arcs now overlap, pairwise, while no stable arc is present. Therefore the continuation of a heteroclinic chain that hits the overlap region is not unique: there are two heteroclinics leaving from there. In other words, the map describing the discrete dynamics is not well-defined and the concept of iterated function systems (IFS) comes into play -- details can be found in~\cite{HellHL} by Hell. The IFS we consider is a collection of maps whose domains of definition cover the Kasner circle. The notion of chaotic discrete dynamical systems is generalized to IFS using the Hausdorff distance between sets. Iterations of a well-defined map are said to be chaotic if three conditions are fulfilled: sensitivity to initial conditions, topological mixing, and density of periodic orbits. In the case of an IFS, a single initial condition generates several trajectories. Therefore each iteration of an IFS is a set instead of a single state. The three above conditions for chaos translate to IFS when distance between iterations is translated to Hausdorff distance between sets of possible iterations. The IFS arising for the HL Bianchi models with $0<v<1/2$ are chaotic in this sense, mostly due to the expansion property. Roughly speaking, this means that there are chaotic realizations of the IFS. This property only concerns the heteroclinic chains. The nearby flow will be investigated elsewhere. Even if the HL models with $0<v<1/2$ are chaotic, like the GR case $v=1/2$, their chaotic structure is quite different. The concept of a BKL era (i.e.\ phases of oscillations between two neighboring expanding arcs) and the growth of their length along heteroclinic chains shape the chaos in the GR case, via the continued fraction expansion. In the HL for $0<v<1/2$, BKL erae do not play such a role: their lengths are bounded above, and the fine structure associated to their rate of growth disappears. The value $v=1/2$ associated to GR can be seen as a critical transitional value where the dynamics bifurcates from nongeneric chaos to generic chaos. The importance of erae is a feature that does not survive below $v=1/2$. Hence the perturbation of GR along the one-parameter family of HL will lead to a breakdown of the BKL conjecture. \\
This concludes our extended survey of some global dynamics aspects, from geometric and algebraic aspects of parabolic global attractors and their blow-up to black hole data and exact chaotic solutions of the Einstein equations in the setting of Bianchi cosmology. Not only many of the particular problems but, much broader, our overall perspective on these problems we owe, deeply, to the unique framework provided by the Sonderforschungsbereich 647 ``Space -- Time -- Matter'' of the Deutsche Forschungsgemeinschaft.
\bibliographystyle{amsplain}
| d494c7cccf60f672bc7721b299750014d5f1d613 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
In the past decade, a new field dubbed topological insulator (TI)\cite{hasan2010colloquium,qi2011topological,kane2005z,bernevig2006quantum} has caused enormous attentions in the condensed matter physics. Different from the earliest topological states in quantum Hall effect\cite{klitzing1980new,thouless1982quantized} which requires a strong magnetic field, TI is time-reversal-symmetry protected, and can be characterized by the $Z_2$ topological invariant\cite{kane2005z,moore2007topological,fu2007topological,fu2007topological2}. In TIs, the bulk insulating states are accompanied by metallic helical Dirac-like boundary states which are related to its bulk topology and protected by the time-reversal symmetry. With the novel properties of the edge states, such as high mobility, absence of backscattering and spin-polarized conductivity channels, TIs provide a promising platform for realizing new electronics and spintronics applications. The discovery of TIs also results in the realization of quantum anomalous Hall effect\cite{yu2010quantized,chang2013experimental}. Moreover, the proximity effect between TI and superconductor hosts non-Abelian Majorana fermion and opens a new venue for topological quantum computations\cite{fu2008superconducting}. On the materials front, all TI materials are narrow-gap semiconductors with inverted band gap at an odd number of time-reversal-invariant momenta (TRIMs)\cite{fu2007topological,fu2007topological2}. Two-dimensional (2D) TIs (also known as quantum spin Hall states) range from HgTe/CdTe\cite{bernevig2006quantum,konig2007quantum} and InAs/GaSb\cite{liu2008quantum,knez2011evidence} quantum well structures to the layered honeycomb lattice materials such as silicene\cite{liu2011quantum,liu2011low}, germanene\cite{liu2011quantum,liu2011low}, stanene\cite{xu2013large,liu2011low,tang2014stable}, their halogenides\cite{xu2013large,si2014functionalized}, and ultrathin Bi films\cite{liu2011stable,wada2011localized}. The 3D TIs materials include Bi$_{1-x}$Sb$_x$ alloys\cite{fu2007topological2,hsieh2008topological}, Bi$_2$Se$_3$-class materials\cite{zhang2009topological,chen2009experimental,xia2009observation}, half-Heusler compounds\cite{chadov2010tunable,lin2010half}, TlBiSe$_2$ family chalcogenides\cite{sato2010direct,kuroda2010experimental,chen2010single}, strained HgTe\cite{fu2007topological2,brune2011quantum}, $\alpha$-Sn\cite{fu2007topological2,barfuss2013elemental}, and bismuth-based III-V semiconductors\cite{huang2014nontrivial}, etc.\cite{autes2015novel,huang2016topological,bansil2016colloquium} Additional ways could convert the normal insulators into TIs, such as external strain\cite{liu2014manipulating}, chemical doping\cite{shi2015converting}.
The discovery of TIs stimulates the identification and searching for more topological quantum states. In TIs, the two branches of helical edge/surface states are related to each other by time-reversal symmetry, and they are degenerate at the TRIMs due to the Kramers' theorem thus forming the Dirac points. In principle, degeneracies can also come from other types of symmetries, such as particle-hole symmetry, crystalline symmetry, etc. Therefore, finding new topological phases protected by other symmetries is a new hot topic in this rapidly developing field\cite{chiu2015classification}. In 2011, Fu\cite{fu2011topological} proposed that crystalline symmetry can protect new types of topological states, which are called topological crystalline insulators (TCIs). The first theoretically predicted and experimentally realized TCI materials are IV-VI semiconductors, with SnTe as a representative\cite{hsieh2012topological,tanaka2012experimental,dziawa2012topological,xu2012observation}. The symmetry responsible for their topological character is the mirror symmetry. Due to the simple crystal structure, SnTe-class TCI materials have caused extensive concerns. Later, some other TCI materials are theoretically proposed in transition metal oxides with a pyrochlore structure\cite{kargarian2013topological} and anti-perovskite materials\cite{hsieh2014topological}; both of these two classes of materials are also topologically protected by mirror symmetry. Additionally, crystalline symmetry has been extended from the point group to non-symmorphic space group and a new class of TCIs named topological non-symmorphic crystalline insulator is proposed\cite{liu2014topological}. Considering the richness and complexity of crystal operations, recently some researchers\cite{jadaun2013topological,slager2013space,dong2015classification} have made great efforts to classify the TCIs systematically. However, these proposals ask for the support of realistic materials, and to date the only TCI materials realized in experiments are the IV-VI semiconductors. In this article, we mainly review the recent research on the SnTe-class TCI materials, which may facilitate future researches to search for more TCI materials and particularly to explore their potential applications.
In the research field of topological materials, the theoretical calculations play an important role in the material predictions and provide a reliable guidance for the experimental observations\cite{bansil2016colloquium}. So in the following (Sections II and III), we will first describe the topological identification of the bulk and surface states properties of SnTe from a computational point of view, and then we introduce the experimental observations. Due to the crystalline symmetry protection and multiple Dirac surface states, TCIs have more tunable properties than TIs. In Sections IV to VI, we will demonstrate the electronic properties of TCIs under various perturbations, carrier types control or superconductivity by chemical doping, strain-tuned surface states and topological phase transition, and thickness-dependent confinement effects on 2D topological states. Some novel quantum transport properties, such as valley-selective filtering and helicity-resolved functionalities for Dirac fermions, are also discussed in Sections V and VI. Finally, we give the conclusions and outlook in Section VII.
\section{Bulk topology of SnTe class of materials}
SnTe-class IV-VI semiconductors have a rocksalt structure [Fig. 1(a)], and its Brillouin zone (BZ) is shown in Fig. 1(b). These materials are narrow-gap semiconductors, with their valence-band maximum (VBM) and conduction-band minimum (CBM) at four equivalent $L$ points of BZ. In the vicinity of $L$ point, the band structures of PbTe and SnTe with $p$ orbital projections of cation and anion based on first-principles calculations are presented in Figs. 1(c) and (d), respectively. It is seen that PbTe has a normal band ordering, i.e. The VBM is primarily derived from Te atoms and the CBM from Pb atoms, suggesting that PbTe can be smoothly connected to the atomic limit. While SnTe has an inverted band structure: the VBM is dominated by the Sn atoms and the CBM by Te atoms. The band inversion of SnTe relative to PbTe indicates that the relative ordering of $L_{6}^{+}$ and $L_{6}^{-}$ states is switched, as shown in Figs. 1(c) and (d). Nonetheless, it should be noted that there are four $L$ points in the whole BZ, leading to that band inversion occurs at an even number of points. So neither SnTe nor PbTe in the rocksalt structure is a TI with non-trivial $Z_2$ topological invariant.
\begin{figure}[ht]
\centering
\includegraphics[width=0.47\textwidth]{FIG1.jpg}
\caption{\label{fig:fig1} (Color online) (a) The crystal structure of SnTe, in which the green plane depicts one of \{110\} mirror planes. (b)The face-centered-cubic Brillouin zone, in which the $\Gamma L_1L_2$ plane is invariant under reflection about the (110) mirror plane. (c) and (d), the band structures of PbTe and SnTe with $p$ orbital projections of cation (red dots) and anion (blue dots). The calculations are performed with optimal lattice constants (6.56 {\AA} for PbTe and 6.40 {\AA} for SnTe). At $L$, the lowest-lying valence and conduction states associated with even and odd parity eigenvalues are labeled by $L_{6}^{+}$ and $L_{6}^{-}$, respectively.}
\end{figure}
However, the band inversions of SnTe can give new non-trivial topology if we take the crystalline symmetry into account. Face-centered-cubic structure has the mirror symmetry with respect to the \{110\} planes, one of which is shown by the green plane in Fig. 1(a), corresponding to the plane $\Gamma L_1L_2$ in momentum space [Fig. 1(b)]. The Hamiltonian of this system under mirror operation $M$ satisfies $MH(k_1,k_2,k_3)M^{-1}=H(-k_1,k_2,k_3)$, in which $k_1$ is along the direction perpendicular to the plane $\Gamma L_1L_2$, and $k_3$ is along the direction of $\Gamma L_1$. On plane $\Gamma L_1L_2$, crystal momenta have $k_1=0$, and the Hamiltonian commutate with the mirror operator, i.e., $[H,M]=0$. Then these two operator have the common eigenstates. The mirror operation satisfies $M^2=-1$ for spin 1/2 electrons, and its eigenvalues are $M=\pm i$. So all the eigenstates of Hamiltonian can be labeled by the eigenvalues $\pm i$ of $M$, i.e., $\Psi_{+i}$ and $\Psi_{-i}$. The Hamiltonian can be written as the form of block diagonalization,
$H=\left(\begin{smallmatrix}
H_{+i} & 0 \\
0 & H_{-i} \\
\end{smallmatrix}\right)$.
Following the definition of spin Chern number\cite{sheng2006quantum}, here each class of eigenstates in subspace has an associated Chern number $C_{\pm i}$. Though the total Chern number is zero, $C=C_{+i}+C_{-i}=0$, due to the time-reversal symmetry, the mirror Chern number defined as $C_M=(C_{+i}-C_{-i})/2$ can be nonzero\cite{teo2008surface}. The crystal may have several mirror planes, and each can define an associated mirror Chern number. A nonzero mirror Chern number defines a topological crystalline insulator protected by the mirror symmetry\cite{hsieh2012topological}.
Hsieh {\it et al.}\cite{hsieh2012topological} introduced a low-energy effective model near the $L$ point to discuss the different topology between PbTe and SnTe, and they found that the band inversion of SnTe at $L$ changes the mirror Chern number by one. Because there are two $L$ points ($L_1$ and $L_2$) in the $\Gamma L_1L_2$ plane and these two $L$ points have the same contributions to the mirror Chern number\cite{hsieh2012topological}, finally the band inversion changes the mirror Chern number for the $\Gamma L_1L_2$ plane by two. As a result, PbTe is topologically trivial, while SnTe is a topological crystalline insulator with the mirror Chern number $C_M=-2$. Actually, there are six equivalent \{110\} mirror planes for SnTe, and each possesses two $L$ points, so the mirror Chern number for every plane is $-2$.
\section{Surface states of SnTe-class TCIs}
The bulk-boundary correspondence indicates the existence of topological surface states. Since band inversion occurs at four $L$ points in bulk BZ of SnTe, there should be four Dirac cones in the surface BZ. However, the gapless surface states of TCIs require the underlying crystalline symmetry to be preserved on the boundary. So not all crystal surfaces of SnTe have gapless surface states; only those surfaces that keep the mirror symmetry with respect to the (110) planes of rocksalt structure are gapless. Three common surface terminations are (001), (111), and (110)\cite{hsieh2012topological,ando2015topological,liu2013two}, and they meet the corresponding symmetry. Interestingly, there are two types of TCI surface states with qualitatively different electronic properties\cite{liu2013two}, which depend on the surface orientation and are schematically shown in Fig. 2. The type-I surface states have the properties that their Dirac points are located at TRIMs, such as the (111) surface states; while the Dirac points of type-II surface states on (001) and (110) surfaces are deviated from TRIMs. The latter types of surface states exhibit a Lifshitz transition as a function of Fermi energy. The dependence of boundary states on surface orientations is an important property of TCIs. Next we will first introduce the (001) surface states, and then the (111) surface states.
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.3\textwidth]{FIG2.jpg}
\caption{\label{fig:fig2} (Color online) The projections of bulk BZ towards (001) and (111) surfaces of the rocksalt crystal structure, and the schematic locations of four Dirac cones in these two surface BZs.\cite{tanaka2013two}}
\end{center}
\end{figure}
\subsection{(001) surface states}
The first-principles calculated band structure of (001) surface of SnTe is shown in Fig. 3(a).\cite{hsieh2012topological} The red lines depict the topological surface band and they are of Dirac-like linear dispersion. Interestingly, the Dirac point is not located at the time-reversal invariant $\bar{X}$ point which is the projection of $L$ point in the bulk BZ, but has a little deviation along $\bar{\Gamma}\bar{X}$. As shown in Fig. 3(b), all four Dirac points in the whole surface BZ are deviated from $\bar{X}$, and they are related to each other by the $C_4$ rotation symmetry. Another important feature is that the change of Fermi surface topology exhibits a Lifshitz transition as a function of the Fermi energy. When the Fermi energy is slightly lower than the Dirac point, the Fermi surface consists of two disconnected hole pockets [red or orange elliptical region in Fig. 3(c)] on either side of the $\bar{X}$ point along $\bar{\Gamma}\bar{X}$; as the Fermi energy decreases, these two pockets enlarge and touch each other, then they reconnect to form a large hole pocket [green or blue region in Fig. 3(c)] and a small electron pocket [gray elliptical region in Fig. 3(c)], both centered at $\bar{X}$. In addition, this transition of Fermi surface is accompanied by a Van Hove singularity in the density of states arising from saddle points in the band structure.\cite{hsieh2012topological}
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.43\textwidth]{FIG3.jpg}
\caption{\label{fig:fig3} (Color online) The (001) surface states of SnTe. (a) Band structure along high-symmetry lines from first-principles calculation. Inset: band structure of $k\cdot p$ model of Eq. (\ref{eq:H}) with 2D momentum near $\bar{X}$. (b) Fermi surface in the whole surface BZ. (c) A series of Fermi surfaces near the $\bar{X}$ point at different energies, displaying a Lifshitz transition.\cite{hsieh2012topological}}
\end{center}
\end{figure}
This type of TCI surface states can be understood as follows\cite{liu2013two,wang2013nontrivial}. From Fig. 2, we can see that two $L$ points ($L_1$ and $L_2$) in bulk BZ of SnTe are projected onto the same time-reversal invariant $\bar{X}$ point on the (001) surface. As a result of band inversions at these two $L$ points, two coexisting gapless Dirac cones will be created at $\bar{X}$. However, the sharp surface introduces intervalley scattering between these two $L$ points, thus producing the hybridization between the two massless surface Dirac fermions at the surface, and forming the novel surface states at $\bar{X}$. The essential properties of the (001) surface states can be captured by the following $k\cdot p$ Hamiltonian\cite{liu2013two,ando2015topological}:
\begin{equation}
H_{\bar{X}}(\textbf{k})=(v_xk_xs_y-v_yk_ys_x)\otimes I+m\tau_x+\delta s_x\tau_y.
\label{eq:H}
\end{equation}
Here, the first term describes two identical surface Dirac fermions associated with $L_1$ and $L_2$ (denoted by $\tau_z=\pm 1$); $\vec{s}$ is a set of Pauli matrices associated with the spin components; the second and third terms describe all possible intervalley hybridizations to zeroth order in $\textbf{k}$, which satisfy all the symmetries of the (001) surface.\cite{liu2013two,ando2015topological} The $m$ term induces the coupling between the two Dirac cones and produces an relative shift between them in energy; the $\delta$ term turns the band crossing of the two Dirac cones into an anticrossing via hybridization and opens a gap, which is, however, strictly forbidden along $\bar{\Gamma}\bar{X}$. Consequently, the bands are gapless along $\bar{\Gamma}\bar{X}$, but gapped along other directions. The band structure of $k\cdot p$ model of Eq. (\ref{eq:H}) with 2D momentum near $\bar{X}$ is shown as an inset in Fig. 3(a). The Hamiltonian above can perfectly repeat the first-principles electronic structure of (001) surface, such as the deviation of Dirac point from $\bar{X}$, the Lifshitz transition of Fermi surface, and the Van Hove singularity in the density of states. The Lifshitz transition is interesting, because it would accompany a marked change in the Dirac-carrier properties and provide another ingredient in the physics of topological materials.
\begin{figure*}[ht]
\begin{center}
\includegraphics[width=0.75\textwidth]{FIG4.jpg}
\caption{\label{fig:fig4} (Color online) The ARPES measurements of the (001) surface states of SnTe-class materials. (a) ARPES intensity for SnTe along the cuts A--E in the surface BZ and mapping in a 2D wave-vector plane at various $E_{\rm B}$.\cite{tanaka2012experimental} (b) The temperature dependence of the ARPES data for the (001) surface of Pb$_{0.77}$Sn$_{0.23}$Se monocrystals. They clearly show the evolution of the gapped surface states (for $T\geq 100$ K) into the Dirac-like states on lowering the temperature ($T=9$ K).\cite{dziawa2012topological} (c) Up panel: ARPES iso-energetic contour mapping ($E_{\rm B}=0.02$ eV) of Pb$_{0.6}$Sn$_{0.4}$Te covering the first surface BZ. Down panel: ARPES dispersion map along the mirror line direction. Inset: Measured spin polarization profile is shown by the green and blue arrows on top of the ARPES iso-energetic contour at binding energy $E_{\rm B}=0.06$ eV.\cite{xu2012observation}}
\end{center}
\end{figure*}
As the cleavage plane, the (001) surface of SnTe and its family materials has been successfully prepared and the corresponding type-II surface states have been confirmed by experiments\cite{tanaka2012experimental,dziawa2012topological,xu2012observation}. Tanaka {\it et al.}\cite{tanaka2012experimental} grew the high-quality single crystal of SnTe, and observed the signature of a metallic Dirac-cone surface band by the angle-resolved photoemission spectrum (ARPES), with its Dirac point slightly away from the edge of the surface BZ [Fig. 4(a)]. In fact, because of the hole-doped nature of SnTe crystal, the Dirac point of surface states is not detected in their work and its energy is estimated to be 0.05 eV above $E_{\rm F}$ from a linear extrapolation of the two dispersion branches. In line with the theoretical prediction, they did not observe such gapless surface states in the cousin material PbTe. About the p-type conductivity of SnTe, we will explain its microscopic origin later in the article. For the Pb$_{1-x}$Sn$_x$Te and Pb$_{1-x}$Sn$_x$Se alloys, the chemical potential can easily be tuned to yield n-type or p-type conductivity by changing the crystals' growth condition, which makes them more suitable for experimental investigations on the TCI state. For example, Dziawa {\it et al.}\cite{dziawa2012topological} prepared the Pb$_{0.77}$Sn$_{0.23}$Se single crystal, and the ARPES of its (001) surface shows a n-type nature of this material. Importantly, they demonstrated that this alloy will undergo a temperature-driven topological phase transition from a trivial insulator to a TCI [Fig. 4(b)]. Xu {\it et al.}\cite{xu2012observation} utilized ARPES and spin-resolved ARPES to study the low-energy electronic structure below and above the band inversion topological transition of Pb$_{1-x}$Sn$_x$Te ($x=0.2$ and $x=0.4$). They demonstrated the observation of spin-polarized surface states in the inverted regime and their absence in the non-inverted regime, in which the ARPES iso-energetic contour mapping in the first surface BZ and spin-resolved topological Dirac surface states of Pb$_{0.6}$Sn$_{0.4}$Te are shown in Fig. 4(c). The interplay between topology and crystal symmetry is the character of TCIs. If the mirror symmetry of SnTe is broken, the gapless surface states will be gapped, i.e., the massless Dirac fermion will acquire mass. Okada {\it et al.}\cite{okada2013observation} reported high-resolution scanning tunneling microscopy studies of Pb$_{1-x}$Sn$_x$Se that reveal the coexistence of zero-mass Dirac fermions protected by crystal symmetry with massive Dirac fermions consistent with crystal symmetry breaking.
\subsection{(111) surface states}
Theoretically, the (111) surface states of SnTe are more simple than those of (001) surface. As shown in Fig. 2, four different $L$ points of bulk BZ are projected to different momenta in surface BZ, i.e., one $\bar{\Gamma}$ and three $\bar{M}$ points. There are no interactions between them. In fact, due to the polarity of (111) surface, it is not so easy to achieve the ideal (111) surface and observe its topological surface states in experiments. On the theoretical computations, there are only some tight-binding calculations at first without regard to the surface dangling bonds. Fig. 5 shows the surface band structure with two different terminations of SnTe (111) surface calculated by Liu {\it et al.}\cite{liu2013two} It is seen that Dirac-like surface band crossing (denoted by the red lines) occurs at $\bar{\Gamma}$ and $\bar{M}$ for both two terminations, whereas the Dirac points are close to the top (bottom) of the valence (conduction) band for Sn (Te) termination. The Dirac fermion at $\bar{\Gamma}$ is isotropic: the Fermi velocity along $\bar{\Gamma}\bar{M}$ is identical to that along $\bar{\Gamma}\bar{K}$; while at $\bar{M}$, the Dirac fermion is anisotropic: the Fermi velocity along $\bar{M}\bar{K}$ is larger than that along $\bar{M}\bar{\Gamma}$. The $k\cdot p$ Hamiltonians at $\bar{\Gamma}$ and $\bar{M}$ are given by $H_{\bar{\Gamma}}(\textbf{k})=v(k_1s_2-k_2s_1)$ and $H_{\bar{M}}(\textbf{k})=v_1k_1s_2-v_2k_2s_1$, where $k_1$ is along the $\bar{\Gamma}\bar{K}$ direction and $k_2$ is along the $\bar{\Gamma}\bar{M}$ direction.\cite{liu2013two}
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.47\textwidth]{FIG5.jpg}
\caption{\label{fig:fig5} (Color online) Band structures of the (111) surface for Sn and Te terminations. There are four Dirac pockets in the surface BZ: one at $\bar{\Gamma}$ and three at $\bar{M}$.\cite{liu2013two}}
\end{center}
\end{figure}
The ideal (111) surface of SnTe-class materials is a polar surface with unpaired surface electrons. In principle, the huge electrostatic potential induced by the dipole accumulation will cause the instability of the surface\cite{tasker1979stability}. Using first-principles calculations, Wang {\it et al.}\cite{wang2014structural} studied the (111) surface morphology and the associated electronic structures of SnTe under different growth conditions. They found that three different surfaces can stably exist, with the increasing of chemical potential of Sn: unreconstructed Te-terminated surface, $(\sqrt{3}\times\sqrt{3})$-reconstructed surface and $(2\times 1)$-reconstructed surface. The electronic structure of unreconstructed (111) surface shows that the nontrivial surface states are affected by the unpaired electrons, however clean type-I surface states can be achieved by hydrogen adsorption. Importantly, they revealed the critical effect of surface reconstruction on the topological surface states: the reconstruction will result in the folding of surface BZ and probably induce the breaking of the mirror symmetries, both of which can change the characteristics of topological surface states. For example, in Fig. 6(a,b), the $(2\times 1)$-reconstructed surface has the type-II surface states at $\bar{\Gamma}$: the Dirac point is deviated from TRIM. The $(2\times 1)$ reconstruction folds the surface BZ (SBZ) from hexagonal SBZ1 to rectangular SBZ2, causing the $\bar{M_1}$ point in SBZ1 to fold to the $\bar{\Gamma}$ point in SBZ2. As a result, two different $L$ points in the bulk BZ are projected to the same momentum $\bar{\Gamma}$ for reconstructed surface, just like the (001) surface. The additional intervalley scattering and the only one symmetric (110) mirror plane remaining create the type-II surface states at $\bar{\Gamma}$, which will also undergo a Lifshitz transition when tuning Fermi energy. Meanwhile the mirror symmetry breaking gaps the surface states at $\bar{M_2}$. In $(\sqrt{3}\times\sqrt{3})$-reconstructed surface, as shown in Fig. 6(c,d), the reconstruction only folds three different $\bar{M}$ in SBZ1 to three new $\bar{M^{\prime}}$ in SBZ3, so four Dirac cones still exist at four TRIMs belonging to type-I surface states. However, due to the different energy of Dirac points at $\bar{\Gamma}$ and $\bar{M}$, their conductive properties can be different when changing the Fermi energy.
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.48\textwidth]{FIG6.jpg}
\caption{\label{fig:fig6} (Color online) (a) Band structure of $(2\times 1)$-reconstructed (111) surface, where the surface states are denoted by red lines. Inset: surface atomic configuration, where only one of original three \{110\} mirror symmetries (indicated by yellow line) remains after reconstruction. (b) The surface Brillouin zones (SBZs): SBZ1 (blue hexagon) for original unreconstructed (111) surface and SBZ2 (red rectangle) for (2$\times$1)-reconstructed surface structure. The iso-energetic contour with energies at 5 meV (30 meV) above the Fermi energy is represented by red circle (green dotted circle and shaded area). (c) Band structure of $(\sqrt{3}\times\sqrt{3})$-reconstructed surface. Three mirror symmetry planes still exist (yellow lines in the inset). (d) SBZ3 and schematic iso-energetic contour with energy between the two Dirac points of $\bar{\Gamma}$ and $\rm{\bar{M}}$.\cite{wang2014structural}}
\end{center}
\end{figure}
In experiments, the observation of (111) surface states is truly more challenging owing to the difficulty in the preparation of non-cleavage surface. To date, epitaxial layer growth is an effective method for obtaining the (111) film. Fig. 7(a) and (b) show the ARPES measurements on the (111) surface of SnTe and Pb$_{1-x}$Sn$_x$Te performed by Tanaka {\it et al.}\cite{tanaka2013two} and Yan {\it et al.}\cite{yan2014experimental}, respectively. For the pure SnTe, both two experiments demonstrate the p-type conductivity, and the spectrum of surface states is covered by the signals of bulk valence band. Tanaka {\it et al.} utilized the second derivatives of the momentum distribution curves and changed the photon energies to recognize the linear-dispersion surface band. While Yan {\it et al.} changed the Pb/Sn ratio and found a topological phase transition in Pb$_{1-x}$Sn$_x$Te. A clear observation of the (111) surface states is done by Polley {\it et al.}\cite{polley2014observation} on the Pb$_{1-x}$Sn$_x$Se alloys. From Fig. 7(c), the Dirac-like band crossing is unchanged when varying the photo energy, revealing the character of surface band. They also observed that there is no relative binding energy difference between the $\bar{\Gamma}$ and $\bar{M}$ Dirac points. In addition, Wang {\it et al.}\cite{wang2015molecular} obtained the metastable rocksalt SnSe (111) films with the molecular beam epitaxy (MBE), and demonstrated that the epitaxial rocksalt SnSe is also a TCI through observing its (111) surface states.
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.46\textwidth]{FIG7.jpg}
\caption{\label{fig:fig7} (Color online) The ARPES measurements of the (111) surface states of SnTe-class materials. (a) Near-$E_{\rm F}$ ARPES intensity of SnTe measured along several cuts around the $\bar{\Gamma}$ point. Blue arrow indicates the top of the band.\cite{tanaka2013two} (b) ARPES intensity in the vicinity of the $\bar{\Gamma}$ point for SnTe, Pb$_{0.25}$Sn$_{0.75}$Te, Pb$_{0.7}$Sn$_{0.3}$Te, and PbTe, respectively.\cite{yan2014experimental} (c) Photon energy dependent ARPES measurements around the $\bar{\Gamma}$ point on a (111) oriented Pb$_{1-x}$Sn$_x$Se film. The bulk band is strongly dependent on the photon energy, while the position of the Dirac-like dispersion is unchanged when varying the photon energy, indicating the nature of surface states. However, its intensity is strongly modulated with the photon energy.\cite{polley2014observation}}
\end{center}
\end{figure}
\section{Native defects and doping in SnTe-class materials}
As mentioned above and observed by experiments, the as-grown SnTe crystals are usually heavily hole doped, which hinders the detection and investigations of the surface states and possible applications. Meanwhile, PbTe, the cousin material, can be either n or p type, depending on the growth conditions. Wang {\it et al.}\cite{wang2014microscopic} studied the role of native point defects in the doping behavior of these two materials and explained the microscopic origin of the p-type conductivity of SnTe.
As shown in Fig. 8(a), the dominant defect in SnTe is the negatively charged cation vacancy (V$_{\rm Sn}$$^{2-}$) under both Sn-rich and Te-rich conditions, and V$_{\rm Sn}$$^{2-}$ has a negative formation energy throughout the bulk gap. As a result, the Fermi level of SnTe has to be pushed under the VBM [i.e., a minus value of Fermi level in Fig. 8(a)] until the formation energy of V$_{\rm Sn}$$^{2-}$ becomes positive. So SnTe is always hole doped. While in PbTe [Fig. 8(b)], the dominant defect is not always the cation vacancy (V$_{\rm Pb}$$^{2-}$), but could be either anion or cation vacancy V$_{\rm Te}$$^{2+}$, V$_{\rm Pb}$$^{2-}$ or antisite Te$_{\rm Pb}$$^{2+}$, depending on the growth conditions. V$_{\rm Te}$$^{2+}$ and Te$_{\rm Pb}$$^{2+}$ are positively charged so that PbTe could be electron doped when these defects dominate.
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.47\textwidth]{FIG8.jpg}
\caption{\label{fig:fig8} (Color online) Formation energy as a function of Fermi level for native point defects in (a) SnTe and (b) PbTe under Sn/Pb rich and Te-rich conditions. The slope of each segment indicates the charge on the defect. Zero of the Fermi level is set to be the VBM of the host, and the Fermi level range spans the bulk gap.\cite{wang2014microscopic}}
\end{center}
\end{figure}
The large difference in the nature of defects between SnTe and PbTe originates from the different work functions or the distinction of the VBM of these two materials, as the formation energies of charged defects depend explicitly on the Fermi level referenced to the VBM. The VBM of SnTe is 0.5 eV higher than that of PbTe, inducing that SnTe has a lower formation energy of the negatively charged defects than PbTe. Meanwhile, alloying SnTe with Pb can continuously lower the VBM and CBM, so as to modify the conductivity type. So tuning the Pb/Sn ratio in Pb$_{1-x}$Sn$_x$Te alloy could obtain either n-type or p-type conductivity, which is a suitable way to realize bulk insulating and observe the topological surface states.
When doping the magnetism into thin films of TIs, the topological electronic states with a Chern number $\pm 1$ can be formed and the quantum anomalous Hall effect (QAHE) can be realized\cite{yu2010quantized}, which is experimentally observed in Cr doped Bi$_2$Te$_3$ class of TIs\cite{chang2013experimental}. Here in TCIs, Fang {\it et al.}\cite{fang2014large} proposed that when an out-of-plane ferromagnetic order is introduced into a TCI film, the QAHE with a tunable large Chern number can be produced. However, magnetic doping in SnTe-class TCIs is not easily achievable. In the early years, some experimental efforts have been made to study various types of magnetism in SnTe crystals doped with 3d transition metals (TMs), which are found to be degenerate magnetic semiconductor with small magnetic moments for these TM atoms\cite{inoue1981various}. These systems show complex magnetic properties: for example, doping with Mn, Cr and Fe act as ferromagnetic behavior, while doping with Co and Ni are magnetically ineffective\cite{inoue1981various}. More peculiarly, a strong effect of carrier concentration on the magnetic properties of these diluted magnetic semiconductors has been demonstrated, and it can be understood on the basis of the Ruderman-Kittel-Kasuya-Yosida (RKKY) interaction mechanism\cite{story1986carrier}. Recently, some calculations investigated the electronic and magnetic properties of V, Cr or Mn-doped SnTe\cite{liu2013magnetism}. However their doping concentration is very high, and it is likely to damage the TCI phase of SnTe. Another theoretical study comes to a conclusion that the doped TM atoms in SnTe have comparatively high formation energies, and predicts that the uniform TM doping with a higher concentration in SnTe will be difficult unless clustering\cite{wang2016defect}. To date, the effective magnetic doping in SnTe-class TCI materials is still a challenge, and the further investigations of magnetic effect on TCIs are required.
Another important doping is In-doped SnTe, which hosts the superconducting behavior\cite{erickson2009enhanced,sasaki2012odd}. In particular, Sn$_{1-x}$In$_x$Te with an indium content of $x=0.045$ preserving the topological surface states has been proved a topological superconducting state by the signals of surface Andreev bound states in the point-contact spectroscopy, and the critical temperature is 1.2 K\cite{sasaki2012odd,sato2013fermiology}. Furthermore it has recently been reported that the superconducting $T_{\rm C}$ of Sn$_{1-x}$In$_x$Te continues to increase with a higher level of doping, such as 4.5 K with In content of $x=0.45$\cite{novak2013unusual,zhong2013optimizing,polley2016observation}. However, the carrier density does not vary monotonically with Indium content. In fact at high Indium content, Indium has a mixed oxidation state in the system, i.e. it is neither In$^{1+}$ nor In$^{3+}$\cite{haldolaarachchige2015resonance}. So superconducting Sn$_{1-x}$In$_x$Te can not be viewed as a simple hole doped semiconducting material.
\begin{figure*}[bt]
\begin{center}
\includegraphics[width=0.8\textwidth]{FIG9.jpg}
\caption{\label{fig:fig9} (Color online) (a, b) The effect of [001]-oriented strain on the SnTe bulk. (a) BZs of SnTe for primitive cell (in black) and conventional cell along the [001] direction (in blue). (b) Band gap of bulk SnTe at the $R$ point as a function of in-plane biaxial strain. (c--f) The effect of [111]-oriented strain on the SnTe bulk. (c) Primitive cell (in black) and conventional cell along the [111] direction (in red). (d) BZs of SnTe for primitive cell (in black) and conventional cell along the [111] direction (in green). (e) and (f), Strain-dependent bulk band-gap evolution at the $A$ (red squares) and $L$ (blue circles) points under in-plane and out-of-plane strains, respectively. In (b), (e) and (f), the pink, light green, and white regions indicate the TCI, TI, and NI phases, respectively. Figures (c)--(f) are adapted with permission from Ref.~84, and copyrighted by the American Physical Society.}
\end{center}
\end{figure*}
\section{Strain effect}
Protected by the crystal symmetry and possessing multiple branches of Dirac surface states, TCIs have a much wider range of tunable electronic properties than TIs under various perturbations. Strain is an important controllable method\cite{si2016strain}. Not only can strain drive topological phase transition in the bulk system, but it can also modulate the electronic properties of the Dirac surface states.
\subsection{Strain effect on bulk: topological phase transition}
Strain is an effective way to tune the band gap of semiconductors. For the narrow-gap semiconductors whose VBM and CBM have the opposite parity or bonding types, the strain will influence their levels differently, and even induce the band inversion\cite{liu2014manipulating}. As has been proposed in Bi$_2$Se$_3$ family compounds, strain can induce the topological phase transition from a normal insulator (NI) to a TI and vice verse\cite{young2011theoretical,liu2011anisotropic,liu2014manipulating}. This manipulation can also be applied to TCIs. As mentioned by Hsieh {\it et al.}\cite{hsieh2012topological}, with the lattice constant decreasing, PbTe can be changed from a NI to a TCI. Such topological phase transition also occurs for other lead chalcogenides under external pressure or volume compression\cite{barone2013pressure,barone2013strain}; while for the volume expansion, SnS and SnSe can be transformed from an ambient pressure TCI to a topologically trivial insulator\cite{sun2013rocksalt}.
As has been known to us, there are four $L$ points in the bulk BZ; the isotropic strain can not distinguish them and will change the band orders of four valleys simultaneously. So the isotropic strain, such as hydrostatic pressure for the rocksalt structure, can only induce the topological phase transition between a NI and a TCI. However, an uniaxial or biaxial strain, which could be applied in the epitaxial films, can offer a diverse manipulation of the electronic properties. A systematic study of SnTe with various in-plane biaxial strains perpendicular to the [001] direction has been made by Qian {\it et al.}\cite{qian2014topological} Since the unit cell fitting for this type of strain is along the [001] direction, there is a folding of bulk BZ. As depicted in Fig. 9(a), four $L$ points in the original BZ (polyhedron encircled by black lines) are folded to $R$ and $R^{\prime}$ in the new BZ (blue tetragonal cell). The in-plane biaxial strain does not break the $C_4$ rotation, so $R$ and $R^{\prime}$ are still equivalent. The calculated band gap at $R$ as a function of biaxial strain is shown in Fig. 9(b). As the strain increases, the energy gap initially decreases, and then opens up again. Above 2.5$\%$ strain, band inversion disappears, and the system is transformed from a TCI to a NI.
A special strain which can distinguish the four $L$ points is the [111]-oriented strain. Zhao {\it et al.}\cite{zhao2015tuning} studied the band evolution of SnTe with the in-plane biaxial strain and out-of-plane uniaxial strain in the [111] direction. The unit cell and BZ are shown in Figs. 9(c) and (d). Due to the enlarged unit cell, the BZ shrinks from the face-centered-cubic case to a hexagonal prism. Consequently, the $L_1$ point in the original BZ folds to the $A$ point in the new BZ, and the band at $A$ will change differently from those at other three $L$ points under strain. The band-gap evolutions at the $A$ and $L$ points under the two types of strains (the in-plane biaxial strain and the out-of-plane uniaxial strain) are shown in Figs. 9(e) and (f), respectively. As the in-plane strain increases, though the band gaps at both $A$ and $L$ reduce to zero then reopen again, the gap closing will first occur at the $A$ point. So except for TCI and NI phases, there is a TI phase for the in-plane strain level between +2$\%$ and +2.5$\%$. For the out-of-plane strain, the band-gap closing and reopening only occur at the A point, so below +10$\%$, there are two topological phase: TCI and TI, where the TI phase could appear in the range of $+5 \% < \varepsilon_{\perp} < +10 \% $.
\subsection{Strain effect on (001) surface}
Here, maintaining the TCI phase, we discuss the effect of strain on the (001) surface states. The Dirac points on the (001) surface of SnTe are not pinned to TRIMs. A mechanical strain can shift the Dirac point positions in the momentum space just like an electromagnetic field acts on an electron\cite{tang2014strain}. Qian {\it et al.}\cite{qian2014topological} studied the strain-dependent electronic properties of SnTe (001) nanomembranes. Here we take the band structure of a 51-layer (001) membrane under the in-plane biaxial strain as an example. As shown in Fig. 10, the compressive strain drives the Dirac points of surface states to move away from the $\bar{X}$ points, while the tensile strain makes them close to the $\bar{X}$ points. In addition, the strain will influence the penetration length of the surface states like the case of TIs. Qian {\it et al.}\cite{qian2014topological} found that the compressive (tensile) strain can reduce (increase) the surface state penetration length, so the tensile strain will cause a hybridization gap for the SnTe thin film, such as the surface band for the 51-layer membrane under the biaxial strain of +1$\%$ [right panel in Fig. 10]. Such shift of Dirac point position in k space under strain has been realized in experiments\cite{zeljkovic2015strain}, which establishes a tunable platform for the strain-based applications.
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.47\textwidth]{FIG10.jpg}
\caption{\label{fig:fig10} (Color online) First-principles band structures of a 51-layer SnTe (001) nanomembrane under biaxial strain of -1$\%$, 0 and +1$\%$. Insets schematically indicate the shift of Dirac points in the surface BZ under strain.}
\end{center}
\end{figure}
The biaxial strain will equally shift the four Dirac points of the (001) surface states. However, a structural distortion with a relative displacement of atoms along the [110] direction on the (001) surface can break one of two mirror planes, and generate a mass to the Dirac cone. As a result, the two Dirac points along [$1\bar{1}0$] direction are gapped, but the other two along [110] remain gapless, which has been demonstrated by the STM measurements of the coexistence of zero-mass Dirac fermions with massive Dirac fermions\cite{okada2013observation}.
\subsection{Strain effect on (111) surface}
The Dirac surface states on the (111) surface are noninteracting and located at well-separated TRIMs (i.e., $\bar{\Gamma}$ and $\bar{M}$) in the surface BZ. Most importantly, differing from the valleys on the (001) surface, the $\bar{\Gamma}$ and $\bar{M}$ valleys appear highly distinct for crystal symmetry, and thus their band topology is sensitively differentiated by mechanical deformation as mentioned above. In addition, strain effect could also alter the valley-based landscape in the (111) surface.
\begin{figure*}[ht]
\begin{center}
\includegraphics[width=0.8\textwidth]{FIG11.jpg}
\caption{\label{fig:fig11} (Color online) (a--c) DFT-calculated surface band structures of SnTe (111) surface under no strain, in-plane biaxial strain of -2$\%$ and out-of-plane uniaxial strain of -2$\%$, respectively. The horizontal dotted lines denote the Fermi levels under different strains, and all bands have set the Fermi level in the unstrained system as the zero reference for energy. (d) Schematic of a strain-induced nanostructure in the SnTe (111) surface grown on a substrate. (e) Schematic of the energy spectra in the strain-induced nanostructure shown in (d). (f) and (g), Transmission probabilities $T(\theta)$ of massless Dirac fermions at the $\bar{\Gamma}$ and $\bar{M}$ valleys at the strain levels of $\varepsilon_{\perp}=-0.25{\%}$ and $\varepsilon_{\perp}=-1.75{\%}$, respectively.\cite{zhao2015tuning}}
\end{center}
\end{figure*}
Maintaining the TCI phase of the entire system, Zhao {\it et al.}\cite{zhao2015tuning} studied the strain-dependent Dirac valley evolution on the SnTe (111) surface under different strains. Figs. 11(a--c) show the DFT-calculated surface band structures under no strain, in-plane biaxial strain of -2$\%$ and out-of-plane uniaxial strain of -2$\%$, respectively. In the absence of strain [Fig. 11(a)], the $\bar{\Gamma}$ and $\bar{M}$ valleys are degenerate in energy; while the compressive strain drives the surface bands at $\bar{\Gamma}$ and $\bar{M}$ to shift oppositely with reference to the Fermi level [Figs. 11(b) and (c)]. In particular, for the out-of-plane compressive strain, the opposite shift of the surface bands at $\bar{\Gamma}$ and $\bar{M}$ also applies with reference to the unstrained Fermi level. Such properties can be used to design the strain-engineered nanodevices for dynamic valley control. As depicted in Fig. 11(d), a local compressive strain in the SnTe (111) surface can be generated by applying perpendicular pressure using a piezoelectric-material-based actuator. Under out-of-plane compressive strain, the Dirac cone at ${\bar{\Gamma}}$ shifts downwards, while the cones at ${\bar{M}}$ shift upwards, as schematically illustrated in Fig. 11(e). Therefore, with increasing strain, the Fermi level in the strained region will approach the Dirac point at $\bar{M}$, then goes across the valence band. The opposite shift in energy of Dirac cones at ${\bar{\Gamma}}$ and $\bar{M}$ in the strained region will lead to different transport behaviors for the massless Dirac fermions at these four valleys. Note that the strain-driven shift of Dirac point in energy also appears in Bi$_2$Se$_3$-class TIs\cite{zhao2012design}, but there is only one Dirac cone located at $\bar{\Gamma}$ for them.
The transmission probabilities $T(\theta)$ through the strain-induced nanostructure with two different strain levels are shown in Figs. 11(f) and (g). It can be seen that for the $\bar{\Gamma}$ valley, high transmission probabilities can always be maintained over a large angular range. This is because the Dirac point shifts downward and the scattering events always stay in the intraband tunneling regime. On the contrary, the Dirac points at the $\bar{M}$ valley shift upward with increasing strain, and the Fermi level in the strained region will gradually approach the Dirac point of zero carrier concentration. Eventually, the scattering of Dirac fermions at the $\bar{M}$ valley will be driven into the interband tunneling regime. Particularly in Fig. 11(g), when the Fermi level exactly crosses the Dirac points at $\bar{M}$ in the strained region, strong beam collimation along three different directions (i.e., $\theta=0$, $\pm 18.2^{\circ}$) can be clearly seen. Deviating from these specific directions, it is possible to achieve the pure current of the $\bar{\Gamma}$ valley. So the strong valley-selective filtering for massless Dirac fermions can be accomplished by dynamically applying local external pressure, which pave the way for strain-engineered nanoelectronic and valleytronic applications in TCIs.
\begin{figure*}[ht]
\begin{center}
\includegraphics[width=0.8\textwidth]{FIG12.jpg}
\caption{\label{fig:fig12} (Color online) (a) Band gap of SnTe at the $X$ point as a function of film thickness. The yellow region indicates a 2D TCI phase. (b) Edge states with mirror eigenvalues of an 11-layer SnTe film (left) and electric-field-induced gap (right). (c) Proposed topological transistor device for using an electric field to tune charge and spin transport. Without the electric field, the spin-filtered edge states are gapless and it is in the ``on'' state (left). Applying a perpendicular electric field, the edge states are gapped and it is in the ``off'' state (right).\cite{liu2014spin}}
\end{center}
\end{figure*}
\section{Thickness engineering}
When the thickness of SnTe film decreases, the top and bottom surface states will hybridize to open up an energy gap at the Dirac points. However, the hybridization is complex and multiple novel properties can emerge.
\subsection{2D TCI in (001) thin films}
The Dirac point of SnTe (001) surface states is deviated from the $\bar{X}$ point in the surface BZ; while at $\bar{X}$, there is an inverted band gap, i.e., the valence band (conduction band) is derived from the cation (anion). When the thickness of SnTe (001) film is below the penetration length of surface states, the top and bottom surface states will hybridize and result in an energy splitting between the bonding and anti-bonding states. The conduction and valence bands of the 2D film at $X$ will come from the bonding state of the anion at energy $E_{\rm A+}(X)$ and anti-bonding state of the cation at energy $E_{\rm C-}(X)$\cite{liu2014spin}. However, the band ordering of $E_{\rm A+}(X)$ and $E_{\rm C-}(X)$ depends on the competition between the hybridization of the two surfaces and the inverted gap of each surface. For thick films, the hybridization is weak so that $E_{\rm A+}(X) > E_{\rm C-}(X)$, which inherits the inverted band gap of the 3D limit. While the strong hybridization in the thin film drives $E_{\rm C-}(X)$ higher than $E_{\rm A+}(X)$, leading to a trivial phase. Considering the SnTe (001) films with an odd number of atomic layers, which are symmetric under the reflection $z \rightarrow -z$ about the (001) plane in the middle, the band gap [$E_{\rm A+}(X)-E_{\rm C-}(X)$] at the $X$ point as a function of film thickness is shown in Fig. 12(a)\cite{liu2014spin}. Above five layers, the gap of SnTe at $X$ is inverted and increases with the thickness. Due to the (001) mirror symmetry of the film, one can define the mirror Chern number of 2D film as in the case of 3D TCI. Liu {\it et al.}\cite{liu2014spin} found that the thick (001) film with an inverted gap at $X$ is a 2D TCI with mirror Chern number $|N_M|=2$ [Fig. 12(a)].
An important result of 2D TCI with mirror Chern number $|N_M|=2$ is that there are two pairs of counter-propagating edge states in the band gap. As shown in the left panel of Fig. 12(b), in a strip structure of 11-layer SnTe thin film, the edge states with opposite mirror eigenvalues cross each other at the edge of BZ, but are not located at TRIM. Here the Dirac-like edge states are protected by the (001) mirror symmetry, rather than the time-reversal symmetry as in the quantum spin Hall insulator. So applying a perpendicular electric field, which breaks the mirror symmetry, will generate a band gap in these edge states, as is shown in the right panel of Fig. 12(b). This unique property that the conductance of edge states is easily and widely tunable by an electric-field-induced gap instead of carrier depletion can be utilized to propose a topological transistor device made of dual-gated TCI thin films, as depicted in Fig. 12(c). With high on/off speed, the device can control the coupled charge and spin transport by purely electrical means.\cite{liu2014spin}
A similar mechanism appears in PbTe/SnTe superlattices along the [001] direction\cite{yang2014weak}. The folding of BZ due to the superlattice structure projects two $L$ points in the original BZ to a single point $L^{\prime}$, and their hybridizations form the bonding and anti-bonding states. The coupling strength between the states at equivalent $L$ points depends on the relative thickness or the ratio $m/2n$ in (PbTe)$_m$(SnTe)$_{2n-m}$ superlattice. For a large range of this ratio, the strong coupling induces a band inversion between the bonding state of conduction band and anti-bonding state of valence band. Thanks to reduction in the number of $L$ points from four to two, the band inversion at two $L^{\prime}$ will lead to a weak TI phase in PbTe/SnTe superlattices\cite{yang2014weak}.
\subsection{Quantum spin Hall states in (111) thin films}
Another important film is the (111) thin film, which has been grown epitaxially in recent experiments\cite{yan2014experimental,polley2014observation,wang2015molecular}. Different from (001) surface states, four Dirac cones of (111) surface states are centered at TRIMs: one is at the $\Gamma$ point and the other three are at the $M$ points, and they are not all equivalent. When the thickness of (111) film decreases, the hybridization strength between the top and bottom surface states at $\Gamma$ and $M$ will be different. In fact, the penetration length of the surface states at $M$ is much larger than that at $\Gamma$, hence the hybridization-induced gap at $M$ is correspondingly larger than the one at $\Gamma$ by orders of magnitude\cite{liu2015electrically}. As a result of this inequivalence between $\Gamma$ and $M$, the properties of SnTe (111) films are mainly determined by the low-energy physics at the $\Gamma$ point. The magnitude and sign of the gap at $\Gamma$ will depend on the film thickness, thus leading to a topological phase transition as a function of the film thickness\cite{liu2015electrically}. Taking the films with an even number of layers as an example [Fig. 13(a)], when the thickness decreases, both the fundamental gap $E_g$ and the gap at the $\Gamma$ point $E_g(\Gamma)$ exhibit a oscillatory behavior, indicating a sign change of the Dirac mass and the $Z_2$ topological phase transition. As shown in the inset of Fig. 13(a), the conduction bands of the 12-layer thin film are mainly derived from Sn orbitals, while they are mainly from Te orbitals for the 14-layer thin film, suggesting that 14-layer thin film is a quantum spin Hall (QSH) state. This result can be confirmed by the edge state calculations. As shown in Fig. 13(b), for a 14-layer thin film, the linear-dispersion gapless states exist in the bulk gap with a Dirac point at $\bar{\Gamma}$, while there are no such edge states in the 12-layer thin film. From Fig. 13(a), all the thin films with an even number of layers between 14 and 30 layers are QSH insulators.\cite{liu2015electrically}
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.48\textwidth]{FIG13.jpg}
\caption{\label{fig:fig13} (Color online) (a) The fundamental gap $E_g$ and the gap at the $\Gamma$ point $E_g(\Gamma)$ as functions of the film thickness with even number layers. The blue region indicates a quantum spin Hall (QSH) state, and the orange region represents a normal insulator (NI). The red (blue) dots in the inset denote the contribution of Te (Sn) atoms. (b) The edge states of a 12-layer thin film (left) and 14-layer thin film (right).\cite{liu2015electrically}}
\end{center}
\end{figure}
The similar results apply to the odd-layer thin film\cite{liu2015electrically}, and even the SnSe (111) film\cite{safaei2015quantum}. The mechanism for realizing the QSH phase in TCI thin films through the intersurface coupling is similar to the thin films of Bi$_2$Se$_3$-class TIs\cite{linder2009anomalous,liu2010oscillatory,lu2010massive}. However, due to the much larger penetration length of the SnTe/SnSe surface states, the hybridization gap is much larger than that in Bi$_2$Se$_3$ class of TIs.
\subsection{Electric field and helicity control}
The Dirac fermion of the surface states in the topological materials has the property of spin-momentum locking, which can also be defined as the helicity degree of freedom. As depicted in Fig. 14(a), in the film of topological materials, two surface Dirac cones located on the top and bottom surfaces are mirror images of each other obeying inversion symmetry and have opposite helicity. Specifically, the top surface has the left-handed (LH) electron and right-handed (RH) hole states, while the bottom surface has the RH electron and LH hole states. For the thinner film [Fig. 14(b)], the two Dirac cones interact with each other and open a hybridization gap. Meanwhile, the twofold helicity-degenerate band structures sustain both LH and RH electron (or hole) states of Dirac fermions. However, the helicity degeneracy can be removed in a controllable way by breaking the spatial inversion symmetry of the film, e.g., by applying an external electric field [Fig. 14(c)]. This feature allows the control over the helicity degree of freedom of Dirac fermions.
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.47\textwidth]{FIG14.jpg}
\caption{\label{fig:fig14} (Color online) A simplified model of electrically tuned band evolution in a thin film of topological material, where the helicity degree of freedom in the system can be controlled by a perpendicular electric field.\cite{zhao2015electronic}}
\end{center}
\end{figure}
Taking an odd-layer SnTe (111) film as examples, Zhao {\it et al.}\cite{zhao2015electronic} studied the helicity-resolved band structures and transports. They found that a moderate electric field can induce a giant helicity splitting to the degenerate bands in the vicinity of the $\bar{\Gamma}$ point. To explore helicity-dependent manipulation of Dirac fermions, a dual-gated heterostructure is exploited in SnTe (111) film, as shown in Fig. 15(a). For Fabry-P\'{e}rot-type model\cite{zhao2015electronic}, when tuning the Fermi level in the dual-gated region to cross the upper Dirac point, a strong direction-dependent helicity-selective transmission is indicated in Fig. 15(b). The RH electron wave has very high transmission probability over a large angular range, whereas a sharp transmission peak appears at normal incidence for the LH electron wave. Such pronounced helicity-selective transmission stems from the fact that the intrahelical scattering dominates over the interhelical one in the transmission events. When changing the width of dual-gated region and tuning the Fermi level, more helicity-resolved functionalities, including helical negative refraction, birefraction, and electronic focusing, are demonstrated\cite{zhao2015electronic}. Like the spin manipulation in thin film of magnetically doped TIs\cite{zhao2013field}, here the electrically tunable helicity-resolved control of Dirac fermions could open up a promising field for helicity-based electronics.
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.47\textwidth]{FIG15.jpg}
\caption{\label{fig:fig15} (Color online) (a) Schematic design of a dual-gated heterostructure tuned by the electric field. The incident helicity-degenerate Dirac electrons emitted from the source, and the transmitted helical electron waves are detected by the STM probe. (b) Pronounced helicity filtering when the Fermi level crosses the upper Dirac point. The colorful arrows illustrate the spin orientations at different transmission angles, where the color scale gives the out-of-plane spin component.\cite{zhao2015electronic}}
\end{center}
\end{figure}
\section{Conclusions and outlook}
In this article, we have reviewed the electronic properties of SnTe-class TCI materials, from the intrinsic topology, orientation-dependent surface states, to tunable properties under various perturbations. Protected by the mirror symmetry and characterized by the mirror Chern number, SnTe-class TCI materials have multiple branches of Dirac surface states, whose types depend on the surface orientations and even the surface reconstructions. Chemical doping can realize the control of carrier types or concentrations, the topological superconductivity, and the large-Chern-number QAHE. Strain is an effective way to tune the electronic properties of IV-VI materials, including driving topological phase transition for the bulk system, shifting the Dirac point position of the surface states in the momentum space or energy space. The latter could be used for the valleytronic applications. For the thin film of SnTe-class materials, the thickness of film can influence the hybridization strength, and create diverse topological phases, for instance, 2D TCI, weak TI, QSH states, and NI. Among them, the (001) thin film with 2D TCI phase could be utilized to design a topological transistor device with a high on/off speed, and more generally, the helicity degree of freedom of Dirac fermions in the thin film can be controlled by an external electric field to realize the helicity-resolved functionalities.
These novel properties and potential applications are not limited to the IV-VI materials. In fact, more TCIs are proposed theoretically\cite{kargarian2013topological,hsieh2014topological,liu2014topological,jadaun2013topological,slager2013space,dong2015classification}, and a great deal of studies are needed to search for new TCI materials and explore their applications. Moreover, the crystal symmetry protection has been extended from the insulators to semimetals, such as the Dirac semimetal\cite{young2012dirac,wang2012dirac,liu2014discovery,wang2013three,liu2014stable,weng2016topological} and nodal-line semimetal\cite{kim2015dirac,burkov2011topological,bian2015topological,bian2015drumhead,weng2016topological}, which possess more exotic effects\cite{hosur2013recent,bian2015topological,bian2015drumhead}. The interplay between the electronic topology and crystal symmetry has broadened the field of topological phase, and paves a path for developing the quantum electronic and spintronic devices.
\providecommand{\noopsort}[1]{}\providecommand{\singleletter}[1]{#1}%
| 339abff211b2e998c99a6ec60ba79a95a339d6c9 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Employing the AdS/CFT correspondence it was shown \cite{Hartnoll:2008vx} that a
holographic superconductor can be builded on a boundary of
a gravity theory consisting of a black hole which acquires scalar hair
at temperature below a critical temperature, while above the critical temperature there is
no scalar hair. A condensate of the charged
scalar field is formed through its coupling to a Maxwell field of
the gravity sector. The study was carried out in the probe limit~\cite{Gubser:2008px}, in which in general the product of the charge of the black hole
and the charge of the scalar field is held fixed while the latter
is taken to infinity and this resulted to neither field to backreact on the metric. Considering fluctuations of the vector
potential, the frequency dependent conductivity was calculated,
and it was shown that it develops a gap determined by the
condensate.
This model was further studied \cite{Hartnoll:2008kx}
beyond the probe limit. If the scalar charge is large then
the backreaction of the scalar field to the spacetime metric has
to be considered. It was found that all the
basic characteristics of the dual superconductor were
persisting even for very small charge. One of the characteristics of these models is
the strong pairing mechanism which is in operation. This is due to strong bounding of Cooper pairs and
this is manifest to the high values of the gap at low temperature. This strong pairing mechanism resulted to $2 \Delta \approx 8.4 T_c$, where $\Delta$ is the condensation gap, which has to be compared to
the BCS prediction $2 \Delta \approx 3.54 T_c$, which is much lower in real materials, due to impurities. The above proposals have inspired much effort on further research and on their possible extensions (see \cite{Horowitz:2010gk,Cai:2015cya} and references therein for a review).
The effects of paramagnetic impurities on
superconductors were studied in \cite{Abrikosov}.
It was found that the transition
temperature decreases sharply with increasing impurity
concentration and goes to zero at a critical concentration.
Measurements
of the energy gap and of the transition temperature
$T_c$ as a function of the concentration of paramagnetic
impurities show that the gap decreases much more
rapidly than does the critical temperature which agrees with the results
of \cite{Abrikosov}. This behaviour was also observed in \cite{Gennes} and
\cite{Phillips}.
Further calculations on the effects of paramagnetic impurities in superconductors were carried out in \cite{skalski},
taking full advantage of the information contained
in the Green's function of the system. The density of states in energy has been computed for different
values of the inverse collision time. The excitation energy gap $\Omega_g$ is defined to be the energy at which the density of states vanishes.
The temperature-dependent order parameter has been computed and the
behaviour of $\Omega_g (T)$ was determined. A comparison with tunneling experiments showed a disagreement of the two parameters.
The real part of the conductivity at $T=0$ is shown to be zero for frequencies less
than $2 \Omega_g$ and proportional to the square of the density of states for vanishingly small frequencies in the gapless
region of impurity concentrations. These detailed calculations showed clearly that in real materials, the essential feature of superconductivity is the correlation of the
electrons and the formation of Cooper pairs, rather than the existence of a gap in the excitation energy spectrum. For even in the absence of a gap, so
long as correlations persist there is an ordered state below a certain critical temperature which displays the usual
properties of a superconductor.
A holographic realization of impurities was discussed in \cite{Hartnoll:2008hs}. Using the AdS/CFT correspondence
two different types of impurities were calculated and the Nernst response for the impure theory was specified.
A model of a gravity dual of a gapless
superconductor was proposed in \cite{Koutsoumbas:2009pa}. The gravity sector was consisting of a charged scalar field
which provided
the scalar hair of an exact black hole
solution~\cite{Martinez:2004nb,Martinez:2005di} below a critical temperature $T_c$, while an electromagnetic perturbation of the background
determined the conductivity giving rise to a gapless superconductor. It was found that the normal component
of the DC conductivity had a milder behaviour than
the dual superconductor in the case of a black hole of flat
horizon~\cite{Hartnoll:2008vx} in which $n_{n}$ exhibited a clear
gap behaviour. The behaviour it was observed in the boundary
conducting theory was attributed to materials with paramagnetic
impurities as it was discussed in \cite{skalski}. Impurities effects were studied in \cite{Ishii:2012hw}, while impurities in the Kondo
Model were considered in \cite{O'Bannon:2015gwa,Erdmenger:2015xpq}.
In this work we extend the model in \cite{Hartnoll:2008vx} by introducing a derivative coupling of the
scalar field to Einstein tensor.
This term belongs to a general class of
scalar-tensor gravity theories resulting from the Horndeski
Lagrangian \cite{Horndeski:1974wa}. These theories, which were
recently rediscovered \cite{Deffayet:2011gz},
give second-order field equations and contain as a subset a theory
which preserves classical Galilean symmetry \cite{Nicolis:2008in,
Deffayet:2009wt, Deffayet:2009mn}. The derivative coupling of the scalar field to Einstein tensor
introduces a new scale in the theory which on short distances allows
to find black hole solutions \cite{Kolyvaris:2011fk,Rinaldi:2012vy,Kolyvaris:2013zfa,Cisterna:2014nua,Brihaye:2016lin} with scalar hair just outside the black hole horizon,
while if one considers the gravitational collapse of a scalar field coupled to the Einstein tensor then the formation of a black hole
takes more time to be formed compared to the collapse of a scalar field minimally coupled to gravity \cite{Koutsoumbas:2015ekk}.
On large distances the presence of the derivative coupling acts as a friction term in the inflationary period of the cosmological
evolution \cite{Amendola:1993uh,Sushkov:2009hk,germani}. Moreover, it was found that at the end of
inflation in the preheading period, there is a suppression of heavy particle production as the derivative coupling is increased. This was attributed to the fast decrease of kinetic
energy of the scalar field due to its wild oscillations \cite{Koutsoumbas:2013boa}.
The above discussion indicates that one of the main effects of the kinematic coupling of a scalar field to Einstein tensor
is that gravity influences strongly the propagation of the scalar field compared to a scalar field minimally coupled to gravity.
We are going to use this behaviour of the scalar field
to holographically simulate the effects of a high concentration of impurities in a material. The presence of impurities in a superconductor is
making the pairing mechanism of forming Cooper pairs less effective and this is happening because the quasiparticles are loosing energy because of strong
concentration of impurities. This holographic correspondence is supported by our finding. We found that as the value of the derivative coupling is increased the critical temperature is degreasing while
the condensation gap $\Delta$ is decreasing faster than the temperature. Also by calculating the perturbation of the scalar potential we found
that the condensation gap for large values of the derivative coupling is not proportional to the frequency of the real part of conductivity
which is characteristic of a superconducting state with impurities \footnote { We note that in \cite{Lin:2015kjk} a holographic
superconductor was considered containing a derivative coupling
for a scalar field in the background of a regular phantom plane symmetric black hole.}.
The work is organized as follows. In Section \ref{review} we review the basic ingredients of the holographic model discussed in \cite{Hartnoll:2008vx}.
In Section \ref{derscalar} we extend the holographic model of Section \ref{review} adding a scalar field coupled to Einstein tensor and we give the equations of motion. In Section \ref{solution} we solve numerically the equations of motion and we find the condensation gap and the critical temperature for various values of the derivative coupling. In Section \ref{conductivity} we study the conductivity while in Section \ref{conc} are our conclusions.
\section{Holographic Superconductor with a scalar field minimally coupled to gravity}
\label{review}
In this section we review in brief the holographic model of \cite{Hartnoll:2008vx}.
To introduce the minimal holographic superconductor model, we consider a Maxwell field and a charged complex scalar field
coupled to gravity with the action
\begin{eqnarray}\label{System}
S=\int d^{4}x\sqrt{-g}\left[\frac{R+\Lambda}{16 \pi G}
-\frac{1}{4}F_{\mu\nu}F^{\mu\nu}-|\nabla\psi - iqA\psi|^{2}
-m^2|\psi|^2 \right] \ .
\end{eqnarray}
If we rescale the fields and the coupling constant as \begin{eqnarray}
A_{\mu} =\frac{\tilde{A}_{\mu}}{q}~,~~~
\psi =\frac{\tilde{\psi}}{q}~,
\end{eqnarray}
then the matter action in (\ref{System}) has a $1/q^2$ in front,
so the backreaction of the matter fields on the metric is
suppressed when $q$ is large and the limit $q\rightarrow\infty$ defines the probe limit.
In this limit
the Einstein equations
admit as a solution the planar Schwarzschild AdS black hole
\begin{eqnarray}\label{schwar}
ds^2= -f(r) dt^2 + \frac{dr^2}{f(r)} + r^2 (dx^2 +
dy^2)\qquad,\qquad
f(r)=r^2 - \frac{r_h^3}{r}~. \end{eqnarray}
Taking the ansatz
$\psi=|\psi|$, $A=\phi dt$ where $\psi$, $\phi$ are both functions
of $r$ only, we can obtain the equations of motion for $\psi,
\phi$
\begin{eqnarray}
\psi^{\prime\prime}+\left(
\frac{f^\prime}{f}+\frac{d-2}{r}\right)\psi^\prime
+\left(\frac{\phi^2}{f^2}-\frac{m^2}{f}\right)\psi=0\,,
\label{Psi}
\end{eqnarray}
\begin{eqnarray}
\phi^{\prime\prime}+\frac{d-2}{r}\phi^\prime-\frac{2\psi^2}{f}\phi=0~.
\label{Phi}
\end{eqnarray}
It was argued in \cite{Gubser:2008px} that there is a critical temperature
below which the
black
hole acquires hair~\cite{Hartnoll:2008vx,Hartnoll:2008kx}.
The equations (\ref{Psi}) and (\ref{Phi}) can be solved
numerically by doing integration from the horizon out to the
infinity taking under consideration the right boundary conditions.
The solutions behave like
\begin{eqnarray}
\psi=\frac{\psi_{-}}{r^{\lambda_{-}}}+\frac{\psi_{+}}{r^{\lambda_{+}}}\,,\hspace{0.5cm}
\phi=\mu-\frac{\rho}{r}\,, \label{infinity}
\end{eqnarray}
with
\begin{eqnarray}
\lambda_\pm=\frac{1}{2}\Big{[}3 \pm\sqrt{(9+4m^{2}L
^2)}~\Big{]}\,,\label{LambdaZF}
\end{eqnarray}
where $L$ is the scale of the AdS space and $\mu$ and $\rho$ are
interpreted as the chemical potential and charge density in the
dual field theory respectively. The coefficients $\psi_{-}$ and
$\psi_{+}$ according to the AdS/CFT correspondence,
correspond to the vacuum expectation values
$\psi_{-}=\langle\mathcal{O}_{-}\rangle/\sqrt{2}$, $\psi_{+}=\langle\mathcal{O}_{+}\rangle/\sqrt{2}$ of an
operator $\mathcal{O}$ dual to the scalar field. We can impose
boundary conditions that either $\psi_{-}$ or $\psi_{+}$ vanishes.
Fluctuations of the vector
potential $A_x$ in the gravity sector gives the
conductivity in the dual CFT as a function of
frequency. The Maxwell equation for
zero spatial momentum and with a time dependence of the form $e^{-
i \omega t}$ reads as
\begin{eqnarray}\label{eq:Axeq} A_x'' + \frac{f'}{f} A_x' +
\left(\frac{\omega^2}{f^2} - \frac{2 \psi^2}{f} \right) A_x = 0
\,. \end{eqnarray}
The above equation can be solved by imposing ingoing wave boundary conditions at the horizon
$A_x \propto f^{-i\omega/3r_0}$. On the other hand, the
asymptotic behaviour of the Maxwell field at boundary is
\begin{eqnarray} A_x = A_x^{(0)} + \frac{A_x^{(1)}}{r} + \cdots. \end{eqnarray}
Then according to AdS/CFT correspondence dual source and expectation
value for the current are given by \begin{eqnarray} A_x = A_x^{(0)} \,, \qquad
\langle J_x \rangle = A_x^{(1)} \,. \end{eqnarray}
Finally, Ohm's law gives the conductivity \begin{eqnarray}\label{eq:conductivity} \sigma(\omega)
= \frac{\langle J_x \rangle}{E_x} = - \frac{ \langle J_x
\rangle}{\dot A_x} = -\frac{ i \langle J_x \rangle}{\omega A_x}
= - \frac{i A_x^{(1)}}{\omega A_x^{(0)}} \,. \end{eqnarray}
\section{Holographic Superconductor with a scalar field kinematically coupled to Einstein Tensor}
\label{derscalar}
In this section, we will consider a complex scalar field which excepts its coupling
to curvature it is also coupled to Einstein tensor with action
\begin{eqnarray}\label{EGBscalaraction}
I=\int d^4x\sqrt{-g}\left[ \frac{R+\Lambda}{16\pi G}-\frac{1}{4}F_{\mu\nu}F^{\mu\nu}-(g^{\mu\nu}+\kappa
G^{\mu\nu})D_\mu\psi (D_\nu\psi)^{*} - m^2|\psi|^2 \right]~,
\end{eqnarray}
where \begin{eqnarray} D_\mu = \nabla_\mu - i e A_\mu \end{eqnarray}
and $e$, $m$ are the charge and the mass of the scalar field and
$\kappa$ the coupling of the scalar field to Einstein tensor of
dimension length squared. For convenience we set
\begin{eqnarray}
\Phi_{\mu\nu} &\equiv& D_{\mu}\psi (D_{\nu}\psi)^*~,\\
\Phi &\equiv& g^{\mu\nu}\Phi_{\mu\nu}~,\\
C^{\mu\nu} &\equiv& g^{\mu\nu} + \kappa G^{\mu\nu}~.
\end{eqnarray}
The field equations resulting from the action
(\ref{EGBscalaraction}) are
\begin{eqnarray} G_{\mu\nu} +\Lambda g_{\mu\nu} = 8\pi T_{\mu\nu}
\ , \ \ \ \ T_{\mu\nu} = T_{\mu\nu}^{(\psi)} +
T_{\mu\nu}^{(EM)} + \kappa\Theta_{\mu\nu}~, \label{einst}\end{eqnarray} where,
\begin{eqnarray}
T_{\mu\nu}^{(\psi)} & = & \Phi_{\mu\nu} + \Phi_{\nu\mu} - g_{\mu\nu}(g^{ab}\Phi_{ab} + m^2 |\psi|^2)~, \\
T_{\mu\nu}^{(EM)} & = & F_{\mu}^{\phantom{\mu} \alpha} F_{\nu
\alpha} - \frac{1}{4} g_{\mu\nu} F_{\alpha\beta}F^{\alpha\beta}~,
\end{eqnarray} and \begin{eqnarray}
\Theta_{\mu\nu} = & -& g_{\mu\nu} R^{ab}\Phi_{ab} + R_{\nu}^{\phantom{\nu}a}(\Phi_{\mu a} + \Phi_{a\mu}) + R_{\mu}^{\phantom{\mu}a} (\Phi_{a\nu} + \Phi_{\nu a}) - \frac{1}{2} R (\Phi_{\mu\nu} + \Phi_{\nu\mu}) \nonumber\\
& - & G_{\mu\nu}\Phi - \frac{1}{2}\nabla^a\nabla_\mu(\Phi_{a\nu} + \Phi_{\nu a}) - \frac{1}{2}\nabla^a\nabla_\nu(\Phi_{\mu a} + \Phi_{a\mu}) + \frac{1}{2}\Box (\Phi_{\mu\nu} + \Phi_{\mu\nu}) \nonumber \\
& + & \frac{1}{2}g_{\mu\nu} \nabla_a\nabla_b (\Phi^{ab} + \Phi^{ba})
+ \frac{1}{2}(\nabla_\mu\nabla_\nu + \nabla_\nu\nabla_\mu) \Phi -
g_{\mu\nu}\Box\Phi~. \label{theta} \end{eqnarray}
The Klein-Gordon equation is
\begin{eqnarray} (\partial_\mu-i e A_\mu) \left[
\sqrt{-g}C^{\mu\nu}(\partial_\nu - i e A_\nu)\psi\right] =
\sqrt{-g} m^2\psi~, \label{glgord} \end{eqnarray}
while the Maxwell equations read \begin{eqnarray} \nabla_\nu F^{\mu\nu} +
C^{\mu\nu} \left[ 2 e^2 A_\nu |\psi|^2 + i e
(\psi^*\nabla_\nu\psi- \psi\nabla_\nu\psi^*)\right]
=0~. \label{max}\end{eqnarray}
\subsection{Solution of equations of motion and phase transition}
\label{solution}
As in \cite{Hartnoll:2008vx} we also take the ansatz
$\psi=|\psi|$, $A=\phi dt$ where $\psi$, $\phi$ are both functions
of $r$ only. In the probe limit, under the metric (\ref{schwar}), the above equations of motion for the matter fields become
\begin{equation}
\left[1+\kappa\left(\frac{f}{r^2}+\frac{f'}{r}\right)\right]\psi^{\prime\prime} + \left[\frac{2}{r} + \frac{f^\prime}{f} + \kappa\left (\frac{3f'}{r^2}+\frac{f'^{2}}{r f}+\frac{f''}{r} \right)\right]\psi^\prime + \left[\frac{e^2\phi^2}{f^2}\left(1+\kappa\left(\frac{f}{r^2}+\frac{f'}{r}\right)\right) - \frac{m^2}{f}\right]\psi = 0 \label{newpsi}~,
\end{equation}
\begin{equation} \phi^{\prime\prime} + \frac{2}{r}\phi^\prime - \frac{2e^2\psi^2}{f}\left[1+\kappa\left(\frac{f}{r^2}+\frac{f'}{r}\right)\right]\phi = 0~. \label{newphi}
\end{equation}
Note that if $\kappa=0$ then we get the equations (\ref{Psi}) and (\ref{Phi}).
The new equations depend now on the second derivative of the function $f$ in contrast to the preview case where we had only its first derivative, due to the additional term of $G^{\mu\nu}$ in the action (\ref{EGBscalaraction}).
With the same study in the minimal case presented in previous section, we can numerically solve the above equations
of motion. We especially study the effect of $\kappa$ on the critical temperature and strength of condensation in the dual
supercondutor. Without loss of generality, we set $e=1$ and $m^2=-2$, so that we have $\lambda_-=1$ while $\lambda_+=2$. We will choose $\psi_-$ as the source which is set to be vanishing, while $\psi_+$ as vacuum expectation values.
\begin{table}
\centering
\begin{tabular}{|c||c|c|c|c|c|c|c|}
\hline
$\kappa$&-0.01&0 & 0.01 &0.05 & 0.1&0.5 &1 \\ \hline
$\frac{T_c}{\sqrt{\rho}}$&$0.1218$ & $0.1184$&$0.1158$&$0.1091$&$0.1043$&$0.09353$&$0.0906$ \\ \hline
$C_1$&$243$ & $140$&$73$&$10$&$2$&$0.01$&$0.003$ \\ \hline
\end{tabular}
\caption{\label{tableTc}Critical temperature $T_{c}$ for the phase transition of $<O_{+}>$ with different $\kappa$.}
\end{table}
\begin{figure}
\centering
\includegraphics[width=.4\textwidth]{condensationO.eps}\hspace{1cm}
\caption{\label{fig-condensation}The strength of condensation VS the temperature with samples of $\kappa$.}
\end{figure}
The critical temperatures with different $\kappa$ are summarized in Table {\ref{tableTc}}. It shows that as the coupling becomes stronger, the critical temperature is lower, meaning that the phase transition is hard to occur as it is expected. This effect of coupling is very different from the influence of higher-derivative coupling between the matters studied in \cite{Kuang:2013oqa,Kuang:2014kha}. As the temperature is decreased, the scalar field can have non-zero solution and the strength of condensation becomes higher. We show this
phenomena in FIG. \ref{fig-condensation}. It is obvious that the strength of condensation is suppressed by the derivative coupling. And with enlarging the coupling\footnote{There is an instability in our numerical solution of the equations of motion for large or small values of the derivative coupling when the temperature is far way from $T_c$. Specially, it is difficult to reach a low temperature $T/T_c\sim 0.1$ when $\kappa>0.5$. For this reason in FIG. \ref{fig-condensation} we restricted the curves to the values of $\kappa$ in the range $ -0.01\leqslant\kappa\leqslant0.5$. However, when we are near $T_c$ our numerics work better, so we could fit relation (\ref{near}) for larger values of $\kappa$. So in order to study the same shift from the critical temperature in the following study we focus on the range $ -0.01\leqslant\kappa\leqslant0.5$. }, the condensation decreases faster than the temperature, so that the condensation gap tends to be zero at low temperatures for the strong coupling. This behaviour from $\kappa$ is coincident with the effect of paramagnetic impurities on superconductors observed in \cite{Abrikosov,Gennes,Phillips}.
We fit the curve near the critical temperatures. As $T\rightarrow T_c$, the condensation is continuous and behaves as
\begin{equation}
\langle\mathcal{O}_{+}\rangle\simeq C_1 T_c^2(1-T/T_c)^{1/2} \label{near}
\end{equation}
where $C_1$ are also listed in Table {\ref{tableTc}}. We see that the coefficient $C_1$ decreases drastically to suppress the condensation when the coupling is increased. Observe that the exponent is always $1/2$ which implies that $\kappa$ does not modify the order of the phase transition and it is always second order. In the next subsection we will study the conductivity caused by the condensation below the critical temperature.
\subsection{Conductivity}
\label{conductivity}
To compute the conductivity in the dual CFT as a function of
frequency we need to solve the Maxwell equation for fluctuations of the vector
potential $A_x$. The Maxwell equation at
zero spatial momentum and with a time dependence of the form $e^{-i \omega t}$ gives
\begin{equation}\label{eq:Axeq}
A_x'' + \frac{f'}{f} A_x' +\left[\frac{\omega^2}{f^2} - \frac{2 e^2\Psi^2}{f}
\left(1+\kappa\left(\frac{f''}{2}+\frac{f'}{r}\right)\right)\right] A_x = 0~.
\end{equation}
We will solve the perturbed Maxwell equation with ingoing wave boundary conditions at the horizon, i.e.,
$A_x \propto f^{-i\omega/3r_0}$. The asymptotic behaviour of the Maxwell field at large radius is $A_x = A_x^{(0)} + \frac{A_x^{(1)}}{r} + \cdots$. Then,
according to AdS/CFT dictionary, the dual source and expectation
value for the current are given by $ A_x = A_x^{(0)} $ and $ \langle J_x \rangle = A_x^{(1)}$, respectively.
Thus, similar to the equation (\ref{eq:conductivity}), the conductivity is also read as
\begin{equation}
\sigma(\omega)= - \frac{i A_x^{(1)}}{\omega A_x^{(0)}}~.
\end{equation}
\begin{figure}[h]
\centering
\includegraphics[width=0.9\textwidth]{conductivity.eps}\hspace{1cm}
\caption{\label{fig-sigma}Conductivity for samples of $\kappa$ at $T/T_c\simeq 0.1$. The values of $\kappa$ are $\kappa=-0.01$(dotted red),$\kappa=0$(black),$\kappa=0.1$(blue) and $\kappa=0.5$(purple).}
\end{figure}
The numerical results of conductivity at low temperature
with $\frac{T}{T_c}\simeq 0.1$ are shown in FIG. \ref{fig-sigma}. Similarly, due to the Kramers-Kronig(KK) relations
\begin{equation}\label{eq:KK}
\mathrm{Im}[\sigma(\omega)]=-\frac{1}{\pi}\mathcal{P}\int_{-\infty}^{\infty}\frac{\mathrm{Re}[\sigma(\omega')]}{\omega'-\omega}d\omega'~,
\end{equation}
where $\mathcal{P}$ denotes the Cauchy principal value, the divergence of the imaginary part at zero frequency indicates a delta function of real part at $\omega=0$.
At very low temperature, the energy gap in the real part of the conductivity measured by the critical temperature (left part of FIG. \ref{fig-sigma}) decreases as the derivative coupling becomes stronger, which agrees well with the behaviour of the condensation as it is depicted in FIG. \ref{fig-condensation} in the last subsection.
In order to show how the conductivity opens a gap as the temperature decreases, in FIG. \ref{fig-sigmak=0.01}, we show the frequency-dependent conductivity with $\kappa=0.01$ at different temperatures. In the left plot, the horizontal line, which is frequency independent, corresponds to temperatures at or above the critical value and there is no condensate. As we lower the temperature in the subsequent curves, we see that a gap $\omega_g$ opens in the real part of the conductivity and the gap becomes wider and deeper as the temperature becomes smaller. In the right plot, we plot the conductivity for temperature lower than
$T_c$ related the left plot by rescaling the frequency by the condensate. It is obvious that the curves tend to a limit in which the width of the gap is proportional to the size of the condensate. i.e., $\omega_g \simeq \sqrt{<O_+>}$. The above features are similar with those in minimal coupling disclosed in \cite{Hartnoll:2008vx} .
Furthermore, the frequency-dependent conductivity with $\kappa=0.5$ at different temperatures is presented in
FIG. \ref{fig-sigmak=0.5}. Similarly to the cases with $\kappa=0.01$ and minimal coupling, lower temperature gives us wider energy gap. By comparing the left plots of FIG. \ref{fig-sigmak=0.01} and FIG. \ref{fig-sigmak=0.5}, we see that at the same scaled temperature $T/T_c$, the gap is narrower for larger coupling, which is consistent with the results shown in FIG. \ref{fig-condensation} and FIG. \ref{fig-sigma}. However, if being scaled by the condensation, see the right plot for
$\kappa=0.5$, the behavior $\omega_g \simeq \sqrt{<O_+>}$ at low $T/T_c$ is violated and it is $\omega_g \simeq 20\sqrt{<O_+>}$ for $T/T_c\simeq 0.1$. The coefficient $20$ is not out of expect. Because if we see carefully the condensation plot
FIG. \ref{fig-condensation} and FIG. \ref{fig-sigma}, we will find $ \sqrt{<O_+>}\simeq 0.2 T_c$ and $\omega_g \simeq 4T_c$, so we get the same relation as before $\omega_g \simeq 20\sqrt{<O_+>}$.
\begin{figure}[h]
\centering
\includegraphics[width=0.8\textwidth]{Conduck-0p01.eps}\hspace{1cm}
\caption{\label{fig-sigmak=0.01}Conductivity with $\kappa=0.01$ as the temperature goes further away from the critical point. In the left plot, from left to right, the temperatures are $T/T_c= 1$, $T/T_c\simeq 0.94$, $T/T_c\simeq 0.80$, $T/T_c\simeq 0.40$, $T/T_c\simeq 0.10$. They are $T/T_c\simeq 0.80$, $T/T_c\simeq 0.40$, $T/T_c\simeq 20$ in the right plot.}
\end{figure}
\begin{figure}[h]
\centering
\includegraphics[width=0.8\textwidth]{Conduck-0p5.eps}\hspace{1cm}
\caption{\label{fig-sigmak=0.5}Conductivity for $\kappa=0.5$ with the frequency rescaled by temperature and the condensation. In the left plot, from left to right, the temperatures are $T/T_c= 1$, $T/T_c\simeq 0.94$, $T/T_c\simeq 0.80$, $T/T_c\simeq 0.40$, $T/T_c\simeq 0.10$. They are $T/T_c\simeq 0.80$, $T/T_c\simeq 0.40$, $T/T_c\simeq 0.10$ in the right plot.}
\end{figure}
\begin{table}[h]
\centering
\begin{tabular}{|c||c|c|c|c|c|c|}
\hline
$\kappa$&-0.01&0 & 0.01 &0.05 & 0.1&0.5 \\ \hline
$\frac{\omega_g}{\sqrt{\langle O_+\rangle}}$&$0.61$ & $0.80$&$1$&$1.9$&$2.7$&$20$ \\ \hline
\end{tabular}
\caption{\label{tablewg}$\frac{\omega_g}{\sqrt{\langle O_+\rangle}}$ far away from the critical temperature ($T/T_c\simeq 0.10$) with different $\kappa$.}
\end{table}
For the systems far away from the critical temperature ($T/T_c\simeq 0.10$), by careful study,
we summarize the values of $\frac{\omega_g}{\sqrt{<O_+>}}$ with some
$\kappa$ in Table \ref{tablewg}. It is obvious that as $|\kappa|$ shifts from the minimal coupling, the relation
$\omega_g \simeq \sqrt{<O_+>}$ is sharply violated, which is one of the main features of a superconducting state with impurities. Note that this effect of possible impurity parameter $\kappa$ is different from that studied in \cite{Ishii:2012hw} where it can not introduce this kind of violation.
We can now explore the behaviour of conductivity at very low frequency. When $T<T_c$, the real part of the conductivity present a delta function at zero frequency and the imaginary part has a pole, which is attributed to the KK relations (\ref{eq:KK}).
More specifically, as $\omega\rightarrow 0$, the imaginary part behaves as $Im(\sigma)\sim n_s/\omega$, and according to Kramers-Kronig relations, the real part has the form $Re(\sigma)\sim\pi n_s\delta(\omega)$. Here the coefficient $n_s$ of the delta function is defined as the superfluid density. By fitting data near the critical temperature, we find that with various couplings, the superfluid density has the behaviour
\begin{equation}\label{eq:ns}
n_s\simeq C_2 T_c(1-T/T_c)~,
\end{equation}
which means that $n_s$ will vanish linearly as $T$ goes to $T_c$. This is consistent with that happens in the minimal coupling. Also we find that the coefficient $C_2$ is not very sensitive to the coupling and the value oscillates around $24$ with $\pm 2$ shifting for the samples of $\kappa$. This insensitivity to the changes of the coupling is reasonable because in the region very close to the critical temperature, the system is marginally shifted from the normal state, so that varying $\kappa$ has very small effect on the behaviour of the solutions. Meanwhile, as proposed in \cite{Hartnoll:2008vx}, the non-superconducting density can be defined as
$n_n=\lim_{\omega\rightarrow 0}Re(\sigma)$. Far from the critical temperature (with $T/T_c\sim 0.1$), we fit and obtain that $n_n$ decays as
\begin{equation}\label{eq:nn}
n_n\sim \exp[-\frac{\Omega_g}{T}]~,
\end{equation}
where $\Omega_g$ can be explained as the energy gap for charged excitation at the corresponding temperature. The values of
$\Omega_g$ close to $T/T_c\sim 0.1$ are summarized in Table \ref{tableWg}. Rough comparison with the numerical energy gap
$\omega_g/T_c$ shown in the left plot of FIG.\ref{fig-sigma} gives us that $\omega_g \simeq 2\Omega_g$, where the factor $2$ suggests that the gaped charged quasiparticles are formed in pairs as addressed in \cite{Hartnoll:2008vx}. Analyzing the data in the table, we see that $\kappa=0.5$ corresponds to $\omega_g \simeq 2\Omega_g \simeq 4 T_c$ which is very near the prediction $\omega_g \simeq 3.54 T_c$ in BCS theory. In this tendency, we believe that BCS prediction may be fulfilled by larger coupling once the numerics are controlable. Thus, in this sense, the derivative coupling somehow mimic the effect of the impurities in a real material.
\begin{table}[h]
\centering
\begin{tabular}{|c||c|c|c|c|c|c|}
\hline
$\kappa$&-0.01&0 & 0.01 &0.05 & 0.1&0.5 \\ \hline
$\frac{\Omega_g}{T_c}$&$4.0$ & $3.8$&$3.5$&$3.0$&$2.8$&$2.0$\\ \hline
\end{tabular}
\caption{\label{tableWg} The energy gap for charged excitation $\frac{\Omega_g}{T_c}$ with different $\kappa$.}
\end{table}
\section{Conclusions}
\label{conc}
We studied a holographic description of a superconductor in which the gravity sector consists of a Maxwell field and a charged scalar field which except its usual minimal coupling to gravity it is also coupled to Einstein tensor. Solving the equations of motion numerically in the probe limit, we found that as the strength of the new coupling is increased,
the critical temperature below which the scalar field condenses is lowering, the condensation gap decreases faster than the temperature, the width of the condensation gap is not proportional to the size of the condensate and
at low temperatures the condensation gap tends to zero for the strong coupling.
Analysing the frequency dependence of conductivity we found that at strong coupling the relation $\omega_g \simeq 2\Omega_g \simeq 4 T_c$ holds, where $\Omega_g$ is the energy gap for charged excitation, which is very near the prediction $\omega_g \simeq 3.54 T_c$ in BCS theory.
We argued that these results suggest that the derivative coupling in the gravity bulk can have a dual interpretation on the boundary corresponding to impurities concentrations in a real material. This correspondence can be understood from the fact that the coupling of a scalar field to Einstein tensor alters the kinematical state of the scalar field a behaviour which it is also exhibited by the quasiparticles moving in a material with impurities.
We believe that the probe limit captures all the essential features of the problem under study. Nevertheless, it would be interesting to extend this study beyond the probe limit. Assuming a spherically symmetric ansatz for the metric the field equations
(\ref{einst}), (\ref{glgord}) and (\ref{max}) have to be solved. However, even their numerical solution is a formidable task mainly because of the presence of the energy-momentum tensor (\ref{theta}) resulting from the coupling of the scalar field to Einstein tensor. A more realistic approach would be to follow
the perturbative methods employed in \cite{Kolyvaris:2011fk,Kolyvaris:2013zfa}.
\begin{acknowledgments}
We thank Olivera Miskovic and George Filios for their contribution at the early stage of this work. X-M.K is partly supported by FONDECYT Grant No.3150006.
\end{acknowledgments}
| 8a5793c178cbe7a6822fb62be9773a665ced90cf | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Heavy quarks (charm and beauty) are abundantly produced at the LHC \cite{LHC} in the early stages of hadronic collisions via hard scattering processes and experience the full evolution of the collision. The measurement of heavy flavours in \ensuremath{\mathrm {p\kern-0.05emp}}\ collisions can be used to test pQCD calculations. In addition, they provide an essential baseline for the studies in nuclear (\ensuremath{\mbox{p--Pb}}\ and \ensuremath{\mbox{Pb--Pb}}) collisions. Their production in nuclear collisions is modified by cold nuclear matter effects (CNM) such as shadowing and energy loss \cite{eloss, cnm}. The knowledge of these effects is fundamental for understanding the interactions of heavy quarks with the deconfined medium formed in heavy-ion collisions where the modified transverse momentum distribution can be used to infer the medium's characteristics. Heavy quark production is experimentally accessible through the measurement of heavy-flavour hadrons (D or B hadrons) via their decay products. \newline
The charged-particle multiplicity is defined as the average number of charged particles produced per event. It is a key observable for multi-parton interactions (MPI) in high-energy hadronic collisions \cite{mpi1, mpi2}. The charged-particle multiplicity may be used to test models of particle production such as PYTHIA \cite{pythia} and PHOJET \cite{phojet}. Finally, the charged-particle multiplicity dependence of open heavy-flavour yields provides insight into mechanisms of particle production in hadronic collisions. In this report we present the relative yield of D mesons as a function of the relative charged-particle multiplicity in \ensuremath{\mathrm {p\kern-0.05emp}}\ collisions at $\sqrt{s}$ = 7 TeV \cite{ppcollisions} and in \ensuremath{\mbox{p--Pb}}\ collisions at $\sqrt{s_{NN}}$ = 5.02 TeV \cite{pPbcollisions}. \newline
At the LHC, ALICE \cite{alice} comprises central detectors, used for the measurement of charged particles as well as photons and a forward muon spectrometer which measures muons. The L3 solenoid magnet provides a field of 0.5 T to the central barrel detectors covering the rapidity -0.9 $<$ $\textit{y}$ $<$ 0.9. The Silicon Pixel Detector (SPD) is used for the determination of the interaction vertex as well as the measurement of charged-particle multiplicity. Particle identification information is given by the Time Of Flight (TOF) and the Time Projection Chamber (TPC). Two arrays of scintillator detectors (V0A and V0C) placed on both sides of the interaction point provide centrality in \ensuremath{\mbox{Pb--Pb}}\ collisions as well as trigger information. They are also used for multiplicity determination. The forward muon spectrometer covers the rapidity range -4 $<$ $\textit{y}$ $<$ -2.5. It consists of a composite absorber, a dipole magnet, five tracking stations, a muon filter and two trigger stations. The absorber reduces background muons from the decays of pions and kaons. The dipole magnet provides a horizontal magnetic field perpendicular to the beam axis and is used for charge and momentum determination. The tracking stations are used to determine the trajectories of the muons traversing the detector. The muon filter is mounted at the end of the spectrometer and filters all particles with momentum $\textit{p}$ $<$ 4 GeV/c. The trigger stations are used for muon identification and triggering.\newline
The reconstruction of D mesons is done via their hadronic decay channels: D$\textsuperscript{0}$ $\rightarrow$ \ensuremath{{\rm K}^{-}}\ensuremath{\pi^{-}} (branching ratio, BR = 3.88$\pm$0.05 \%), D$\textsuperscript{+}$ $\rightarrow$ \ensuremath{{\rm K}^{-}}\ensuremath{\pi^{+}}\pip (BR = 9.13$\pm$0.19 \%), D$\textsuperscript{*+}$ $\rightarrow$ D$\textsuperscript{0}$\ensuremath{\pi^{+}}\ (BR = 67.7$\pm$0.05 \%) followed by D$\textsuperscript{0}$ $\rightarrow$ \ensuremath{{\rm K}^{-}}\ensuremath{\pi^{+}} \cite{branchingratios} and their charge conjugates at central rapidity. Further details on D meson and charged-particle multiplicity measurements are reported in \cite{ppcollisions} and \cite{ppcollisions1}. The measurement of prompt and non-prompt (produced via the decay of beauty hadrons) J/$\psi$ $\rightarrow$ e$\textsuperscript{+}$e$\textsuperscript{+}$ (BR = 5.97$\pm$0.03 \%) and J/$\psi$ $\rightarrow$ $\mu$$\textsuperscript{+}$$\mu$$\textsuperscript{+}$ (BR = 5.96$\pm$0.03 \%) at central and forward rapidity are reported in \cite{jpsi}.
\
\section{Results}
\subsection{$\textbf{D-meson production in $\textit{pp}$ collisions}$}
The results of the relative yields (ratio between the yield in a given multiplicity interval normalised to the
multiplicity-integrated one) of D mesons, D$\textsuperscript{0}$, D$\textsuperscript{+}$, D$\textsuperscript{*+}$, measured at central rapidity as a function of relative charged-particle multiplicity (multiplicity of charged particles normalised to the average value) in \ensuremath{\mathrm {p\kern-0.05emp}}\ collisions at $\sqrt{s}$ = 7 TeV are presented in Figure \ref{results1}. In Figure \ref{fig:first_1}, the prompt D-meson relative yield is compared with that of inclusive J/$\psi$ as a function of the relative multiplicity. The yields of D mesons and inclusive J/$\psi$ show a similar increase with the multiplicity, in particular for J/$\psi$ measured at central rapidity. In Figure \ref{fig:first_2} the yield of D mesons is compared to that of non-prompt J/$\psi$ (produced in the decay of beauty hadrons), which provides an indirect measurement of the multiplicity dependence of beauty hadrons. The trend is similar to that observed in Figure \ref{fig:first_1}. We observe no difference, within uncertainty, in the multiplicity dependence of charm and beauty hadron yields at central rapidity.
\begin{figure}[!h]
\centering
\subfigure[]
{
\includegraphics[width=200px,height=180px]{jpsidmeson.pdf}
\label{fig:first_1}
}
\subfigure[]
{
\includegraphics[width=200px,height=180px]{2015-Aug-21-AverageD_NonPromtJpsi_Pt2to4.pdf}
\label{fig:first_2}
}
\caption{Average D meson yield as a function of the relative charged-particle multiplicity in \ensuremath{\mathrm {p\kern-0.05emp}}\ collisions at $\sqrt{s}$ = 7 TeV, measured at central rapidity in the interval 2 $<$ $\textit{p}$$_\text{T}$ $<$ 4 GeV/c compared to the yields of inclusive J/$\psi$ (a) and non-prompt J/$\psi$ (b) for $\textit{p}$$_\text{T}$ $>$ 0 GeV/c \cite{ppcollisions, jpsi}. The vertical bars and the boxes represent the statistical and systematic uncertainties respectively. The bottom panels show the B feed-down fraction uncertainty.}
\label{results1}
\end{figure}
\begin{figure}[!h]
\centering
\end{figure}
\subsection{$\textbf{D-meson production in \ensuremath{\mbox{p--Pb}}\ collisions}$}
Figure \ref{results3} shows the prompt D-meson yield as a function of the relative multiplicity measured at central rapidity (-0.9 $<$ $\textit{y}$ $<$ 0.9) compared with J/$\psi$ in \ensuremath{\mbox{p--Pb}}\ collisions at $\sqrt{s_{NN}}$ = 5.02 TeV at forward (p going) and backward (Pb going) rapidity. The yield of D mesons increases faster than the J/$\psi$ yield towards high multiplicity, particularly for J/$\psi$ measured at forward rapidity.
\begin{figure}[!h]
\centering
\includegraphics[width=200px,height=180px]{pPbJPsiandDmeson}
\label{fig:first_4_1}
\caption{Relative yields of the average of the D-meson species and J/$\psi$ in the forward (p going) and backward (Pb going) rapidity as a function of the relative charged-particle multiplicity \cite{pPbcollisions, jpsi}. The vertical bars and the boxes represent the statistical and systematic uncertainties respectively. The bottom panel shows the B feed-down fraction uncertainty.}
\label{results3}
\end{figure}
\subsection{$\textbf{Comparison of D-meson measurements in $\textit{pp}$ and \ensuremath{\mbox{p--Pb}}\ collisions}$}
The yield of D mesons in \ensuremath{\mathrm {p\kern-0.05emp}}\ collisions at $\sqrt{s}$ = 7 TeV is compared to the yield in \ensuremath{\mbox{p--Pb}}\ collisions at $\sqrt{s_{NN}}$ = 5.02 TeV as a function of charged-particle multiplicity in Figure \ref{results5}. The D-meson yield increases faster than linear with multiplicity for both \ensuremath{\mathrm {p\kern-0.05emp}}\ and \ensuremath{\mbox{p--Pb}}\ collisions. The increase is independent of the colliding system.
\begin{figure}[!h]
\begin{center}
\includegraphics[width=200px,height=180px]{hfvsmultpppPb.jpeg}
\caption{The relative yields of the averages of the three D-meson species as a function of the relative charged-particle multiplicity in \ensuremath{\mathrm {p\kern-0.05emp}}\ and \ensuremath{\mbox{p--Pb}}\ collisions at $\sqrt{s}$ = 7 TeV and $\sqrt{s_{NN}}$ = 5.02 TeV, respectively \cite{pPbcollisions}. The vertical bars and the boxes represent the statistical and systematic uncertainties respectively. The bottom panel shows the B feed-down fraction uncertainty.}
\label{results5}
\end{center}
\end{figure}
\subsection{$\textbf{Comparison of results with theoretical models}$}
The comparison of the measurements in \ensuremath{\mathrm {p\kern-0.05emp}}\ and \ensuremath{\mbox{p--Pb}}\ collisions with theoretical calculations is shown in Figure \ref{results6}. In \ensuremath{\mathrm {p\kern-0.05emp}}\ collisions the yield of D mesons as a function of the charged-particle multiplicity is compared with theoretical predictions based on percolation \cite{percolation, percolation1}, EPOS \cite{epos} with and without hydrodynamical effects and PYTHIA 8 \cite{pythia8}. All the models predict an increase of the relative yield of D mesons with the relative multiplicity for both \ensuremath{\mathrm {p\kern-0.05emp}}\ and \ensuremath{\mbox{p--Pb}}\ collisions. However, percolation and EPOS+hydro predict a faster than linear increase while PYTHIA 8 and EPOS are consistent with the data at lower multiplicity. In \ensuremath{\mbox{p--Pb}}\ collisions, the results are compared with EPOS+hydro and EPOS. Both models predict an increase of the relative yield with the multiplicity. EPOS+hydro predicts a faster than linear increase.
\begin{figure}[!h]
\centering
\subfigure[]
{
\includegraphics[width=190px,height=190px]{ppDmesonmodels.pdf}
\label{fig:first_5}
}
\subfigure[]
{
\includegraphics[width=190px,height=190px]{2015-Sep-23-D_data_theory_summary_17092015.pdf}
\label{fig:first_6}
}
\caption{Relative yields of D mesons in different $\textit{p}$$_\text{T}$ intervals as a function of the relative charged-particle multiplicity in \ensuremath{\mathrm {p\kern-0.05emp}}\ collisions at $\sqrt{s}$= 7 TeV (a) and \ensuremath{\mbox{p--Pb}}\ collisions at $\sqrt{s}$ = 5.02 TeV (b). The lines represent various theoretical models \cite{percolation} - \cite{pythia8}.}
\label{results6}
\end{figure}
\section{Conclusion}
The ALICE Collaboration measured the relative yield of D mesons as a function of the charged-particle multiplicity in \ensuremath{\mathrm {p\kern-0.05emp}}\ and \ensuremath{\mbox{p--Pb}}\ collisions and compared the results with analogous measurements of inclusive and non-prompt J/$\psi$. In both colliding systems the relative yield of D mesons increases with multiplicity and the increase is faster than linear. The increase is independent of the $\textit{p}$$_\text{T}$ interval considered. J/$\psi$ yields show a similar increase in \ensuremath{\mathrm {p\kern-0.05emp}}\ collisions, however in \ensuremath{\mbox{p--Pb}}\ collisions D-meson yields increase faster than J/$\psi$, particularly at forward rapidity. The results were also compared with theoretical calculations. All models considered in the study predict an increase of heavy-flavour yields with multiplicity. The multiplicity dependence of D-meson production is described better by calculations that include hydrodynamics, especially at higher multiplicity.
\section*{References}
| fb3684b62d10a08ac9454f8bd7ad3899125aed32 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction\label{sec:intro}}
In the United States, bladder cancer is the fourth and ninth most
widespread cancer among males and females, respectively, whereas in
Europe, approximately 50,000 people die from this disease each year
\cite{Alcaraz:2007}. Due to the high recurrence rate of $50\%$,
lifetime monitoring of patients is required after surgical removal of
cancer tumors \cite{Holzbeierlein:2004}. During a cystoscopy, a
flexible or rigid endoscope is inserted into the bladder through the
urethra, and a video of the epithelium is visualized on a screen. The
clinician (urologist or surgeon) scans the epithelium by maintaining
the distal tip of the cystoscope close to the internal organ wall. The
resulting highly resolved videos yield accurate visualization of the
wall through small field of view (FOV) images. However, neither
complete regions of interest (multifocal lesions, scares, etc.) nor
anatomical landmarks (urethra, ureters, air bubbles) are visible in
these images. Therefore, lesion diagnosis and follow-up may be
difficult and time-consuming. Extended FOV mosaics can be obtained by
superimposing the regions which can be jointly observed in small FOV
images. Not only do they facilitate lesion diagnosis and follow-up,
but they are also useful for data archiving (videos are highly
redundant data) and medical traceability (videos are not easy to
interpret by a clinician having not performed the acquisition).
\subsection{Bladder image mosaicing}
Bi-dimensional (2D) bladder image mosaicing algori\-thms were proposed
for the standard white-light modality \cite{Bergen13,Miranda2008} and
for the fluorescence modality \cite{Behrens:2008}. A complete
mosaicing algorithm consists of \emph{(i)} the registration of
consecutive images of the video sequence~\cite{Hernandez:2010},
\emph{(ii)} the correction of image misalignment in the mosaic leading
to bladder texture discontinuities \cite{Weibel:2012a}, and
\emph{(iii)} the contrast, intensity and/or color correction to handle
image blurring, shading effects and/or instrument viewpoint changes
\cite{Behrens:2010,Weibel2012b}. In \cite{Weibel:2012a}, it was shown
that mosaics covering large bladder surfaces (\emph{e.g.}, half of the
organ) can be computed. Although such 2D mosaics (see
Fig.~\ref{fig:1}) facilitate lesion diagnosis and follow-up, these
representations can be improved. Indeed, only the first image of the
mosaic has the original image resolution, the resolution of the other
images depending on the instrument trajectory and orientation.
Moreover, these images are strongly distorted during their placement
into the 2D map due to perspective changes of the endoscope. Both
image distortion and changes of resolution lead to mosaics with
spatially-varying visual quality. At last, urologists or surgeons
represent themselves the bladder in the three-dimensional (3D)
space. Mosaics in 3D (\emph{i.e.}, bladder wall surfaces with
superimposed 2D image textures) not only match this mental 3D
representation, but allow for a bladder texture representation without
any image distortion and with preserved resolution. Moreover, virtual
navigation inside the reconstructed organ part is possible after
clinical examinations using 3D mosaics. The interest of 2D and 3D
endoscopic image mosaicing in various medical fields is discussed in
the overview paper~\cite{Bergen16}.
\subsection{Previous work in 3D endoscopic data mosaicing}
In various fields of medical endoscopy, 3D mosaics are built following
three kinds of approaches.
\begin{figure}[t]
\centering
\psfig{figure=fig1-eps-converted-to.pdf, width=\linewidth}
\caption{Bladder mosaic (clinical data) obtained with the registration
algorithm~\cite{Weibel:2012a} and using the
algorithm~\cite{Weibel2012b} for correction of illumination
discontinuities. The first image of the sequence is at the top
right. From this image position, the endoscope first moves downwards
in the bright regions (dashed bounding box) with weak
textures. Then, its trajectory goes up towards the last image at the
top left. Strong perspective and scale changes arise between the
first and last images.}
\label{fig:1}
\end{figure}
In the first approach, analytically known surface shapes (representing
organs) are used both to limit the 2D registration errors and
visualize textured surfaces. In \cite{Behrens:2009a}, bladder
hemispheres were approximated by hemi\-cubes whose five sides
represent quasi planar bladder regions. A video sequence was then
partitioned into image groups, each group corresponding to some
hemicube side. The five reconstructed 2D sub-mosaics were then
projected onto the hemisphere to obtain a 3D bladder representation.
As for the inner esophagus wall~\cite{Carroll:2009}, the organ was
approximated by tubular shapes. The registration method aligns
endoscopic images using jointly the image colors and a mosaicing
surface of known geometry to reduce distortions. This algorithm
provides both the endoscope trajectory and a 2D mosaic that can be
projected onto the inner cylinder surfaces. Such approach based on 3D
prior models is well-suited to the esophagus which can indeed be
modelled by tubes, but is less appropriate for 3D bladder
mosaicing. Indeed, the bladder shape is hardly predictable, notably
since it is deformed by neighboring organs putting pressure on
it. Moreover, grouping images with respect to hemicube sides is
time-consuming and difficult.
In the second approach, no 3D prior information is imposed,
\emph{i.e.}, only video-images are taken into account. Different
passive vision methods were proposed based on shape from shading (SfS
\cite{Wu:2010}, bone surface reconstruction), structure from motion
(SfM \cite{Soper2012}, cystoscopy), shape from motion associated to
SfS (\cite{Kaufman:2008}, colon surface reconstruction),
SfM-factorization based on surgical instrument tracking
(\cite{Wu:2007}, surgery) or simultaneous localization and mapping
methods (SLAM \cite{Mountney2009}, laparoscopy for minimal invasive
surgery).
The lack of intensity variation in standard bladder images impedes
SfS. However, SfM or SLAM methods may be appropriate in urology. In
their 3D bladder mosaicing feasibility study~\cite{Soper2012}, Soper
\emph{et al} replaced the endoscope standardly used in urology by an
ultrathin fiber. The fiber was translated and rotated in such a way
that a complete bladder scan is performed. The acquired image sequence
was first used to find point correspondences in image pairs using
homologous feature points. Then, the 3D surface was constructed by
displacing a set of points initially located on a sphere.
Specifically, a bundle adjustment method was used to iteratively
determine both fiber tip trajectory and pig bladder phantom surface.
The main limitation of this approach for standard cystoscopy is that
salient texture points must be extracted from the video-images. In
large human bladder regions, textures may be missing or very weakly
contrasted for both healthy (see the bottom of the mosaic in
Fig.~\ref{fig:1}) and damaged tissues, \emph{e.g.}, due to scares. In
such images, SfM or SLAM techniques based on SIFT or SURF approaches
for extracting and matching textures thus suffer from a lack of
robustness (see \cite{Ali:2013,Hernandez:2010} for a detailed
discussion on feature based techniques in cystoscopy).
Feature point extraction and matching techniques is also difficult
because of illumination changes occurring between images, which are
due to strong perspective changes and/or the vignetting effect of
endoscopes. Another major limitation of feature-based methods is blur,
which is commonly encountered in standard cystoscopy images.
A less investigated approach in 3D endoscopy is based on active
vision, where surface points are reconstructed in the small FOV of
modified endoscopes. In \cite{Chan2003}, a two-channel endoscope was
used to reconstruct the inner surface of the mouth. The first channel
projects a structured-light pattern whereas the second acquires
video-images. Classical triangulation methods were used to reconstruct
3D points. Tests on phantoms have shown that submillimetre point
reconstruction accuracy can be reached with a small baseline of 2
mm. Another acquisition setup \cite{Penne2009} consists of a classical
CCD-camera and a time of flight (ToF) camera providing a depth map of
the FOV. A calibration procedure provides for each pixel both a color
and a 3D position with millimetre accuracy.
In \cite{Penne2009}, a two-channel endoscope (laparoscope) was used to
reconstruct points in a pig stomach. It is noticeable that such camera
pair can be used as well for single channel endoscopes such as
cystoscopes. Indeed, a ToF camera uses infrared signals that
do not interfere with the visible light of a color camera.
Shevchenko \emph{et al} proposed a structured-light approach for
obtaining extended 3D bladder surfaces~\cite{Shevchenko:2012}.
The prototype in~\cite{Shevchenko:2012} consists of an external
navigation system (which localizes the endoscope position in the
examination room) coupled with an endoscopic system projecting a grid
of points on the inspected surface. The 3D surface was built by
computing a 2D mosaic from the CCD camera images, and exploiting the
information provided by the external system. Knowing the endoscope
position and the 2D image correspondence, the 2D mosaic has to be
projected onto the reconstructed surface.
It is noticeable that in \cite{Agenant:2013}, the authors also
proposed a navigation system based method to document the position of
a lesion into the bladder during a first cystoscopy. The aim was not
3D image mosaicing, but rather to facilitate follow-up by performing
fast localization of lesions in upcoming cystoscopies.
In the conference paper \cite{Ben-Hamadou:2010}, we proposed the first
active vision based approach for 3D bladder mosaicing. Based on a
similar structured light principle as that described later in
\cite{Shevchenko:2012}, the proposed solution did however not require
any external navigation system for endoscope localization. The 3D
surface construction algorithm was based on 2D image registration and
on the reconstruction of 3D points in the coordinate system of the
endoscope camera. However, the proof of concept of 3D cystoscopy was
only partly established in \cite{Ben-Hamadou:2010}. The algorithm was
not tested using data (2D images and 3D points) acquired with the
structured light principle embedded in a cystoscope. The related data
were acquired for a large baseline (distance between the focal lengths
of the camera and the light projector) of about 150 mm. This approach
was not yet evaluated in the less favorable case of small baselines of
2 to 3 mm, like in cystoscopy. Furthermore, the image processing
algorithm in \cite{Ben-Hamadou:2010} needed to be improved. It was
based on a registration step, where the squared difference between
image grey-levels was minimized. This approach is not robust when
textures are weakly contrasted or with high inter- and intra-patient
variability. Such similarity measure is also not appropriate to the
illumination changes occurring in cystoscopic data.
\begin{figure*}[t]
\begin{tabular}{cc}
\hspace{-1mm}\includegraphics[width=0.45\textwidth]{fig2a-eps-converted-to.pdf}
&
\hspace{9mm}\includegraphics[width=0.45\textwidth]{fig2b-eps-converted-to.pdf}
\\
(a)& (b)
\end{tabular}
\caption{Acquisition principle~(a) and available data~(b) for 3D
mosaicing. The acquisition principle is implemented using the
prototype of Section~\ref{sec:ProtoResultsDiscussion}.\, (a) The
$i$-th ``projector ray'' is shown in green. The corresponding
``camera ray'' is the black line originating from point
$P_{3D}^{i,k}$ of the surface (intersection of the projector and
camera rays) and passing through the optical center $O_{c_k}$ of the
camera. $P_{2D}^{i,k}$ is the intersection of the camera ray with
the plane supporting the acquired image $I_k$.\, (b) The available
data are represented for the first and $k$-th acquisitions. The
camera coordinate system $\{c_k$\} reads $(O_{c_k},
\protect\overrightarrow{x_{c_k}}, \protect\overrightarrow{y_{c_k}},
\protect\overrightarrow{z_{c_k}})$ and moves together with the
prototype. The black line in bold represents the camera focal
distance $f$, calibrated as described in \cite{Ben-Hamadou:2013}.}
\label{fig:2}
\end{figure*}
\subsection{Objectives\label{sec:objectives}}
The medical objective of the scanning fiber endoscope described in
\cite{Soper2012} was to design a procedure where bladder walls can be
fully acquired by nurses or ancillary care providers. This allows for
sparing the time of urologists who can post-operatively review the
bladder data. Such procedure is currently not standard but may be
complementary to the usual examination with flexible or rigid
cystoscopes. In the present paper, we aim to establish the proof of
concept of bladder mosaicing using 3D cystoscopes based on active
vision, which is a promising approach as shown
in~\cite{Shevchenko:2012}. In \cite{Soper2012}, the distal tip of the
fiber follows a spiral shaped trajectory and the distance from the
optical center to the bladder wall is in average larger than the
acquisition distance with cystoscopes. This allows for the use of the
SfM approach. This is no longer true with cystoscopy, since depth
disparity is often not guaranteed. Moreover, the lack of textures (as
in the bounding box of Fig. \ref{fig:1}) also complicates SfM
approaches, which require robust and accurate homologous point
correspondence between images.
In the present 3D bladder mosaicing feasibility study, an active
vision solution is chosen. As compared to \cite{Shevchenko:2012}, our
goal is to show that a 3D mosaicing algorithm guided by image
registration is useful for retrieving the endoscope displacement
between two acquisitions without using external navigation systems.
From the instrumentation viewpoint, we show that 3D bladder mosaicing
is feasible by simply modifying standard cystoscopes. Although the
proposed prototype cannot currently be used in clinical situation, our
3D point reconstruction approach can be implemented on cystoscopes.
Furthermore, it was shown that 3D point reconstruction is accurate
enough, even for small baselines of point triangulation
\cite{Ben-Hamadou:2013,Chan2003}.
From the image processing viewpoint, we propose a 3D image mosaicing
algorithm for which no assumption has to be made regarding the
endoscope trajectory. The unknown endoscope displacement between
consecutive image acquisitions is a general combination of a 3D
rotation and a 3D translation. Although real-time mosaicing can be of
interest, the computation time is not the most critical point in
urology. As argued in \cite{Miranda2008,Weibel:2012a}, a mosaicing
time of some tens of minutes is acceptable for a second (finer)
diagnosis carried out to confirm or modify the first diagnosis during
the real-time visualization of the small FOV video-images. Real-time
is neither a critical factor for lesion follow-up because the lesion
evolution assessment is performed with data from a cystoscopy
scheduled some weeks or months after the cystoscopy in progress.
The paper is organized as follows. Section \ref{sec:AcquisitionData}
introduces the data acquisition principle (for readability reasons,
the description of the prototype implementing this measurement
principle is deferred to Section~\ref{sec:ProtoResultsDiscussion}).
Section~\ref{sec:3Dmosaicing} details the 3D mosaicing algorithm.
In Section~\ref{sec:ProtoResultsDiscussion}, the algorithm is
tested on phantoms with various shapes allowing for a validation
of the algorithm with data acquired by the proposed 3D cystoscope prototype.
In Section~\ref{sec:SimulatedDataResultsDiscussion}, simulated data allow
for qualitative evaluation on more realistic bladder data, in terms of
both 3D surface shape and bladder texture.
Section \ref{sec:3DRegistrationError} exhibits the 3D registration
errors obtained for both synthetic data and data related to the
prototype. Concluding remarks and perspectives are given in
Section~\ref{sec:Conclusion}.
\section{Acquisition principle and available data \label{sec:AcquisitionData}}
The principle of our laser based active vision system is sketched in
Fig.~\ref{fig:2}(a). The prototype is equipped with a color camera and
a structured laser light projector. The basic idea is to project laser
rays, such as the green line of Fig.~\ref{fig:2}(a), onto the bladder
epithelial surface. Then, each projector ray is being reflected by the
surface. This generates a so-called ``camera ray'' (the black line of
Fig.~\ref{fig:2}(a)) passing through the green laser dots visible in
image $I_k$. In Fig.~\ref{fig:2}(a), the quadrangle with black
background corresponds to diffractive optics, \emph{i.e.}, holographic
binary phase lens, designed to project eight laser rays. The latter
are located on a cone originating from the projector optical
center. Each projector ray passes through one of the eight white dots
of the diffractive optics. For readability matters, only one ray is
shown in Fig.~\ref{fig:2}(a).
The calibration procedure in \cite{Ben-Hamadou:2013} is used to
compute the intrinsic and extrinsic camera and projector
parameters. The parameters of the projector ray equations are computed
in the camera coordinate system $(O_{c_k}, \overrightarrow{x_{c_k}},
\overrightarrow{y_{c_k}}, \overrightarrow{z_{c_k}})$, referred to as
$\{c_k$\}, with $O_{c_k}$ the optical center of the camera. Because
the projector and camera are rigidly fixed together, the equations of
the projector rays in $\{c_k$\} do not depend on $k$. The equations of
the camera rays (passing all through $O_{c_k}$) are also computed in
$\{c_k$\} given the location $P_{2D}^{i,k}$ of the green dots in image
$I_k$ ($i$ stands for the index of the green dots). The green dots can
be easy segmented since the hue of the bladder is always reddish or
orange, see the real images of Figs.~\ref{fig:1} and
\ref{fig:2}(b). The $i$-th laser point on the bladder surface
($P_{3D}^{i,k}$) is the intersection of the projector ray with its
corresponding camera ray.
The data available for 3D mosaicing are related to a video-sequence
with $K$ frames. Each acquisition $k \in \{1,\ldots,K\}$ corresponds
to a cystoscope viewpoint and yields: \emph{(i)}~an image $I_k$;
\emph{(ii)}~eight laser dots $P_{2D}^{i,k}=(x_{2D}^{i,k},
y_{2D}^{i,k})^T$ given in the coordinate system $(O_{I_k},
\overrightarrow{x_{I_k}},\\ \overrightarrow{y_{I_k}}$) of the plane
supporting $I_k$; \emph{(iii)}~eight laser points
$P_{3D}^{i,k}=(x_{3D}^{i,k}, y_{3D}^{i,k}, z_{3D}^{i,k})^T$ given in
the camera coordinate system $\{c_k$\} (see Fig.~\ref{fig:2}(b)). The
barrel distortions in images $I_k$ are corrected using the known
intrinsic camera parameters
\cite{Ben-Hamadou:2013}. The laser rays ($i \in \{1,\ldots,M\}$ with
$M=8$) are projected in the periphery of the circular FOV of the
images to minimize the loss of bladder texture and to allow for
diagnosis. Although only eight $P_{3D}^{i,k}$ points are available per
viewpoint, the video acquisition leads to a large number of points
overall due to the high acquisition speed (25 images/s) and the slow
endoscope displacements (some mm/s).
The algorithm proposed in Section~\ref{sec:3Dmosaicing} aims at
placing all points $P_{3D}^{i,k}$ in a common coordinate system, and
then computing a polygonal mesh representing the smooth bladder surface,
without corners nor sharp edges. The textures of images $I_k$ will
finally be projected onto the polygonal mesh.
\section{Construction of three-dimensional textured surfaces \label{sec:3Dmosaicing}}
The proposed 3D mosaicing method takes as inputs a set of
distortion-corrected images $I_k$ and a set of $M=8$ points per
viewpoint $k$. Each $P_{3D}^{i,k}$ point is given in the local
(moving) coordinate system $\{c_k$\} of the camera. Moreover, their 2D
projections $P_{2D}^{i,k}$ in the $k$-th image of the video-sequence
($i=1,\ldots,M$) are known as well. The algorithm for reconstructing
the $P_{3D}^{i,k}$ points from the knowledge of their projections
$P_{2D}^{i,k}$ can be found in~\cite{Ben-Hamadou:2013}.
The mosaicing algorithm requires some geometric prerequisites, which
are presented now.
\subsection{Geometrical considerations \label{sec:geometry}}
The first geometrical transformation involves the intrinsic camera
parameters linking the 3D camera coordinate system $\{c_k$\} to the 2D
image plane coordinate system $(O_{I_k}, \overrightarrow{x_{I_k}},
\overrightarrow{y_{I_k}}$). In~\eqref{eq:matricePerspective}, the
perspective projection matrix ${\mathbf{K}}$ is fully defined from the
knowledge of the projection $(u, v)$ of the camera optical center in
image $I_k$, the focal length $f$, and the CCD-sensor size $l_x$ and
$l_y$ along the $\overrightarrow{x_{I_k}}$ and
$\overrightarrow{y_{I_k}}$ axes (see Fig.~\ref{fig:2}(b)):
\begin{align}
\left[
\setlength\arraycolsep{3.0pt}
\begin{array}{c} P_{2D}^{i,k} \\ 1
\end{array}\right]
=\,
\frac{1}{z_{3D}^{\,i,k}} \, {\mathbf{K}} \, P_{3D}^{i,k} \,\,\,
\textrm{with} \,\,\,\,
\label{eq:matricePerspective}
{\mathbf{K}}=
\left[
\setlength\arraycolsep{1.0pt}
\begin{array}{ccc}
f/l_x & 0 & u \\
0 & f/l_y & v \\
0 & 0 & 1
\end{array}\right].
\end{align}
The second transformation is a 3D rigid transformation linking
consecutive camera viewpoints. It is parameterized by a $3\times 4$
matrix $\mathbf{T}_{3D}^{k-1,k}$. $\mathbf{T}_{3D}^{k-1,k}$ is
composed of a $3\times 3$ rotation matrix $\mathbf{R}^{k-1,k}$ and a
$3\times 1$ translation vector $D^{k-1,k}$:
\begin{align}
\hat{P}_{3D}^{i,k-1}
=
\mathbf{T}_{3D}^{k-1,k} \,
\!\!\left[\begin{array}{c} P_{3D}^{\,i,k}\\1
\end{array}\right],
\label{eq:matriceTrigidea}%
\end{align}
with
\begin{align}
\mathbf{T}_{3D}^{k-1,k}
=
\left[
\setlength\arraycolsep{2pt}
\begin{array}{c|c}
\mathbf{R}^{k-1,k} & D^{k-1,k}
\end{array}\right].
\label{eq:matriceTrigideb}%
\end{align}
The coefficients of the rotation matrix are defined from the
well-known Euler angles, denoted by $\theta_1^{k-1,k}$,
$\theta_2^{k-1,k}$ and $\theta_3^{k-1,k}$. The so-called
``$x$-convention'' is used for these angles: $\theta_1^{k-1,k} \in
[0,2\pi]$ is the angle around the $\overrightarrow{z_{c_k}}$-axis,
$\theta_2^{k-1,k}\in [0, \pi]$ is around the new (rotated)
$\overrightarrow{x_{c_k}}^{\,'}$-axis, while $\theta_3^{k-1,k}\in
[0,2\pi]$ is around the rotated
$\overrightarrow{z_{c_k}}^{\,''}$-axis. Superscripts $'$ and $''$
respectively refer to the new location of axes
$\overrightarrow{x_{c_k}}$ and $\overrightarrow{z_{c_k}}$ after the
first and second rotations. The matrix $\mathbf{T}_{3D}^{k-1,k}$
displaces point $P_{3D}^{i,k}$ in $\{c_k\}$ to
$\hat{P}_{3D}^{i,k-1}=(\hat{x}_{3D}^{i,k-1}, \hat{y}_{3D}^{i,k-1},
\hat{z}_{3D}^{i,k-1})^T$ in $\{c_{k-1}$\}. It is worth noticing that
$\hat{P}_{3D}^{i,k-1}$ are points brought wi\-thin $\{c_{k-1}$\}
whereas $P_{3D}^{i,k-1}$ are ``data points'' reconstructed in
$\{c_{k-1}\}$ using the calibration parameters and the segmented dots
$P_{2D}^{i,k-1}$. Thus, $P_{3D}^{i,k-1}$ and $\hat{P}_{3D}^{i,k-1}$
represent different laser points.
The third relationship required by the proposed me\-thod is a
homography linking two consecutive images $I_{k}$ and $I_{k-1}$. This
2D relationship is established under several working
assumptions. First, the distal tip of the cystoscope is usually close
to the inner bladder epithelium. The FOV being very limited, the
surfaces viewed in the images are usually small and are assumed to be
planar. Second, the bladder is filled by an isotonic saline solution
which rigidifies the bladder wall. The acquisition speed being high
(25 image/s) and the endoscope displacement speed low (some mm/s), we
assume that the common surface part viewed in images $I_{k-1}$ and
$I_{k}$ does not warp itself between two consecutive
acquisitions. Based on these working assumptions, the homography is
well-suited to model the dependence between $I_{k}$ and
$I_{k-1}$. This choice was practically validated for numerous
algorithms for 2D/2D registration of bladder images: feature-based
methods in \cite{Behrens:2009a} (fluorescence modality) and
\cite{Bergen13,Soper2012} (white-light modality), the graph-cut method
in \cite{Weibel:2012a} and two optical flow methods in
\cite{Ali16b,Ali16a}. The homography matrix is defined by:
\begin{align}
\mathbf{T}_{2D}^{k-1,k} = \left[\begin{array}{ccc}
\underbrace{\alpha^{k-1,k}\cos\varphi^{k-1,k}}_{a^{k-1,k}_{1,1}} &
\underbrace{-s_x^{k-1,k}\sin\varphi^{k-1,k}}_{a^{k-1,k}_{1,2}} &
\underbrace{t^{k-1,k}_{x, 2D}}_{a^{k-1,k}_{1,3}} \\
\underbrace{s_y^{k-1,k}\sin\varphi^{k-1,k}}_{a^{k-1,k}_{2,1}} &
\underbrace{\alpha^{k-1,k}\cos\varphi^{k-1,k}}_{a^{k-1,k}_{2,2}} &
\underbrace{t^{k-1,k}_{y, 2D}}_{a^{k-1,k}_{2,3}} \\
a_{3,1}^{k-1,k} &
a_{3,2}^{k-1,k} & 1
\end{array}\right]
\label{eq:matriceHomography}
\end{align}
where the parameters $\alpha$, $\varphi$, $(s_x, s_y)$ and $(t_{x,2D},
t_{y,2D})$ respectively denote the scale factor, in-plane rotation,
shearing and 2D translation changes. In the following and for
simplicity reasons, the matrix $\mathbf{T}_{2D}^{k-1,k}$ will be
simply described by means of its coefficients $a^{k-1,k}_{r,s}$.
Image $I_k$ is registered with $I_{k-1}$ when
$\mathbf{T}_{2D}^{k-1,k}$ superimposes homologous points $P_{2D}^{k}$
and $P_{2D}^{k-1}$ of both images, \emph{i.e.}, when
\begin{align}
\left[\begin{array}{c} P_{2D}^{k-1} \\ 1
\end{array}\right]
&\!\!=\!
\frac{1}{\beta^{k-1,k}} \,\, \mathbf{T}_{2D}^{k-1,k}
\left[\begin{array}{c} P_{2D}^{k} \\ 1
\end{array}\right].
\label{eq:PointsHomologues}
\end{align}
The normalizing factor $\beta^{k-1, k}$ deduces from the perspective
parameters $a_{3,1}^{k-1,k}$ and $a_{3,2}^{k-1,k}$:
\begin{align}
\beta^{k-1,k}=[a_{3,1}^{k-1,k},a_{3,2}^{k-1,k}]^TP_{2D}^{k}+1.
\label{eq:beta}
\end{align}
Let us stress that \eqref{eq:PointsHomologues} is met for all
homologous texture pixels of images $I_k$ and $I_{k-1}$
($P_{2D}^{k-1}$ and $P_{2D}^{k}$), but not for the laser points
$P_{2D}^{i,k}$ and $P_{2D}^{i,k-1}$. Indeed, $P_{2D}^{i,k}$ and
$P_{2D}^{i,k-1}$ are not homologous points because the active vision
system is moving from viewpoint $k-1$ to viewpoint $k$. Nevertheless,
\eqref{eq:PointsHomologues} will be exploited to estimate the
parameters of a candidate homography $\mathbf{T}_{2D}^{k-1,k}$
corresponding to a candidate $\mathbf{T}_{3D}^{k-1,k}$
transformation. Moreover, the quality of image superimposition will be
assessed based on \eqref{eq:PointsHomologues}, applied to all
homologous texture pixels of $I_{k-1}$ and $I_k$.
\setcounter{equation}{12}
\begin{table*}[bp]
\begin{equation}
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\left[
\setlength\arraycolsep{8pt}
\begin{array}{cccccccc}
%
x_{2D}^{1,k} & y_{2D}^{1,k} & 1 & 0 & 0 & 0 &
-c^{k-1,k}_{1,1}x_{2D}^{1,k} & -c^{k-1,k}_{1,1}y_{2D}^{1,k} \\[.2cm]
%
0 & 0 & 0 & x_{2D}^{1,k} & y_{2D}^{1,k} & 1 &
-c^{k-1,k}_{2,1}x_{2D}^{1,k} & -c^{k-1,k}_{2,1}y_{2D}^{1,k} \\
%
\vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \vdots &
\vdots \\
%
x_{2D}^{M,k} & y_{2D}^{M,k} & 1 & 0 & 0 & 0 &
-c^{k-1,k}_{1,M}x_{2D}^{M,k} & -c^{k-1,k}_{1,M}y_{2D}^{M,k} \\[.2cm]
%
0 & 0 & 0 & x_{2D}^{M,k} & y_{2D}^{M,k} & 1 &
-c^{k-1,k}_{2,M}x_{2D}^{M,k} & -c^{k-1,k}_{2,M}y_{2D}^{M,k} \\
\end{array}\right]
%
%
\left[
\setlength\arraycolsep{0pt}
\begin{array}{c}
a_{1,1}^{k-1,k} \\[.1cm] a_{1,2}^{k-1,k} \\[.1cm] a_{1,3}^{k-1,k} \\[.1cm] a_{2,1}^{k-1,k} \\[.1cm]
a_{2,2}^{k-1,k} \\[.1cm] a_{2,3}^{k-1,k} \\[.1cm] a_{3,1}^{k-1,k} \\[.1cm] a_{3,2}^{k-1,k}
\end{array}\right]
%
=
%
%
\left[
\setlength\arraycolsep{0pt}
\begin{array}{c}
c_{1,1}^{k-1,k} \\[.2cm]
c_{2,1}^{k-1,k} \\
\vdots \\
c_{1,M}^{k-1,k} \\[.2cm]
c_{2,M}^{k-1,k}
\end{array}\right]
\label{eq:Lien2D3DMatricielFinal}
%
\end{equation}
\end{table*}
\setcounter{equation}{6}
\subsection{Consecutive viewpoint registration\label{sec:registration}}
\subsubsection{Principle of the registration algorithm}
Let us define the displaced projected laser points
$\hat{P}_{2D}^{i,k-1}$ as the 2D projections of the laser points
$P_{3D}^{i,k}$, displaced from $\{c_{k}$\} to $\{c_{k-1}$\}.
Combining~\eqref{eq:matricePerspective} and~\eqref{eq:matriceTrigidea}
yields:
\begin{align}
\setlength\arraycolsep{0.0pt}
\left[\begin{array}{c}
\hat{P}_{2D}^{i,k-1} \\ 1
\end{array}\right] & = \frac{1}{ \hat{z}_{3D}^{i,k-1}}\;
{\mathbf{K}} \mathbf{T}_{3D}^{k-1,k} \left[
\begin{array}{c} P_{3D}^{i,k}\\ 1
\end{array}\right].
\label{eq:PointsTransforme}
\end{align}
Now, consider an ideal (and searched) $\mathbf{T}_{3D}^{k-1,k}$
transformation, whose related homography $\mathbf{T}_{2D}^{k-1,k}$
superimposes the pixels of $I_k$ and their homologous pixels in
$I_{k-1}$. Such $\mathbf{T}_{2D}^{k-1,k}$ transformation not only
superimposes homologous texture pixels, but also relates each laser
point $P_{2D}^{i,k}$ to its displaced instance $\hat{P}_{2D}^{i,k-1}$:
\begin{align}
\setlength\arraycolsep{0.0pt}
\left[\begin{array}{c} \hat{P}_{2D}^{i,k-1} \\ 1
\end{array}\right] = \frac{1}{\beta^{k-1,k}_i}\:\mathbf{T}_{2D}^{k-1,k}
\left[\begin{array}{c} P_{2D}^{i,k} \\ 1
\end{array}\right],
\label{eq:PointsHomologuesModif}
%
\end{align}
where the factor $\beta^{k-1,k}$ appearing in
\eqref{eq:PointsHomologues} has been replaced by $\beta^{k-1,k}_i$ to
stress the dependence upon $i$.
Combining~\eqref{eq:PointsTransforme} and
\eqref{eq:PointsHomologuesModif} yields:
\begin{align}
\setlength\arraycolsep{0.0pt}
\frac{1}{\beta^{k-1,k}_i}\;\mathbf{T}_{2D}^{\,k-1,k}\left[
\begin{array}{c}
P_{2D}^{\,i,k}\\1
\end{array}\right]
&=\frac{1}{\hat{z}_{3D}^{\,i,k-1}}\,{\mathbf{K}}\;\mathbf{T}_{3D}^{k-1,k}
\left[\begin{array}{c} P_{3D}^{\,i,k}\\1
\end{array}\right]
\label{eq:Lien2D3DMatriciel}
\end{align}
which involves the 2D and 3D laser points related to viewpoint $k$ only.
For a given candidate $\mathbf{T}_{3D}^{k-1,k}$,
$\hat{z}_{3D}^{i,k-1}$ is known from the last equation
of~\eqref{eq:PointsTransforme} and the knowledge of $P_{3D}^{i,k}$.
Similarly, $\beta^{k-1,k}_i$ can be deduced from
$\mathbf{T}_{2D}^{k-1,k}$ using~\eqref{eq:beta}. Therefore,
$\mathbf{T}_{3D}^{k-1,k}$ and $\mathbf{T}_{2D}^{k-1,k}$ are the only
unknowns in~\eqref{eq:Lien2D3DMatriciel}. As we will see,
$\mathbf{T}_{2D}^{k-1,k}$ can actually be deduced from
$\mathbf{T}_{3D}^{k-1,k}$ and the knowledge of the laser points.
The proposed registration scheme is designed in such a way that some
similarity measure (to be detailed) between both images $I_k$ and
$I_{k-1}$ is maximized with respect to the $\mathbf{T}_{3D}^{k-1,k}$
parameters under the constraint formulated
in~\eqref{eq:Lien2D3DMatriciel}. Each iteration of the optimization
algorithm includes four steps related to the update and evaluation of
the candidate $\mathbf{T}_{3D}^{k-1,k}$:
\begin{enumerate}
\item Update the rigid transformation $\mathbf{T}_{3D}^{k-1,k}$.
\item Compute matrix $\mathbf{T}_{2D}^{k-1,k}$ induced by
$\mathbf{T}_{3D}^{k-1,k}$ using \eqref{eq:Lien2D3DMatriciel} and the
location of the $M$ laser points $(P_{3D}^{i,k},P_{2D}^{i,k})$.
\item Superimpose $I_{k-1}$ with $\mathbf{T}_{2D}^{k-1,k}(I_{k})$,
which denotes the transformation of image $I_k$ using the computed
homography $\mathbf{T}_{2D}^{k-1,k}$; see \eqref{eq:PointsHomologues}.
\item Evaluate the similarity measure between images $I_{k-1}$ and
$\mathbf{T}_{2D}^{k-1,k}(I_{k})$ (the pixels of both images
corresponding to laser dots are not taken into account).
\end{enumerate}
This process terminates when the similarity measure between the
registered images is maximal. This algorithm simultaneously provides
the homography $\mathbf{T}_{2D}^{k-1,k}$ and the rigid transformation
$\mathbf{T}_{3D}^{k-1,k}$ linking $\{c_{k}$\} to $\{c_{k-1}$\}.
Hereafter, we detail how $\mathbf{T}_{2D}^{k-1,k}$ is computed from
$\mathbf{T}_{3D}^{k-1,k}$. Then, we define the similarity measure used
for assessing the 2D image superimposition quality. Finally, an
overview of the optimization algorithm is given
(Algorithm~\ref{alg:transformationDetermination}).
\subsubsection{Matrix relationship between $\mathbf{T}_{2D}^{k-1,k}$ and $\mathbf{T}_{3D}^{k-1,k}$\label{sec:matrixrelationship}}
For a point correspondence $(P_{3D}^{i,k}, P_{2D}^{i,k})$,
\eqref{eq:Lien2D3DMatriciel} is a system of three
equations. Rearranging the first two equations and replacing
$\beta^{k-1,k}_i$ and $\hat{z}_{3D}^{\,i,k-1}$ by their expressions,
we obtain:
\begin{align}
\left\{
\begin{array}{ccc}
\frac{\mathbf{T}_{2D,1\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}^{k-1,k}\left[
\setlength\arraycolsep{1.2pt}
\begin{array}{c}P_{2D}^{i,k}\\1\end{array}\right]}
{\mathbf{T}_{2D,3\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}^{k-1,k}\left[\begin{array}{c}P_{2D}^{i,k}\\1\end{array}\right]}
& =
\frac{
{\mathbf{K}}_{1\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}\;
\mathbf{T}_{3D}^{k-1,k}\left[
\setlength\arraycolsep{1.2pt}
\begin{array}{c}P_{3D}^{i,k}\\1\end{array}\right]}
{\mathbf{T}_{3D,3\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}^{k-1,k}\left[
\setlength\arraycolsep{1.2pt}
\begin{array}{c}P_{3D}^{i,k}\\1\end{array}\right]}
\\[1.0cm]
\frac{\mathbf{T}_{2D,2\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}^{k-1,k}\left[
\setlength\arraycolsep{1.2pt}
\begin{array}{c}P_{2D}^{i,k}\\1\end{array}\right]}
{\mathbf{T}_{2D,3\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}^{k-1,k}\left[
\setlength\arraycolsep{1.2pt}
\begin{array}{c}P_{2D}^{i,k}\\1\end{array}\right]}
& =
\frac{
{\mathbf{K}}_{2\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}\;
\mathbf{T}_{3D}^{k-1,k}\left[
\setlength\arraycolsep{1.2pt}
\begin{array}{c}P_{3D}^{i,k}\\1\end{array}\right]}
{\mathbf{T}_{3D,3\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}^{k-1,k}\left[
\setlength\arraycolsep{1.2pt}
\begin{array}{c}P_{3D}^{i,k}\\1\end{array}\right]}
\end{array}
\right.
\label{eq:Lien2D3DNonMatriciel1}
\end{align}
where the subscripts $1\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}$, $2\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}$ and $3\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}$
respectively index the first, second and third matrix rows and all
column indices. Denoting by $c^{k-1,k}_{1,i}$ and $c^{k-1,k}_{2,i}$
the right-hand sides in~\eqref{eq:Lien2D3DNonMatriciel1},
\eqref{eq:Lien2D3DNonMatriciel1} rewrites:
\begin{align}
\left\{\begin{array}{ccc}
\mathbf{T}_{2D,1\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}^{k-1,k}\left[\begin{array}{c}P_{2D}^{i,k}\\1\end{array}\right]
& = &
c^{k-1,k}_{1,i}\;\mathbf{T}_{2D,3\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}^{k-1,k}\left[\begin{array}{c}P_{2D}^{i,k}\\1\end{array}\right]
\\[20pt]
\mathbf{T}_{2D,2\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}^{k-1,k}\left[\begin{array}{c}P_{2D}^{i,k}\\1\end{array}\right]
& = &
c^{k-1,k}_{2,i}\;\mathbf{T}_{2D,3\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}^{k-1,k}\left[\begin{array}{c}P_{2D}^{i,k}\\1\end{array}\right]
\end{array} \right. \label{eq:Lien2D3DNonMatriciel2}
\end{align}
with
\begin{align*}
c^{k-1,k}_{1,i} & = \left (
\textbf{K}_{1\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}\;
\mathbf{T}_{3D}^{k-1,k}\left[
\setlength\arraycolsep{0.0pt}
\begin{array}{c}P_{3D}^{i,k}\\1\end{array}\right]
\right )/
\left (\mathbf{T}_{3D,3\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}^{k-1,k}\left[
\setlength\arraycolsep{0.0pt}
\begin{array}{c}P_{3D}^{i,k}\\1\end{array}\right]
\right ) \\
c^{k-1,k}_{2,i} & = \left (
\textbf{K}_{2\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}\;
\mathbf{T}_{3D}^{k-1,k}\left[
\setlength\arraycolsep{0.0pt}
\begin{array}{c}P_{3D}^{i,k}\\1\end{array}\right]\right )
/
\left (
{\mathbf{T}_{3D,3\raisebox{-0.8ex}{\scalebox{1.8}{$\cdot$}}}^{k-1,k}\left[
\setlength\arraycolsep{0.0pt}
\begin{array}{c}P_{3D}^{i,k}\\1\end{array}\right]}\right )
\end{align*}
Replacing the points $P_{3D}^{i,k}$ and $P_{2D}^{i,k}$ by their
coordinates, \eqref{eq:Lien2D3DNonMatriciel2} yields a linear system
of two equations involving the coefficients $a_{r,s}^{k-1,k}$
in~\eqref{eq:matriceHomography}:
\begin{equation}
\left\{
\setlength\arraycolsep{1pt}
\begin{array}{l}
x_{2D}^{i,k}a_{1,1}^{k-1,k}+ y_{2D}^{i,k}a_{1,2}^{k-1,k} +
a_{1,3}^{k-1,k}- \ldots \medskip
\\
\hspace*{.2cm}c^{k-1,k}_{1,i}x_{2D}^{i,k}a_{3,1}^{k-1,k} -
c^{k-1,k}_{1,i}y_{2D}^{i,k}a_{3,2}^{k-1,k} = c^{k-1,k}_{1,i} \bigskip\\
x_{2D}^{i,k}a_{2,1}^{k-1,k} +y_{2D}^{i,k}a_{2,2}^{k-1,k} + a_{2,3}^{k-1,k}- \ldots\medskip \\
\hspace*{.2cm}c^{k-1,k}_{2,i}x_{2D}^{i,k}a_{3,1}^{k-1,k}-c^{k-1,k}_{2,i}y_{2D}^{i,k}a_{3,2}^{k-1,k}
=c^{k-1,k}_{2,i}
\end{array}
\right.
\label{eq:Lien2D3DNonMatriciel3}
\end{equation}
Both equations are valid for all laser points $P^{i,k}_{3D}$ of
viewpoint $k$. This yields a linear system of $2M$ equations with
eight unknown parameters $a_{r,s}^{k-1,k}$ related to transformation
$\mathbf{T}_{2D}^{k-1,k}$: see~\eqref{eq:Lien2D3DMatricielFinal}.
For our prototype, $M=8$, so \eqref{eq:Lien2D3DMatricielFinal} is an
over-determined system of 16 equations. Notice that when $M \geq 4$,
the homography $\mathbf{T}_{2D}^{k-1,k}$ induced by
$\mathbf{T}_{3D}^{k-1,k}$ is unique since there are $2M\geq 8$
equations in~\eqref{eq:Lien2D3DMatricielFinal}.
\setcounter{equation}{13}
\begin{algorithm*}
\caption{Data registration of viewpoints $k$ and $k-1$.}
\label{alg:transformationDetermination}
\KwIn{\begin{itemize}{}
\item Source image $I_k$ and target image $I_{k-1}$.
\item $M$ laser points of viewpoint $k$ given in 3D and 2D:
$\{P_{3D}^{i,k}$, $P_{2D}^{i,k}\}|_{i\in \{1,\ldots,M\}}$.
\item Initial simplex given in~\eqref{eq:initialSimplex}, and
related similarity values $\varepsilon(V_e)$, $e \in \{1,\ldots,7\}$.
\end{itemize}}
\KwOut{Geometrical transformation linking viewpoints $k$ and $k-1$:
\begin{itemize}{}
\item homography matrix $\mathbf{T}_{2D}^{k-1,k}$.
\item translation vector $D^{k-1,k}$ and angles
$\theta_1^{k-1,k}$, $\theta_2^{k-1,k}$ and $\theta_3^{k-1,k}$
induced by the estimated $3\times 4$ rigid transformation
matrix $\mathbf{T}_{3D}^{k-1,k}$.
\end{itemize}}
\Repeat{the variation of the simplex volume leads to a mean displacement
value of $I_{k}$ on $I_{k-1} < 0.1$ pixel}
{
%
\underline{\em Step 1}: Replace the vertex of smallest
$\varepsilon(V_e)$ value by a new vertex according to the
symmetry rules of the simplex algorithm.
%
Then, update the entries of $\mathbf{T}_{3D}^{k-1,k}$
(see~\eqref{eq:matriceTrigideb}) from the knowledge of
$\theta_1^{k-1,k}$, $\theta_2^{k-1,k}$, $\theta_3^{k-1,k}$,
and of the 3D translation vector $D^{k-1,k}$ for the new
vertex\;
\underline{\em Step 2}: Given~\eqref{eq:Lien2D3DMatricielFinal}
and the $M$ points $(P_{3D}^{i,k}, P_{2D}^{i,k})$, compute the 8
parameters of the homography $\mathbf{T}_{2D}^{k-1,k}$ induced
by the rigid transformation $\mathbf{T}_{3D}^{k-1,k}$ determined
in {\em Step 1}\;
\underline{\em Step 3}: Superimpose $I_k$ on $I_{k-1}$
using~\eqref{eq:PointsHomologues} and the homography
$\mathbf{T}_{2D}^{k-1,k}$ obtained in {\em step 2}\;
\underline{\em Step 4}: For the new vertex, assess the
registration quality $\varepsilon(V_e)$
using~\eqref{eq:valVertex} and the images superimposed in {\em step 3}\;
}
\vspace*{.6mm}
%
Store the six parameters of the vertex having the largest
$\varepsilon(V_e)$ value in $D^{k-1,k}$, $\theta_1^{k-1,k}$,
$\theta_2^{k-1,k}$, and $\theta_3^{k-1,k}$\;
%
\vspace*{0mm}
%
Fill matrix $\mathbf{T}_{3D}^{k-1,k}$ with the rigid
transformation parameters computed from $D^{k-1,k}$,
$\theta_1^{k-1,k}$, $\theta_2^{k-1,k}$, and $\theta_3^{k-1,k}$:
see~\eqref{eq:matriceTrigideb};
\vspace*{0mm}
%
Fill matrix $\mathbf{T}_{2D}^{k-1,k}$ with the homography
parameters corresponding to $\mathbf{T}_{3D}^{k-1,k}$.
\end{algorithm*}
\subsubsection{Similarity measure between
$I_{k-1}$ and $\mathbf{T}_{2D}^{k-1,k}(I_{k})$ \label{sec:similarityMeasure}}
The homography $\mathbf{T}_{2D}^{k-1,k}$ determined with
\eqref{eq:Lien2D3DMatricielFinal} is now used to superimpose image
$I_k$ on $I_{k-1}$. The hue being constant in bladder images
\cite{Miranda2008}, the pixel values of $I_k$ and $I_{k-1}$ are
converted into grey-levels for the registration step. As shown
in~\cite{Hernandez:2010}, homologous features cannot be systematically
extracted from the images due to blur, weak textures or lacking
textures in some image parts. Therefore, iconic data are directly used
to measure the similarity of the overlapping parts of $I_k$ and
$I_{k-1}$. The mutual information, denoted by ${\cal
MI}(I_{k-1},\mathbf{T}^{k-1,k}_{2D}(I_{k}))$, is a robust similarity
measure \cite{Pluim:2003} which was successfully used for 2D bladder
image mosaicing~\cite{Miranda2008}. This statistical measure is
notably robust since it is computed with the whole pixels of the
common regions of the superimposed images. The mutual information is
defined from the grey-level entropies ${\cal H}(I_{k-1})$ and ${\cal
H}(\mathbf{T}^{k-1,k}_{2D}(I_{k}))$ and the joint grey-level entropy
${\cal H}_J(I_{k-1},\mathbf{T}^{k-1,k}_{2D}(I_{k}))$ of source image
$I_{k}$ and target image $I_{k-1}$:
\begin{eqnarray}
{\cal MI}\left(I_{k-1},\mathbf{T}^{k-1,k}_{2D}(I_{k})\right) =
{\cal H}(I_{k-1}) + \nonumber \qquad \qquad \\
\qquad \qquad {\cal H}(\mathbf{T}^{k-1,k}_{2D}(I_{k}))
- {\cal H}_J(I_{k-1},\mathbf{T}^{k-1,k}_{2D}(I_{k}))
\label{eq:MutualInformation}
\end{eqnarray}
with
\begin{align*}
\left\{\begin{array}{l}
{\cal H}(I_{k-1}) = -{\displaystyle\sum_{z_{k-1}} p(z_{k-1})\log p(z_{k-1})}\\
{\cal H}(\mathbf{T}^{k-1,k}_{2D}(I_{k})) = -{\displaystyle\sum_{
z_k} p(z_k)\log p(z_k)} \\
{\cal H}_J(I_{k-1},\mathbf{T}^{k-1,k}_{2D}(I_{k})) = \\ \qquad \quad
-{\displaystyle\sum_{z_{k-1}}\sum_{z_k}
p_J(z_{k-1},z_k)\log p_J(z_{k-1},z_k)}.
\end{array} \right.
\end{align*}
The grey-levels $z_{k-1}$ and $z_k$ range in $[0, 255]$ because images
are single-byte encoded. $p(z_{k-1})$ and $p(z_k)$ are the probability
density functions of the overlapping parts of images $I_{k-1}$ and
$\mathbf{T}_{2D}^{k-1,k}(I_{k})$ whereas $p_J(z_{k-1},z_k)$ is a joint
probability density function. The mutual information is maximal when
both images are registered, \emph{i.e.}, when $I_{k-1}$ is
statistically correlated with $\mathbf{T}_{2D}^{k-1,k}(I_{k})$.
\subsubsection{Registration algorithm\label{sec:simplex}}
The mutual information is maximized with respect to the rotation
angles $\theta_1^{k-1,k}$, $\theta_2^{k-1,k}$, $\theta_3^{k-1,k}$ and
the translation parameter $D^{k-1,k}$ appearing
in~\eqref{eq:matriceTrigideb}. The optimization task is done using the
simplex algorithm \cite{Nelder:1965} since the cost
function~\eqref{eq:MutualInformation} is non-differentiable. Because
there are six unknowns, the simplex is composed of seven vertices
$V_e$ (with $e \in \{1,\ldots,7\}$) lying in $\mathbb{R}^6$. The
similarity measure
\begin{align}
\varepsilon(V_e) = {\cal MI}(I_{k-1},\mathbf{T}^{k-1,k}_{2D}(I_{k}))
\label{eq:valVertex}
\end{align}
is computed for each vertex $V_e$, where the $\mathbf{T}_{2D}^{k-1,k}$
matrix deduces from $V_e$ as detailed in
\S~\ref{sec:matrixrelationship}. At each iteration, the vertex of
smallest $\varepsilon(V_e)$ value is replaced by a new vertex whose
location in the parameter space is computed with the simplex symmetry
rules. The angles and translations associated to the new vertex are
used to determine the rigid transformation matrix, compute the related
homography matrix, and then measure the image registration quality
using~\eqref{eq:valVertex}. By repeating this process, the simplex
vertices gradually converge towards the maximizer of
\eqref{eq:MutualInformation}. The optimization principle is detailed
in Algorithm \ref{alg:transformationDetermination}. The initial
simplex size is experimentally set relative to the 25 images/s
acquisition rate and the low cystoscope speed, leading to
sub-millimetric translations and to rotations smaller than $1^{\circ}$
between two acquisitions. The universal initial simplex is built with
0.3 mm translations and $0.4^{\circ}$ rotation values:
\begin{align}
\begin{array}{lclccccr}
V_1 & = & [\,0 & 0 & 0 & 0 &0 &0\,] \\
V_2 & = & [\,0.3 & 0 & 0 & 0 &0 &0\,] \\
V_3 & = & [\,0 & 0.3 & 0 & 0 &0 &0\,] \\
V_4 & = & [\,0 & 0 & 0.3 & 0 &0 &0\,] \\
V_5 & = & [\,0 & 0 & 0 & 0.4 &0 &0\,] \\
V_6 & = & [\,0 & 0 & 0 & 0 &0.4 &0\,] \\
V_7 & = & [\,0 & 0 & 0 & 0 &0 &0.4\,]
\end{array}
\label{eq:initialSimplex}
\end{align}
Each vertex gathers the three translation parameters (first three
entries in $D$), and then $\theta_1$, $\theta_2$ and $\theta_3$. The
iterative process of Algorithm~\ref{alg:transformationDetermination}
is stopped when the mean displacement of the corners of image $I_k$
(displaced in the $I_{k-1}$ domain) becomes smaller than 0.1 pixel.
When the optimization task is completed, the geometrical links between
consecutive coordinate systems $\{c_{k}$\} and $\{c_{k-1}$\} and
consecutive images ($I_k$, $I_{k-1}$) are fully defined by
$\mathbf{T}^{k-1,k}_{3D}$ and $\mathbf{T}^{k-1,k}_{2D}$, respectively.
\begin{figure*}[!t]
\centering
\begin{tabular}{c}
\includegraphics[width=0.8\textwidth]{fig3a-eps-converted-to.pdf}\\
(a)\\
\includegraphics[width=0.8\textwidth]{fig3b-eps-converted-to.pdf}\\
(b)
\end{tabular}
\caption{Active vision prototype. (a) Principle of the
structured-light prototype. The trajectory of the beam emitted by
the laser source is precisely controlled by two mirrors. The
adjustable mirror orientations ensure that the laser beam falls on
the center of the diffractive lens. This lens projects eight laser
rays onto the scene. The laser wavelength and power were set to 532
nm and 10 mW. The green color allows us to have a good contrast
between the image textures and the green laser points. The 10 mW
power was actually much too strong, therefore optical attenuators
were used. The triangulation principle is sketched in
Fig.~\ref{fig:2} and described in
Section~\ref{sec:AcquisitionData}. The prototype (camera and
projector) was calibrated using the generic method described in
\cite{Ben-Hamadou:2013}.
(b) Prototype corresponding to the sketch in (a). Mirrors 1 and 2
are fixed onto planar metal supports. The orientation of the planar
supports is precisely set using three screws. The planar support
and two screws are not visible for mirror 2.}
\label{fig:3}
\end{figure*}
\subsection{Data mosaicing\label{sec:mosaicing}}
The 3D laser points from all viewpoints $k$ have to be expressed in a
common coordinate system. Choosing the camera coordinate system of the
first acquisition, the points of viewpoint $k$ are displaced from
$\{c_{k}$\} to $\{c_{1}$\} using the ``global'' (g) transformation
matrix:
\begin{equation}
\mathbf{T}_{3D}^{k,g} = \prod_{j=2}^{k} \mathbf{T}_{3D}^{j-1,j} = \mathbf{T}_{3D}^{1,2}
\times \mathbf{T}_{3D}^{2,3}\times \hdots \times \mathbf{T}_{3D}^{k-1,k}.
\label{eq:discreteSurface}
\end{equation}
The whole set of points in $\{c_{1}$\} is then exploited to compute a
3D mesh with triangular faces representing the bladder
surface~\cite{Kazhdan:2006}. Finally, the textures of images $I_k$ can
be projected onto the 3D mesh.
Specifically, the color of each triangular face in the 3D mesh is
obtained by copying the color of the pixel in the first image of the
video in which the triangular face can be seen. This simple choice
could be improved using more involved strategies in order to reduce
possible texture discontinuities in the 3D mosaic or to smooth the 3D
surface by thin-plate interpolation. See \emph{e.g.},
\cite{Weibel2012b} in the 2D case.
Similar to the 3D case, a 2D mosaic is built by taking the coordinate
system of the first image as reference and using the global 2D
matrices:
\begin{equation}
\mathbf{T}_{2D}^{k,g} = \prod_{j=2}^{k} \mathbf{T}_{2D}^{j-1,j} = \mathbf{T}_{2D}^{1,2}
\times \mathbf{T}_{2D}^{2,3}\times \hdots \times \mathbf{T}_{2D}^{k-1,k}.
\label{eq:2DMosaic}
\end{equation}
It has to be noticed that the green laser points cover a small surface
in the images and are easy to segment since a simple thresholding of
the H (hue) component in the HSV color space robustly separates the
green laser points from the reddish or orange bladder epithelium
pixels (see \cite{daul:2000} for a detailed description of the HSV
color space family). Since images $I_k$ and $I_{k-1}$ include large
common scene parts, the missing bladder textures in $I_{k-1}$ can be
replaced in the mosaic by the homologous texture pixels of $I_k$. The
correspondence between pixel textures hidden by laser dots in image
$I_{k-1}$ and the corresponding visible texture pixels in image $I_k$
is given by homography $T_{2D}^{k-1,k}$. This correspondence is used
to build 2D and 3D mosaics without laser dot pixels.
\begin{figure*}[!t]
\centering
\begin{tabular}{ccc}
\includegraphics[height=4.3cm]{fig4a-eps-converted-to.pdf}&
\includegraphics[height=4.3cm]{fig4b-eps-converted-to.pdf}&
\includegraphics[height=4.3cm]{fig4c-eps-converted-to.pdf}\\
(a)&(b)&(c)
\end{tabular}
\caption{Three real phantoms. The planar phantom (a) is used to assess
the accuracy of the 3D registration algorithm. The half cylinder~(b)
and wave~(c) phantoms allow us to evaluate the ability of the
registration method to construct convex and nonconvex surface parts,
respectively.}
\label{fig:4}
\end{figure*}
\section{Quantitative results for prototype data \label{sec:ProtoResultsDiscussion}}
The goal of this section is to validate on real data that the
mosaicing approach of Section~\ref{sec:3Dmosaicing} indeed leads to
coherent surface construction results. To do so, an active vision
prototype is first proposed. Then, 3D mosaicing results are presented
for data acquired with the prototype using three real phantoms, on
which pig bladder texture images were stuck. This allows for
a first quantitative assessment of the mosaicing accuracy based on ground
truths. As mentioned before, our prototype cannot be used in clinical
situations yet. It was rather designed for validation of the
structured-light approach in combination with a cystoscope and to show
the feasibility of 3D cartography. The applicability to clinical
situations will be further discussed in
Sections~\ref{sec:SimulatedDataResultsDiscussion} and
\ref{sec:Conclusion} together with complementary tests on realistic
simulated data.
\subsection{Prototype and test phantom description \label{sec:ProtoPhantomDescription}}
Our active vision prototype is illustrated in Fig.~\ref{fig:3}. The
acquisition channel consists of a Karl Storz cystoscope, and the image
sequences are acquired with a Basler scout model camera. The image
resolution is 768 $\times$ 576 pixels. The diffractive lens is fixed
nearby the wide FOV lens of the cystoscope distal tip. The principle
of the laser ray projection is sketched in Fig.~\ref{fig:3}(a). This
prototype uses the triangulation principle described in
Section~\ref{sec:AcquisitionData} and has a geometry equivalent to
that of Fig.~\ref{fig:2}(a). The baseline (distance between the camera
and projector optical centers, see Fig.~\ref{fig:3}(a)) available for
triangulation is about 3 mm.
The laser projector wavelength is equal to 520 nm, and the 10 mW power
should be adapted for further clinical deployment.
High resolution images of an incised and flattened pig bladder were
acquired to simulate human bladder textures (see
Fig.~\ref{fig:4}(b)). The images were printed on paper sheets and
stuck on surfaces with various shapes. First, the planar surface of
Fig.~\ref{fig:4}(a) is used to test the 3D mosaicing algorithm. Here,
the underlying assumption in the proposed algorithm is perfectly
fulfilled: the surfaces viewed in consecutive images are planar. Then,
the half cylinder of Fig.~\ref{fig:4}(b) is used to simulate a convex
surface. The half cylinder has a radius of 35 mm and allows for
simulating a plausible bladder curvature in the direction orthogonal
to the main cylinder axis (a non-warped bladder can be approximately
thought of as an ellipsoid with a major axis from 80 to 100 mm and
minor axes from 30 to 40 mm). The bladder is in contact with other
organs is often warped. In this case, the surface includes concave
parts due to troughs in the organ wall. As shown in
Fig.~\ref{fig:4}(c), a wave phantom was built to simulate warped
bladder parts with troughs between waves. The length of the wave (a
period of the phantom) is 40 mm and the wave depth (from the trough to
the crest of the wave) is 20 mm. This phantom simulates extreme
situations since consecutive troughs of large magnitude never arise in
practice.
All phantoms are positioned and displaced relative to the endoscope
prototype using a micrometric positioning table to generate ground
truth 3D transformations.
\begin{figure*}[!t]
\centering
\includegraphics[width=0.7\linewidth]{fig5-eps-converted-to.pdf}
\caption{Plane construction results. Bottom: 2D mosaic. The green
arrows refer to the 8 laser point displacements (constant
translation, no rotation) throughout image acquisition. Center: 3D
points placed in the common coordinate system $(O_{c_1},
\protect\overrightarrow{x_{c_1}}, \protect\overrightarrow{y_{c_1}},
\protect\overrightarrow{z_{c_1}})$. Top: related mesh with
triangular faces on which the 2D mosaic image can be projected.\label{fig:5}}
\end{figure*}
\subsection{Point reconstruction accuracy and surface construction results
\label{sec:ResultsRegisMosaics}}
Once the prototype has been calibrated with the method described
in~\cite{Ben-Hamadou:2013}, 3D point reconstruction is performed in
the local camera coordinate system $\{c_k\}$, and $M=8$ $P_{3D}^{i,k}$
points are then available per viewpoint $k$.
In order to evaluate the laser point reconstruction accuracy, a first
data acquisition was carried out using the wave phantom. The
cystoscope axis was held orthogonal to the translation plane and
$K=200$ images were acquired with 1 mm translation steps. The range of
$z_{c_k}$-values that were found (in between 20 and 40 mm) for the
$P_{3D}^{i,k}$ points matches the wave height, equal to 20 mm. It was
further verified that for prototype distal tip-to-surface distances
ranging from some millimeters to 5 centimeters, the average
reconstruction error on the $z_{c_k}$ coordinate (depth) is close to
0.5 mm and remains submillimetric for all points. When normalizing
this 3D reconstruction error by the tip-to-surface distance, the
errors are always lower than $3~\%$. This accuracy is by far
sufficient for bladder surface mosaicing since the constructed surface
shape must only be representative of the organ wall (a precise
dimensional measure does not make sense in urology).
In the following surface construction tests and without loss of
generality, the phantoms were displaced in front of the fixed
prototype. This is in contrast with the setup of
Section~\ref{sec:3Dmosaicing}, where the instrument is moving inside
the organ. However, it is easy to see that the proposed mosaicing
algorithm remains valid and unchanged in both settings whatever the
moving object (surface or instrument). The simultaneous 2D and 3D data
mosaicing results are illustrated in Fig.~\ref{fig:5}
for the planar phantom. Here, 100 images were acquired for 99 constant
translations of 0.3 mm (with the three components of $D^{k-1,k}$
different from 0) and without 3D rotations. Thus, the magnitude of 3D
displacements is equal to:
\begin{align}
\|D^{k-1,k}\|=0.3 \label{eq:criterionTranslations}
\end{align}
where $\|\,.\,\|$ stands for the Euclidean norm.
\begin{figure*}
\centering
\psfig{figure=fig6-eps-converted-to.pdf, width=0.8\linewidth}
%
\caption{Wave construction results. Center: 3D points placed in the
common camera coordinate system $(O_{c_1},
\protect\overrightarrow{x_{c_1}},
\protect\overrightarrow{y_{c_1}},
\protect\overrightarrow{z_{c_1}})$ and representing 2.5 wave
periods in the phantom of Fig.~\ref{fig:4}(c). Top: mesh surface
reconstructed from the first 100 acquisitions (zoom in
rectangle). Bottom: textured surface corresponding to the first
100 acquisitions. The cystoscope prototype is roughly oriented
along the $\protect\overrightarrow{z_{c_1}}$ axis. The local
phantom surface corresponding to the left edge of the zoom in
rectangle is the most distant from the camera in contrast with the
local surface related to the right edge.\label{fig:6}}
\end{figure*}
The average translation magnitude and standard deviation computed with
the registration algorithm for the 99 consecutive image pairs are 0.27
$\pm$ 0.015 mm, whereas the angles $\theta_1^{k-1,k}$,
$\theta_2^{k-1,k}$ and $\theta_3^{k-1,k}$ indeed tend towards
$0^{\circ}$. So, the average magnitude of displacement is slightly
underestimated with respect to the ground truth (0.3 mm), whereas the
$0.015$ mm standard deviation value is very low. These results are a
first indication that the nature (translation) of the displacement can
be successfully determined using the registration algorithm.
Fig.~\ref{fig:5} confirms that accurate results are obtained for the
planar phantom. The 2D mosaic built from the estimated
$\mathbf{T}_{2D}^{k-1,k}$ homographies is visually coherent,
\emph{i.e.}, without texture discontinuities between images. In the
textured image of Fig.~\ref{fig:5}, the green arrows represent the
trajectory of the eight laser points throughout the video-image
acquisition. As could be expected, these trajectories are affine
because the camera movement is a pure translation. The 3D points,
placed in the common coordinate system $\{c_{1}$\}, all lie on a
plane. The constructed mesh with triangular faces is displayed in the
top image of Fig.~\ref{fig:5}.
\begin{figure*}
\begin{tabular}{cc}
\setlength{\tabcolsep}{0.05cm}
\begin{tabular}{c}\includegraphics[width=10cm]{fig7a-eps-converted-to.pdf}\end{tabular}&
\begin{tabular}{c}\includegraphics[width=6cm]{fig7b-eps-converted-to.pdf}\end{tabular}\\
(a)&(b)
\end{tabular}
\caption{Half cylinder reconstruction results. (a)~3D points placed in
the global coordinate system $\{c_{1}$\}. Their locations are
consistent with the trajectories sketched in green on the half
cylinder phantom~(b). Each green curve represents the trajectory of
two very close laser points.}
\label{fig:7}
\end{figure*}
In the following surface construction tests, we aim to address a case where the
geometrical link between the surfaces in the FOV of two images is not
exactly an homography (working assumption of the 3D registration
algorithm) since the local surfaces are not planar anymore. The data
of Fig.~\ref{fig:6} were acquired for the wave phantom with the same
displacements as before: acquisition of $K=360$ images for 359
constant translations of 0.3 mm and without any rotation. For this
phantom, the local surface orientation with respect to the cystoscope
optical axis strongly varies throughout the sequence. To quantify the
algorithm accuracy, we used again
criterion~\eqref{eq:criterionTranslations}. The average magnitude of
the 3D translations computed with the registration algorithm for
consecutive image pairs is $0.29 \pm 0.03$ mm, and
$\theta_1^{k-1,k}=\theta_2^{k-1,k}=\theta_3^{k-1,k} \approx
0^{\circ}$. The magnitude of translations (0.29 mm) is very close to
the true average displacement (0.3 mm) whereas the related standard
deviation (0.03 mm) remains very close to the ideal zero value. These
results show that the translation of the cystoscope prototype can be
successfully determined for non-planar surfaces and for varying angles
between the local surface and the cystoscope optical axis. The
variation of this angle combined with changes of phantom-to-camera
distances (see Fig.~\ref{fig:6}) impact the perspective
($a_{3,1}^{k-1,k}$ and $a_{3,2}^{k-1,k}$), shearing ($s_x^{k-1,k}$ and
$s_y^{k-1,k}$), and scale ($\alpha^{k-1,k}$) parameters,
see~\eqref{eq:matriceHomography}.
The blue points in Fig.~\ref{fig:6} represent again the 3D laser
points placed in the common coordinate system $(O_{c_1},
\overrightarrow{x_{c_1}}, \overrightarrow{y_{c_1}},
\overrightarrow{z_{c_1}})$. The 360 viewpoints correspond to 2.5 wave
periods, \emph{i.e.}, a length of 100 mm. Qualitatively, the global
surface indeed corresponds to waves although after 1.5 wave period,
the wave surface bends itself instead of remaining straight. However,
the results are satisfactory since a wave period represents an organ
pushing against the bladder (\emph{e.g.}, a trough), and two or three
consecutive troughs are an extreme situation never arising for
cystoscopic data.
In Fig.~\ref{fig:7}, the cystoscope prototype was approximately held
at a 30 mm distance from the half cylinder surface (of radius 190 mm)
during the acquisition of 130 images. The surface covered by the
cystoscope corresponds to a circular arc of aperture $130^\circ$. The
displacements between consecutive acquisitions $k-1$ and $k$ mainly
consist of $1^\circ$ rotations around $\overrightarrow{x_{c_1}}$
combined with small translations and rotations around
$\overrightarrow{y_{c_1}}$ and $\overrightarrow{z_{c_1}}$.
Fig.~\ref{fig:7} shows that the reconstructed points indeed lay on a
cylindrical surface. The depth of the cylinder part was used to
quantify the surface construction accuracy. Ideally, the dashed line in
Fig.~\ref{fig:7}(a) should circumscribe an arc of aperture
$130^\circ$. The real depth of the cylinder part corresponding to the
$130^\circ$ phantom aperture is 25 mm, whereas the depth computed from
the reconstructed data is 21 mm. This error does not alter the visual
quality of the surface representation: the phantom shape is indeed
cylindrical and no dimensional analysis is required.
\subsection{Overview of the results related to prototype
data \label{sec:DiscussionProtoDataResults}}
The results obtained so far are a first validation that the prototype
displacement (\emph{i.e.}, the displacement of the camera coordinate
system in between two acquisitions) can be recovered using only the
video-image and 8 laser points per acquisition. Indeed, for both
phantom types, the deviation of the estimated 3D displacement from the
ground truth is small.
{\em - Local surface planarity.} Planarity is one of the main
differences between the plane and wave phantoms. Indeed, the planar
assumption (made for registering two images using an homography) is
not fulfilled anymore for the surface parts of the wave. We found that
the camera displacements were computed with similar accuracy for both
planar and non-planar surfaces. It was already proved in the frame of
2D bladder mosaicing that image registration with homographies is
accurate even for regions where the bladder is
non-planar~\cite{Weibel:2012a} or contains lesions such as
polyps~\cite{Hernandez:2010}. The results of
\S~\ref{sec:ResultsRegisMosaics} indicate that homographies linking 2D
images are appropriate as well for 3D mosaicing.
{\em - 3D point reconstruction accuracy.} Inaccurate 3D laser
point reconstruction impacts the precision of 3D camera displacement
estimation. The results obtained with real phantoms show that 3D
point reconstruction is accurate enough to enable camera motion
assessment and 3D mosaicing, even for active vision prototypes
having small baselines (3 mm here).
{\em - Cystoscope axis orientation.} For tests with wave phantoms,
the angle between the cystoscope optical axis and the normal to the
local surface is varying: at extreme points (points of minimum or
maximum depth) of the wave, the angles are small ($\approx
0^\circ$), whereas for medium points, their value is maximum
($45^{\circ}$). The accuracy of the reconstructed camera
displacement is not strongly impacted by the variation of angles.
The tests with the plane and wave phantoms involved 3D translations,
whereas the viewpoint changes arising in the half-cylinder
experiment combined both rotations and translations. In the latter
case, no ground truth displacements were available. However, the
cylinder depth that was estimated for an $130^\circ$ aperture shows
that the surface shape can be accurately recovered.
In all tests, images were acquired for a distal tip-to-surface
distance ranging from 10 to 40 mm and for surface areas of one to
several square centimetres. These dimensions are consistent with
real bladder image acquisitions in terms of acquisition distance
variation and size of the area viewed in images. For this distance
interval, the laser points were systematically visible in the
images, and their position could then be reconstructed in the camera
coordinate system.
\begin{figure}[!t]
\centerline{\includegraphics[width=0.8\linewidth]{fig8-eps-converted-to.pdf}}
\caption{Simulated bladder volume and simulation of video sequence
acquisition with known 3D endoscope displacements. The cone
corresponds to the field of view of the camera for one viewpoint.
The green dashed curve refers to the trajectory of the virtual
cystoscope. The simulated video sequence then contains images
related to both convex and nonconvex parts of the virtual bladder. }
\label{fig:8}
\end{figure}
\begin{figure*}[!t]
\centering
\begin{tabular}{ccc}
\includegraphics[width=0.3\linewidth]{fig9a-eps-converted-to.pdf} &
\includegraphics[width=0.3\linewidth]{fig9b-eps-converted-to.pdf} &
\includegraphics[width=0.3\linewidth]{fig9c-eps-converted-to.pdf} \\[.2cm]
(a)&(b)&(c)
\end{tabular}
\caption{Extraction of an image from each of the three simulated
video-sequences. (a) Image with weakly contrasted texture. (b)
Image with highly contrasted vessels. (c) Image with blurry
texture.}
\label{fig:9}
\end{figure*}
\section{Results for realistic simulated data \label{sec:SimulatedDataResultsDiscussion}}
The results presented so far are a first validation of the feasibility
of 3D bladder mosaicing. However, similar to the contributions of
Soper \emph{et al}~\cite{Soper2012} and Shevchenko \emph{et
al}~\cite{Shevchenko:2012}, the method was not tested on patient
data (our active vision prototype was not built for patient data
acquisition at this stage). In this section, realistic bladder data
are simulated to perform complementary mosaicing tests. Not only the
bladder volume and texture are closer to human data, but also the
simulated endoscope trajectory can be controlled to reproduce all
types of cystoscope displacements.
\subsection{Simulation of textured bladder data}
To simulate a more realistic bladder shape, we considered a (convex)
oval volume whose width, height and depth are equal to 110, 100, and
70 mm, respectively (the standard capacity/volume of the bladder
ranges in between 400 and 600 ml). The ovoid was then locally deformed
to create a nonconvex surface part. This surface, illustrated in
Fig.~\ref{fig:8}, was defined in agreement with the Institut de
Canc\'erologie de Lorraine. Human bladder textures were extracted from
large FOV mosaics built using 2D mosaicing algorithms
\cite{Miranda2008,Weibel:2012a}. Three such mosaics were considered
(see Fig.~\ref{fig:1}) and mapped on the inner wall of the simulated
3D bladder surface.
Blender (a 3D graphics modeling and rendering
software~\cite{blender:2014}) was used to simulate the camera and the
laser projector while reproducing the geometrical configuration of our
prototype. Such 3D representation can be found in Fig.~\ref{fig:8}.
Both camera and laser projector parameters (calibrated
using~\cite{Ben-Hamadou:2013}) are handled. Moreover, Blender allows
us to simulate complex cystoscope trajectories combining rotations and
translations. Similar to clinical examinations, the distance between
the cystoscope distal tip and the surface is changing. The resulting
area seen in images varies from 1 to a few square
centimeters. Fig.~\ref{fig:9} displays a sample of the simulated
video-sequences obtained for each of the three numerical phantoms
using the 3D surface of Fig.~\ref{fig:8}. Each image sequence includes
120 images and covers $24$ cm$^2$ of the phantom surface. Importantly,
the texture variability is large from one video-sequence to another.
\begin{figure*}
\centering
{
\setlength{\tabcolsep}{-0.2cm}
\begin{tabular}{ccc}
\includegraphics[width=0.36\textwidth]{fig10a-eps-converted-to.png} &
\includegraphics[width=0.36\textwidth]{fig10b-eps-converted-to.png} &
\includegraphics[width=0.36\textwidth]{fig10c-eps-converted-to.png} \\[-.4cm]
(a)&(b)&(c)\\[.4cm]
\includegraphics[width=0.36\textwidth]{fig10d-eps-converted-to.png} &
\includegraphics[width=0.36\textwidth]{fig10e-eps-converted-to.png} &
\includegraphics[width=0.36\textwidth]{fig10f-eps-converted-to.png} \\[-.4cm]
(d)&(e)&(f)
\end{tabular}
}
\caption{Results obtained for the three numerical phantoms related to
Fig.~\ref{fig:9}(a,b,c). Top: constructed 3D point cloud (in black)
and cystoscope trajectory (magenta curve). The point cloud is
displayed along with the ground truth surfaces (grey mesh) to
visually assess their overlapping quality. The green dashed curve
refers to the ground truth cystoscope trajectory. For readability
reasons, the meshed surface generated from the reconstructed 3D
points is not shown. Bottom: 3D reconstructed mosaics seen from
above (for improved readability). }
\label{fig:10}
\end{figure*}
\subsection{Surface construction results for the simulated data \label{sec:ResNumericalPhantoms}}
The proposed 3D mosaicing method was applied to the three simulated
data sequences. Fig.~\ref{fig:10}(a,b,c) displays the cloud of
reconstructed points $P_{3D}^{i,k}$ for each sequence in the common
coordinate system $\{c_1\}$ together with the trajectory of the
optical center of the cystoscope prototype (the computed trajectory
appears in magenta). The textured surfaces generated from the cloud of
3D points can be found in Fig.~\ref{fig:10}(d,e,f).
The surface construction accuracy is assessed by computing the average
Euclidean distance between the reconstructed 3D points in $\{c_1\}$
and their ground truth locations on the phantom surface. These
distances are ideally equal to 0. The numerical errors obtained for
the three phantoms are 0.79$\pm$0.34 mm, 0.78$\pm$0.33 mm, and
0.80$\pm$0.34 mm, respectively. On Fig. \ref{fig:10}, it is noticeable
that the shape of the reconstructed surface part (containing the black
points) is visually consistent with that of the true surface (grey
mesh). The mean error is small ($<1$ mm), confirming that the
reconstructed extended FOV surfaces remain accurate for both convex
and nonconvex surface parts.
\begin{table*}[t!]
\begin{center}
\begin{tabular}{lccccc}
\hline
{\bf Phantom} & Planar phantom & Wave phantom & First numerical & Second numerical & Third numerical \\ \vspace*{1mm}
{\bf type} & (Fig.~\ref{fig:5}) & (Fig.~\ref{fig:6}) & phantom (Fig.~\ref{fig:10}(a)) & phantom (Fig.~\ref{fig:10}(b)) & phantom (Fig.~\ref{fig:10}(c)) \\ \cline{2-6}
{\bf Mean} $\epsilon_{3D}^{k-1,k}$ & 0.016$\pm$0.002 & 0.05$\pm$0.01 & 0.021$\pm$ 0.0023 & 0.023$\pm$ 0.0025& 0.023$\pm$0.0027 \\ \cline{2-6}
{\bf Mean} $\epsilon_{2D}^{k-1,k}$ & $0.30\pm$0.29 & 0.85$\pm$0.8 & 0.40$\pm$0.39 & 0.42$\pm$0.37 & 0.43$\pm$0.39 \\ \hline
\end{tabular}
\caption{Registration errors for five phantoms. The mean (over $k$)
and standard deviation ($mean \pm std$) of criteria
$\epsilon_{3D}^{k-1,k}$ and $\epsilon_{2D}^{k-1,k}$ are in
millimetres and pixels, respectively. \label{Tab:RegistrationError}}
\end{center}
\end{table*}
\subsection{Discussion on the simulated data results}
Several conclusions can be drawn. First, the bladder surface of
Fig.~\ref{fig:8} was accurately constructed for video-sequences
containing various human textures. Second, the average 3D error
remains small (less than 1 mm) in all cases. The proposed algorithm
was also able to deal with the pig bladder textures (shown in
Section~\ref{sec:ProtoResultsDiscussion}), which is another indication
of its robustness against bladder texture variations. The use of
mutual information as a similarity measure is the main reason behind
this robustness.
For long video sequences, it may occur that the 3D points
$P_{3D}^{i,k}$ substantially diverge from the ground truth surface.
This is due to the accumulation of 3D registration errors over the
sequence. Error accumulation is a common issue when 2D or 3D
transformations between consecutive frames are sequentially
estimated. This problem was already encountered for 2D mosaicing
algorithms, and efficient solutions were proposed, based on a global
texture discontinuity correction over the processed image
sequence~\cite{Weibel:2012a}.
\section{Assessment of registration accuracy}
\label{sec:3DRegistrationError}
The quantitative results of Sections \ref{sec:ResultsRegisMosaics} and
\ref{sec:ResNumericalPhantoms} are respectively related to the
accuracy of the endoscope displacement and surface estimation. This
section focusses on the registration accuracy between viewpoints.
\subsection{Registration accuracy criteria}
In~\eqref{eq:3DRegistrationCriterion}, $\epsilon_{3D}^{k-1,k}$ defines
the 3D registration error compu\-ted with the known ground truth
transformation $\mathbf{T}_{3D,gt}^{k-1,k}$ and the transformation
$\mathbf{T}_{3D}^{k-1,k}$ estimated with the proposed registration
algorithm. Criterion $\epsilon_{3D}^{k-1,k}$ computes the mean
Euclidean distance in millimetres between the exact and estimated
locations of points $P_{3D}^{k,i}$, displaced in the coordinate system
$\{c_{k-1}\}$ of viewpoint $k-1$:
\begin{align}
\epsilon_{3D}^{k-1,k} = \frac{1}{N}\sum_{i=1}^{N}\left \|
\hat{P}_{3D,gt}^{i,k-1}-\hat{P}_{3D}^{i,k-1}
\right
\| \label{eq:3DRegistrationCriterion}
\end{align}
where
the displaced locations are respectively denoted by
$\hat{P}_{3D,gt}^{i,k-1}$ and $\hat{P}_{3D}^{i,k-1}$. They are
computed from $\hat{P}_{3D}^{i,k}$ using the ground truth and
estimated transformations $\mathbf{T}_{3D,gt}^{k-1,k}$ and
$\mathbf{T}_{3D}^{k-1,k}$; see~\eqref{eq:matriceTrigidea}. Ideally,
$\epsilon_{3D}^{k-1,k}$ equals 0.
Similarly, in~\eqref{eq:2DRegistrationCriterion}, criterion
$\epsilon_{2D}^{k-1,k}$ computes the mean Euclidean distance in pixels
between the projections in image $I_{k-1}$ of the displaced 3D points
$\hat{P}_{3D,gt}^{i,k-1}$ and $\hat{P}_{3D}^{i,k-1}$:
\begin{align}
\epsilon_{2D}^{k-1,k}
\hspace*{-.11cm}=\hspace*{-.11cm}\frac{1}{N}\sum_{i=1}^{N}\left \|\frac{1}{\hat{z}_{3D,gt}^{i,k-1}}
\mathbf{K}\hat{P}_{3D,gt}^{i,k-1}-\frac{1}{\hat{z}_{3D}^{i,k-1}}
\mathbf{K}\hat{P}_{3D}^{i,k-1}\right \|
\label{eq:2DRegistrationCriterion}
\end{align}
$\mathbf{K}\hat{P}_{3D,gt}^{i,k-1}/\hat{z}_{3D,gt}^{i,k-1}$
and $\mathbf{K}\hat{P}_{3D}^{i,k-1}/\hat{z}_{3D}^{i,k-1}$ gather the
coordinates in pixel unit of the ground truth and estimated
point locations, where $\mathbf{K}$ is the projective matrix coding for
the intrinsic camera parameters;
see~\eqref{eq:matricePerspective}. Ideally, $\epsilon_{2D}^{k-1,k}$
also equals 0.
\subsection{Registration results and accuracy discussion}
Table \ref{Tab:RegistrationError} gives the registration errors, \emph{i.e.}, the mean and standard deviation of
the $\epsilon_{3D}^{k-1,k}$ and $\epsilon_{2D}^{k-1,k}$ values computed with the data of viewpoint pairs ($k-1$, $k$) for five phantoms.
The results obtained for the planar phantom correspond to the
inherent accuracy of the algorithm (best possible registration results
obtained with the planarity assumption perfectly fulfilled). The mean
(0.016 mm) and standard deviation (0.003 mm) of the
$\epsilon_{3D}^{k-1,k}$ values show that in such ideal situation, the
3D registration errors between viewpoints are indeed very small.
The 2D sub-pixel registration error is $\epsilon_{2D}^{k-1,k}=0.30\pm0.29$ pixels for the planar phantom.
The planarity assumption is not fulfilled for the wave phantom: the
field of view of the images is wide and visualizes large wave parts
whose surfaces strongly deviate from a planar shape. In such an
extreme situation, the mean registration error remains small (0.05
mm). Due to these small 3D registration errors, the surface in
Fig.~\ref{fig:6} has a correct global wave shape. Moreover, when
scanning the wave surface, the angle between the 3D cystoscope
prototype and the normal to the local surface takes values ranging in
the whole interval [0-45] degrees. The small standard deviation of
criterion $\epsilon_{3D}^{k-1,k}$ ($0.01$ mm) indicates that the
registration algorithm is robust towards changes
of viewing angle. The mean of the 2D registration error (0.85 pixels)
remains also small.
The 2D and 3D registration errors are quite close for all three
numerical phantoms. The unique difference in the three numerical
phantoms was the human bladder texture used for the data acquisition
simulations (the surface shape and cystoscope trajectory was the same
for all simulations). These results show that the registration
algorithm is robust towards human bladder texture variability (the
mutual information based similarity measure systematically led to
comparable registration accuracy).
The $\epsilon_{3D}^{k-1,k}$ errors measured for the simulated phantoms
are always larger than the $\epsilon_{3D}^{k-1,k}$ score obtained for
the planar phantom (most favorable situation) and smaller than the
$\epsilon_{3D}^{k-1,k}$ score for the wave phantom (most difficult
situation). This is consistent with the fact that the phantom of
Fig.~\ref{fig:5} is exactly planar, the wave phantom is highly
non-planar, whereas the degree of planarity of the numerical phantoms
is intermediate. We would like to stress that the
$\epsilon_{3D}^{k-1,k}$ and $\epsilon_{2D}^{k-1,k}$ scores obtained
for the simulated phantoms are significantly closer to the error
obtained for the plane rather than those of the wave phantom. This is
an indication of the ability of the algorithm to reconstruct bladder
shapes.
\section{Conclusion\label{sec:Conclusion}}
The proposed 3D bladder mosaicing algorithm is guided by 2D image
registration. One advantage of the proposed approach is that no strong
assumption is made on surface shapes. On the one hand, the choice of
some initial surface shape is not required. On the other hand, smooth
surfaces with both convex and nonconvex parts can be accurately
constructed, the latter corresponding to warped bladder
parts. Furthermore, the results obtained for pig bladder data and
human bladder images highlight the robustness of the approach, since
the mutual information based registration can cope with high texture
variability.
An important feature of our endoscope prototype is that the
triangulation geometry for 3D data acquisition and the baseline length
of 3 mm are very close to those of the 3D cystoscopes which might be
built for clinical use. A current trend in endoscopy is the ``chip on
the tip'' technology (\emph{i.e.}, sensors fixed on the distal tip of
endoscopes). In this technology, the laser based solution can be used
for both rigid and flexible cystoscopes (both the CCD matrix and the
diffractive optics can be fixed on the distal tip): the calibration
procedure described in \cite{Ben-Hamadou:2013} and the mosaicing
algorithm proposed here remain unchanged. For rigid cystoscopes, the
calibration and 3D point reconstruction methods are the same for
different lens orientations ($0^\circ$ straight looking lens,
$30^\circ$ forward-oblique looking lens or $70^\circ$ lateral looking
lens).
Except for the optics to be implemented on the cystoscope for 3D point
reconstruction, the modifications to be done for clinical translation
are rather light. This is a key advantage over alternative solutions
which rely on external
devices~\cite{Shevchenko:2012,Agenant:2013}. This should lead to a
drastic reduction of the overall cost of cystoscopy equipment together
with a simplification of the technical installations in the
examination rooms. Actually, the ``chip on the tip'' technology should
allow us to simplify the prototype in future years: the cylindrical
tube of Fig.~\ref{fig:3} cannot be used in clinical situations, and it
will not be used anymore.
Prior to the construction of a 3D cystoscope usable in clinical
situation, it was required to validate both the measurement principle
and the mosaicing algorithm on phantoms. Our work is the first which
shows on phantoms with realistic bladder textures that 3D cystoscopy
and mosaicing is feasible while minimally modifying standard
endoscopes. This result is an important step in the process towards
clinical 3D cystoscopy.
The proposed mosaicing algorithm may be improved in several situations.
\indent 1) When cystoscope trajectories are crossing, texture and 3D
surface misalignments will be visible in the mosaic due to small error
accumulation during the mosaicing process. Such crossings have to be
automatically detected so that the surface, texture and color
discontinuities can be corrected. Global 2D mosaic
correction methods (such as the one proposed in~\cite{Weibel:2012a})
may be adapted to improve the visual quality of 3D textured
surfaces, but such approaches will not improve the 3D mosaics
from the dimensional (accuracy) point of view.
2) It was shown in recent studies \cite{Ali16b,Ali16a} that optical
flow can provide an accurate dense correspondence between homologous
points of bladder images with strong illuminations variations, large
displacements or perspective changes, blur or weak textures. Thus, a
perspective of this work is to elaborate an optical flow method with
energies allowing for simultaneous
optimization of the vector field and the parameters of the transformation $T_{3D}^{k-1,k}$.
3) The implementation of our algorithm may be improved as well in
specific cases. When texture features are available and well spread
over the images, they can be fastly detected \cite{Ali:2013}. Then,
the distance between homologous texture points can be used as a
similarity measure for a given $\mathbf{T}_{2D}^{k-1,k}$, instead of
the mutual information in (14).
When textures are lacking as in Fig.~\ref{fig:1}, the mutual
information has to be used. Second, the processing time can obviously
be reduced by replacing the sequential implementation of image
registration (viewpoints $k-1$ and $k$ for $k=1,2,\ldots$) by a
parallel implementation \emph{e.g.}, on a GPU graphical card. By
coupling fast implementations with future clinical endoscopes based on
the current proof-of-concept, we expect that the image processing time
will remain moderate as compared to the times for acquiring images,
for file transfers, and for archiving and retrieving data.
The registration algorithm was designed for bladder lesion
diagnosis. It can be adapted to minimal invasive surgery in the
abdominal cavity, \emph{e.g.}, for computing extended FOV surfaces of
the stomach. Indeed, the external stomach wall is smooth and contains
textures close to bladder textures. Moreover, a laparoscope associated
to a structured-light projector can be calibrated similar to a
cystoscope. Increasing the FOV and providing 3D information
facilitates a surgical intervention. For this application, real-time
mosaicing must be reached.
\begin{acknowledgements}
This work was sponsored by the R\'egion Lorraine and the Centre
National de la Recherche Scientifique (CNRS) under contract PEPS
Biotechno et Imagerie de la Sant\'e. The authors would also like to
thank the Centre Hospitalier Universitaire Nancy-Brabois for
providing pig bladders and Prof. Fran\c{c}ois Guillemin from the
Institut de Canc\'erologie de Lorraine for his expertise in urology.
\end{acknowledgements}
\bibliographystyle{spmpsci}
| 146a43e5be59abe2c63837179ee3a0d5933d8c32 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Custodial symmetry in the SM and the THDM}
\label{Sec:CustSym}
The custodial symmetry is an approximate global \linebreak${SU(2)_L\times SU(2)_R}$ symmetry of the SM which is responsible for the tree-level value of the $\rho$ parameter \cite{Weinberg:1979bn,Susskind:1978ms,Sikivie:1980hm}. Since the Higgs potential respects the remaining $SU(2)_{L+R}$ after electroweak symmetry breaking the $\rho$ parameter is protected from large radiative corrections in the Higgs mass. In the gauge interaction the custodial symmetry is only approximate since it is broken by the hypercharge coupling $g_1$. Moreover, the custodial symmetry is broken by the Yukawa interaction which leads to large corrections to the $\rho$ parameter for large mass differences between quarks in the same doublet \cite{Veltman:1977kh,Chanowitz:1978mv,Chanowitz:1978uj}. A detailed review can be found for example in \cite{Willenbrock:2004hu}.
\subsection{Custodial symmetry in the SM}
As already mentioned the custodial symmetry is a global symmetry of the potential
\begin{equation}
V_\text{SM}\left(\Phi\right)=-\mu^2\Phi^\dagger\Phi+\lambda\left(\Phi^\dagger\Phi\right)^2,
\end{equation}
with the complex doublet
\begin{equation}
\Phi=\begin{pmatrix}
\phi^+ \\
\phi^0
\end{pmatrix}.
\end{equation}
To make the symmetry apparent, it is useful to introduce the complex matrix field
\begin{equation}
\mathcal{M}=\left(\widetilde{\Phi}|\Phi\right)=
\begin{pmatrix}
\phi^{0\ast} & \phi^+ \\
-\phi^- & \phi^0
\end{pmatrix}
\label{Eq:SMmatrix}
\end{equation}
where
\begin{equation}
\widetilde{\Phi}=i\sigma_2\Phi^\ast=
\begin{pmatrix}
0 & 1 \\
-1 & 0
\end{pmatrix}
\begin{pmatrix}
\phi^- \\
\phi^{0\ast}
\end{pmatrix}.
\end{equation}
With this matrix field the potential can be expressed by
\begin{equation}
V_\text{SM}\left(\mathcal{M}\right)=-\mu^2\frac{1}{2}\Tr{\mathcal{M}^\dagger\mathcal{M}}+\lambda\left(\frac{1}{2}\Tr{\mathcal{M}^\dagger\mathcal{M}}\right)^2.
\end{equation}
In addition to the global version of the $SU(2)_L$ gauge symmetry, which transforms $\mathcal{M}$ according to
\begin{equation}
\mathcal{M} \rightarrow L \mathcal{M}
\end{equation}
the potential is also invariant for $SU(2)_R$ transformations of the form
\begin{equation}
\mathcal{M} \rightarrow \mathcal{M} R^\dagger.
\end{equation}
While after electroweak symmetry breaking the vaccum expectation value
\begin{equation}
\langle \mathcal{M} \rangle = \frac{1}{2}
\begin{pmatrix}
v & 0 \\
0 & v
\end{pmatrix}
\end{equation}
breaks both symmetries
\begin{equation}
L \langle \mathcal{M} \rangle \neq \langle \mathcal{M} \rangle;\quad \langle \mathcal{M} \rangle R^\dagger \neq \langle \mathcal{M} \rangle,
\end{equation}
the potential is still invariant under the subgroup \linebreak$SU(2)_{L+R}$ of simultaneous $SU(2)_L$ and $SU(2)_R$ transformations with $L=R$, since
\begin{equation}
L \langle \mathcal{M} \rangle L^\dagger = \langle \mathcal{M} \rangle.
\end{equation}
However the custodial symmetry is not an exact symmetry of the SM. It is broken by the hypercharge coupling $g_1$ in the kinetic term of the Higgs Lagrangian which can be written with the matrix field $\mathcal{M}$ as
\begin{equation}
\frac{1}{2}\Tr{\left(D_\mu\mathcal{M}\right)^\dagger \left(D^\mu\mathcal{M}\right)}
\end{equation}
with the covariant derivative
\begin{equation}
D_\mu\mathcal{M}=\left(\partial_\mu\mathcal{M}+i\frac{g_2}{2}\vec{\sigma}\cdot \vec{W}_\mu \mathcal{M}-i\frac{g_1}{2}B_\mu\mathcal{M}\sigma_3\right).
\end{equation}
When neglecting $g_1$ the kinetic term is invariant under the custodial symmetry since $\vec{W_\mu}$ transforms as a triplet under the global $SU(2)_L$,
\begin{equation}
\vec{\sigma}\cdot\vec{W}_\mu\rightarrow L\vec{\sigma}\cdot\vec{W}_\mu L^\dagger.
\end{equation}
\subsection{Custodial symmetry in the THDM}
A scalar potential with two doublets leads to additional terms which can violate the custodial symmetry. A lot of work has been dedicated to investigations of how the custodial symmetry can be restored in the THDM \cite{Haber:1992py,Pomarol:1993mu,Gerard:2007kn,Grzadkowski:2010dj,Haber:2010bw,Nishi:2011gc}, since there are several possibilities to implement the ${SU(2)_L\times SU(2)_R}$ transformations for two doublets. One way is to introduce matrices similar to \eqref{Eq:SMmatrix} for the two original doublets in \eqref{Eq:Doublets}. The potential is then custodial invariant for $m_{H^\pm}=m_{A^0}$ \cite{Haber:1992py,Pomarol:1993mu}. Different implementations of the custodial transformations were found in \cite{Pomarol:1993mu,Gerard:2007kn}; these require $m_{H^\pm}=m_{H^0}$ in order to obtain a custodial-symmetric potential. However, as shown by \cite{Grzadkowski:2010dj,Haber:2010bw,Nishi:2011gc} these different implementations of the $SU(2)_L \times SU(2)_R$ transformations are dependent on the selected basis of the two doublets and can be related to each other by a unitary change of the basis. Since the two doublets have the same quantum numbers, such a change of basis maintains the gauge interaction but modifies the form of the potential and the Yukawa interaction.
We will demonstrate how the custodial symmetry can be imposed on the potential for the basis of $\Phi_\text{SM}$ and $\Phi_\text{NS}$ as defined in \eqref{Eq:SMdoublet} and \eqref{Eq:NSdoublet}.
This choice of basis corresponds to the so-called Higgs basis as defined for example in \cite{Botella:1994cs,Davidson:2005cw} in which only one of the doublets has a non-vanishing vacuum expectation value in its neutral component. Note that the definition of the Higgs basis is only specified up to a rephasing of the second doublet. As explained in \cite{Haber:2010bw}, the only two possible definitions for matrix fields
which preserve the custodial $SU(2)_{L+R}$ after electroweak symmetry breaking are
\begin{equation}
\mathcal{M}_\text{SM}
=\begin{pmatrix}
\widetilde{\Phi}_\text{SM}|\Phi_\text{SM}
\end{pmatrix}
\end{equation}
and
\begin{equation}
\mathcal{M}_\text{NS}
=
\begin{pmatrix}
\widetilde{\Phi}_\text{NS} | \Phi_\text{NS}
\end{pmatrix}.
\end{equation}
Following \cite{Gerard:2007kn,Haber:2010bw} we write the transformations under the $SU(2)_L\times SU(2)_R$ as
\begin{equation}
\mathcal{M}_\text{SM}\rightarrow L\mathcal{M}_\text{SM}R^\dagger,\quad
\mathcal{M}_\text{NS}\rightarrow L\mathcal{M}_{\text{NS}}R^{\prime\dagger},
\label{Eq:Cust_Trans}
\end{equation}
with $L\in SU(2)_L$ and $R,R^\prime\in SU(2)_R$. Since both doublets transform in the same way under the weak $SU(2)_L$ gauge transformations, they have the same transformation matrix $L$ in \eqref{Eq:Cust_Trans}. The same requirement does not hold for transformations under $SU(2)_R$. As explained in \cite{Gerard:2007kn,Haber:2010bw,Nishi:2011gc}, the matrices $R$ and $R^\prime$ are only related by the fact that the doublets $\Phi_\text{SM}$ and $\Phi_\text{NS}$ have the same hypercharge and that the $U(1)_Y$ is a subgroup of the $SU(2)_R$. When writing $R=\exp{\left(i\theta n^a T^a_R\right)}$ in terms of an unit vector $n^a$ and the generators $T^a_R=\sigma^a/2$ ($a=1,2,3$), the hypercharge operator for the matrix fields is
\begin{equation}
Y=\text{diag}(-1,1)=2T^3_R.
\end{equation}
In order to obtain the same hypercharge transformations for $\mathcal{M}_\text{SM}$ and $\mathcal{M}_\text{NS}$ the matrices $R$ and $R^\prime$ are related by
\begin{equation}
R=X^{-1} R^\prime X,
\end{equation}
with
\begin{equation}
X\exp{(i\theta Y)}X^{-1}=\exp{(i\theta Y)}.
\end{equation}
This requires the matrix $X$ to have the form
\begin{equation}
X=\begin{pmatrix}
e^{-i\chi} & 0 \\
0 & e^{i\chi}
\end{pmatrix}, \quad 0\leq\chi\leq 2\pi.
\end{equation}
A scalar potential is invariant under the transformations in \eqref{Eq:Cust_Trans} if it contains only the invariant combinations
\begin{eqnarray}
&\Tr{\mathcal{M}_\text{SM}^\dagger\mathcal{M}_\text{SM}}=2\Phi^\dagger_\text{SM}\Phi_\text{SM},\\
&\Tr{\mathcal{M}_\text{NS}^\dagger\mathcal{M}_\text{NS}}=2\Phi^\dagger_\text{NS}\Phi_\text{NS},
\end{eqnarray}
and
\begin{equation}
\Tr{\mathcal{M}_\text{SM}^\dagger\mathcal{M}_\text{NS}X}=e^{-i\chi}\Phi_\text{NS}^\dagger\Phi_\text{SM}+e^{i\chi}\Phi_\text{SM}^\dagger\Phi_\text{NS}.
\end{equation}
The parts $V_{I}$ and $V_\text{II}$ of the potential in \eqref{Eq:PotAlign} are clearly custodial invariant. The parts $V_\text{III}$ and $V_{IV}$ are in general not invariant under the transformations in \eqref{Eq:Cust_Trans}. In order to restore the custodial symmetry the parameters have to be adjusted depending on the value of $\chi$. Since we assumed a $CP$ conserving potential with real parameters this is only possible for $\chi=0$ and $\chi=\pi/2$, as we will show in the following.
\subsubsection{\boldmath Custodial symmetry for \texorpdfstring{$\chi=0$}{chi=0}}
\label{Sec:CustTrans1}
For $\chi=0$, we have $R=R^\prime$ and therefore
\begin{eqnarray}
\mathcal{M}_\text{SM}&\rightarrow L\mathcal{M}_\text{SM}R^\dagger,\\
\mathcal{M}_\text{NS}&\rightarrow L\mathcal{M}_\text{NS}R^\dagger.
\end{eqnarray}
This leads to the invariant quantity
\begin{align}
\Tr{\mathcal{M}^\dagger_\text{SM}\mathcal{M}_\text{NS}X}&=\Tr{\mathcal{M}^\dagger_\text{SM}\mathcal{M}_\text{NS}}\notag\\
&=\Phi_\text{NS}^\dagger\Phi_\text{SM}+\Phi_\text{SM}^\dagger\Phi_\text{NS}.
\end{align}
The part $V_{IV}$ from the potential in \eqref{Eq:PotAlign} is invariant under this custodial transformation since it can be written as follows:
\begin{equation}
\begin{aligned}
V_{IV}=&\frac{1}{2t_{2 \beta }}\left(\frac{2 m_{H^0}^2}{v^2}-\lambda _5\right)\cdot\\
&\cdot\Tr{\mathcal{M}_\text{NS}^\dagger\mathcal{M}_\text{NS}}\;\Tr{\mathcal{M}^\dagger_\text{SM}\mathcal{M}_\text{NS}}
\end{aligned}
\end{equation}
If we set $m_{A^0}=m_{H^\pm}$ we can also write $V_\text{III}$ in terms of the invariant quantities,
\begin{align}
V_{III}&\xrightarrow{m_{A^0}=m_{H^\pm}}& \notag\\
&\frac{m_{H^0}^2-m_{H^\pm}^2}{2v^2} \, \left(\Tr{\mathcal{M}_\text{SM}^\dagger\mathcal{M}_\text{NS}}\right)^2\notag\\
&+\left(\frac{2m_{H^\pm}^2+m_{h^0}^2}{v^2}-\lambda_5\right)\cdot\notag\\
&\cdot\frac{1}{4}\Tr{\mathcal{M}_\text{SM}^\dagger\mathcal{M}_\text{SM}}\;\Tr{\mathcal{M}_\text{NS}^\dagger\mathcal{M}_\text{NS}}
\end{align}
Consequently custodial invariance in the potential can be restored for $m_{A^0}=m_{H^\pm}$.
\subsubsection{\boldmath Custodial symmetry for \texorpdfstring{$\chi=\frac{\pi}{2}$}{chi=pi/2}}
\label{Sec:CustTrans2}
For $\chi=\frac{\pi}{2}$ we have
\begin{equation}
X=
\begin{pmatrix}
-i & 0\\
0 & i \\
\end{pmatrix}
\end{equation}
and
\begin{equation}
\Tr{\mathcal{M}_\text{SM}^\dagger\mathcal{M}_\text{NS}X}=-i\Phi_\text{NS}^\dagger\Phi_\text{SM}+i\Phi_\text{SM}^\dagger\Phi_\text{NS}.
\label{Eq:Invquant}
\end{equation}
Invariance of $V_\text{III}$ under this custodial transformation is obtained for $m_{H^0}^2=m_{H^\pm}^2$:
\begin{align}
V_\text{III}&\xrightarrow{m_{H^0}=m_{H^\pm}} & \notag\\
&\frac{m_{A^0}^2-m_{H^\pm}^2}{2v^2} \, \left(\Tr{\mathcal{M}_\text{SM}^\dagger\mathcal{M}_\text{NS}X}\right)^2\notag\\
&+\left(\frac{2m_{H^\pm}^2+m_{h^0}^2}{v^2}-\lambda_5\right)\cdot\notag\\
&\cdot\frac{1}{4}\Tr{\mathcal{M}_\text{SM}^\dagger\mathcal{M}_\text{SM}}\;\Tr{\mathcal{M}_\text{NS}^\dagger\mathcal{M}_\text{NS}}.
\end{align}
However, the part $V_{IV}$ in the potential cannot be written in
terms of the invariant quantity specified in
\eqref{Eq:Invquant}. Consequently, it has to vanish in the case of a
potential invariant under this custodial transformation.
This can be achieved by setting
\begin{equation}
\frac{2m_{H^0}^2}{v^2}=\lambda_5
\end{equation}
or
\begin{equation}
t_\beta=1.
\end{equation}
\section{Feynman rules for the counterterm vertices}
\label{App:Feynmanrules}
In the counterterm vertices in the diagrams in figure~\ref{Fig:VV_t_CTDiagramme} and figure~\ref{Fig:VV_NS_CTDiagramme} we keep only the counterterms which survive the gauge-less limit. These are the mass counterterms of the top quark and the non-standard scalars and the renormalization constant $\delta s_W^2$ which has a remaining contribution in the gauge-less limit (see section~\ref{Sec:Renormalization}). All field counterterms are dropped since they either cancel in the full result or vanish in the gauge-less limit. In the vertices all the momenta are considered as incoming.
Dropping field renormalization, the scalar--scalar two-point vertex counterterm takes the form
\begin{center}
\parbox[c]{0.55\linewidth}{
\includegraphics[width=0.25\textwidth]{SSCT.pdf}
}
\parbox[c]{0.35\linewidth}{\vspace{2.5mm}$=-i\delta m_{S}^2$}
\end{center}
for $S=h^0,H^0,A^0,H^\pm,G^0,G^\pm$.
The Feynman rule involving the top-quark mass counterterm is given by
\begin{center}
\parbox[c]{0.55\linewidth}{
\includegraphics[width=0.25\textwidth]{ttCT.pdf}\vspace{2.5mm}
}
\parbox[c]{0.35\linewidth}{\vspace{2.5mm}$=-i\delta m_t$.\vspace{2.5mm}}
\end{center}
The renormalization conditions for the mass counterterms are given in section~\ref{Sec:Renormalization}.
For the coupling of a massive gauge boson to two scalars we obtain the following counterterm vertices:
\begin{center}
\fbox{
\parbox[c]{0.45\linewidth}{
\includegraphics[width=0.25\textwidth]{VSSCT.pdf}
}
\parbox[c]{0.5\linewidth}{
$=\left(k_{1}-k_{2}\right)_\mu C \left[V_\mu,S_1,S_2\right]$
}
}
\begin{footnotesize}
\begin{align}
C[Z_\mu,h^0,G^0]&=\frac{e}{4 s_W c_W}\frac{\left(s_W^2-c_W^2\right)}{c_W^2}\frac{\delta s_W^2}{s_W^2},\notag\\
C[Z_\mu,H^0,A^0]&=-\frac{e}{4 s_W c_W}\frac{\left(s_W^2-c_W^2\right)}{c_W^2}\frac{\delta s_W^2}{s_W^2},\notag\\
C[Z_\mu,H^-,H^+]&=-i\frac{e}{4 s_W c_W^3}\frac{\delta s_W^2}{s_W^2},\notag\\
C[Z_\mu,G^-,G^+]&=-i\frac{e}{4 s_W c_W^3}\frac{\delta s_W^2}{s_W^2},\notag\\
C[W_\mu^\pm,h^0,G^\mp]&=\pm i\frac{e}{4 s_W }\frac{\delta s_W^2}{s_W^2},\notag\\
C[W_\mu^\pm,H^0,H^\mp]&=\mp i\frac{e}{4 s_W }\frac{\delta s_W^2}{s_W^2},\notag\\
C[W_\mu^\pm,A^0,H^\mp]&=-\frac{e}{4 s_W }\frac{\delta s_W^2}{s_W^2},\notag\\
C[W_\mu^\pm,G^0,G^\mp]&=-\frac{e}{4 s_W }\frac{\delta s_W^2}{s_W^2},\notag
\end{align}
\end{footnotesize}
\end{center}
\newpage
For the coupling of two massive gauge bosons to two scalars the counterterm vertices are
\begin{center}
\fbox{
\parbox[c]{0.45\linewidth}{
\includegraphics[width=0.25\textwidth]{VVSSCT.pdf}
}
\parbox[c]{0.45\linewidth}{
$= g_{\mu\nu} C \left[V_\mu,V_\nu,S_1,S_2\right]$
}
}
\begin{footnotesize}
\begin{align}
C[Z_\mu,Z_\nu,h^0,h^0]&=i\frac{e^2}{2 s_W^2 c_W^2}\frac{s_W^2-c_W^2}{c_W^2}\frac{\delta s_W^2}{s_W^2},\notag\\
C[Z_\mu,Z_\nu,H^0,H^0]&=i\frac{e^2}{2 s_W^2 c_W^2}\frac{s_W^2-c_W^2}{c_W^2}\frac{\delta s_W^2}{s_W^2},\notag\\
C[Z_\mu,Z_\nu,A^0,A^0]&=i\frac{e^2}{2 s_W^2 c_W^2}\frac{s_W^2-c_W^2}{c_W^2}\frac{\delta s_W^2}{s_W^2},\notag\\
C[Z_\mu,Z_\nu,H^+,H^-]&=i\frac{e^2}{2 s_W^2 c_W^2}\frac{s_W^2-c_W^2}{c_W^2}\frac{\delta s_W^2}{s_W^2},\notag\\
C[Z_\mu,Z_\nu,G^0,G^0]&=i\frac{e^2}{2 s_W^2 c_W^2}\frac{s_W^2-c_W^2}{c_W^2}\frac{\delta s_W^2}{s_W^2},\notag\\
C[Z_\mu,Z_\nu,G^+,G^-]&=i\frac{e^2}{2 s_W^2 c_W^2}\frac{s_W^2-c_W^2}{c_W^2}\frac{\delta s_W^2}{s_W^2},\notag\\
C[W_\mu^+,W_\nu^-,h^0,h^0]&=-i\frac{e^2}{2 s_W^2}\frac{\delta s_W^2}{s_W^2},\notag\\
C[W_\mu^+,W_\nu^-,H^0,H^0]&=-i\frac{e^2}{2 s_W^2}\frac{\delta s_W^2}{s_W^2},\notag\\
C[W_\mu^+,W_\nu^-,A^0,A^0]&=-i\frac{e^2}{2 s_W^2}\frac{\delta s_W^2}{s_W^2},\notag\\
C[W_\mu^+,W_\nu^-,H^+,H^-]&=-i\frac{e^2}{2 s_W^2}\frac{\delta s_W^2}{s_W^2}.\notag\\
C[W_\mu^+,W_\nu^-,G^0,G^0]&=-i\frac{e^2}{2 s_W^2}\frac{\delta s_W^2}{s_W^2},\notag\\
C[W_\mu^+,W_\nu^-,G^+,G^-]&=-i\frac{e^2}{2 s_W^2}\frac{\delta s_W^2}{s_W^2},\notag
\end{align}
\end{footnotesize}
\end{center}
The counterterm vertices between the gauge bosons and the fermions in figure~\ref{Fig:VV_t_CTDiagramme} are given by
\begin{center}
\fbox{
\parbox[c]{0.45\linewidth}{
\includegraphics[width=0.25\textwidth]{VffCT.pdf}
}
\parbox[c]{0.45\linewidth}{
$= \left(C_- \omega_- + C_+ \omega_+\right)\gamma_\mu$,
}
}
\begin{footnotesize}
\begin{align}
&W^{+}\overline{t} b:\quad C_+=0,\quad C_-=i\frac{e}{2\sqrt{2}s_W}\frac{\delta s_W^2}{s_W^2},\notag\\
&W^{-} \overline{b} t:\quad C_+=0,\quad C_-=i\frac{e}{2\sqrt{2}s_W}\frac{\delta s_W^2}{s_W^2},\notag\\
&Z \overline{t} t:\quad C_+=i\frac{e}{3c_W^2}\frac{s_W}{c_W}\frac{\delta s_W^2}{s_W^2}, \quad C_-=\frac{ie}{4c_W s_W}\left(1+\frac{1}{3}\frac{s_W^2}{c_W^2}\right)\frac{\delta s_W^2}{s_W^2}\notag.
\end{align}
\end{footnotesize}
\end{center}
\subsection{The Inert-Higgs-Doublet model}
\label{Sec:IHDM}
We now discuss our result in the context of a special version of the THDM, the Inert-Higgs-Doublet Model (IHDM).
Originally proposed in~\cite{Deshpande:1977rw}, it has received topical attention in the light of
neutrino mass phenomenology and dark-matter searches (see e.g.~\cite{Arhrib:2013ela}
for a recent comprehensive analysis and more references).
In the IHDM the SM scalar sector is extended by a second complex doublet with the special feature that the Lagrangian has an exact $Z_2$ symmetry under which all the SM particles are even while the second doublet is odd. This $Z_2$ symmetry has several interesting consequences. The requirement that it stays unbroken forbids a vacuum expectation value of the second doublet. Therefore the doublets of the IHDM are
\begin{eqnarray}
H_1=&
\begin{pmatrix}
G^\pm \\
\frac{1}{\sqrt{2}}\left(v+h^0+iG^0\right)
\end{pmatrix},\\
H_2=&
\begin{pmatrix}
H^\pm\\
\frac{1}{\sqrt{2}}\left(H^0+iA^0\right)
\end{pmatrix}.
\end{eqnarray}
The doublet $H_1$ is identical to the scalar doublet in the SM. It consists of the physical SM-like Higgs boson $h^0$ and the Goldstone bosons $G^0$ and $G^\pm$. The second doublet $H_2$ transforms under the $Z_2$ as $H_2\rightarrow -H_2$ and contains the $CP$-even scalar $H^0$, the $CP$-odd scalar $A^0$ and the charged scalars $H^\pm$. All the terms in the Lagrangian in which the SM particles couple to a single scalar of $H_2$ are forbidden by the $Z_2$ symmetry and the lightest of the scalars from $H_2$ is stable. If this is one of the neutral states $H^0$ or $A^0$ the IHDM provides a dark matter candidate.
\newcommand{\VIHDM}[1][{}]{V^\text{IHDM}_{#1}}
The most general scalar potential which is renormalizable, gauge invariant and respects the $Z_2$ symmetry is given by (see for example \cite{Arhrib:2015hoa})
\begin{align}
\VIHDM=&\mu _1^2 \left(H_1^{\dagger }H_1\right)+\mu
_2^2 \left(H_2^{\dagger } H_2\right)+\Lambda _1 \left(H_1^{\dagger } H_1\right){}^2 \notag\\
&+\Lambda _2\left(H_2^{\dagger } H_2\right){}^2 +\Lambda _3 \left(H_2^{\dagger } H_2\right) \left(H_1^{\dagger } H_1\right)\notag\\
&+\Lambda _4 \left(H_1^{\dagger } H_2\right) \left(H_2^{\dagger } H_1\right)\notag\\
&+\frac{1}{2} \Lambda _5
\left(\left(H_1^{\dagger } H_2\right){}^2+\left(H_2^{\dagger } H_1\right){}^2\right) .
\end{align}
To avoid $CP$-violation all the parameters in the potential are chosen to be real. The minimization condition can be used to eliminate one of the parameters from the potential. From the remaining six parameters four can be expressed by the scalar masses (see \cite{Arhrib:2015hoa} for the explicit relations between the masses and the potential parameters). If we choose $\mu_2^2$ and $\Lambda_2$ for the remaining two parameters we can express the potential by
\begin{align}
\VIHDM=&\VIHDM[I]+\VIHDM[II]+\VIHDM[III] , \\[0.2cm]
\VIHDM[I]=&\frac{m_{h^0}^2 }{2 v^2}\left(H_1^{\dagger } H_1\right){}^2-\frac{1}{2} m_{h^0}^2 \left(H_1^{\dagger } H_1\right) , \\
\VIHDM[II]=&\mu _2^2 \left(H_2^{\dagger } H_2\right)+\Lambda _2 \left(H_2^{\dagger } H_2\right){}^2 , \\
\VIHDM[III]=&\frac{ \left(m_{A^0}^2-2 m_{H^\pm}^2+m_{H^0}^2\right)}{v^2}\left(H_1^{\dagger } H_2\right) \left(H_2^{\dagger } H_1\right)\notag\\
&+\frac{\left(m_{H^0}^2-m_{A^0}^2\right)}{2 v^2}\left[\left(H_1^{\dagger } H_2\right){}^2+\left(H_2^{\dagger } H_1\right){}^2\right]\notag\\
&+\frac{2 \left(m_{H^\pm}^2-\mu _2^2\right)}{v^2}\left(H_1^{\dagger } H_1\right) \left(H_2^{\dagger } H_2\right).
\end{align}
We see that the SM-like doublet $\Phi_\text{SM}$ in the aligned THDM is identical to the doublet $H_1$ in the IHDM. The non-standard doublet $\Phi_\text{NS}$ in \eqref{Eq:NSdoublet} differs from the doublet $H_2$ of the IHDM by the overall sign in front of the $CP$-even scalar $H^0$. However, our result is independent on this overall sign, since the $CP$-even scalar $H^0$ appears only as an internal particle in the calculated self-energies. Therefore we can identify the doublet $H_2$ with the doublet $\Phi_\text{NS}$. By using this identifications we can relate the potential between the IHDM and the more general THDM in the alignment limit in order to interpret our results in the context of the IHDM:
\begin{itemize}
\item There is no non-standard correction to $\Delta\rho$ from the top-Yukawa interaction, since the interaction of the fermions with the non-standard scalars is forbidden by the $Z_2$ symmetry.
\item The part $\VIHDM[I]$ has the same structure as the scalar potential of the SM and will not lead to contributions to the $\rho$ parameter since it is invariant under the custodial symmetry (see section~\ref{Sec:CustSym}).
\item In the IHDM all the quartic couplings between four non-standard scalars are proportional to $\Lambda_2$. However, in our calculation in the aligned THDM we found that all the contributions to $\Delta\rho$ from couplings between four non-standard scalars vanish (see section~\ref{Subsec:drhoNSHiggs}). The responsible arguments can also be transferred to the IHDM.
\item When we identify $H_1$ with $\Phi_\text{SM}$ and $H_2$ with $\Phi_\text{NS}$ we see that the part $V_{III}$ of the potential in the aligned THDM can be obtained by the replacement
\begin{equation}
\mu_2^2 = \frac{1}{2}\lambda_5 v^2- \frac{ m_{h^0}^2}{2}
\label{Eq:mu2relation}
\end{equation}
in $\VIHDM[III]$. Consequently for the calculation of the $\rho$ parameter in the IHDM we get corrections corresponding to $\Delrho[NS]{1}$ and $\delrho[H,Mix]{2}$. The one-loop correction $\Delrho[NS]{1}$ is identical in the IHDM since it is independent of $\lambda_5$. The two-loop part $\delrho[H,Mix]{2}$ can be written in terms of the IHDM parameter $\mu_2^2$ by using \eqref{Eq:mu2relation}.
\item As mentioned in section~\ref{Subsec:drhoNSHiggs}, the correction $\delrho[H,NS]{2}$ contains the interaction between three of the non-standard scalars $H^0$, $A^0$ and $H^\pm$ which follows from the part $V_{IV}$ of the potential in \eqref{Eq:PotAlign}. In the IHDM couplings between three non-standard scalars are forbidden because of the exact $Z_2$ symmetry. As a consequence, corrections to the $\rho$ parameter which would correspond to $\delrho[H,NS]{2}$ are absent in the IHDM.
\end{itemize}
\section{One- and two-loop integrals}
\subsection{Scalar one-loop integrals}
\label{App:1LIntegrals}
Here we list all the one- and two-loop scalar integrals that are used in our calculation.
They are evaluated in dimensional regularization \cite{tHooft:1972fi,Bollini:1972ui,Ashmore:1972uj}
with dimension $D$ of the integrated momentum and the associated mass parameter $\mu_D$,
\begin{equation}
\int d^4q\rightarrow\mu^{4-D}_D\int d^Dq \, .
\end{equation}
The scalar integrals are expanded in $\delta=(D-4)/2$ and the divergencies appear as poles in $\delta$.
The reduction of the one-loop tensor integrals to scalar integrals and
their classification is following the work of
\cite{Passarino:1978jh,tHooft:1978xw} (for more details and notation see \cite{Denner:1991kt}).
The only one-loop integrals which are needed for the evaluation of the self-energies are
\begin{equation}
A_0(m^2)=\int\frac{d^Dq}{i\pi^2} \frac{(2\pi \mu_D)^{(4-D)}}{(q^2-m^2+i\epsilon)},
\end{equation}
\begin{multline}
B_0(p^2,m_1^2,m_2^2)\\
=\int \frac{d^Dq}{i\pi^2}\frac{(2\pi \mu_D)^{(4-D)}}{(q^2-m_1^2+i\epsilon)((p+q)^2-m_2^2+i\epsilon)}.
\end{multline}
We need an expansion up to order $\delta$ of the scalar integrals; analytic expressions can be found in \cite{Nierste:1992wg,Hollik:2014bua}.
\subsection{Scalar two-loop integrals}
\label{App:2LIntegrals}
The notation of the two-loop integrals follows the conventions of \cite{Weiglein:1993hd}.
For vanishing external momentum
all two-loop integrals in the self energies can be reduced to the scalar integral
\begin{multline}
T_{134}\left(m_1^2,m_2^2,m_3^2\right)=\left(\frac{(2\pi \mu_D)^{(4-D)}}{i\pi^2}\right)^2\cdot\\
\int \frac{ \text{d}^Dq_1 \text{d}^Dq_2}{\left(k_1^2-m_1^2+i \epsilon\right)\left(k_3^2-m_2^2+i \epsilon\right)\left(k_4^2-m_3^2+i \epsilon\right)}
\end{multline}
with $k_1=q_1$, $k_3=q_2-q_1$ and $k_4=q_2$.
This integral can be calculated analytically and the result can be found in \cite{Berends:1994ed,Davydychev:1992mt}.
\section{Approximations and outline of the calculation}
\label{Sec:Outline}
In order to evaluate the leading two-loop contributions from the
Yukawa sector and from the Higgs self-interactions
a number of approximations can be made.
\subsection{Gauge-less limit and top-Yukawa approximation}
Since our focus is on the corrections to $\Delta\rho$ originating from
the top-Yukawa and the Higgs self-couplings in the THDM we neglect all
other couplings. This means that we work in the gauge-less limit
(as done in \cite{Haestier:2005ja} for the MSSM)
in which the electroweak gauge couplings $g_{1,2}$ are put to zero and
thus the gauge-boson masses are also equal to zero
\begin{equation}
M_W^2=\frac{g_2^2 v^2}{4}\rightarrow 0,\quad
M_Z^2=\frac{\left(g_1^2+g_2^2\right)v^2}{4}\rightarrow 0 \, ,
\end{equation}
while their ratio in $c_W$ and $s_W$ stays constant.
Moreover, the masses of the Goldstone bosons are zero,
\begin{equation}
m_{G^0}=m_{G^\pm}=0 ,
\end{equation}
in the gauge-less limit.
In addition we are using the top-Yukawa approximation in which all the
fermion masses with the exception of the top-quark mass are
neglected. Especially for the bottom quark, which appears in some of
the diagrams for the $\order{\alpha_t^2}$ contributions, we set $m_b=0$.
Differently from the top-Yukawa coupling, which is universal in
all of the four models, the Yukawa coupling of the bottom quark is
model specific. In models of type-I and type-X, the bottom- and
top-Yukawa interactions have the same structure, and the
additional contributions to $\Delta\rho$ from the $b$~quark are negligible due to the
small value of $m_b$. In models of type-II or type-Y, the $b$-Yukawa
coupling can be enhanced by $t_\beta$, and the top-Yukawa approximation
is justified in these models as long as we do not consider large
values of $t_\beta$. For large $t_\beta$ values additional constraints from
flavor physics would have to be taken into account as well.
\subsection{The alignment limit}\label{subsec:Align}
Due to the fact that a scalar particle with a mass of approximately
125 GeV has been observed at the LHC~\cite{Aad:2012tfa,Chatrchyan:2012ufa}
we can identify one of the $CP$-even scalars with the observed resonance.
Choosing $h^0$ (without loss of generality) corresponds to setting
\begin{equation}
m_{h^0}=125\text{ GeV}.
\label{Eq:Mh0value}
\end{equation}
Furthermore the analysis of the Higgs couplings by \mbox{ATLAS} \cite{Aad:2015gba} and CMS \cite{Khachatryan:2014jba} indicate no significant deviations from the couplings of the Higgs boson in the SM. Therefore we choose to work in the alignment limit \cite{Dev:2014yca,Bernon:2015qea}, in which
the angles are correlated via
\begin{equation}
\alpha=\beta-\frac{\pi}{2}
\label{Eq:Alignalpha}
\end{equation}
and the couplings of $h^0$ to the vector bosons and fermions are identical to the corresponding couplings of the Higgs boson in the SM. In this limit the $CP$-even Higgs states are obtained by
\begin{equation}
\begin{pmatrix}
H^0 \\
h^0
\end{pmatrix}
=
\begin{pmatrix}
s_\beta & -c_\beta \\
c_\beta & s_\beta \\
\end{pmatrix}
\begin{pmatrix}
\eta_1 \\
\eta_2
\end{pmatrix}
\end{equation}
and the two doublets can be rewritten as
\begin{eqnarray}
\Phi_1&=c_\beta\Phi_\text{SM}-s_\beta\Phi_\text{NS},\\
\Phi_2&=s_\beta\Phi_\text{SM}+c_\beta\Phi_\text{NS}
\end{eqnarray}
with
\begin{eqnarray}
\Phi_\text{SM}
&=
\begin{pmatrix}
G^+\\
\frac{1}{\sqrt{2}}\left(v+h^0+iG^0\right)
\end{pmatrix},\label{Eq:SMdoublet}\\
\Phi_\text{NS}
&=
\begin{pmatrix}
H^+\\
\frac{1}{\sqrt{2}}\left(-H^0+iA^0\right)
\end{pmatrix}.
\label{Eq:NSdoublet}
\end{eqnarray}
Moreover, the relations for $\lambda_1$, $\lambda_2$ and $\lambda_3$ simplify to
\begin{eqnarray}
\lambda_1=&\frac{e^2}{8M_W^2s_W^2}\frac{m_{H^0}^2}{c_\beta^2 }
+\frac{1}{4}\left(1-t_\beta^2\right)\lambda_5
\\
\lambda_2=&\frac{e^2}{8M_W^2s_W^2}
\frac{m_{H^0}^2}{s_\beta^2}+\frac{1}{4}\left(1-\frac{1}{t_\beta^2}\right)\lambda_5
\\
\lambda_3=&\frac{e^2}{8M_W^2s_W^2}\left(m_{h^0}^2-m_{H^0}^2\right)-\frac{\lambda_5}{4}.
\end{eqnarray}
The potential can be rewritten in terms of the doublets given in \eqref{Eq:SMdoublet} and \eqref{Eq:NSdoublet}. For the classification of the different contributions to $\Delta\rho$ we split it in the four parts
\begin{align}
V=&V_{I}+V_{II}+V_{III}+V_{IV};\label{Eq:PotAlign}\\
%
V_{I}=&\frac{m_{h^0}^2 }{2 v^2}\left(\Phi_{\text{SM}}^{\dagger } \Phi _{\text{SM}}\right){}^2+\frac{1}{2}m_{h^0}^2\left(\frac{1}{4} v^2 - \Phi_{\text{SM}}^{\dagger } \Phi_{\text{SM}}\right),\label{Eq:V_I}\\
%
V_{II}=&\frac{1}{2 v^2}\left(\Phi_{\text{NS}}^{\dagger } \Phi _{\text{NS}}\right){}^2 \left(m_{h^0}^2+\frac{4 m_{H^0}^2}{t_{2 \beta }^2}-\frac{2 \lambda _5 v^2}{t_{2 \beta }^2}\right)
\notag\\
&+\frac{1}{2} \left(\lambda _5 v^2-m_{h^0}^2\right) \left(\Phi_{\text{NS}}^{\dagger } \Phi _{\text{NS}}\right),\label{Eq:V_II}\\
%
V_{III}=&\frac{\left(m_{A^0}^2-2 m_{H^\pm}^2+m_{H^0}^2\right)}{v^2}
\left(\Phi_{\text{SM}}^{\dagger } \Phi _{\text{NS}}\cdot\Phi_{\text{NS}}^{\dagger } \Phi_{\text{SM}}\right) \notag\\
&+\frac{\left(m_{H^0}^2-m_{A^0}^2\right)}{2 v^2}\left( \left(\Phi_{\text{NS}}^{\dagger } \Phi _{\text{SM}}\right){}^2+\left(\Phi_{\text{SM}}^{\dagger } \Phi _{\text{NS}}\right){}^2\right)\notag\\
&+ \left(\frac{2 m_{H^\pm}^2+m_{h^0}^2}{v^2}-\lambda _5\right)\left(\Phi_{\text{NS}}^{\dagger } \Phi _{\text{NS}} \cdot \Phi_{\text{SM}}^{\dagger } \Phi _{\text{SM}}\right),\label{Eq:V_III}\\
V_{IV}=&\frac{1}{t_{2 \beta }}\left(\frac{2 m_{H^0}^2}{v^2}-\lambda _5\right)\cdot\notag\\
&\left( \Phi_{\text{NS}}^{\dagger } \Phi _{\text{NS}}\cdot \Phi_{\text{NS}}^{\dagger } \Phi _{\text{SM}}+\Phi_{\text{NS}}^{\dagger } \Phi _{\text{NS}}\cdot \Phi_{\text{SM}}^{\dagger } \Phi_{\text{NS}}\right)\label{Eq:V_IV}.
\end{align}
Imposing \eqref{Eq:Alignalpha} on the top-Yukawa interaction given in
\eqref{Eq:topLagrangian} one finds that the resulting coupling between the SM-like scalar $h^0$ and the top quark is identical to the top-Yukawa coupling in the SM, while the couplings to the non-standard Higgs states $A^0$, $H^0$ and $H^\pm$ receive an additional factor of $t_\beta^{-1}$.
The various types I, II, X, Y of THDMs coincide within the approximations made in this paper.
\subsection{Outline of the calculation}
All needed diagrams and amplitudes are generated with the help of the Mathematica package \texttt{FeynArts} \cite{Hahn:2000kx}. The
evaluation of the one-loop amplitudes and the calculation of the renormalization constants is done with the help of
the package \texttt{FormCalc} \cite{Hahn:1998yk}, which is also employed to generate a Fortran expression of the result.
In the numerical analysis of the one-loop result the integrals are evaluated with the program \texttt{LoopTools} \cite{Hahn:1998yk}.
The package \texttt{TwoCalc} \cite{Weiglein:1993hd,Weiglein:1992rj} is applied to deal with the Lorentz and Dirac algebra of the two-loop
amplitudes and to reduce the tensor integrals to scalar integrals. In the gauge-less limit the external momenta of all the two-loop diagrams
are equal to zero and the result depends only on the one-loop functions $A_0$ and $B_0$ (see appendix~\ref{App:1LIntegrals}) and on the
two-loop function $T_{134}$ (see appendix~\ref{App:2LIntegrals}) for which analytic expressions are known~\cite{Davydychev:1992mt,Berends:1994ed}
and Fortran functions are encoded in the program \texttt{FeynHiggs} \cite{Heinemeyer:1998yj,Hahn:2009zz}.
For the automation of the calculation and the implementation of the result in Fortran, the techniques from \cite{Hahn:2015gaa} are
employed.
\section{Renormalization scheme}
\label{Sec:Renormalization}
For our calculation we are using the on-shell renormalization scheme
with the conventions from \cite{Denner:1991kt} in which the masses and
couplings are related to physical parameters. For the renormalization
of the Higgs sector the parameters in the Higgs potential can be
replaced by bare parameters $\hat{v}_{i,0}$ and $\lambda_{i,0}$. Also
the vacuum expectation values $v_1$ and $v_2$ are renormalized in
order to correct for shifts in the minimum of the Higgs potential through
radiative corrections. The resulting renormalization constants can be
translated into counterterms for the masses and mixing angles and for
the tadpoles of $h^0$ and $H^0$. For the subloop renormalization
in the two-loop self-energies we need the following parameters and counterterms:
\begin{align}
M^2_{W,0} & =M_W^2+\delta M_W^2,\label{Eq:MWbare}\\
M^2_{Z,0} & =M_Z^2+\delta M_Z^2,\label{Eq:MZbare}\\
m_{f,0} & =m_f+\delta m_f, \label{Eq:dMfren}\\
m_{S,0}^2 & = m_S^2 +\delta m_S^2;\quad(S=h^0,H^0,A^0,H^\pm) \label{Eq:dMSren},\\
T_{h,0} & =T_{h}+\delta T_{h},\\
T_{H,0} & =T_{H}+\delta T_{H}.
\end{align}
The tadpole counterterms are fixed such that they cancel all the tadpole diagrams of $h^0$ and $H^0$.
The resulting renormalization conditions are given by
\begin{align}
\delta T_{h}&=-T^{(1)}_{h},\\
\delta T_{H}&=-T^{(1)}_{H},
\end{align}
where $T^{(1)}_{h,H}$ denote the sum of the respective one-loop Higgs tadpole graphs.
The tadpole counterterms determine the mass counterterms for the Goldstone bosons,
\begin{equation}
\begin{aligned}
\delta m_{G^0}^2&=\delta m_{G^\pm}^2\\
&=-\frac{e}{2 M_W s_W} \left(s_{\beta-\alpha} \delta T_{h}+c_{\beta-\alpha} \delta T_{H}\right) ,
\end{aligned}
\end{equation}
following from \eqref{Eq:Goldstonemass}. In the alignment limit this simplifies to
\begin{equation}
\delta m_{G^0}^2=\delta m_{G^\pm}^2=-\frac{e}{2 M_W s_W} \delta T_{h}.
\end{equation}
In the on-shell scheme mass renormalization is done by the requirement that the renormalized masses are equal to the pole masses,
defined by the real part of the poles of the corresponding propagators.
Therefore, the mass counterterms
have to absorb the corrections from the self-energies. In terms of the gauge-boson self-energies ($V=W,Z$)
\begin{multline}
\Sigma_{V}^{\mu\nu}\left(p^2\right)=\left(g^{\mu\nu}-\frac{p^\mu p^\nu}{p^2}\right)\Sigma_{V,T}\left(p^2\right)\\
+\frac{p^\mu p^\nu}{p^2}\Sigma_{V,L}\left(p^2\right) ,
\label{transverseselfenergy}
\end{multline}
the fermion self-energies
\begin{equation}
\Sigma_f\left(p^2\right)=\slashed p \omega_-\Sigma^L_f\left(p^2\right)+\slashed p \omega_+ \Sigma^R_f\left(p^2\right)+m_f\Sigma^S_f\left(p^2\right),
\label{fermionselfdec}
\end{equation}
and the scalar self-energies $\Sigma_S(p^2)$,
the on-shell renormalization conditions yield the mass counterterms
\begin{align}
\delta M_W^2=&\Re\Self[W,T]{(1)}{M_W^2},\\
\delta M_Z^2=&\Re\Self[Z,T]{(1)}{M_Z^2},\\
\delta m_f=& \frac{m_f}{2}\big[\Re\Sigmaf{L}{m_f^2}+\Re\Sigmaf{R}{m_f^2}\notag\\
&\hphantom{\frac{m_f}{2}\big[}+2\Re\Sigmaf{S}{m_f^2}\big],\\
\delta m_S^2 =&\Re\Self[S]{}{m_S^2}\quad (S=h^0,H^0,A^0,H^\pm).
\end{align}
The upper index of the gauge-boson self-energies indicates the loop order, since we need also the two-loop contribution to the gauge-boson
self-energies in the calculation of $\Delta\rho$. For all the other
quantities, one-loop renormalization is sufficient and we drop the
loop index. Furthermore, we will write $\Sigma_{V}\equiv\Sigma_{V,T}$ ($V=W,Z$)
for the transverse part of the gauge-boson self-energies.
In the on-shell scheme the definition of the electroweak mixing angle by \eqref{Eq:CW_Definition} and \eqref{Eq:SW_Definition} is valid to all orders in perturbation theory.
Inserting the bare masses from \eqref{Eq:MWbare} and \eqref{Eq:MZbare} yields
\begin{equation}
s_{W,0}^2=1-c_{W,0}^2=1-\frac{M_{W,0}^2}{M_{Z,0}^2}
\label{Eq:SWbare}
\end{equation}
and expanding the ratio of the bare masses up to one-loop order leads to the counterterm
\begin{equation}
\frac{\delta s_W^2}{s_W^2}=-\frac{c_W^2}{s_W^2}\frac{\delta c_W^2}{c_W^2}=\frac{c_W^2}{s_W^2}\left(\frac{\delta M_Z^2}{M_Z^2}-\frac{\delta M_W^2}{M_W^2}\right) .
\end{equation}
In the gauge-less limit the ratios $\dMVsq{V}/M_V^2$ have remaining
contributions, since the gauge couplings of $\order{g_{1,2}^2}$ in the
self-energies cancel with those contained in the gauge-boson
masses. The resulting one-loop counterterms in the
gauge-less limit are thus given by
\begin{equation}
\frac{\delta M_W^2}{M_W^2}=\frac{\Re\Self[W]{(1)}{0}}{M_W^2},\quad \frac{\delta M_Z^2}{M_Z^2}=\frac{\Re\Self[Z]{(1)}{0}}{M_Z^2}.
\label{Eq:dMVgless1L}
\end{equation}
Renormalization of the electric charge is not needed in the gauge-less limit.
Moreover, we do not need field renormalization because all the
field counterterms drop out in our calculation.
\section{Numerical results}
\label{Sec:Results}
In this part we present the numerical results of the two-loop corrections to the $\rho$ parameter.
We study the dependence on the various parameters of the aligned THDM
and compare the non-standard two-loop contributions with the one-loop result
which is part of existing calculations of electroweak precision observables so far.
In this way the parameter regions emerge where the one-loop calculations
are insufficient and bounds on parameters derived from experimental precision data
will be significantly changed when the two-loop terms are taken into account.
The values for the SM input parameters are \cite{Olive:2016xmw}
\begin{eqnarray}
M_W =&80.385 \text{ GeV}, \\
M_Z =&91.1876 \text{ GeV}, \\
m_t =&173.21 \text{ GeV}.
\end{eqnarray}
For the mass of the SM-like Higgs state $h^0$ we take over the value $m_{h^0} = 125\, \text{GeV}$.
The effect of non-standard corrections to electroweak observables is often parametrized in terms of the parameter
set $S$, $T$, $U$, originally defined in \cite{Peskin:1990zt,Peskin:1991sw}.
Following the conventions of~\cite{Olive:2016xmw}, the quantity $T$ is related
to the correction $\Delta\rho$ via
\begin{equation}
\Delta\rho=\hat{\alpha}\left(M_Z\right)T
\end{equation}
with the running electromagnetic fine structure constant\cite{Olive:2016xmw}
\begin{equation}
\hat{\alpha}\left(M_Z\right)^{-1}=127.950\pm0.017 \, .
\end{equation}
The current value of $T$ ~\cite{Olive:2016xmw},
determined from experimental data,
\begin{equation}
\label{Eq:Texp}
T=0.08\pm0.12 \, ,
\end{equation}
can be translated into bounds for $\Delta\rho$ according to
\begin{equation}
-0.000313\leq \Delta\rho \leq 0.00156,
\end{equation}
which can be used for a quick estimate of the effect of the
higher-order contributions to $\Delta\rho$ in view of current
experimental constraints.
\subsection{Results for the top-Yukawa contribution}
\label{Sec:NStResults}
\begin{figure}[ht]
\begin{center}
\includegraphics[width=\linewidth]{drhoNStdecpl.pdf}
\includegraphics[width=\linewidth]{drhoNStoptbvari.pdf}
\caption{Analysis of $\delrho[t,NS]{2}$. The upper panel presents a variation of the degenerate masses $m_{H^0}$, $m_{A^0}$ and $m_{H^\pm}$ up to large values. The solid lines correspond to different values of $t_\beta$.
In the lower panel $\delrho[t,NS]{2}$ is plotted as a function of $t_\beta$ for different values of $m_H^\pm$. The masses of $H^0$ and $A^0$ are fixed at $m_{H^0}=350\text{ GeV}$ and $m_{A^0}=300\text{ GeV}$. The value of the two-loop top-Yukawa correction in the SM, $\delta\rho^{(2)}_{\rm t,SM} = -1.60\cdot 10^{-4}$, is shown by the black dashed line for comparison.}
\label{Fig:NStopResults}
\end{center}
\end{figure}
\noindent
We start with the analysis of the contribution $\delrho[t,NS]{2}$ which is originating from the coupling between the top quark and the non-standard scalars. As a first test of our result we examine the behaviour in the so-called decoupling limit \cite{Gunion:2002zf}, in which the masses of the non-standard scalars are much larger than $m_h^0$. In this limit the scalar sector of the THDM can be described by an effective theory which is identical to the SM Higgs sector. Consequently we expect $\delrho[t,NS]{2}$ to vanish for large, equal non-standard Higgs masses. The decoupling scenario is investigated in the upper panel of figure~\ref{Fig:NStopResults}, where $\delrho[t,NS]{2}$ is shown for degenerate masses of the non-standard scalars. The solid lines represent results for different values of $t_\beta$. Since the top-Yukawa coupling breaks the custodial symmetry this contribution is still non-zero, even if the custodial symmetry in the Higgs potential is restored by equal masses of the charged and neutral Higgs states. As expected it approaches zero when the masses increase. Moreover, we can see that larger values of $t_\beta$ suppress the correction. The reason is that the coupling of the top quark to the scalars $H^0$, $A^0$ and $H^\pm$ scales with $t_\beta^{-1}$ in the alignment limit (see section~\ref{subsec:Align}).
The influence of $t_\beta$ is visualised on the lower panel of figure~\ref{Fig:NStopResults} with $\delrho[t,NS]{2}$ for the mass configurations as described by the legend, showing the decrease of the contribution with $t_\beta$. In addition different mass splittings between charged and neutral scalars yield noticable deviations in the result and can even lead to different signs.
In general, the top-Yukawa contribution is of the order of the SM value $\delta\rho^{(2)}_{\rm t,SM}$
or smaller.
In order to test the validity of the top-Yukawa approximation, we repeated our calculation including also the contribution from the bottom-Yukawa coupling.
In the THDM of type-I and type-X the additional corrections from the bottom-Yukawa coupling are
negligibly small, as expected from their suppression by the $b$-quark mass
(see section \ref{Sec:Outline}).
In the type-II and type-Y models, the contribution from the bottom-Yukawa coupling can be enhanced for large values of $t_\beta$ since the coupling of the $b$-quark to the non-standard scalars carries a factor $t_\beta$ in the alignment limit.
Additional two-loop contributions from finite $m_b$ that reach the level of
$\delrho[t,SM]{2}$, require $t_\beta\simeq 40-50$.
For such large values of $t_\beta$, however, one has to prevent the non-standard scalar
self-couplings from becoming non-perturbative by
restricting the parameter $\lambda_5$ to be very close to
$\lambda_5 v^2= 2m_{H^0}^2$\cite{Akeroyd:2000wc,Das:2015qva}.
Moreover, the constraints from flavour physics give further significant restrictions
for large values of $t_\beta$
(see for example \cite{Deschamps:2009rh,Enomoto:2015wbn}).
\subsection{Results for the non-standard scalar contribution}
\label{Sec:NSResults}
\begin{figure}[ht]
\begin{center}
\includegraphics[width=\linewidth]{drhoNSHmasssplit.pdf}
\caption{Effect of mass differences between neutral and charged scalars on $\delrho[H,NS]{2}$ for $\lambda_5=\pm1$. The neutral masses are fixed at $m_{H^0}=350\text{ GeV}$ and $m_{A^0}=400\text{ GeV}$. The mass of $H^\pm$ is varied from $250\text{ GeV}$ to $500\text{ GeV}$. The solid lines represent different values of $t_\beta$ as explained in the legend. The blue dashed line shows the non-standard one-loop correction $\Delrho[NS]{1}$ for comparison. The grey area depicts the bounds from the experimental limits of the $T$ parameter.}
\label{Fig:drhoNSHmasssplit}
\end{center}
\end{figure}
\begin{figure}[ht]
\begin{center}
\includegraphics[width=\linewidth]{drhoNSHtbvari.pdf}
\caption{Influence of a variation of $t_\beta$ on $\delrho[H,NS]{2}$ for the specified mass configurations. The result is plotted for different values of $\lambda_5$. The blue dashed line gives the value of the non-standard one-loop correction $\Delrho[NS]{1}$ for the specified masses. The grey area depicts the bounds from the experimental limits of the $T$ parameter.}
\label{Fig:drhoNSHtbvari}
\end{center}
\end{figure}
We now discuss the numerical results of the contribution $\delrho[H,NS]{2}$ which originates from the coupling between three non-standard scalars as described in section~\ref{Subsec:drhoNSHiggs}. The influence of a mass splitting between charged and neutral scalars is presented in figure~\ref{Fig:drhoNSHmasssplit}. The two panels show results for $m_{H^0}=350 \text{ GeV}$, $m_{A^0}=400\text{ GeV}$ and $\lambda_5=\pm1$. The variation of $m_{H^\pm}$ is performed such that it yields similar mass differences for the specified parameter settings. The different lines correspond to different values of $t_\beta$ as defined in the legend. For comparison the blue dashed line displays the result for the one-loop non-standard correction $\Delrho[NS]{1}$. The grey area indicates the bounds from the $T$ parameter in~(\ref{Eq:Texp}).
We see that the contribution $\delrho[H,NS]{2}$ can give corrections to the $\rho$ parameter which are comparable in size or even larger than the one-loop correction. The reason are the new couplings between three non-standard scalars which enter for the first time in the two-loop contribution. Adding the two-loop corrections to the one-loop result can lead to noticeable modifications of the parameter region allowed by the constraints on $T$.
The triple non-standard scalar couplings arise from the term $V_{IV}$ of the potential in \eqref{Eq:PotAlign}, when the vacuum expectation value
\begin{equation}
\langle\Phi_\text{SM}\rangle=\frac{1}{\sqrt{2}}
\begin{pmatrix}
0 \\
v
\end{pmatrix}
\end{equation}
is inserted for the doublet $\Phi_\text{SM}$. Since they enter quadratically in all the diagrams in figure~\ref{Fig:NSHiggs2Loop}, the contribution $\delrho[H,NS]{2}$ is proportional to (see \eqref{Eq:V_IV})
\begin{equation}
\frac{1}{4}\left(\frac{1}{t_\beta}-t_\beta\right)^2\left(\frac{2m_{H^0}^2}{v^2}-\lambda_5\right)^2.
\label{Eq:Prefactor}
\end{equation}
The prefactor explains the strong influence of $t_\beta$ on the results in figure~\ref{Fig:drhoNSHmasssplit}. The enhancement of the coupling can be weakened for positive values of $\lambda_5$ (see the lower panel of figure~\ref{Fig:drhoNSHmasssplit}) or increased for negative values of $\lambda_5$ (see the upper panel of figure~\ref{Fig:drhoNSHmasssplit}).
The effect of the custodial transformations described in section~\ref{Sec:CustSym} is also visible in figure~\ref{Fig:drhoNSHmasssplit}. The one-loop contribution $\Delrho[NS]{1}$ is zero for $m_{H^0}=m_{H^\pm}$ and $m_{A^0}=m_{H^\pm}$ since it originates only from the part $V_{III}$ of the potential which is custodial symmetric for these two mass settings.
As explained in section~\ref{Sec:CustTrans1} the part $V_{IV}$ is
invariant under the custodial transformation for
$\chi=0$. Consequently $\delrho[H,NS]{2}=0$ for $m_{A^0}=m_{H^\pm}$ since all
the involved couplings are custodial invariant for this mass degeneracy.
However, for $m_{H^0}=m_{H^\pm}$ we have $\delrho[H,NS]{2}\neq 0$ since in that case $V_{III}$
is invariant only under custodial transformations for $\chi=\frac{\pi}{2}$,
but then $V_{IV}$ is not invariant and
the triple couplings between three non-standard scalars
hence break the custodial symmetry
(see section~\ref{Sec:CustTrans2}).
The dependence of $\delrho[H,NS]{2}$ on $t_\beta$ is visualized
directly in figure~\ref{Fig:drhoNSHtbvari} for different values of $\lambda_5$,
displaying the increase with $t_\beta$
and the modification by the choice of $\lambda_5$ according to~\eqref{Eq:Prefactor}.
\subsection{Results for the mixed scalar contribution}
\begin{figure}[ht]
\begin{center}
\includegraphics[width=\linewidth]{drhomixsplit.pdf}
\caption{Influence of mass splitting between charged and neutral
scalars on $\delrho[H,Mix]{2}$. The two plots show different values
of $m_{H^0}$ and $m_{A^0}$, and the variation of $m_{H^\pm}$ leads to
comparable mass differences for the different mass
configurations. The results are independent of $t_\beta$.
The different lines represent different values of $\lambda_5$.
The blue dashed line shows the result of the non-standard one-loop
correction $\Delrho[NS]{1}$ for comparison.
The grey area depicts the bounds from the experimental limits of the $T$ parameter.}
\label{Fig:drhomixsplit}
\end{center}
\end{figure}
In the last part we discuss the contribution $\delrho[H,Mix]{2}$ from the interaction of the SM-like scalars $h^0$, $G^0$, $G^\pm$ with the non-standard scalars $H^0$, $A^0$, $H^\pm$. Similar to the one-loop correction $\Delrho[NS]{1}$ it originates only from the part $V_{III}$ of the potential in \eqref{Eq:PotAlign}. Consequently it is independent of $t_\beta$ (see \eqref{Eq:V_III}).
In figure~\ref{Fig:drhomixsplit} we analyse the influence of a mass splitting between the charged and neutral scalars. We show two scenarios for different values of $m_{H^0}$ and $m_{A^0}$, while the mass of $m_{H^\pm}$ is varied in such a way that the mass splittings are comparable. The three solid lines present the results for different values of $\lambda_5$. The blue dashed line gives the one-loop contribution $\Delrho[NS]{1}$ for comparison.
The results of figure~\ref{Fig:drhomixsplit} can again be explained with the help of the custodial symmetry. As discussed in section~\ref{Sec:CustSym} there are the two possible ways,
\begin{equation}
m_{H^0}=m_{H^\pm}
\end{equation}
or
\begin{equation}
m_{A^0}=m_{H^\pm} \, ,
\end{equation}
to restore a custodial symmetry in $V_{III}$. For these two mass configurations $\Delrho[NS]{1}$ and $\delrho[H,Mix]{2}$ vanish, since they do not contain any additional custodial-symmetry breaking couplings.
While the one-loop contribution originates only from the coupling of the non-standard scalars to the Goldstone bosons, new couplings between $h^0$ and the non-standard scalars enter the two-loop diagrams in figure~\ref{Fig:Mix2Loop}. These are proportional to the combination
\begin{equation}
2 m_{S}^2+m_{h^0}^2-\lambda_5 v^2
\label{Eq:h0NScoupling}
\end{equation}
where $S$ can be either of $H^0$, $A^0$ or $H^\pm$, depending on which scalar couples to $h^0$. The effect of these new couplings is clearly visible in the numerical results. By comparing the upper and the lower panel of figure~\ref{Fig:drhomixsplit} we see that larger masses of the non-standard scalars yield larger values of $\delrho[H,Mix]{2}$.
In addition the couplings can be enhanced or suppressed by negative or positive values of $\lambda_5$,
which explains the variation between the different solid lines representing different values of $\lambda_5$.
Since the correction $\delrho[H,Mix]{2}$ is independent of $t_\beta$ it will be the dominant scalar two-loop correction to the $\rho$ parameter for $t_\beta\approx 1$ where $\delrho[H,NS]{2}$ is small. However, for $m_{H^0}=m_{H^\pm}$ both the one-loop correction $\Delrho[NS]{1}$ and $\delrho[H,Mix]{2}$ vanish independently of $t_\beta$,
and $\delrho[H,NS]{2}$ is the only
remaining scalar correction to the $\rho$ parameter (for $t_\beta \neq 1$).
For the Inert-Higgs-Doublet-Model (IHDM),
as explained in section~\ref{Sec:IHDM},
the only non-standard two-loop correction to the $\rho$ parameter is equivalent to $\delrho[H,Mix]{2}$.
Conventionally, the parameter $\mu_2^2$ is often used as a free input parameter. The results in
fig.~\ref{Fig:drhomixsplit} can easily be interpreted in the IHDM
by means of the relation~\eqref{Eq:mu2relation} to trade $\lambda_5$ for $\mu_2^2$.
\subsection{Results for a light pseudoscalar}
A light pseudoscalar with $m_{A^0}<125\text{ GeV}$ can still be possible in the THDM (for a detailed analysis see \cite{Bernon:2014nxa}).
The non-standard top-Yukawa contribution is similar to the case discussed in
section~\ref{Sec:NStResults}.
For the scalar contributions, a general feature of a light $A^0$ boson
consists in a large splitting of the two zeros of the dashed line
in figures~\ref{Fig:drhoNSHmasssplit} and~\ref{Fig:drhomixsplit}.
The area around the zero at $m_{H^\pm}=m_{A^0}$ is
excluded by the absence of light charged Higgs bosons. Hence, only the other
zero at $m_{H^\pm}=m_{H^0}$ is phenomenologically acceptable and deserves a closer inspection.
The one-loop contribution $\Delrho[NS]{1}$ and the two-loop contribution $\delrho[H,Mix]{2}$
are both independent of~$t_\beta$; they are displayed in figure~\ref{Fig:drhoMixlightA0} where one can see
that $\delrho[H,Mix]{2}$ follows the direction of $\Delrho[NS]{1}$ and thus amplifies the
dependence on the mass splitting between $H^0$ and $H^\pm$,
disfavoring the case $m_{H^\pm} < m_{H^0}$.
The purely non-standard scalar contribution $\delrho[H,NS]{2}$ vanishes for $t_\beta=1$, but otherwise has a strong
variation with~$t_\beta$ (and $\lambda_5$). It is shown in figure~\ref{Fig:drhoHNSlightA0}, the analogous plot
to figure~\ref{Fig:drhoNSHmasssplit}, now with a light $A^0$. Since the common zero of all curves
corresponds to $m_{A^0}$, the two-loop contribution $\delrho[H,NS]{2}$ is always negative for $m_{H^0, H^\pm} > m_{A^0}$
and thus can diminish $\Delrho[NS]{1}$ substantially for $m_{H^\pm} > m_{H^0}$ when $t_\beta$ increases.
Again, the situation $m_{H^\pm} < m_{H^0}$ is disfavored.
For $m_{A^0}<m_{h^0}/2$, the coupling of $h^0$ to two pseudoscalars
has to be small to suppress the decay channel
$h^0\rightarrow A^0 A^0$ \cite{Bernon:2014nxa}.
In the alignment limit this requires to restrict the value of $\lambda_5$
to $\lambda_5 v^2 \simeq 2m_{A^0}^2+m_{h^0}^2$
(see \eqref{Eq:h0NScoupling}).
Scenarios with a light $A^0$ are especially interesting in the THDM,
since Barr-Zee type two-loop diagrams can provide an explanation for
the $3\sigma$ difference between the SM prediction and the measured
value of the muon anomalous magnetic moment $a_\mu$
\cite{Chang:2000ii}. An improved agreement between theory and
experiment consistent with several theoretical and experimental
constraints can be achieved in a type-X model with very large values of $t_\beta$
(see \cite{Broggio:2014mna,Chun:2016hzs} and references
therein). Usually $m_{H^\pm}=m_{H^0}$ is assumed, to fulfill the constraints
from electroweak precision observables. For the $\rho$ parameter this means
vanishing contributions from $\Delrho[NS]{1}$ and
$\delrho[H,Mix]{2}$. Furthermore, the top-Yukawa contribution
$\delrho[t,NS]{2}$ is strongly suppressed.
However, for such large values of $t_\beta$, the non-standard scalar contribution
$\delrho[H,NS]{2}$ would completely run out of control unless the
scalar self-coupling is kept small by adjusting $\lambda_5$ very close to
$\lambda_5 =2m_{H^0}^2/v^2$.
An additional aspect of type-X models with very large $t_\beta$ is the enhanced
Yukawa coupling of the $\tau$ lepton. This could yield a further
two-loop contribution to the $\rho$ parameter, which we did not consider in this work.
\begin{figure}[ht]
\includegraphics[width=\linewidth]{drhoMixlightA0.pdf}
\caption{
Results for $\delrho[H,Mix]{2}$ for $m_{A^0}=60\text{ GeV}$. The mass of
$H^0$ is fixed at $350\text{ GeV}$.
The different solid lines correspond to different values of $\lambda_5$, and the results are independent of $t_\beta$. The non-standard one-loop correction $\Delrho[NS]{1}$ is shown by the blue dashed line.
The grey area corresponds to the bounds from the experimental limits of the $T$ parameter.
}
\label{Fig:drhoMixlightA0}
\end{figure}
\begin{figure}[ht]
\includegraphics[width=\linewidth]{drhoHNSlightA0.pdf}
\caption{
Results for $\delrho[H,NS]{2}$ for $m_{A^0}=60\text{ GeV}$. The mass of $H^0$ is fixed at $350\text{ GeV}$.
The different solid lines correspond to different values of $t_\beta$. The non-standard one-loop correction $\Delrho[NS]{1}$ is shown by the blue dashed line.
The grey area corresponds to the bounds from the experimental limits of the $T$ parameter.
}
\label{Fig:drhoHNSlightA0}
\end{figure}
\section{Conclusions}
\label{Sec:conclusions}
We have given an overview over the calculation of the two-loop contributions
to the $\rho$-parameter in the $CP$-conserving Two-Higgs-Doublet Model
where one of the $CP$-even scalars ($h^0$) is identified with the scalar resonance
at 125 GeV observed by the LHC experiments ATLAS and CMS.
The approximation of the gauge-less limit and massless fermions except the top quark
yield the leading contributions from the top-Yukawa coupling and the self-couplings of the
Higgs bosons, which can be separated into standard and non-standard contributions.
As already at the one-loop level, the non-standard contributions from the scalar self-interactions
are particularly sensitive to mass splittings between neutral and charged scalars.
As a new feature, the two-loop contributions have a significant dependence on the
parameters $\tan\beta$ and $\lambda_5$, the coefficient of the THDM scalar potential
that is not fixed by the masses of the neutral and charged Higgs bosons, and thus can
modify the one-loop result substantially. Moreover, this significant dependence on
the additional parameters can be exploited to get more indirect information on the
Higgs potential from electroweak precision data than with the
currently available one-loop calculations.
The loop correction $\Delta\rho$ to the $\rho$-parameter is an important entry
in the calculation of electroweak precision observables, parametrizing dominant
universal contributions from particles with mass splitting in isospin doublets,
in the THDM in particular from
neutral and charged Higgs bosons.
For an estimate of the impact of a shift in $\Delta\rho$
on the prediction of the $W$ mass and the effective weak mixing angle
$\sin^2 \theta_{\rm eff}$ at~$M_Z$, one can use the approximate expressions
\begin{eqnarray}
\Delta M_W \simeq \frac{M_W}{2} \, \frac{c_W^2}{c_W^2-s_W^2} \, \Delta\rho, \\
\Delta \sin^2 \theta_{\rm eff} \simeq - \frac{c_W^2 s_W^2}{c_W^2-s_W^2} \, \Delta\rho ,
\end{eqnarray}
to translate the two-loop contribution to $\Delta\rho$ from the non-standard Higgs sector
obtained in this paper into shifts of the observables.
An accurate evaluation of the precision observables and implications from comparisons with
experimental data requires a more detailed study, which will be presented in a forthcoming publication.
\begin{acknowledgements}
This work was supported in part by the Deutsche Forschungsgemeinschaft (DFG)
under Grant No.~EXC-153
(Excellence Cluster {\it Structure and Origin of the Universe}).
We thank Georg Weiglein for useful discussions and Thomas Hahn and Sebastian Pa\ss ehr for
their helpful support in the installation and handling of the two-loop calculational tools.
\end{acknowledgements}
\section{Introduction}
High-precision experiments at electron-positron and hadron colliders
together with highly accurate measurements at low energies
have imposed
stringent tests on the Standard Model (SM) and possible extensions.
The experimental accuracy
in the electroweak observables is sensitive to the quantum effects
and requires the highest standards
on the theoretical side as well.
A sizeable amount of theoretical work has
contributed over more than two decades to a steadily rising
improvement of the SM predictions and also for
specific new physics scenarios like supersymmetric extensions.
The highly accurate measurements and theoretical predictions,
at the level of 0.1\% precision and better, provide unique tests of
the quantum structure of the SM, which has been impressively
confirmed by the discovery of a Higgs particle
by ATLAS~\cite{Aad:2012tfa} and CMS~\cite{Chatrchyan:2012ufa}.
Moreover, it opens the possibility to obtain indirect informations
on potential heavy new physics beyond the SM,
in particular on the not yet sufficiently explored scalar sector.
With the meanwhile very precisely measured Higgs-boson
mass~\cite{Aad:2015zhl} of
$M_H = 125.09 \pm 0.24 \; {\rm GeV}$
the SM input is now completely determined
and the SM predictions for the set of precision observables are unique,
being in overall good agreement with the data.
This improves the sensitivity to physics beyond the SM and
makes constraints on parameters of extended models quite severe.
Models with two scalar doublets in the Higgs sector are among the simplest extensions of
the Standard Model (a review on theory and phenomenology can be found in~\cite{Branco:2011iw}).
They fulfill the relation $\rho = 1$ at lowest order for the $\rho$ parameter as favored by experimental data
for electroweak precision observables allowing only small deviations from unity.
Such small deviations $\Delta\rho$ naturally originate exclusively from quantum effects in models with Higgs
doublets, with special sensitivity on mass splittings between different isospin components of fermions and scalars.
$\Delta\rho$ can be related to the vector-boson self-energies
and plays the most prominent role in the higher-order calculation
of precision observables, constituting the leading process independent
loop corrections to accurately measured quantities
like the $W$-$Z$ mass correlation and the effective weak mixing angle
$\sin^2\theta_{\rm eff}$.
The calculation of electroweak precision observables in the general THDM has a long
history ~\cite{Bertolini:1985ia,Hollik:1986gg,Hollik:1987fg,Denner:1991ie,Froggatt:1991qw,Chankowski:1999ta,Grimus:2007if,Grimus:2008nb,LopezVal:2012zb}.
Details of the one-loop renormalization of the THDM have been discussed in various
papers~\cite{Santos:1996vt,Arhrib:2003ph,LopezVal:2009qy,Kanemura:2014bqa,Kanemura:2015mxa,Krause:2016oke,Denner:2016etu},
with the emphasis of the more recent ones on the Higgs sector aiming
at loop-improved predictions for Higgs-boson observables.
The current status of precision observables is given by complete one-loop calculations which can
be augmented in the subset of the SM loop contributions by
incorporating the known higher-order terms of the SM; the non-standard contribution
is of one-loop order and systematic two-loop calculations have not
been done (ref.~\cite{LopezVal:2012zb} contains some higher-order
terms by means of effective couplings in the one-loop Higgs contributions).
This is different from the supersymmetric version of the THDM, the MSSM,
where the non-standard one-loop corrections to precision observables
have been improved by the two-loop contributions to $\Delta\rho$
resulting from the strong and Yukawa interactions
\cite{Djouadi:1996pa,Djouadi:1998sq,Heinemeyer:2002jq,Haestier:2005ja}.
In order to achieve a similar quality of the theoretical predictions also in the general THDM,
the first step consists in getting the two-loop contributions to $\Delta\rho$
from those sectors where the one-loop effects are large, i.e.\ from
the top-Yukawa interaction and the self-interaction of the extended Higgs scalars.
In this paper we present the leading two-loop corrections to $\Delta\rho$
in the $CP$-conserving THDM which result from the top-Yukawa coupling and
the self-couplings of the Higgs bosons. Technically they are obtained
in the approximation of the gauge-less limit where the electroweak gauge
couplings are set to zero (and thus the gauge-boson masses, but keeping $M_W/M_Z$ fixed).
With the assumption that one
of the $CP$-even neutral scalars represents the scalar boson observed
by the LHC experiments, with SM properties, the two-loop non-standard
contributions in $\Delta\rho$ can be separated from the SM ones.
These contributions are of particular interest since they involve corrections
proportional to $m_t^4$, or increase with mass splittings between
non-standard Higgs bosons and can be additionally enhanced by $\tan\beta$
and $\lambda_5$, an additional free coefficient of the Higgs potential.
The paper is organized as follows. Section~\ref{Sec:THDM}
contains the basic features
of the THDM and specifies the notations, and section~\ref{Sec:Outline}
describes the simplifications made for our two-loop calculation.
Aspects of custodial symmetry in the context of the THDM
are considered in section~\ref{Sec:CustSym} which provide a deeper
understanding of the various higher-order contributions to $\Delta\rho$.
The renormalization scheme is specified in section~\ref{Sec:Renormalization}, and
the calculation of $\Delta\rho$ is described in section~\ref{Sec:rhoparameter}.
The appendix contains the Feynman rules for the counterterm vertices and the definitions of the one- and two-loop scalar integrals.
The numerical analysis is the content of section~\ref{Sec:Results},
and conclusions are given in section~\ref{Sec:conclusions}.
\input{THDM}
\input{Outline}
\input{CustodialSymmetrie}
\input{Renormalization}
\input{rhoparameter}
\input{Results}
\begin{appendix}
\input{Feynmanrules}
\input{Integrals}
\end{appendix}
\bibliographystyle{spphys}
\section{\boldmath Corrections to the \texorpdfstring{$\rho$}{rho} parameter}
\label{Sec:rhoparameter}
The $\rho$ parameter
\begin{equation}
\rho=\frac{G_{NC}}{G_{CC}}
\label{Eq:rhodefinition}
\end{equation}
was originally introduced \cite{Ross:1975fq} for four-fermion processes at low momentum
as the strength $G_{NC}$ of the effective neutral current coupling normalized by the charged current coupling $G_{CC}$.
In the electroweak theory both classes of processes are mediated by the exchange of a heavy gauge boson,
the $Z$ boson for NC and the $W^\pm$ boson for CC processes.
In the effective theory for low momentum transfer we can approximate
the propagators by $1/M_V^2$ ($V=W,Z$).
Therefore the effective couplings at the tree level are given by
\begin{align}
\frac{G_{NC}}{\sqrt{2}}&=\frac{e^2}{8s_W^2c_W^2M_Z^2},\\
\frac{G_{CC}}{\sqrt{2}}&=\frac{e^2}{8s_W^2M_W^2},
\end{align}
which results in
\begin{equation}
\rho=\frac{M_W^2}{c_W^2 M_Z^2}=1.
\end{equation}
Including higher-order processes in the calculation of the effective couplings $G_{NC}$ and $G_{CC}$ leads to a deviation $\Delta \rho$ from unity,
\begin{equation}
\label{Eq:rhoparameter}
\rho=\frac{1}{1-\Delta\rho} ,
\end{equation}
where
\begin{equation}
\label{Eq:delrhoperturbative}
\Delta\rho = \Delta\rho^{(1)} + \Delta\rho^{(2)} + \cdots
\end{equation}
can be calculated in the loop-order expansion. Although conceptually defined at low-momentum scales,
the quantity $\Delta\rho$ represents an important ingredient for electroweak precision observables
as the leading universal correction, with a substantial impact e.g.\ on the effective electroweak mixing angle and the $W$ mass.
Vertex and box-diagram corrections to charged and neutral current
processes do not contribute in the gauge-less limit
and for vanishing masses of the external fermions, as well as $\gamma$-$Z$ mixing in the neutral current interaction.
Consequently, only corrections from the gauge-boson self-energies arise, of the form
\begin{equation}
\frac{\Self[V]{}{0}}{M_V^2}\quad(V=W,Z).
\label{Eq:Selfratio}
\end{equation}
Due to a Ward identity in the gauge-less limit \cite{Barbieri:1992dq,Fleischer:1994cb} these quantities can be calculated also by the relations
\begin{align}
\label{Eq:ZWardId}
\frac{\Self[Z]{}{0}}{M_Z^2} & = -\Self[G^0]{\prime}{0}, \qquad
\frac{\Self[W]{}{0}}{M_W^2} = -\Self[G^\pm]{\prime}{0},
\end{align}
where the Goldstone self-energies are decomposed according to
\begin{equation}
\Self[G]{}{p^2}=\Self[G]{}{0}+p^2 \Self[G]{\prime}{p^2}, \quad (G=G^0,G^\pm).
\end{equation}
We use this Ward identity as a test for our result. Moreover, the origin of a specific contribution in $\Delta\rho$ is not always directly visible in the calculation based on the gauge-boson self-energies due to the cancellation of the gauge couplings in the ratio (\ref{Eq:Selfratio}). In these cases, the couplings involved can be identified with the help of the Ward identity.
\subsection{One-loop corrections in the SM and the THDM}
\begin{figure*}
\begin{center}
\includegraphics{ZWNS1Loop.pdf}
\caption{Non-standard contributions from the THDM scalars to the $Z$ and $W$ boson self-energies in the alignment limit at the one-loop level.}
\label{Fig:ZW_Selfenergy_NS_1Loop}
\end{center}
\end{figure*}
The calculation of the effective coupling strengths at the one-loop order in the gauge-less limit results in
\begin{equation}
\frac{G_{NC}}{\sqrt{2}}=\frac{e_0^2}{8s_{W,0}^2c_{W,0}^2M_{Z,0}^2}\left[1+\frac{\Self[Z]{(1)}{0}}{M_Z^2}\right]
\end{equation}
and
\begin{equation}
\frac{G_{CC}}{\sqrt{2}}=\frac{e_0^2}{8s_{W,0}^2M_{W,0}^2}\left[1+\frac{\Self[W]{(1)}{0}}{M_W^2}\right].
\end{equation}
The expansion of the bare parameters cancel in the $\rho$ parameter. Therefore, the correction to the $\rho$ parameter at one-loop order is given by
\begin{equation}
\Delta \rho^{(1)}=\frac{\Self[Z]{(1)}{0}}{M_Z^2}-\frac{\Self[W]{(1)}{0}}{M_W^2}.
\label{Eq:drho1Loopdef}
\end{equation}
In the THDM with the assumptions described in section~\ref{Sec:Outline} we can split the one-loop correction
\begin{equation}
\Delta \rho^{(1)}=\Delrho[t]{1}+\Delrho[NS]{1}
\label{Eq:drho1Loopsplit}
\end{equation}
into two independent parts originating from the top-Yukawa coupling and the scalar sector. The first part arises from the large mass splitting between the top and the bottom quark and is identical to the dominant part of the one-loop corrections to $\Delta\rho$ in the SM \cite{Veltman:1977kh,Chanowitz:1978mv,Chanowitz:1978uj}. When neglecting the mass of the bottom quark one obtains the one-loop result
\begin{equation}
\Delrho[t]{1}=\frac{3\alpha_\text{em}}{16\pi M_W^2s_W^2}m_t^2.
\label{Eq:drhoSM1Loop}
\end{equation}
In the SM no contributions to $\Delta\rho^{(1)}$ arise from scalar
loops due to the custodial symmetry of the Higgs potential. In a similar way there is no correction from the SM-like scalars $h^0$,
$G^0$ and $G^\pm$ in the THDM with the assumptions from section~\ref{Sec:Outline}, since the part $V_{I}$ of the potential in
\eqref{Eq:PotAlign} is custodial invariant. The contributions from the
SM-like scalars to the gauge-boson self-energies yield the SM result
in dimensional regularization with dimension $D$, expressed in terms
of $A_0$ in \ref{App:1LIntegrals},
\begin{equation}
\frac{\Self[V,\text{SM}]{(1)}{0}}{M_V^2}=\frac{\alpha_\text{em}}{16 \pi s_W^2 M_W^2}\frac{(D-4) }{ D }A_0\left(m_{h^0}^2\right),
\label{Eq:VVSM1Loop}
\end{equation}
for both $V=W,Z$.
They cancel in the difference for $\Delta\rho^{(1)}$ in \eqref{Eq:drho1Loopdef}.
However, the extended scalar sector of the THDM gives additional
scalar contributions to $\Delta\rho$
\cite{Bertolini:1985ia,Hollik:1986gg,Hollik:1987fg,Denner:1991ie,Grimus:2007if,gunion:1990}.
In the alignment limit the additional correction follows from the scalars $H^0$, $A^0$ and $H^\pm$. The gauge-boson self-energies from the diagrams in figure~\ref{Fig:ZW_Selfenergy_NS_1Loop} give rise to the non-standard one-loop part
\begin{align}
\Delrho[NS]{1}=&\frac{\alpha_\text{em}}{16 \pi s_W^2 M_W^2 D }\Big\{4 m_{A^0}^2 B_0\left(0,m_{A^0}^2,m_{H^\pm}^2\right)\notag\\
&+4 m_{H^0}^2 B_0\left(0,m_{H^0}^2,m_{H^\pm}^2\right)\notag\\
&-4 m_{A^0}^2 B_0\left(0,m_{A^0}^2,m_{H^0}^2\right)\notag\\
&+(8-2 D) A_0\left(m_{H^\pm}^2\right)-4A_0\left(m_{H^0}^2\right)\Big\}
\label{Eq:drhoNS1L_arbD}
\end{align}
which simplifies to
\begin{align}
\Delrho[NS]{1} \xrightarrow{D\rightarrow4}&\frac{\alpha_\text{em}}{16 \pi s_W^2 M_W^2}\Bigg\{
\frac{m_{A^0}^2 m_{H^0}^2 }{m_{A^0}^2-m_{H^0}^2}\log \left(\frac{m_{A^0}^2}{m_{H^0}^2}\right)\notag\\
&-\frac{m_{A^0}^2 m_{H^{\pm }}^2 }{m_{A^0}^2-m_{H^{\pm}}^2}\log \left(\frac{m_{A^0}^2}{m_{H^{\pm }}^2}\right)\notag\\
&-\frac{m_{H^0}^2 m_{H^{\pm }}^2 }{m_{H^0}^2-m_{H^{\pm }}^2}\log \left(\frac{m_{H^0}^2}{m_{H^{\pm }}^2}\right)\notag\\
&+m_{H^{\pm }}^2\Bigg\}.
\label{Eq:drhoNS1Loop}
\end{align}
in four dimensions.
It increases quadratically with the mass difference between the charged and the neutral Higgs states, and it vanishes for
\begin{equation}
m_{H^0}=m_{H^\pm}
\end{equation}
or
\begin{equation}
m_{A^0}=m_{H^\pm}.
\end{equation}
The reason is that this correction contains only couplings between the Goldstone bosons and the non-standard scalars $H^0$, $A^0$ and $H^\pm$ which are determined by the part $V_{III}$ of the potential. As explained in section~\ref{Sec:CustSym}, the custodial symmetry in this part can be restored for equal charged and neutral Higgs masses.
Note that in the alignment case the entire non-standard one-loop
contribution to $\Delta\rho$ is exclusively given by
the expression~(\ref{Eq:drhoNS1Loop}),
corresponding to the gauge-less limit.
\subsection{Higher-order corrections in the THDM}
As mentioned above, it is sufficient to
keep only the corrections from the gauge-boson self-energies in the calculation of the effective neutral and charged current interaction of the four fermion processes.
The two-loop results of the effective couplings are
\begin{align}
\frac{G_{NC}}{\sqrt{2}}
=&\frac{e_0^2}{8s_{W,0}^2c_{W,0}^2M_{Z,0}^2}\left[1+\frac{\Self[Z]{(1)}{0}}{M_{Z}^2}-\frac{\delta M_Z^2}{M_Z^2}\frac{\Self[Z]{(1)}{0}}{M_{Z}^2}\right.\notag\\
&\left.+\left(\frac{\Self[Z]{(1)}{0}}{M_{Z}^2}\right)^2+\frac{\Self[Z]{(2)}{0}}{M_Z^2}\right]
\end{align}
and
\begin{align}
\frac{G_{CC}}{\sqrt{2}}
=&\frac{e_0^2}{8s_{W,0}^2M_{W,0}^2}\left[1+\frac{\Self[W]{(1)}{0}}{M_{W}^2}-\frac{\delta M_W^2}{M_W^2}\frac{\Self[W]{(1)}{0}}{M_{W}^2}\right.\notag\\
&\left.+\left(\frac{\Self[W]{(1)}{0}}{M_{W}^2}\right)^2+\frac{\Self[W]{(2)}{0}}{M_W^2}\right].
\end{align}
With the renormalization condition \eqref{Eq:dMVgless1L} for the gauge-boson mass
counterterms in the gauge-less limit the
products of one-loop corrections in the brackets cancel.
The calculation of $\rho$ as defined by \eqref{Eq:rhodefinition}
then yields the deviation $\Delta\rho$ in \eqref{Eq:rhoparameter} as follows,
\begin{align}
\label{Eq:delroZW}
\Delta \rho=& \left(\frac{\Self[Z]{(1)}{0}}{M_Z^2}-\frac{\Self[W]{(1)}{0}}{M_W^2}\right)\notag\\
&-\frac{\Self[Z]{(1)}{0}}{M_Z^2}\left(\frac{\Self[Z]{(1)}{0}}{M_Z^2}-\frac{\Self[W]{(1)}{0}}{M_W^2}\right)\notag\\
&+\left(\frac{\Self[Z]{(2)}{0}}{M_Z^2}-\frac{\Self[W]{(2)}{0}}{M_W^2}\right)\\
=&\Delta\rho^{(1)}+\Delta\rho^{(2)},
\end{align}
where the two-loop part is given by
\begin{equation}
\Delta\rho^{(2)}\equiv-\frac{\Self[Z]{(1)}{0}}{M_Z^2}\Delrho{1}+\left(\frac{\Self[Z]{(2)}{0}}{M_Z^2}-\frac{\Self[W]{(2)}{0}}{M_W^2}\right) .
\label{Eq:Deltarho2}
\end{equation}
$\Delta\rho^{(1)}$ summarizes the one-loop corrections as given by
\eqref{Eq:drho1Loopdef} and \eqref{Eq:drho1Loopsplit}. The self-energy of the $Z$
boson in the first term consists of all the corrections from the
top quark and the scalars as internal particles. Note that it contains
also the part from the SM-like scalars in \eqref{Eq:VVSM1Loop}, which
cancel in $\Delrho{1}$. The second part of \eqref{Eq:Deltarho2} follows
from the two-loop corrections to the gauge-boson self-energies. In
addition to the part from the genuine two-loop diagrams (labeled as $\delrho{2Loop}$)
it also includes one-loop diagrams with counterterm insertions for the subloop
renormalization (labeled as $\Delrho{CT}$).
With the assumptions from section~\ref{Sec:Outline} we have two sources for the two-loop contribution $\Delrho{2}$: the top-Yukawa interaction and the scalar self-interaction.
Due to the alignment limit we can subdivide the top-Yukawa corrections into two parts. The first one is identical to the two-loop top-Yukawa contribution in the SM and is discussed in section~\ref{Subsec:drhoSMt}. The second one originates from the coupling between the top quark and the non-standard scalars $H^0$, $A^0$ and $H^\pm$ and is described in more detail in section~\ref{Subsec:drhoNSt}.
A similar separation can be made for the additional corrections to the $\rho$ parameter from the scalar self-interaction. The part $V_I$ of the potential (see \eqref{Eq:V_I}), which describes only the interaction between $h^0$ and the Goldstone bosons $G^0$, $G^\pm$, is invariant under the custodial symmetry and the corresponding contributions to the vector-boson self-energies in $\Delta\rho$ cancel each other. The remaining part of the potential gives rise to two finite subsets in $\Delrho{2}$. One follows from the interaction between the SM-like scalars $h^0$, $G^0$, $G^\pm$ and the non-standard scalars $H^0$, $A^0$, $H^\pm$ and is discussed in section~\ref{Subsec:drhoMix}. The other one contains only the non-standard scalars $H^0$, $A^0$ and $H^\pm$ as internal particles in the gauge-boson self-energies and is described in section~\ref{Subsec:drhoNSHiggs}.
With this categorization we subdivide the contribution from the
genuine two-loop diagrams (without subloop renormalization) to the
vector-boson self-energies into different parts, according to their origin,
\begin{multline}
\delrho{2Loop}=\delrho[t,SM]{2Loop}+\delrho[t,NS]{2Loop}\\
+\delrho[H,NS]{2Loop}+\delrho[H,Mix]{2Loop}
\label{Eq:dr2Loopsplit}
\end{multline}
which are classified by the participating couplings:
\begin{itemize}
\item $\delrho[t,SM]{2Loop}$ originates from the coupling between the top quark and the SM-like scalars $h^0$, $G^0$ and $G^\pm$ (see section~\ref{Subsec:drhoSMt});
\item $\delrho[t,NS]{2Loop}$ is the part which follows from the top-Yukawa interaction of the non-standard scalars $H^0$, $A^0$ and $H^\pm$ (see section~\ref{Subsec:drhoNSt});
\item $\delrho[H,NS]{2Loop}$ contains the scalar self-coupling between the non-standard scalars (see section~\ref{Subsec:drhoNSHiggs});
\item $\delrho[H,Mix]{2Loop}$ follows from the interaction between the SM-like scalars and the non-standard scalars (see section~\ref{Subsec:drhoMix}).
\end{itemize}
\begin{figure}[ht]
\begin{center}
\includegraphics[width=0.45\textwidth]{VVtCTdiag.pdf}
\caption{Generic diagrams for the gauge-boson self-energies containing quarks with counterterm insertions. $V=\{W,Z\}$; $f,f^\prime=\{t,b\}$.}
\label{Fig:VV_t_CTDiagramme}
\end{center}
\end{figure}
\begin{figure*}[ht]
\begin{center}
\includegraphics{VVNSCTdiag.pdf}
\caption{Generic diagrams for the gauge-boson self-energies $V=\{W,Z\}$ containing scalars with counterterm insertions.
The contribution from the diagrams can be divided into two parts: one part with only the SM-like scalars ($S,S^\prime=\{h^0,G^0,G^\pm\}$)
and one part with only the non-standard scalars ($S,S^\prime=\{H^0, A^0, H^\pm\}$).}
\label{Fig:VV_NS_CTDiagramme}
\end{center}
\end{figure*}
For one-loop subrenormalization we need the diagrams shown in figure~\ref{Fig:VV_t_CTDiagramme} for the
self-energies with the top quarks and in figure~\ref{Fig:VV_NS_CTDiagramme} for the scalar contribution. In the gauge-less limit only two types of
renormalization constants survive: the counterterm $\delta s_W^2$ from the counterterm insertions in the vertices, and the mass counterterms in
the propagators of the internal particles. All field counterterms of the internal particles drop out in the calculation,
and all other counterterms are zero in the gauge-less limit.
From the diagrams of figure~\ref{Fig:VV_t_CTDiagramme} we obtain the part of
the subloop renormalization from the top quark. The renormalization
of the weak mixing angle is contained in the vertex counterterms
(see \ref{App:Feynmanrules}) and yields the term
\begin{equation}
\frac{s_W^2}{c_W^2}\frac{\delta s_W^2}{s_W^2}\frac{\Self[Z,\text{t}]{(1)}{0}}{M_Z^2}-\frac{\delta s_W^2}{s_W^2}\Delrho[t]{1} .
\label{Eq:VertexCTtop}
\end{equation}
From the diagrams with counterterms in the propagators in figure~\ref{Fig:VV_t_CTDiagramme} we obtain the term
\begin{equation}
\delrho[t]{CT}=-\frac{3 \alpha_\text{em} (D-4) (D-2)^2 A_0\left(m_t^2\right)}{16 \pi D M_W^2 s_W^2}\frac{\delta m_t}{m_t}.
\label{Eq:drhotCT}
\end{equation}
Due to the alignment limit we can split the result of the top mass counterterm into a SM-like and a non-standard part. We use this for the seperation
\begin{equation}
\delrho[t]{CT}=\delrho[t,SM]{CT}+\delrho[t,NS]{CT} \, ,
\end{equation}
where the two parts are defined as follows:
\begin{itemize}
\item the part $\delrho[t,SM]{CT}$ contains the correction to the top-mass counterterm from the SM-like scalars $h^0$, $G^0$, $G^\pm$ as shown in the self-energy diagrams in figure~\ref{Fig:topSMCT};
\item the second part $\delrho[t,NS]{CT}$ contains the part of $\delta m_t$ which comes from the top quark self-energy corrections from the non-standard scalars as depicted in figure~\ref{Fig:topNSCT}.
\end{itemize}
For the subloop renormalization diagrams in figure~\ref{Fig:VV_NS_CTDiagramme} with the SM-like scalars $h^0$, $G^0$ and $G^\pm$ we find that the mass counterterms drop out in the difference of the $W$ and $Z$ self-energy, due to custodial symmetry. From the vertex counterterms we obtain the contribution
\begin{equation}
\frac{s_W^2}{c_W^2}\frac{\delta s_W^2}{s_W^2}\frac{\Self[Z,\text{SM}]{(1)}{0}}{M_Z^2}
\label{Eq:VertexCTSM}
\end{equation}
with the one-loop self-energy from \eqref{Eq:VVSM1Loop}.
The diagrams in figure~\ref{Fig:VV_NS_CTDiagramme} with the possible
insertions of the non-standard scalars for $S$ and $S^\prime$ give the
last part of the subloop renormalization. With the Feynman rules of
\ref{App:Feynmanrules} the counterterms in the vertices yield the contribution
\begin{equation}
\frac{\Self[Z,\text{NS}]{(1)}{0}}{M_Z^2}\frac{s_W^2}{c_W^2}\frac{\delta s_W^2}{s_W^2}-\frac{\delta s_W^2}{s_W^2}\Delrho[NS]{1},
\label{Eq:VertexCTNS}
\end{equation}
where the $Z$ self-energy in the first term contains just the contribution of the non-standard scalars.
The correction from the mass counterterms $\delta m_{H^0}^2$, $\delta m_{A^0}^2$ and $\delta m_{H^\pm}^2$ in the diagrams in figure~\ref{Fig:VV_NS_CTDiagramme} is denoted by $\delrho[H]{CT}$. It is identical to
\begin{equation}
\begin{aligned}
\delrho[H]{CT}=\Big(&\delta m_{A^0}^2\frac{\partial}{\partial m_{A^0}^2}+\delta m_{H^0}^2\frac{\partial}{\partial m_{H^0}^2}\\
&+\delta m_{H^\pm}^2\frac{\partial}{\partial m_{H^\pm}}\Big)\Delrho[1]{NS},
\label{Eq:drhoNSCT}
\end{aligned}
\end{equation}
with the one-loop contribution from \eqref{Eq:drhoNS1L_arbD}.
By splitting up the mass counterterms we will classify three different parts
\begin{equation}
\delrho[H]{CT}=\delrho[H,t]{CT}+\delrho[H,NS]{CT}+\delrho[H,Mix]{CT} \, ,
\label{Eq:drhoHCT}
\end{equation}
which are defined as follows:
\begin{itemize}
\item $\delrho[H,t]{CT}$ contains the non-standard scalar mass
counterterms originating from the top-Yukawa coupling. The corresponding diagrams are shown in figure~\ref{Fig:SStopCT}.
\item $\delrho[H,NS]{CT}$ labels the part which contains only
non-standard scalars in the calculation of $\delta m_{H^0}^2$, $\delta m_{A^0}^2$ and $\delta m_{H^\pm}^2$.
The diagrams are displayed in figure~\ref{Fig:SSNSCTs}.
\item $\delrho[H,Mix]{CT}$ incorporates the contribution to the mass counterterms of $H^0$, $A^0$ and $H^\pm$ which originates from the couplings of the non-standard scalars to the SM-like scalars. The corresponding self-energy diagrams are presented in figure~\ref{Fig:SSMixCTs}.
\end{itemize}
When we combine the various parts from the subloop renormalization,
their overall contribution to $\Delrho{2}$ can be written as follows:
\begin{equation}
\begin{aligned}
\Delrho{CT}=&\frac{s_W^2}{c_W^2}\frac{\delta s_W^2}{s_W^2}\frac{\Self[Z]{(1)}{0}}{M_Z^2}\\
&-\frac{\delta s_W^2}{s_W^2}\left(\Delrho[t]{1}+\Delrho[NS]{1}\right)+\delrho{CT}.
\end{aligned}
\label{Eq:DeltarhoCT}
\end{equation}
The first term incorporates all parts from \eqref{Eq:VertexCTtop}, \eqref{Eq:VertexCTSM} and \eqref{Eq:VertexCTNS}
involving a single $Z$-boson self-energy; the remaining terms from the renormalization of $s_W$ in
\eqref{Eq:VertexCTtop} and \eqref{Eq:VertexCTNS} are kept separately in the second term. The last term
\begin{equation}
\delrho{CT}=\delrho[t]{CT}+\delrho[H]{CT}
\label{Eq:drhoCT}
\end{equation}
collects the various parts resulting from the mass counterterms of the internal particles.
The two-loop correction to the $\rho$ parameter in \eqref{Eq:Deltarho2} can be further simplified,
since the counterterm of the weak mixing angle reduces to
\begin{equation}
\frac{\delta s_W^2}{s_W^2}=\frac{c_W^2}{s_W^2}\left(\frac{\Self[Z]{(1)}{0}}{M_Z^2}-\frac{\Self[W]{(1)}{0}}{M_W^2}\right)=\frac{c_W^2}{s_W^2}\Delrho{1}{}.
\label{Eq:dSWgless}
\end{equation}
in the gauge-less limit (see \eqref{Eq:dMVgless1L}).
Combined with \eqref{Eq:DeltarhoCT} the first term in \eqref{Eq:Deltarho2} is canceled and we obtain
\begin{equation}
\Delta\rho^{(2)}=-\frac{c_W^2}{s_W^2}\left(\Delrho{1}\right)^2+\delrho{2} .
\label{Eq:withreduciblepart}
\end{equation}
In this notation, the genuine two-loop part
\begin{equation}
\label{Eq:delro2genuine}
\delrho{2}=\delrho{CT}+\delrho{2Loop}
\end{equation}
contains $\delrho{CT}$ resulting exclusively from the insertions of the mass counterterms,
and the contribution $\delrho{2Loop}$ from the pure two-loop diagrams
for the $Z,W$ self-energies (without subloop renormalization) in \eqref{Eq:Deltarho2}.
The appearance of the reducible term $\left(\Delrho{1}\right)^2$ in $\Delrho{2}$ is a consequence of the parameterisation of $v^2$ by
\begin{equation}
\frac{1}{v^2}=\frac{e^2}{4 s_W^2 M_W^2}
\end{equation}
together with the on-shell renormalization of $s_W$. A different parameterisation in terms of the Fermi constant~$G_F$
can be introduced with the help of the relation
\begin{equation}
\sqrt{2} G_F=\frac{e^2}{4 M_W^2 s_W^2}\left(1+\Delta r\right) ,
\end{equation}
where the quantity $\Delta r$ describes the higher-order corrections. In the gauge-less limit the one-loop contribution is given by
\begin{equation}
\Delta r =-\frac{\delta s_W^2}{s_W^2}.
\end{equation}
Consequently, the reparameterisation of the one-loop result $\Delrho{1}$ in terms of $G_F$ induces a two-loop shift
originating from $\Delta r$, which effectively cancels the reducible
term in $\Delrho{2}$ in~\eqref{Eq:withreduciblepart}.
Hence, in the $G_F$ expansion, the two-loop contribution in $\Delta\rho$
is identified as the irreducible two-loop part $\delta\rho^{(2)}$ in~\eqref{Eq:delro2genuine}.
In this way, the same pattern for $\rho$ is found
as in the SM \cite{Consoli:1989fg}.
The structure of the irreducible quantity $\delta\rho^{(2)}$ in \eqref{Eq:delro2genuine}
with $\delta\rho^{({\rm 2Loop})}$ defined in \eqref{Eq:dr2Loopsplit}
allows us to divide it into four finite subsets of different origins,
\begin{equation}
\label{Eq:aufteilung}
\delta\rho^{(2)} = \delta\rho^{(2)}_{\rm t, SM} + \delta\rho^{(2)}_{\rm t, NS}
+ \delta\rho^{(2)}_{\rm H,NS} + \delta\rho^{(2)}_{\rm H, Mix} \, ,
\end{equation}
which we describe now in more detail.
\begin{figure*}[ht]
\begin{center}
\includegraphics{NStGeneric.pdf}
\caption{Generic two-loop diagrams for the top-Yukawa corrections to the vector-boson self-energies with $V=\{W,Z\}$ and $f=\{t,b\}$. The standard contribution $\delrho[t,SM]{2Loop}$ follows from $S,S^\prime=\{h^0, G^0, G^\pm\}$. The non-standard contribution $\delrho[t,NS]{2Loop}$ is obtained by all possible insertions of $S,S^\prime=\{H^0, A^0, H^\pm\}$.}
\label{Fig:NStop2Loop}
\end{center}
\end{figure*}
\subsubsection{Standard model corrections from the top-Yukawa coupling}
\label{Subsec:drhoSMt}
\begin{figure}[ht]
\begin{center}
\includegraphics{ttSM1Loop.pdf}
\caption{One-loop diagrams for the standard contribution to the top-mass counterterm $\delta m_t$.}
\label{Fig:topSMCT}
\end{center}
\end{figure}
The first contribution under investigation are the two-loop corrections from the top-Yukawa coupling. In the alignment limit this corrections can be split into two independent subsets. From the coupling of the top quark to the SM-like scalars $h^0$, $G^0$ and $G^\pm$ we obtain the finite correction
\begin{equation}
\delrho[t,SM]{2}=\delrho[t,SM]{CT}+\delrho[t,SM]{2Loop}.
\end{equation}
$\delrho[t,SM]{2Loop}$ are the pure two-loop contributions, which are depicted by the generic diagrams in figure~\ref{Fig:NStop2Loop} for $S,S^\prime={h^0,G^0, G^\pm}$. Its divergences are cancelled by the part $\delrho[t,SM]{CT}$ which is the part of \eqref{Eq:drhotCT} with the top-mass counterterm calculated from the diagrams in figure~\ref{Fig:topSMCT}. $\delrho[t,SM]{2}$ is identical to the already known SM contribution from the top-Yukawa interaction. First the result was calculated in the approximation $M_{H}=0$ \cite{vanderBij:1986hy} and as an expansion for large values of $M_{H}$ \cite{vanderBij:1983bw}. Later the full result for arbitrary Higgs masses was obtained \cite{Barbieri:1992dq,Fleischer:1993ub,Fleischer:1994cb}. We checked that our calculation leads to the same result.
\subsubsection{Non-standard corrections from the top-Yukawa coupling}
\label{Subsec:drhoNSt}
\begin{figure}[t]
\begin{center}
\includegraphics{ttNS1Loop.pdf}
\caption{One-loop diagrams for the non-standard contribution to the top-mass counterterm $\delta m_t$.}
\label{Fig:topNSCT}
\end{center}
\end{figure}
\begin{figure}[t]
\begin{center}
\includegraphics{SStopLoop.pdf}
\caption{One-loop diagrams for the top-Yukawa contribution to the non-standard scalar mass counterterms.}
\label{Fig:SStopCT}
\end{center}
\end{figure}
\noindent
More interesting is the additional contribution due to the coupling of the top quark to the non-standard scalars $H^0$, $A^0$ and $H^\pm$, which is given by
\begin{equation}
\delrho[t,NS]{2}=\delrho[t,NS]{CT}+\delrho[H,t]{CT}+\delrho[t,NS]{2Loop}.
\end{equation}
$\delrho[t,NS]{2Loop}$ denotes the pure two-loop part, represented by
the generic diagrams shown in figure~\ref{Fig:NStop2Loop} with
$S,S^\prime=\{H^0,A^0,H^\pm\}$. The result does not only consist of terms
of $\order{\alpha_t^2}$, which originate only from the top-Yukawa
interaction, but also of contributions of $\order{\alpha_t \lambda_i}$ which contain the scalar self-couplings in addition to the top-Yukawa coupling. The divergences from the $\order{\alpha_t^2}$ part are canceled by $\delrho[t,NS]{CT}$ which originates from the subloop renormalization diagrams of figure~\ref{Fig:VV_t_CTDiagramme} with the top-mass counterterm calculated from the diagrams in figure~\ref{Fig:topNSCT}. The divergences of $\order{\alpha_t\lambda_i}$ are cancelled by $\delrho[H,t]{CT}$ with the mass counterterms calculated from the diagrams in figure~\ref{Fig:SStopCT}.
In the calculation by means of the gauge-boson self-energies the
separation between the $\order{\alpha_t^2}$ and the $\order{\alpha_t \lambda_i}$ contributions is obscured.
Using the Ward identity in \eqref{Eq:ZWardId} can help to disentangle the
two different finite contributions of $\order{\alpha_t^2}$ and $\order{\alpha_t\lambda_i}$.
\subsubsection{Scalar corrections from the interaction of the non-standard scalars}
\label{Subsec:drhoNSHiggs}
\begin{figure}[ht]
\begin{center}
\includegraphics{VVNS2Loop.pdf}
\caption{Generic two-loop diagrams for the vector-boson self-energies from the interaction from the non-standard scalars. $V=\{W,Z\}$; $S=\{H^0, A^0, H^\pm\}$.}
\label{Fig:NSHiggs2Loop}
\end{center}
\end{figure}
\begin{figure}[ht]
\begin{center}
\includegraphics{SSNS1Loop.pdf}
\caption{One-loop diagrams for the non-standard scalar mass counterterms from the interaction between the non-standard scalars.
For the $H^0$ self-energy: $S=S^\prime=H^0,A^0,H^\pm$. For the $A^0$ self-energy: $S=A^0$ and $S^\prime=H^0$. For the $H^\pm$ self-energy: $S=H^\pm$ and $S^\prime=H^0$.}
\label{Fig:SSNSCTs}
\end{center}
\end{figure}
The interaction between the non-standard scalars gives another finite subset. When inspecting this contribution we found that all the corrections from a coupling between four non-standard scalars are cancelled. The two-loop diagrams which contain such a coupling can be written as a product of two scalar one-loop integrals. The mass counterterms in the subloop renormalization lead to the same product from the corrections to the scalar self-energies, but with an opposite sign. Consequently the two terms cancel each other.
The remaining contribution
\begin{equation}
\delrho[H,NS]{2}=\delrho[H,NS]{CT}+\delrho[H,NS]{2Loop}
\end{equation}
comes from all the diagrams which include a triple scalar coupling between $H^0$, $A^0$ and $H^\pm$. $\delrho[H,NS]{2Loop}$ is the result for the vector-boson self-energies of the generic two-loop diagrams in figure~\ref{Fig:NSHiggs2Loop}. For the subloop renormalization we need the corrections from the triple non-standard scalar coupling to the scalar self-energies, as shown in figure~\ref{Fig:SSNSCTs}. Inserting the corresponding mass counterterms into \eqref{Eq:drhoNSCT} leads to the result of $\delrho[H,NS]{CT}$.
\subsubsection{Scalar corrections from the interaction of the non-standard scalars with the SM scalars}\label{Subsec:drhoMix}
\begin{figure*}[ht]
\begin{center}
\includegraphics{VVMix2Loop.pdf}
\caption{Generic two-loop diagrams from the interaction between the SM-like scalars $S=h^0$, $G^0$, $G^\pm$ and the non-standard scalars $S^\prime=H^0$, $A^0$, $H^\pm$. $V=\{W,Z\}$}
\label{Fig:Mix2Loop}
\end{center}
\end{figure*}
\begin{figure*}[ht]
\begin{center}
\includegraphics{SSMix1Loop.pdf}
\caption{One-loop diagrams for the non-standard scalar mass counterterms
from the interaction between the SM-like scalars $S=h^0, G^0, G^\pm$ and the non-standard scalars $S^\prime=H^0,A^0,H^\pm$.}
\label{Fig:SSMixCTs}
\end{center}
\end{figure*}
As already mentioned another finite subset of two-loop corrections to the $\rho$ parameter comes from the interaction between the scalars $h^0$, $G^0$, $G^\pm$ with the non-standard scalars $H^0$, $A^0$, $H^\pm$. This interaction follows only from the part $V_{III}$ of the potential (see \eqref{Eq:V_III}) which is custodial-symmetry breaking. We denote the resulting contribution by
\begin{equation}
\delrho[H,Mix]{2}=\delrho[H,Mix]{CT}+\delrho[H,Mix]{2Loop},
\end{equation}
where $\delrho[H,Mix]{2Loop}$ is the part from the two-loop diagrams shown in figure~\ref{Fig:Mix2Loop}.
The divergences are canceled by $\delrho[H,Mix]{CT}$ from \eqref{Eq:drhoHCT}, which is obtained by calculating the mass counterterms in \eqref{Eq:drhoNSCT} from the diagrams in figure~\ref{Fig:SSMixCTs}.
\input{IHDM}
\section{The Two-Higgs-Doublet model}
\label{Sec:THDM}
The THDM Higgs sector consists of two complex $SU(2)_L$ doublet scalar fields with hypercharge $Y=1$:
\begin{equation}
\Phi_1=
\begin{pmatrix}
\phi_1^+ \\
\phi_1^0
\end{pmatrix}, \quad
\Phi_2=
\begin{pmatrix}
\phi_2^+ \\
\phi_2^0
\end{pmatrix}.
\label{Eq:Doublets}
\end{equation}
For our calculation we are using the parameterisation of the potential from \cite{gunion:1990}
\begin{equation}
\begin{aligned}
V=& \lambda_1\left(\Phi_1^\dagger\Phi_1-\frac{\hat{v}_1^2}{2}\right)^2+\lambda_2\left(\Phi_2^\dagger\Phi_2-\frac{\hat{v}_2^2}{2}\right)^2 \\
& +\lambda_3\left[\left(\Phi_1^\dagger\Phi_1-\frac{\hat{v}_1^2}{2}\right)+\left(\Phi_2^\dagger\Phi_2-\frac{\hat{v}_2^2}{2}\right)\right]^2\\
& +\lambda_4\left[\left(\Phi_1^\dagger\Phi_1\cdot\Phi_2^\dagger\Phi_2\right)-\left(\Phi_1^\dagger\Phi_2\cdot\Phi_2^\dagger\Phi_1\right)\right] \\
& +\lambda_5\left[\operatorname{Re}\left(\Phi_1^\dagger\Phi_2\right)-\frac{\hat{v}_1 \hat{v}_2}{2}\right]^2 \\
&+\lambda_6\left[\operatorname{Im}\left(\Phi_1^\dagger\Phi_2\right)\right]^2,
\end{aligned}
\label{Eq:potential}
\end{equation}
in which all the parameters appearing in the potential are chosen real.
The form of \eqref{Eq:potential} represents the most general potential involving two doublets and is $CP$-conserving, gauge-invariant, renormalizable and subject to a discrete $Z_2$ symmetry ($\Phi_2\rightarrow - \Phi_2$) which is only softly violated by dimension-two terms. The vacuum expectation values are written as
\begin{equation}
\langle\Phi_i\rangle=
\begin{pmatrix}
0 \\
\frac{v_i}{\sqrt{2}}
\end{pmatrix};\quad i=1,2,
\end{equation}
such that the electromagnetic gauge-symmetry $U(1)_\text{em}$ is preserved. The two doublets can be expanded around the vacuum expectation values
\begin{equation}
\Phi_i=
\left(\begin{array}{c}
\phi_i^+ \\
\frac{1}{\sqrt{2}}\left(v_i+\eta_i+i\chi_i\right)
\end{array}\right);\quad i=1,2.
\label{scalarparam}
\end{equation}
Inserting this decomposition into the potential yields for the linear
and quadratic terms the following component form:
\begin{align}
V =& -T_{\eta_1}\eta_1-T_{\eta_2}\eta_2 \notag\\
&+
\begin{pmatrix}
\phi_1^- & \phi_2^-
\end{pmatrix}
\hat{\mathbf{M}}^\phi
\begin{pmatrix}
\phi_1^+ \\
\phi_2^+
\end{pmatrix}\notag\\
&+\frac{1}{2}
\begin{pmatrix}
\chi_1 & \chi_2
\end{pmatrix}
\hat{\mathbf{M}}^\chi
\begin{pmatrix}
\chi_1 \\
\chi_2
\end{pmatrix}\notag\\
&+\frac{1}{2}
\begin{pmatrix}
\eta_1 & \eta_2
\end{pmatrix}
\hat{\mathbf{M}}^\eta
\begin{pmatrix}
\eta_1 \\
\eta_2
\end{pmatrix}+\dots.
\label{Eq:Vmass}
\end{align}
The explicit expressions for the tadpoles and the mass matrices in \eqref{Eq:Vmass} can be written in a compact form with the help of the quantities
\begin{eqnarray}
T_{1}=& \lambda _1 \left(v_1^2-\hat{v}_1^2\right),\\
T_{2}=& \lambda _2 \left(v_2^2-\hat{v}_2^2\right),\\
T_{3}=& \lambda _3 \left(v_1^2+v_2^2-\hat{v}_1^2-\hat{v}_2^2\right),\\
T_{5}=&\frac{1}{2} \lambda _5 \left(v_1 v_2-\hat{v}_1 \hat{v}_2\right).
\end{eqnarray}
The tadpoles are given by
\begin{eqnarray}
T_{\eta_1}&=-T_5 v_2-\left(T_1 + T_3\right) v_1,\label{Eq:eta1tadpole} \\
T_{\eta_2}&=-T_5 v_1-\left(T_2+T_3\right) v_2.
\label{Eq:eta2tadpole}
\end{eqnarray}
The mass matrices can be decomposed as
\begin{equation}
\hat{\mathbf{M}}^X=\mathbf{M}^X+\mathbf{M}^{T};\quad X={\phi,\chi,\eta},
\end{equation}
with
\begin{equation}
\mathbf{M}^T=\begin{pmatrix}
T_1+T_3 & T_5 \\
T_5 & T_2+T_3 \\
\end{pmatrix}
\end{equation}
and
\begin{align}
\mathbf{M}^\eta=&
\begin{pmatrix}
2 \left(\lambda _1+\lambda _3\right) v_1^2+\frac{v_2^2}{2} \lambda _5 & v_1 v_2 \left(2 \lambda _3+\frac{\lambda _5}{2}\right) \\
v_1 v_2 \left(2 \lambda _3+\frac{\lambda _5}{2}\right) & 2 v_2^2 \left(\lambda _2+\lambda _3\right) +\frac{v_1^2}{2} \lambda _5 \\
\end{pmatrix},\\
\mathbf{M}^\chi=&\frac{1}{2}
\begin{pmatrix}
v_2^2 \lambda _6 & - v_1 v_2 \lambda _6 \\
- v_1 v_2 \lambda _6 & v_1^2 \lambda _6 \\
\end{pmatrix},\\
\mathbf{M}^\phi=& \frac{1}{2}
\begin{pmatrix}
v_2^2 \lambda _4 & - v_1 v_2 \lambda _4 \\
- v_1 v_2 \lambda _4 & v_1^2 \lambda _4 \\
\end{pmatrix}.
\end{align}
The requirement that the tadpoles in \eqref{Eq:eta1tadpole} and \eqref{Eq:eta2tadpole} must vanish results in the minimum conditions
\begin{equation}
\hat{v}_{1,2}=v_{1,2}.
\label{Eq:Mincondition}
\end{equation}
With the help of these minimum conditions the mass matrices take the form $\mathbf{M}^\eta$, $\mathbf{M}^\chi$ and $\mathbf{M}^\phi$ which have to be diagonalized in order to obtain the physical Higgs states. The unitary transformations
\begin{eqnarray}
\begin{pmatrix}
G^\pm \\
H^\pm
\end{pmatrix}
=&
\Rot{\beta}
\begin{pmatrix}
\phi_1^\pm \\
\phi_2^\pm \\
\end{pmatrix}
\label{Eq:chargrot},\\
\begin{pmatrix}
G^0 \\
A^0
\end{pmatrix}
=&
\Rot{\beta}
\begin{pmatrix}
\chi_1 \\
\chi_2 \\
\end{pmatrix},
\label{Eq:chirot}\\
\begin{pmatrix}
H^0 \\
h^0
\end{pmatrix}
=&
\Rot{\alpha}
\begin{pmatrix}
\eta_1 \\
\eta_2
\end{pmatrix},
\label{Eq:etarot}
\end{eqnarray}
with
\begin{equation}
\Rot{x}=
\begin{pmatrix}
\cos x & \sin x \\
-\sin x & \cos x \\
\end{pmatrix}
\end{equation}
lead to five physical mass eigenstates: two $CP$-even states $h^0$ and $H^0$, a $CP$-odd state $A^0$ and a charged pair $H^\pm$. $G^0$ and $G^\pm$ are the usual Goldstone bosons associated with the longitudinal modes of the gauge bosons. The mixing angles are determined by
\begin{equation}
\tan\beta=\frac{v_2}{v_1}
\end{equation}
and
\begin{equation}
\sin2\alpha=\frac{2\mathbf{M}^\eta_{12}}{\sqrt{\left(\mathbf{M}^\eta_{11}-\mathbf{M}^\eta_{22}\right)^2+4(\mathbf{M}^\eta_{12})^2}}.
\end{equation}
From now on we will use the short notation $\sin x=s_x$, $\cos x=c_x$ and $\tan x=t_x$ for all the appearances of the mixing angles.
The kinetic terms of $\Phi_{1,2}$ in the Lagrangian describe the interactions between the scalar fields and the gauge fields and give rise to the gauge-boson masses
\begin{equation}
M_W^2=\frac{g_2^2 (v_1^2+v_2^2)}{4},
\end{equation}
\begin{equation}
M_Z^2=\frac{\left(g_1^2+g_2^2\right)(v_1^2+v_2^2)}{4}.
\end{equation}
$g_1$ is the gauge coupling of $U(1)_Y$ and $g_2$ is the gauge coupling of the $SU(2)_L$. They are also contained in the definition of the electroweak mixing angle $\theta_W$ by
\begin{eqnarray}
\cos^2\theta_W&=c_W^2=\frac{M_W^2}{M_Z^2},\label{Eq:CW_Definition}\\
\sin^2\theta_W&=s_W^2=1-c_W^2\label{Eq:SW_Definition}
\end{eqnarray}
and the electric charge
\begin{equation}
e=\sqrt{4 \pi \alpha_\text{em}}=\frac{g_1 g_2}{\sqrt{g_1^2+g_2^2}}
\end{equation}
with the electromagnetic fine structure constant $\alpha_\text{em}$.
After electroweak symmetry breaking the combination $v^2=v_1^2+v_2^2$ is fixed by the masses of the gauge bosons. The other seven free parameters of the Higgs potential can be rewritten in terms of the Higgs masses, the mixing angle $\alpha$, the ratio of the vacuum expectation values $t_\beta$ and the remaining self coupling parameter $\lambda_5$:
\begin{align}
\lambda_1=&\frac{e^2}{8M_W^2s_W^2}\left[\frac{c_{\beta-\alpha}s_\alpha}{s_\betac_\beta^2 }m_{h^0}^2
+\frac{s_{\beta-\alpha}c_\alpha}{s_\betac_\beta^2 }m_{H^0}^2\right]\notag\\
&+\frac{1}{4}\left(1-t_\beta^2\right)\lambda_5,
\\
\lambda_2=&\frac{e^2}{8M_W^2s_W^2}\left[\frac{c_{\beta-\alpha}c_\alpha}{s_\beta^2c_\beta }m_{h^0}^2
-\frac{s_{\beta-\alpha}s_\alpha}{s_\beta^2c_\beta }m_{H^0}^2\right]\notag\\
&+\frac{1}{4}\left(1-\frac{1}{t_\beta^2}\right)\lambda_5,
\\
\lambda_3=&-\frac{e^2}{8M_W^2s_W^2}\frac{s_{2\alpha}}{s_{2\beta}}\left(m_{h^0}^2-m_{H^0}^2\right)-\frac{\lambda_5}{4},
\\
\lambda_4=&\frac{e^2m^2_{H^\pm}}{2M_W^2s_W^2},
\\
\lambda_6=&\frac{e^2m^2_{A^0}}{2M_W^2s_W^2}.
\end{align}
The terms in \eqref{Eq:Vmass} can be written in the mass eigenstate basis by applying the rotations from \eqref{Eq:chargrot}, \eqref{Eq:chirot} and \eqref{Eq:etarot} before employing the minimum conditions. From the terms linear in the $CP$-even components $\eta_1$ and $\eta_2$ we obtain the tadpoles
\begin{eqnarray}
T_{h}=&-s_\alpha T_{\eta_1} + c_\alpha T_{\eta_2},\\
T_{H}=&c_\alpha T_{\eta_1} +s_\alpha T_{\eta_2},
\end{eqnarray}
of the fields $h^0$ and $H^0$. From the matrix $\mathbf{M}^T$ in the quadratic terms we obtain mass terms for the Goldstone bosons $G^0$ and $G^\pm$ with
\begin{equation}
m_{G^0}^2=m_{G^\pm}^2=-\frac{e}{2 M_W s_W} \left(s_{\beta-\alpha} T_{h}+c_{\beta-\alpha} T_{H}\right)
\label{Eq:Goldstonemass}
\end{equation}
and off-diagonal elements in the mass matrices in the mass eigenstate basis. The minimum condition in \eqref{Eq:Mincondition} is then equivalent to the requirement that the tadpole coefficients $T_{\{h^0,H^0\}}$ and all off-diagonal elements in the mass matrices must vanish. An additional consequence is that the Goldstone bosons receive no masses from the Higgs potential.
The couplings between the scalars and the fermions are restricted by the experimental limits on tree-level flavor changing neutral currents (FCNCs). It has been shown in \cite{Glashow:1976nt} and \cite{Paschos:1976ay} that a necessary and sufficient condition to avoid the FCNCs by neutral Higgs exchange at tree level is that not more than one of the doublets couples to fermions of a given charge. This has lead to four main models which are discussed in the literature
\begin{itemize}
\item type-I: all leptons and quarks couple only to the doublet $\Phi_2$;
\item type-II: the up-type quarks couple to the doublet $\Phi_2$, while all the down-type quarks and leptons couple to the doublet $\Phi_1$;
\item type-X or lepton specific model: all quarks couple to $\Phi_2$ and all leptons couple to $\Phi_1$;
\item type-Y or flipped model: the up-type quarks and leptons couple to the doublet $\Phi_2$ while the down-type quarks couple only to $\Phi_1$.
\end{itemize}
In the top-Yukawa approximation all the Yukawa couplings besides the one of the top quark, are neglected. Since the top-Yukawa coupling is given in all models by the interaction of the up-type quarks with the doublet $\Phi_2$, our result is valid in each of the four models. Since we are assuming a diagonal CKM matrix the top-Yukawa term of the Lagrangian takes the form
\begin{align}
\mathcal{L}_{Y,t}=&-m_t\overline{\psi}_t\psi_t-\frac{e m_t}{2 M_W s_W}\overline{\psi}_t\psi_t\left(\frac{s_\alpha}{s_\beta}H^0+\frac{c_\alpha}{s_\beta}h^0\right)\notag\\
&+i\frac{e m_t}{2 M_W s_W}\overline{\psi}_t\gamma_5\psi_t\left(G^0+\frac{1}{t_\beta}A^0\right)\notag\\
&+\frac{e m_t}{\sqrt{2} M_W s_W}\overline{\psi_t}\omega_-\psi_b\left(G^+ +\frac{1}{t_\beta}H^+\right)+h.c.
\label{Eq:topLagrangian}
\end{align}
where $m_t$ is the mass of the top quark, $\psi_{t,b}$ are the Dirac spinors of the top and bottom quarks and $\omega_-=(1-\gamma_5)/2$ is the projector on the left-handed spinor states.
Models with a more general structure for the Higgs-fermion interactions are usually refered to as type-III models \cite{Hou:1991un,Chang:1993kw,Atwood:1996vj} and allow couplings of all the SM fermions to both Higgs doublets. The more general Higgs-fermion couplings are then strongly restricted by the absence of FCNCs. | 0c870621eabe49e88f5a15bf42c8972f001a2121 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{\label{sec:level1}Introduction\protect\\}
Contamination of a cylindrically or toroidally trapped electron plasma by a fractional population of ions can, under certain conditions, disrupt the equilibrium of the trapped electron cloud. The related instability, known as the ion resonance instability, manifests itself in the form of growing poloidal modes (diocotron modes\cite{rdav1,dub}) of the plasma\cite{megh3}. The primary cause for such contamination (either planned or inadvertent) in most electron plasma trap experiments is the process of ionization of the neutral background gas in the trap by the electrons. A brief history of analytical modelling\cite{levy,rdav1, rdavh1,rdavh2,faj,rdavh3}, numerical analysis\cite{chen}, and experiments\cite{eck,per,bet,kab,bet2,mark,stone,lach} performed to understand this destabilization process can be found in Sec I of Reference 3\cite{megh3}. Of the theoretical models for the ion resonance instability, the one developed by Davidson and Uhm\cite{rdavh1} aptly describes the collisionless dynamics of the instability in the present set of numerical experiments performed with an initially cold, partially neutralized electron cloud in cylindrical confinement. The model describes the ion resonance instability as a rotating two-stream instability between the electron component and the partially neutralizing ion component of the plasma. The differential rotation between the components in mutual equilibrium drives the instability; the rotation itself being a result of the radial self-electric field of the nonneutral plasma. Davidson and Uhm's linearised model\cite{rdavh1} works for any cold equilibrium of the two component plasma in cylindrical confinement. It shows that depending on the equilibrium parameters, one or more exponentially growing diocotron modes may be excited on the plasma profile with the equilibrium condition determining the fastest growing diocotron mode in the system. An abridged version of the analytical procedures for this model can be found in Sec III of Reference 3\cite{megh3}.
In an earlier work\cite{megh3} we performed numerical experiments to investigate the full nonlinear evolution of the ion resonance instability in partially neutralized electron clouds in cylindrical traps. The numerical experiments were performed for a set of different initial unstable equilibria of the two component plasma (defined by the equilibrium parameters), and a well benchmarked\cite{megh, megh2} 2D Electrostatic Particle-in-cell (PIC) code was utilized for these simulations. Results\cite{megh3} from the initial linear growth phase of the instability were in excellent agreement with the linearised analytical model of Davidson and Uhm\cite{rdav1} which served as an added benchmark for the code. The simulations delved further in time beyond the linear growth phase and revealed several new and interesting nonlinear phenomena associated with the instability. For example, a process of simultaneous wave breaking of the excited poloidal mode on the ion cloud and pinching of the poloidal perturbations on the electron cloud occurring in the nonlinear phase of the instability was reported\cite{megh3}. This simultaneous nonlinear dynamics of the two components was found to be associated with an energy transfer process from the electrons to the ions. At later stages of the nonlinear phase, a heating induced cross-field transport of the heavier ions and tearing across the pinches on the electron cloud followed by an inverse cascade of the torn sections was also demonstrated\cite{megh3}.
These ion resonance instability simulations of the previous work\cite{megh3} were performed in an ideal scenario with a perfect vacuum as a background. For the present work we have taken these simulations a notch closer to a usual experimental scenario by including the effect of elastic and excitation collisions of the electrons with background neutrals that are invariably present at experimental low pressures inside the trap. The first set of numerical experiments of this paper (Sec III) are PIC-with-MCC simulations that investigate how collisional relaxation of the electron cloud's profile can influence dynamics of the ion resonance instability happening due to the ion contamination of the cloud.
In the second set of numerical experiments of this paper (Sec IV) we initiate electron clouds in stable $l=1$ ($l$ being the diocotron mode number) mode orbits of very small radius, and utilize the MCC to simulate non-ionizing elastic and excitation collisions of these electrons with a neutral background in course of the simulations. The objective of this set of numerical experiments is to test if such non-ionizing collisions can destabilize the stable motion of the pure electron clouds. It had been theorized by Davidson and Chao\cite{chao} that elastic collisions between electrons and background neutrals will not only cause relaxation of the electron cloud's profile but can also destabilize any small-amplitude, stable $l=1$ mode azimuthal asymmetry present in the cloud into a growing $l=1$ mode by virtue of their capability of extracting potential energy from the cloud. Contrary to these theoretical predictions, our simulations show that such non-ionizing collisions between electrons and background neutrals can not by themselves destabilize a stable configuration of the cloud. We believe that this has got to do with the manner in which potentially energy of the cloud gets reduced by such collisions. A descriptive comparison of the potential energy extraction process between non-ionizing collisions and a resistive wall instability\cite{white,bet3} that is known to be capable of destabilizing the cloud, will make this point clear. This is presented in Sec IV of this paper.
A brief outline of the following Sections of this manuscript is as follows. Sec. II is a description of the 2D3v PIC code with newly incorporated MCC facility. The PIC-with-MCC simulations of partially neutralized electron clouds and initially axially asymmetric pure electron clouds will be presented in Sec. III and Sec. IV respectively. The numerical MCC scheme used to produce the results of Sec. III and IV is validated statistically and numerically in Sec. V. The MCC verification procedure is carried out for one of the experiments of Sec. III. A summary of the simulation results and related discussions make up the last section of the paper, Sec. VI.
\section{The 2D3v PIC code with MCC facility}
Different technical aspects of our PIC-with-MCC code, and their signifance are described in the following sub-sections.
\subsection{From the 2D PIC code to the 2D3v version}
A previously developed 2D Electrostatic PIC has been upgraded to a 2D3v PIC code with facility for Monte-Carlo-Collisions for the purpose of simulating the collisional interactions of nonneutral plasmas with background neutrals along with the collisionless dynamics of the plasma in cylindrical trap cross-sections. The 2D PIC code\cite{megh3, megh, megh2} (earlier version) was developed in FORTRAN-90 using Cartesian co-ordinates, and parallelized with OPEN-MP. It can simulate cross sections of multi-component plasmas of varying neutrality, confined within any perfectly conducting closed boundary curve. The size, shape, and toroidal aspect ratio of the boundary can be manoeuvred as per requirements. The numerical schemes employed in the development of different parts of the code along with references to the sources\cite{bird,hal,ket,cz,lal,chin} of these numerical procedures is described Sec. II of Reference 3\cite{megh3}. The 2D code has passed rigorous benchmarking with analytical results for cylindrically trapped pure electron plasmas\cite{megh,megh2} as well as partially neutralized electron plasmas\cite{megh3}. The 2D3v PIC-with-MCC version of the code retains all the features of the original 2D code. In subsections B to D we will discuss only those new features of the 2D3v PIC-with-MCC version that have been applied for the numerical experiments of this paper. For convenience of describing these numerical features of the code we will be explicitly referring to the computational particles of simulation as pseudo-electrons/ions throughout Sec. II instead of loosely calling them electrons/ions as in the rest of the paper. The last two subsections, E and F, describe the setup of the numerical experiments of Sec. III-IV and the numerical diagnostics used in the simulations respectively.
\subsection{Giving the pseudo-electrons an axial velocity component perpendicular to the 2D PIC plane}
In addition to the 2D Cartesian components of velocity the pseudo-electrons now have a $3^{rd}$ velocity component representing their axial velocities. A common fixed magnitude of this component is initially given to all pseudo-electrons. The axially up or down direction for this component is selected randomly using a pseudo random number generator. The axial velocity does not get modified by the 2D PIC dynamics of the pseudo-electron. It can get modified only when the pseudo-electron is scattered by a collisional event. Thus the extra axial component, with a suitable choice of magnitude, will have a very similar effect on the collisional dynamics of the pseudo-electrons in simulation, as the axial bouncing of electrons between electrostatic end plugs\cite{malm} have on the collisional dynamics of electrons in cylindrical trap experiments.
\subsection {The MCC procedure}
The MCC is a technique of making computational particles (here pseudo-electrons) undergoing guided motion in a simulation collide at suitable intervals (collision time step) with background matter that is not represented in any form in the simulation (here neutral Argon atoms)\cite{vah}. The method is useful when we are interested only in the effect (through collisions) of the latter on the former and not vice versa. To set up the MCC within the framework of a PIC code, the computational particles (here pseudo-electrons) of the simulation have to be treated as real particles (here electrons) inside the MCC algorithm\cite{cbird}. This implies that mass of a pseudo particle entering the MCC routine gets divided by its representation number. MCC can be implemented using a pseudo random number generator if the collision cross-section of a particle (here electrons) is available in analytical or numerical form as a function of its kinetic energy\cite{frig}. Taking the average kinetic energy of a particle (particle here being a pseudo-electrons represented as an electron by scaling down its mass) in the collision time step its collision probability in that interval is determined from the available analytical/numerical collision cross-section. Then the pseudo random number generator churns out a pseudo random number that decides whether the particle undergoes a collision or not based on its collision probability (the Monte-Carlo method). If there are more than one kind of collision possible between the particle and the background then the generated random number can also decide between different possible types of collision and the possibility of no collision for the particle\cite{nan2}. Of course the probability of each type of collision in the collision time step should be available\cite{frig}. For the present set of experiments only elastic collisions and the most abundant first level excitation collisions of background atom has been considered.
If the fate of the particle is found to be no collision then its velocity components are left untouched else the code proceeds to the next step of the MCC which decides how the particle gets scattered by the collision. If the collision happens to be of the elastic type then using a set of numerical formulas relying on pseudo random numbers the scattered speed and direction of the particle from the collision is determined\cite{nan1,okh}. These formulas are fed with the incident kinetic energy of the colliding particle and also the fixed temperature and pressure of the background gas to produce the required outputs. In our MCC code the set of formulas for elastic collision execution are very realistic. They implement the change in kinetic energy of the particle as a result of the elastic collision as well as the natural anisotropy of the scattering with respect to the incident direction of the particle. The scattering angle is determined from a very accurate formula for differential scattering cross section introduced in the year 2002\cite{okh}. This formula is actually a correction on the differential scattering cross section formula that had been in use earlier\cite{vah, nan1}. The colliding neutral though massive (in comparison with electron) is not approximated as a stationary body in the collision but its inherent thermal velocity is also taken into account in the scattering procedure through pseudo random numbers\cite{nan1}.
If on the other hand, the collision happens to be a first excitation of the background (Ar atom), then the MCC follows a two step algorithm to simulate the collision. First the kinetic energy of the particle (pseudo electron with mass scaled down to electron mass) is reduced by an amount equal to the first excitation energy (= 11.55 eV for Ar atom)\cite{vah}. Next the particle with reduced velocity goes through an elastic scattering procedure as explained above. The first-level excitation collisions have been grouped together with, and treated as elastic collisions in the numerical experiments of this paper. This is because there is an elastic collision happening in every exciting collision adding to the relaxation of the electron component profile which is the subject of interest here.
After the MCC procedure is complete the masses of the computational particles are again scaled up to their original values for the regular PIC part of the code.
The procedure adopted for the MCC in the PIC-with-MCC code can be summarized as follows:
\begin{itemize}
\item{Scale down the mass of the pseudo electron to that of an electron}
\item{Now calculate the electron's kinetic energy}
\item{Calculate the probability of elastic collision, probability first level excitation collision, and probability of no collision for that kinetic energy of electron}
\item{Generate a pseudo random number to decide the fate of the particle}
\item{a) For elastic collision use pseudo random number based formulas to decide kinetic energy after scattering and scattered direction of the electron b) For first level excitation reduce energy of the electron by a magnitude equal to first level excitation energy of the neutral and then use the elastic collision formulas to determine final scattered speed and direction c) For no collision leave the velocity components of the particle untouched}
\item{Scale up the mass of the particle to the representative mass of pseudo electron for the regular PIC part of the code}
\end{itemize}
\subsection {The collision time interval}
The interval of time representing the collision step is a critical parameter which determines how closely and how efficiently the MCC simulates the collisional interaction of particles with the background. The first criterion the collision time step should satisfy is that its value should be so small that possibility of simulation particles colliding more than once with the background matter in this interval is practically removed all through the simulation\cite{vah}. Otherwise the MCC routine will overlook possible collisions leading to divergence of the simulation from a real experiment. It can be calculated that a collision time step in which the highest (among particles) probability of a collision remains below the value $0.095$ will satisfy this first criterion\cite{vah}.
At the same time one should avoid diminishing the size of the collision time step beyond the point at which average (over particles) probability of collision in the interval becomes less than $1/N$, where $N$ is the size of the colliding pseudo-particle population. This is because such a choice will result in too many executions of the MCC routine in the simulation period with zero or very few pseudo particles actually colliding in each execution. Hence it will bring down the speed of execution of the PIC-with-MCC code. However improving efficiency of the MCC by increasing size of the collision time step should only be attempted as long as the primary criterion (related to accuracy of simulation) is satisfied for all pseudo particles.
In our simulations the pseudo electrons are loaded in rigid rotor equilibrium with a fixed angular velocity and a common fixed magnitude of the axial velocity component. Hence a pseudo electron lying at the half-radius of the cylindrical, uniform density electron cloud will have the mean speed and the average collision cross section for electrons in the cloud. Hence in order to ensure that the first criterion is satisfied for our chosen collision time step we checked that the collision probability of the particles at half-radius of the cloud is well below the $0.095$ limit\cite{vah}. Also the size of the collision time step was adjusted so that on a average a fraction of $1.197\times10^{-3}$ from the total pseudo electron population of $87834$ collides in every collision step, thus satisfying the second criterion as well.
\subsection{The numerical experimental setup and its PIC parameters}
All simulation experiments described in sections III and IV of this paper were carried out for typical parameters found in experiments of cylindrical traps\cite{malm,dub}, in particular, a wall radius, $R_w=0.125m$, and axial magnetic field, $B_z=0.015T$, and radial extent of the uniform density plasma, $r_p=0.5\times{R_w}$ . The simulation time step, ${\delta}t=10^{-11}sec$ is chosen much smaller than the cyclotron time period, $T_{ce}=2.38\times10^{-9}sec$ of electrons, such that the code can well resolve the cyclotron motion of both ions and electrons. The collision time step ${\Delta}t_c$ is adjusted according to the chosen pressure and temperature of the background gas. The magnitude of axial velocity with which the electrons are loaded is $1.02727\times10^7 m s^{-1}$ which implies that for a typical cylindrical plasma column of length of $60 cm$\cite{dris2} the bounce frequency of electrons is roughly $8.5 $ MHz. This value of the axial bounce frequency is higher than the typical experimental bounce frequency ($<0.2$ MHz\cite{dris2}) of a $60 cm$ long cylindrical plasma column. By loading the electrons with high axial speeds we have made the electrons more energetic and thereby maintain the average total non-ionizing (elastic + excitation) collision cross section between the moderate orders of $10^{-21}\,m^2$ and $10^{-20}\,m^2$ (see Fig. 3 of Reference 30\cite{vah}) in the numerical experiments of Sec III and IV. It must be re-emphasized here that the loaded value of the axial velocity of electrons in the simulations serves only to adjust the collisional probabilities and does not partake in any of the regular PIC dynamics of the plasma.
PIC-parameters used in the simulations of Sec. III are $87834$ pseudo particles for each plasma component, on a $70\times70$ grid. As the actual population of the ions is a fraction, $f$ of the electron population, the number of real ions represented by each ion pseudo-particle has also been scaled down to a fraction, $f$ of the representation value of electron pseudo-particles. The ion species (H+ ion) is $1836$ times heavier than an electron and the background atom (Ar atom) is $72820.77$ times heavier than an electron.
In Sec. IV, the simulations are for pure electron plasmas represented by $87834$ pseudo electrons on a $70\times70$ grid. The background species is again Argon.
\subsection{Diagnostics for the simulations}
The diagnostics that will be used in the experiments of Sec. III and IV are 4 potential probes which are basically 4 azimuthally equispaced Cartesian cell nodes inside the cylindrical trap that are all located at a common radius that is very close to the wall. As per their location on the trap cross-section the 4 potential probes have been named as left probe, right probe, top probe, and bottom probe. The potential probes signal are used to study the dynamics of the plasma, and find the frequencies and growth rates of diocotron modes excited on the plasma\cite{megh3,megh}. Besides the potential probes the energetics (kinetic and potential energies versus time) of both components of the plasma, and the radius of the centre-of-mass of the electron component as a function of time have also been used as diagnostics in Sec. III and IV.
\section{Ion resonance instability in presence of electron-neutral elastic collisions}
For the numerical experiments of this Section we loaded two uniform density 2-component (e- and H+) nonneutral plasma profiles in their respective rigid rotor equilibrium at different background gas pressures. The first profile was loaded with a value of electron component Brillouin ratio, $f_b=0.02$ and a fractional neutralization, $f=0.15$ provided by the uniformly mixed H+ ions. The second profile had values $f_b=1.0$ and $f=0.1$. In both cases the electron and the ion components of the cylindrical, uniform density nonneutral plasma are rotating in the slow mode of the mutual rigid rotor equilibrium\cite{megh3}. Hence the electron component is rotating with angular velocity $\omega_{re}^{-}$ while the ion component is rotating with angular velocity $\omega_{ri}^{-}$ in mutual equilibrium\cite{megh3}.
\begin{figure}
\centering
\includegraphics[scale=0.7]{sfnew0}
\captionsetup{justification=raggedright,
singlelinecheck=false
}
\caption{Readings of the left potential probe, $V_p$ for the $f_b=0.02$, $f=0.15$ equilibrium at background pressure, $P_{bg}$ values of $0$ and $2\times10^{-6} Torr$. $V_p$ readings of the equilibrium at the other two simulated background pressures of $2\times10^{-7} Torr$ and $2\times10^{-8} Torr$ (not shown here) also overlap nearly perfectly with the plotted readings in this figure. $T_{bg}=300 K$ is the chosen temperature of the background gas in this set of runs for all $P_{bg}>0$. Normalized time, $\tau$ is in units of electron cyclotron time, \textit{i.e.} $\tau=t/T_{ce}$. Readings upto the growth phase (till $\tau=5000$) have been zoomed in here for clarity while the inset has the complete readings upto the end of the simulations. The other three potential probes also recorded almost perfectly overlapping readings for all the four values of $P_{bg}$.}
\end{figure}
\begin{figure}
\centering
\includegraphics[scale=1.0]{sfnew}
\captionsetup{justification=raggedright,
singlelinecheck=false
}
\caption{Snapshots of pseudo particles (electrons in red and ions in green) for the $f_b=1.0$, $f=0.1$ equilibrium load at zero pressure ($P_{bg}=0$) that excites a nonlinear $l=1$ mode . Below each snap, the time elapsed is mentioned in normalised units of electron cyclotron time, \textit{i.e.} $\tau=t/T_{ce}$.}
\end{figure}
\begin{figure*}
\includegraphics[scale=1.0]{sfnew2}
\captionsetup{justification=raggedright,
singlelinecheck=false
}
\caption{Potential-probe signal analysis of the $f_b=1.0$, $f=0.1$ equilibrium load at zero pressure ($P_{bg}=0$). The analysis of the left probe's signal is shown here. The other three potential probes also give the same results from their data analysis: (a) $V_p$ is the electrostatic potential recorded by the left probe. Normalized time, $\tau$ is in units of electron cyclotron time, \textit{i.e.} $\tau=t/T_{ce}$. (b) is a zoomed in plot of the same signal upto $\tau=7600$. This part of the signal has been exponentially fitted for a linear approximation of the growth of the $l=1$ mode, and has been Fourier analysed for an estimate of the mode's frequency. (c) is an exponential fit on the readings of (b). $V_{p0}$ is the left probe's reading at $\tau=0$ . The y-axis is in log scale while time axis is in linear scale. (d) A FFT is performed on the readings of (b). Here the x-axis is in rad/sec unit and the y-axis is the power factor in arbitrary unit.}
\end{figure*}
\begin{figure}
\centering
\includegraphics[scale=0.7]{sfnew3}
\captionsetup{justification=raggedright,
singlelinecheck=false
}
\caption{Energetics of the $f_b=1.0$, $f=0.1$ equilibrium load at zero pressure ($P_{bg}=0$): $W$ represents the energy components in the legend, $E_0$ is the initial total energy, and $\tau=t/T_{ce}$. $J_e$ and $J_i$ are the potential energies of electrons and ions respectively. $K_e$ and $K_i$ are the kinetic energies of electrons and ions respectively. $E$ is the total energy of the system. The energy components are normalised by $E_0$ and plotted as a function of time. The time axis is normalised by the cyclotron time period of electrons, $T_{ce}$.}
\end{figure}
\begin{figure*}
\includegraphics[scale=1.0]{sfnew4}
\captionsetup{justification=raggedright,
singlelinecheck=false
}
\caption{Comparison of diagnostic results between different background pressure, $P_{bg}$, for the $f_b=1.0$, $f=0.1$ equilibrium load: With $\tau=t/T_{ce}$ the set of plots are zoomed in showing their respective diagnostic readings upto $\tau=8500$ for clarity. $T_{bg}$ is the temperature of the background gas for all $P_{bg}>0$. (a) The radius of the centre-of-mass of the electron component, $R_{cme}$ normalised by the wall radius $R_W$ is plotted as a function of time. \textit{Inset}: The same set of plots is extended upto the end of the simulations at $\tau=22700$ (b) The left potential probe's reading, $V_p$ as a function of time at different background pressures. (c) The Potential Energy of the electron component, $J_e$ normalised by the initial total energy of the 2-component plasma, $E_0$ is plotted as a function of time.(d) The Potential Energy of the ion component, $J_i$ normalised by the initial total energy of the 2-component plasma, $E_0$ is plotted as a function of time. }
\end{figure*}
Davidson and Uhm's linear analytical theory\cite{rdavh1}, when applied on the first equilibrium ($f_b=0.02$, $f=0.15$) at zero background pressure (absence of neutrals) predicts growth of an unstable $l=1$ azimuthal mode ($l$ being the mode number) on both components with angular frequency $2.827\times10^6 rad/sec$ and growth rate $9.47\times10^5 rad/sec$. Our simulation of the same equilibrium sans neutrals are in excellent agreement with the linearised theory in the linear phase of the instability. We get an unstable $l=1$ mode with angular frequency $2.792\times10^6 rad/sec$ and growth rate $9.6\times10^5 rad/sec$. A linear perturbation analysis of the second equilibrium ($f_b=1.0$, $f=0.1$) at zero background pressure predicts that this equilibrium should be stable\cite{rdavh1}. However the corresponding simulation reveals that this equilibrium infact destabilizes very slowly in the the form of a nonlinear $l=1$ mode accompanied by a few higher weaker modes\cite{megh3}. A linearly fitted growth rate of the impure (mixed with a few higher weaker modes) $l=1$ mode is obtained as $4.0\times10^5 rad/sec$ while its angular frequency comes out to be $2.733\times10^7 rad/sec$\cite{megh3}. The slower nonlinear destabilization of the second equilibrium in absence of neutrals can be attributed to nonlinear terms in the perturbed equations of motion that have been neglected in its linear analysis. That the destabilization of the second equilibrium is not a numerical artefact has been verified in our previous work\cite{megh3} by repeating the experiment with greater number of pseudo particles for both components. The repeated simulations also gave results in perfect agreement with original simulation at zero background pressure.
We now subjected these two unstable equilibria to three different values of neutral background pressures, $P_{bg}$. The values of $P_{bg}$ chosen are $2\times10^{-8} Torr$, $2\times10^{-7} Torr$, and $2\times10^{-6} Torr$ at a common fixed background temperature $T_{bg}=300 K$. The collision time step, ${\Delta}t_c$ for the lowest background pressure was chosen as $6\times10^{-6} sec$ and it was reduced by an order of magnitude and two orders of magnitude for the $2\times10^{-7} Torr$ and $2\times10^{-6} Torr$ background pressures respectively. This kind of scaling of ${\Delta}t_c$ between the three background pressures helped in maintaining uniformity of the MCC routine's collision resolving capacity among the different experimental pressures. Ofcourse all the three values of ${\Delta}t_c$ satisfy the two criteria for collision time interval in their corresponding experiments. With the help of these PIC-with-MCC simulations we were able to study the effect of electron-neutral elastic (and exciting) collisions on two ion resonance instabilities that are growing via the fundamental $l=1$ mode at different rates. A point to be noted here is that in these sets of experiments the loaded ions are not of the same species as the background gas while in typical electron plasma traps the destabilizing ions are created from the ionization of the background atoms by electrons. Simulations of the latter kind is beyond the scope of this paper and has been reserved for our next work\cite{megh4}. For the present work we chose the ion species H+ for which the ion resonance instability is a theoretically and numerically well understood phenomenon\cite{rdav1,rdavh1,megh3}. The background gas with which the electrons make non-ionizing collisions is Argon. In our upcoming work we will be studying the effects of electron impact ionization of the Argon atoms on the dynamics of the plasma\cite{megh4}. As there are no ionizing collisions in the present set of numerical experiments the neutrals will only influence the instability through collisional relaxation of the electron cloud's profile. So the experiments of this Section can be described as a qualitative investigation of how such a process of profile relaxation of the electron cloud due to an arbitrary neutral species influences the ion-resonance instability due to an arbitrary ion contaminant mixed in the electron cloud.
The detailed linear and nonlinear dynamics of the first ($f_b=0.02$, $f=0.15$) equilibrium in the absence of any colliding neutrals has been described in detail in Sec. IV - subsection A of Reference 3 along with corresponding snapshots of the two-component plasma in course of the simulation (Fig 1 of Reference 3)\cite{megh3}. In brief, this unstable equilibrium leads to an in-phase excitation of the $l=1$ mode on both components of the plasma. The two components gradually get separated azimuthally (growing phase separation) as the orbital radius of the $l=1$ mode increases for both components. During the nonlinear growth phase of the mode, ions gain potential energy and are also collisionlessly heated by the instability. The gain in potential energy and kinetic energy of the ion component comes at the expense of the potential energy of electrons. The heated ion component develops a filamentation structure on its surface which eventually makes contact with the grounded wall during the unstable $l=1$ orbital motion. Soon after the entire ion population gets transported and lost to the grounded wall through this filamentation structure. The $l=1$ orbit of the electron component however stabilizes very close to grounded wall with the electron component profile deforming into an elliptical shape. Hence the final state of the ion resonance instability for this particular unstable equilibrium is an elliptical pure electron cloud making stable orbits very close to the grounded wall.
Now we load this equilibrium in the presence of an Argon neutral background at the three experimental background pressures. For the three PIC-with-MCC simulations the electrons are also loaded with an added axial velocity component enabling them for 3D elastic and exciting collisions with background neutrals. What we observe from our diagnostics comprising of potential probes (Fig 1), radius of centre-of-mass of the electron component, and the potential energy of the electron component, is that there is negligible variation in the dynamics of this instability brought about by the electron-neutral non ionizing collisions for the above set of background pressures. The diagnostics record values with progression of the simulation that are identical to the zero pressure experiment for all the three experimental background pressures. More specifically for all the experimental $P_{bg}$ values including $P_{bg}=0$, the potential probes record a similar exponential growth and saturation of $l=1$ mode, the potential energy of the electron cloud decreases and saturates similarly with time, and the radial location of the electron cloud also increases and saturates similarly with time. As a illustrative example, we have shown in Fig. 1 the left potential probe's reading for the lowest and the highest background pressures ($P_{bg}=0$ and $P_{bg}=2\times10^{-6} Torr$). It can bee seen that the the two signals lie nearly on top of one another. All these diagnostic results bring us to the conclusion that for this particular equilibrium at the chosen set of background pressures the electron-neutral elastic and exciting collisions can not influence the evolution of the ion resonance instability on the e- H+ cloud. We will get back to reason for this in a later comparison with the results for the second equilibrium.
Next we come to the set of experiments performed at $f_b=1.0$, $f=0.1$. Loading this equilibrium at zero pressure produces a slower growing $l=1$ mode than the $f_b=0.02$, $f=0.15$ equilibrium. Fig. 2 is a set of snapshots showing the evolution of the 2-component plasma profile for this unstable equilibrium at zero background pressure. The snapshots show that the $l=1$ mode is eventually excited after the plasma profile goes through a few higher weaker modes (the $l=3$ mode being the most visible in Fig. 2d). The $l=1$ mode gradually takes over as the most dominant mode (Fig. 2h-i) and towards the end of the simulation it is the only excited mode on the profile executing stable orbits (Fig. 2j-l) after having saturated its growth. We analysed the growth and frequency of the $l=1$ mode using our potential probes. Fig. 3a shows the signal recorded by the left potential probe for this simulation at zero background pressure. As explained earlier for this particular equilibrium the excited $l=1$ mode is nonlinear and mixed with other modes from the outset. However it shows a slow growth rate right up to its saturation phase (Fig. 3b). So it is worthwhile to get an estimate of the growth of this impure $l=1$ mode with a linear exponential fit on the growing part of the potential probe signal (Fig. 3c). This growth rate comes out to be $4.0\times10^5 rad/sec$ . We also performed a FFT on the truncated reading of Fig. 3b to obtain an estimate of the angular frequency of the mode which comes to be $2.733\times10^7 rad/sec$ (Fig. 3d). The energetics of the simulation at zero background pressure is shown in Fig 4. Here we can see that the nonlinear growth of the mode is associated with collisionless heating of the electrons by the instability at the expense of their potential energy. A smaller fraction of the potential energy of electron component is also pumped into the potential energy of the ion component during the growth of the mode. The energy exchange process stops with the saturation of the mode.
Now the $f_b=1.0$, $f=0.1$ equilibrium is loaded at the same experimental background pressures as the $f_b=0.02$, $f=0.15$ equilibrium. The electron now have cylindrical trap like axial velocities and are capable of 3D elastic and excitation collisions with the background neutrals. The collisions relax the profile of the electron cloud and the resultant changes in the electron density profile influences the ongoing collisionless ion resonance instability. Specifically collisional relaxation of the electron cloud reduces the average density of the electron cloud and thereby increases the fractional density of the ions mixed in it. The dynamically changing electron density and fractional density of ions feed back on the ongoing ion-resonance (two-stream) instability between the two components of the nonneutral cloud and produce deviations in the paths of progression of the instability that are uncorrelated at different background gas pressures. As the variations in the path of evolution of the plasma at the experimental background pressures are small in comparison to average evolution over time it is sufficient to identify and discuss these variations only qualitatively without quantifying their magnitudes with direct measurements. We will discuss these variations with the help of the diagnostic results of Fig. 5.
First let us look at radial location of the electron component as a function of time which directly shows how the $l=1$ mode grows in orbital radius and saturates at different background pressures for the electron component (Fig. 5a). It can be seen from Fig. 5a that the $P_{bg}=0$ load has the slowest increase in orbital radius of the electron cloud with the growth rate increasing the in the order of $P_{bg}=2\times10^{-7} Torr$, $P_{bg}=2\times10^{-6} Torr$, and $P_{bg}=2\times10^{-8} Torr$. After the growth period the $l=1$ mode for the lower background pressures of $P_{bg}=0$ and $P_{bg}=2\times10^{-8} Torr$ saturate at slightly smaller orbital radii as compared to the two higher background pressures of $P_{bg}=2\times10^{-7} Torr$ and $P_{bg}=2\times10^{-6} Torr$ (see inset of Fig. 5a). The potential probe readings also mirror the same information about the growth of the mode (Fig. 5b).
Next we have also compared the time evolution of two energy components between the different background pressures, viz the potential energy energy of the electron cloud and the potential energy of the ion cloud. We have already seen in Fig. 4 that for the $P_{bg}=0$ load the potential energy of the electron component decreases while that of ion component increases during the growth of the instability. Fig. 5c and 5d show the potential energies of the electron component and ion component respectively upto the the growth phase of the instability for the experimental background pressures. It can be seen from Fig. 5c-d that the rates of decrease/increase in potential energy of electrons/ions increases with increase in background pressure.
Hence we can conclude that changing the collisionality of the electron cloud by changing the background gas densities has brought about interesting subtle variations in the dynamics of the ion-resonance instability that are unique to each experimental background gas pressure. It is also to be noted that once the growth of the collisionless ion resonance instability saturates there is no further destabilization caused by the continuing collisions (see inset of Fig. 5a) which further emphasizes that the collisions are only influencing the dynamics of the cloud through feedback of electron cloud's profile relaxation on the growing ion resonance instability.
We have seen that elastic and excitation collision of electrons with background neutrals influence the dynamics of the ion resonance instability for the $f_b=1.0$, $f=0.1$ equilibrium while they have no effect on the instability dynamics for the $f_b=0.02$, $f=0.15$ equilibrium. From the nature of the growth of instability in the two equilibria it is evident that this is a due to the different lengths of growth phase in the two equilibria. The excited $l=1$ mode in the $f_b=0.02$, $f=0.15$ equilibrium grows faster and saturates quicker (Fig. 1) than the $f_b=1.0$, $f=0.1$ equilibrium (Fig. 3a). So at a particular background pressure there will be more elastic and excitation collisions occurring during the growth phase of the instability for the $f_b=1.0$, $f=0.1$ equilibrium as compared to the $f_b=0.02$, $f=0.15$ equilibrium. Hence there is more feedback on the instability by the collisional relaxation of the electron cloud for the $f_b=1.0$, $f=0.1$ equilibrium. The feedback process brings about the subtle changes in the evolution of this equilibrium at the experimental background pressures. Furthermore we have already seen that once saturation of the mode is achieved, the ongoing collisions can not further destabilize the cloud. Hence a smaller time window of influence of the electron-neutral collisions for the $f_b=0.02$, $f=0.15$ equilibrium is the precise reason why we do not observe any deviation in dynamics at different experimental background pressures for this equilibrium.
\section{Dynamics of $l=1$ perturbed pure electron clouds in presence of collisional background}
\begin{figure}
\centering
\includegraphics[scale=0.7]{sfnew5}
\captionsetup{justification=raggedright,
singlelinecheck=false
}
\caption{Potential probe data for the perturbed load (off-axis by $0.1\times{R_W}$) of the $f_b=0.02$ and the$f_b=0.9$ pure electron clouds at a common background pressure, $P_{bg}=2\times10^{-6} Torr$: $V_p$ is the reading of the left potential probe and $V_{p0}$ represents its corresponding initial value in the two simulations. The y axis is plotted in logscale for clarity. $T_{bg}=300 K$ is the temperature of the background gas in this set of runs. Normalized time, $\tau$ is in units of electron cyclotron time, \textit{i.e.} $\tau=t/T_{ce}$. Hence it is clear from the potential probe's data that there is no growth of the initially implanted $l=1$ mode in the simulation. The other three potential probes also gave similar results for this set of simulations.}
\end{figure}
In Sec III we have explained that the elastic and exciting collisions can influence the dynamics of the plasma through a feed back of the profile relaxation on the ongoing instability. But they can not by themselves destabilize the stable orbital motion of the plasma once saturation of the instability is reached. In this section we will put this reasoning to the test and verify that these collisions are indeed incapable of exciting an instability on the plasma profile. For this purpose we loaded two uniform density cylindrical pure electron profiles with a very small $l=1$ perturbation seeded from the outset \textit{i.e.} the profiles' central axis were shifted from the trap's central axis by a very small distance of $0.1\times{R_W}$. The Brillouin ratios of the profiles chosen were $f_b=0.9$ and $f_b=0.02$. It is well understood that in the absence of any other influence these profiles will continue to make stable $l=1$ obits at their preset orbital radius with their characteristic $l=1$ mode frequency\cite{rdav1,megh}. In our simulations we loaded these profiles at the highest experimental background pressure used in Sec III \textit{i.e.} $P_{bg}=2\times10^{-6} Torr$ to ensure maximum exposure of the profiles to elastic and excitation collisions during the simulations. However as expected there was no destabilization of the stable orbital motion caused by the collisions for both the high and the low Brillouin ratios. Fig. 6 shows the squared and logscaled plot of the reading of the left potential probe for these two simulations. The signals have been plotted in such a manner for accurate exponential fitting of any observable growth of the signals (see also Fig. 3d) and also for the purpose of depicting on the same graph the two signals with vast difference in their mean values as well as their amplitudes. Fig. 6 shows that both the signals remain perfectly stable throughout the simulations.
Our simulation results also contradict a theoretical model by Davidson and Chao\cite{chao} for the effects of electron-neutral elastic collisions on the dynamics of the cloud. This model predicts that the non-ionzing collisions should destabilize the initially stable $l=1$ orbital motion in the above two simulations into a growing $l=1$ mode with growth rate, $\gamma=2{\pi}{\nu_{en}}{\omega_{0}}/{\omega_{ce}}$\cite{chao}, where ${\nu_{en}}$, ${\omega_{ce}}$, and ${\omega_{0}}$ are the non-ionizing collision frequency in $s^{-1}$, cyclotron frequency in $rad/s$ and characteristic frequency of the $l=1$ mode\cite{rdav1} in the loaded electron cloud expressed in $rad/s$. The actual theory considered only electron-neutral elastic collisions contributing to the collision frequency ${\nu_{en}}$ . In applying the model to our numerical experiments we have suitably included the effectively elastic-like excitation collisions along with the regular elastic collisions in estimating an experimental value of ${\nu_{en}}$. The theoretical model\cite{chao} also assumed a constant temperature of the plasma profile maintained by electron-electron collisions in the derivation of the above expression for the growth rate, $\gamma$ due to electron-neutral collisions. We do not have electron-electron collisions happening in our simulations. But the plasma is loaded at zero temperature and does not undergo any collsionless heating process, although there is some amount of heating by the non ionizing electron-neutral collisions. So applying the above expression for $\gamma$ to our numerical experiments will include the approximation of a constant temperature of the plasma profile.
The procedure for obtaining the growth rates from Davidson of Chao's\cite{chao} model when applied to our numerical experiments is as follows: From the initial cold loads of the plasma profiles it is simple to find an average velocity magnitude of electrons which we approximate as the average velocity magnitude all through the simulation. The total non-ionizing (elastic + excitation) collision cross-section can then be estimated using this average velocity magnitude, which leads to a value for the collision frequency ${\nu_{en}}$. This value comes out to be $2.3985\times10^4$ per sec for the $f_b=0.02$ load and $1.7548\times10^4$ per sec for the $f_b=0.9$ load. Plugging in the other fixed frequencies in the above expression for the growth rate, $\gamma$, we can estimate the growth rate from Davidson and Chao's model to be $1.8838\times10^2$ rad/sec for the $f_b=0.02$ load and $6.2021\times10^3$ rad/sec for the $f_b=0.9$ load.
These growth rates imply that at the end of the simulations period of $5.4\times10^{-5}$ seconds, the $l=1$ orbital radius should have increased by $1.02\%$ for the $f_b=0.02$ load and by $39.78\%$ for the $f_b=0.9$ load according to the model of Davidson and Chao\cite{chao}. Even if we neglect the theoretical small increase in orbital radius for the $f_b=0.02$ load, the theoretically predicted increase in orbtial radius is substantial for the $f_b=0.9$ load and should be detected by the simulation diagnostics. However as we have discussed earlier using the potential probe diagnostic (Fig 6) the stability and initial amplitude of the $l=1$ mode remains completely unaltered in both the simulations, $f_b=0.02$ and $f_b=0.9$. A new theoretical explanation in support of our experimental results and their contradiction with the model given by Davidson and Chao\cite{chao} is provided below.
The electron plasma in diocotron motion, \textit{e.g.} a stable $l=1$ orbit, can be thought of as having its electrostatic potential energy in two parts. The first part is the electrostatic energy of the profile without the surface wave/ diocotron mode and the other part which is a negative quantity is the electrostatic energy associated with the diocotron mode. Any process that further draws out energy from the mode \textit{i.e.} makes the mode's electrostatic energy even more negative and thereby also reduces the total potential energy of the profile, will bring about growth of the mode. A process that only reduces the first part of the potential energy \textit{i.e.} the profile's electrostatic energy by virtue of its internal configuration without any surface wave, can not cause growth of the dioctron mode. Through our simulations we found that the elastic and excitation collisions can only reduce this first part of the potential energy by means of profile relaxation. However these collisions can not directly influence the energy associated with diocotron mode. On the other hand a process like a resistive wall instability can directly takes away energy from the diocotron mode by changing surfaces charge configuration on the grounded wall (or alternatively the virtual image charge configuration) that interacts with the diocotron mode of the plasma. So while a resistive wall may cause growth of negative energy dioctron modes on the electron plasma the elastic and excitation collisions can not directly influence diocotron motion by virtue of their potential energy reduction capability. Hence contrary to the theory by Davidson and Chao\cite{chao} which predicts that elastic collisions can reduce both parts of the potential energy of an electron cloud and destabilize them, our simulations show that elastic (and excitation) collisions can only cause reduction in the first part of the potential energy through relaxation of the profile.
\section{Validation of the MCC numerical scheme}
\begin{table*}
\captionsetup{justification=raggedright,
singlelinecheck=false
}
\caption{\label{tab:table1} Cartesian velocity components and kinetic energies of arbitrarily chosen colliding electrons (pseudo electrons which are scaled down in mass inside the MCC routine) just before and just after the numerical collision operation: $v_{x0}$ , $v_{y0}$, and $v_{z0}$ are three components of the electron's velocity just before the collision, with $v_{z0}$ being the axial directions. $v_{x1}$ , $v_{y1}$, and $v_{z1}$ are the same velocity components just after the collision. ${\epsilon}_{k0}$ and ${\epsilon}_{k1}$ represent the kinetic energy of the electron just before and just after the collision. The time of collision, $t_{col}$ and the type of collision, $CT$ - ({\bf EL})astic or ({\bf EX})citation are also tabulated.}
\begin{ruledtabular}
\begin{tabular}{cccccccccc}
$v_{x0} (m/s)$ & $v_{y0} (m/s)$ & $v_{z0} (m/s)$ & ${\epsilon}_{k0} (eV)$ & $t_{col} ({\mu}s)$ & $CT$ & $v_{x1} (m/s)$ & $v_{y1} (m/s)$ &$v_{z1} (m/s)$ & ${\epsilon}_{k1} (eV)$\\
\hline
\\
\tiny{$-20251700.4777294$} & \tiny{$1175073.05198364$} & \tiny{$-10272700.0000000$} & \tiny{$1469.84998176886$} & \tiny{$0.6$} & \tiny{$EL$} & \tiny{$-19516351.0169898$} & \tiny{$2245227.82545585 $} & \tiny{$-11450392.4387800$} & \tiny{$1469.84986180629$}\\
\\
\tiny{$-40854485.1073843$} & \tiny{$13575274.2347969$} & \tiny{$10272700.0000000$} & \tiny{$5568.80333388360$} & \tiny{$27.6$} & \tiny{$EL$} & \tiny{$-39096793.4565210$} & \tiny{$4628854.69327025$} & \tiny{$20221704.5292680$} & \tiny{$ 5568.79622455448$}\\
\\
\tiny{$1316893.20280337$} & \tiny{$-26955331.0768088$} & \tiny{$10494034.6948208$} & \tiny{$2383.55397402810 $} & \tiny{$45.6$} & \tiny{$EL$} & \tiny{$710343.788248104 $} & \tiny{$-27151109.5770355$} & \tiny{$10038002.6573024 $} & \tiny{$ 2383.55395005079$}\\
\\
\\
\tiny{$5140270.17678625$} & \tiny{$-13367586.5916318$} & \tiny{$10272700.0000000$} & \tiny{$883.100541131483 $} & \tiny{$0.6$} & \tiny{$EX$} & \tiny{$7203428.86747464 $} & \tiny{$-13903403.3192488 $} & \tiny{$7834977.10927241 $} & \tiny{$871.550540940748$}\\
\\
\tiny{$365257.220980826$} & \tiny{$-4167743.83521545$} & \tiny{$-10272700.0000000$} & \tiny{$349.756692587647$} & \tiny{$27.6$} & \tiny{$EX$} & \tiny{$-2560205.78911724 $} & \tiny{$ -5110140.42010097 $} & \tiny{$-9289835.85222759 $} & \tiny{$338.206692396912$}\\
\\
\tiny{$18744576.1081227$} & \tiny{$-19513541.1212450$} & \tiny{$752752.461213745$} & \tiny{$2082.94122670152$} & \tiny{$45.6$} & \tiny{$EX$} & \tiny{$17956984.2621484 $} & \tiny{$-20049772.6621182$} & \tiny{$2048400.16132038$} & \tiny{$2071.39097050021$}\\
\end{tabular}
\end{ruledtabular}
\end{table*}
In order to verify the correctness and accuracy of our simulation results in Sec. III and IV we provide here a statistical and numerical validation of the MCC routine employed in these numerical experiments. The PIC part of the code is already very well benchmarked\cite{megh, megh2, megh3}. For the MCC routine validation procedure we have picked the PIC-with-MCC simulation of the $f_b=1.0$ ,$f=0.1$ equilibrium at $P_{bg}=2\times10^{-7} Torr$ described in Sec III.
We will now verify that the total number of elastic and excitation collisions simulated by the MCC routine in the simulation time period are correct numbers coherent with a real experiment with same initial conditions. For this purpose we will proceed to make a rough theoretical estimate of the number of collisions that should happen in course of the simulation. We already know that electron cloud is loaded with an equilibrium rigid rotation and a given initial magnitude of axial velocity. Taking the mean radial co-ordinate of electrons as half the radial extent of the cloud we can multiply this radius with the electron cloud's angular velocity to obtain an estimate of the average linear velocity of electrons in the plane of the PIC simulation. The magnitude of the average velocity in the plane of PIC simulation and the fixed initial magnitude of the axial component of velocity can be used to determine the average magnitude of resultant velocity of the loaded electron cloud. The initial average kinetic energy of electrons can also be calculated from this average speed of electrons.
Next we make a simplifying assumption that average kinetic energy of electrons remains near about its calculated initial value all through the simulation, or atleast its order of magnitude remains unchanged (see the plot of the total kinetic energy of electrons as a function of time in Fig. 4). So the initial average kinetic energy of electrons can be taken as rough estimate of the average kinetic energy throughout the simulation. With this approximate value of average kinetic energy of electrons we calculated the average cross sections for elastic and excitation collisions of electrons with neutrals from the available analytical formulas for collision cross sections. Using these mean collision cross sections and the average speed of electrons calculated earlier the average probability of elastic and excitation collision of an electron in the collision interval is calculated for the given background pressure. These probabilities multiplied by the total number of real electrons considered in the simulation give the average number of elastic and excitation collisions occurring in the collision interval. A further multiplication by the total number of collision intervals in the simulation time period gives an estimate of the total number of elastic collisions and excitation collisions that occurs in the simulation time period. These numbers come out to be $1.158\times10^{12}$ elastic collisions and $1.130\times10^{11}$ excitation collisions.
In our simulation we recorded the total number of elastic and excitation collisions for pseudo electrons in the simulation time period. Multiplying these numbers by the representation number for pseudo electrons gives us the total number of collisions by real electrons that were represented by the pseudo electron collisions. These numbers are obtained as $1.317\times10^{12}$ elastic collisions , $1.293\times10^{11}$ excitation collisions. So the collisions statics from simulation being in good agreement with corresponding theoretical estimates show that the MCC operates correctly in this aspect.
We will now test the correctness of the dynamics of collisions implemented by the MCC routine. For this purpose we randomly chose three elastic collisions and three excitation collisions occurring at different moments in course of the simulation. Recall that inside the MCC routine the pseudo electron is for all intents and purposes a real electron. Noting down the velocity components, and kinetic energy of the colliding electron prior to and after the collision in Table 1 we can check from the kinetic energy columns of Table 1 that for elastic collisions only a tiny fraction of the kinetic energy of the projectile electron gets transferred to the massive thermally moving neutral when they collide. For the excitation collision Table 1 shows that $11.55 eV$ of the projectile's kinetic energy is used up for excitation of the neutral and a tiny fraction of the remaining kinetic energy is transferred as kinetic energy to the neutral. Hence the dynamics of collisions are implemented correctly by the MCC. From the values of the velocity components before and after the collision we also see that electron is scattered by the collision which is the underlying cause of the profile relaxation induced by elastic and excitation collisions. The common value of projectile axial velocity $v_{z0}=\pm1.02727\times10^7$ m/s for some of the collisions is because these particles did not undergo any collisions before this and hence still carry the common loaded magnitude of axial velocity (recall that axial velocities can only be altered by collisions and not by the PIC).
\section{Conclusions and discussions}
In this paper we have addressed how the dynamics of a trapped electron cloud is influenced by elastic collisions with an inert neutral gas that is always present in an experimental trap, usually at very low pressures. Our simulations reveal the true nature of the interaction of the electron-neutral elastic collisions with the dynamics of the cloud. We have investigated how a ongoing collisionless ion-resonance instability due to ion impurities mixed with the cloud is dynamically influenced by the feedback of the electron cloud's collisional profile relaxation on the instability. The effect of feedback is visible in the energetics and growth rate of the instability. As this feedback is a highly dynamical nonlinear process it effects also do not follow any particular trends with increase/decrease of the background gas pressure. We also observed that the feedback can influence a growing ion resonance instability but can not alter the stable dynamics of the cloud once the instability has saturated. So if the period of growth of the instability is very short, then there may not be sufficient number of collisions in the short growth phase to influence the dynamics of the instability. Such a case of a quickly saturating ion resonance instability being unaltered by the elastic collisions was also shown in our simulations.
We have also demonstrated that contrary to an existing theory\cite{chao} on influence of elastic collisions between electrons and background neutrals, the collisions themselves can not destabilize an otherwise stable electron cloud. This, we have shown, is because of the nature of extraction of potential energy from the cloud by the collisions.
Now that we have built some understanding of the effects of elastic and excitation collisions of trapped electrons with background neutrals, we will explore in our next work\cite{megh4}, how ionizing collisions of electrons with background neutrals together with elastic and excitation collisions influence the dynamics of a trapped electron cloud.
\begin{acknowledgements}
The computer simulations described in this paper were performed on the UDBHAV cluster maintained by the Computer Centre of IPR, and the DAE grid UTKARSH maintained by BARC, Mumbai. We would like to thank staff of both these offices for keeping these clusters up and running at all times.
\end{acknowledgements}
\nocite{*}
\providecommand{\noopsort}[1]{}\providecommand{\singleletter}[1]{#1}%
| 9ce9f67d4831d7bf4123bc469134913bd97055ff | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Radio galaxies and quasars have radio-emitting structure that can be conveniently divided into two
categories: extended structure, the dimensions of which range
from $10^{3}$ pc to even $10^{6}$ pc, and compact structure,
with dimensions typically ranging from 1 pc to 100 pc \citep{kel88}.
Extragalactic radio sources with compact structure are used to
realize the fundamental Celestial Reference Frame with axis
stability at the level of ten microarcseconds ($\mu$as) by very
long baseline interferometry (VLBI) observations \citep{ma98, fey15}.
Given that the typical distance to these sources is at the level of $10^{9}$~pc, the
compact structure should have angular dimensions of
0.2--20 milliarcseconds (mas), as shown in images of
astrometric sources from astrophysical imaging studies
\citep[e.g.,][]{cha90a, ojh04, ojh05, pin07, lis09, cha10, lis13}. For example,
survey images of 91 compact sources obtained from VLBI observations at 5 GHz
by \citet{tay94} showed that only eight sources had a structure
smaller than one milliarcsecond. The effects of source structure on source
position determined from VLBI observations were studied and
demonstrated in a series of studies \citep[e.g.,][]{whi71, fey97, fey00,
fei03, mac07, mal08, moo11}. Recently, by observing four close radio
sources in the second
realization of the International Celestial Reference Frame (ICRF2)
for five times over one year, \citet{fom11} found that the
radio flux intensity maximum could follow a jet component rather
than stay close to the radio core. Their study suggests that if
the jet component gets fainter than the radio core or if they get completely
separated at some time, significant position variations will occur
at the level of 0.1 mas yr$^{-1}$ or even larger.
The study of the source structure effect on VLBI observables was
pioneered by \citet{tho80}. A significant
effort was made by \citet{cha90b}, who modeled the source structure
corrections for VLBI group delay and phase rate observables based on
the brightness distributions of the sources. Many studies then
attempted to introduce the theoretical model of the structure effect
into astrometric VLBI data analysis based on images of sources
\citep[e.g.,][]{cam88, cha88, tan88, ulv88, zep91, cha93, gon93, fey96, pet07}.
For example, \citet{sov02} applied it to a series of
ten of the Research and Development VLBI (RDV) sessions,
and the results showed that the weighted delay residuals could be reduced.
An example of the application of the theoretical
model to the European geodetic VLBI sessions was tested by
\citet{tor07}.
There are, however, several points that presently limit the
application of this model for the correction of the structure effect. First, the source
structure effect is very sensitive to a slight change in the brightness
distribution. Unfortunately, the time histories of available images for most
sources are quite sparse, and in the foreseeable future it is almost impossible
to make images on regular basis at intervals of much less than a year
for so many sources in the astrometric catalog unless astrometric/geodetic observations
themselves will be scheduled in a suitable way and sufficient efforts of making images
will be made. Secondly, even when images made several
months apart from each other are available, the stationary
reference point in these images can be hard to recognize
if the radio flux intensity maximum observed by VLBI is dominated
by a jet component. Consequently, in standard astrometric/geodetic VLBI data
analysis, the source structure effect has not actually been handled
so far. The source structure effect is still very important and
challenging for the astrometric VLBI, as shown in
simulation studies \citep{sha15, pla16}. If VLBI is to achieve its full potential
of the realization of the extragalactic Celestial Reference Frame with accuracy of the microarcsecond
level and that of the Terrestrial Reference Frame with accuracy of the millimeter level,
it is necessary to study and handle the source structure effect more effectively based on
the astrometric observations themselves.
These are the purposes of this paper.
In this paper we perform an initial analysis to determine how well
source structure can be determined directly from the geodetic VLBI
observables themselves \footnote{\added{Geodetic/astrometric VLBI observables are the baseline-based group delays and phase rates determined per scan within a geodetic VLBI experiment. The International VLBI Service for Geodesy and Astrometry (IVS) coordinates archives of geodetic VLBI experiment observables (see http://ivscc.gsfc.nasa.gov/products-data/data.html), but visibility datasets are normally not made available for analysis.}}. We aim to develop an alternative method for
studying the structure effect that should be simple, easy to
implement, and applicable for general historical and future geodetic
VLBI observations, including many of the oldest observations (back to
the 1970s) for which the visibility datasets are no longer available.
Although a self-calibration and Fourier imaging analysis of the
visibility data can give superior results for determining source
structure, that approach is time and computing resource intensive, it
requires large amounts of software not currently implemented in
geodetic analysis packages, it requires that the observations were
conducted in a manner suitable for imaging, which is frequently not
the case for historical geodetic VLBI sessions, and it will be
difficult, yield sub-optimal results, or even be impossible for the
historical experiments that no longer have archived visibility
datasets.
Therefore we defer our structure analysis based on imaging for a
future publication, and
we make use of the closure delay, the sum of the delays around a closed
loop of baselines, as a new observable and propose
a method to use this new observable for the determination of
the source structure effect on the astrometric VLBI observable.
We calculate the closure delays, investigate the
characteristics of the source structure, and
then solve for the source structure effect on each observable.
The source structure can be finally
obtained and the source structure effect can be determined.
The source 0642+449, one of the ICRF2 defining
sources, is selected as a demonstration case for this method.
The systematical analysis of closure delay requires a consistent definition and
a careful discussion of closure delay,
which are presented along with its calculating model
in Section 2. The data used here, the CONT14 observations, and the overall
statistics of the closure delay of source 0642+449 are
introduced in Section 3. Section 4 discusses the method that was
used to solve for the source structure effect on each observable
based on the knowledge from Section 3. The results, describing
the structure of this source, are shown in Section 5,
and the final model is presented in Section 6. Conclusions and discussion are given
in the last Section.
\section{Model of closure delay}
The closure phase,
unaffected by instrumental and atmospheric instabilities,
was recognized as a good observable for the study of source structure
first by \citet{jen58} and later by \citet{rog74}.
A method for
recovering the brightness distributions of compact radio sources
from VLBI observations of closure phase,
together with the measured visibility amplitudes, was developed to make images
at the scale of milliarcseconds \citep{rea78} and
used to obtain valuable maps for 45 objects by \citet{pea81, pea88}
in the early stage of VLBI imaging. The methods using the closure phase
for recovering the brightness distribution of a radio source were called \emph{hybrid-mapping based on closure quantities},
which was widely applied for the study of source structure \citep[see][and references therein]{pea84}.
In terms of independence from instrumental and atmospheric instabilities and
sensitivity to source structure,
closure delay has similar characteristics to closure phase.
The closure delay was used for evaluating the performance of the structure models
in the past \citep{cha90b}, but our method will use it in a more direct way.
For the first time, the definition and the model of closure delay is given here.
For a triangle of 3 stations, $a$, $b$, and $c$, the closure delay is defined for an individual
wavefront by
\begin{equation}
\label{eq_closure}
\tau_{abc}\equiv\tau_{ab}+\tau_{bc}+\tau_{ca},
\end{equation}
where, for instance, $\tau_{ab}$ is the delay observable from station $a$ to
station $b$, and $\tau_{bc}$ is the delay observable from station $b$ to
station $c$, for the same wavefront received by three stations.
Equation (\ref{eq_closure}) depicts the scenario in which
the same wavefront passes stations $a$, $b$, and $c$ in some arbitrary
sequence. The closure delay, for an ideal point source, is independent of any
station-based or source-based effect and should be zero if there is
no observational noise. For sources with detectable structure, as expected
even for most astrometric sources, the group delay observable depends on the
observing frequency and the baseline length and orientation differently
from the group delay behavior for a point source. In general, the change in
the group delay will be different for each baseline in the closure delay triangle
dependent upon the source structure, and the resulting closure delay will be non-zero.
Therefore, variations in the closure delays sufficiently far away from zero that they are
unlikely to be caused by random measurement noise should, in principle, only be observed
for sources with significant structure.
In geodetic VLBI measurements, by convention, the time tag of group delay, phase, and phase rate observables is
referred to the epoch when the wavefront passes the first station in the
baseline. They are implemented in the VLBI processor such that all observables in one scan have
exactly the same time tag value. This means that the observables in
one scan are usually not related to the same wavefront and allowance
must be made for the delay of the arrival of the same wavefront at the different stations.
Moreover, the baseline name is made up of two stations' names in the alphabetical order,
so that $\tau_{ca}$
should be replaced with the actual geodetic VLBI observables on that baseline, $-\tau_{ac}$.
Therefore, to the accuracy of the
second order in delay, the closure delay at reference epoch $t$ is calculated from
\begin{equation}
\label{eq_closure2}
\tau_{abc}(t)=\tau_{ab}(t)+[\tau_{bc}(t)+\dot{\tau}_{bc}(t)\cdotp{\tau^{\prime}_{ab}(t)}+\frac{1}{2}\ddot{\tau}_{bc}(t)\cdotp{{\tau^{\prime}_{ab}(t)}^2}]-\tau_{ac}(t),
\end{equation}
where a prime on a delay symbol indicates an absence of dependence
on station clock offset, that is, referring to the geometric delay, and
a superposed dot and double superposed dots denote differentiation
with respect to time once and twice, respectively.
For a goal of 1 ps accuracy of the closure delay,
the third term inside the bracket of Equation (\ref{eq_closure2})
is smaller than 0.1 ps and is therefore negligible,
since for ground-based VLBI the magnitude of the group delay is at the level of 0.02~s and that of the second order derivative is at the
level of 10~ps~s$^{-2}$. Since the
magnitude of the first order derivative is at the level of 1~$\mu$s~s$^{-1}$, the
second term inside the bracket has the magnitude of about ten nanoseconds (ns) and should be
calculated as accurately as possible. The $\dot{\tau}_{bc}(t)$ term can be
calculated from the theoretical model, which may have to ignore the
rate of change of the propagation delay; it can be derived more accurately from
the phase rate observable and the ionospheric delay rate,
both of which are available for International VLBI Service for
Geodesy and Astrometry \citep[IVS,][]{sch12} VLBI observations. All clock offsets or
jumps larger than 0.1 $\mu$s should be taken into account for the
geometric delay, $\tau^{\prime}_{ab}(t)$, in Equation (\ref{eq_closure2}) if it is calculated
from the group delay observable.
Following the conventions in geodetic VLBI measurements of the time tag and of
the baseline's name, the closure delay in this paper will be
referred to the same epoch as that of the three observables in the
triangle and be labeled by the names of the three stations in
alphabetical order. In this case, the baseline between the
first station and the third station in a triangle will always
contribute to the closure delay calculated from Equation
(\ref{eq_closure2}), in contrast to other two
baselines, with a negative sign.
\section{Closure delay of source 0642+449}
The data from CONT14\footnote{http://ivscc.gsfc.nasa.gov/program/cont14/}
observations \citep{not15} at X band were used.
CONT14 is a campaign of continuous VLBI observations, conducted by the IVS from
00:00:00 UT on 2014 May 6 to 00:00:00 UT on 2014 May 21. It is a
continuation of the series of continuous VLBI observations over 15 days that were
observed every three years since 2002. With a network of ten
stations in the northern hemisphere and seven in the southern
hemisphere at sixteen sites\footnote{Two stations, HOBART12 and HOBART26,
are located at the Hobart site in Australia.}, this campaign was intended to
acquire state-of-the-art VLBI data over a time period of 15 days
with the highest accuracy that the then existing VLBI
system was capable. In these continuous observations there are
several sources with about one thousand observables per day, which makes
the CONT14 observations considerably valuable for the study of
source structure.
\begin{figure}
\begin{center}
\includegraphics[width=0.48\textwidth]{net_0642}
\caption{The baseline lengths of the observing network of source 0642+449 in CONT14. There were in total 15 stations
observing this source, and 96 baselines were correlated.}
\label{base_leng}
\end{center}
\end{figure}
The radio source 0642+449 was
selected as the target of our study. It was observed in 512 scans
by 15 stations in the CONT14 campaign (only the two southernmost
stations at the Hobart site could not observe this source), and has in total 11~027 pairs of
usable (quality codes\footnote{http://lupus.gsfc.nasa.gov/global/ngs-doc.html}
smaller than 8) group delay and delay rate
observables at X band. In total, 22~154 closure delays were
calculated from this data set and provided important statistics of
the performance of this individual source. As the number of triangles
is about two times that of group delay observables, each individual
group delay on average is involved in six triangles. The resulting
triangles have 350 different kinds of geometry. There are in total
96 baselines with lengths ranging from 900 km to 12~600~km as shown in
Figure \ref{base_leng}. This widely-spanned baseline length provides
a good $uv$ coverage, eventually facilitating the detection of the source structure
at a variety of scales.
\begin{figure}
\begin{center}
\includegraphics[width=0.48\textwidth]{numb0642}
\caption{The absolute magnitude distribution of closure delay of the source 0642+449.
The subplot $a$ shows the distribution of closure delay in
the whole range from 0 ns to 28 ns with the bin width of 1 ns,
and the subplot $b$ shows, with a smaller bin width of 20 ps to
demonstrate more detail, the distribution in the main range from 0 ns to 1 ns
which has been out of the axis limit as one bin in the subplot $b$.}
\label{nemb0642}
\end{center}
\end{figure}
Multiband group delay ambiguities of 100 ns or 50 ns, determined by inspecting the closure
delay, were fixed for three group delays
for the baseline WESTFORD--YARRA12M on the 6th, 10th, and 11th of May. After this change to the CONT14 observations, the 22~154 closure
delays were then recalculated. Six closure delays
with absolute value larger than 28 ns were subsequently
excluded as outliers in the analysis here. No other changes to the data
were made and no additional points were excluded from analysis.
Figure \ref{nemb0642} shows the
distribution of the absolute magnitudes of the 22~148 closure delays.
Approximately 49$\%$ of the triangles have closure
delays of absolute value smaller than 20 ps, and 67$\%$ of triangles have closure delays of that
smaller than 40 ps. As Figure \ref{nemb0642} shows,
there is a rather flat distribution in each of the subplots: the one
in subplot $a$ lies in the range larger than
1 ns containing 759 closure delays; and the one in subplot $b$ lies
in the range of 0.1 ns to 0.9 ns, which
has $16\%$ of triangles. The mean value and the
standard deviation of closure delays for the whole set of triangles is
$-$0.02 ns and 2.668 ns, respectively, and 0.3~ps and 139 ps, respectively,
for the set in the range 0 to 1.0 ns. The statistics and the
distribution of closure delay demonstrate that source 0642+449 to
some extent performs well for a geodetic source showing closure delays of less than 20 ps for half of the triangles,
but it is not as compact as a point-like source.
Furthermore,
there are about 759 closure delay magnitudes above 1~ns, suggesting that about
127 group delays (based on the average number of triangles each single
group delay is part of),
about 1~\% of the total observables for this source, sense exceptionally large structure effects.
For a comparison to demonstrate the measurement noise in VLBI observables, the
standard deviations of closure delays for unresolved sources, such as 0016+731 and 0727-115,
were calculated as well. Source 0016+731 has about 23~300 closure delays and 0727-115 has about
11~200 closure delays. The standard
deviation for source 0016+731, which showed
a little resolved structure, is about 11 ps, and that for source 0727-115 is about 8.0 ps.
Figure \ref{s0727} shows the closure delay distribution over 15 days.
Geodetic VLBI observations are scheduled with
different observing durations for each station and each expected source brightness to achieve
a uniform signal to noise ratio for all observations, so we assume
that the 8.0/1.7=4.6~ps standard deviation for $0727-115$ represents the upper
limit to the typical measurement noise for all sources in CONT14 observations, and the vast
majority of the 139~ps scatter in the $0642+449$ closure delays cannot be
explained by the measurement noise.
\begin{figure}
\begin{center}
\includegraphics[width=0.47\textwidth]{0727-115}
\caption{The closure delays of 0727-115 over 15 days. There is absolutely no triangle with closure delay of magnitude larger than 70 ps, and most triangles are with closure delays of magnitude smaller than 30 ps.}
\label{s0727}
\end{center}
\end{figure}
\subsection{Triangles with three short baselines}
The closure delays of triangles with all three baseline lengths\footnote{Total baseline length is used
throughout the paper as a proxy for projected baseline length.}
smaller than 7100 km were investigated. In Figure
\ref{small_net1}, the closure delays of these triangles are shown as
red circles connected by dashed lines. The time in hours on the X
axis is the GMST time of the observation
for each triangle epoch to show observations over the fifteen days in an
overlapping 24-hour plot. The same technique is applied in the following figures of closure delays.
There are in total 6611 closure delays for these
small triangles, and the standard deviation is about 21.7~ps.
Figure \ref{small_net2} geographically shows these small triangles\footnote{We should notice that there are more short baselines
than shown in Figure \ref{small_net2}, since there are some
isolated short baselines unable to form such a small triangle.}.
The magnitudes of the
closure delays of these triangles are mostly smaller than 40 ps.
There are 24 closure delays with magnitudes
larger than 0.1 ns, and if we constrain the small
triangles to having baseline lengths shorter than 7000 km,
the magnitudes of the closure delay are all smaller than 0.1~ns.
\begin{figure}
\begin{center}
\includegraphics[width=0.50\textwidth]{s0642_s52}
\caption{The closure delays of triangles with all three baselines
shorter than 7100 km, which in total have 42 different geometries. All points are connected by dashed lines to
show whether there are points out of the chosen axis limit. This is applied to all figures that show the closure delay or
the source structure effect in the paper. For this case, not a single closure
delay is out of the axis limit. }
\label{small_net1}
\end{center}
\end{figure}
\begin{figure}
\begin{center}
\includegraphics[width=0.48\textwidth]{s0642_s52_net}
\caption{The triangles with all three baselines shorter than 7100 km.}
\label{small_net2}
\end{center}
\end{figure}
\subsection{Triangles with one long and common baseline}
The closure delays of triangles that contain one long baseline and
two short baselines were investigated. In the CONT14 network, six stations are densely
located in the European region so that one can make use of them to
construct a good set of triangles for this investigation. Figure
\ref{small_net2} suggests that the long baseline should be BADARY--WESTFORD,
since the European region locates almost in the middle of
it. The length of baseline BADARY--WESTFORD is approximately 8700
km. ZELENCHK connects to BADARY with the baseline length of
approximately 4400 km, but to WESTFORD about 7800 km, a little longer than 7100 km;
the other five European stations connect to both
with baselines shorter than 7100 km.
Figure \ref{s0642_bawe} shows the closure delays with respect to the GMST time of the resulting six kinds of
geometry using six colors. Figure \ref{s0642_bawe_net}
shows the geographic positions of these eight stations and the
triangles with these six different geometries.
Aside from the points outside the $Y$ axis limits, these
six sets of closure delays roughly have the same pattern, two peaks
happening around 03:00 and 15:00 with magnitude of approximately 0.5--0.7
ns. When WESTFORD is the third station in the triangle, the sign of
the peaks is opposite to that when WESTFORD is the second\footnote{By the naming
convention, in this case, WESTFORD can never be the first station in a baseline, and BADARY is always the first station.}.
Considering that baseline BADARY--WESTFORD is the only common one in these triangles and that triangles
with baselines shorter than 7100~km show no closure delays larger in magnitude than 0.2~ns,
it is reasonable and obvious to conclude that the systematic
variation in the closure delays of these six kinds of
geometry is predominately related to the baseline BADARY--WESTFORD.
In addition, the closure delays of the nanosecond order and
larger, which are out of the axis limit in the plot, only happened
around the same GMST time of the peaks.
\begin{figure}[tb]
\begin{center}
\includegraphics[width=0.49\textwidth]{s0642_bawe}
\caption{The closure delays of triangles with BADARY and WESTFORD and one of the six European stations.
Each color corresponds to the closure delays of triangles of one kind of geometry, with the name shown on the upper right corner in the same color
for the European station in that triangle. Points for the triangles with the same
geometry are connected with dashed lines.}
\label{s0642_bawe}
\end{center}
\end{figure}
\begin{figure}[tb]
\begin{center}
\includegraphics[width=0.48\textwidth]{s0642_bawe_net}
\caption{The geographic locations of BADARY, WESTFORD, and six central European stations. The black line
is BADARY--WESTFORD; the six pairs of baselines connecting the stations BADARY and
WESTFORD are shown with the same colors with that of the closure delays that they contribute to in Figure \ref{s0642_bawe}. }
\label{s0642_bawe_net}
\end{center}
\end{figure}
\subsection{Triangles with the shortest baseline}
As we know, the atmospheric and clock errors limit our ability to examine the structure effect on a single baseline on its own.
The investigation in Section 3.2 above demonstrates the effect of source structure
introduced by one long baseline by using two short baselines in the closure triangle
to cancel out the atmospheric errors, clock errors, and so on, in the long baseline.
The closure delays of all triangles including the smallest
baseline in the CONT14 network, ONSALA60--WETTZELL, were
investigated. They are shown in Figure \ref{s0642_onwe_2} by two
subplots with different scales. As there were 13 stations available
to form a triangle with baseline ONSALA60--WETTZELL, we have 13 different
kinds of geometry for this case. The closure delays were
classified and sorted by the geometry of each triangle. In this figure, the names of 13
different geometries are presented in the numerical order of
the longest baseline length in the geometry, and the colors for the
names of the 13 stations correspond to those of the closure delay in
each of the subplots. Most closure delays of these triangles are very
small, at the level of tens of picoseconds, if the longest baseline in the
triangle is shorter than 7600 km. The magnitudes of the closure
delays become larger as the baseline lengths in the triangles continue
to increase, and eventually systematic variations appear. In this
case, the variations are much more complicated than that for
triangles with only one long baseline as shown in Figure
\ref{s0642_bawe}. Closure delays larger than 1 ns appear only when
there are long baselines in the triangle and where apparent systematic
variations for short baseline triangles occurred.
\begin{figure}
\begin{center}
\includegraphics[width=0.48\textwidth]{s0642_onwe_2}
\caption{The closure delays of all triangles that contain the baseline ONSALA60--WETTZELL.
Since 15 stations observed this source, there are 13 station available to join these two stations to construct a triangle.
Subplot a shows the closure delays with seven different third station in different colors, while
subplot b shows with a larger axis limit, closure delays from triangles using the remaining six stations. The names of 13 different kinds
of geometry in both subplots are presented in the numerical order of their longest baseline lengths. For each geometry,
the name of the third station is presented in the color of its closure delay. }
\label{s0642_onwe_2}
\end{center}
\end{figure}
\section{Method to solve triangles}
As discussed in Section 2, the measured closure delay is due to the observational
noise and the source structure effect that is baseline
dependent. Then
\begin{equation}
\label{eq_closure3}
\tau_{abc}=\delta \tau_{ab}+\delta \tau_{bc}- \delta \tau_{ac} + \sum_{i=1}^{3} \epsilon_{i},
\end{equation}
where $\delta \tau_{ab}$, $\delta \tau_{bc}$, and $\delta \tau_{ac}$
are the source structure effects on the three baselines, and
$\epsilon_{i}$ is the measurement noise in the $i$-th baseline
of that triangle. It is reasonable to assume here that
measurement noises are random at the level of a few picoseconds, which has been demonstrated
by the closure delays of unresolved sources.
In an array of $N$ stations, there are at most $N(N-1)/2$ baselines
and $N(N-1)(N-2)/6$ closure delay relations. But only $(N-1)(N-2)/2$ of these
relations are independent \citep{{pea84}}. There are therefore $(N-1)$ too few closure delays to
determine the source structure effect for each standard baseline-delay observable, and
an independent estimate of the structure delay on $(N-1)$ baselines has to be derived.
The investigations in Section 3 give insights to this and
the proposed method uses the assumption that source 0642+449 is
a point-like source with respect to the baselines shorter than a
certain value. We then choose 7100 km as the threshold and
select a necessary and minimum number of $<$ 7100 km
baselines to connect as many stations as possible in each scan.
The structure effects on these selected baselines are assumed to be zero. Taking
a fifteen-station array as an example, ideally there are
fourteen baselines shorter than 7100 km connecting fifteen
stations as a complete connection shown in Figure \ref{s0642_datum}.
Setting the source structure effect on
these selected baselines to be zero allows us to solve for the structure effects
on other baselines utilizing the closure delays.
\begin{figure}[tb]
\begin{center}
\includegraphics[width=0.46\textwidth]{s0642_datum}
\caption{An example of a complete connection of the fifteen-station network by a minimal number
of $<$ 7100 km long baselines. To solve for the baseline delays using closure delay in one scan, these selected
baselines are set to have zero source structure effect in that scan.
This kind of connection and its role is designated as connection in the paper.}
\label{s0642_datum}
\end{center}
\end{figure}
To find such a kind of a complete connection could be very challenging in some
cases. This is the case for the southern stations. Three southern
stations, KATH12M, YARRA12M, and WARK12M, can be connected to the
whole network through such a connection only if at least one of the
baselines KOKEE--WARK12M (6600 km) and KATH12M--TSUKUB32
(5500 km) is available. If both of these two baselines get lost in a
scan, it is thus impossible
to solve all the big triangles related to these three
stations. This is also one of the reasons for choosing 7100 km
as the threshold for the baseline length of the connection, considering
that the shortest baselines related to station HART15M, are
HART15M--MATERA (7033 km) and FORTLEZA--HART15M (7025 km).
Next step is to solve as many triangles as possible based on the
connection with the threshold of 7100 km scan by scan.
\section{Results of source structure effect}
Finally, 16~941 triangles were solved and the source structure effects
on 8492 baseline delays were determined based on Equation
(\ref{eq_closure3}). In this solution, 2179
observables of short ($<$ 7100 km long) baselines were
assumed to have no source structure effect to build up the connection. Figure \ref{uv0642_all} shows the
$uv$ coverage with color marking the magnitude of the derived source
structure effect on each point. In general, the source structure has
a strong effect on long baselines at two opposite directions of approximately the $u$ axis.
The results are divided into two groups for the
detailed study of learning the source structure effect.
\begin{figure}[tb]
\begin{center}
\includegraphics[width=0.50\textwidth]{uv0642_all}
\caption{Coverage of the Fourier domain ($uv$ coverage) of the CONT14 observations of 0642+449 at
8.4 GHz ($\lambda$ = 3.6 cm), plotted in units of Mega lambda. Color marks the absolute magnitude of
the source structure effect on each of 8492 observables
derived from closure delays by the method introduced in Section 4. }
\label{uv0642_all}
\end{center}
\end{figure}
First, apart from the observables selected to form the connection, the structure
effects on the remaining 3443 observables of short baselines were
estimated along with long baselines in the solution. The
estimated source structure effects on these baselines with respect to $uv$ position angle of baseline in the $uv$ plane
(measured North-through-East) are shown in
Figure \ref{s0642_s71}.
All these observables without exception have
very small source structure effects, and the standard deviation is
about 22.7 ps, which is close to that of the closure delays of
triangles with short baselines shown in Figure \ref{small_net1}. It
may be slightly larger than that of random measurement noises in
VLBI measurements, partly because the measurement noise in 2179
zeroed observables were propagated into the estimated ones.
There is no point larger than 0.2 ns. Therefore, the result
supports the assumption of no significant structure on short baselines
that has been used for the connection.
\begin{figure}[tb]
\begin{center}
\clearpage
\includegraphics[width=0.48\textwidth]{s0642_s71}
\caption{Derived source structure effects on 3443 observables of short baselines ($<$ 7100 km) from the
closure delays, plotted as a function of the $uv$ position angle of baseline. }
\label{s0642_s71}
\end{center}
\end{figure}
Secondly, the source structure effects for eight long baselines with
significant variations are presented one by one in Figure \ref{s0642_model} in blue dots.
The patterns shown in these plots are incomplete over one circle of $uv$ position angle,
especially for the longest baselines, due to both the loss of common
visibility of the source and the vulnerability of losing link to
the connection. This indeed makes the identification of the source
structure extremely challenging. However, it is beneficial to
combine together the patterns in these plots. Two peaks for the
baselines with length from 8400 km to 9600 km can be
identified, appearing at the $uv$ position angles of approximately 90$\degr$
and 270$\degr$. The peak structure effect for these baselines is 0.7~ns.
Moreover, there are two characteristics of the source structure that can be identified.
First, apart from the points outside of the axis limit, the scatter of
the structure effect is rather small along the track of the
variation curve, and this should imply that the source has multiple
point-like components rather than a flat brightness distribution
over the extended structure. Second, the variation is symmetric over
the $uv$ position angle of one circle, and the separation between the two
peaks is 180$\degr$. As a result, the source should have a symmetric
brightness distribution, for example, the multiple components are in
a straight line. Finally, points out of the axis limit happen exclusively
on baselines longer than 9000~km where significant source
structure effects occur.
\begin{figure*}
\begin{center}
\includegraphics[width=0.45\textwidth]{s0642_modelTSUWET}
\includegraphics[width=0.45\textwidth]{s0642_modelBADWES}
\includegraphics[width=0.45\textwidth]{s0642_modelMATTSU}
\includegraphics[width=0.45\textwidth]{s0642_modelTSUWES}
\includegraphics[width=0.45\textwidth]{s0642_modelTSUYEB}
\includegraphics[width=0.45\textwidth]{s0642_modelKOKWET}
\includegraphics[width=0.45\textwidth]{s0642_modelKATYEB}
\includegraphics[width=0.45\textwidth]{s0642_modelKATWES}
\caption{Derived source structure effects from the closure delay and the pattern of structure effects calculated from the two-component model as a function of the $uv$ position angle for eight baselines with significant variations. The baseline name along with the baseline length in brackets is shown for each of the plots. The blue dots correspond to the observation points, and the complete patterns of structure effects over one circle calculated from the model are shown in red dash lines.}
\label{s0642_model}
\end{center}
\end{figure*}
\section{Modeling the source structure of 0642+449} \label{sec:method}
The examples of a double-component source and a triple-component
source have been analyzed by \citet{tho80}, and a double-component
source has been studied in great detail by some simulations of
\citet{cha90b}. Based on these studies, we can determine that the
source exhibits a structure with two compact components on
baselines with length of around 9000 km. The
ratio of the flux densities of the weaker component to
the stronger one, $K$, is about 0.92, since the peak structure effect
for these baselines is approximately 0.7 ns.
\citet{tho80} referred the source position in his model to the
middle point of the separation between the two components, while
\citet{cha90b} took the centroid point of brightness as the
reference point. The reference point does not matter for the closure
delay, but we apply the centroid model here, because
we assume that observables of short baselines have zero source
structure effect, and in practice they are referred very closely to
the center of brightness. According to this model, the structure
delay $\tau_s$ on baseline $\mathbf{B}$ is given as
\begin{equation}
\label{eq_structure}
\tau_{s}=\frac{K(1-K)[1-cos(2 \pi R)]R}{f(1+K)[K^{2}+2Kcos(2\pi R)+1]},
\end{equation}
where $f$ is the observing frequency, $R = \mathbf{B} \cdot \mathbf{S_{12}}/\lambda$, $\lambda$ is the wavelength, and
$\mathbf{S_{12}}$ is the relative position vector in the $uv$ plane of the weaker
component $P_2$ with respect to the stronger one $P_1$. This
expression only accounts for the change of the projected baseline on the $uv$ plane, i.e., the sky fringes.
It assumes that the flux densities of two components do not
change with frequency and observations are made on an infinitesimal bandwidth.
Since the closure delay is not
sensitive to the reference point in the source structure
at all, this study lets all observables refer
to an unknown point, in this case to the centroid of brightness by the choice of the model, per scan.
This reference point will finally be determined through VLBI data analysis.
We further assume that the source structure does not change within the fifteen days of observations.
A ``guess'' estimation for the unknown parameters in Equation (\ref{eq_structure}) was made from the results
of baselines with lengths from 8400~km to 9600~km derived in Section 5.
The final estimation was done
by model-fitting all the closure delays with magnitude smaller than 1.0 ns
based on the a priori values from the guess estimation.
The flux-density ratio is then estimated to be 0.916 $\pm$ 0.012, and the relative position
vector be $-$426 $\pm$ 12 $\mu$as in right ascension
and $-$66 $\pm$ 19 $\mu$as in declination. According to the
detected morphology, a baseline with length of
7100~km has $R =$ 0.41 and the peak structure effect is only 13~ps, which explains the
foundation of the assumption short baselines have no structure effect used
for the connection.
One can easily compute the source structure effect for each
observable based on Equation (\ref{eq_structure}) and the two-component model.
The results for the eight baselines are shown in Figure
\ref{s0642_model} in red dash lines. By comparison, we found
that the model fits all the baselines with length smaller than 10~000~km,
but it does not fit so well for some of the longest baselines, for
example baseline KATH12M--WESTFORD shown in Figure
\ref{s0642_model}, where the model and the estimated
structure effect just have different variations. After using the
model to correct the structure effect, the standard
deviation of the closure delay was decreased from 139 ps to
90 ps for all triangles with closure delay less than 1 ns, which
is a significant improvement.
\begin{figure}[tb]
\begin{center}
\includegraphics[width=0.48\textwidth]{uv0642_angle}
\caption{Distribution of the 117 observables with contribution to the closure delay larger than 1 ns. }
\label{uv0642_angle}
\end{center}
\end{figure}
\begin{table}
\begin{center}
\caption{Statistics of the 117 observables with contribution to the closure delay larger than 1 ns.} \label{table:1} \centering
\begin{tabular}{lrr}
\hline\hline Baseline & Number & Length (km)\\
\hline
FORTLEZA--KOKEE & 1 & 11063 \\
HART15M--YARRA12M & 1 & 7848 \\
KATH12M--NYALES20 & 1 & 10410 \\
KOKEE--WESTFORD & 1 & 7676 \\
WESTFORD--WARK12M & 1 & 11534 \\
WESTFORD--YARRA12M & 1 & 12638 \\
FORTLEZA--ZELENCHK & 3 & 8649 \\
BADARY--FORTLEZA & 4 & 11154 \\
FORTLEZA--TSUKUB32 & 4 & 12252 \\
KATH12M--ONSALA60 & 4 & 10928 \\
KATH12M--WETTZELL & 4 & 11026 \\
TSUKUB32--WESTFORD & 4 & 9505 \\
BADARY--WESTFORD & 5 & 8671 \\
KATH12M--WESTFORD & 5 & 12164 \\
KATH12M--YEBES40M & 5 & 11697 \\
KOKEE--YEBES40M & 5 & 10687 \\
KATH12M--MATERA & 6 & 10953 \\
KOKEE--MATERA & 6 & 10894 \\
KOKEE--WETTZELL & 9 & 10357 \\
KOKEE--ONSALA60 & 15 & 9792 \\
KOKEE--YARRA12M & 32 & 9498 \\
\hline\hline
\end{tabular}
\end{center}
\end{table}
Another solution was done to trace the observables that cause 749
triangles to have closure delays larger than 1 ns. It is much
easier to locate these observables actually, since the magnitude is
so large that all triangles related to them apparently have quite
large closure delays. The threshold of baseline length was then
reset to a little larger than the previous one, 8000 km for the
connection. The baseline with a length within this threshold should have a peak
structure effect of 0.12 ns, which will
corrupt the determination of the source structure
effect for many observables, but facilitate locating the points
with large contribution to the closure delay. This solution allows
21~700 triangles to be solved, and 117 observables have been
identified with large contributions to the closure delay. Figure
\ref{uv0642_angle} shows the estimated closure delays of these 117
observables as a function of the $uv$ position angle. Table \ref{table:1}
shows the baselines lengths of these observables. All these points
systematically lie in three ranges of $uv$ position angle, 30$\degr$--60$\degr$, 220$\degr$--250$\degr$,
and 280$\degr$--310$\degr$, and refer to baselines
with lengths larger than 7600 km, most of them larger than 9000~km. It leads us to conclude that these large closure delays are
more likely caused by the source structure as well. It is worth
noting that except for the baseline clock offset used to absorb
constant biases between baselines, no other parameters in VLBI data
analysis can absorb any part of the closure delay, which will finally impact the residuals. The
observables with a large contribution to the closure delay
are always identified as
outliers during the data analysis. The study here shows that these
observables are not outliers but are systematically affected by source structure.
The limitations of the application of equation (\ref{eq_structure}) to the multiband
group delay observable
need to be discussed here. Finite bandwidth for multiband group delay contributes to
errors in the modeled structure effects based on equation~(\ref{eq_structure})
because the structure changes with frequency and/or the
change in resolution $R$ as a function with frequency cause
$\tau_s$ to change with frequency. The X~band
observations in CONT14 use 8 channels spread over 720~MHz, for
a fractional bandwidth of 0.085. We have performed numerical
simulations covering a large range of component separations,
baseline lengths and orientations, and K values to study this
effect. For the case of $0642+449$, with $K \sim 0.9$ and the
separation about 0.5~mas, we find a median absolute deviation
between $\tau_s$ computed at the top and bottom of the
observed X~band frequencies, over the observed baseline
lengths and orientations, to be 2.6~fs, and the RMS deviation
to be 11~ps, for two components with the same spectral index.
With one component having a spectral index of 0 (typical for
an AGN core) and the other component having a spectral index
of $-0.7$ (typical for an optically thin jet component), the
median absolute deviation and RMS are 40~fs and 14~ps,
respectively. Hence, the finite bandwidth of the CONT14
observations analyzed here does not present a significant
violation of the infinitesimal bandwidth assumption of
equation~(\ref{eq_structure}) for this source. Furthermore, our simulations
show that having reasonable different spectral indices for the
two components contributes significantly less to differences
in the structure delay with frequency than the changes in
interferometer resolution ($R$) as a function of frequency
(14~ps for spectral index and $R$ changes together, compared
to 11~ps for $R$ changes alone).
\section{Conclusion and discussion}
Closure delay analysis has several advantages for geodetic VLBI. First, it
directly studies structure effects from the geodetic observables---the
multiband group delays---themselves, the quantities actually
used to determine station position and motions, Earth
orientation parameters, and other astrometric/geometric parameters.
Second, closure delay analysis can serve as
an indicator to evaluate the performance of any structure model,
whether that model is determined by fitting closure delays or through
an imaging analysis. Third, model fitting of closure delays can
identify the strongest components of a source while ignoring weak
components that do not significantly alter the group delay,
simplifying the analysis procedure. In contrast to many astrophysical
studies where the weak components are of interest, geodetic VLBI
analysis can often ignore such components. Fourth, a closure delay
analysis can save time and effort in both processing and software
development for geodetic purposes compared to standard VLBI imaging.
We showed that closure delays can determine the magnitude of the
measurement noise in the geodetic VLBI observables, and thereby also
determine the magnitude of structure effects in the geodetic VLBI
observables. We also showed that large closure delays, even closure delays
in excess of 10~ns, are related to source structure effects, and that
the underlying delay measurements are not caused by simple measurement
errors. We demonstrated for the first time that source structure can
be obtained from the closure \emph{delays} as opposed to closure
phases or closure amplitudes from visibility data. For sources that
are reasonably compact on short VLBI baselines, we can
simply and directly solve for (not fit) structure effects for the
entire VLBI network of baselines without any additional a priori
information. We also apply model fitting to determine source
structure, showing how closure delays can yield structure information
without the need for sources to be unresolved on short baselines.
This method is relatively simple to implement
within existing geodetic analysis tools, uses input data
from the standard geodetic database files, and does not require
significant computational resources. For example, we can
compute structure models for all sources in the 15-day CONT14
campaign in a fraction of an hour, whereas our current imaging
analysis from the raw visibility data requires about 16~hours
to process one 24-hour segment of the CONT14 campaign on a
similar computer.
In an array of $N$ antennas, with $N(N-1)/2$ interferometer baselines, there are
$N(N-1)/2$ unknown structure effects to be determined and $(N-1)(N-2)/2$
independent closure delays as observables. Therefore, the fraction of the total
structure delay information available in the closure delays is
$(N-2)/N$. The ratio shows the benefits to be gained by increasing the number
of antennas in the array; with only 4 antennas, 50\% of the structure delay information
is available, while for 15 antennas, as the case of CONT14 observations of source 0642+449,
the ratio increases to 87\%. From these observations, the source structure effect
is demonstrated at the level of each individual VLBI group delay for the first time.
The study reveals that at X band
(8.4 GHz) during the CONT14 sessions the source had two point-like components with a
flux-density ratio of 0.92, that is, almost equally bright. The position of the weaker component
with respect to the stronger one is estimated to be $-$426 $\pm$ 12 $\mu$as in
right ascension and $-$66 $\pm$ 19 $\mu$as in declination. Finally,
the standard deviation of the corrected closure delay was reduced by
36$\%$. This structure model agrees with the estimated source
structure effect on baselines with lengths less than 10~000~km
very well, but does not fit some of the longest baselines.
There are mainly four reasons for that: (1) source
structure effects on these longest baselines with $R \ga$ 0.7 are
much more sensitive to the relative position of the two components;
(2) such long baselines only have a few observations over one circle of $uv$ position angle, making it statistically
difficult to identify the variation caused by the source structure; (3) there can be structure at smaller scale that
shows structure effect on longest baselines but none on baselines shorter than 10~000~km;
and (4) the model of Equation (\ref{eq_structure}) is
the derivative of the structure phase with respect to the observing frequency,
while the multiband group delay is derived from the linear estimation of the observed phase
over 8 channels spanning about 0.7 GHz---application of this model
formultiband group delays introduces errors in the structure effect, which may have larger impacts when
the baseline length is longer than 11~000~km with $R \ga$ 0.7.
Due to this inadequateness of the model for multiband group delays, the flux-density ratio $K$ may have been
underestimated.
In 1992 and 1995, this source was observed to have a compact ``core-jet"
morphology with the resolution of several milliarcseconds by \citet{gur92} and
\citet{xu95}, respectively. Recently, space VLBI (RadioAstron)
observations of this source at 1.6 GHz in 2013 with a resolution of
0.8 mas, $\sim$ 4 times better than that of ground VLBI images at
this wavelength, found that this source has two compact cores
separated by 0.76 mas with a position angle of 81$\degr$ in the sky plane \citep{lob15}.
Since space VLBI observations were made fourteen months earlier than CONT14 observations, one
may not expect they were observing the same blob, but the position angle of two components
should be approximately in the same direction.
Our result domnstrates that the two components are in the direction of position angle about
261.2$\degr$, which is the same direction detected by space VLBI.
The source 0642+449 did not exhibit a significant structure effect due to a frequency
dependence of the flux densities of the two components, which has a completely different
pattern, such as more peaks over 24 hours for baselines with $R \approx
0.5$. Our study shows a similar
structure of this source with a resolution comparable to that
of the space VLBI, demonstrating the feasibility of the application
of astrometric observables for the \deleted{direct} study of the source structure
with this method.
From the study by \citet{ber11}, we expect polarization
leakage to affect the multiband group delay by less than 1.6 ps for
90~\% of the observations. General leakage of LCP into RCP for the
geodetic observations will result in a baseline-dependent bias. For
the LCP part of the Stokes I emission that leaks into RCP, these
biases are constant in time and baseline orientation for a given
station pair, and do not explain the large, systematic, and source-dependent effects. \citet{hom06}
showed that at VLBI resolutions, the fractional circular polarization
of AGN core and jet components is typically less than about 1~\%.
Supposing that different baseline orientations constructively add/subtract
the phases from two components that are circularly polarized at the 1~\%
level, the change in delay caused by circularly polarized source
structure would only be 2~\% of the change in delay for the Stokes
I emission. We therefore expect that polarization effects are
negligible for this study, although they may be important to reach picosecond accuracies.
The large closure delays have also been effectively traced, which
reveals that most observables erroneously identified as outliers in VLBI
data analysis are in fact exposed to the source structure effect and
this effect could be at the level of tens of ns in some occasions
even for a radio source with a rather flat spectral index. This
cannot be explained yet by the model. This
needs to be studied in the future to find the explanation from
astrophysics, while for astrometric VLBI we should schedule routine
observations more effectively to exclude this kind of radio source,
or to observe it without such long baselines only if the two components
do not move with each other.
It is still challenging to implement the identified source structure
to correct the effect in VLBI data analysis. First, an accurate
model for multiband group delays to the level of at least 10 ps needs
to be derived. This model should be able to reduce the magnitudes of
the closure delays of triangles with the longest baselines to the
level of that of small triangles, a few tens of ps. Second, a careful re-study of the
linear combination of S and X band data with the presence of source
structure would be essential to have an accurate correction for the
source structure effect on the combined S-X observable.
Moreover, astrometric/geodetic VLBI observables at S band are one order of
magnitude noisier than that at X band,
which makes the source structure at S band almost unrecoverable, and
the structures at S and X bands are different. In general, structures
at S band are much more resolved than that at X band.
How could our method be improved for the study of source structure?
First, there should be more effective ways
of deriving the structure effects on $(N-1)$ baselines.
\replaced{Without the limitation from the assumption used for the connection,
our method by using the closure delay can be applied for
the source consisting of compact components with any separation.
And this will allow us to directly correct the structure delay on each single delay observable
for geodetic VLBI data analysis.}{Due to the limitation from the assumption used for the connection, our method would confine to a small fraction of radio sources. But if one can develop a new way to break this limitation, our method may allow us to directly correct the structure delay on each single delay observable for geodetic VLBI data analysis.}
Second, one might develop a new method of image reconstruction in an iterative way
other than modeling the structure delay. Then a non-linear estimation of
the structure parameters from the closure delay
has to be developed. The method can thus be extended to be used for
more general cases, complex or resolved sources. The rigorous method to correct the structure effect is to make images based on the standard VLBI imaging from the same observations and to correct the raw visibility phases for source structure in the geodetic VLBI analysis software prior to the multiband group delay fitting.\added{Even though this will need more work and resource compared to the current procedure of the routine VLBI data analysis, the geodetic VLBI should move onto it in the near future.} We are working on making images of CONT14 observations and the results will be presented in another paper.
This method could be of
great help to monitor the performance of radio sources for the historical VLBI observations and the VLBI
Global Observing System \citep[VGOS;][]{pet09}. In VGOS, there is a
global network of well-distributed stations and particularly
several twin telescopes. A wider range of baseline lengths from
hundreds of meters will be available, which then will allow
the source structure with a wide separation of compact components to be detectable.
Moreover, if the point-like sources that are more likely to be observed in astrometric VLBI
begin to demonstrate structure, it is likely to roughly model their structures
as consisting of compact components rather than a flat brightness
distribution. The source structure effect, as one of the main and
inevitable problems for the goals of the VGOS, can be expected to be
handled by this method to some extent. Besides the astrometric VLBI,
the method can provide benefits as imaging for the astrophysical
study of the source structure from continuous observations within
VGOS.
\acknowledgments
We thank the IVS for the CONT14 data used in this work, and David Gordon
(Goddard Space Flight Center, USA) and Brian Corey (MIT Haystack Observatory, USA) for the helpful discussions
of calculating closure delay from geodetic VLBI observations. This work was done while M.H.X.
worked as a guest scientist at GFZ, Germany, and supported by the National
Natural Science Foundation of China (Grant No. 11473057 and 11303077).
| bc2307d32b6377d0716c6d80009f5a776321707b | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Let $X_{1},X_{2},....$ be a sequence of independent random variables with common distribution function $F$ and for each integer $n\geq 1$, let
X_{1,n}\leq ...\leq X_{n,n}$ denote the order statistics based on the first
n$ of these random variables. Cs\"{o}rg\H{o} and Mason (1985, 1986) have recently shown among other results that if
\begin{equation}
1-F(x)=L^{\ast }(x)x^{-a}\text{ }as\text{ }x\longrightarrow \infty, \label{r1}
\end{equation}
\noindent where $L^{*}$ is a slowly varying function at infinity and $a\geq 2$, or if $F$ has exponential-like upper tails, meaning
\begin{equation}
\int_{x}^{+\infty }(1-F(y))dy/(1-F(x))\rightarrow c\text{ }as\text{
x\longrightarrow \infty , \label{r2}
\end{equation}
\noindent where $0<c<+\infty$, then for any sequence of integers satisfying
\begin{equation*}
1\leq k_{n}\leq n,\text{ }k_{n}\rightarrow +\infty \text{ }and\text{ }k_{n}/n\rightarrow 0\text{ }as\text{ }n\rightarrow +\infty, \tag{K}
\end{equation*}
\noindent there exist sequences $A_{n}>0$ of normalizing constants and $C_{n}$ of
centering constants such tha
\begin{equation}
A_{n} \left(\sum_{i=1}^{k_{n}}X_{n-i+1n}-C_{n}\right)\overset{d}{\longrightarrow }N(0,1
\text{ }as\text{ }n\rightarrow +\infty. \label{r3}
\end{equation}
\noindent The case \eqref{r1} is contained in the theorem of Cs\"{o}rg\H{o} and Mason (1986) and the case \eqref{r2} is Theorem 1.5 of Cs\"{o}rg\H{o} and Mason (1985). An application of Theorem 2.4.1 of de Haan (1970) (Lemma \ref{l1} below) combined with Fact 1.4 of Cs\"{o}rg\H{o} and Mason (1985) shows that \eqref{r2} implies the existence of sequences of normalizing constants $a_{n}$ and centering constants $b_{n}$, such that
\begin{equation}
a_{n}^{-1}(X_{n,n}-b_{n})\overset{d}{\longrightarrow }G\text{ }as\text{
n\rightarrow +\infty , \label{r4}
\end{equation}
\noindent where $G$ is a Gumbel random variable with distribution functio
\begin{equation*}
P(G\leq x)=\exp (-e^{-x})\text{ for }-\infty <x<+\infty .
\end{equation*}
\noindent Whenever such sequences of constants can be chosen so that \eqref{r4} holds,
we say that $F$ is in the domain of attraction of a Gumbel law, written
F\in D(\Lambda)$.\\
\noindent One of the purposes of this note is to show that \eqref{r3} holds more generally than under condition \eqref{r2}, that is, $F\in D(\Lambda)$ is sufficient for \eqref{r3} to hold. This will be a consequence of our main results stated in the next section. We shall also obtain some further extensions of the results of Cs\"{o}rg\H{o} and and Mason. The proofs are given in Section \ref{sec3}.\\
\section{Statement of main results} \label{sec2}
First we introduce some notations. Let
\begin{equation*}
Q(s)=inf\{x:F(x)\geq s),\text{ for }0<s\leq 1,
\end{equation*}
\bigskip \noindent with $Q(0)=Q(0+)$, denote the inverse or quantile function of $F$. Write
\begin{equation*}
\sigma ^{2}=\int_{1-s}^{1}\int_{1-s}^{1}(\min (u,v)-uv)dQ(u)dQ(v),\text{ for
}0\leq s\leq 1.
\end{equation*}
\noindent For any $0<\beta <\infty $, set
\begin{equation*}
c(s,\beta )=s^{-\beta }\int_{1-s}^{1}(1-u)^{\beta }dQ(u),\text{ for }0\leq
s\leq 1.
\end{equation*}
\noindent For convenience, when $\beta =1$, we set $c(s)=c(s,1)$. (Refer to the next
section for our integral convention).\\
\noindent Let $D^{\ast }(\Lambda )$ denote the subclass of $D(\Lambda )$ consisting of
all distribution functions $F$ whose quantile function $Q$ satisfie
\begin{equation*}
Q(1-s)=a+\int_{s}^{1}u^{-\beta }r(u)du,
\end{equation*}
\noindent for all $s\geq 0$ sufficiently small, $a$ is a fixed constant and $r$ is a
strictly positive function slowly varying at zero. The fact that $D^{\ast
}(\Lambda )$ is a subclass of $D(\Lambda )$ follows from Theorem 2.4.1 of de
Haan (1970).\\
\noindent For any sequence of positive integers $k_{n}$, such that $(K)$ holds and
F\in D(\Lambda )$, set for $n=1,2,...,$
\begin{equation*}
\mu _{n}(k_{n})=\int_{1-k_{n}/n}^{1}Q(s)ds.
\end{equation*}
\noindent The following theorem contains our main results.\\
\noindent \textbf{Theorem}.
\textit{On a rich enough probability space, there exist a sequence of independent
random variables $X_{1},X_{2},....,$ with common distribution function $F$
and a sequence of Brownian bridges $B_{1},$ $B_{2},$, . . . , such that for
any sequence $k_{n}$ satisfying $(K)$, whenever $F\in D(\Lambda )$},
\begin{eqnarray}
&&k_{n}^{1/2}c(k_{n}/n)^{-1}\left\{ \sum_{i=1}^{k_{n}}X_{n-i+1n}-\mu
_{n}(k_{n})\right\} \label{r5} \\
&=&-(n/k_{n})^{1/2}c(k_{n}/n)^{-1
\int_{1-k_{n}/n}^{1}B_{n}(s)dQ(s)+o_{P}(1):=Z_{n}+o_{P}(1), \notag
\end{eqnarray
\textit{and whenever} $F\in D^{\ast }(\Lambda ),
\begin{eqnarray}
&&k_{n}^{-1/2}c(k_{n}/n)^{-1}\left\{
\sum_{i=1}^{k_{n}}X_{n-i+1n}-Q(1-k_{n}/n)\right\} \label{r6} \\
&=&-(n/k_{n})^{1/2}c(k_{n}/n)^{-1}B_{n}(1-k_{n}/n)+o_{P}(1):=Y_{n}+o_{P}(1),
\notag
\end{eqnarray}
\noindent \textit{and}
$$
k_{n}^{-1/2}c(k_{n}/n)^{-1}\left\{
\sum_{i=1}^{k_{n}}X_{n-i+1n}-k_{n}X_{n-k_{n},n}-
\int_{1-k_{n}/n}^{1}r(1-s)ds)\right\}
$$
\begin{equation}
=Z_{n}-Y_{n}+o_{P}(1). \label{r7}
\end{equation}
\noindent\textit{ Furthermore, the random variables on the left side of \eqref{r5}, \eqref{r6}
and \eqref{r7}, respectively, converge in distribution to $N(0,2)$, $N(0,1)$
and $N(0,1)$, respectively, as $n\rightarrow +\infty$.}\\
\noindent \textbf{Remark}. With the choice $A_{n}=\left( 2k_{n}\right) ^{-1/2}c(k_{n}/n)^{-1}$ and $C_{n}=\mu _{n}(k_{n}),$ we see that our theorem implies \eqref{r3} whenever $D(\Lambda)$. Our theorem also extends Theorem 1.5, 1.7 and 2.1 and Corollary 2.5 of Cs\"{o}rg\H{o} and Mason (1985). The random variable on the left side of \eqref{r7} is related to the Hill (1975) estimator of the tail index of a distribution for this random variable was motivated by the work of Mason (1982) (see also Deheuvels, Haeusler and Mason (1988)).
\section{Proof of the theorem} \label{sec3}
We use the following integral convention : When $0\leq a\leq
b\leq 1,$ g is left-continuous and f is right-continuous,
\begin{equation*}
\int_{b}^{a}fdg=\int_{(a,b)}fdg\text{ and }\int_{a}^{b}gdf=\int_{(a,b)}gdf
\end{equation*}
\noindent whenever these integrals make sense as Lebesgue-Stieltjes integrals. In this
case, the usual integration by parts formula
\bigskip
\begin{equation*}
\int_{a}^{b}fdg+\int_{a}^{b}gdf=g(b)f(b)-g(a)f(a)
\end{equation*}
\noindent is valid.\\
\noindent The proof of our theorem will follow closely the proofs of the results of
Cs\"{o}rg\H{o} and Mason (1985), substituting their technical lemmas concerning
properties of the quantile functions of distribution functions satisfying \eqref{r2}, by those describing properties of the quantile functions of $F\in
D(\Lambda ).$ We therefore begin with these technical lemmas.\\
\bigskip
\begin{lemma} \label{l1} $F\in D(\Lambda )$ if and only if for each choice of $0\leq x,y,w,z<\infty $
\ fixed, $y\neq w,
\begin{equation}
\frac{Q\left( 1-sx\right) -Q\left( 1-sz\right) }{Q\left( 1-sy\right)
-Q\left( 1-sw\right) }\rightarrow \frac{\log x-\log z}{\log y-\log w}\text{
\ as s}\downarrow 0. \label{r8}
\end{equation}
\end{lemma}
\noindent This is Theorem 2.4.1 of de Haan (1970).\\
\begin{lemma} \label{l2} Whenever $F\in D(\Lambda ),$ $c(s,\beta )$ is slowly varying at
zero for each choice of $0<\beta <\infty $ .
\end{lemma}
\noindent \textbf{Proof}. We have to show that for each $0<\lambda <\infty $ and
0<\beta <\infty ,
\begin{equation}
c(\lambda s,\beta )/c(s,\beta )\rightarrow 1\ \text{as \ }s\downarrow 0.
\label{r9}
\end{equation}
\noindent Choose any $0<\lambda <\infty $ \ and $0<\theta <1.$ Then for all $s>0$ small enough we hav
\begin{eqnarray}
\int_{1-\lambda s}^{1}\left( 1-u\right) ^{\beta }dQ\left( u\right)
&=&\sum\limits_{i=0}^{\infty }\int_{1-\lambda s\theta ^{i}}^{1-\lambda
s\theta ^{i+1}}\left( 1-u\right) ^{\beta }dQ\left( u\right) \notag \\
&\leq &\sum\limits_{i=0}^{\infty }\lambda ^{\beta }s^{\beta }\theta
^{i\beta }\left\{ Q\left( 1-\lambda s\theta ^{i+1}\right) -Q\left( 1-\lambda
s\theta ^{i}\right) \right\} \label{r10}
\end{eqnarray}
\noindent Applying Lemma \ref{l1} gives
\begin{equation}
\frac{Q\left( 1-\lambda \theta u\right) -Q\left( 1-\lambda u\right) }
Q\left( 1-\theta u\right) -Q\left( 1-u\right) }\rightarrow 1\text{ as s
\downarrow 0. \label{r11}
\end{equation}
\noindent Select any \ $0<\varepsilon <\infty$. From \eqref{r11} we have that, for all $s>0$ sufficiently small, expression \eqref{r10} is
\begin{eqnarray*}
&\leq &\left( 1+\varepsilon \right) \sum\limits_{i=0}^{\infty }\lambda
^{\beta }s^{\beta }\theta ^{i\beta }\left\{ Q\left( 1-\lambda s\theta
^{i+1}\right) -Q\left( 1-\lambda s\theta ^{i}\right) \right\} \\
&\leq &\frac{\left( 1+\varepsilon \right) \lambda ^{\beta }}{\theta ^{\beta
}\sum\limits_{i=0}^{\infty }\theta ^{\left( i+1\right) \beta
}\int_{1-s\theta ^{i}}^{1-s\theta ^{i+1}}dQ\left( u\right) \\
&\leq &\frac{\left( 1+\varepsilon \right) \lambda ^{\beta }}{\theta ^{\beta
}\int_{1-s}^{1}\left( 1-u\right) dQ\left( u\right).
\end{eqnarray*}
\noindent Thus for all $s>0$ sufficiently small,
\begin{equation}
c(\lambda s,\beta )\leq \frac{\left( 1+\varepsilon \right) }{\theta ^{\beta
}c(s,\beta ). \label{r12}
\end{equation}
\noindent Observing that for all $s>0$ small enough
\begin{equation*}
\int_{1-\lambda s}^{1}\left( 1-u\right) ^{\beta }dQ\left( u\right) \geq
\sum\limits_{i=0}^{\infty }\lambda ^{\beta }s^{\beta }\theta ^{\left(
i+1\right) \beta }\left\{ Q\left( 1-\lambda s\theta ^{i+1}\right) -Q\left(
1-\lambda s\theta ^{i}\right) \right\} ,
\end{equation*}
\noindent we see that by an argument very much like the one just given, we have for
all $s>0$ sufficiently small,
\begin{equation}
c(\lambda s,\beta )\geq \left( 1+\epsilon \right) \theta ^{\beta }c(s,\beta). \label{r13}
\end{equation}
\noindent Assertion \eqref{r9} now follows from inequalities \eqref{r12} and \eqref{r13}
by the fact that $\theta $ can be chosen arbitrarily close to one and $\varepsilon$ arbitrarily close to zero. This completes the proof of Lemma \ref{l2}.\\
\noindent The following lemma is related to Theorem 1.4.3.d of de Haan (1970) and its proof is based on a modification of the techniques used to prove this theorem. For details see Deheuvels \textit{et al.} (1986).\\
\begin{lemma} \label{l3} Whenever $F\in D(\Lambda )$, there exists a constant \ $-\infty <b<+\infty $
\ such that for all \ $0<s\leq 1/2,$
\begin{equation}
Q\left( 1-s\right) =b-c(s)+\int_{s}^{1}u^{-1}c(u)du. \label{r14}
\end{equation}
\end{lemma}
\bigskip
\begin{lemma} \label{l4}
Whenever $F\in D(\Lambda )$, for each \ $0<x<+\infty ,$
\begin{equation}
\lim_{s\downarrow 0}\frac{Q\left( 1-xs\right) -Q\left( 1-s\right) }{c(s)
=-\log x. \label{r15}
\end{equation}
\end{lemma}
\bigskip \noindent \textbf{Proof}. Applying Lemma \ref{l3}, we have for any $0<x<+\infty$ and for all
$s$ sufficiently smal
\begin{equation*}
\frac{Q\left( 1-xs\right) -Q\left( 1-s\right) }{c(s)}=\frac{c(s)-c(xs)}{c(s)
+\int_{xs}^{s}\frac{c(u)}{u}du.
\end{equation*}
\noindent Since $c$ is slowly varying at zero, both
\bigskip
\begin{equation*}
\inf \left\{ \frac{c(u)}{c(s)}:u\in I(s)\right\} \rightarrow 1
\end{equation*}
\noindent and
\begin{equation*}
\sup \left\{ \frac{c(u)}{c(s)}:u\in I(s)\right\} \rightarrow 1
\end{equation*}
\noindent as s$\downarrow 0$ \ where $I(s)$ is the closed interval formed by $xs$ and
s$. From these two facts the proof of Lemma \ref{l4} follows immediately.
\begin{lemma} \label{l5} Whenever $F\in D(\Lambda )$, for each \ $0<\beta <+\infty
\begin{equation}
c(s,\beta )/c(s)\rightarrow 1/\beta \text{ as s}\downarrow 0. \label{r16}
\end{equation}
\end{lemma}
\bigskip \noindent \textbf{Proof}. Let $\widehat{Q}(1-s)=$ $Q(1-s^{1/\beta })$. Since by Lemma \ref{l1} for any
choice of $0<x,y<\infty ,$ $y\neq 1,
\begin{equation*}
\frac{\widehat{Q}(1-xu)-\widehat{Q}(1-u)}{\widehat{Q}(1-yu)-\widehat{Q}(1-u)}=\frac{{Q}(1-x^{1/\beta }u^{1/\beta
})-Q(1-u^{1/\beta })}{Q(1-y^{1/\beta }u^{1/\beta })-Q(1-u^{1/\beta })}
\end{equation*}
\noindent converges as $u\downarrow 0$ to $\log x/\log y$, we conclude that $\widehat{Q}\in
D(\Lambda)$. Let
\begin{equation*}
\widehat{c}(s)=s^{-1}\int_{1-s}^{1}\left( 1-u\right) d\widehat{Q},\text{ \ \ \ for }0<s<1.
\end{equation*}
\noindent A change of variables shows that $\widehat{c}(s^{\beta})=c(s,\beta )$ \ for $0<s<1$. Thus
\begin{eqnarray*}
\frac{c(s,\beta )}{c(s)} &=&\frac{\widehat{c}(s^{\beta })}{c(s)}
\\
&=&\frac{Q(1-2s)-Q(1-s)}{c(s)}\times \frac{\widehat{Q}(1-\left(
2s\right) ^{\beta })-\widehat{Q}(1-s^{\beta })}{Q(1-2s)-Q(1-s)}
\\
&&\times \frac{\widehat{c}(s^{\beta })}{\widehat{Q}(1-\left( 2s\right) ^{\beta })-\widehat{Q}(1-s^{\beta })},
\end{eqnarray*}
\noindent which by Lemmas \ref{l1} and \ref{l4} converges to $1/\beta $ as s$\downarrow 0$,
completing the proof of Lemma \ref{l5}.
\bigskip
\begin{lemma} \label{l6}
Whenever $F\in D(\Lambda )$
\begin{equation}
\sigma ^{2}(s)/\left( 2sc^{2}(s)\right) \rightarrow 1\text{, \ \ as
s\downarrow 0. \label{r17}
\end{equation}
\end{lemma}
\bigskip \noindent \textbf{Proof. }The proof is based on Lemma \ref{l5} and follows almost exactly as
the proof of Lemma 3.3 of Cs\"{o}rg\H{o} and Mason (1985). Therefore, the details
are omitted. The proof of the following lemma is an easy consequence of the
Karamata representation for a slowly varying function.
\begin{lemma} \label{l7}
Let $a_{n}$ be any sequence of positive constants such that
a_{n}\rightarrow 0$ and \ $na_{n}\rightarrow \infty .$ Also let $L$ be any
slowly varying function at zero. Then for any $0<\beta <\infty ,
\begin{equation}
n^{-\beta }L\left( 1/n\right) /\left( \left( a_{n}\right) ^{\beta }L\left(
a_{n}\right) \right) \rightarrow 0\text{ \ \ as }n\downarrow \infty.
\label{r18}
\end{equation}
\end{lemma}
\bigskip \noindent We now describe the probability space on which the assertions of the theorem
are assumed to hold. M. Cs\"{o}rg\H{o}, S. Cs\"{o}rg\H{o} , Horv\'{a}th and Mason (1986) have
constructed a probability space $(\Omega ,\mathcal{A},\mathbb{P})$ carrying
a sequence $U_{1},U_{2},...$ of independent random variables uniformly
distributed on $(0,1)$ and a sequence $B_{1},B_{2},...$ of Brownian
bridges such that for the empirical process
\begin{equation*}
\alpha _{n}(s)=n^{1/2}\left\{ G_{n}\left( s\right) -s\right\} ,\text{ }0\leq
s\leq 1,
\end{equation*}
\noindent and the quantile process
\begin{equation*}
\beta _{n}(s)=n^{1/2}\{s-U_{n}(s)\},0\leq s\leq 1,
\end{equation*
where
\begin{equation*}
G_{n}(s)=n^{-1}\{k:1\leq k\leq n,U_{k}\leq s\},
\end{equation*}
\noindent and, with $U_{1,n}\leq \ldots \leq U_{n,n}$ denoting the order statistics
corresponding to $U_{1},\ldots ,U_{n}.$,
\begin{equation*}
U_{n}(s)=\left\{
\begin{tabular}{lll}
$U_{k,n}$ & $if$ & $(k-1)/n<s\leq k/n,k=1,\ldots ,n,$ \\
$U_{1,n}$ & $if$ & $s=0
\end{tabular
\right.
\end{equation*}
\noindent we have
\begin{equation}
\sup_{0\leq s\leq 1}n^{\nu _{1}}|\alpha _{n}(s)-\bar{B}_{n}(s)|/(1-s)^{-\nu
_{1}+1/2}=O_{p}(1) \label{r19}
\end{equation}
\noindent with $\bar{B}_{n}(s)=B_{n}(s)$ for $1/n\leq s\leq 1-1/n$ and zero elsewhere and
\begin{equation} \label{r20}
\sup_{0\leq s \leq 1-1/n }n^{\nu_2}|\beta_n(s)-B_n(s)|/(1-s)^{-\nu_2 +1/2}
=O_p(1)
\end{equation}
\noindent where $\nu_1$, and $\nu_2$ are any fixed number such that $0\leq \nu_1<1/4$
and $0 \leq \nu_2\leq 1/2$. The statement in \eqref{r19} follows from Theorem
2.1, while the statement in \eqref{r20} is easily inferred from Corollaries
2.1 and 4.2.2 of the above paper.\\
\noindent Thoughout the remainder of the proof of our theorem, we assume that we are
on the probability space of Cs\"{o}rg\H{o} et al.(1986). Since the sequence of
random variables $X_1, X_2,\ldots,$ is equal in distribution to $Q(U_1),
Q(U_2,),\ldots$, we can and do assume that the first sequence is equal to
the second.\\
\noindent First assume $F\in D(\Lambda )$. We shall establish \eqref{r5}. Applying
integration by parts we see that the left side of \eqref{r5} equals
$$
-(n/k_n)^{1/2}c(k_n/n)^{-1}\int_{1-k_n/n}^1\alpha_n(s)dQ(s)
$$
$$
+nk_n^{-1/2
\int_{U_n-k_n/n}^{1-k_n/n}(1-G_n(s)-k_n/n )\frac{dQ(s)}{c(k_n/n)}
$$
$$
:=\Delta_{1,n}+\Delta_{2,n}.
$$
\noindent We shall first show that
\begin{equation*}
\Delta_{1,n}=Z_n+R_n
\end{equation*}
\noindent with $R_n=o_p(1)$. From \eqref{r19} we have for any $0<\nu<1/4$,
\begin{equation} \label{r21}
\sup_{ 0\leq s\leq 1 }n^{\nu_1}|\alpha_n(s)-\bar{B}_n(s)|/(1-s)^{-\nu_1
+1/2}=O_p(n^{-\nu}).
\end{equation}
\noindent Notice that for any such $\nu$,
\begin{eqnarray*}
&& |R_n|\leq \sup_{ 0\leq s\leq 1 } \frac{|\alpha_n(s)-\bar{B}_n(s)|}
(1-s)^{-\nu_1 +1/2}}\left( \frac{n}{k_n}\right)^{1/2}\int_{1-k_n/n}^1(
1-s)^{-\nu_1 +1/2}dQ(s)/c(k_n/n) \\
&&+ \left\vert \left( \frac{n}{k_n}\right)^{1/2}\int_{1-1/n}^1
B_n(s)dQ(s)/c(k_n/n)\right \vert :=R_{1,n}+R_{2,n}.
\end{eqnarray*}
\noindent From \eqref{r21}, we obtain
\begin{equation*}
R_{1,n}=O_p(n^{-\nu}n^\nu \frac{c(k_n/n,1/2-\nu)}{c(k_n/n)}k_n^{-\nu},
\end{equation*}
\noindent which by Lemma \ref{l5} equals $o_p(1)$. Also
\begin{equation*}
E R_{2,n}^2=\sigma^2(1/n)/\left(\frac{k_n}{n}c_n^2\left(\frac{k_n}{n}\right)
\right)
\end{equation*}
\noindent which by Lemma \ref{l6} is
\begin{equation*}
\sigma ^{2}(1/n)/\sigma ^{2}(k_{n}/n)\ \ as\ n\rightarrow \infty.
\end{equation*}
\noindent From Lemmas \ref{l2} and \ref{l6} we infer that $\sigma^2(s)$ is regularly varying of
exponent one at zero. Hence, by Lemma \ref{l7},
\begin{equation*}
\sigma ^{2}(1/n)/\sigma ^{2}(k_{n}/n)\rightarrow 0\ \ as\ n\rightarrow
\infty
\end{equation*}
\bigskip \noindent which yields $R_{2,n}=o_p(1)$. Thus we have proved $R_{n}=o_p(1)$.\newline
Next we show that $\Delta_{2,n}=o_p(1)$. Choose any $1 <\lambda<\infty$ and
set
\begin{equation*}
T_{n}(\lambda
)=nk_{n}^{-1/2}c(k_{n}/n)^{-1}|1-G_{n}(1-k_{n}/n)-k_{n}/n|\{Q(r_{n}^{+}
\lambda ))-Q(r_{n}^{-}(\lambda ))\},
\end{equation*
\begin{equation*}
r_{n}^{-}(\lambda ))=1-\frac{\lambda k_{n}}{n}\ \ and\ r_{n}^{+}(\lambda
))=1-\frac{k_{n}}{\lambda n}.
\end{equation*}
\noindent Notice that since for all $s$ in the closed interval formed by
U_{n-k_{n},n}$ and $1-k_{n}/n$,
\begin{equation*}
|1-G_{n}(s)-k_{n}/n|\leq |1-G_{n}(1-k_{n}/n)-k_{n}/n|
\end{equation*}
\noindent we have for any $1<\lambda <\infty $
\begin{equation*}
\liminf_{n\rightarrow \infty }P(|\Delta _{2,n}|\leq T_{n}(\lambda )|\geq
\liminf_{n\rightarrow \infty }P(r_{n}^{-}(\lambda )\leq U_{n-k_{n},n}\leq
r_{n}^{+}(\lambda )).
\end{equation*}
\noindent Since $(K)$ implies (cf. Balkema and de Haan (1975)) that
\begin{equation}
n(l-U_{n-k_{n},n)/k_{n}}\rightarrow ^{P}1\text{ as } n\rightarrow \infty:
\label{r22}
\end{equation}
\noindent the lower bound in the above inequality equals one. Hence for each
1<\lambda <\infty $,
\begin{equation}
\lim_{n\rightarrow }P(|\Delta _{2,n}|\leq |T_{n}(\lambda )|)=1. \label{r23}
\end{equation}
\noindent Observe for each $1<\lambda <\infty $
\begin{equation*}
E(T_{n}(\lambda )\leq \left\{ Q\left( 1-\frac{k_{n}}{\lambda _{n}}\right)
-Q\left( 1-\frac{\lambda k_{n}}{n}\right) \right\} /c(k_{n}/n).
\end{equation*}
\noindent Applying Lemma \ref{l4} we see that this last expression converges to $2\log
\lambda $, which yields
\begin{equation}
\lim_{\lambda \downarrow 1}\limsup_{n\rightarrow \infty }E(T_{n}(\lambda )=0.
\label{r24}
\end{equation}
\noindent The fact that $\Delta _{2,n}=o_{p}(1)$ now follows by an elementary argument
based on \eqref{r23} and \eqref{r24}. This completes the proof of \eqref{r5}.\\
\noindent Next consider \eqref{r6}. Notice that since $F\in D^{\star }(\Lambda )$,
\begin{equation*}
c(s)=s^{-1}\int_{1-s}^1r(1-u)du.
\end{equation*}
\noindent Thus since $r>0$ and slowly varying at zero Theorem 1.2.1 of de Haan (1970)
gives
\begin{equation} \label{r25}
r(s)/c(s) \rightarrow 1 \ \ \mbox{as}\ \ s\downarrow 0
\end{equation}
\noindent The left side of \eqref{r6} equals
\begin{eqnarray*}
&& -\frac{k_n^{1/2}}{c(k_n/n}\int_{k_n/n}^{1-U_{n-k_n/n}}\frac{r(u)}{u
du=-k_n^{1/2}\frac{r(k_n/n)}{c(k_n/n)}\{\log(1-U_{n-k_n/n}\} \\
&& \ \ \ \ - \ \ \frac{k_n^{1/2}}{c(k_n/n
\int_{k_n/n}^{1-U_{n-k_n/n}}(r(u)-r(k_n/n))\frac{du}{u}:=\Delta_{1,n}^\star
+\Delta_{2,n}^\star
\end{eqnarray*}
\noindent The same argument based on \eqref{r20} as given in Cs\"{o}rg\H{o} and Mason (1985)
shows that
\begin{equation*}
-k_n^{1/2}(\log(1-U_{n-k_n/n})-\log(k_n/n))=Y_n+o_p(1).
\end{equation*}
\noindent Therefore by \eqref{r25} and the fact that $Y_n=o_p(1)$ we have
\begin{equation*}
\Delta_{1,n}^\star =Y_n +o_p(1).
\end{equation*}
\noindent Since $r$ is slowly varying at zero, we get for each $1<\lambda <\infty$ as
n \rightarrow \infty$,
\begin{equation}
\sup\left\{|r(s)-r(k_n/n)/c(k_n/n):\frac{k_n}{\lambda_n}\leq s \leq \frac
\lambda k_n}{n} | \right\} \rightarrow 0 \label{r26}
\end{equation}
\noindent The fact that $\Delta_{2,n}^\star=o_p(1)$ now follows easily from $Y_n=o_p(1)
$, \eqref{r22},\eqref{r26} completing the proof of \eqref{r6}.\\
\noindent Since $F\in D^\star(\Gamma)$ we have
\begin{equation*}
\mu_n(k_n)-k_nQ(1-k_n/n)=\int_{1-k_n/n}^1r(1-s)ds.
\end{equation*}
\noindent Assertion (7) is now a direct consequence of \eqref{r5} and \eqref{r6}.\\
\noindent Finally we prove the convergence in distribution of $Z_n$, $Y_n$, and $Z_n-
Y_n$,, to $N(0,2), N(0,1)$ and $N(0,1)$, respectively, as $n \rightarrow
\infty$. Notice that the $Z_n$, random variable in \eqref{r5} is normal with
mean zero and second moment $\sigma^2(k_n/n)/(k_nc^2(k_n)/n),$, which by
Lemma \ref{l6} converges to $1$ as $n\rightarrow \infty$. The $Y_n$, random
variable in \eqref{r6} is normal with mean zero and second moment $1-k_n/n
\rightarrow 1$ as $n\rightarrow \infty$.\\
\noindent The $Z_{n}-Y_{n}$ random variable in \eqref{r7} is normal with mean zero.
Applying Lemmas \ref{l5} and \ref{l6} it is easy to verify that $E(Z_{n}-Y_{n})^{2
\rightarrow 1$ and $n\rightarrow \infty $. This completes the proof of the
theorem.
\bigskip \noindent{\textbf{Acknowledgement}}.\newline
\noindent The author is extremely grateful to David Mason, Erich Haeusler
and Paul Deheuvels, for helping him to complete this work.
| 02f5c1a1917612665f91158e3a98497b31cfad05 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section*{abstract}
Light possesses both spin and orbital angular momentum (AM). While spin AM is determined by helicity of circular-polarization, orbital AM is characterized by topological charge of vortex beam. Interaction of AM with optical beam orbit leads to optical spin Hall or orbital Hall effect, exhibited as spin-dependent or topological charge-dependent transverse shift of optical beam. Conservation of AM enables spin-to-orbital AM conversion, where circular-polarized Gaussian beam is converted to opposite-helicity circular-polarized vortex beam with topological charge $\pm 2$, an example of controlling spatial beam profiling by spin flip. However, the resultant vortex beam has the beam center of gravity unchanged, the same as that of incident Gaussian beam, meaning a null transverse shift. Here we introduce a cyclic group symmetric metasurface to demonstrate generation of vortex beam exhibiting spin-dependent transverse shift, namely, spin- and orbital-Hall effect, attributed to an alteration of dynamical phase of scattered beam according to the order $n$ of cyclic group while keeping geometric phase constant. Capability of spin-controlled spatial beam profiling with a transverse shift via spin- and orbital-Hall effect has important implications for spatial demultiplexing in optical communication utilizing orbital AM mode division multiplexing as well as for optical vortex tweezer and signal processing involving vortex beams.
\section*{Introduction}
Circularly polarized (CP) light possesses spin angular momentum (SAM) with a helicity defined as its projection to linear momentum of propagating optical wave. In addition to SAM, optical wave can be rendered to possess orbital angular momentum (OAM) by constructing a vortex beam having helical wavefront.\cite{allen1992orbital} Both SAM and OAM are utilized to carry informations in optical communication via polarization-division or OAM-division multiplexing. Differently from 2 modes of SAM with spins of $\pm 1$, the number of modes of OAM is not restricted since the topological charge of vortex beam is determined by twisting number of the helical wavefront. Owing to the availability of a large number of modes, vortex beams have been widely employed in applications to increase the signal channels for optical communications.\cite{bozinovic2013terabit,ren2016chip,yang2016wavelength}
In those applications it is necessary to be able to manipulate and control the intensity and phase distributions of vortex beam.
Anisotropic planar structures such as metasurface are important optical elements for such applications when SAM is utilized for shaping vortex beam, since interaction of SAM and OAM with optical beam orbit is enhanced.\cite{bliokh2015spin}
There are several well-known means to create vortex beams, for example, a pair of cylindrical lenses, spiral phase plate, fork-shaped grating, and spatial light modulator as well as photonic integrated devices.\cite{allen1999iv,padgett2004light,cai2012integrated,li2015orbital}
Another important means to create vortex beams is to make use of geometric phase by constructing space-variant Pancharatnam-Berry phase optical elements including
subwavelength diffraction grating\cite{biener2002formation} and liquid crystal q-plates in spin-to-orbital AM conversion (SOC) process.\cite{marrucci2006optical}
Related to metasurface, a high-efficiency SOC has been demonstrated to generate vortex beams with high topological charges in the visible wavelength based on dielectric metasurfaces.\cite{devlin2016spin}
Generation of vortex beam by SOC is an example of spin-controlled spatial beam profiling since an incident Gaussian beam converts to a vortex beam with donut-shaped spatial beam profile by spin flip. In case of a rotational symmetric Pancharatnam-Berry phase optical element, SOC-generated vortex beams with opposite signs of topological charges are degenerate in spatial beam profiles, which results from conservation of angular momentum consisting of spin AM and orbital AM. In other words, the beam centers of gravity of vortex beams are the same at that of the incident Gaussian beam.
Now we address the question how the degeneracy of spatial beam profiles of SOC-generated vortex beams can be lifted in SOC to provide spin-controlled spatial beam profiling such that the beam centers of gravity of vortex beams with opposite signs of topological charges are spatially separated by transverse-shifting in opposite directions.
Optical Hall effect is one means to achieve spatial separation by use of spin-dependent transverse shift originating from SAM- or OAM-orbit interaction,
which is also technically important when spatially demultiplexing OAM-mode division multiplexed superposition states.
Here, we introduce a cyclic group symmetric metasurface (CGSM) to generate vortex beam exhibiting a spin- and orbital-AM dependent transverse shift by identifying SOC in terms of geometric phase acquired by an optical beam in cross-polarization scattering from ring structure of nano-antennas.
By designing CGSMs belonging to the cyclic group $C_n$, dynamical phase of cross-polarization scattered beam is altered according to the order $n$ of cyclic group while keeping geometric phase constant. When $n$-fold rotational symmetry of azimuthal dynamical phase gradient is an odd integer $n$, there results a spin- and orbital-Hall effect.\cite{bliokh2010angular}
\section{Geometric and dynamical phases in optical Hall effect}
\label{sec:geometirc}
Intrinsic optical spin or orbital Hall effect refers to spin-dependent or vortex-charge dependent transverse shift, a consequence of the coupling between SAM or OAM with optical beam orbit originating from transversality of electromagnetic wave.\cite{shitrit2013spin,cardano2015spin}. In a general framework of physical description, it is an effect of fast degree of freedom (SAM or OAM) on slow degree of freedom (optical beam orbit), described by momentum-space Lorentz equation of motion in the presence of a Berry curvature of topological monopole. \cite{Berry-BookChapter,onoda2004hall,bliokh2008geometrodynamics}
Spin-dependent transverse shift of optical beam is also achieved by geometric Pancharatnam-Berry (PB) phase in a linear array of nano-antennas, called extrinsic optical spin Hall effect, to distinguish from the transverse shift originating from topological monopole.
The amount of shift is related to PB phase as shown in \textbf{Fig.~\ref{constantPB} a, d, e}.\cite{luo2015photonic}
In a circular array of nano-antennas with uniform width and length, even though not exhibiting spin-dependent transverse shift, spin-dependent PB phase is utilized to generate vortex beam by SOC in \textbf{Fig.~\ref{constantPB} b, f, g}.\cite{biener2002formation,marrucci2006optical, brasselet2013topological,karimi2014generating,osorio2015k}
Polarization states in Poincar\'e sphere go through a circular trajectory twice providing the solid angle $\pm8\pi$ corresponding to geometric phase $\pm4\pi$, generating vortex beam of $l=\pm 2$, as shown in \textbf{Fig.~\ref{constantPB} c}.
Optical wave scattered from the circular array acquires a phase factor, entirely geometric phase coming from a circular closed path $\textbf{C}$ in Stokes parameter space, leading to $\oint d\gamma(\textbf{C}) = \pm 4\pi$ with a constant azimuthal PB phase gradient $\nabla_\phi \Phi_{PB} = \pm 2$.
Therefore, the intensity profiles of $I_{-+}(r,\phi)=|{E}_{-+}|^2$ and $I_{+-}(r,\phi)=|{E}_{+-}|^2$ are degenerate even though helical phases are opposite for $l=\pm 2$, where $\{+-\}$ stands for left-circular polarization (LCP, $\sigma_{+}$) scattering / right-circular polarization (RCP, $\sigma_{-}$) incidence and $\{-+\}$ vice-versa.
In fact, ${E}_{-+}(r,\phi) = J_2(k_0r)e^{-{k_0}^2r^2}\exp(i2\phi)$ and ${E}_{+-}(r,\phi) = J_{-2}(k_0r)e^{-{k_0}^2r^2}\exp(-i2\phi)$, where a constant $\nabla_\phi \Phi_{PB}$ leads to $\phi$-independent amplitude $J_{\pm 2}(r)e^{-{k_0}^2r^2}$ from the fact that each two neighboring nano-antennas provide an equal amount of PB phase during one full cycle, owing to rotational invariance of circular array of nano-antennas.
It has been reported that a spin-dependent transverse shift of vortex beam in SOC is observed by lifting degeneracy in OAM though rotational symmetry breaking without
clarifying the underlying role of azimuthal phase gradient in giving rise to extrinsic spin- and orbital-Hall effect.\cite{shitrit2011optical,shitrit2013rashba,liu2015photonic,ling2015giant}
Now we examine how the azimuthal phase gradient is closely related to the existence of extrinsic spin- and orbital-Hall effect.
Consider a circular array of nano-antennas with different sizes from $1$ to $16$ in the range of $0 \le \phi \le 2\pi$, as shown in \textbf{Fig.~\ref{nonconstantPB} a}.
Scattering wave from the circular array acquires a phase factor originating from both geometric phase $\gamma(\textbf{C})$ and dynamical phase $\Phi_D$,\cite{Berry45} where $\gamma(\textbf{C})$ is wavelength-independent while $\Phi_D$ has a frequency dispersion related to plasmonic resonances of each nano-antenna (see Supplementary Information).
When rotational symmetry is broken, azimuthal dynamical phase gradient $\nabla_\phi \Phi_{D}(\phi)$ is $\phi$-dependent, which results in $\phi$-dependent amplitude of scattered wave, generating additional vortex beams with $l \neq 2$.\cite{kotlyar2014asymmetricModes,kotlyar2014asymmetricBeams}
\begin{eqnarray}
E(r,\phi)_{-+} &=& BG_2(r)f(r,\phi) \exp\{i2\phi+ i\Phi_D^{-+}(\phi)\}\nonumber\\
&=& \sum_l^\infty \tilde{a}_l BG_l(r)e^{il\phi}\nonumber
\end{eqnarray}
where scattering wave is partial-wave expanded in term of vortex beams of topological charge $l$ with $BG_l (r) \equiv J_l(k_0r)e^{-{k_0}^2r^2}$, and expansion coefficients $\tilde{a}_l={a}_le^{i\psi_l}$ are complex, and $f(r,\phi)=1$ when $\nabla_\phi \Phi_{D}^{-+}$ vanishes. A similar expression holds for $E(r,\phi)_{+-}$.
Owing to the difference in scattering amplitudes from nano-antennas with different sizes, polarization states of scattered waves form a spiral trajectory on Poincar\'e sphere as shown in \textbf{Fig.~\ref{nonconstantPB} b}. In other words, solid angles subtended by polarization states between two neighboring nano-antennas are not constant.
Rotational symmetry breaking introduced by nano-antennas with different sizes yields a spin-dependent non-constant azimuthal PB phase gradient, $\nabla_\phi \Phi_{PB}$, permitting spin-dependent deflections in SOC as shown in \textbf{Fig.~\ref{nonconstantPB} c}.
Analysis of PB phase in rotation symmetry-broken circular array of nano-antennas shows that
$\phi$-dependence of beam profiles $I_{-+}(r,\phi)$ and $I_{+-}(r,\phi)$ is determined by the total phase $\Phi(\phi)^{-+} = 2\phi + \Phi_D^{-+}(\phi)$ and $\Phi(\phi)^{+-} = -2\phi + \Phi_D^{+-}(\phi)$ acquired by cross polarization scattering beam, where non-constant azimuthal dynamical phase gradients $\nabla_\phi \Phi_{D}$ are responsible for $\phi$-dependence behavior of beam profiles. The occurrence of spin-dependent transverse shift, equivalent to non-vanishing $\Delta I \equiv I_{-+}-I_{+-}$, depends on the relation between $\Phi_D^{-+}(\phi)$ and $\Phi_D^{+-}(\phi)$ dictated by symmetry property of the circular array of nano-antennas with different sizes.
\section{Sample design and measurement}
\label{sec:sample}
For a systematic study of relation between symmetry property of metasurface and spin-dependent transverse shift, tapered arc (TA) antennas are arranged in a circle to
form metasurfaces belonging to cyclic group $C_n$, called tapered arc cyclic group symmetric metasurface (TA-CGSM), where in-plane inversion symmetry is determined by the order $n$ of cyclic group, as shown in \textbf{Fig.~\ref{samplestructure} a}.
Six different TA-CGSMs are fabricated, each TA-CSGM being composed of multiple TA antennas with varying width from 45$nm$ to 150$nm$ organized in $8$ azimuthal segments of concentric rings repeated with 600$nm$ radial spacing.
We used 1$mm$ thick borosilicate glass round substrates with diameter of one inch (UQG Optics). First, the substrates were cleaned in sequential bathes of acetone and isopropanol, assisted by ultrasonicating. Then the substrates were rinsed in deionized water and dried by nitrogen flow followed by oxygen radical plasma treatment in a barrel reactor (Nanoplas, France) to activate the glass surface.
Second, an electron beam resist (PMMA, ARP 679) and a conductive polymer (SX AR PC 5000/90.2 from Allresist, Germany) layers were successively spin-coated. The thickness of the resulting layer was chosen to be about 70$nm$ and it was measured by a contact profilometer (DektakXT, Bruker, Germany) after e-beam exposition and resist development.
Third, we used an e-beam lithography (EBL) tool (Pioneer, Raith, Germany) for sample patterning. We used 20$kV$ acceleration voltage of the electron gun, and the beam current was 0.016$nA$ for the aperture of 7.5$\mu m$. The typical working distance was about 5$mm$, and the nominal exposition dose was chosen in the range from 100 to 180$\mu C/cm^2$.
Forth, two successive metal layers (Cr, 3$nm$ for seeding and Au, 27$nm$) were deposited by thermal evaporation (Auto 306, Edwards, UK). The thickness was monitored by quartz crystal microbalance during the deposition and measured by a contact profilometer after lift-off process. The lift-off was done in ethyl lactate solution, followed by rinsing and nitrogen drying.
Finally, the samples were observed by optical and scanning electron microscopy (SEM). For SEM observation, we used low acceleration voltage of 3$kV$ in order to decrease the sample charging because the glass substrate is insulating. (For more information, see Supplementary Information)
\textbf{Fig.~\ref{samplestructure} b} shows optical microscope images of TA-CGSMs possessing $C_\infty$, $C_1$, $C_2$, $C_3$, $C_4$, $C_5$, and $C_6$ symmetries, and SEM images of $C_\infty$ and $C_1$ TA-CGSMs are shown in \textbf{Fig.~\ref{samplestructure} c}.
Among experimental measurements in \textbf{Fig.~\ref{samplestructure} e, f}, $C_\infty$ TA-CGSM exhibits degenerate $\phi$-independent intensity profiles for $I_{-+}$ and $I_{+-}$,
while $C_n (n\neq\infty)$ TA-CGSMs exhibits $\phi$-dependent intensity profiles of $I_{-+}$ and $I_{+-}$, owing to non-constant azimuthal PB phase gradient $\Phi_{PB}$ coming from plasmonic resonance of TAs.
Difference $\Delta I = I_{-+}-I_{+-}$ in \textbf{Fig.~\ref{samplestructure} g} corresponds to spin- and orbital-Hall shift, observed only in $C_1$, $C_3,$ and $C_5$ TA-CGSMs (odd order $n$ of $C_n$), where degeneracy of geometric structure is lifted by in-plane inversion symmetry breaking.\cite{slobozhanyuk2016enhanced}
Vortex charge of cross-polarization scattering beam from $C_n (n\neq\infty)$ TA-CGSM is identified by an interference pattern between scattering vortex and incidence Gaussian beams,
and counter-clockwise and clockwise twisted fringes confirm topological charges of $l=2$ and $l=-2$ of $I_{-+}$ and $I_{+-}$, respectively. See \textbf{Fig.~S2} in Supplementary Information. FDTD calculation is in a perfect agreement with experimental results. See \textbf{Fig.~S1} in Supplementary Information.
\section{Partial wave decomposition}
\label{sec:partialwave}
Vortex beam with topological charge $l$ is described by Bessel-Gaussian beam,\cite{mendoza2015laguerre}
\begin{eqnarray}{{J}_{l}}(k_0r){{e}^{-{{{k_0}^2r}^{2}}}}\exp(i l \phi) = {BG}_{l}(r)\exp(i l \phi).\nonumber
\end{eqnarray}
In order to clarify how the addition of dynamical phase to geometric phase in SOC alters $\phi$-dependence of phases $\Phi^{-+}(\phi)$ and $\Phi^{+-}(\phi)$, experimentally measured intensity profiles in \textbf{Fig.~\ref{samplestructure} e, f} are fit as a linear superposition of vortex beams, as shown in \textbf{Fig.~\ref{Results1cal} a, b} along with phases in \textbf{Fig.~\ref{Results1cal} d, e}.
Spin-dependent vortex beams with asymmetric helical wave front are described by
\begin{eqnarray}
{{{E}}(r,\phi)}=\sum\limits_{l}{{{\tilde{a}}_{l}}}{{{BG}}_{l}(r)}{{{e}}^{i l\phi}}= A(r,\phi )e^{i{\Phi }_{\rm tot}},\nonumber
\end{eqnarray}
where $\tilde{a}_l$ is the $l$-th order complex coefficient of the Fourier expansion, i.e., $\tilde{a}_l={a}_l e^{i\psi_l}$, $BG_l (r) = J_l(k_{0}r)e^{-{k_{0}}^2r^2}$, $A(r,\phi)=\left|E(r,\phi)\right|$ is the amplitude, ${{\Phi }_{\rm tot}}= \gamma (\mathbf{C})+{\Phi}_{D} = \tan^{-1}(\textit{Im}(E)/\textit{Re}(E))$ is total phase. It shows that the far fields are determined by the local topological features of interfering fields.
The fit values of partial-wave expansion coefficient $\tilde{a}_l$ are listed in \textbf{Table 1}.
We observe two important features.
First, phases of $E_{-+}(\phi)$ and $E_{+-}(\phi)$ are opposite, i.e., $\Phi^{-+}(\phi)= - \Phi^{+-}(\phi)$, which is from the opposite senses of rotation of polarization state trajectories on Poincar\'e sphere for $\{-+\}$ and $\{+-\}$. Second, intensity profile $I_{+-}(\phi)$ is the same as that of in-plane inversion symmetry operated $I_{-+}(\phi)$, i.e., $I_{+-}(\phi) = I_{-+}(\phi+\pi)$, as a result of $\Phi^{-+}(\phi)= - \Phi^{+-}(\phi)$. The two features immediately explain why spin- and orbital-Hall effect takes place only for $C_n$ with odd $n$. For even $n$, $\Delta I$ vanishes identically, since geometric structures possess in-plane inversion symmetry.
Now we further clarify the contributions of dynamical phase to spin-controlled spatial beam profiling in SOC.
First, we expect a correlation between intensity profile and $\phi$-dependence of phase. As shown in \textbf{Fig.~\ref{matlab}} for $C_1$ TA-CGSM,
$E_{-+}(\phi)$ and $E_{+-}(\phi)$ have a local maximum at the azimuthal angles $\phi$ where the absolute value $|\nabla_\phi\Phi_D|$ is a local minimum. Helical wavefronts with azimuthal geometric phase gradient $\pm2$ are altered by introduction of non-constant azimuthal dynamical phase gradient $\nabla_\phi\Phi_D$, to build focused spots along azimuthal direction, similar to wavefront modification when a plane wave goes through a medium with spatially varying index.\cite{saleh1991fundamentals} Refer to Supplementary Information for $C_n$ TA-CGSMs with $n$ other than $1$. The intensity profile possesses a feature of OAM superposition states, useful for quantum information systems.\cite{franke2008orbital}
Second, the phase $\psi_l$ of partial-wave expansion coefficient $\tilde{a}_l$ is related to dynamical phase $\Phi_D(\phi)$, determining interference behavior between partial-wave vortex beams having difference topological charge $l$s. $C_4$ and $C_5$ TA-CGSMs are closely examined in the left panel (\textbf{Fig.~\ref{Results1cal} f, g, h, i}) and right panel (\textbf{Fig.~\ref{Results1cal} j, k, l, m}).
In $C_4$ TA-CGSM
\begin{eqnarray}
E_{-+} = BG_2 \exp\{i 2 \phi\}+ 0.24 BG_{-2}\exp\{-i2\phi\} \exp\{+i{3\pi\over 2}\}\nonumber
\end{eqnarray}
and
\begin{eqnarray}
E_{+-} = BG_{-2}\exp\{-i2\phi\} + 0.24 BG_{2}\exp\{i 2 \phi\} \exp\{-i{3\pi\over 2}\},\nonumber
\end{eqnarray}
where azimuthal angles for constructive and destructive interferences are the same for $E_{-+}$ and $E_{+-}$, resulting in $\Delta I = 0$.
In $C_5$ TA-CGSM
\begin{eqnarray}
E_{-+} = BG_2 \exp\{i2\phi\}+ BG_{-3}\exp\{-i3\phi\} \exp\{+i\pi\}\nonumber
\end{eqnarray}
and
\begin{eqnarray}
E_{+-} = BG_{-2}\exp\{-i2\phi\} + BG_{3}\exp\{i3\phi\} \exp\{-i\pi\},\nonumber
\end{eqnarray}
where azimuthal angles for constructive and destructive interferences differ by $0.2\pi$ for $E_{-+}$ and $E_{+-}$, and $\Delta I \neq 0$ leading to non-vanishing spin-dependent transverse shift.
Third, a frequency dispersion of dynamical phase due to the plasmonic resonance of TA renders intensity profiles to be dependent on the incidence beam wavelength, confirmed to be true.
\section{Wavelength dependence of extrinsic spin- and orbital-Hall effect}
\label{sec:wavelength}
The role of dynamical phase can be identified by examining wavelength dependence of the spin- and orbital- Hall effect since plasmonic resonance of nano-antennas is wavelength-dependent.
We measured the spin- and orbital-dependence of far field intensity $C_1$ metasurface by employing lasers with different wavelengths as shown in \textbf{Fig. \ref{Results2}}.
Single-mode fiber pigtailed laser diodes with wavelengths $\lambda =1,310nm$, $\lambda =730nm$, and $\lambda =660nm$ are adopted as normal-incidence illumination sources.
Direction of the transverse shift goes through clock-wise rotation as the wavelength gets shorter. While geometric phase is wavelength-independent, dynamical phases originating from multiple TA antennas with varying widths give rise to wavelength- and $\phi$-dependent far-field intensity distributions.
On the other hand, ${C_1}'$ metasurface possessing $\sigma_{yz}$ symmetry, i.e., reflection symmetry with respect to $yz$-plane, allows for splitting of vortex beam along $x$-direction only as shown in \textbf{Fig. \ref{Results2} g-i}, and direction of the transverse shift is fixed along $x$-direction as the wavelength get shorter, exhibiting no wavelength-dependence. We note that in ${C_1}'$ only $y$-direction dynamical phase gradient gets involved owing to breaking of $y$-direction inversion-symmetry. Normalized horizontal $(X)$ and vertical $(Y)$ spin- and orbital-Hall effect profiles, $\Delta I (X)$ and $\Delta I (Y)$ along the dashed coordinate axes, are displayed in two right panel, where $\Delta I $=$I_{-+}(r,\phi)-I_{+-}(r,\phi)$.
\section{Conclusion}
\label{sec:conclusion}
In summary we demonstrated spin- and orbital-Hall effect in metasurfaces possessing cyclic point group symmetry $C_n$.
Lifting degeneracy in orbital angular momentum is achieved by selectively controlling the symmetry order $n$.
By identifying the role of dynamical phase in giving rise to spin-dependent transverse shift in SOC, spatial separation and modification of vortex beam profile are made possible.
This new approach provides significant advantages in applications such as vortex multiplexing in communication device and vortex beam analyzer as well as fundamental understating of interactions among angular momenta of light in metasurface.
\begin{table*}[t]
\centering
\label{my-label}
\begin{tabular}{|c|
>{\columncolor[HTML]{9B9B9B}}c ||c|c|c|c|c|c|c||c|c|c|c|c|c|c|}
\hline & \multicolumn{1}{c||}{\cellcolor[HTML]{C0C0C0}\textbf{ }}
& \multicolumn{7}{c||}{\cellcolor[HTML]{C0C0C0}\textbf{$E_{-+}$}} & \multicolumn{7}{c|}{\cellcolor[HTML]{C0C0C0}\textbf{$E_{+-}$}} \\ \hline
& \textbf{$l$} & \cellcolor[HTML]{9B9B9B}\textbf{-4} & \cellcolor[HTML]{9B9B9B}\textbf{-3} & \cellcolor[HTML]{9B9B9B}\textbf{-2} & \cellcolor[HTML]{9B9B9B}\textbf{-1} & \cellcolor[HTML]{9B9B9B}\textbf{0} & \cellcolor[HTML]{9B9B9B}\textbf{1} & \cellcolor[HTML]{9B9B9B}\textbf{2} & \cellcolor[HTML]{9B9B9B}\textbf{-2} & \cellcolor[HTML]{9B9B9B}\textbf{-1} & \cellcolor[HTML]{9B9B9B}\textbf{0} & \cellcolor[HTML]{9B9B9B}\textbf{1} & \cellcolor[HTML]{9B9B9B}\textbf{2} & \cellcolor[HTML]{9B9B9B}\textbf{3} & \cellcolor[HTML]{9B9B9B}\textbf{4} \\ \hline\hline
& \textbf{${{{a}}_{l}}$} & 0 & 0 & 0 & 0.02 & 0.01 & 0.1 & 1 & 1 & 0.1 & 0.01 & 0.02 & 0 & 0 & 0 \\ \cline{2-16}
\multirow{-2}{*}{$C_1$}
& \textbf{$\psi_{l}$}
& 0 & 0 & 0 & 0 & $\frac{\pi}{2}$ & $\pi$ & 0 & 0 & $-\pi$ & $-\frac{\pi}{2}$ & 0 & 0 & 0 & 0 \\ \hline
& \textbf{${{{a}}_{l}}$} & 0 & 0 & 0.12 & 0 & 0.02 & 0 & 1 & 1 & 0 & 0.02 & 0 & 0.12 & 0 & 0 \\ \cline{2-16}
\multirow{-2}{*}{$C_2$} & \textbf{$\psi_{l}$}
& 0 & 0 & $\frac{3\pi}{2}$ & 0 & $\frac{\pi}{2}$ & 0 & 0 & 0 & 0 & $-\frac{\pi}{2}$ & 0 & $-\frac{3\pi}{2}$ & 0 & 0
\\ \hline
& \textbf{${{{a}}_{l}}$}
& 0 & 0 & 0 & 0.05 & 0 & 0 & 1 & 1 & 0 & 0 & 0.05 & 0 & 0 & 0 \\ \cline{2-16}
\multirow{-2}{*}{$C_3$} & \textbf{$\psi_{l}$}
& 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \\ \hline
& \textbf{${{{a}}_{l}}$}
& 0 & 0 & 0.24 & 0 & 0 & 0 & 1 & 1 & 0 & 0 & 0 & 0.24 & 0
& 0 \\ \cline{2-16}
\multirow{-2}{*}{$C_4$} & \textbf{$\psi_{l}$}
& 0 & 0 & $\frac{3\pi}{2}$ & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & $-\frac{3\pi}{2}$ & 0 & 0 \\ \hline
& \textbf{${{{a}}_{l}}$} & 0 & 1 & 0 & 0 & 0 & 0 & 1 & 1 & 0 & 0 & 0 & 0 & 1 & 0 \\ \cline{2-16}
\multirow{-2}{*}{$C_5$} & \textbf{$\psi_{l}$}
& 0 & $\pi$ & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & $-\pi$ & 0 \\ \hline
& \textbf{${{{a}}_{l}}$}
& 1.74 & 0 & 0 & 0 & 0 & 0 & 1 & 1 & 0 & 0 & 0 & 0 & 0 & 1.74 \\ \cline{2-16}
\multirow{-2}{*}{$C_6$} & \textbf{$\psi_{l}$}
& $\frac{\pi}{2}$ & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & 0 & $-\frac{\pi}{2}$ \\ \hline
\end{tabular}
\centering
\caption{Fit values of partial-wave expansion coefficient $\tilde{a}_l$, the $l$-th order complex coefficient}
\end{table*}
\clearpage
\section*{Acknowledgement}
JWW acknowledges the support from the Ministry of Science, ICT \& Future Planning, Korea (2015001948, 2014M3A6B3063706). Nanofabrication processes were performed in PLANETE cleanroom facility, CT PACA.
| ade7eb14b08b55267ea7e071d08cf0f07e6cddad | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
\label{sec:intro}
The precise measurement of the polarization of the Cosmic Microwave Background (CMB) is an active area of experimental physics, motivating a variety of current and upcoming polarimeter projects.
The CMB temperature anisotropy, measured to extreme precision over the last two decades, has provided evidence for the standard model of Big Bang cosmology and hints at an inflationary period. In the same way, the CMB polarization, and in particular the curl `B-mode' pattern, contains a wealth of cosmological information on many scales and continues to be targeted by experimentalists. At arcminute scales on the sky, the dominant B-mode signal is primarily due to the gravitational lensing of the much brighter, curl-free `E-mode' signal. This lensing B-mode signal probes the large scale structure of the universe that sits between the CMB surface and ourselves and has recently been detected by P\textsc{olarbear}\cite{pbear2014}, SPTpol\cite{sptpol2015}, and \bicep2/\textit{Keck}\,\textit{Array}\cite{BKI,BKVI}.
At larger, degree-scales on the sky, the B-mode signal due to inflationary gravitational waves (IGW) is predicted to peak with an as-yet-unmeasured amplitude.
During inflation, quantum fluctuations are expected to produce both density and gravitational wave perturbations, the latter of which imprints a B-mode pattern on the CMB. The ratio of the amplitude of the primordial tensor spectrum to the scalar spectrum, parametrized as $r$,
is a measurable quantity that will constrain inflationary models.
To this end, experiments aim to continually lower the upper-limit on, and eventually detect, the IGW B-mode signal and the resulting $r$ parameter, a recent summary of which is given in Ref.~\citenum{kam2016}.
The B\textsc{icep}/\textit{Keck}\ series of South Pole-based CMB polarimeters has focused efforts on this IGW B-mode signal, taking the unique approach of using compact, refracting telescope receivers that only resolve the larger degree-scale features necessary. This approach precludes science at the smaller scales but has several advantages for IGW signal sensitivity.
The first experiment, \bicep1, observed for three seasons from 2006--2008 using 98 feedhorn-coupled bolometers (split primarily between 100 and \SI{150}{GHz} band centers) and found a 95\% confidence upper limit of $r<0.70$\cite{barkats2014}.
\bicep2 followed with three observing seasons from 2010--2012, using 500 antenna-coupled transition edge sensor (TES) bolometers centered at 150 GHz. It detected on-sky B-mode signal,
later found to be consistent with galactic dust foregrounds, for a combined limit $r<0.12$ at 95\% confidence with \textit{Keck}\,\textit{Array}\ and \textit{Planck} data\cite{BKI,bkp2015}.
While \bicep2 was still operating, \textit{Keck}\,\textit{Array}\ began observing on a nearby telescope mount, with first science observations in 2012.
\textit{Keck}\,\textit{Array}\ is made up of 5 \bicep2-class receivers sharing a common boresight. It is currently in its fifth observing season and has now included receivers centered on multiple frequencies (95, 150, and \SI{220}{GHz}) in order to help distinguish polarized galactic foregrounds from the IGW signal. The latest results give $r<0.09$ at 95\% confidence\cite{BKVI}, or $r<0.07$ when combined with external datasets.
\bicep3\ replaces \bicep2\ and is now in its second observing season (and first science season) during the 2016 austral winter, with a full complement of 2560 TES bolometers at 95 GHz.
With its larger aperture, faster optics, and higher detector count,
\bicep3\ acts as a pathfinder receiver for the 3rd-generation of the B\textsc{icep}/\textit{Keck}\ series. Its relatively compact design with on-axis optics again lends itself to multi-frequency observations, in the same way that \bicep2\ was evolved into the \textit{Keck}\,\textit{Array}. Such an expansion to an array is highly motivated by the continued need for increasing observation depth combined with discrimination of galactic foregrounds, and is easily realized by upgrading the \textit{Keck}\,\textit{Array}\ with \bicep3-class receivers.
In the following sections, these proceedings first give an overview of the \bicep3 instrument (Sec.~\ref{sec:inst}) and its South Pole observing location and strategy (Sec.~\ref{sec:obs}). It then summarizes \bicep3's current, second-season status and performance (Sec.~\ref{sec:perf}) before finally outlining plans for a future multi-frequency array of \bicep3-class receivers (Sec.~\ref{sec:array}).
\section{Instrument Overview}
\label{sec:inst}
Like its B\textsc{icep}/\textit{Keck}\ predecessors described in Sec.~\ref{sec:intro}, \bicep3\ is composed of a single cryostat, also called the receiver, which contains all of the detectors and optical components required for its measurement of the CMB. This compact design, which keeps all optics at cryogenic temperatures, maintains the practical and scientific advantages of the previous B\textsc{icep}/\textit{Keck}\ experiments. Namely,
\begin{itemize}
\item the cryostat lends itself to complete assembly and disassembly by a small crew of experienced people;
\item the compact, on-axis optics allow for symmetric rotation of the telescope as a whole, an important method for probing systematics in the polarization-based measurement;
\item the compact optics allow for ground-based characterization in the optical far-field;
\item the compact receiver size allows installation of a comoving absorptive forebaffle that terminates beam sidelobes and reduces stray pickup;
\item and the ability to cryogenically cool the optical elements and internal baffling minimizes the thermal photon noise seen by the detectors and leads to excellent stability.
\end{itemize}
This section provides a short overview of the \bicep3 instrument in its current, 2016 observing season state, offered here as both a recap before giving a status update in Sec.~\ref{sec:perf}, and to highlight features that will transfer over to a future \bicep3-class array as outlined in Sec.~\ref{sec:array}.
A more detailed overview before the first (2015) season deployment is found in Ref.~\citenum{ahmed2014}, and an in-depth design and operation summary during the 2015 season is provided in Ref.~\citenum{thesis:wu}.
\subsection{Cryostat Receiver}
The \bicep3 receiver is a custom aluminum dry cryostat, \SI{2.4}{m} tall along the optical axis with a \SI{0.73}{m} outer diameter and totalling roughly \SI{550}{kg} without attached readout electronics and forebaffle.
Fig.~\ref{fig:b3_rx_cutaway} shows a cutaway view of the \bicep3 CAD model in this state.
The top of the receiver is capped by a \SI{31.75}{mm} thick high-density polyethylene (HDPE) window through which light is collected, while the bottom end is used to interface to electronic subsystems. From the side protrudes a vacuum shell extension that houses the single PT415\footnote{Cryomech Inc., Syracuse, NY 13211, USA (\texttt{www.cryomech.com})} two-stage pulsetube cryocooler, which provides continuous cooling to $<$\SI{4}{K}.
\begin{figure} [t]
\begin{center}
\begin{tabular}{c}
\includegraphics[width=.8\textwidth]{img/b3_rx_cutaway}
\end{tabular}
\end{center}
\caption[b3rxcutaway]
{ \label{fig:b3_rx_cutaway}
Cutaway view of the \bicep3 cryostat receiver, with key elements labeled.
}
\end{figure}
The cryogenic design can be simply described as concentric cylindrical aluminum shells, each an increasingly cold and thermally isolating radiation shield that together protect the
sub-Kelvin detectors from warm radiation.
Moving inward from the room-temperature vacuum shell are the nominal `\SI{50}{K}' and `\SI{4}{K}' stages, respectively cooled by the 1st and 2nd stages of the pulsetube. The pulsetube is thermally linked to the \SI{50}{K} and \SI{4}{K} stages in the main body of the cryostat via oxygen-free high-conductivity (OFHC) copper heat straps, including flexible foil and braid sections to minimize transmission of vibrational energy from the pulsetube. Within the \SI{4}{K} shield is a series of stacked planar `sub-Kelvin' stages (\SI{2}{K}, \SI{350}{mK}, \SI{250}{mK}) culminating in the \SI{250}{mK} focal plane (Fig.~\ref{fig:subk}). The various temperature stages are structurally supported and thermally isolated from each other by trusses of low thermal-conductivity materials (G10 fiberglass and carbon fiber).
\begin{figure} [t]
\begin{center}
\begin{tabular}{c}
\includegraphics[width=.7\textwidth]{img/subk}
\end{tabular}
\end{center}
\caption[subk]
{ \label{fig:subk}
The \bicep3 sub-Kelvin stages exposed during assembly, with various components labeled.
}
\end{figure}
A three-stage ($^4$He/$^3$He/$^3$He) helium sorption fridge\footnote{Chase Research Cryogenics Ltd., Sheffield, S10~5DL, UK (\texttt{www.chasecryogenics.com})} is mounted on the \SI{4}{K} stage near the pulsetube and provides non-continuous cooling for all three sub-Kelvin stages. The three stages of the fridge contain \SI{55}{L} of $^4$He and \SI{33}{L}/\SI{4}{L} of $^3$He at standard temperature and pressure, able to cool \bicep3's detector modules and large copper focal plane from \SI{4}{K} to \SI{250}{mK} and then maintain operating temperatures for more than 48 hours
before needing to re-cycle.
The \SI{250}{mK} stage is separated into two planar copper stages to provide thermal stability for the detectors. The lower level, closest to the \SI{350}{mK} stage, is actively temperature-controlled to \SI{269}{mK} by a resistive heater and acts as a buffer for the more isolated focal plane. Stainless steel blocks separate the two levels and provide low-pass thermal filtering. The focal plane itself is additionally temperature-controlled to \SI{274}{mK}.
\subsection{Detectors and Readout}
\bicep3's focal plane is populated by planar arrays of polarization-sensitive slotted antennas, coupled to transition edge sensor (TES) bolometers and centered at \SI{95}{GHz}, the same technology also fabricated by JPL/Caltech for \bicep2/\textit{Keck}\,\textit{Array}\ and detailed in Ref.~\citenum{dets2015}.
Each silicon detector tile contains an $8\times8$ array of pixels with each pixel made up of two co-located, orthogonally-polarized sub-antenna networks and two TES bolometers.
A Gaussian taper of each pixel's beam is achieved by tuning the summing tree that couples each bolometer to its sub-antenna network\cite{obrient2012}.
A major difference from \bicep2/\textit{Keck}\,\textit{Array}\ is the installation method of detector tiles on the focal plane.
In \bicep2/\textit{Keck}\ receivers, four detector tiles are directly wire bonded to the focal plane, and the supporting readout electronics sit adjacent at the focal plane's outer radius.
\bicep3\ instead packages each detector tile and its electronics in a more easily installed compact detector module. The concurrently presented Ref.~\citenum{spie:howard} describes these detector modules in more detail.
Twenty such detector modules are tiled onto the \bicep3 focal plane, each containing 128 TES bolometers, for a total of 2560. Each detector tile contains 8 `dark' TES bolometers that are intentionally left uncoupled to antennas, 4 of which are read out as non-optical probes, giving a total of 2400 optically-coupled bolometers or, equivalently, 1200 polarization-sensitive detector pairs centered at \SI{95}{GHz}. This contrasts to a single \bicep2/\textit{Keck}\,\textit{Array}\ receiver at \SI{150}{GHz}, which has four detector tiles for a total of 248 polarization-sensitive detector pairs. At \SI{95}{GHz}, a single \textit{Keck}\,\textit{Array}\ receiver provides an even sharper comparison with 136 polarization-sensitive detector pairs, owing to its slower optics and smaller aperture than \bicep3.
\bicep3 continues to use time-domain multiplexed (TDM) readout of the TES bolometers, via SQUID MUX chips developed by NIST, in order to reduce the total number of lines that must enter the cryostat and their associated heat load.
The signal current through each voltage-biased TES is inductively coupled to its own 1st-stage SQUID series array, forming one column of one multiplexed `row'. A pair of 11-row MUX chips is wire bonded in series in order to allow multiplexing through 22 rows. Each detector module contains 6 pairs of MUX chips, forming 6 columns of 22 rows, or 132 total 1st-stage channels. Each column's multiplexed output is inductively coupled to a separate SQUID series array on the \SI{4}{K} stage for additional amplification before exiting the cryostat.
\bicep3 uses MUX11d chips, a newer generation than those used by \bicep2/\textit{Keck}\,\textit{Array}\cite{dekorte2003} that has a different row-switching mechanism. Details of the new design are found in the concurrently presented Ref.~\citenum{spie:howard}.
Control of the MUX system and feedback-based readout of the TES data are via room temperature Multi-Channel Electronics (MCE) systems developed by the University of British Columbia\cite{mce2008}, again similar to those used in \bicep2/\textit{Keck}\,\textit{Array}.
\subsection{Optics and IR Filtering}
\label{subsec:op}
Following the other B\textsc{icep}/\textit{Keck}\ experiments, \bicep3 targets degree-scale polarization features by using a compact, on-axis, two-refractor optical design. Its $f/1.7$ system with a \ang{27.4} field of view (FOV) is faster and wider than the $f/2.2$, FOV$\sim$\ang{15} optics of \bicep2/\textit{Keck}\,\textit{Array}. It has a significantly larger aperture diameter of \SI{520}{mm}, vs. \SI{264}{mm} for \bicep2/\textit{Keck}\,\textit{Array}, defined by a millimeter-wave absorbing aperture stop. Both refracting lenses and the aperture stop are heatsunk to the \SI{4}{K} stage, with the objective lens (warmest and furthest from the focal plane) operating at \SI{4.9}{K}. The lenses are made of 99.6\% pure alumina ceramic, which allows a thinner design and smaller thermal gradients than the HDPE plastic used for previous B\textsc{icep}/\textit{Keck}\ optics, crucial advantages for \bicep3's faster system and increased infrared (IR) thermal loading.
The much larger diameter of the \bicep3 receiver window ($\sim$\SI{670}{mm} clear), necessitated by the increase in aperture size, creates a large IR radiative heat load. IR power entering the cryostat is in excess of \SI{100}{W}, more than $2\times$ the cooling power of the 1st stage of the single PT415 pulsetube cooler. The first level of IR filtering therefore attempts to reject the majority of IR loading before it reaches the \SI{50}{K} stage,
and consists of a stack of 10 metal-mesh IR-reflective filters mounted just within the receiver window at room temperature. These filters are made up of $\mathcal{O}$(\SI{10}{$\mu$m}) capacitive squares created by laser ablation of aluminized thin plastic films (\SI{3.5}{$\mu$m} Mylar or \SI{6}{$\mu$m} polypropylene (PP)/polyethylene (PE) blend), as introduced in Ref.~\citenum{mesh2014}. In all, through both reflection and absorption/reemission, the metal-mesh filter stack allows $\sim$\SI{19}{W} to reach the \SI{50}{K} stage. One additional metal-mesh filter is mounted just behind the \SI{50}{K} alumina filter. The remaining power is absorbed by a series of three filters (and thereby removed by the pulsetube) in order to shield the sub-Kelvin detector stage: a \SI{10}{mm} thick alumina filter on the \SI{50}{K} stage, between the metal-mesh filter stack and the objective lens; a \SI{5}{mm} thick nylon filter on the \SI{4}{K} stage, between the two lenses; and a second \SI{4}{K} nylon filter, \SI{9.5}{mm} thick, between the field lens and the focal plane. These IR filters and lenses are labeled in Fig.~\ref{fig:b3_rx_cutaway}.
Finally, low-pass metal-mesh edge filters\cite{ade2006} with a \SI{4}{cm$^{-1}$} (\SI{120}{GHz}) cutoff are mounted onto each detector module to prevent any remaining out-of-band sub-millimeter blue leaks from coupling to the detector bolometers.
\section{Observing Site and Strategy}
\label{sec:obs}
The \bicep3 receiver is integrated with its telescope mount and various subsystems in the Dark Sector Laboratory (DSL) building at the South Pole, from where it observes the sky during the austral winter.
Support and logistics are managed through the US National Science Foundation's Amundsen-Scott South Pole Station, located \SI{1}{km} away, including satellite-based daily transfer of the full science data stream. This cold and remote observing site sits at an elevation of over \SI{2800}{m} on the Antarctic Plateau and provides extremely dry and stable atmospheric conditions, optimal for \bicep3's millimeter-wave observations.
The warmed indoor environment of DSL is extended beyond the roof level by a flexible insulating boot so that only the receiver window is exposed to the outdoor environment (Fig.~\ref{fig:b3_roof}).
A dry nitrogen gas system fills a protective membrane above the receiver window to prevent snow accumulation, and a cylindrical comoving microwave-absorptive forebaffle extends beyond the receiver window to intercept stray light. A reflective, stationary ground shield is fixed to the roof of DSL and acts as a second barrier against signal contamination from nearby ground sources.
\begin{figure} [t]
\begin{center}
\begin{tabular}{c}
\includegraphics[width=.5\textwidth]{img/b3_roof}
\end{tabular}
\end{center}
\caption[b3roof]
{ \label{fig:b3_roof}
View of \bicep3 from the roof of DSL, showing the green insulating boot, comoving forebaffle, and reflective ground shield. The larger \textit{Keck}\,\textit{Array}\ ground shield is visible in the background and the main Amundsen-Scott station extends along the horizon.
}
\end{figure}
\bicep3 continues to use the original three-axis \bicep1 mount in DSL that also supported \bicep2. By removing some non-essential structures, the mount was modified to fit \bicep3's larger size while retaining the same functions: ranges of \textgreater\ang{360} in azimuth, \ang{50}--\,\ang{90} (zenith) in elevation, and boresight rotation about the optical axis (Fig.~\ref{fig:b3_axes}). Due to the larger diameter of \bicep3 and the protruding vacuum shell extension that houses the pulsetube cooler, boresight rotation is limited by mechanical interference to a range of \ang{255}.
All electrical cabling, high-pressure helium gas hoses for the pulsetube, and other necessary lines are routed through cable carriers for the three axes of motion, with the exception of a 4-channel high-pressure rotary joint\footnote{Dynamic Sealing Technologies, Inc., Andover, MN 55304, USA (\texttt{www.dsti.com})} used to bypass the two helium hoses at the azimuth stage. By pressurizing two outer channels to act as buffers to atmosphere, the 4-channel rotary joint achieves much better leak performance than the 2-channel unit used during \bicep3's 2015 season; the optimal pulsetube system pressure is maintained by small, non-disruptive refills to the compressor less than once a month.
\begin{figure} [t]
\begin{center}
\begin{tabular}{c}
\includegraphics[width=.6\textwidth]{img/b3_axes}
\end{tabular}
\end{center}
\caption[b3axes]
{ \label{fig:b3_axes}
\bicep3 observing on the original B\textsc{icep}\ mount, with the three axes of motion labeled.
}
\end{figure}
\bicep3 focuses its CMB observations on the same sky patch as the \textit{Keck}\,\textit{Array}\ and previous B\textsc{icep}\ experiments, spanning right ascension $\ang{-60}<\textrm{RA}<\ang{60}$ and declination $\ang{-70}<\delta<\ang{-40}$.
Its larger optical FOV results in an effective sky area of $\sim$ \SI{600}{deg$^2$}, larger than the $\sim$ \SI{400}{deg$^2$} area covered by \bicep2/\textit{Keck}.
The fundamental observing block is a constant-elevation `scanset' of 50 back-and-forth \SI{2.8}{deg\per s} fixed-center azimuth scans spanning \ang{64.4}. Each scanset is bookended by a \SI{26}{s} elevation nod of \ang{1.3} peak to peak for airmass-based detector gain calibration and $\sim$ \SI{1}{minute} of partial TES loadcurves (non-superconducting) to monitor detector parameters, for a total duration of $\sim$ 50 minutes. The azimuth center is adjusted every other scanset to track the changing RA of the sky patch, using that drift in combination with the field of view to fill in the full range in RA. Elevation is staggered by twenty \ang{0.25} offsets amongst the scansets to fill in coverage of the detector beam pattern.
Observations are on a two-sidereal day cadence, including 40 CMB scansets and 7 scansets aimed at the galactic plane. Each two-day cycle uses one of four boresight angles within the allowed boresight envelope to help constrain polarization systematics: two sets of \ang{180} opposing orientations, offset from each other by \ang{45}, and clocked to optimize coverage symmetry of both the Stokes $Q$ and $U$ polarizations on the sky. The first \SI{6.5}{hrs} of each cycle is spent re-cycling the helium sorption fridge to keep the detectors cold for the rest of the two days.
\section{Current Performance}
\label{sec:perf}
\bicep3's first season performance, during the 2015 austral winter, is summarized in Ref.~\citenum{wu2015}. During that initial engineering season, 9 of the possible 20 detector modules were installed, for a total of 540 polarization-sensitive detector pairs centered at \SI{95}{GHz}. Yield after accounting for known hardware issues (e.g. suspected broken wire bonds, SQUID MUX chip yield) was 436 good polarization-sensitive pairs, or 81\%. 51 of the 104 lost pairs were attributed to a single anomalously low-yield detector module, potentially due to wire bond or other damage to the detector tile during installation and receiver cool-down.
Median per-detector CMB noise-equivalent temperature (NET) for the 2015 season is estimated at \SI{449}{$\mu$K$\sqrt{\textrm{s}}$}, based on polarization pair-differenced timestream data as shown in Fig.~\ref{fig:netspec} and explained later in this section.
\bicep3 underwent a number of improvements during the 2015/16 austral summer in preparation for its second season in the 2016 austral winter, which is the first dedicated science season. As originally planned, the focal plane was populated with the full complement of 20 detector modules, including 12 new detector modules and the removal of the anomalously low-yield module mentioned above.
Additionally, several important changes were made in response to data and operations knowledge collected during the 2015 season.
First, cryogenic improvements have added over 20\% to the helium sorption fridge duty cycle and allow the current two-day observing schedule cadence. During the 2015 season, continuous observing time was limited by the fridge to \SI{12}{hrs}. This cryogenic fix came in two parts, both vetted by parallel lab testing in North America during \bicep3's 2015 season observations. Lowering of the helium sorption fridge bath temperature by \SI{0.6}{K}, to \SI{3.5}{K}, was achieved by increasing thermal conduction through the copper heatstrap to the pulsetube's 2nd stage and thereby allowed increased helium condensation during the fridge re-cycling. Furthermore, a foil shielding included on a small number of electrical cables was measured to give an excess \SI{200}{$\mu$W} parasitic conductive heat load to the nominal \SI{350}{mK} stage, at least tripling the total expected load. Removal of this incorrectly specified shielding resulted in a large \SI{280}{mK} reduction on that stage, allowing it to reach \SI{360}{mK}, which in turn allowed lowering the focal plane operating temperature from \SI{313}{mK} to \SI{274}{mK} for the 2016 season.
Second, two faulty metal-mesh IR-reflective filters were replaced for 2016 and have significantly reduced broad and polarization-dependent scattering of the 95 GHz signal. This has been measured both as a per-detector 26\% median reduction of loading from the comoving forebaffle, and as an improvement in detector loading median mismatch from $\sim$~10\% to $<3$\% between co-located orthogonally polarized detectors when observing broad sources (forebaffle and atmosphere). Lab inspection has shown these faulty filters to be plagued by incomplete and inconsistent laser ablation of the aluminization on micron scales, now easily remedied with replacements and improved screening.
Finally, 2015 season data showed \bicep3 to be sensitive to \SI{450}{MHz} interference from the main South Pole Station radio system, owing to its larger receiver aperture than \bicep2/\textit{Keck}\,\textit{Array}\ (Sec.~\ref{subsec:op}) and correspondingly lower high-pass waveguide cutoff frequency. The interference caused a mostly ground-fixed signal, but also contributed to transients in the feedback-based detector readout. For the 2016 season, internal RF shielding improvements at the focal plane have given a $\sim$~\SI{10}{dB} reduction in RF signal sensitivity, and changes at the main station (reduced output power and a directional antenna) have reduced the signal strength at DSL by $\sim$~\SI{35}{dB}, for a total $\sim$~\SI{45}{dB} reduction.
As is routine for the B\textsc{icep}/\textit{Keck}\ experiments, \bicep3 was subject to several calibration measurements during the 2015/16 austral summer in order to characterize the instrument before the 2016 observation season.
Per-detector spectra were measured for all 20 modules with a custom Martin-Puplett Fourier Transform Spectrometer (FTS)\cite{karkare2014} mounted to the window of \bicep3 while in the mount.
Optical efficiency through the receiver was measured with an aperture-filling liquid nitrogen load mounted above the receiver window, which also acted as an ambient temperature source when unfilled.
Results from both measurements can be found in the concurrently presented Ref.~\citenum{spie:howard}.
Far-field beam characteristics were measured over several weeks using a chopped thermal source mounted high above the nearby Martin A. Pomerantz Observatory building (the location of \textit{Keck}\,\textit{Array}) and are covered in detail in the concurrently presented Ref.~\citenum{spie:kirit}.
As of writing in early June 2016, \bicep3 has observed for $\sim2.5$ months in its second season and accumulated roughly 1224 receiver-hours of new CMB data.
Overall detector yield due to known hardware issues is 951 good polarization-sensitive pairs, or 79\%. 180 of the 249 lost pairs are due to electrical open circuits that developed on 7 SQUID MUX row-select lines shared by a group of 5 detector modules. The cause of the bad connections is yet to be determined and only became present after cryogenic cooling of the receiver. Because these row-select lines are run in series, it is possible that poor wire bond adhesion within a single detector module could be the culprit.
The number of lost detector pairs is magnified due to polarization pairing across adjacent MUX rows (pairs are wired along the same MUX column, rather than row, sharing common parameters such as SQUID and TES biases).
The median noise spectrum and per-detector, per-scanset noise distribution is shown in Fig.~\ref{fig:netspec}, based on the timestreams for all CMB data so far during the 2016 season. The timestream data has been differenced between polarization pairs, with basic cuts applied so that only good pairs remain (based on both the known hardware yield and preliminary data-quality cuts). Pair-summed spectra are also shown to demonstrate the $1/f$ noise rejection of the pair-difference polarization data. Conversion to CMB temperature is done by calibration of a \bicep3 CMB temperature map to \SI{100}{GHz} Planck data.
The histogram shows science-band average noise after filtering the timestreams with a 3rd-order polynomial during every unidirectional azimuth scanning movement (100 moves per scanset) to mimic final map making.
\bicep3's 2015 season data is also shown for comparison.
An improvement in per-detector sensitivity is clear, summarized by the improved per-detector median band-average NET of \SI{347}{$\mu$K$\sqrt{\textrm{s}}$} for the 2016 season versus \SI{449}{$\mu$K$\sqrt{\textrm{s}}$} from the 2015 season data (right panel of Fig.~\ref{fig:netspec}). However, \textit{Keck}\,\textit{Array}\ \SI{95}{GHz} data shows a median \SI{280}{$\mu$K$\sqrt{\textrm{s}}$} with the same treatment. Reduction of scattering and the resulting detector photon load from warm surfaces, along with the reduction of RF interference, both detailed earlier in this section, can explain \bicep3's improvement in the 2016 season, but a detailed quantification and an understanding of the performance relative to \textit{Keck}\,\textit{Array}\ detectors is ongoing.
Combining the band-average NET values across all detectors for each scanset gives a \bicep3 2016 median array NET of \SI{9.91}{$\mu$K$\sqrt{\textrm{s}}$}. This value accounts for hardware yield (79\%) and for preliminary data-quality cuts (currently a $\sim$45\% loss). \bicep3 2015 data gives a much higher median array NET of \SI{24.6}{$\mu$K$\sqrt{\textrm{s}}$} due to the higher per-detector NET, the only partially populated focal plane, and a higher ($\sim$58\%) loss to data-quality cuts.
\begin{figure} [t]
\begin{center}
\begin{tabular}{c}
\includegraphics[width=.82\textwidth]{img/netspec_p0left}
\end{tabular}
\end{center}
\caption[netspec]
{ \label{fig:netspec}
\textit{Left}: Median per-detector noise spectra for \bicep3 2015 and 2016 season data, from both pair-summed and pair-differenced minimally processed timestreams (showing $1/f$ noise rejection of the differenced polarization measurement).
\textit{Right}: Histogram of the per-detector per-scanset noise, applying 3rd-order polynomial timestream filtering and averaged across the 0.1-\SI{1}{Hz} science band. Median values of \SI{449}{$\mu$K$\sqrt{\textrm{s}}$} and \SI{347}{$\mu$K$\sqrt{\textrm{s}}$} are marked by vertical dashed lines for the 2015 and 2016 data, respectively.
Significantly increased detector population and observing duty cycle explain the much larger 2016 histogram amplitude, which will continue to grow as the season progresses.
}
\end{figure}
All above quoted NET values in this section are based on timestream data. This is an important distinction and the values are not meant to be directly comparable to map-based NET estimates.
Specifically, Ref.~\citenum{wu2015} and Ref.~\citenum{spie:howard} quote map-based \bicep3 per-detector NET estimates of \SI{395}{$\mu$K$\sqrt{\textrm{s}}$} (2015 season) and \SI{333}{$\mu$K$\sqrt{\textrm{s}}$} (2016 season).
The difference arises from details of the two NET estimations.
The map-based value is an estimate based on the standard deviation across all map pixels of a polarization noise map, accounting for the weighted integration time of each pixel, and is therefore insensitive to details of individual detector spectra and assumes a Gaussian noise distribution that will not necessarily reproduce a median of the distributions shown in Fig.~\ref{fig:netspec}.
Additionally, the timestream-based value averages across 0.1--\SI{1}{Hz}, not the full range of scales considered in the map-based method described.
The two methods converge more closely for the total \bicep3 array NET, for which the map-based method gives \SI{9.72}{$\mu$K$\sqrt{\textrm{s}}$}\cite{spie:howard}.
\section{Future Multi-Receiver Array}
\label{sec:array}
Plans have begun to upgrade the \textit{Keck}\,\textit{Array}\ to a 3rd-generation, or ``Stage 3'', experiment with \bicep3-class receivers. The \textit{Keck}\,\textit{Array}, the culmination of the \bicep2/\textit{Keck}\ 2nd-generation, demonstrates the scalability of
\bicep2-class receiver technology and the efficiency of frequency expansion through an array model in which the optics and detectors of each receiver can be tuned to a single frequency band. For the 2014 observing season, two of the five \SI{150}{GHz} \textit{Keck}\,\textit{Array}\ receivers were replaced with \SI{95}{GHz} detectors and matching optics. That additional data allowed a more extensive model of the polarized galactic foreground, and for the first time showed B-mode constraints to be the most powerful on the IGW signal (beyond constraints derived from CMB temperature and other evidence)\cite{BKVI}.
In the following 2015 observing season, two more \SI{150}{GHz} \textit{Keck}\,\textit{Array}\ receivers were refitted, this time with \SI{220}{GHz} detectors and optics, further broadening the spectral coverage of the combined data set by reaching to the higher-frequency polarized dust signal.
Most recently, now that \bicep3 is fully populated at \SI{95}{GHz} for the current 2016 observing season, the two \SI{95}{GHz} \textit{Keck}\,\textit{Array}\ receivers were also refitted to \SI{220}{GHz}, for a total of four at \SI{220}{GHz}.
With its 5$\times$ higher per-receiver detector count, the \bicep3 design acts as a building block for this 3rd-generation multi-frequency CMB polarimeter that we have chosen to name B\textsc{icep}\,Array. In order to continue to further constrain the IGW signal and aim for primordial B-mode detection, B\textsc{icep}\,Array will focus its increased total detector count ($\sim$~12$\times$ greater than \textit{Keck}\,\textit{Array})
on the same sky patch from the South Pole. This shared sky patch with all other B\textsc{icep}/\textit{Keck}\ experiments covers $\sim$1\% of the sky, sufficient area to search for the degree-scale (multipole $l=80$) recombination bump B-mode signal. With the large amount of accumulated observations, it remains the preferred patch
to continue with deeper, multi-frequency observations for targeted degree-scale IGW signal detection and improved galactic foreground separation.
B\textsc{icep}\,Array will be 4 new \bicep3-class receivers with frequency band centers spanning 35 to \SI{270}{GHz}. The design of a \bicep3-class receiver was detailed in Sec.~\ref{sec:inst}, showing how \bicep3 expanded to higher detector count and larger aperture while maintaining the advantages of the compact, on-axis, cold refractor design shared by the B\textsc{icep}/\textit{Keck}\ program. A number of changes are expected to improve on the \bicep3 design. Testing of the metal-mesh IR-reflective filter design has been ongoing since its first development for \bicep3, and potential changes include double-sided aluminization and laser patterning so that fewer filters are required to achieve the same IR rejection. Heatsinking the metal-mesh filter stack to \SI{50}{K}, rather than the room temperature vacuum shell as in \bicep3, could also reduce IR loading from the emissive substrate films.
Some changes are also expected for the detector modules used in B\textsc{icep}\,Array. The four receivers will separately have focal planes centered on 35, 95, and \SI{150}{GHz}, and lastly a 220/\SI{270}{GHz} hybrid. Each focal plane will be made up of \bicep3-style detector module packages but will use 6'' rather than 4'' silicon wafer processes and require many fewer than 20 modules to fill the focal plane, thereby reclaiming focal plane area lost to the module borders.
Table~\ref{tab:numbers} shows the nominal number of detectors that will fit into each focal plane of B\textsc{icep}\,Array.
The 220/\SI{270}{GHz} dust-control receiver will have detectors at the two frequencies in a checkerboard array on the focal plane, based on the current \SI{220}{GHz} \textit{Keck}\,\textit{Array}\ detectors and a lab-tested \SI{270}{GHz} design that will be introduced into \textit{Keck}\,\textit{Array}\ for the 2017 season.
The extremely high 220/\SI{270}{GHz} detector count motivates frequency domain multiplexing readout schemes as a possible alternative to \bicep3's SQUID-based TDM readout (which will continue to be used for the lower frequency receivers).
The \SI{35}{GHz} receiver will help constrain galactic synchrotron emission at the level required for B\textsc{icep}\,Array's sensitivity; prototyping and testing of the \SI{35}{GHz} detectors is currently underway.
Fig.~\ref{fig:bkhistory} shows a comparison of the proposed B\textsc{icep}\,Array design to its predecessors, including the focal plane size and frequency mix.
\begin{table}
\centering
\begin{tabular}{l c c c}
\toprule
Experiment & Frequency Band (GHz) & TES Detector Count & Survey Weight Per Year \\
& & Per Receiver & Per Receiver ($\mu$K$^{-2}$\,yr$^{-1}$) \\
\toprule
\textit{Keck}\,\textit{Array} & \phantom{0}95 & \phantom{0}\textbf{288} & \phantom{0}\textbf{24000} \\
& 150 & \phantom{0}\textbf{512} & \phantom{0}\textbf{30000} \\
& 220 & \phantom{0}\textbf{512} & \phantom{00}\textbf{2000} \\
& 270 & \phantom{0}512 & \phantom{000}800 \\
\midrule
\bicep3 & \phantom{0}95 & \textbf{2560} & \phantom{*}213000\footnotemark\\
\midrule
B\textsc{icep}\,Array & \phantom{0}35 & \phantom{0}384 & \phantom{0}37000 \\
& \phantom{0}95 & 3456 & 287000 \\
& 150 & 7776 & 453000 \\
& 220 & 9408 & \phantom{0}37000 \\
& 270 & 9408 & \phantom{0}15000 \\
\bottomrule
\end{tabular}
\caption{Nominal number of TES detectors and survey weight per year for each of the four B\textsc{icep}\,Array receivers (\SI{220}{GHz} and \SI{270}{GHz} make up one receiver), and a comparison to \bicep3 and \textit{Keck}\,\textit{Array}.
Bolded detector counts are from fielded receivers.
Maximum \textit{Keck}\,\textit{Array}\ (B\textsc{icep}\,Array) population across 5 (4) receivers is 2560 (30432) detectors.
Bolded survey weights are achieved results, indicating performance taking into account all real world observing inefficiencies (see text for details). Other survey weights are scaled from the bolded values.
}
\label{tab:numbers}
\end{table}
B\textsc{icep}\,Array will use a new mount at the South Pole to hold all 4 \bicep3-class receivers, each arranged around a common boresight, as the increased receiver size cannot be accommodated by the existing \textit{Keck}\,\textit{Array}\ mount (originally built for the D\textsc{asi} experiment). The use of high-pressure helium rotary joints, operating successfully with \bicep3, will help reduce the total size of the new mount by avoiding a significant volume of helium hose in the azimuth and theta stages. B\textsc{icep}\,Array will replace the \textit{Keck}\,\textit{Array}\ and observe simultaneously with the ongoing \bicep3 program. The \SI{35}{GHz} receiver is under development and is planned to be the first of the B\textsc{icep}\,Array, beginning as early as the 2018 observing season. Three \textit{Keck}\,\textit{Array}\ receivers outfitted at \SI{270}{GHz} will fill the remaining spots of the new mount. Before the second observing season, the 95 and \SI{150}{GHz} receivers will be installed, and finally the 220/\SI{270}{GHz} receiver will complete the array for the third and subsequent observing seasons.
A useful metric to quantify the CMB polarization sensitivity and the statistical sensitivity to $r$ that will be achieved with B\textsc{icep}\,Array is the survey weight $W=2A/N^2$ (units of $\mu$K$^{-2}$), where $A$ is the map area and $N$ is the polarization map depth (e.g. units of $\mu$K$\cdot$degrees).
Conveniently, survey weight scales linearly with detector count and integration time, and Table~\ref{tab:numbers} shows a comparison of survey weights per year for \textit{Keck}\,\textit{Array}\ and B\textsc{icep}\,Array receivers.
The \textit{Keck}\,\textit{Array}\ \SI{95}{GHz} survey weight is based on the 2014 season published result\cite{BKVI},
while the \SI{150}{GHz} survey weight is based on both \bicep2 performance\cite{BKI} and the latest published 3 receiver-years of \textit{Keck}\,\textit{Array}\cite{BKVI}.
The \textit{Keck}\,\textit{Array}\ \SI{220}{GHz} survey weight is based on preliminary results from the first season (2015) of receivers at that frequency.
These achieved values are then simply scaled by the squared ratio of nominal per-detector NET values (taking into account e.g. detector parameters, loading from optical elements and the atmosphere) to 35 and \SI{270}{GHz}, and the B\textsc{icep}\,Array values are further scaled by the ratio of detector counts.
In this way, all values take into account all real world observing inefficiencies of the \bicep2/\textit{Keck}\ experiments.
From this treatment it is clear that the B\textsc{icep}\,Array represents an order of magnitude upgrade in total raw experimental sensitivity versus \textit{Keck}\,\textit{Array}.
\footnotetext{
The quoted \bicep3 survey weight is scaled from the \textit{Keck}\,\textit{Array}\ \SI{95}{GHz} value, reflecting the eventual expectation for \bicep3. Adjusting by the \bicep3 and \textit{Keck}\,\textit{Array}\ median per-detector band average NET values quoted in Sec.~\ref{sec:perf} gives an estimated \bicep3 survey weight of \SI{139000}{$\mu$K$^{-2}$} for 2016.
}
\begin{figure} [t]
\begin{center}
\begin{tabular}{c}
\includegraphics[width=1\textwidth]{img/bkhistory}
\end{tabular}
\end{center}
\caption[bkhistory]
{ \label{fig:bkhistory}
Summary of the 2nd and 3rd-generations of the B\textsc{icep}/\textit{Keck}\ program. The bottom row shows the beam patterns on the sky with a common scale.
}
\end{figure}
\section{Conclusion}
These proceedings have summarized the instrument design and observation strategy of the \bicep3 polarimeter, and reported on changes and current performance for its ongoing second season.
Important changes included full detector population of the focal plane, cryogenic improvements that led to increased observation duty cycle, replacement of under-performing IR filters, and a significant reduction in RF interference.
Based on the first $\sim1224$~hours of new CMB data this season (2016), \bicep3 has a timestream-based median per-detector NET of \SI{347}{$\mu$K$\sqrt{\textrm{s}}$} and an array NET of \SI{9.91}{$\mu$K$\sqrt{\textrm{s}}$} for the whole instrument, an improvement over the 2015 per-detector median of \SI{449}{$\mu$K$\sqrt{\textrm{s}}$}.
Plans were also outlined for an upgrade to the multi-frequency \textit{Keck}\,\textit{Array}, named B\textsc{icep}\,Array.
B\textsc{icep}\,Array will be four \bicep3-class receivers spanning 35 to \SI{270}{GHz} and begin a staggered deployment as soon as 2018.
It will use a new mount at the existing South Pole \textit{Keck}\,\textit{Array}\ site, and have more than an order of magnitude greater survey weight per receiver versus \textit{Keck}\,\textit{Array}.
\section{MISCELLANEOUS FORMATTING DETAILS}
\acknowledgments
This work is supported by the National Science Foundation (grant nos. 1313158, 1313010, 1313062, 1313287, 1056465, 0960243), the SLAC Laboratory Directed Research and Development Fund, the Canada Foundation for Innovation, Science and Technology Facilities Council Consolidated Grant (ST/K000926/1), and the British Columbia Development Fund. The development of detector technology was supported by the JPL Research and Technology Development Fund and grants 06-ARPA206-0040, 10-SAT10-0017, and 12-SAT12-0031 from the NASA APRA and SAT programs. The development and testing of detector modules was supported by the Gordon and Betty Moore Foundation.
| 7f2ff211bca5f88957ff4cbb4ef5c5b18b408707 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Radio galaxies (RGs) are radio loud active galaxy nuclei (AGN) exhibiting clear structure of a compact central source, large-scale jets and lobes. These sources are also of interest due to the close proximity to the earth affording us an excellent opportunity for studying the physics of relativistic outflows. They have been widely studied from radio wavelengths to MeV-GeV $\gamma$-rays and recently at very-high-energy (VHE) by Imagine Atmospheric Cherenkov Telescope (IACT). RGs are generally described by the standard non-thermal synchrotron self-Compton (SSC) model \citep{2010ApJ...719.1433A,2012ApJ...753...40F,1998ApJ...509..608T}. In SSC framework, low-energy emission; radio through optical, originates from synchrotron radiation while HE photons; X-rays through $\gamma$-rays, come from inverse Compton scattering emission. However, this model with only one population of electrons, predicts a spectral energy distribution (SED) that can be hardly extended up to higher energies than a few GeVs \citep{2005ApJ...634L..33G,2008A&A...478..111L}. In addition, some authors have suggested that the emission in the GeV - TeV energy range may have origins in different physical processes \citep{2011MNRAS.413.2785B,2005ApJ...634L..33G}.\\
The Telescope Array (TA) experiment, located in Millard Country (Utah), was designed to study ultra-high-energy cosmic rays (UHECRs) with energies above 57 EeV \citep{2012NIMPA.689...87A}. TA observatory, with a field of view covering the sky region above -10$^\circ$ of declination, detected a cluster of 72 UHECRs events centered at R. A.=146$^\circ$.7, Dec.=43$^\circ.2$. It had a Li-Ma statistical significance of 5.1$\sigma$ within 5 years of data taking \citep{2014arXiv1404.5890T}. The Pierre Auger observatory (PAO), located in the Mendoza Province of Argentina, was designed to determine the arrival directions and energies of UHECRs using four fluorescence telescope arrays and the High Resolution Fly's eye (HiRes) experiment, located in the west-central Utah desert, consisted of two detectors that observed cosmic ray showers via the fluorescence light. PAO studying the composition of the high-energy showers found that the distribution of their properties was located in somewhere between pure proton (p) and pure iron (Fe) at 57 EeV\citep{2008ICRC....4..335Y, 2008APh....29..188P, 2007AN....328..614U}, although the latest results favored a heavy nuclear composition \citep{2010PhRvL.104i1101A}. By contrast, HiRes data were consistent with a dominant proton composition at this energy range \citep{2007AN....328..614U, 2008ICRC....4..385E}. PAO detected 27 UHECRs, with two of them associated with Centaurus A (Cen A), being reconstructed inside a circle centered at the position of this FR I with a aperture of 3.1$^{\circ}$.\\
The IceCube detector located at the South Pole was designed to record the interactions of neutrinos. Encompassing a cubic kilometer of ice and almost four years of data taking (from 2010 to 2014), the IceCube telescope reported with the High-Energy Starting Events (HESE)\footnote{http://icecube.wisc.edu/science/data/HE-nu-2010-2014} catalog a sample of 54 extraterrestrial neutrino events in the TeV - PeV energy range. Arrival directions of these events are compatible with an isotropic distribution. The neutrino flux is compatible with a high component due to extragalactic origin \citep{2014PhRvD..90b3010A, 2015ApJ...815L..25G}. For instance, gamma-ray bursts \citep{2013PhRvL.111l1102M, 2013PhRvD..88j3003R, 2014MNRAS.437.2187F, 2013ApJ...766...73L, 2015JCAP...09..036T, 2014MNRAS.445..570P, 2016JHEAp...9...25F} and AGN (\cite{2014JHEAp...3...29D,2013PhRvD..88d7301S, 2015MNRAS.448.2412P, 2015MNRAS.452.1877P, 2014arXiv1410.8549M, 2015APh....71....1F, 2015APh....70...54F}), etc.\\
Hadronic processes producing VHE neutrinos and photons through the acceleration of cosmic rays in AGNs have been explored by many authors \citep{2001PhRvL..87v1102A, 2008PhR...458..173B, 2014arXiv1410.8124K, 2001APh....15..121M, 2008PhRvD..78b3007C, 2009NJPh...11f5016D, 2015MNRAS.451.1502T, 2015ApJ...806..159K, 2007ApJ...670L..81H, 2003ApJ...593..169H}. In particular, GeV-TeV $\gamma$-ray fluxes interpreted as pion decay products from photo-hadronic interactions occurring close to the core of RGs have been also discussed for different sources \citep{2012PhRvD..85d3012S, 2012ApJ...753...40F, 2015arXiv150407592K, 2016MNRAS.455..838K, 2014MNRAS.441.1209F, 2014A&A...562A..12P,2015IAUS..313..175F}.\\
In this work we introduce a leptonic and hadronic model to describe the broadband SED of the closest RGs. For the leptonic model, we present the SSC model to explain the SED up to dozens of GeV and for the hadronic model, we propose that the proton-photon (p$\gamma$) interactions occurring close to the core could describe the $\gamma$-ray spectrum at the GeV - TeV energy range. Correlating the TeV $\gamma$-ray, UHECR and neutrino spectra through p$\gamma$ interactions, we estimate their fluxes and number of events expected in PAO, TA and HiRes experiments and IceCube telescope, respectively. For this correlation, we have assumed that the proton spectrum is extended through a simple power law up to UHEs.
\section{The closest RGs: Cen A, M87 and NGC1275}
In this section, we are going to present a brief description of the three RGs studied in this work and the set of data used in our analysis.
\subsection{Cen A}
Cen A, at a distance of d$_z\simeq3.8$ Mpc (z=0.00183), has been one of the best studied radio galaxies. It is characterized by having an off-axis jet estimated in $\sim 15^\circ - 80^\circ$ \citep[see, e.g.] [and reference therein]{2006PASJ...58..211H} and two giant radio lobes. Cen A has been imaged in radio wavelengths, optical bands \citep{1975ApJ...199L.139W,1976ApJ...206L..45M,1970ApJ...161L...1B,1981ApJ...244..429B}, X-rays, sub-GeV$\gamma$-rays and VHE $\gamma$-rays \citep{2003ApJ...593..169H,1999APh....11..221S,2009ApJ...695L..40A, 2010ApJ...719.1433A, 1998A&A...330...97S, 2010ApJ...719.1433A,2013ApJ...770L...6S}. This radio galaxy was observed during a period of 10 months by the Large Area Telescope (LAT) on board the Fermi. The sub-GeV $\gamma$-ray flux collected was described with a power law (F$\propto E^{-\alpha}$ with $\alpha= 2.67\pm 0.10_{\rm stat} \pm 0.08_{\rm sys}$; \cite{2010ApJ...719.1433A}). In addition, Cen A was also detected for more than 120 hr by the High Energy Stereoscopic System (HESS) \citep{2005A&A...441..465A, 2009ApJ...695L..40A}. The observed spectrum ($>$ 250 GeV) was fitted using a simple power law with a spectral index of 2.7$\pm\, 0.5_{\rm stat} \pm 0.2_{\rm sys}$ and an apparent luminosity of $\simeq 2.6 \times 10^{39}$ erg s$^{-1}$. No significant flux variability was detected.
\subsection{M87}
M87, at a distance of d$_z\simeq16$ Mpc (z=0.0043), is located in the Virgo cluster of galaxies and hosts a central black hole (BH) of $(3.2\pm 0.9)\times 10^9$ solar masses \citep{1996ApJ...467..597B}. The jet, inclined at an angle of $\sim$ 15$^\circ$ - 20$^\circ$ relative to the observe's line of sight, has been imaged in its base down to $\sim$0.01 pc resolution \citep{1999Natur.401..891J, 2007ApJ...660..200L}. As one of the nearest radio galaxies to us, M87 is amongst the best-studied of its source class. It has been detected from radio wavelengths to VHE $\gamma$ rays. Above 100 MeV, M87 has been detected by LAT-Fermi \citep{2009ApJ...707...55A}, HESS, MAGIC and the Very Energetic Radiation Imaging Telescope Array System (VERITAS) telescopes \citep{2012ApJ...746..151A, 2009ApJ...706L.275A, 2012ApJ...746..141A}. In 2004, the HESS telescope detected the active galaxy M87 in its historical minimum. The TeV flux measurement was well fitted with a simple power law. The differential energy spectrum was $(2.43\pm0.75)\times10^{-13} \,\, {\rm TeV^{-1}\,cm^{-2}\,s^{-1}}$ with a photon index of $2.62\pm0.35$ \citep{2006Sci...314.1424A}. No indications for short-term variability were found.
\subsection{NGC1275}
NGC 1275, also known as Perseus A and 3C 84, is an elliptical/radio galaxy located at the center of the Perseus cluster at d$_z\simeq76$ Mpc ($z=0.0179$). This source has a strong, compact nucleus which has been studied in detail with Very Long Baseline Interferometry (VLBI)\citep{1994ApJ...430L..41V, 2006MNRAS.368.1500T, 2000ApJ...530..233W, 2006PASJ...58..261A} and Array (VLBA) \citep{2014ApJ...785...53N}. These observations reveal a compact core and a bowshock-like souther jet component moving steadily outwards at 0.3 mas/year \citep{2004ApJ...609..539K,2009AJ....137.3718L}. The norther counter jet is also detected, though it is much less prominent due to Doppler dimming, as well as to free-free absorption due an intervening disk. Walker, Romney and Berson (1994) derive from these observations that the jet has an intrinsic velocity of $0.3c - 0.5c$ oriented at an angle $\approx$ 10$^\circ$ - 35$^\circ$ to the line of sight. Polarization has recently been detected in the southern jet \citep{2006MNRAS.368.1500T}, suggesting increasingly strong interactions of the jet with the surrounding environment.\\
Due to its brightness and proximity, this source has been detected from radio to TeV $\gamma$-ray bands \citep{2009ApJ...699...31A}. This radio galaxy was observed at energies above 100 MeV by Fermi-LAT from 2008 August 4 to 2008 December 5. The average flux and photon index measured, which remained relatively constant during the observing period were $(2.10\pm0.23)\times 10^{-7}\, {\rm ph\, cm^{-2}\, s^{-1}}$ and $2.17\pm 0.05$, respectively. This source has been detected by MAGIC telescope with a statistical significance of $6.6\,\sigma$ above 100 GeV in 46 hr of stereo observations carried out between August 2010 and February 2011. The measured differential energy spectrum between 70 GeV and 500 GeV was described by a power law with a steep spectral index of $-4.1\pm0.7_{stat}\pm0.3_{\rm sys}$, and an average flux of $(1.3\pm 0.2_{stat}\pm0.3_{\rm sys})\times 10^{-11}\,{\rm cm^{-2}\,s^{-1}}$ \citep{2012A&A...539L...2A}. The light curve ($>$ 100 GeV) did not show hints of variability on a month time scale.
\section{Theoretical Model}
We propose that the broadband SED of RGs can be described as the superposition of SSC emission and $\pi^0$ decay product resulting from p$\gamma$ interactions.
\subsection{Leptonic Model}
Fermi-accelerated electrons are injected in an emitting region with radius $r_d$ which moves at ultra-relativistic velocities with Doppler factor $\delta_D$. Relativistic electrons are confined in the emitting region by a magnetic field, therefore it is expected non-thermal photons by synchrotron and Compton scattering radiation.
\subsubsection{Synchrotron radiation}
The electron population is described by a broken power-law given by \citep{1994hea2.book.....L}
\begin{equation}
\label{espele}
N_e(\gamma_e) = N_{0,e}
\cases {
\gamma_e^{-\alpha_e} & $\gamma_{\rm e,min}<\gamma_e < \gamma_{\rm e,c}$,\cr
\gamma_{e,c} \gamma_e^{-(\alpha_e+1)} & $\gamma_{\rm e,c} \leq \gamma_e<\gamma_{\rm e,max}$,\cr
}
\end{equation}
\noindent where $N_{0,e}$ is the proportionality electron constant, $\alpha_e$ is the spectral power index of the electron population and $\gamma_{e,i}$ are the electron Lorentz factors. The index $i$ is min, c or max for minimum, cooling and maximum, respectively. By considering that a fraction of total energy density is given to accelerate electrons $U_e=m_e \int\gamma_e N_e(\gamma_e)d\gamma_e$, then the minimum electron Lorentz factor and the electron luminosity can be written as
\begin{eqnarray}\label{gamma_m}
\gamma_{\rm e,min}&=&\frac{(\alpha_e-2)}{m_e(\alpha_e-1)}\,\frac{U_e}{N_e}\,,
\end{eqnarray}
and
\begin{equation}\label{Le}
L_e=4\pi\, \delta^2_D\,r^2_d\,U_e\,,
\end{equation}
respectively, where $m_e$ is the electron mass. The electron distribution in the emitting region permeated by a magnetic field $B=\sqrt{8\pi\,U_B}$ cools down following the cooling synchrotron time scale $t'_c=\frac{3m_e}{4\sigma_T}\,\,U^{-1}_B\,\gamma^{-1}_e$, with $\sigma_T=6.65 \times 10^{-25}\, {\rm cm^2}$ the Compton cross section. Comparing the synchrotron time scale with the dynamic scale $t'_d\simeq r_d/\delta_D$, we get that the cooling Lorentz factor is
\begin{equation}\label{gamma_b}
\gamma_{\rm e,c}= \frac{3\,m_e}{4\,\sigma_T}\,(1+Y)^{-1}\, \delta_D\, U_B^{-1}\,r_d^{-1}\,,
\end{equation}
where the Compton parameter is
{\small
\begin{equation}\label{Y1}
Y\equiv\frac{L_{\rm \gamma, IC}}{L_{\rm syn}}=\frac{U_{\rm rad}}{U_B}=\frac{U_{\rm syn}}{U_B}=\frac{\eta\, U_e}{(1+Y)\,U_B}\,.
\end{equation}
}
Here, $L_{\rm \gamma, IC}$ is the observed luminosity around the second SSC peak, $U_{\rm syn}$ and $U_B$ are the energy density of synchrotron radiation and magnetic field, respectively. Solving equation (\ref{Y1}), the two interesting limits are
{\small
\begin{eqnarray}
\label{Y2}
Y=\cases{
\frac{\eta\,U_e}{U_B}\,, & if $\frac{\eta\,U_e}{U_B}\ll 1$ ,\cr
\left(\frac{\eta\,U_e}{U_B}\right)^{1/2}\,, & if $\frac{\eta\,U_e}{U_B}\gg 1$,\cr
}
\end{eqnarray}
}
with $\eta=(\gamma_{\rm e,c}/\gamma_{\rm e,min})^{2-\alpha_e}$ given for slow cooling and $\eta=1$ for fast cooling \citep{2001ApJ...548..787S}. By considering that acceleration time scale $t'_{acc}\simeq\sqrt\frac\pi2m_e/q_e\, U^{-1/2}_B\,\gamma_e$ and cooling time scale are similar, it is possible to write the maximum electron Lorentz factor as
\begin{eqnarray}\label{gamma_max}
\gamma_{\rm e,max}&=&\biggl(\frac{9\, q_e^2}{8\pi\, \sigma_T^2}\biggr)^{1/4}\,U_B^{-1/4}\,,
\end{eqnarray}
\noindent where $q_e$ is the electric charge. Taking into account the synchrotron emission ${\small \epsilon_\gamma(\gamma_{e,i})=\sqrt{\frac{8\pi q_e^2}{m_e^2}}\,\delta_D\, U_B^{1/2}\, \gamma^2_{e,i}}$ and eqs. (\ref{gamma_m}), (\ref{gamma_b}) and (\ref{gamma_max}), the synchrotron break energies are
\begin{eqnarray}\label{synrad}
\epsilon^{\rm syn}_{\rm \gamma,m} &=& \frac{\sqrt{8\pi}\,q_e}{m_e}\,\delta_D\,U_B^{1/2}\,\gamma^2_{\rm e,min}\cr
\epsilon^{\rm syn}_{\rm \gamma,c} &=&\frac{ 9\sqrt{2\pi}\,q_e\,m_e}{8\,\sigma_T^2}\, (1+Y)^{-2}\,\delta_D^3\, U_B^{-3/2}\, r_d^{-2}\cr
\epsilon^{\rm syn}_{\rm \gamma, max} &=&\frac{3\,q_e^2}{m_e\,\sigma_T}\, \delta_D\,.
\end{eqnarray}
The synchrotron spectrum is computed through the electron distribution (eq. \ref{espele}) rather than synchrotron radiation of a single electron. Therefore, the photon energy radiated in the range $\epsilon_\gamma$ to $\epsilon_\gamma + d\epsilon_\gamma$ is given by electrons between $E_e$ and $E_e + dE_e$; then we can estimate the photon spectrum through emissivity $\epsilon_\gamma N_\gamma(\epsilon_\gamma) d\epsilon_\gamma=(- dE_e/dt)\,N_e(E_e)dE_e$. Following \cite{1994hea2.book.....L} and \cite{1986rpa..book.....R}, one can show that if electron distribution has spectral indexes $\alpha_e$ and $(\alpha_e-1)$, then the photon distribution has spectral indexes $(\alpha_e-1)/2$ and $\alpha_e/2$, respectively. The observed synchrotron spectrum can be written as
{\small
\begin{eqnarray}
\label{espsyn}
\left[\epsilon^2_\gamma N(\epsilon_\gamma)\right]_{\rm \gamma, syn}= A_{\rm \gamma,syn}\cases{
(\frac{\epsilon_\gamma}{\epsilon^{\rm syn}_{\rm \gamma,m}})^\frac43 & $\epsilon_\gamma < \epsilon^{\rm syn}_{\rm \gamma,m}$,\cr
(\frac{\epsilon_\gamma}{\epsilon^{\rm syn}_{\rm \gamma,m}})^{-\frac{\alpha_e-3}{2}} & $\epsilon^{\rm syn}_{\rm \gamma,m} < \epsilon_\gamma < \epsilon^{\rm syn}_{\rm \gamma,c}$,\cr
(\frac{\epsilon^{\rm syn}_{\rm \gamma,c}}{\epsilon^{\rm syn}_{\rm \gamma,m}})^{-\frac{\alpha_e-3}{2}} (\frac{\epsilon_\gamma}{\epsilon_{\rm \gamma,c}})^{-\frac{\alpha_e-2}{2}}, & $\epsilon^{\rm syn}_{\rm \gamma,c} < \epsilon_\gamma < \epsilon^{\rm syn}_{\rm \gamma,max}$\,,\cr
}
\end{eqnarray}
}
\noindent where $A_{\rm \gamma,syn}$ is the proportionality constant of synchrotron spectrum. This constant can be estimated through the total number of radiating electrons in the volume of emitting region, ${\small n_e=N_e/V=4\pi N_e\,r_d^3/3}$, the maximum radiation power ${\small P_{\rm \nu,max}\simeq \frac{dE_e/dt}{\epsilon_\gamma(\gamma_e)}}$ and the distance d$_z$ from the source. Therefore, the proportionality constant can be written as
\begin{eqnarray}
\label{Asyn}
A_{\rm \gamma,syn}&=& \frac{P_{\rm \nu,max} n_e}{4\pi d_z^2}\,\epsilon^{\rm syn}_{\rm \gamma,m} \cr
&=&\frac{4\,\sigma_T}{9}\, d^2_z\,\delta^3_D\,U_B\,r_d^3\,N_e\,\gamma^2_{\rm e,min}\,.
\end{eqnarray}
\subsubsection{Compton scattering emission}
Fermi-accelerated electrons in the emitting region can upscatter synchrotron photons up to higher energies as
\begin{equation}\label{ic}
\epsilon^{\rm ssc}_{\gamma,{\rm (m,c,max)}}\simeq \gamma^2_{e,{\rm (min,c,max)}} \epsilon^{\rm syn}_{\gamma,{\rm (m,c, max)}}\,.
\label{ic}
\end{equation}
From the electron Lorentz factors (eqs. \ref{gamma_m}, \ref{gamma_b}, \ref{gamma_max}) and the synchrotron break energies (eq. \ref{synrad}), we get that the Compton scattering break energies are given in the form
\begin{eqnarray}\label{icrad}
\epsilon^{\rm ssc}_{\rm \gamma,m} &=& \frac{\sqrt{8\pi}\,q_e}{m_e}\,\delta_D\,U_B^{1/2}\,\gamma^4_{\rm e,min},\cr
\epsilon^{\rm ssc}_{\rm \gamma,c} &=&\frac{ 81\sqrt{2\pi}\,q_e\,m_e^3}{128\,\sigma_T^4}\,(1+Y)^{-4}\,\delta_D^5\, U_B^{-7/2}\, r_d^{-4},\cr
\epsilon^{\rm ssc}_{\rm \gamma, max} &=&\frac{9\,q_e^3}{2\sqrt{2\pi}\,m_e\,\sigma^2_T}\, \delta_D\,U_B^{-1/2}\,.
\end{eqnarray}
The Compton scattering spectrum obtained as a function of the synchrotron spectrum (eq. \ref{espsyn}) is
{\small
\begin{eqnarray} \label{espic}
\left[\epsilon^2_\gamma N(\epsilon_\gamma)\right]_{\rm \gamma, ssc}=A_{\rm \gamma,ssc} \cases{
(\frac{\epsilon_\gamma}{\epsilon^{\rm ssc}_{\rm \gamma,m}})^\frac43 & $\epsilon_\gamma < \epsilon^{ssc}_{\gamma,m}$,\cr
(\frac{\epsilon_\gamma}{\epsilon^{\rm ssc}_{\rm \gamma,m}})^{-\frac{\alpha_e-3}{2}} & $\epsilon^{\rm ssc}_{\rm \gamma,m} < \epsilon_\gamma < \epsilon^{\rm ssc}_{\rm \gamma,c}$,\cr
(\frac{\epsilon^{\rm ssc}_{\rm \gamma,c}}{\epsilon^{ssc}_{\gamma,m}})^{-\frac{\alpha_e-3}{2}} (\frac{\epsilon_\gamma}{\epsilon^{ssc}_{\gamma,c}})^{-\frac{\alpha_e-2}{2}}, & $\epsilon^{\rm ssc}_{\rm \gamma,c} < \epsilon_\gamma < \epsilon^{\rm ssc}_{\rm \gamma,max} $
}\cr
\end{eqnarray}
}
where $A_{\rm \gamma,ssc}=Y\,[\epsilon^2_\gamma N_\gamma(\epsilon_\gamma)]_{\rm max}^{\rm syn}$ is the proportionality constant of Compton scattering spectrum.
\subsection{Hadronic Model}
RGs have been proposed as a powerful accelerator of charged particles through the Fermi acceleration mechanism or/and magnetic reconnection \citep{2015arXiv150407592K}. We consider a proton population described as a simple power law given by
\begin{equation}\label{prot_esp}
\left(\frac{dN}{dE}\right)_p=A_p\,E_p^{-\alpha_p}\,,
\end{equation}
with $A_p$ the proportionality constant and $\alpha_p$ the spectral power index of the proton population. From eqs. (\ref{prot_esp}) we can compute that the proton density can be written as
\begin{equation}
U_p=\frac{L_p}{4\pi\,\delta^2_D\,r^2_d}\,,
\end{equation}
with the proton luminosity given by
\begin{equation}\label{lum}
L_p= 4\,\pi\,d^2_z A_p\,\int \,E_p\,E_p^{-\alpha_p} dE_p\,.
\end{equation}
Fermi-accelerated protons lose their energies by electromagnetic channels and hadronic interactions. Electromagnetic channels such as proton synchrotron radiation and inverse Compton will not be considered here, we will only assume that protons will be cooled down by p$\gamma$ interactions at the emitting region of the inner jet. Charged ($\pi^+$) and neutral ($\pi^0$) pions are obtained from p$\gamma$ interaction through the following channels
\begin{eqnarray}
p\, \gamma &\longrightarrow&
\Delta^{+}\longrightarrow
\left\{
\begin{array}{lll}
p\,\pi^{0}\ && \mbox{fraction }2/3, \\
n\, \pi^{+} && \mbox{fraction }1/3\,.\nonumber
\end{array}\right. \\
\end{eqnarray}
After that neutral pion decays into photons, $\pi^0\rightarrow \gamma\gamma$, carrying $20\% (\xi_{\pi^0}=0.2)$ of the proton's energy $E_p$. The efficiency of the photo-pion production is \citep{1968PhRvL..21.1016S, PhysRevLett.78.2292}
{\small
\begin{equation}\label{eficiency}
f_{\pi^0} \simeq \frac {t_{\rm dyn}} {t^{-1}_{\pi^0}} =\frac{r_d}{2\gamma^2_p}\int\,d\epsilon\,\sigma_\pi(\epsilon)\,\xi_{\pi^0}\,\epsilon\int dx\, x^{-2}\, \frac{dn_\gamma}{d\epsilon_\gamma} (\epsilon_\gamma=x)\,,
\end{equation}
}
where $dn_\gamma/d\epsilon_\gamma$ is the spectrum of seed photons, $\sigma_\pi(\epsilon_\gamma)$ is the cross section of pion production and $\gamma_p$ is the proton Lorentz factor. Solving the integrals we obtain
{\small
\begin{eqnarray}
f_{\pi^0} \simeq \frac{\sigma_{\rm p\gamma}\,\Delta\epsilon_{\rm res}\,\xi_{\pi^0}\, L_{\rm \gamma,IC}}{4\pi\,\delta_D^2\,r_d\,\epsilon_{\rm pk,ic}\,\epsilon_{\rm res}}
\cases{
\left(\frac{\epsilon^{\pi^0}_{\gamma}}{\epsilon^{\pi^0}_{\gamma,c}}\right)^{\beta_h-1} & $\epsilon_{\gamma} < \epsilon^{\pi^0}_{\gamma,c}$\cr
\left(\frac{\epsilon^{\pi^0}_{\gamma}}{\epsilon^{\pi^0}_{\gamma,c}}\right)^{\beta_l-1} & $\epsilon^{\pi^0}_{\gamma,c} < \epsilon_{\gamma}$\,,\cr
}
\end{eqnarray}
}
where $\beta_h$ and $\beta_l$ are the high-energy and low-energy photon index, respectively, ($L_{\rm \gamma, IC}$) is the observed luminosity around the second SSC peak, $\Delta\epsilon_{\rm res}$=0.2 GeV, $\epsilon_{\rm res}\simeq$ 0.3 GeV, $\epsilon_{\rm pk,ic}$ is the energy of the second SSC peak and $\epsilon^{\pi^0}_{\gamma,c}$ is the break photon-pion energy given by
\begin{equation}
\epsilon^{\pi^0}_{\gamma,c}\simeq 31.87\,{\rm GeV}\, \delta_D^2\, \left(\frac{\epsilon_{\rm pk,ic}}{ {\rm MeV}}\right)^{-1}\,.
\label{pgamma}
\end{equation}
It is worth noting that the target photon density and the optical depth can be obtained through the equations
\begin{equation}
n_{\gamma}\simeq\frac{L_{\rm \gamma,IC}}{4\pi r^2_d\,\epsilon_{\rm pk,ic}}\,,
\label{den}
\end{equation}
and
\begin{equation}
\tau_{\gamma}\simeq\frac{L_{\rm \gamma,IC}\,\sigma_T}{4\pi r_d\,\delta_D\epsilon_{\rm pk,ic}}\,.
\end{equation}
respectively. Taking into account that photons released in the range $\epsilon_\gamma$ to $\epsilon_\gamma + d\epsilon_\gamma$ by protons in the range $E_p$ and $E_p + dE_p$ are $f_{\pi^0}\,E_p\,(dN/dE)_p\,dE_p=\epsilon_{\pi^0,\gamma}\,(dN/d\epsilon)_{\pi^0,\gamma}\,d\epsilon_{\pi^0,\gamma}$, then photo-pion spectrum is given by
{\small
\begin{eqnarray}
\label{pgammam}
\left[\epsilon^2_\gamma N(\epsilon_\gamma)\right]_{\rm \gamma, \pi^0}= A_{\rm p\gamma} \left(\frac{\epsilon^{\pi^0}_{\gamma,c}}{\epsilon_{0}}\right)^{-\alpha_p+2} \cases{
\left(\frac{\epsilon^{\pi^0}_{\gamma}}{\epsilon^{\pi^0}_{\gamma,c}}\right)^{\beta_h+1-\alpha_p} & $ \epsilon_{\gamma} < \epsilon^{\pi^0}_{\gamma,c}$\cr
\left(\frac{\epsilon^{\pi^0}_{\gamma}}{\epsilon^{\pi^0}_{\gamma,c}}\right)^{\beta_l+1-\alpha_p} & $\epsilon^{\pi^0}_{\gamma,c} < \epsilon_{\gamma}$\,,\cr
}
\end{eqnarray}
}
\noindent where the proportionality constant $A_{\rm p\gamma}$ is in the form
\begin{equation}\label{Apg}
A_{\rm p\gamma}= \frac{L_{\rm \gamma,IC}\,\sigma_{\rm p\gamma}\,\Delta\epsilon_{\rm res}\,\epsilon^2_0\,\left(\frac{2}{\xi_{\pi^0}}\right)^{1-\alpha_p}}{4\pi\,\delta_D^2\,r_d\,\epsilon_{\rm pk,ic}\,\epsilon_{\rm res}}\,A_p\,.
\end{equation}
The value of $A_p$ is determined through the TeV $\gamma$-ray flux (eq. \ref{Apg}).
\section{High energy neutrino expectation}
Photo hadronic interactions in the emitting region (see subsection 3.2) also generate neutrinos through the charged pion decay products ($\pi^{\pm}\rightarrow \mu^\pm+ \nu_{\mu}/\bar{\nu}_{\mu} \rightarrow e^{\pm}+\nu_{\mu}/\bar{\nu}_{\mu}+\bar{\nu}_{\mu}/\nu_{\mu}+\nu_{e}/\bar{\nu}_{e}$). Taking into account the distances of RGs, the neutrino flux ratio (1 : 2 : 0 ) created on the source will arrive on the standard ratio (1 : 1 : 1 ) \citep{2008PhR...458..173B}. The neutrino spectrum produced by the photo hadronic interactions is
{\small
\begin{eqnarray}
\label{pgammam}\label{espneu1}
\left[E_\nu^2 N(\epsilon_\nu)\right]_\nu= A_\nu \epsilon^2_0 \cases{
\left(\frac{E_\nu}{\epsilon_0 }\right)^{\beta_h} & $ E_\nu < E_{\nu,c}$\cr
\left(\frac{E_\nu}{\epsilon_0 }\right)^{\beta_l+1-\alpha_{\nu}} & $E_{\nu,c} < E_\nu$\,,\cr
}
\end{eqnarray}
}
where the factor, A$_{\nu}$, normalized through the TeV $\gamma$-ray flux is \citep[see, Julia Becker] [and reference therein]{2007Ap&SS.309..407H}
\begin{equation}
A_\nu=A_{\rm p\gamma}\,\epsilon_0^{-2}\, 2^{-\alpha_p}\,.
\end{equation}
The previous equation was obtained from solving the integral terms {\small $\int \frac{dN_{\nu}}{dE_\nu}\,E_\nu\,dE_\nu=\frac14\int \frac{dN_\gamma}{dE_\gamma}\,E_\gamma\,dE_\gamma$}, considering that the spectral indices for neutrino and that $\gamma$-ray spectra are similar $\alpha_p\simeq \alpha_\nu$ \citep{2008PhR...458..173B} and each neutrino (photon) bring 5\% (10\%) of the initial proton energy.
The neutrino flux is detected when it interacts inside the instrumented volume. Considering the probability of interaction for a neutrino with energy $E_\nu$ in an effective volume ($V_{eff}$) with density ($\rho_{ice}$), the number of expected neutrino events after a period of time $T$ is
{\small
\begin{equation}
N_{\rm ev} \approx\,T \rho_{\rm ice}\,N_A\,\epsilon_0\int_{E_{\rm \nu,th}} V_{\rm eff}(E_\nu) \sigma_{\rm \nu N}(E_\nu) \,A_\nu\left(\frac{E_{\nu}}{\epsilon_0}\right)^{-\alpha_p}\, dE_\nu\,,
\label{numneu1}
\end{equation}
}
where $N_A$ is the Avogadro number, $\sigma_{\nu N}(E_\nu)$ is the the charged current cross section and $E_{\nu,th}$ is the energy threshold. It is worth noting that the effective volume $V_{eff}$ is obtained for a hypothetical Km$^{3}$ neutrino telescope through the Monte Carlo simulation.
\section{Ultra-high-energy cosmic rays}
TeV $\gamma$-ray observations from low-redshift AGN have been proposed as good candidates for studying UHECRs \citep{2012ApJ...749...63M, 2009NJPh...11f5016D, 2012ApJ...745..196R}. We consider that the proton spectrum is extended up to UHEs and then calculate the number of events expected in PAO, TA and HiRes experiments.
\subsection{Hillas Condition}
By considering that super massive BHs have the power to accelerate particles up to UHEs through Fermi processes, protons accelerated in the emitting region are confined by the Hillas condition \citep{1984ARA&A..22..425H}. Although this requirement is a necessary condition and acceleration of UHECRs in AGN jets \citep{2012ApJ...749...63M,2012ApJ...745..196R,2010ApJ...719..459J}, it is far from trivial (see e.g., \cite{2009JCAP...11..009L} for a more detailed energetic limits). The Hillas criterion is defined as the maximum proton energy achieved in a region with radius $r_d$ and magnetic field $B$. It can be written as
\begin{equation}\label{Emax}
E_{\rm p,max}\simeq \frac{Zq_e}{\phi}\,B\,r_d\,\Gamma\,.
\end{equation}
Here $Z$ is the atomic number, $\phi\simeq$ 1 is the acceleration efficiency factor and $\Gamma$ is the bulk Lorentz factor given by
\begin{equation}
\Gamma=\frac{1\pm\sqrt{1-(1-\cos^2\theta)(1+\delta^2_D\cos^2\theta)}}{\delta_D(1-\cos^2\theta)}\,,
\end{equation}
where $\theta$ is the viewing angle.
\subsection{Deflections}
The magnetic fields play important roles on cosmic rays. UHECRs traveling from source to Earth are randomly deviated by galactic (B$_G$) and extragalactic (B$_{EG}$) magnetic fields. By considering that magnetic fields are quasi-constant and homogeneous, the deflection angle due to the B$_G$ and B$_{EG}$ \citep{1997ApJ...479..290S} are
\begin{equation}\label{thet_G}
\psi_{\rm G}\simeq 3.8^{\circ}\left(\frac{E_{p,th}}{57 EeV}\right)^{-1} \int^{L_G}_0 | \frac{dl}{{\rm kpc}}\times \frac{B_G}{4\,{\rm \mu G}} |\,,
\end{equation}
and
\begin{equation}\label{thet_EG}
\psi_{\rm EG}\simeq 4^{\circ}\left(\frac{E_{p,th}}{57 EeV}\right)^{-1} \,\left(\frac{L_{\rm EG}}{100\, {\rm Mpc}}\right)^{1/2}\,\left(\frac{l_c}{1\, {\rm Mpc}}\right)^{1/2}\,\left( \frac{B_{\rm EG}}{1\,{\rm nG}} \right)\,,
\end{equation}
respectively, where L$_{\rm G}$ corresponds to the distance of our Galaxy (20 kpc), $l_c$ is the coherence length and $E_{\rm p,th}$ is the threshold proton energy. Due to the strength of extragalactic ($B_{\rm EG}\simeq$ 1 nG) and galactic ($B_{\rm G}\simeq$ 4 $\mu$G) magnetic fields, UHECRs are deflected; firstly, $\psi_{\rm EG}\simeq 4^{\circ}$ and after $\psi_G\simeq 3.8^{\circ}$, between the true direction to the source and the observed arrival direction, respectively. Estimation of the deflection angles could associate the transient UHECR sources with the HE neutrino and $\gamma$-ray fluxes. Taking into account of extragalactic and galactic magnetic fields, it is reasonable to correlate UHECRs lying within $\sim5^\circ$ of a source.
\subsection{Expected number of events}
\paragraph{Telescope Array observatory}. Located in Millard Country (Utah), TA experiment was designed to study UHECRs with energies above 57 EeV. With an area of $\sim$ 700 km$^2$, it is made of a scintillator surface detector (SD) array and three fluorescence detector (FD) stations \citep{2012NIMPA.689...87A}. The TA exposure is given by $\Xi\,t_{op}\, \omega(\delta_s)/\Omega$, where $\Xi\,t_{op}=(5)\,7\times10^2\,\rm km^2\,yr$, $t_{op} $ is the total operational time (from 2008 May 11 and 2013 May 4), $\omega(\delta_s)$ is an exposure correction factor for the declination of the source \citep{2001APh....14..271S} and $\Omega\simeq\pi$.
\paragraph{Pierre Auger observatory}. The PAO, located in the Mendoza Province of Argentina at latitude $\simeq 36^\circ$S, was designed to determine the arrival directions and energies of UHECRs using four fluorescence telescope arrays and 1600 surface detectors spaced 1.5 km apart. The large exposure of its ground array, combined with accurate energy has provided an opportunity to explore the spatial correlation between cosmic rays and their sources in the sky. The PAO exposure is given by $\Xi\,t_{op}\, \omega(\delta_s)/\Omega_{60}$, where $\Xi\,t_{op}=9\times10^3\,\rm km^2\,yr$, $t_{op} $ is the total operational time (from 1 January 2004 until 31 August 2007), $\omega(\delta_s)$ is an exposure correction factor and $\Omega_{60}\simeq\pi$ is the Auger acceptance solid angle \citep{2007Sci...318..938P, 2008APh....29..188P}.
\paragraph{The High Resolution Fly's eye experiment}. The HiRes experiment, located atop two hills 12.6 km apart in the west-central Utah desert, consisted of two detectors that observed cosmic ray showers via the fluorescence light. Both detectors, called HiRes-I and HiRes-II, consisted of 21 and 42 telescopes, respectively, each one composed of a spherical mirror of 3.8 m$^2$. The HiRes exposure is (3.2 - 3.4) $\times 10^3$ km$^2$ year sr. HiRes experiment measured the flux of UHECRs using the stereoscopic air fluorescence technique over a period of nine years (1997 - 2006) \citep{2005ApJ...622..910A, 2009APh....32...53H}.\\
The expected number of UHECRs above an energy $E_{p,th}$ yields
\begin{equation}
N_{\rm \tiny UHECR}= ({\rm Expos.})\times \,N_p,
\label{num}
\end{equation}
where $N_p$ is calculated from the proton spectrum extended up to energies higher than $E_{p,th}$ (eq. \ref{prot_esp}). The expected number can be written as
\begin{eqnarray}
N_{\rm \tiny UHECR}=\frac{\Xi\,t_{\rm op}\, \omega(\delta_s)}{(\alpha_p-1)\Omega}\,A_p \int_{E_{\rm p,th}} E_{p}^{-\alpha_p} dE_p\,,
\label{nUHE1}
\end{eqnarray}
where the value of $A_p$ is normalized with the TeV $\gamma$-ray fluxes (eq. \ref{Apg}).
\section{Results and Discussion}
We have presented a lepto-hadronic model to describe the broadband SED of the closest RGs, supposing that electrons and protons are co-accelerated at the emitting region of the jet. In the leptonic model, we have required the SSC model to explain the spectrum up to dozens of GeV and in the hadronic scenario, we have evoked the p$\gamma$ interactions occurring close to the core of the RGs to interpret the TeV $\gamma$-ray fluxes. The SSC model depends basically on magnetic field ($B$), electron density ($N_e$), size of emitting region ($r_d$) and Doppler factor ($\delta_D$). To reproduce the electromagnetic spectrum up to dozens of GeV, we have used an electron distribution described by a broken power law (eq. \ref{espele}) with the minimum, cooling and maximum Lorentz factors given by eqs. (\ref{gamma_m}), (\ref{gamma_b}), (\ref{gamma_max}), respectively. The minimum Lorentz factor is obtained through electron density and electron energy density, the cooling Lorentz factor is computed through the synchrotron and dynamical time scales, and the maximum Lorentz factor is calculated considering the synchrotron and the acceleration time scales. Considering the break Lorentz factors, the synchrotron and Compton scattering break energies are obtained (eqs. \ref{synrad} and \ref{icrad}). The synchrotron (eq. \ref{espsyn}) and Compton scattering (eq. \ref{espic}) spectra are estimated through the density and distribution of radiating electrons confined inside emitting region. In the p$\gamma$ interaction model, we have considered Fermi-accelerated protons described by a simple power law (eq. \ref{prot_esp}) which are accelerated close to the core and furthermore interact with the photon population at the second-peak SED photons. The spectrum generated by this hadronic process (eq. \ref{pgammam}) depends on the proton luminosity (through $A_p$), the observed luminosity around the second SSC peak ($L_{\rm \gamma, IC}$), the energy of the second SSC peak ($\epsilon_{\rm pk,ic}$), the size of emitting region, the Doppler factor, and the high-energy and low-energy photon index, respectively. The efficiency of the photo-production (eq. \ref{eficiency}) is calculated through the photo-pion cooling and dynamical time scales for $\beta_h\sim2$ and $\beta_l\sim1$. Evoking these interactions, we have interpreted naturally the TeV $\gamma$-ray spectra as $\pi^0$ decay products.\\
We have required the data used by the Fermi collaboration for Cen A \citep{2010ApJ...719.1433A}, M87 \citep{2009ApJ...707...55A} and NGC1275 \citep{2009ApJ...699...31A}. In the case of NGC1275, the TeV $\gamma$-ray data have been added \citep{2012A&A...539L...2A}. Using the method of Chi-square $ \chi^2$ minimization as implemented in the ROOT software package \citep{1997NIMPA.389...81B}, we get the break energies, spectral indexes and normalization of the SSC and photo-pion spectrum as reported in Table 1 and Figures \ref{cenA}, \ref{m87} and \ref{ngc1275} for Cen A , M87 and NGC1275, respectively. For the sake of simplicity, the process was integrated into a python script that called the ROOT routines via the pyroot module. The whole data of the closest RGs were read from a file and written into an array which was then fitted with pyROOT. The fitted data with the PyROOT module and from eqs. (\ref{espsyn}), (\ref{espic}) and (\ref{pgammam}) were plotted with a smooth sbezier curve through the gnuplot software \footnote{gnuplot.sourceforge.net}. The sbezier option first renders the data monotonic (unique) and then applies the Bezier algorithm\footnote{Bezier curves are widely used in computer graphics to model smooth curves using the Casteljau algorithm which is a method to split a single Bezier curve into two Bezier curves at an arbitrary parameter value. http://web.mit.edu/hyperbook/Patrikalakis-Maekawa-Cho/node13.html}.
\begin{center}\renewcommand{\arraystretch}{1.3}\addtolength{\tabcolsep}{4pt}
\begin{center}
\scriptsize{\textbf{Table 1. Values obtained after fitting the whole spectrum of the RGs with our lepto-hadronic model.}}\\
\end{center}
\begin{tabular}{ l c c c c}
\hline \hline
\scriptsize{Parameter} &\scriptsize{Symbol} & \scriptsize{Cen A}& \scriptsize{M87} & \scriptsize{NGC1275}\\
\hline
\hline
\multicolumn{2}{c}{Leptonic model} \\
\cline{1-2}
\scriptsize{$ A_{\rm syn,\gamma}\,\, ({\rm MeV\,cm^{-2}\,s^{-1}})$} &\scriptsize{[0]} & \scriptsize{$(0.25\pm 0.05)\times 10^{-4}$}& \scriptsize{$(0.66\pm 0.08)\times 10^{-4}$} & \scriptsize{$(1.25\pm 0.26)\times 10^{-5}$} \\
\scriptsize{$\alpha_e$} & \scriptsize{[1]} & \scriptsize{3.50$\pm$ 0.02} & \scriptsize{3.21$\pm$ 0.02} & \scriptsize{2.81$\pm$ 0.05} \\
\scriptsize{$ \epsilon^{syn}_{\rm \gamma,m}\,\, ({\rm eV})$} & \scriptsize{[2]} & \scriptsize{ $0.046\pm 0.002$}& \scriptsize{$(2.26\pm 0.25)\times 10^{-2}$} & \scriptsize{$(1.25\pm 0.05)\times 10^{-2}$} \\
\scriptsize{$\epsilon^{syn}_{\rm \gamma,c}\,\, ({\rm eV})$} & \scriptsize{[3]} & \scriptsize{0.15$\pm$ 0.02} & \scriptsize{$0.15\pm 0.02$} & \scriptsize{ $0.33\pm 0.02$} \\
\hline
\scriptsize{$ A_{\rm ssc,\gamma}\,\, ({\rm MeV\,cm^{-2}\,s^{-1}})$} &\scriptsize{[4]} & \scriptsize{$(6.65\pm 0.67) \times 10^{-4}$} & \scriptsize{ $(1.31\pm 0.75) \times 10^{-5}$} & \scriptsize{ $(3.99\pm 0.86) \times 10^{-4}$} \\
\scriptsize{$ \epsilon^{\rm ssc}_{\rm \gamma,m}\,\, ({\rm keV}) $} & \scriptsize{[5]} & \scriptsize{ $67.1\pm 1.51$} & \scriptsize{ $11.2\pm 0.9$} & \scriptsize{$3.15\pm 0.16$} \\
\scriptsize{$\epsilon^{\rm ssc}_{\rm \gamma,c}\,\, ({\rm MeV})$} & \scriptsize{[6]} & \scriptsize{$0.53\pm 0.02$} & \scriptsize{$0.41\pm 0.15$} & \scriptsize{$2.1\pm 0.1$} \\
\hline
\multicolumn{2}{c}{Hadronic model} \\
\cline{1-2}
\scriptsize{$ A_{\rm p\gamma} \,\, ({\rm MeV\,cm^{-2}\,s^{-1}})$} & \scriptsize{[7]} & \scriptsize{$(3.98\pm 0.18)\times 10^{-7}$}& \scriptsize{$(1.17\pm 0.24)\times 10^{-7}$} & \scriptsize{$(3.69\pm 0.91)\times 10^{-7}$} \\
\scriptsize{$\alpha_p$} & \scriptsize{[8]} & \scriptsize{2.81$\pm$ 0.05}& \scriptsize{2.80$\pm$ 0.02} & \scriptsize{ $3.81\pm 0.64$} \\
\hline
\end{tabular}
\end{center}
From eqs. (\ref{synrad}), (\ref{Asyn}), (\ref{icrad}) and the values reported in Table 1, we have obtained the values of magnetic field, electron density, size of emitting region and Doppler factor that describe the broadband SED of Cen A, M87 and NGC1275. For this fit, we have considered the effect of the electromagnetic background light (EBL) absorption \citep{2008A&A...487..837F} and adopted the typical values reported in the literature such as viewing angles, observed luminosities, distances of the closest RGs, minimum Lorentz factors and energy normalizations \citep{2007ApJ...654..186F,1991ApJ...373L...1T, 1992ApJS...83...29S, 2010ApJ...719.1433A, 1996ApJ...467..597B, 2006PASJ...58..261A, 2009ApJ...695L..40A, 2006Sci...314.1424A, 2012A&A...539L...2A, 2009ApJ...707...55A, 2009ApJ...699...31A, 1998A&A...330...97S, 2009ApJ...690..367A}. Other quantities such as bulk Lorentz factors, proton and electron luminosities, magnetic field, proton and electron densities, etc, are derived from these parameters. Table 2 shows all the parameter values obtained, used and derived in and from the fit.\\
The maximum proton Lorentz factors were estimated with the maximum electron Lorentz factors as {\small $\gamma_{p,max}=\frac{m_p}{m_e}\gamma_{e,max}$} \citep{2014A&A...562A..12P}. Requiring that electron and proton number densities are similar ($N_e\simeq N_p$), we have calculated the minimum proton Lorentz factors $\gamma_{p,min}$. These values could increase/decrease considering that proton number density is smaller/higher than the electron number density. For instance, if we assume that ($N_p\simeq b\, N_e$) with b=10 (0.1) the minimum proton Lorentz factors are $\gamma_{p,min}= 9.1\times10^3$ ($1.6\times10^6$), $3.2\times10^5$ ($6.7\times10^7$) and $7.1\times10^5$ ($9.2\times10^6$) for Cen A, M87 and NGC1275, respectively. Due to the fact that the minimum Lorentz factor cannot be determined just assuming the standard scenario of injection and acceleration, we have used the condition that electron and proton number densities are similar. Otherwise, for $\gamma_{p,min}=1$ the proton luminosities becomes $>5\times 10^{47}\, {\rm erg/s}$. Considering the values of magnetic field, electron and proton energy densities, and their rates; $\lambda_{e,B}=\frac{U_e}{U_B}$ ($\lambda_{p,B}=\frac{U_p}{U_B}$)= 2.92 (4.90), 24.1 (26.9) and 117.81 (382.56) for Cen A, M87 and NGC1275, respectively, we can see that energy densities could be related through principle of equipartition.\\
\begin{center}\renewcommand{\arraystretch}{1.1}\addtolength{\tabcolsep}{4pt}
\vspace{4cm}
\begin{center}
\scriptsize{\textbf{Table 2. Parameters obtained, derived and used of lepton-hadronic model to fit the spectrum of Cen A, M87 and NGC1275.}}\\
\end{center}
\begin{tabular}{ l c c c c c}
\hline
\hline
\normalsize{} & \normalsize{Cen A}& \normalsize{M87}&\normalsize{NGC1275} \\
\hline
\hline
\multicolumn{2}{c}{Obtained quantities} \\
\cline{1-2}
\scriptsize{$\delta_d$} & \scriptsize{1.0} & \scriptsize{2.8}& \scriptsize{2.2} \\
\scriptsize{$B$ (G)} & \scriptsize{3.6}& \scriptsize{1.61}& \scriptsize{2.01} \\
\scriptsize{$r_d$ (cm)} & \scriptsize{$5.2 \times 10^{15}\,$} & \scriptsize{$2.1 \times 10^{15}\,$} & \scriptsize{$2.26 \times 10^{15}\,$} \\
\scriptsize{$N_e$ (cm$^{-3}$)} & \scriptsize{$1.1\times 10^3$} & \scriptsize{$2.3\times 10^3$}& \scriptsize{$3.11\times 10^4$} \\\hline
\multicolumn{2}{c}{Used quantities} & & & \small{References}\\
\cline{1-2}
\scriptsize{$d_z\,\,({\rm Mpc})$} & \scriptsize{3.7} & \scriptsize{16}& \scriptsize{76} & \scriptsize{(1,2,3)} \\
\scriptsize{$\theta$}\,\,({\rm degree}) & \scriptsize{30}& \scriptsize{17}& \scriptsize{20} & \scriptsize{(4,5,6)} \\
\scriptsize{$\epsilon_0$}\,\,({\rm TeV}) & \scriptsize{1} & \scriptsize{1}& \scriptsize{1} & \scriptsize{(7,8,9)} \\
\scriptsize{$\gamma_{\rm e,min}$} & \scriptsize{$1.2\times10^3$}& \scriptsize{$0.67\times10^3$}& \scriptsize{$0.5\times10^3$} & \scriptsize{(4,10,11)} \\
\scriptsize{$L_{\rm \gamma, IC}\,\, {\rm (erg/s)}$} & \scriptsize{$5\times 10^{42}$}& \scriptsize{$5\times 10^{42}$}& \scriptsize{$5\times 10^{42}$} & \scriptsize{(12, 13)}\\\hline
\multicolumn{2}{c}{Derived quatities} \\
\cline{1-2}
\scriptsize{$\Gamma$} & \scriptsize{7.0}& \scriptsize{6.47}& \scriptsize{6.29} \\
\scriptsize{$f_{\pi^0}$} & \scriptsize{$1.05\times10^{-6}$ }& \scriptsize{ $6.68\times10^{-8}$ }& \scriptsize{ $1.55\times10^{-8}$ } \\
\scriptsize{$\tau_{\gamma}$} & \scriptsize{$7.89\times10^{-6}$ }& \scriptsize{ $1.40\times10^{-6}$ }& \scriptsize{ $2.56\times10^{-7}$} \\
\scriptsize{$\gamma_{e,max}$} & \scriptsize{$3.64\times10^7$}& \scriptsize{$3.70\times10^7$}& \scriptsize{$3.29\times10^7$} \\
\scriptsize{$\gamma_{p,min}$} & \scriptsize{$1.0\times10^5$}& \scriptsize{$7.0\times10^6$}& \scriptsize{$2.4\times10^6$} \\
\scriptsize{$^a\gamma_{p,max}$} & \scriptsize{$6.73\times10^{10}$}& \scriptsize{$6.76\times10^{10}$}& \scriptsize{$6.04\times10^{10}$} \\
\scriptsize{$\epsilon_{\rm pk, ic}\,\, ({\rm MeV}) $ } & \scriptsize{$0.1$}& \scriptsize{$0.5$}& \scriptsize{$3.5$} \\
\scriptsize{$\epsilon_{\pi^0,\gamma,c}\,\,({\rm TeV})$} & \scriptsize{$0.32$}& \scriptsize{$0.54$}& \scriptsize{$0.05$} \\
\scriptsize{$U_B\,\, {\rm (erg/cm^3)}$} & \scriptsize{$0.52$}& \scriptsize{$0.10$}& \scriptsize{$0.16$} \\
\scriptsize{$U_e\,\, {\rm (erg/cm^3})$} & \scriptsize{$1.52$}& \scriptsize{$2.41$}& \scriptsize{$28.45$} \\
\scriptsize{$U_p\,\, {\rm (erg/cm^3})$} & \scriptsize{$2.55$}& \scriptsize{$2.69$}& \scriptsize{$61.21$} \\
\scriptsize{$L_p\,\, {\rm (erg/s)}$} & \scriptsize{$3.74\times 10^{43}$}& \scriptsize{$4.89\times 10^{43}$}& \scriptsize{$2.29\times 10^{44}$} \\
\scriptsize{$L_e\,\, {\rm (erg/s})$} & \scriptsize{$1.55\times 10^{43}$}& \scriptsize{$3.17\times 10^{43}$}& \scriptsize{$2.27\times 10^{44}$} \\
\scriptsize{$E_{\rm p,max}\,\, {\rm (EeV)}$} & \scriptsize{$40.1$} & \scriptsize{$6.55$}& \scriptsize{$7.92$} \\
\scriptsize{$N_{\rm \tiny UHECRs}$} & \scriptsize{$1.52$} & \scriptsize{$0.41$}& \scriptsize{$2.63\times 10^{-6}$} \\
\hline
\hline
\end{tabular}
\end{center}
\begin{flushleft}
\scriptsize{
Notes.\\
$^a$ $\gamma_{p,max}=\frac{m_p}{m_e}\gamma_{e,max}$.\\
\textbf{References}. (1) \cite{2007ApJ...654..186F} (2) \cite{1991ApJ...373L...1T} (3) \cite{1992ApJS...83...29S} (4) \cite{2010ApJ...719.1433A} (5) \cite{1996ApJ...467..597B} (6) \cite{2006PASJ...58..261A} (7) \cite{2009ApJ...695L..40A} (8) \cite{2006Sci...314.1424A} (9) \cite{2012A&A...539L...2A} (11) \cite{2009ApJ...707...55A} (10) \cite{2009ApJ...699...31A} (12) \cite{1998A&A...330...97S} (13) \cite{2009ApJ...690..367A}.}
\end{flushleft}
We plot in a sky-map the 54 neutrino events detected by the IceCube collaboration, the 72 and 27 UHECRs collected by TA and PAO experiments, respectively, and also a circular region of 5$^\circ$ around the closest RGs, as shown in Figure \ref{Skymap}. This figure shows that whereas there are not neutrino track events associated to Cen A, M87 and NGC1275, two UHECRs are only enclosed around Cen A.\\
By assuming that the inner part of the RG jet has the potential to accelerate particles up to UHEs, we can see that protons at the emitting region can achieve maximum energies of 40.1, 6.55 and 7.92 EeV for Cen A, M87 and NGC1275, respectively, as shown in Table 2. Therefore, it is improbable that protons can be accelerated to energies as high as 57 EeV, although it is plausible for heavier accelerated ions. In this case, they can be disintegrated by infrared photons from the core. It is worth noting that any small variation in the strength of magnetic field and/or size of emitting region would allow that protons could achieve a maximum energy of 57 EeV for Cen A. Similarly, supposing that the BH jet has the power also to accelerate particles up to UHEs through Fermi processes, then during flaring intervals (for which the apparent isotropic luminosity can reach $\approx 10^{46}$ erg s$^{-1}$ and from the equipartition magnetic field $\epsilon_B$) the maximum particle energy of accelerated UHECRs can achieve values as high as {\small $E_{\rm max}\approx 3.0\times10^{20}\,\frac{Zq_e\,\epsilon^{1/2}_B}{\phi\, \Gamma}\,\left(\frac{L}{10^{46}\,{\rm erg/s} }\right)^{1/2}\, {\rm eV}$} \citep{2009NJPh...11f5016D}. Describing the TeV gamma-ray spectra through p$\gamma$ interaction and extrapolating the interacting proton spectra up to energies higher than 1 EeV, we plot the UHE proton spectra expected for these three RGs (Cen A, M87 and NGC1275) with the UHECR spectra collected with PAO \citep{2011arXiv1107.4809T}, HiRes \citep{2009APh....32...53H} and TA \citep{2013ApJ...768L...1A} experiment as shown in Figure \ref{uhecr_flux}. We can see that as energy increases the discrepancy between the UHE proton fluxes and the observed UHECR spectra increases. Taking into account the TA and PAO exposures, we estimate the number of UHECRs above 57 EeV, as shown in Table 2. The number of UHECRs computed with our model are 1.52, 0.41 and $2.63\times 10^{-6}$ for Cen A, M87 and NGC1275, respectively. Due to extragalactic (eq. \ref{thet_EG}) and galactic (eq. \ref{thet_G}) magnetic fields, UHECRs are deflected between the true direction to the source, and the observed arrival direction. Regarding these considerations, the total deflection angle could be as large as the mean value of $<\theta_T>\simeq$ 15$^\circ $\citep{2010ApJ...710.1422R}. Therefore, it is reasonable to assume a circular region with $5^\circ$ (eqs. \ref{thet_EG}) centered around each source (see fig. \ref{Skymap}). Taking into account these regions, we can see two UHECRs associated to Cen A and none to M87 and NGC1275. We can see that number of UHECRs calculated with our model is consistent with those reported by the TA and PAO collaborations, although the maximum proton energies are less than 57 EeV. It is worth noting that the latter results reported by PAO suggested that UHECRs are heavy nuclei instead of protons \citep{2010PhRvL.104i1101A}. If UHECRs have a heavy composition, then a significant fraction of nuclei must survive photodisintegrations in their sources \citep{2010MNRAS.405.2810H,2009MNRAS.393.1041H}. In this case for Z$\gtrsim$2, the emitting region can achieve maximum energies for heavy nuclei of $\gtrsim$ 80, 13 and 16 EeV for Cen A, M87 and NGC1275, respectively.\\
After fitting the TeV $\gamma$-ray spectra of the RGs with our hadronic model, from eq. (\ref{numneu1}) we obtain the neutrino fluxes and events expected in a hypothetical Km$^{3}$ neutrino telescope through the Monte Carlo simulations. We consider a point source neutrino emitters at Cen A, M87 and NGC1275 positions with a energy range spanning from 10 GeV to 10 PeV. The neutrino spectra are normalized from the observed TeV $\gamma$-ray spectra, assuming that the TeV $\gamma$-ray fluxes from the closest RGs are interpreted through the $\pi^0$ decay products from the p$\gamma$ interactions. The values of magnetic field, Doppler factor and emitting radius (see Table 2) were calculated as the result of fitting the SED with SSC model up to dozens of GeV. Changes in these observables would vary the photon density generated by synchrotron radiation, and thus the number of photons scattered by inverse Compton scattering. As neutrino fluxes were computed from the photo-hadronic interactions between Fermi-accelerated protons and the seed photons around the first and second SSC peaks, then variations in the magnetic field, Doppler factor and emitting radius would affect the target photon densities (eq. \ref{den}) and then the neutrino fluxes. For the simulated neutrino telescope we additionally calculate the atmospheric and cosmic neutrinos expected from the portion of the sky inside a circular region centered in each source and covering $1^{\circ}$ square. The atmospheric neutrino flux is described using the Bartol model \citep{2004PhRvD..70b3006B,2006PhRvD..74i4009B} for the range of energy considered in this analysis. The cosmic diffuse neutrinos signal has been discussed by Waxman and Bahcall \citep{2001PhRvD..64b3002B, wax98} and the upper bound for this flux is $E^{2}_{\nu}\,d\Phi/dE_{\nu} <2\times 10^{-8}\, {\rm GeV\,cm^{-2}\,s^{-1}\,sr^{-1}}$ \citep{1999PhRvD..59b3002W}. We rule out the atmospheric muon ``background'' from this analysis due to earth filtration and to the softener spectrum with respect to the neutrino signal and ``backgrounds'' considered. We plot the neutrino spectra expected for CenA, M87 and NGC1275, as shown in figure \ref{neutr_flux}. In this figure we have considered Fermi-accelerated protons interacting with the seed photons around the first and second SSC peaks, for this reason neutrino spectra are exhibited, firstly, by declining power laws and after by peaking around 1 EeV as expected \citep{2008PhRvD..78b3007C}. By computing the signal to noise ratio for one year of observation in Km$^3$ neutrino telescope, the neutrino IceCube flux \citep{2014PhRvL.113j1101A} and the upper limits set by IceCube (IC40; \cite{2011PhRvD..83i2003A}), PAO \citep{2011PhRvD..84l2005A}, RICE \citep{2012PhRvD..85f2004K} and ANITA \citep{2010PhRvD..82b2004G} we confirm the impossibility to observe HE and UHE neutrinos from RGs. No visible neutrino excess (under the assumption of p$\gamma$ interaction model) respect to atmospheric and cosmic neutrinos is expected for the three RGs considered, as it is shown in figure \ref{neut_event}. In our model, the proton and neutrino spectra are normalized with the TeV $\gamma$-ray fluxes from the closest RGs, therefore to obtain the diffuse flux from other RGs would be needed to model the TeV $\gamma$-ray flux of each RG which is outside of the scope of this paper. It is important to say that if UHECRs are heavy as suggested by PAO \citep{2010PhRvL.104i1101A}, HE neutrinos from UHE nuclei are significant lower than the neutrino flux obtained by UHE protons \citep{2010PhRvD..81l3001M}.
\vspace{1cm}
\section{Summary and conclusions}
We have proposed a leptonic and hadronic model to explain the broadband SED spectrum observed in the closest RGs. In the leptonic model, we have used the SSC emission to describe the SED up to dozens of GeV. To explain the spectrum from hundreds of GeV up to a few TeV, we have introduced the hadronic model assuming that accelerated protons in the inner jet interact with the photon population at the SED peaks. Evoking these interaction, we have interpreted the TeV $\gamma$-ray spectra as $\pi^0$ decay products in Cen A, M87 and NGC 1275.\\
Correlating the TeV $\gamma$-ray and HE neutrino fluxes through p$\gamma$ interactions, we have computed the HE and UHE neutrino fluxes, and the neutrino event rate expected in a kilometric scale neutrino detector. The neutrino event rate was obtained through MC simulation by considering a region of 1$^{\circ}$ around the source position and assuming a hypothetical Km$^3$ Cherenkov telescope. We found that the neutrino fluxes produced by p$\gamma$ interactions close to the core of RGs cannot explain the astrophysical flux and the expected $\nu_{\mu}$ events in a neutrino telescope are consistent with the nonneutrino track-like associated with the location of the closest RGs \citep{2013PhRvL.111b1103A, 2013Sci...342E...1I, 2014PhRvL.113j1101A, 2015ATel.7856....1S}. In addition, the atmospheric muon neutrino background is also shown.\\
Extrapolating the proton spectrum by a simple power law up to UHEs, we have computed the number of UHECRs expected from Cen A, M87 and NGC1275. We found that those UHECRs obtained with our model is in agreement with the TA and PAO observations. Although UHECRs from Cen A can hardly be accelerated up to the PAO energy range at the emitting region (E$_{max}$= 40 EeV), they could be accelerated during the flaring intervals, with small changes in the strength of magnetic field and/or emitting region and in the giant lobes. It is very interesting the idea that UHECRs could be accelerated partially in the jet at energies ($<40\times 10^{19}$ eV) and partially in the Lobes at ($E>40\times 10^{19}$ eV) \citep{2014ApJ...783...44F}. If UHECRs have a heavy composition as suggested by PAO \citep{2010PhRvL.104i1101A}, UHE heavy nuclei in Cen A could be accelerated at energies greater than $\sim$ 80 EeV, thus reproducing the detections reported by PAO. It is worth noting that if radio Galaxies are the sources of UHECRs, their on-axis counterparts (i.e. blazars, and flat-spectrum radio quasars) should be considered a more powerful neutrino emitters \citep{2001PhRvL..87v1102A, 2014JHEAp...3...29D,2014PhRvD..90b3007M}. In fact, a PeV neutrino shower-like was recently associated with the flaring activity of the blazar PKS B1424-418 \citep{2016arXiv160202012K}. \\
In summary, we have showed that leptonic SSC and hadronic processes are required to explain the $\gamma$-ray fluxes at GeV- TeV energy range. We have successfully described the TeV $\gamma$-ray \citep{2009ApJ...695L..40A, 2006Sci...314.1424A, 2010ApJ...710..634A}, HE neutrinos \citep{2013PhRvL.111b1103A, 2013Sci...342E...1I, 2014PhRvL.113j1101A, 2015ATel.7856....1S} and UHECRs \citep{2011arXiv1107.4809T, 2013ApJ...768L...1A} around the closest RGs.
\acknowledgments
We thank the anonymous referee for a critical reading of the paper and valuable suggestions that helped improve the quality and clarity of this work. We also thank to Charles Dermer, Tyce DeYoung, Markus B\"{o}tcher, Bin Zhang for useful discussions.
| 74abc0a9da6f5ee75201a26a53581751d0f3b4af | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Bulk acoustic wave (BAW) piezoelectric devices are widely used in research and industry to make temperature sensors, acoustic cavities or filters for several applications. Recently it has been demonstrated that quartz acoustic cavities can exhibit very high quality factors, up to a few billions for the best ones, at liquid helium temperature~\cite{galliouScRip2013}. In these conditions they become very attractive for various experiments in fundamental physics~\cite{Muller2016, Goryachev:2014aa}, as well as for hybrid quantum systems~\cite{Goryachev:2013aa, AspelmeyerNature2010}, optomechanics~\cite{Aspelmeyer2014}, etc. This interest is amplified by the fact that commercial cryorefrigerators are now easy to use. In the field of ultra stable frequency standards, the advent of superconducting devices also offers new interesting opportunities. The coupling of those cavities exhibiting extremely low mechanical losses with SQUID (Superconducting QUantum Interference Device) is a first step towards cryogenic ultra-stable frequency sources~\cite{GoryachevSQUID2014}. Nevertheless, progress in this field suffer of the lack of knowledge regarding the physical constants of the acoustic cavity material, the crystalline quartz, at cryogenic temperatures. The very first topic of interest is the behavior of elastic constants versus temperature. Although a set of temperature coefficients of the elastic constants of quartz already exists since 1975~\cite{Smagin}, it covers the very large temperature range $4 {\rm K}-400 {\rm K}$, and it turns out that it is not reliable within the very useful range $4 {\rm K}-15 {\rm K}$ commonly achieved by standard cryocoolers. In addition, these coefficients have been determined from the fast and slow (quasi-)shear propagation modes usually denoted B and C respectively, and not from the (quasi-)longitudinal mode, denoted A, the one which exhibits the highest quality factors at low temperatures.
Several methods can be used to determine the thermal coefficients. It can be from phase velocity measurements by pulse echo overlap methods, resonant ultrasound spectroscopy or combined resonance techniques, or by means of laser ultrasonic techniques. However, application of these methods at low temperature arises technical problems. Alternatively, one way consists in probing the temperature sensitivity of a resonance frequency from a dedicated acoustic device. Indeed, the acoustic resonance frequency of a mechanical system is a function of its elastic coefficients which are themselves temperature dependent. This method is particularly relevant for a quartz crystal device: the temperature sensitivity of the elastic coefficients of the quartz material can be determined by identifying the frequency-to-temperature coefficients of a plano-convex quartz resonator, i.e. a trapped energy resonator, whose frequency of overtone order $n$ derives from the basic relationship $f_n=\frac{n}{2h}\sqrt{\frac{C_{e}}{\rho}}$ where $2h$ is a thickness, $C_{e}$ an elastic coefficient as described below, and $\rho$ the quartz density. This latter approach is used here to extract the temperature coefficients of quartz stiffness within $4 {\rm K}-15 {\rm K}$ from various quartz-cut with different orientations relative to the crystallographic reference.
\section{Method}
The Eulerian formalism can be used to determine the thermal coefficients of the elastic coefficients $C_{ijkl}$ because the linear thermal expansion coefficients of quartz are known at the operating temperatures~\cite{Barron}.
Strictly speaking, according to the Eulerian description, the problem should be solved in the so-called final state~\cite{Dulmet2001},~\cite{Dulmet2001bis} by adding the mechanical strains caused by the acoustic vibration to the temperature strains induced by the temperature change from room temperature to cryogenic temperature. However, as an available approximation, the problem can be solved in the intermediate state by taking into account the strains induced by the very small vibration-caused displacement magnitude, typically in the order of a nanometer.
In the Euler formalism the strain-stress relationship is given by the Cauchy tensor expressed, for the intermediate state, by $T_{ij}=C_{ijkl}\frac{\partial u_k}{\partial x_l}$ where $C_{ijkl}$ is the elastic coefficient tensor and $u_k$ a component of the displacement vector. The eigenfrequencies of the acoustical modes of the mechanical structure obviously depend on the elastic coefficients $C_{ijkl}$ that are temperature dependent.
To bring up the thermal coefficients, the elastic coefficients can be written in the form of a Taylor series:
\begin{equation}
\begin{array}{l}
C_{ijkl}(T)=C_{ijkl}(T_{ref})\left[1+\sum_k T^{(k)} C_{ijkl} \right. \vspace{8pt}\\
\hspace{48pt} \left.(T-T_{ref})^k\right]
\end{array}
\label{eq:elastic_coef}
\end{equation}
where $C_{ijkl}(T_{ref})$ are the elastic coefficients at the reference temperature $T_{ref}$, and $T^{(k)} C_{ijkl}$ the order $k$ thermal coefficients defined by:
\begin{equation}
T^{(k)} C_{ijkl} = \frac{1}{k!}\frac{1}{C_{ijkl}(T_{ref})}\frac{d^{k}C_{ijkl}}{dT^{k}}
\label{eq:Taylor_coef}
\end{equation}
The eigenfrequencies of the plano-convex resonators measured in this study (see Fig.~\ref{fig:plano_convex_scheme}) can be expressed according to the Stevens-Tiersten model~\cite{Tiersten1986}:
\begin{equation}
\begin{array}{l}
f_{nmp}^2=f_n^2\left[1+\frac{1}{n\pi}\sqrt{\frac{2h}{R_0}}\left(\sqrt{\frac{M_n}{C_e}}(2m+1)\right.\right. \vspace{8pt}\\
\hspace{32pt} \left.\left.+\sqrt{\frac{P_n}{C_e}}(2p+1)\right)\right]
\end{array}
\label{eq:Euler_eigenfrequency}
\end{equation}
where index $n$ is the overtone number and $m,p$ the number of modal lines in the resonator plane, $M_n$ and $P_n$ the dispersion constants, $2h$ the resonator thickness, $R_0$ the curvature radius of the convex face, and $f_n$ the eigenfrequencies of the corresponding infinite flat plate. As shown in Fig.~\ref{fig:eigenfrequencies_AT-cut_plano_convex}, the latter can usually be computed accurately from the eigenfrequencies of the plano-convex resonator by combining both anharmonic modes $n02$ and $n20$, and the corresponding quasi-harmonic one $n00$ as follows:
\begin{equation}
f_n^2=\frac{(6f_{n00}^2-f_{n02}^2-f_{n20}^2)}{4}
\label{eq:Plate_convex_eigenfrequency}
\end{equation}
\begin{figure}[t!]
\centering
\includegraphics[width=3.25in]{resonator_scheme_SG}
\captionof{figure}{Plano-convex resonator. In such a trapped-energy device the vibration displacement $u_2$ along the Y-axis has a gaussian distribution \cite{Tiersten1986}.}
\label{fig:plano_convex_scheme}
\end{figure}
\begin{figure}[t!]
\centering
\includegraphics[width=3.25in]{mode_AT5_plan-convex_plaque_plane}
\captionof{figure}{Fractional frequency deviation of $f_5$ $f_{502}$ and $f_{520}$ versus the temperature offset with respect to $T_{ref}=8 \rm{K}$, in the case of the single-rotated Y-cut called AT-cut, as an example. Plots of the $5^{th}$ order resonance frequency $f_5$ of the infinite flat plate calculated from the plano-convex resonator frequencies $f_{5mp}$ are in very good agreement with \eqref{eq:Plate_convex_eigenfrequency}.}
\label{fig:eigenfrequencies_AT-cut_plano_convex}
\end{figure}
As a consequence, the temperature coefficients $T^{(k)}C_{ijkl}$ of the elastic constants can just be determined from the temperature coefficients of the eigenfrequencies of the corresponding infinite flat plate. These eigenfrequencies are given by \cite{Tiersten1986} :
\begin{equation}
{f_n^\xi}^2(T)=\frac{n^2}{4h^2(T)}\frac{\overline{C_{e}^\xi}(T)}{\rho (T)}\left(1-\frac{8k_\alpha^2(T)}{n^2 \pi^2}-r \right)
\label{eq:Plate_eigenfrequency}
\end{equation}
where $\overline{C_{e}^\xi}$ is the effective elastic coefficient, $k_\alpha^2$ the coupling electromechanical coefficient, $h$ the resonator thickness (about 1 mm; the electrode thickness is neglected because it is small enough, a few hundreds of nanometers, to neglect the wave propagation inside the electrodes), $\rho$ the mass density of quartz, $n$ the overtone range and $r$ the ratio between masses of the resonator and electrodes (typically $10^{-3}$). $\xi$ denotes A, B or C, referring to the extensional propagation mode, the high-speed thickness mode, and the low-speed thickness mode respectively.
Actually, terms inside the brackets of~\eqref{eq:Plate_eigenfrequency} can be reduced to unity in the $T^{(k)}C_{ijkl}$ calculation. Indeed, the third term $r$ is a constant independent of temperature, and then is not involved in this calculation. Regarding the second term inside the brackets of~\eqref{eq:Plate_eigenfrequency}, values of the electromechanical coefficients $k_\alpha^2$ are known only at room temperature, but are close to $10^{-4}$. In addition, as any physical constant, their thermal sensitivity are smaller at low temperature than at room temperature. Thus, this second term can also be reasonably neglected especially for overtone orders greater than 3.
The effective elastic coefficients $\overline{C_{e}^\xi}$ are eigenvalues in the Christoffel theory, given by \cite{Auld} in the case of a piezoelectric material:
\begin{equation}
\overline{\Gamma_{il}}=C_{ijkl}^E n_j n_k + \frac{(e_{pij} n_p )(e_{qkl} n_q )}{(\epsilon_{jk}^S n_j n_k )}
\label{eq:Christoffel_1}
\end{equation}
where $C_{ijkl}^E$ are the elastic coefficients measured at constant electric field, $e_{pij}$ and $e_{qkl}$ the piezoelectric constants, $\epsilon_{jk}^S$ the dielectric constant at constant strain and $n_\beta$ (with $\beta=p,q,j$ or $k$) the directions of propagation. The second term of~\eqref{eq:Christoffel_1} is linked to the electromechanical coefficient, and it turns out that it can also be neglected in comparison with the first one.
To sum up, it results from the simplifications described above that piezoelectricy is not involved anymore. According to these hypotheses the issue is then limited to its pure mechanical nature, and it turns out that: $\overline{C_{e}^\xi}\approx C_{e}^\xi$ and $\overline{\Gamma_{il}} \approx \Gamma_{il}$.
Coefficients $C_{e}^{\xi}$ are eigenvalues of the Christoffel tensor\cite{Royer}:
\begin{equation}
\Gamma_{il} {}^\circ u_l = \rho V^2 {}^\circ u_i
\end{equation}
with $\Gamma_{il} = C_{ijkl} n_j n_k$, and where $V$ and ${}^\circ u_l$ are respectively the phase velocity and polarization of the acoustic wave.
Terms of the symmetrical tensor $\Gamma_{il}$ are defined in the crystalline class 32 of quartz as:
\begin{equation}
\begin{array}{lll}
\Gamma_{il} = C_{i11l} n_1^2+C_{i22l} n_2^2+C_{i33l} n_3^2+(C_{i12l}+ \vspace{4pt} \\
\hspace{25pt} C_{i21l})n_1 n_2+ (C_{i13l}+C_{i31l}) n_1 n_3+(C_{i23l}+ \vspace{4pt} \\
\hspace{25pt} C_{i32l}) n_2 n_3
\end{array}
\end{equation}
where the index $i$ and $l$ varies from 1 to 3.
The cubic shape of the frequency-temperature characteristic observed experimentally implies that the thermal coefficients of the elastic constants should be provided up to the $3^{rd}$ order. These thermal sensitivities can be obtained by differentiating equation~\eqref{eq:Plate_eigenfrequency} with respect to temperature:
\begin{equation}
\begin{array}{lll}
T^{(1)}C_{e} \approx 2T^{(1)}f+T^{(1)}\rho+2T^{(1)}h \vspace{6pt} \\
T^{(2)}C_{e} \approx \frac{1}{2}[T^{(1)}C_{e}]^2-[T^{(1)}f]^2-\frac{1}{2}[T^{(1)}\rho]^2- \vspace{6pt} \\
\hspace{42pt} [T^{(1)}h]^2+2T^{(2)}f+T^{(2)}\rho+2T^{(2)}h \vspace{6pt} \\
T^{(3)}C_{e} \approx 2T^{(3)}f-2T^{(1)}fT^{(2)}f+\frac{2}{3}T^{(1)}f+ \vspace{6pt} \\
\hspace{42pt} \frac{1}{3}[T^{(1)}\rho]^3-\frac{1}{3}[T^{(1)}C_{e}]^3+\frac{2}{3}[T^{(1)}h]^3- \vspace{6pt} \\
\hspace{42pt} T^{(1)}\rho T^{(2)}\rho-2T^{(1)}hT^{(2)}h+2T^{(3)}h+ \vspace{6pt} \\
\hspace{42pt} T^{(1)}C_{e}T^{(2)}C_{e}+T^{(3)}\rho
\end{array}
\label{eq:Thermal_coeff}
\end{equation}
where $T^{(k)}\gamma=\frac{1}{\gamma !}\frac{d^{k}\gamma}{dT^{k}}$, with $\gamma = C_e, h, \rho$, or $f$.
The temperature coefficients of the thickness $2h$ are simply the expansion coefficients of the quartz material within $4.5 K-15 K$, with $T_{ref}=8 K$. Their numerical values have been computed by a third order fitting of data provided by~\cite{Barron} (see Tab.~\ref{tab:table4}), and according to the proper resonator cut angles.
\begin{table} [h!]
\caption{\label{tab:table4}Interpolated values of expansion coefficients from ~\cite{Barron} for the quartz material over the cryogenic range [4.5K-15K]. $\alpha_{\perp}$ and $\alpha_{//}$ are the linear expansion coefficients along the X-axis and Z-axis respectively.}
\begin{ruledtabular}
\begin{tabular}{cccc}
& $\alpha^{(1)}, 10^{-8}$ & $\alpha^{(2)}, 10^{-8}$ & $\alpha^{(3)}, 10^{-8}$ \\
\hline
$\alpha_{\perp}$ $(\alpha_{11})$ & $0.9651$ & $0.5160$ & $0.1416$ \\
$\alpha_{//}$ $(\alpha_{33})$ & $-0.5134$ & $-0.1202$ & $0.0241$ \\
\end{tabular}
\end{ruledtabular}
\end{table}
The temperature coefficients of the mass density $\rho$ also depend on the previous expansion coefficients, and can be expressed as follows~\cite{Dulmet}:
\begin{equation}
\begin{array}{lll}
T^{(1)}\rho = -(2\alpha_{11}^{(1)}+\alpha_{33}^{(1)}) \vspace{6pt} \\
T^{(2)}\rho = -(2\alpha_{11}^{(2)}+\alpha_{33}^{(2)}-3(\alpha_{11}^{(1)})^2-\vspace{6pt} \\
\hspace{44pt} (\alpha_{33}^{(1)})^2-2\alpha_{11}^{(1)}\alpha_{33}^{(1)}) \vspace{6pt} \\
T^{(3)}\rho = -(2\alpha_{11}^{(3)}+\alpha_{33}^{(3)}-2\alpha_{11}^{(1)}(3\alpha_{11}^{(2)}+\vspace{6pt} \\
\hspace{44pt} \alpha_{33}^{(2)})-2\alpha_{33}^{(1)}(\alpha_{11}^{(2)}+\alpha_{33}^{(2)})+ \vspace{6pt} \\
\hspace{44pt} 3(\alpha_{11}^{(1)})^2 \alpha_{33}^{(1)}+2\alpha_{11}^{(1)}(\alpha_{33}^{(1)})^2+ \vspace{6pt} \\
\hspace{44pt} 4(\alpha_{11}^{(1)})^3+(\alpha_{33}^{(1)})^3)
\end{array}
\end{equation}
Regarding the frequency differentiation, it is obtained from a $3^{rd}$ order fitting of a set of experimental frequency-temperature plots recorded from quartz resonators of various cut angles described in Tab.~\ref{tab:table3}. In the Eulerian description, six frequency-temperature graphs are necessary to determine the three temperature coefficients of each of the seven elastic coefficients of the tensor $C_{ijkl}$ written in matrix notation as:
\begin{equation}
\begin{pmatrix}
\centering
C_{11} & C_{12} & C_{13} & C_{14} & 0 & 0 \\
C_{12} & C_{11} & C_{13} & -C_{14} & 0 & 0 \\
C_{13} & C_{13} & C_{33} & 0 & 0 & 0 \\
C_{14} & -C_{14} & 0 & C_{44} & 0 & 0 \\
0 & 0 & 0 & 0 & C_{44} & C_{14} \\
0 & 0 & 0 & 0 & C_{14} & C_{66} \\
\end{pmatrix}
\label{eq:matrix_Cijkl}
\end{equation}
In~\eqref{eq:matrix_Cijkl}, the elastic coefficients are given according to the specific rules of index contraction:
\begin{table} [h!]
\begin{tabular}{p{2.5cm}p{2.5cm}c}
\centering $(11) \leftrightarrow 1$ & \centering $(22) \leftrightarrow 2$ & $(33) \leftrightarrow 3$ \vspace{6pt} \\
\centering $(23)=(32) \leftrightarrow 4$ & \centering $(13)=(31) \leftrightarrow 5$ & $(12)=(21) \leftrightarrow 6$ \\
\end{tabular}
\end{table}
Cuts and resonance modes used for the calculation of the temperature coefficients are summarized in Tab.~\ref{tab:table1}.
The temperature coefficients $C_e^{\xi}$ can be linked to the temperature behavior of the elastic coefficients by:
\begin{equation}
T^{(k)}C_e^{\xi}=\left[A_{\xi j}\right]^{-1} T^{(k)}C_{j}
\end{equation}
Where $\left[A_{\xi j} \right] = \frac{C_{e}^{\xi}}{C_{j}}\frac{\partial C_{e}^{\xi}}{\partial C_{j}}$ and $C_j$ are the elastic coefficients $C_{ij}$ arranged in a column vector where index $j$ defines the component location in this vector. Values of the elastic coefficients $C_{ij}$ of $\left[A_{ij}\right]$ are those provided by \cite{Bechmann}. Although these values are given at room temperature, they can also be used at cryogenic temperature with a relative accuracy better than $10^{-3}$, the fractional frequency shifts observed from room temperature to cryogenic temperature operations being typically lower than $10^{-3}$.
\begin{table} [h]
\caption{\label{tab:table3}Elastic coefficients and quartz-cuts used for their respective determination. the X, Y, and Z-cuts are cuts whose planes are normal to the X, Y, and Z cristalline axis respectively. The AT-cut is a single rotated Y-cut, i.e. a cut whose cut plane, originally normal to the Y-axis, is rotated around the X-axis. The SC-cut is a doubly-rotated cut, i.e. a AT-cut rotated around $Z'$ the new Z-axis after the first rotation. The elastic coefficient $C_{66}$ is related to $C_{11}$ and $C_{12}$ as $C_{66}=(C_{11}-C_{12})/2$ \cite{Royer}.}
\begin{ruledtabular}
\begin{tabular}{ccc}
$\textit{Elastic coefficient}$ & $\textit{Quartz cut}$ & $\textit{Comments}$\\
\hline
$C_{11}$ & X & \\
$C_{12}$ & X & Calculated from $C_{11}$ and $C_{66}$ \\
$C_{13}$ & SC & \\
$C_{14}$ & AT & \\
$C_{33}$ & SC & \\
$C_{44}$ & Z & \\
$C_{66}$ & Y & \\
\end{tabular}
\end{ruledtabular}
\end{table}
\begin{table} [h!]
\caption{\label{tab:table1}Cuts of crystal plates and modes used to measure the corresponding resonance-frequency coefficients from a $3^{rd}$ order polynomial fit, with $T_{ref}=8$K. A, B and C modes are the (quasi-)longitudinal, fast (quasi-)shear and slow (quasi-)shear thickness modes respectively.}
\begin{ruledtabular}
\begin{tabular}{c p{1.8cm} ccc}
$\textit{Quartz cut}$ & \centering{$\textit{Mode}$} & $T^{(1)}f, 10^{-7}$ & $T^{(2)}f, 10^{-8}$ & $T^{(3)}f, 10^{-9}$ \\
\hline
AT & \centering{$C_{300}$, $C_{500}$} & 0.769 & 0.748 & 0.105 \\
SC & \centering{$A_{300}$, $A_{500}$} & -1.178 & -2.189 & -2.474 \\
SC & \centering{$B_{300}$, $B_{302}$, $B_{320}$, $B_{500}$} & -0.879 & -1.672 & -1.832 \\
X & \centering{$A_{100}$, $A_{1300}$, $A_{1500}$} & -0.143 & -0.451 & -0.581 \\
Y & \centering{$C_{300}$, $C_{500}$} & 1.339 & 1.869 & 1.289 \\
Z & \centering{$C_{300}$, $C_{500}$} & -1.465 & -2.662 & -3.571 \\
\end{tabular}
\end{ruledtabular}
\end{table}
\section{Experiment}
The temperature coefficients of the elastic tensor of the quartz material at liquid helium temperature have been calculated from the set of experimental frequency-temperature characteristics of various cut angles. Actually the resulting frequency coefficients in Tab.~\ref{tab:table1}, used to determine these temperature coefficients, are averaged values from measurements made with several overtone numbers of each tested cut.
The general overview of the measurement procedure can be summed up by the following steps:
$1/$ Resonators are realized in a set of plates of various cut angles according to the conventional process: plate orientation with respect to the quartz crystal axis, plate cutting, rounding, lapping and polishing to obtain a 13 mm diameter plano-convex device, gold-electrode coating for piezo-electrically exciting/probing the acoustic cavity. The thickness of these resonators is such that the $3^{rd}$ overtone of the C-mode is close to 10 MHz at room temperature.
Cuts used to take into account the quartz anisotropy comply with the IEEE-std 176-1978 \cite{IEEE}: AT ($0^\circ, -35.3^\circ$), SC ($22^\circ, -34^\circ$), X ($30^\circ, 0^\circ$), Y ($0^\circ, 0^\circ$), Z ($0^\circ, 90^\circ$). $2/$ To be tested, the resonators are embedded inside an oxygen-free cooper block screwed on the cold stage of a pulse-tube cryorefrigerator (a Sumitomo cryocooler SRP082B). This block is cooled down in the vicinity of 4 K and temperature controlled within $\pm 3$ mK around the set temperature value when the steady state is reached. $3/$ For a given temperature, the resonator admittance $Y$ is measured around the interesting resonance frequencies, and over a frequency span properly chosen, by means of a network analyzer. However a calibration procedure should be followed before recording, to cancel the influence of the coaxial cable feeding the resonator inside the vacuum chamber from the network analyzer: this calibration involves three identical coaxial cables respectively ended by a standard 50 ohms resistor, an open circuit and a short circuit. The network analyzer (HP4195A) is locked to a hydrogen maser to get the frequency-stability performances of the latter. The analyzer provides a $0.001$ dB and $0.01^\circ$ resolution in magnitude and phase respectively, and its frequency resolution reaches $1$ mHz. In addition, amplitude-to-frequency nonlinear effects are avoided by limiting the excitation power. $4/$ The post-processing consists in plotting the admittance circle $Y=G+jB$ from the experimental data, magnitude and phase of $Y$, to get the resonance frequency with the best accuracy. Fig.~\ref{fig:adm_circle_plot} shows an example of such a G-B plot.
\begin{figure}[t!]
\centering
\includegraphics[width=3.25in]{G-B_plot_X_600micron_B1_4_5K_adm_circle}
\captionof{figure}{\rm{a)} G-B plot from the network analyzer for the $1^{st}$ OT of a X-cut at 4.5K. $\Delta f$ is with respect to $f_{ref} = 4.580957 \rm{Hz}$. \rm{b)} Admittance circle from the experimental data above.}
\label{fig:adm_circle_plot}
\end{figure}
\section{Results}
As an example, Fig.~\ref{fig:F_T_curves} shows experimental results of the fractional frequency deviation versus temperature for a X-cut operating around $T_{ref}=8K$. The geometrical properties of this device are $h = 600 \mu m$ and $R_0 = 250$ mm. It can be observed that all the A-mode overtones exhibit similar behaviors.The temperature coefficients of the frequency are extracted from such experimental data (see Tab.~\ref{tab:table1}).
As a final result, Tab.~\ref{tab:table2} gives the thermal coefficients $T^{(k)}C_{ij}$ of the elastic coefficients $C_{ij}$ of quartz, up to the $3^{rd}$ order, determined according to the Eulerian description.
\begin{figure}[t!]
\centering
\includegraphics[width=3.25in]{Coupe_X_A1_A13_A15_points}
\captionof{figure}{Resonance fractional frequency shift $\frac{f(T)-f(T_{ref})}{f(T_{ref})}$ for a X-cut as a function of temperature offset $\delta T = T-T_{ref}$, where $T_{ref} = 8 {\rm K}$.}
\label{fig:F_T_curves}
\end{figure}
\begin{table} [h!]
\caption{\label{tab:table2}$1^{st}$, $2^{nd}$, and $3^{rd}$ order temperature coefficients of the elastic constants of crystalline quartz within 4 {\rm K} - 15 {\rm K}, with $T_{ref} = 8 {\rm K}$.}
\begin{ruledtabular}
\begin{tabular}{ccccc}
$C_{ij}$ & $T^{(1)}C_{ijkl}, 10^{-7}$ & $T^{(2)}C_{ijkl} 10^{-8}$ & $T^{(3)}C_{ijkl}, 10^{-9}$\\
\hline
$C_{11}$ & $-0.235$ & $1.28$ & $-3.25$ \\
$C_{12}$\footnotemark[1] & $-6.39$ & $-10.56$ & $-4.22$ \\
$C_{13}$ & $-7.18$ & $-7.42$ & $-0.877$ \\
$C_{14}$ & $-1.35$ & $-1.31$ & $-7.93$ \\
$C_{33}$ & $-4.62$ & $-5.65$ & $-7.62$ \\
$C_{44}$ & $-3.17$ & $-4.37$ & $-11.6$ \\
$C_{66}$ & $3.08$ & $5.92$ & $0.486$ \\
\end{tabular}
\end{ruledtabular}
\footnotetext[1]{Calculated coefficient}
\end{table}
\section{Conclusion}
Various cut-angles of synthetic quartz have been measured at cryogenic temperature in order to determine the temperature coefficients of its elastic coefficients. The Eulerian formalism have been used to provide these $T^{(k)}C_{ij}$ up to the $3^{rd}$ order. Uncertainties on the elastic coefficient values are not specified because most of the uncertainties linked to the many steps of measurements and calculation are difficult to bring out. Actually, in addition to the cut angle uncertainty, one of the higher uncertainties is the result of the Tiersten model which is a theoretical approach modelling plano-convex resonators. However, it can be stated that the uncertainty on the temperature sensitivity of the elastic coefficient $C_{13}$ is higher than that of other coefficients because of its very slight involvement in calculations.
\section*{Acknowledgements}
Thanks to Philippe Abb\'{e} and Xavier Vacheret for their help for making resonators and experiments. This work is supported by Conseil R\'{e}gional de Franche-Comt\'{e}. J.B. is thankful to Minist\`{e}re de l'Enseignement Sup\'{e}rieur et de la Recherche, France, for his grant.
| 2f1d086b770e05b526c08c4cb4594f8a6224d5e6 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
\label{intro}
Deep neural networks (DNNs) have been successful in modeling many aspects of
natural language including word meanings \cite{mikolov}\cite{le},
machine translation \cite{sutskever},
syntactic parsing \cite{vinyals_grammar}, language modeling \cite{jozefowicz} and
image captioning \cite{vinyals_caption}.
Given sufficient training data, DNNs are
highly accurate and can be trained end-to-end without the need for
intermediate knowledge representations or explicit feature extraction.
With recent interest in conversational user interfaces such as virtual
assistants and chatbots, the application of DNNs to facilitate
meaningful conversations is an area where more progress is needed.
While sequence to sequence models based on recurrent
neural networks (RNNs) have shown initial promise in creating intelligible
conversations \cite{vinyals_conversation}, it has been noted that more work is needed
for these models to fully capture larger aspects of human communication including
conversational goals, personas, consistency, context, and word knowledge.
Since discourse analysis considers language at
the conversation-level, including its social and psychological context,
it is a useful framework for guiding the extension of end-to-end neural conversational models.
Drawing on concepts from discourse analysis such as \textit{coherence}
and \textit{cohesion} \cite{halliday},
we can codify what makes conversations more intelligent
in order to design more powerful neural models that reach beyond the
sentence and utterance level.
For example, by looking for features that indicate deixis, anaphora,
and logical consequence in the machine-generated utterances we can benchmark the level of coherence and
cohesion with the rest of the conversation, and then make improvements to
models accordingly.
In the long run, if neural models can encode the long-range structure
of conversations, they may be able to express conversational discourse similar to the way the human brain does, without the need for
explicitly building formal representations of discourse theory into the model.
To that end, we explore RNN-based sequence to
sequence architectures that can
capture long-range relationships between multiple utterances in conversations
and look at their ability to exhibit discourse relationships.
Specifically, we look at 1) a baseline RNN encoder-decoder with attention mechanism and 2) a model with an
additional discourse RNN that encodes a sequence of multiple utterances.
Our contributions are as follows:
\begin{itemize}
\item We examine two RNN models with attention mechanisms to model
discourse relationships across different utterances
that differ somewhat compared to what has been done before
\item We carefully construct controlled experiments
to study the relative merits of different models on multi-turn conversations
\item We perform a sensitivity analysis on how the amount of context
provided by previous utterances affects model performance
\item We quantify how neural conversational models display coherence by measuring the prevalence of specific
syntactical features indicative of deixis, anaphora, and logical consequence.
\end{itemize}
\section{Related Work}
\label{relatedwork}
Building on work done in machine translation, sequence to sequence models based on RNN encoder-decoders were
initially applied to generate conversational outputs given a single previous
message utterance as input\cite{shang}\cite{vinyals_conversation}.
In \cite{sordoni} several models were presented that included a
``context'' vector (for example representing another previous utterance)
that was combined with the message utterance via various encoding
strategies to initialize or bias a single decoder RNN.
Some models have also included an additional RNN tier to capture the
context of conversations. For example,
\cite{serban} includes a hierarchical ``context RNN'' layer to summarize
the state of a dialog, while
\cite{yao} includes an RNN ``intension network'' to model conversation
intension for dialogs involving two participants speaking in turn.
Modeling the ``persona'' of the participants in a conversation by
embedding each speaker into a $K$-dimensional embedding was shown
to increase the consistency of conversations in \cite{li_persona}.
Formal representations such as Rhetorical Structure Theory (RST) \cite{mann}
have been developed to identify discourse structures in written text.
Discourse parsing of cue phrases \cite{marcu} and coherence modeling
based on co-reference resolution of named-entities \cite{barzilay}\cite{kibble} have been applied to tasks such as
summarization and text generation. Lexical chains \cite{morris} and
narrative event chains \cite{chambers} provide directed graph models of text
coherence by looking at thesaurus relationships and subject-verb-temporal
relationships, respectively.
Recurrent convolutional neural networks have been used to classify
utterances into discourse speech-act labels \cite{kalchbrenner} and
hierarchical LSTM models have been evaluated for generating coherent
paragraphs in text documents \cite{li} .
Our aim is to develop end-to-end neural conversational models that exhibit
awareness of discourse without needing a formal representation of
discourse relationships.
\subsection{Models}
\label{models}
Since conversations are sequences of utterances and utterances are sequences of words, it is natural to use models based on an RNN encoder-decoder
to predict the next utterance in the conversation given $N$
previous utterances as source input.
We compare two types of models: \textbf{seq2seq+A}, which applies an attention mechanism
directly to the encoder hidden states, and \textbf{Nseq2seq+A}, which
adds an additional RNN tier with its own attention mechanism to model discourse relationships between $N$ input utterances.
In both cases the RNN decoder predicts the output utterance and
the RNN encoder reads the sequence of words in each input utterance.
The encoder and decoder each have their own vocabulary
embeddings.
As in \cite{vinyals_grammar} we compute the attention vector at each
decoder output time step $t$
given an input sequence $(1,...,T_{A})$ using:
\begin{eqnarray}
u_{i}^{t} & = & v^{T}tanh(W_{1} h_{i} + W_{2} d^{t}) \nonumber \\
a_{i}^{t} & = & softmax(u_{i}^{t}) \nonumber \\
c^{t} & = & \sum_{i=1}^{T_{A}} a_{i}^{t} h_{i} \nonumber
\end{eqnarray}
where the vector $v$ and matrices $W_{1}$, and $W_{2}$ are learned parameters.
$d^{t}$ is the decoder state at time $t$ and is concatenated with $c^{t}$ to
make predictions and inform the next time step. In \textbf{seq2seq+A} the $h_{i}$ are the
hidden states of the encoder $e_{i}$, and for \textbf{Nseq2seq+A} they are the
$N$ hidden states of the discourse RNN (see Fig. \ref{schematic}.)
Therefore, in
\textbf{seq2seq+A} the attention mechanism is applied at the
word-level, while in \textbf{Nseq2seq+A} attention is applied at the utterance-level.
\begin{figure*}[t]
\centerline{\includegraphics[width=12cm]{Figure1.pdf}}
\caption{Schematic of \textbf{seq2seq+A} and \textbf{Nseq2seq+A}
models for multiple turns of conversation. An attention mechanism is
applied either directly to the encoder RNN or to an intermediate
discourse RNN.}
\label{schematic}
\end{figure*}
\subsection{seq2seq+A}
\label{seq2seq}
As a baseline starting point we use an attention mechanism to help model
the discourse by a straightforward adaptation of
the RNN encoder-decoder conversational model discussed in \cite{vinyals_conversation}.
We join multiple source utterances using the \textit{EOS} symbol as a delimiter, and feed them into the
encoder RNN as a single input sequence.
As in \cite{sutskever}, we reversed the order of the tokens
in each of the individual utterances but preserved the order of
the conversation turns.
The attention mechanism is able to make connections to any of the words used in earlier
utterances as the decoder generates each word in the output response.
\subsection{Nseq2seq+A}
\label{Nseq2seq}
Since conversational threads are ordered sequences of utterances,
it makes sense to extend an RNN encoder-decoder by adding another
RNN tier to model the discourse as the turns of the conversation progress.
Given $N$ input utterances, the RNN encoder is applied to each
utterance one at a time as shown in Fig. \ref{schematic} (with tokens
fed in reverse order.)
The output of the encoder from each of the input utterances forms
$N$ time step inputs for the discourse RNN.
The attention mechanism is then applied to the $N$ hidden states of the
discourse RNN and fed into the decoder RNN.
We also considered a model where the output of the encoder is
also combined with the output of the discourse RNN and fed into the
attention decoder,
but found the purely hierarchical architecture performed better.
\subsection{Learning}
\label{learning}
For each model we chose identical optimizers, hyperparameters, etc. in our experiments
in order to isolate the impact of specific differences in the network
architecture, also taking computation times and available GPU resources into account.
It would be straightforward to perform
a grid search to tune hyperparameters, try LSTM cells, increase layers
per RNN, etc. to further improve
performance individually for each model beyond what we report here.
For each RNN we use one layer of Gated Recurrent Units (GRUs) with
512 hidden cells.
Separate embeddings for the encoder and decoder, each with dimension
512 and vocabulary size of 40,000, are trained on-the-fly without using predefined
word vectors.
We use a stochastic gradient descent (SGD) optimizer with $L2$ norms
clipped at $5.0$, an initial learning rate of $0.5$, and a learning rate decay
factor of $0.99$ is applied when needed. We trained with mini-batches of 64
randomly selected examples, and ran training for approximately 10
epochs until validation set loss converged.
\section{Experiments}
\label{experiments}
We first present results comparing
our neural discourse models trained on a large set of conversation
threads based on the OpenSubtitles dataset \cite{tiedemann}.
We then examine how our models are able to produce
outputs that indicate enhanced coherence by searching for discourse markers.
\subsection{OpenSubtitles dataset}
\label{opensubtitles}
A large-scale dataset is important if we want to model all the
variations and nuances of human language.
From the OpenSubtitles corpus we created a training set and validation
set with 3,642,856 and 911,128 conversation fragments,
respectively\footnote{The training and validation
sets consisted of $320M$ and $80M$ tokens, respectively}.
Each conversation fragment consists of 10 utterances from
the previous lines of the movie dialog leading up to a target utterance.
The main limitation of the OpenSubtitles dataset is that it is derived
from closed caption style subtitles, which can be noisy, do not include labels for
which actors are speaking in turn, and do not show conversation
boundaries from different scenes.
We considered cleaner datasets such as the Ubuntu dialog corpus
\cite{lowe}, Movie-DiC dialog corpus \cite{banchs}, and SubTle corpus
\cite{ameixa} but found they all contained orders of magnitude fewer
conversations and/or many fewer turns per conversation on average.
Therefore, we found the size of the OpenSubtitles dataset outweighed the benefits of
cleaner smaller datasets. This echoes a trend in neural networks
where large noisy datasets tend to perform better than small clean datasets.
The lack of a large-scale clean dataset of conversations is an open
problem in the field.
\subsection{Results}
\label{results}
We compared models and performed a sensitivity analysis by varying the
number of previous conversation turns fed into the encoder during
training and evaluation.
\begin{table*}[!htbp]
\caption{Results on OpenSubtitles dataset. Perplexity vs. number of
previous conversation turns.}
\label{perplexity}
\begin{center}
\begin{tabular}{cll}
\hline
\textbf{Previous conversation turns} & \textbf{seq2seq+A} & \textbf{Nseq2seq+A} \\
\hline
$N=1$ & 13.84$\pm$0.02 & 13.71$\pm$0.03 \\
$N=2$ & 13.49$\pm$0.03 & 13.40$\pm$0.04 \\
$N=3$ & 13.44$\pm$0.05 & 13.31$\pm$0.03 \\
$N=5$ & - & 13.14$\pm$0.03 \\
$N=7$ & - & 13.08$\pm$0.03 \\
\hline
\end{tabular}
\end{center}
\end{table*}
\begin{figure*}[t]
\centerline{\includegraphics[width=11cm]{Figure2.pdf}}
\caption{Sensitivity analysis of perplexity vs. number of previous conversations turns.}
\label{sensitivity}
\end{figure*}
In Table \ref{perplexity} we
report the average perplexity\footnote{We use perplexity as our performance metric, because it is simple to
compute and correlates with human judgements, though it has well-known limitations.} on the validation set at convergence for each model.
For $N=1,2,3$ we found that \textbf{Nseq2seq+A}
shows a modest but significant performance improvement over the baseline
\textbf{seq2seq+A}. We only ran \textbf{Nseq2seq+A} on larger values
of $N$, assuming it would continue to outperform.
In Fig. \ref{sensitivity} we show that increasing the amount
of context from previous conversation turns significantly improves
model performance, though there appear to be diminishing returns.
\subsection{Discourse analysis}
\label{discourse}
Since a large enough dataset tagged with crisp discourse
relationships is not currently available, we seek a
way to quantitatively compare relative levels of coherence and cohesion.
As an alternative to a human-rated evaluation we performed
simple text analysis to search for specific discourse markers \cite{fraser}
that indicate enhanced coherence in the decoder output as follows:
\begin{itemize}
\item \textbf{Deixis:} contains words or phrases\footnote{here, there, then, now, later, this, that} referring
to previous context of place or time
\item \textbf{Anaphora:} contains pronouns\footnote{she,
her, hers, he, him, his, they, them, their, theirs}
referring to entities mentioned in previous utterances
\item \textbf{Logical consequence:} starts with a cue
phrase\footnote{so, after all, in addition, furthermore, therefore,
thus, also, but, however, otherwise, although, if, then}
forming logical relations to previous utterances
\end{itemize}
\begin{table*}[!htbp]
\caption{Discourse analysis of \textbf{Nseq2seq+A} decoder
output. Likelihood of discourse markers vs. number of previous
conversation turns used as input.}
\label{cohesion}
\begin{center}
\begin{tabular}{llllll}
\hline
& $N=1$ & $N=2$ & $N=3$ & $N=5$ & $N=7$ \\
\hline
deixis & 4.0\% & 3.4\% & 16.3\% & 5.1\% & 5.0\% \\
anaphora & 4.4\% & 6.1\% & 9.9\% & 7.2\% & 9.3\% \\
logical consequence & 0.03\% & 0.05\% & 0.08\% & 0.34\% & 0.12\% \\
\hline
\end{tabular}
\end{center}
\end{table*}
In Table \ref{cohesion} we show how $N$, the number of
previous conversation turns used as input,
affects the likelihood that these discourse
markers appear in the decoder output. The percentage of output
utterances containing discourse markers related to deixis, anaphora, and logical
consequence are reported from a sample of 100,000 validation set
examples.
In general, we find that more context leads to a higher likelihood of
discourse markers indicating that long-range discourse relationships are indeed
being modeled.
The results show a potentially interesting sensitivity to the value of $N$, require further study, and are
likely to be dependent on different conversational styles and domains.
\subsection{Examples}
\label{examples}
In Table \ref{cohesion} we show a few examples comparing decoder outputs
of the \textbf{Nseq2seq+A} model using either 1 or 5 previous
conversation turns as input. Qualitatively we can see that this
neural discourse model is capable of producing increased cohesion when provided with
more context.
\begin{table*}[!htbp]
\caption{Example responses comparing \textbf{Nseq2seq+A} with either 1
or 5 previous conversation turns as input. Output words exhibiting cohesion with
previous conversation turns are indicated in bold.}
\label{responses}
\begin{center}
\begin{tabular}{|l|l|}
\hline
\textbf{Previous Lines of Conversation Input} & \textbf{Decoder Output} \\
\hline
\hline
can ' t see a number plate , even . & \textit{i ' m sorry .} \\
\hline
then there ' s this . & \\
five streets away from UNK ' s flat , it ' s the night we broke in . & \\
it ' s him , it ' s the man we disturbed . & \\
we can ' t identify him from this . & \\
can ' t see a number plate , even . & \textit{we ' re not going to get rid of
\textbf{him} .} \\
\hline
\hline
but he was such a puppy ! & \textit{i was so scared .} \\
\hline
how time flies !& \\
it only seems like yesterday since i put 00 candles on his birthday
cake . & \\
he was such a tall boy !& \\
nearly six feet . & \\
but he was such a puppy ! & \textit{he was a \textbf{young man} !} \\
\hline
\hline
now just two months ago right here & \textit{and the other ?} \\
\hline
would you swear to that ?& \\
yes , sir , we swear to that . & \\
there you are , an UNK alibi .& \\
serena robbins , you come down here ! & \\
now just two months ago right here & \textit{\textbf{we ' re} on the \textbf{way down} .} \\
\hline
\end{tabular}
\end{center}
\end{table*}
\section{Conclusions}
\label{conclusions}
We studied neural discourse models that can capture long distance
relationships between features found in different utterances of a conversation.
We found that a model with an additional discourse RNN outperforms the
baseline RNN encoder-decoder with an attention mechanism.
Our results indicate that providing more context from previous
utterances improves model performance up to a point.
Qualitative examples illustrate how the discourse RNN produces
increased coherence and cohesion with the rest of the conversation,
while quantitative results based on text mining of discourse markers
show that the amount of deixis, anaphora, and logical consequence
found in the decoder output can be sensitive to the size of the
context window.
In future work, it will be interesting to train discourse models on even
larger corpora and compare conversations in different domains.
By examining the attention weights it should be possible to study what
discourse markers the models are ``paying attention to'' and possibly
provide a powerful new tool for analyzing discourse relationships.
By applying multi-task sequence to sequence learning techniques as in \cite{luong}
we may be able to combine the conversational modeling
task with other tasks such as discourse parsing and/or world knowledge modeling achieve better
overall model performance. Not just for conversations, neural discourse modeling could also be applied to
written text documents in domains with strong patterns of discourse such as news, legal, healthcare.
| 80092f558d83c636b8f26890ba6e55a07c2c4ca1 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
\label{section.introduction_chap_fb}
The Fermi bubbles are important Galactic structures that were recently discovered by the \textit{Fermi Gamma-ray Space Telescope} \citep{su_etal10}. The bubbles are two diffuse lobes of material extending $\sim$50$\arcdeg$ above and below the Galactic plane ($\approx$10 kpc at the Galactic center). Their surface brightness shows little variation on the sky, their gamma-ray spectrum follows a power law with $dN/dE \propto E^{-2}$ between $\approx$1 and 200 GeV, and they have a counterpart in microwaves, known as the $Wilkinson\ Microwave\ Anisotropy\ Probe$ ($WMAP$) haze \citep{dobler_finkbeiner08, dobler_etal10, su_etal10, ackermann_etal14}. It is still unclear what produces the gamma rays, but all plausible mechanisms imply that energetic cosmic-ray particles exist within the bubbles. This inference combined with the bubbles' size and location on the sky suggests that they are affiliated with a massive energy injection event near the Galactic center.
The bubbles' morphology is similar to wind-blown bubbles observed in other galaxies, indicating that they formed from either a period of enhanced nuclear star formation or a Sgr A* outburst event (see \citealt{veilleux_etal05} for a review). Star formation can drive outflows through a combination of stellar winds from young stars and multiple type-II supernova explosions \citep[e.g., ][]{leitherer_etal99}, while black hole accretion episodes can produce energetic jets or winds that inflate a cavity with thermal and non-thermal particles \citep[e.g., ][]{mcnamara_nulsen07,yuan_narayan14}. Both of these scenarios are critical events in galaxy evolution, as they both can deposit significant amounts of energy into the rest of the galaxy on scales $\gtrsim$10 kpc (see \citet{mcnamara_nulsen07} for a review). However, the details of these ``feedback'' effects (mass displacement, energy transport, etc.) are poorly understood since we observe them in external galaxies. The Fermi bubbles are a unique laboratory for understanding these processes since we can spatially resolve the bubbles across multiple wavebands.
A popular strategy to probe these effects and bubbles' origins has been the use of magnetohydrodynamic (MHD) simulations to reproduce the bubbles’ global morphology and non-thermal properties. Simulations produce cosmic rays either from a black hole accretion event \citep{zubovas_etal11,guo_mathews12a,guo_mathews12b,yang_etal12,yang_etal13,zubovas_nayakshin12,mou_etal14,mou_etal15}, from nuclear star formation activity \citep{crocker12,crocker_etal14,crocker_etal15,sarkar_etal15,ruszkowski_etal16}, or in-situ as the bubbles evolve \citep{cheng_etal11,cheng_etal15b,mertsch_sarkar11,fujita_etal14,lacki14,sasaki_etal15}, and compare the non-thermal emission to the bubbles' gamma-ray emission. All of these origin scenarios can reproduce the bubbles' morphology, but they imply significantly different input energetics and timescales required to inflate the bubbles ($\dot{E} \gtrsim 10^{41}$ erg s$^{-1}$, $t \lesssim 5$ Myr for black hole accretion compared to $\dot{E} \lesssim 5 \times 10^{40}$ erg s$^{-1}$, $t \gtrsim 50$ Myr for star formation). This variation in the feedback rate is a significant uncertainty in how the bubbles impact the Galaxy, but there are additional factors that can constrain the characteristic bubble energetics.
Constraining the bubbles' thermal gas distribution is a promising avenue to solve this problem, since the characteristic densities and temperatures should be significantly different depending on the bubble energetics. In the framework of expanding galactic outflows and shocks \citep[e.g., ][]{veilleux_etal05}, a higher energy input rate leads to a higher plasma temperature and a larger expansion rate for a fixed bubble size and ambient density. Thus, the plasma temperature at the interface between the bubbles and surrounding medium encodes information on the bubbles' shock strength, expansion properties, and overall energy input rate. A generic prediction from simulations and observations of galactic outflows is that the bubbles are overpressurized and hotter than the surrounding medium ($\gtrsim 2 \times 10^6$ K), implying that the bubbles' thermal gas should have signatures at soft X-ray energies. Indeed, the bubbles appear to be bounded by X-ray emission seen in the \textit{ROSAT} 1.5 keV band \citep{bh_cohen03, su_etal10}; however, these observations do not constrain the bubbles' intrinsic thermal gas structure since the broad-band images are a weak temperature diagnostic. Spectral observations with current X-ray telescopes are a much better temperature diagnostic for this type of environment.
Initial efforts to observe the bubbles in soft X-rays with \textit{Suzaku} and \textit{Swift} and constrain their temperature and shock strength were carried out by several groups \citep{kataoka_etal13,kataoka_etal15,tahara_etal15}. \citet{kataoka_etal15} extracted soft X-ray background (SXRB) spectra in the 0.5--2.0 keV band for 97 sight lines that pass through the Fermi bubbles, and fit the spectra with thermal plasma models. They consistently measured plasma temperatures of $kT$ = 0.3 keV for these sight lines, which is systematically higher than the characteristic temperature measured in sight lines away from the Galactic center \citep[$kT \approx$0.2 keV; ][]{hs13}. From this temperature ratio, they inferred a shock Mach number of $\mathcal{M} \approx $0.3 keV / 0.2 keV = 1.5, and corresponding expansion rate of $\approx$300 km s$^{-1}$. This is a valuable attempt to constrain these quantities, but the analysis assumes that the Fermi bubble plasma dominates the hotter spectral component. In practice, there are other known emission sources that contribute to the SXRB spectrum, and accounting for this emission can change the inferred thermal gas temperature.
The Milky Way hosts a hot gas distribution with $T \approx 2 \times 10^6$ K extending on scales $\gtrsim$10 kpc based on shadowing experiments from \textit{ROSAT} all-sky data \citep{snowden_etal97, kuntz_snowden00}. This plasma is believed to dominate any SXRB spectrum, with \ion{O}{7}\ and \ion{O}{8}\ being the characteristic observed line transitions \citep[e.g., ][]{mccammon_etal02, yoshino_etal09, hs12}. The structure of this extended plasma distribution has been debated in the literature, but numerous studies on but numerous studies on both absorption and emission line strengths indicate that the plasma is spherical and extends to at least $r \sim 50$ kpc \citep{fang_etal06, bregman_ld07, gupta_etal12, fang_etal13, miller_bregman13, miller_bregman15}. In particular, Miller \& Bregman (\citeyear{miller_bregman15}, defined as \citetalias{miller_bregman15} henceforth) modeled a set of 648 \ion{O}{8}\ emission line intensities from Henley \& Shelton (\citeyear{hs12}, defined as \citetalias{hs12} henceforth), and found that a hot gas density profile with $n \propto r^{-3/2}$ extending to the virial radius reproduces the observed emission line intensities. These modeling studies have placed useful constraints on the Galactic-scale hot gas distribution, but also highlight the fact that this extended plasma is likely the dominant emission source in all 0.5--2.0 keV band spectra.
In this study, we expand the analysis Kataoka et al. (\citeyear{kataoka_etal15}, defined as \citetalias{kataoka_etal15} henceforth) analysis by modeling the combined emission from the Fermi bubbles and hot gas halo present in \ion{O}{8}\ emission line measurements. We modify the Galactic-scale hot gas models from \citetalias{miller_bregman15} to include a geometry, density, and temperature structure for the Fermi bubbles. Given a set of model parameters, we predict the contribution to the emission made by the Fermi bubbles and hot gas halo along any sight line. This results in a more careful comparison between the Fermi bubbles' emission and the total observed emission in any SXRB measurement.
The \ion{O}{8}\ observations used in our analysis consist of published \textit{XMM-Newton} measurements from \citetalias{hs12}, and a new \textit{Suzaku} data set produced for this work. The \textit{XMM-Newton} data are mostly the same measurements used in \citetalias{miller_bregman15}, but we now include data near the Fermi bubbles. We supplement these data with archival \textit{Suzaku} measurements of SXRB spectra, which more than doubles the number of emission line measurements projected near the bubbles. These data are processed in a similar way to the \textit{XMM-Newton} data reduction outlined in \citetalias{hs12}, resulting in a uniformly processed data set of emission line intensities from the SXRB.
Following the methodology from \citetalias{miller_bregman15}, we constrain the Fermi bubbles' density and temperature structure by finding the parametric model that is most consistent with the observed emission line intensities. We measure the characteristic bubble temperature from analyzing the distribution of observed \ion{O}{8}\ /\ion{O}{7}\ line ratios near the bubbles, and the characteristic bubble density from explicitly modeling the \ion{O}{8}\ emission line intensities. We infer a similar bubble shock strength compared to the \citetalias{kataoka_etal15} analysis, and discuss the systematic differences between our approaches and results and theirs. We also estimate the bubbles' age and energy input rate, and these with the possible formation mechanisms discussed above.
The rest of the paper is outlined as follows. Section~\ref{section.data_chap_fb} discusses how we compiled our emission line sample, including an overview of the \textit{XMM-Newton} data set and the \textit{Suzaku} data processing. Section~\ref{section.model_chap_fb} definesdefines our parametric density and temperature model and discusses our line intensity calculation. Section~\ref{section.results_chap_fb} discusses our model fitting routine and results. Section~\ref{section.discussion_chap_fb} discusses our constraints on the Fermi bubbles in the context of galactic outflows, previous X-ray studies, and simulations. Section~\ref{section.conclusions_chap_fb} summarizes our results.
\section{Emission Line Data}
\label{section.data_chap_fb}
Our data set includes \ion{O}{7}\ (He-like triplet at $E \approx 0.56$ keV) and \ion{O}{8}\ (Ly$\alpha$ transition at $E \approx 0.65$ keV) emission lines, which are the dominant ions for thermal plasmas with temperatures between $T \sim 10^{5.5}$ and $10^{7}$ K \citep{sutherland_dopita93}. For an optically thin plasma in collisional ionization equilibrium, the emission line intensity depends on the plasma density and temperature as $I\ \propto\ n^2 \epsilon(T)$, where $\epsilon(T)$ is the volumetric line emissivity. This implies that the line strength ratio is a temperature diagnostic and the individual ion line strengths can be used to estimate the plasma density. Large, all-sky samples of emission line measurements in particular have been instrumental in constraining the Milky Way's global density distribution of $\sim 10^6$ K gas \citepalias{miller_bregman15}.
The full data set used in our modeling analysis is a combination of published \textit{XMM-Newton} emission line measurements from \citetalias{hs12} and a complementary sample of \textit{Suzaku} measurements compiled specifically for this project. The \textit{XMM-Newton} sample from \citetalias{hs12} contains $\sim$1000 emission line measurements distributed across the sky, making it a valuable starting point for our modeling work. While \textit{XMM-Newton} has more collecting power than \textit{Suzaku} near the emission lines (collecting area $\times$ field of view $\approx$140,000 cm$^{2}$ arcmin$^{2}$ for the MOS1 camera compared to $\approx$70,000 cm$^{2}$ arcmin$^{2}$ for the XIS1 detector), we include a supplemental \textit{Suzaku} data set for two reasons. \textit{Suzaku's} low Earth orbit results in a lower and more stable particle background than \textit{XMM-Newton's}, often resulting in measurements with higher signal-to-noise ratio at soft X-ray energies. Also, there are many valuable archival \textit{Suzaku} observations projected near the Fermi bubbles, including 14 observations dedicated to observing the bubbles' edge. In practice, \textit{Suzaku} data should have a comparable signal-to-noise ratio to the \textit{XMM-Newton} data, while probing the crucial region in and near the Fermi bubbles.
Our goal is to create a clean sample of uniformly processed emission line measurements by reducing the \textit{Suzaku} data in a similar way to how \citetalias{hs12} reduced the \textit{XMM-Newton} data. The main steps include: the removal of bright point sources, light curve filtering, spectral fitting, and filtering for solar wind charge-exchange (SWCX) emission. The following sections summarize how the \textit{XMM-Newton} data were produced, and detail our \textit{Suzaku} data reduction steps. After applying all these data reduction methods, our final sample includes 683 useful \textit{XMM-Newton} measurements and 58 useful \textit{Suzaku} measurements.
\subsection{\textit{XMM-Newton} Observations}
\label{subsection.xmm_chap_fb}
We summarize the \textit{XMM-Newton} emission line sample compilation here, but we refer the reader to \citetalias{hs12} for a more detailed description of their data reduction methods. Their initial sample included 5698 observations that had any EPIC-MOS exposure time. They reduced the data using the \textit{XMM-Newton} Extended Source Analysis Software \footnote{\url{http://heasarc.gsfc.nasa.gov/docs/xmm/xmmhp_xmmesas.html}} \citep[\textit{XMM}-ESAS;][]{kuntz_snowden08, snowden_kuntz11}, which includes screening the 2.5--12 keV band count rate for soft proton flares. They also removed point sources from the spectral extraction regions using data from the Second \textit{XMM-Newton} Serendipitous Source Catalog \footnote{\url{http://xmmssc-www.star.le.ac.uk/Catalogue/2XMMi-DR3/}} \citep{watson_etal09}, as well as visual inspection for bright sources in the images. The authors attempted to reduce geocoronal SWCX emission by excluding observing periods with high solar wind proton flux measurements (see Section~\ref{subsection.swcx_chap_fb} for the details of this procedure), which they called their ``flux-filtered'' sample. These reduction methods resulted in 1868 total observations and 1003 flux-filtered observations with $\geq$5 ks of good observing time. Each of these observations includes an \ion{O}{7}\ and \ion{O}{8}\ emission line intensity measurement, and they have been used to analyze the known emission sources (i.e., Local Bubble (LB), extended hot halo, SWCX).
This sample has been used before to successfully model the Milky Way's global hot gas structure, thus motivating its use to model the Fermi bubbles. \citetalias{miller_bregman15} compiled a subset of the \citetalias{hs12} flux-filtered sample with additional spatial screening criteria to reduce any residual emission from sources other than the LB and Galactic hot halo. To achieve this, they removed observations within 0.5$\arcdeg$ of potential bright X-ray sources (see Table 1 in \citetalias{miller_bregman15} for the types of sources) or within 10$\arcdeg$ of the Galactic plane, and sight lines near the Fermi bubbles ($|l| \leq\ 22\arcdeg$, $|b| \leq\ 55\arcdeg$). This resulted in a subsample of 649 observations from the \citetalias{hs12} flux-filtered sample.
The \textit{XMM-Newton} data used in this study are the same as those used in \citetalias{miller_bregman15}, but including the sight lines near the Fermi bubbles. We start with the \citetalias{hs12} flux-filtered sample and still exclude sight lines near bright X-ray sources or within 10$\arcdeg$ of the Galactic plane. These screening criteria result in a total of 683 \textit{XMM-Newton} measurements distributed across the sky, with 34 measurements passing near or through the bubbles' gamma-ray edge.
\subsection{\textit{Suzaku} Observations and Data Reduction}
\label{subsection.suz_chap_fb}
\begin{figure*}
\begin{center}
\includegraphics[width = 1.0\textwidth, keepaspectratio=true]{lc_100011010_401051010.pdf}
\caption[Example light curve and count rate histogram]{Example 0.4--10.0 keV light curves (left panels) and count-rate histograms (right panels) with fitted Gaussian distributions for two observations in our initial sample. Observing periods within the 2$\sigma$ limits (red dashed lines) were kept while periods outside of these limits were excluded (gray points in the left panels). The top panels show a Galactic bulge observation (Obs. ID 100011010) with a well-behaved, Gaussian light curve that we retained in our final sample. The bottom panels show an observation toward the X-ray binary 4U1822-37 (Obs. ID 401051010), but excluding the point source from the extraction region. We excluded this observation from the sample because the light curve shows clear episodic variations due to residual X-ray binary emission.
}
\label{figure.lc_regfilter_chap_fb}
\end{center}
\end{figure*}
\begin{figure*}
\begin{center}
\includegraphics[width = 1.0\textwidth, keepaspectratio=true]{lc_100011010_507002010_fluxfiltered.pdf}
\caption[Solar wind proton flux and X-ray light curves]{Example solar wind proton flux light curves (top panels) and 0.4--10.0 keV light curves (bottom panels) light curves for two observations in our initial sample. Observing periods when the solar wind proton flux exceeded $2 \times 10^8$ cm$^{-2}$ s$^{-1}$ (black dashed lines in the {top} panels) were excluded. We represent these periods with a gray background in the {top} panels and gray points in the {bottom} panels. The {left} panels show the same Galactic bulge observation from Figure~\ref{figure.lc_regfilter_chap_fb} (Obs. ID 100011010), which had some observing time removed, but was retained in the sample because it had $\geq$5 ks of good observing time. The {right} panels show an observation from the analysis of \citet{kataoka_etal13} (FERMI\_BUBBLE\_N2; Obs. ID 507002010). We rejected this observation from the sample since most of it occurred during a period of high solar wind proton flux.
}
\label{figure.lc_fluxfilter_chap_fb}
\end{center}
\end{figure*}
We compiled an initial \textit{Suzaku} target list of all observations that were publicly available as of 2015 January and near the Fermi bubbles. This included any observations with Galactic coordinates $|l| \leq\ 25\arcdeg$ and $10\arcdeg \leq |b| \leq\ 55\arcdeg$. There were 143 observations in this region of the sky, which we inspected for usable spectra.
Each observation includes data from the three active X-ray Imaging Spectrometer (XIS) detectors on board \textit{Suzaku} \citep{koyama_etal07}. The detectors each have an $18\arcmin \times 18\arcmin$ field of view and a point spread function of $\approx 2 \arcmin$. We only considered data from the back-illuminated XIS1 detector since it has a larger collecting area below 1 keV than the other detectors.
We processed all XIS1 data using HEAsoft version 6.17 and calibration database (CALDB) files from 2015 January. We followed the standard data reduction procedure described in \textit{The Suzaku Data Reduction Guide}.\footnote{\url{https://heasarc.gsfc.nasa.gov/docs/suzaku/analysis/abc/}} This includes recalibrating the raw data files, screening for flickering or bad pixels, energy-scale reprocessing, and building good time interval (GTI) files. Fortunately, the FTOOL script \texttt{aepipeline} performs all these tasks for standard \textit{Suzaku} observations. We used \texttt{aepipeline} version 1.1.0 to generate reprocessed images, light curves, and spectra for our analysis.
We extracted all data products using \texttt{xselect} version 2.4. Each data set combined data from $3 \times 3$ and $5 \times 5$ editing modes where applicable. We did not impose any non-standard criteria on the data extraction steps with the exception of the cutoff rigidity (COR) of the Earth's magnetic field. This parameter varies throughout \textit{Suzaku's} low Earth orbit, and larger COR constraints result in fewer particle background counts. A higher cutoff than the standard COR $>$ 4 GV has been used in previous SXRB spectral fits, and typically results in higher signal-to-noise ratio between 0.5 and 2.0 keV \citep{smith_etal07, kataoka_etal15}. Here, we follow the suggested value from \citet{smith_etal07} and use a constraint of COR $>$ 8 GV.
Since the observational goal of this study is to measure \ion{O}{7}\ and \ion{O}{8}\ intensities from SXRB spectra, we removed point sources from each observation before we extracted spectra. We inspected each XIS1 image for point sources to remove with self-defined region files. Observations of exceptionally bright sources (i.e., where there was clear emission extending over more than $\approx5\arcmin$), extended sources (galaxy clusters, star clusters, etc.), or other anomalous features were rejected from the data set. We also rejected any observations not taken in the standard observing window mode due to the reduction in field-of-view collecting area. For any remaining visible point sources, we defined circular exclusion regions between 1$\arcmin$ and 4$\arcmin$ in radius centered on each source. We then re-extracted the data products for each observation with the point source region excluded.
The next step in our data cleaning process was light curve inspection and filtering. We extracted 0.4--10.0 keV light curves and constructed count-rate histograms for each observation. Our default screening criterion was to remove observing periods that were $> 2 \sigma$ from the mean count rate. This led to a small reduction in observing time since most count-rate histograms followed an approximately Gaussian distribution. We flagged observations that did not have an approximately Gaussian count-rate distribution, which is indicative of additional soft proton flares or residual point-source emission that may have been variable throughout the observation. Example light curves with the various filtering tasks can be seen in Figure~\ref{figure.lc_regfilter_chap_fb}. We also expand on this analysis step in Section~\ref{subsection.swcx_chap_fb} where we exclude observing periods with high solar wind proton flux measurements. This light curve filtering created new GTI files, which we used to compile our initial data products.
The procedures for point-source exclusion and light curve filtering procedures outlined above were the primary stages in our initial sample catalog. To summarize the main screening criteria, we excluded observations that showed anomalous features in either their images or light curves or that had exceptionally bright point sources, and we filtered the images for removable point sources. After these screening criteria, we kept observations with $\geq$5 ks of good observing time. This resulted in 112 observations out of the original 143.
\subsection{SWCX Filtering}
\label{subsection.swcx_chap_fb}
SWCX emission can occur in any X-ray observation, but it is difficult to predict or attribute the amount of SWCX emission in individual SXRB observations \citep[e.g., ][]{carter_sembay08, carter_etal11, galeazzi_etal14, hs15}. The emission can occur either at the interface between the solar system and interstellar medium (ISM) (heliospheric emission) or as solar wind ions pass near the Earth's neutral atmosphere (geocoronal emission). Heliospheric emission is thought to vary with the overall solar cycle, the observed direction relative to the Sun's orbit and ecliptic plane, and the hydrogen and helium density in the neutral ISM \citep{cravens_etal01, robertson_cravens03_a, koutroumpa_etal06, koutroumpa_etal07, koutroumpa_etal11, galeazzi_etal14, hs15}. Geocoronal emission is thought to depend on the solar wind proton flux and the observed direction relative to the magnetosheath \citep{snowden_etal04, wargelin_etal04, fujimoto_etal07, carter_sembay08, carter_etal10, carter_etal11, ezoe_etal10, ezoe_etal11, ishikawa_etal13, hs15}. It is still unclear what the typical amount of SWCX emission is in a given X-ray observation, and models predict a wide range of \ion{O}{7}\ and \ion{O}{8}\ intensities depending on the parameters listed above \citep{robertson_cravens03_b,koutroumpa_etal06, koutroumpa_etal07, koutroumpa_etal11,robertson_etal06}. For the purpose of this project, SWCX emission is considered to be contamination, and our goal is to reduce the amount of potential emission as much as possible.
Following the work of \citetalias{hs12}, we filter the observations for periods of high solar wind proton flux in an effort to reduce geocoronal SWCX. For each observation, we gathered solar wind data from the OMNIWeb database,\footnote{\url{http://omniweb.gsfc.nasa.gov/}} which includes data from the \textit{Advanced Composition Explorer} and \textit{Wind} satellites. The database includes solar wind densities and velocities, which we convert to fluxes. We cross-correlated each solar wind proton flux light curve with the X-ray light curves. Periods with solar wind flux values $> 2 \times 10^8 $ cm$^{-2}$ s$^{-1}$ were flagged and considered to potentially include SWCX emission. We illustrate how this screening works for several example spectra in Figure~\ref{figure.lc_fluxfilter_chap_fb}. We made new GTI files incorporating these filtering periods and the $> 2 \sigma$ count rate periods discussed above. These GTI files were used for the final spectral extraction used in the fitting analysis.
We point out that this filtering procedure is designed to reduce geocoronal SWCX emission, not heliospheric SWCX emission. The models suggesting that heliospheric SWCX varies with ecliptic latitude imply that applying an ecliptic latitude cut to an observing sample may help reduce that emission. Indeed, \citet{hs13} discuss this effect and employ this screening criterion for their study on fitting SXRB spectra. However, the analysis in \citetalias{miller_bregman15} argues that there does not appear to be a significant enhancement of \ion{O}{7}\ or \ion{O}{8}\ line emission within 10$\arcdeg$ of the ecliptic plane, part of which passes through the Fermi bubbles. Therefore, heliospheric SWCX is likely present at softer X-ray energies, but it does not appear to be a significant emission source for the oxygen lines of interest.
This additional screening procedure can only reduce the good exposure time in a given observation. Some observations occurred entirely during a period of high solar wind proton flux, in which case the observation was removed from the sample. Other observations occurred entirely during a period of low solar wind proton flux, in which case the observation was unaffected. The rest of the observations were partially contaminated, leading to a reduction of observing time. We enforced the same good exposure time requirement noted above of $>$5 ks to keep observations in the sample. After the default screening outlined in Section~\ref{subsection.suz_chap_fb} and this additional SWCX filtering, our sample includes 58 of the original 143 observations.
\begin{figure}
\begin{center}
\includegraphics[width = .5\textwidth, keepaspectratio=true]{702028010_spectrum_zoom.pdf}
\caption[Example SXRB spectrum for Observation ID 702028010]{Binned SXRB spectrum showing our spectral model components (Obs. ID 702028010). The solid magenta line represents our NXB model, the green dashed line is the absorbed CXB power law, the blue dashed line is the absorbed hot gas continuum without the oxygen lines, the red dashed lines show the \ion{O}{7}\ and \ion{O}{8}\ lines as Gaussian components, and the solid red line represents the total model spectrum. The oxygen lines dominate the spectrum between 0.5 and 0.7 keV.
}
\label{figure.spectrum_chap_fb}
\end{center}
\end{figure}
\begin{figure*}[t]
\begin{center}
\includegraphics[width = 1.0\textwidth, keepaspectratio=false]{fb_map_obs_combinedfit.pdf}
\caption[Line intensity maps for \textit{XMM-Newton} and \textit{Suzaku} data]{All-sky Aitoff projections ({left} panels) and a projection near the Fermi bubbles ({right} panels) of our \ion{O}{8}\ and \ion{O}{7}\ emission line samples ({top} and {bottom} panels respectively). The squares represent measurements from \textit{XMM-Newton} \citepalias{hs12}, the circles represent our new \textit{Suzaku} measurements, and the dashed lines represent the Fermi bubbles' gamma-ray edge. We use the \ion{O}{8}\ data in our model fitting process.
}
\label{figure.maps_chap_fb}
\end{center}
\end{figure*}
\subsection{Spectral Modeling}
\label{subsection.spec_mod_chap_fb}
This section outlines our spectral fitting procedure, including the response files used, our spectral model, and resultant data products. We extracted spectra in the 0.4--5.0 keV band, which is broad enough to model the known SXRB emission sources. Each observation had its own particle spectrum, or non X-ray background (NXB) spectrum, and response files. We used \texttt{Xspec} version 12.9.0 for all spectral fitting, where we used the Cash statistic as our fit statistic \citep{cash79}. Figure~\ref{figure.spectrum_chap_fb} shows an example observed spectrum and best-fit multi-component model. Our final result includes \ion{O}{7}\ and \ion{O}{8}\ line intensities for each observation.
We generated response matrices and auxiliary response files using standard \textit{Suzaku} scripts. The script \texttt{xisrmfgen} was used to generate response matrices (RMF files) for each observation. We used the ray-tracing program \texttt{xissimarfgen} to generate ancillary response files (ARFs) assuming a uniformly emitting source and a 20$\arcmin$ radius circle for a simulated source region. The size of the source region acts as a normalization in our spectral fit values.
Each observation has an NXB spectrum collected by \textit{Suzaku} observations of the Earth at night. We generated NXB spectra using the script \texttt{xisnxbgen} \citep{tawa_etal08}. This creates an NXB spectrum using a weighted sum of \textit{Suzaku} of the Earth at night based on the exposure times. The only unique parameter we supplied to the script is the same constraint of COR $>$8 GV that we applied to the initial data extraction. Different SXRB studies have multiple treatments for these NXB spectra, with groups either subtracting the NXB counts from the observed spectrum \citepalias[e.g., ][]{kataoka_etal15} or including the NXB spectrum as additional data and simultaneously modeling its contribution to the full spectrum \citep[e.g., ][]{smith_etal07}. We follow the latter methodology, meaning we fit both our observed spectrum and the NXB spectrum as one process.
Our spectral model includes the following components: NXB spectrum, an absorbed cosmic X-ray background (CXB) or extragalactic power law, an absorbed hot gas continuum component with no oxygen emission lines, and the \ion{O}{7}\ and \ion{O}{8}\ lines of interest. The absorbed components were attenuated using the \texttt{phabs} model in Xspec \citep{bc_mccammon92, yan_etal98} and had column densities fixed to values from the LAB survey \citep{kalberla_etal05}. We also assume metal abundances from \citet{asplund_etal09} unless otherwise stated. The rest of this section details the spectral model assumed for each source.
The NXB model includes a contribution from particles hitting the XIS detectors that are not focused by the telescope and three instrumental lines. For the particle spectrum, we include a power law where both the normalization and slope can vary. The instrumental lines include an Al K line at 1.49 keV, an Si K line at 1.74 keV, and an Au M line at 2.12 keV.\footnote{\url{https://heasarc.gsfc.nasa.gov/docs/suzaku/prop_tools/suzaku_td/}} We model each of these lines as Gaussians with widths and normalizations left to vary as free parameters and the centroids fixed. This model component is only folded through the RMF response (as opposed to both the RMF and ARF), and it contributes to both the observed and NXB spectra.
The CXB spectrum is typically modeled as an absorbed power law or multiple broken power laws, and is thought to be due to unresolved active galactic nuclei (AGNs). The differences between these spectral shapes have minimal effects on the measured oxygen line intensities, since the CXB contributes $\lesssim10\%$ to the total SXRB flux below $\approx$1 keV. Therefore, we adopt the spectral shape used by \citetalias{hs12}, which is a power law with fixed spectral index of 1.46 \citep{chen_etal97}. These authors discuss the uncertainty in the CXB power law normalization, but argue for a nominal value of 7.9 photons cm$^{-2}$ s$^{-1}$ sr$^{-1}$ keV$^{-1}$ at 1 keV after accounting for CXB sources with $F_X < 5 \times 10^{-14}$ erg cm$^{-2}$ s$^{-1}$ in the 0.5--2.0 keV band \citep{moretti_etal03}. We allow the CXB normalization to be a free parameter in our spectral model, but place $\pm$30\% hard boundaries on the parameter to allow for field-to-field variation.
Although the oxygen lines are the measurement of interest, we still account for hot gas continuum emission in our model. We model this component as an absorbed thermal APEC plasma \citep{smith_etal01} with fixed solar metallicty and without the oxygen lines. We achieve the latter by setting the oxygen line emissivities to zero in the standard Xspec APEC files \citep{lei_etal09, hs12}. The normalization and temperature were left as free parameters in the model. We expected the fitted plasma temperatures to be between 0.1 and 0.3 keV, but these temperatures are typically most sensitive to the oxygen lines that we disabled. Therefore, we let the plasma temperature vary outside this range, but with hard boundaries between 0.05 and 5 keV.
Our final spectral model components are the \ion{O}{7}\ and \ion{O}{8}\ emission lines. We model each of these components as Gaussian features with the widths fixed to the instrumental resolution. We fixed the \ion{O}{8}\ centroid to its laboratory value of 0.6536 keV, but let the \ion{O}{7}\ centroid vary since the line is an unresolved blend of the resonance, forbidden, and intercombination lines. The Gaussian normalizations were also free parameters, because these represent the line strength measurements. We point out that this spectral fitting method measures the \textit{total} emission line strengths from all emission sources (residual SWCX, LB, absorbed hot gas halo, or Fermi bubble) for each sight line. This is why our model line intensities in Section~\ref{subsection.intensity_chap_fb} include all emission sources along each sight line.
\subsection{The Exclusion of the North Polar Spur Region}
\label{subsection.nps}
The North Polar Spur (NPS) is an extended region of enhanced X-ray emission that is cospatial with part of a larger region of enhanced radio continuum emission known as Loop I \citep{berkhuijsen_etal71,borken_iwan77,snowden_etal97}. The X-ray enhancement is strongest at lower Galactic latitudes near $l, b \approx$ 25\arcdeg, 25\arcdeg, and gradually decreases in intensity toward the Galactic pole. Spectral observations with \textit{XMM-Newton} and \textit{Suzaku} indicate that the plasma is hotter than the surrounding medium with $kT \approx 0.3$ keV, is depleted in C, O, Mg, Ne, Fe, and enhanced in N \citep{willingale_etal03,miller_etal08,ursino_etal16}. These measurements alone do not constrain the NPS distance, which makes it unclear whether the enhancement is associated with the Fermi bubbles.
Several independent methods place constraints on the NPS distance, but there is tension between the results. \citet{sofue15} analyzed how the NPS X-ray emission varies with extinction near the Aquila Rift, and concluded that the NPS must be behind the rift and $>$1 kpc from the Sun. \citet{puspitarini_etal14} compared three-dimensional models of the Milky Way's ISM and found no evidence for a low-density, hot cavity capable of producing the NPS X-ray emission within $\approx$200 pc of the Sun. Alternatively, \citet{sun_etal15} mapped the NPS in polarized radio emission and constrained the distance using maps of Faraday rotation measure and depth. They argue that the NPS emission at $b \gtrsim$50\arcdeg\ is likely within a few hundred parsecs of the Sun, while emission at lower latitudes can be either local or distant depending on the sign of the Milky Way's large-scale magnetic field.
These distance discrepancies are problematic because the NPS could be a region of compressed circumgalactic medium (CGM) material due to the Fermi bubbles' expansion, and thus a probe of their kinematics. This scenario was suggested long before the bubbles' discovery, and models of ``biconical hypershells'' expanding away from the Galactic center can reproduce the X-ray and radio morphologies of the NPS \citep{sofue77,sofue00,sofue_etal16}. Alternatively, models of stellar winds and supernova remnants from the Sco--Cen OB association can also reproduce the X-ray and radio morphologies, in which case the emission occurs $\approx$200 pc away from the Sun \citep{egger_aschenbach95,wolleben07}. In this scenario, the NPS has no affiliation from the Fermi bubbles, so analyzing its emission would lead to inaccurate measured bubble properties.
In light of these uncertainties, we choose to exclude observations that are projected through the NPS in our model fitting analysis. This is a conservative approach that prevents us from fitting hot gas emission that may be unrelated to our emission model. Furthermore, the NPS appears to be a unique feature in terms of its location on the sky. There should be similarly enhanced X-ray-emitting regions in the other quadrants of the sky if the NPS were due to the Fermi bubbles, but these are not seen in the \textit{ROSAT} maps. Thus, modeling the NPS emission could lead to biased and inaccurate results.
We exclude observations near the NPS in two square regions. Any observations within $l$ = 20\arcdeg--25\arcdeg, $b$ = 26\arcdeg--34\arcdeg, or $l$ = 4\arcdeg--20\arcdeg, $b$ = 40\arcdeg--57\arcdeg\ were excluded. This led to the exclusion of 27 observations from our model fitting procedure.
\begin{deluxetable*}{l r r r c r r}[t!]
\input{data_table.tex}
\end{deluxetable*}
\subsection{Data Summary}
\label{subsection.data_summary_chap_fb}
The final \textit{Suzaku} sample with spectral fitting results can be seen in Table~\ref{table.suz_data_chap_fb}. We include the observation ID, the sight line in Galactic coordinates, the good XIS exposure time, and the oxygen line strengths with their 1$\sigma$ uncertainties in Line Units (L.U.). The emission line measurements presented here are designed to have as little SWCX as possible, while containing only emission from astrophysical sources of interest (i.e., Galactic hot gas halo and Fermi bubbles). We also outlined data reduction, extraction, and cleaning prodecures as similar as possible to the work by \citetalias{hs12}, such that this sample and the \textit{XMM-Newton} data are processed in a uniform way.
Our total data sample used in our astrophysical modeling combines the \textit{XMM-Newton} and \textit{Suzaku} measurements. There are 683 \textit{XMM-Newton} measurements in total distributed across the sky, with 34 projected near the Fermi bubbles. The \textit{Suzaku} data are exclusively projected near the Fermi bubbles, and there are 58 measurements included here. Figure~\ref{figure.maps_chap_fb} shows all-sky maps of the oxygen emission line strengths.
\section{Model Overview}
\label{section.model_chap_fb}
In this section, we define our parametric astrophysical models and assumptions. The SXRB is known to have at least two plasma sources---a ``local'' source within $\approx$300 pc from the Sun and a ``distant'' source at $\gtrsim$5 kpc from the Sun. These sources have all been modeled in different ways, resulting in different inferences on their underlying emission properties. The Fermi bubbles have not been considered in most SXRB modeling studies, with the exception of recent studies by \citetalias{kataoka_etal15}. Here, we identify all emission sources in our model, and justify our choices for the underlying source distributions.
We point out that this work is an advancement over the modeling work presented in \citetalias{miller_bregman15}, who used the same \textit{XMM-Newton} data discussed in Section~\ref{subsection.xmm_chap_fb} to constrain the structure of the Milky Way's hot gas halo. The model used in that study is identical to the model outlined below, with the exception of the Fermi bubble emission source. We summarize these models in Sections~\ref{subsection.lb_mod_chap_fb} and ~\ref{subsection.halo_mod_chap_fb}, but we refer the reader to \citetalias{miller_bregman15} for additional explanation of the model choice.
\subsection{LB / Residual SWCX Model}
\label{subsection.lb_mod_chap_fb}
The ``local'' emission source has been argued to include emission from both the LB and SWCX based on numerous shadowing experiments \citep{galeazzi_etal07,koutroumpa_etal07,koutroumpa_etal11,smith_etal07} and studies of the \textit{ROSAT} 1/4 keV band \citep{kuntz_snowden00, galeazzi_etal14}. As discussed in Section~\ref{subsection.swcx_chap_fb}, SWCX emission is difficult to predict or quantify; however, the flux-filtering techniques are designed to reduce its contribution to our measured line strengths. The physical properties of the LB have also been debated, with some studies arguing that the LB is volume-filled with $\sim10^6$ K gas \citep{smith_etal07} and others arguing that the emission comes primarily from the bubble edges about 100--300 pc away \citep{lallement_etal03,welsh_shelton09}. Regardless of these differences, our goal is to choose a parameterization that characterizes the emission from this source.
We parameterize the LB as a volume-filled plasma with a constant density and temperature and a size varying between 100 and 300 pc. This follows interpretation from \citet{smith_etal07}, who conducted SXRB modeling with \textit{Suzaku} on the nearby molecular cloud MBM12. Under the assumption of a volume-filled plasma, these authors concluded that the LB has a temperature of $1.2 \times 10^6$ K and a density of 1--4$\times 10^{-3}$ cm$^{-3}$. In our model, we fix the plasma temperature to this value and let the density, $n_{LB}$, be a free parameter.
While we include an Local Bubble emission source in our model for completeness, it is unlikely to have a significant impact on our results. The ``local'' plasma source is known to contribute more to the \textit{ROSAT} 1/4 keV band than to the \textit{ROSAT} 3/4 keV band \citep[e.g., ][]{snowden_etal97,kuntz_snowden00}. This implies that the ``local'' emission source should produce more \ion{O}{7}\ than \ion{O}{8}\ in a given observation. Shadowing spectroscopic observations verify this, and show there is minimal ($\lesssim$0.5 L.U.) \ion{O}{8}\ due to ``local'' sources \citep{koutroumpa_etal07,koutroumpa_etal11,smith_etal07}. Furthermore, \citetalias{miller_bregman15} showed that this LB model effectively has no contribution to the \ion{O}{8}\ emission lines from the \textit{XMM-Newton} sample discussed in Section~\ref{subsection.xmm_chap_fb}. Our modeling work below focuses on \ion{O}{8}\ emission lines, so we do not believe that this LB parameterization will affect our results.
\subsection{Hot Halo Model}
\label{subsection.halo_mod_chap_fb}
We assume that the Milky Way's ``extended'' hot gas plasma structure is dominated by a spherical, volume-filling halo of material extending to the virial radius, as opposed to the alternative assumption of an exponential disk morphology with scale height between 5 and 10 kpc. The latter structure is believed to form from supernovae in the disk \citep[e.g., ][]{norman_ikeuchi89, joung_maclow06, hill_etal12} and can reproduce X-ray absorption and emission line strengths in several individual sight lines \citep{yao_wang05,yao_wang07,yao_etal09_b,hagihara_etal10}. However, numerous studies have shown that a spherical, extended morphology due to shock-heated gas from the Milky Way's formation reproduces a multitude of observations \citep[e.g., ][]{white_frenk91,cen_ostriker06, fukugita_peebles06}. These include ram-pressure stripping of dwarf galaxies \citep{blitz_robishaw00,grcevich_putman09,gatto_etal13}, the pulsar dispersion measure toward the Large Magellanic Cloud \citep{anderson_bregman10,fang_etal13}, and the aggregate properties of oxygen absorption and emission lines distributed in multiple sight lines across the sky \citep{bregman_ld07,gupta_etal12,miller_bregman13,miller_bregman15,faerman_etal16}. This distribution has been proven to reproduce most of the \ion{O}{8}\ emission line intensities from the \textit{XMM-Newton} portion of the sample, thus justifying its use in this modeling work.
Our parameterized density distribution follows a spherical $\beta$-model, which assumes that the hot gas is approximately in hydrostatic equilibrium with the Milky Way's dark-matter potential well. The $\beta$-model has also been used to fit X-ray surface brightness profiles around early-type galaxies \citep[e.g., ][]{osullivan_etal03} and massive late-type galaxies \citep{anderson_bregman11,dai_etal12,bogdan_etal13b,bogdan_etal13a,anderson_etal16}. The model is defined as
\begin{equation}
\label{eq.beta_model_chap_fb}
n(r) = {n_o}({1 + ({r}/{r_c})^2})^{-{3\beta}/{2}},
\end{equation}
\noindent where $r$ is the galactocentric radius, $n_\circ$ is the central density, $r_c$ is the core radius ($\lesssim$5 kpc), and $\beta$ defines the slope (typically between 0.4 and 1.0). The previous modeling by \citetalias{miller_bregman15} was limited to using an approximate form of this model in the limit where $r \gg r_c$, since they specifically did not include observations near the expected $r_c$. This resulted in constraints on a power-law density distribution:
\begin{equation}
\label{eq.beta_model_approx_chap_fb}
n(r) \approx \frac{n_or_c^{3\beta}}{r^{3\beta}}.
\end{equation}
\noindent The emission line sample in this study includes 33 sight lines that pass within 20$\arcdeg$ of the Galactic center, so we present model results assuming both distributions. The net effect of this will be for the power-law model to produce more halo emission for sight lines near the Galactic center than the usual $\beta$-model since the density continues to increase at small $r$ instead of approach $n_o$ for $r \lesssim r_c$.
We assume the halo gas is isothermal with a temperature of log($T_{halo}$) = 6.30, or $T_{halo} = 2 \times 10^6$ K. This temperature is characteristic of the Milky Way's virial temperature, and thus consistent with the picture in which the ``extended'' plasma is spherical and extended to $r_{vir}$. This temperature is also constrained by observations. \citet{hs13} provide the strongest observational constraints on the plasma temperature, as they fit SXRB spectra for 110 high-latitude ($|b| > 30\arcdeg$) sight lines from the \citetalias{hs12} sample. They fit the spectra with thermal APEC plasma models and found a narrow range of plasma temperatures (median and interquartile range of $2.22 \pm 0.63 \times 10^6$ K). These results suggest that the plasma is nearly isothermal, thus validating our assumption.
\subsection{Fermi Bubble Geometry}
\label{subsection.geometry_chap_fb}
\begin{figure*}[t!]
\begin{center}
\includegraphics[width = 1.0\textwidth, keepaspectratio=true]{fb_geometry.pdf}
\caption[Fermi bubble geometry in physical and projected coordinates]{Outlines of our volume-filled (blue lines) and shell (green lines) model distributions in physical galactocentric coordinates ({left} and {center}) and projected Galactic coordinates ({right}). The {left} panel shows a side view of the structure in the $l = 0\arcdeg$ plane (the $\odot$ symbol represents the Sun), while the {center} panel shows a face-on view at the Galactic center. The {right} panel indicates that our volume-filled distribution creates a projection consistent with the bubbles' observed gamma-ray outline (black dashed line).
}
\label{figure.geometry_chap_fb}
\end{center}
\end{figure*}
Our Fermi bubble structure includes two components: a volume-filled ellipsoid and a shell surrounding the ellipsoid. These components are designed to parameterize the outer regions of galactic outflows, where the volume-filled structure includes hot, shocked wind material, and the shell includes shocked ISM/CGM material (see Section~\ref{subsubsection.wind_chap_fb} for an overview of galactic wind morphology). Hard X-ray emission bounds the bubbles at low Galactic latitudes, verifying that there is a distinct shell structure of thermal gas surrounding the bubbles \citep{bh_cohen03,su_etal10}. The flat gamma-ray intensity distribution on the sky indicates that non-thermal particles fill the bubbles in a quasi-spherical volume, and we include a thermal gas component in this region.
We define the bubble volume as a three-dimensional ellipsoid designed to match the bubbles' projected gamma-ray edge on the sky. Each bubble (positive and negative Galactic latitudes) is centered at $|z|$ = 5 kpc away from the Galactic plane, has a semi-major axis of 5 kpc, and has both minor axes set to 3 kpc. We also tilt each bubble $5\arcdeg$ toward negative longitudes to match the slight asymmetry observed in the bubble shape. Figure~\ref{figure.geometry_chap_fb} shows this bubble volume in physical and projected coordinates.
The shell volume is defined in the same way as the bubble volume, but with a thickness of $\approx$1 kpc away from the bubble surface. This implies that the shell ellipsoids are also centered at $|z|$ = 5 kpc from the Galactic plane and have semimajor axes of 6 kpc and minor axes of 4 kpc. The region inside this surface but outside the bubble surface is considered to be the shell region. We note that this parameterization is different from the modeling work presented by \citetalias{kataoka_etal15}, who considered Fermi bubble emission from only two angled shells (one for each bubble) with inner and outer radii of 3 and 5 kpc. However, Figure~\ref{figure.geometry_chap_fb} indicates that our bubble volume is consistent with the projected bubble outline, and the expected galactic wind morphology suggests that there should be at least two distinct outflow regions we can observe (the shocked wind and shocked ISM/CGM). Therefore, we feel that our choice of bubble volume is reasonable given the observational constraints available at this time.
\subsection{Fermi Bubble Density and Temperature}
\label{subsection.fb_nt_chap_fb}
We assume that the bubble and shell components each have constant electron densities, defined as $n_{FB}$ and $n_{shell}$. This parameterization is useful since it is simple, yet still allows us to constrain the average thermal gas densities. Simulations suggest that the bubbles have some thermal gas substructure \citep[e.g., ][]{yang_etal12}, and we did explore more sophisticated models with density gradients away from the Galactic plane or from the bubble edges. However, the data did not provide statistically significant constraints with these profiles (any gradient parameter was consistent with a constant-density profile within the 1$\sigma$ uncertainties). These constant-density models should be considered a valuable first step when analyzing the bubbles' thermal gas structure.
The bubble and shell are likely hotter than the surrounding medium, so we assume that each component has a characteristic temperature $\geq 2 \times 10^6$ K. Like the bubble and shell densities, each component has a constant temperature ($T_{FB}$ and $T_{shell}$). However, these temperatures are each initially fixed to $3 \times 10^6$ K during the model fitting process. The temperatures are not free parameters in our models because the calculated line intensity scales with density and temperature as $I \propto n^2 \epsilon(T)$, where $\epsilon$ has a temperature dependence. Since we explicitly model a sample of \ion{O}{8}\ emission line intensities, the density and temperature parameters would be degenerate with each other. Section~\ref{section.results_chap_fb} discusses how we constrain the bubble and shell temperatures by looking at the distribution of \ion{O}{8}\ /\ion{O}{7}\ line ratios for different assumed temperatures.
Our modeling also implicitly assumes that the Fermi bubble and shell plasmas have solar metallicities. This implies that our bubble density parameters are degenerate with the assumed metallicity since the plasma emission measure scales linearly with metallicity. We make this choice because there are no direct observational constraints on the plasma metallicity inside the bubbles. The SXRB modeling from \citet{kataoka_etal13} advocates for a sub-solar bubble metallicity of $Z \approx 0.2 Z_{\odot}$, but this value is weakly constrained due to photon statistics, and the spectral fits represent the emission measure-weighted spectrum from the bubbles and Galactic hot halo (see discussion in Section~\ref{subsubsection.softxray_chap_fb}). Thus, it is difficult to interpret whether the bubbles' plasma is enriched or sub-solar. The bubbles' abundance ratios can be diagnostics for how they formed \citep{inoue_etal15}, but a detailed abundance analysis is beyond the scope of this work.
\subsection{Line Intensity Calculation}
\label{subsection.intensity_chap_fb}
\begin{figure*}[t]
\begin{center}
\includegraphics[width = 1.\textwidth, keepaspectratio=true]{fit_bin_ellipse_onlyratio.pdf}
\caption[\ion{O}{8}\ /\ion{O}{7}\ ratios binned on the sky]{
{Left:} measured \ion{O}{8}\ /\ion{O}{7}\ ratios folded into one quadrant of an Aitoff projection with the circles and squares representing \textit{Suzaku} and \textit{XMM-Newton} observations respectively. The black dashed lines represent the observed Fermi bubble outline while the black dotted lines represent edges used to bin the data. The line ratios are systematically larger for sight lines passing near the Fermi bubbles compared to sight lines near the Galactic pole or anti-center.
{Right:} median and interquartile ranges of line ratios binned on the sky. The bin edges are the black dotted lines in the {left} panel, with the first bin including all observations within the Fermi bubbles. The top of the gray band represents the line ratio expected for a plasma with $T = 2 \times 10^6$ K, and the bottom includes a contribution from a cooler LB plasma source or SWCX. The observations in the first bin have significantly larger line ratios than the expected range in the gray band, indicating the presence of a plasma at $> 2 \times 10^6$ K.
}
\label{figure.map_quarterbin_chap_fb}
\end{center}
\end{figure*}
Calculating model line intensities depends on the density and temperature profile along each line of sight. For any given Galactic coordinate ($l,b$), we divide the line of sight into cells extending to the virial radius ($r_{vir}$ = 250 kpc). Each cell position along the line of sight ($s$) is converted to Galactic coordinates ($R$, $z$, $r$) by the standard equations:
\begin{equation}
R^2 = R_{\odot}^2 + s^2\cos(b)^2 - 2sR_{\odot}\cos(b)\cos(l)
\end{equation}
\vspace{-.3cm}
\begin{equation}
z^2 = s^2\sin(b)^2
\end{equation}
\vspace{-.3cm}
\begin{equation}
r^2 = R^2 + z^2,
\end{equation}
\noindent where $R_{\odot}$ = 8.5 kpc is the Sun's distance from the Galactic center. We assign a density and temperature to each cell based on its set of Galactic coordinates and the assumed model parameters. The hot halo profile described in Section~\ref{subsection.halo_mod_chap_fb} sets the density and temperature for cells outside the shell volume. The parameters $n_{shell}$ and $T_{shell}$ set the density and temperature for cells within the shell volume, while $n_{FB}$ and $T_{FB}$ set the density and temperature for cells within the bubble volume. Therefore, sight lines not passing through the bubbles include only halo emission, while sight lines passing through the bubbles include emission from the hot gas halo, bubble, and shell.
We assume an optically thin plasma in collisional ionization equilibrium to calculate all line intensities. Given a line-of-sight density and temperature profile, the model line intensity is defined as
\begin{equation}
\label{eq.intensity_thin_chap_fb}
I (l,b) = \frac{1}{4 \pi} \int n_e(s)^2 \epsilon(T(s)) ds,
\end{equation}
\noindent where $n_e(s)$ is the line-of-sight electron density, $T(s)$ is the line-of-sight temperature profile, and $\epsilon(T)$ is the volumetric line emissivity for a thermal APEC plasma. We use AtomDB version 2.0.2 for all line emissivities \citep{foster_etal12}, and characteristic values for \ion{O}{8}\ (in photons cm$^3$ s$^{-1}$) are $\epsilon(T_{halo}) = 1.45 \times 10^{-15}$ and $\epsilon(3 \times 10^6\ \text{K}) = 3.84 \times 10^{-15}$. Although believed to be minimal, the intensity contribution from the LB is defined as
\begin{equation}
I_{LB} (l,b) = \frac{n_{LB}^2 L(l,b) \epsilon(T_{LB})}{4 \pi},
\end{equation}
\noindent where $L(l,b)$ defines the LB path length ($\approx$100--300 pc; see \citet{lallement_etal03} and \citetalias{miller_bregman15}). The total line intensity is thus defined as
\begin{equation}
I_{total} (l,b) = I_{LB} + e^{-\sigma N_{HI}}(I_{halo} + I_{FB} +I_{shell}),
\end{equation}
\noindent where the exponential term accounts for attenuation due to neutral hydrogen in the disk, $N_{HI}$ is the same neutral hydrogen column assumed for each sight line in in the spectral fitting procedure, and $\sigma$ is the \ion{H}{1} absorption cross section \citep{bc_mccammon92, yan_etal98}. Thus, our model line intensities are comparable to the total observed line intensities.
\begin{figure*}
\begin{center}
\includegraphics[width = 1.\textwidth]{residual_zoom_o8.pdf}
\caption[\ion{O}{8}\ residual maps near the Fermi bubbles]{Observed emission line intensities ({left} panel), model emission line intensities ({center} panel), and residuals ({right} panel) for a model without a bubble or shell emission component. The hot gas halo dominates the model \ion{O}{8}\ emission in this case. Sight lines passing through the bubbles have significant ($\gtrsim 3 \sigma$) positive residual \ion{O}{8}\ emission, which we attribute to the bubbles and their interaction with the ambient hot halo medium.
}
\label{figure.residuals_chap_fb}
\end{center}
\end{figure*}
\section{Results}
\label{section.results_chap_fb}
Our results include a discussion of the \ion{O}{7}\ and \ion{O}{8}\ line intensity distributions along with a parametric modeling analysis. Section~\ref{subsection.ratios_chap_fb} presents the line strength and ratio distributions on the sky for the combined \textit{XMM-Newton} and \textit{Suzaku} sample. The latter provides model-independent evidence that the bubbles contain gas at higher temperatures than the surrounding medium ($> 2 \times 10^6 $ K). Section~\ref{subsection.obs_model_chap_fb} builds on this evidence and the modeling work from \citetalias{miller_bregman15} to constrain the characteristic thermal gas densities and temperatures associated with the bubbles.
\subsection{Emission Line Ratios}
\label{subsection.ratios_chap_fb}
The observed \ion{O}{8}\ /\ion{O}{7}\ ratios in our sample can be used as crude temperature diagnostics. If the observed emission lines come from a single, cospatial plasma source, Equation~\ref{eq.intensity_thin_chap_fb} indicates that the \ion{O}{8}\ /\ion{O}{7}\ ratio is a direct temperature diagnostic because $I_{OVIII}/I_{OVII} \propto n^2 \epsilon_{OVIII}(T)/n^2 \epsilon_{OVII}(T) = \epsilon_{OVIII}(T)/ \epsilon_{OVII}(T)$. The observations from SXRB spectra are more complicated since we know that multiple plasma sources exist along each line of sight. This implies that the total observed \ion{O}{8}\ /\ion{O}{7}\ line ratio probes the emission measure-weighted temperature due to the various plasma sources. However, we discussed in Section~\ref{section.model_chap_fb} how the LB is believed to produce little \ion{O}{8}\ emission with a variable amount of \ion{O}{7}\ emission, and the hot halo plasma is believed to be nearly isothermal at $\approx 2 \times 10^6$ K. The expected \ion{O}{8}\ /\ion{O}{7}\ line ratio for a thermal plasma at this temperature is $\approx 0.25$, so the observed line ratios in our sample would be $\lesssim 0.25$ if they included only emission from the LB and hot halo.
We explore this idea by examining the \ion{O}{8}\ /\ion{O}{7}\ distribution on the sky from our total observation sample. Figure~\ref{figure.map_quarterbin_chap_fb} shows our line intensity ratios on the sky. Inspecting the sky projection alone suggests that the line intensity ratios are systematically higher for sight lines that pass through or near the Fermi bubbles ($\approx$0.5) than for those farther away from the Galactic center ($\approx$0.2). To quantify this, we bin the sight lines on the sky and calculate the median and interquartile range for the line ratios in each bin. The bin edges are defined as ellipses in $l,b$ space, where the first bin includes sight lines that pass though the Fermi bubbles and subsequent bins include sight lines extending farther into the halo (see dotted lines in Figure~\ref{figure.map_quarterbin_chap_fb}). Figure~\ref{figure.map_quarterbin_chap_fb} shows the line ratio median and interquartile range for observations in each bin. These results clearly show that the line ratios are systematically higher for sight lines in the first bin, and are also higher than the characteristic ratio expected if the observations included just LB and hot gas halo emission (gray shaded band in Figure~\ref{figure.map_quarterbin_chap_fb}).
These systematically larger line ratios near the Galactic center indicate the presence of hotter gas than the ambient $2 \times 10^6$ K plasma. This interpretation is model-independent and builds upon the fact that we know the Fermi bubbles occupy a significant volume above and below the Galactic center. While this is a useful result that relies only on observations, the observed line ratios do not encode the bubbles' detailed temperature structure due to additional emission from the LB and hot gas halo. Nevertheless, this result motivates the modeling work below and validates the assumption that the bubbles contain gas hotter than $2 \times 10^6$ K.
\subsection{Comparing Models with Data}
\label{subsection.obs_model_chap_fb}
As a preliminary test, we explore an emission model including only contributions from the LB and hot gas halo. This model assumes that the bubble and shell volumes contribute no line emission, or equivalently that $n_{FB} = n_{shell} = 0$. For the other emission components, we assume a parametric model distribution from \citetalias{miller_bregman15}. This includes an LB density of $n_{LB} = 4 \times 10^{-3}$ cm$^{-3}$ and a hot gas density profile described by Equation~\ref{eq.beta_model_approx_chap_fb} with $n_or_c^{3\beta}$ = $1.35 \times 10^{-2}$ cm$^{-3}$ kpc$^{3\beta}$ and $\beta$ = 0.5. This model likely \textit{overestimates} any halo emission since it assumes a power law all the way to the Galactic center, as opposed to having a flat core density. We calculate model \ion{O}{8}\ emission line intensities for this limiting case and compute the residual emission defined as $(I_{observed} - I_{model})/I_{error}$. Figure~\ref{figure.residuals_chap_fb} shows how the residual emission varies on the sky, with a particular emphasis on the strong ($\gtrsim 3 \sigma$) positive residuals near the Fermi bubbles. We interpret these residuals as missing emission due to the Fermi bubbles, which motivates the modeling procedure outlined below.
The goal of our modeling procedure is to find the density model that is most consistent with our observed data set, including contributions from the Fermi bubble and shell components. We quantify this consistency with the model $\chi^2$ or likelihood ($\mathcal{L} \propto$ exp$(-\chi^2 / 2)$). We use the publicly available Markov chain Monte-Carlo (MCMC) Python package \texttt{emcee} \citep{foreman_mackey_etal13} to explore our model parameter space and find the parameters that minimize the model $\chi^2$, or maximize the model ln($\mathcal{L}$). The output chains for each model parameter are treated as marginalized posterior probability distributions. We define ``best-fit'' parameters as the median values for each binned chain distribution, which yields identical results to the Gaussian-fitting procedure outlined by \citetalias{miller_bregman15}, assuming the distributions are approximately Gaussian. Thus, these best-fit parameters maximize the model likelihood, given the data.
We considered several different model parameterizations in our model fitting process. These included hot gas halo density models described by either a power law (Equation~\ref{eq.beta_model_approx_chap_fb}) with two free parameters (the normalization and $\beta$) or a full $\beta$-model (Equation~\ref{eq.beta_model_chap_fb}) with $r_c$ and $\beta$ left to vary. We did not let $n_o$ vary independently in this model since the previous modeling work from \citetalias{miller_bregman15} effectively constrained the halo normalization parameter $n_or_c^{3\beta}$. Our modeling procedure keeps this quantity fixed to $n_or_c^{3\beta} = 1.35 \times 10^{-2}$ cm$^{-3}$ kpc$^{3\beta}$, while letting the core radius vary as the free parameter. We also experimented with fixing the hot gas halo profile with the fit values from \citetalias{miller_bregman15} or with $r_c$ = 3 kpc, but we found that this made little difference in the best-fit parameters for either the halo density profile or the bubble/shell densities.
\begin{deluxetable*}{c c c c c c c c}
\input{results_table.tex}
\end{deluxetable*}
Table~\ref{table.mcmc_results_chap_fb} summarizes our best-fit model parameters, including 1$\sigma$ uncertainties encompassing the 68\% probability ranges from the posterior probability distributions. There are several trends to note from these results. The LB density parameter is consistent with zero, validating the assumption that the LB contributes little emission to the \ion{O}{8}\ data. The halo density profile results are consistent with those reported in \citetalias{miller_bregman15} when considering the same power-law density parametrization ($n_or_c^{3\beta} = 1.35 \times 10^{-2}$ cm$^{-3}$ kpc$^{3\beta}$, $\beta$ = 0.5). This implies that the hot gas density profile constraints are not biased due to observations near the Fermi bubbles. We also find characteristic best-fit core radii of 2--3 kpc, which is expected. The parameters of interest, $n_{FB}$ and $n_{shell}$, have characteristic best-fit densities of (5--8)$\times 10^{-4}$ cm$^{-3}$ assuming a temperature of log($T$) = 6.50. The inferred densities are lower if we assume a power-law model for the halo gas density than if we assume a $\beta$-model. We expect to see this trend since the power-law model produces more halo emission near the Galactic center than a $\beta$-model with a core radius/density, thus resulting in less Fermi bubble/shell emission being required to produce the observed emission. After weighing these effects, we define our fiducial model to be one with a $\beta$-model and $r_c$ fixed to 3 kpc. This results in best-fit parameters of $n_{FB} = 7.2 \pm 0.2 \times 10^{-4}$ cm$^{-3}$, and $n_{shell} = 7.7 \pm 0.2 \times 10^{-4}$ cm$^{-3}$. Figure~\ref{figure.pdf_chap_fb} shows the marginalized posterior probability distributions and contour plots from our MCMC analysis assuming this parametric model \citep[generated using the Python code \texttt{corner.py}; ][]{foreman_mackey_corner16}.
\begin{figure}
\begin{center}
\includegraphics[width =.5\textwidth, keepaspectratio=true]{pdf_bs_iso_fixhalo3_rc3_tilt.pdf}
\caption[Probability distribution functions for the Fermi bubble and shell parameters]{Our model fitting results for the Fermi bubbles' volume-filled and shell components represented as marginalized posterior probability distributions and a two-dimensional contour plot. The dashed lines and white cross represent the best-fit model values and the contour ranges include 1$\sigma$, 2$\sigma$, and 3$\sigma$. This model assumes that the bubble and shell components have a temperature of log($T$) = 6.50.
}
\label{figure.pdf_chap_fb}
\end{center}
\end{figure}
We also explore models with either the bubble or shell distributions to determine each component's significance in our model fitting procedure. A bubble-only model is equivalent to setting the shell thickness to 0 kpc and not including $n_{shell}$ as a model parameter parameter. A shell-only model is the same as our original emission model ($r_c$ fixed to 3 kpc) but with $n_{FB}$ fixed to zero. When we refit the data with these models, we find that each density component increases to compensate for the lack of emission from the other component. For example, $n_{FB}$ increased to $7.7\times 10^{-4}$ cm$^{-3}$ in the bubble-only model, and $n_{shell}$ increased to $10.0\times 10^{-4}$ cm$^{-3}$ in the shell-only model. These best-fit models lead to changes in the overall fit quality, where our initial best-fit $\chi^2_{r}$ (dof) = 2783 (739). The bubble-only model leads to a marginal improvement in the overall fit quality ($\chi^2_{r,\ bubble-only}$ (dof) = 2741 (740)), while the shell-only model leads to a worse quality of fit ($\chi^2_{r,\ shell-only}$ (dof) = 3241 (740)). This implies that the volume-filled structure is more important than the shell structure, although we point out that this exercise fixes every other component of the emission model (halo emission, bubble/shell geometry, etc.). Thus, we still assume that the bubble and shell structures are each present in our discussion and temperature analysis.
In order to constrain the bubble and shell temperatures, we compare best-fit model line ratios for different temperature distributions with the observed \ion{O}{8}\ /\ion{O}{7}\ line ratios near the Fermi bubbles. To do this, we change the bubble and shell temperatures while keeping the product $n^2 \epsilon_{OVIII}(T)$ fixed from the best-fit model results. This fixes the \ion{O}{8}\ emission coming from the bubble and shell, but changes the model \ion{O}{7}\ emission because $\epsilon_{OVII}(T)$ decreases faster than $\epsilon_{OVIII}(T)$ with increasing temperature. Thus, increasing the assumed temperature leads to an increase in the model line ratios, an increase in the inferred best-fit densities ($\epsilon_{OVIII}$ decreases for $T > 3 \times 10^6$ K), and a constant contribution to the \ion{O}{8}\ emission.
A model temperature distribution with log($T_{FB}$) = 6.60 and log($T_{shell}$) = 6.70 leads to a model line ratio distribution most consistent with the observed line ratios near the Fermi bubbles. This changes the inferred best-fit densities to $n_{FB} = 8.2\times 10^{-4}$ cm$^{-3}$ and $n_{shell} = 1.0\times 10^{-3}$ cm$^{-3}$ in order to keep the product $n^2 \epsilon_{OVIII}(T)$ fixed for each component. Figure~\ref{figure.ratio_hist_chap_fb} shows histograms of the observed and new best-fit model line ratios for sight lines that pass within $\approx 5 \arcdeg$ of the projected bubble edge ($\sim$100 sight lines). These densities and temperatures produce an \ion{O}{8}\ /\ion{O}{7}\ ratio median and interquartile range of 0.52 (0.41--0.60), consistent with the observed median and interquartile range of 0.49 (0.38--0.62). We treat these densities and temperatures as the characteristic physical properties for the bubble and shell components in our subsequent analysis.
\begin{figure}
\begin{center}
\includegraphics[width = .5\textwidth, keepaspectratio=true]{ratio_hist_bs_iso_fixhalo3_rc3_tilt_histonly.pdf}
\caption[Observed and model line ratio histograms near the Fermi bubbles]{Histograms of observed (shaded green areas) and model (black hatched areas) \ion{O}{8}\ /\ion{O}{7}\ line ratios for sight lines passing through the Fermi bubbles. The model assumes log($T_{FB}$) = 6.60 and log($T_{shell}$) = 6.70, which produces line ratios that are most consistent with the observations.
}
\label{figure.ratio_hist_chap_fb}
\end{center}
\end{figure}
\section{Discussion}
\label{section.discussion_chap_fb}
In this section, we discuss how our constraints fit in with our current picture for the Fermi bubbles and Milky Way. This includes an overview of the constrained thermal gas structure, and how this compares to the surrounding hot medium. We extend these constraints to infer the bubbles' characteristic shock strength, current expansion velocity, energy input rate, age, and likely formation scenario. We also discuss how our constraints compare with previous Fermi bubble analyses.
Table~\ref{table.inferred_prop_chap_fb} summarizes our most important inferred quantities discussed below for the best-fit densities and temperatures discussed above. The density uncertainties follow directly from our MCMC results summarized in Table~\ref{table.mcmc_results_chap_fb}. We use less strict criteria for the temperature uncertainties since we did not directly fit the \ion{O}{8}\ /\ion{O}{7}\ line ratios. The temperature limits represent where the differences between the observed and model line ratio medians are less than 1$\sigma$ of the corresponding uncertainties in the sample median. Uncertainties on all subsequent calculated quantities (masses, expansion rates, ages, etc.) use the density and temperature uncertainties listed in Table~\ref{table.inferred_prop_chap_fb}.
\subsection{Inferred Bubble Structure}
\label{subsection.structure_chap_fb}
We discuss our inferred bubble densities and temperatures in this section, and compare them to the assumed ambient structure. Overall, our constraints indicate that the bubbles are hotter and overpressurized compared to the surrounding medium, consistent with previous observations of the bubble \citep{su_etal10, kataoka_etal15}. Figure~\ref{figure.fit_slice_chap_fb} shows our best-fit model as two-dimensional maps of density, temperature, and pressure projected at the Galactic center. This visualizes the comparison with the surrounding medium that we discuss in the rest of the section.
The derived structure indicates that the best-fit densities and temperatures for the bubble and shell components are nearly identical to each other. This suggests that there may not be two distinct outflow regions, which contradicts the global morphology predicted from Galactic outflow simulations \citep[e.g., ][]{yang_etal12,sarkar_etal15,sofue_etal16}. This inference could be due to our assumed geometry, in terms of both the volume-filled shape and the shell thickness. Our emission model effectively constrains each component's emission measure, so changes in the bubble/shell path lengths along different sight lines could change the inferred densities. Additional substructure could also be present inside the volume-filled component, which might explain why we infer a relatively high density permeating the entire bubble volume. Modeling this substructure is beyond the scope of this work, but our modeling results still provide valuable constraints on the average bubble and shell properties.
\begin{deluxetable}{l c c c}[t]
\input{properties_table.tex}
\end{deluxetable}
The bubble and shell densities have characteristic values of $\sim 10^{-3}$ cm$^{-3}$, which are comparable to the surrounding medium at low $z$. Including a core radius for the hot gas halo of 3 kpc in our fiducial model implies a core density of $2.6 \times 10^{-3}$ cm$^{-3}$, assuming a fixed power-law normalization of $n_o r_c^{3\beta} = 1.35 \times 10^{-2}$ cm$^{-3}$ kpc$^{3\beta}$. This suggests $n_{shell} \sim n_{halo}$ within $|z| \lesssim $5 kpc. The hot gas halo density decreases by about a factor of 6 between $r = $1 and 10 kpc, meaning that our bubble and shell densities are larger than the surrounding CGM density farther away from the Galactic plane. We also note that $n_{FB} \approx n_{shell}$ from our model fitting results, making it difficult to distinguish between volume-filling emission and limb-brightened emission. This might be due to our choice to parameterize the structures with constant densities and temperatures, but our constraints still probe the average densities associated with the bubbles.
Our inferred bubble and shell temperatures of log($T_{FB,\ shell}$) = 6.60--6.70 are hotter than the surrounding medium (log($T_{halo}$) = 6.30). This is broadly consistent with the bubbles injecting enough energy to shock-heat the surrounding medium, although simulations predict a wide range of shock strengths and bubble temperatures as high as $\sim 10^8$ K \citep[e.g., ][]{guo_mathews12a, yang_etal12}. It is possible that the bubbles contain gas at this high temperature, but plasma at this temperature would not produce observable \ion{O}{8}\ emission. Our modeling results indicate that on average the bubble and shell are hotter than the surrounding medium, but still at low enough temperatures to produce observable signatures in the data.
Combining the density and temperature constraints indicates that the bubbles' are overpressurized compared to the surrounding medium. Our best-fit bubble and shell parameters indicate thermal gas pressures of $P_{FB} = 4.5 \times 10^{-13}$ dyn cm$^{-2}$, $P_{shell} = 6.9 \times 10^{-13}$ dyn cm$^{-2}$, or $P/k \approx $3000--5000 cm$^{-3}$ K. The surrounding thermal gas pressure varies with $r$ and $z$ due to the decreasing density profile, with characteristic values of $\approx$5000 cm$^{-3}$ K near the Galactic center and $\approx$1000 cm$^{-3}$ K at $r$ = 10 kpc. In this picture, the bubbles are in approximate pressure equilibrium at lower $z$, but become overpressurized with increasing height away from the Galactic plane. These estimates are also a lower limit to how overpressurized the bubbles actually are, because they do not account for non-thermal or magnetic pressure contributions. Nevertheless, these constraints indicate that the bubbles are generally overpressurized, and thus expanding into the surrounding medium.
We use these quantities to infer a characteristic shock strength and instantaneous expansion velocity associated with the bubbles. The classic treatment of shocks propagating through the ISM yields specific pre- and post-shock jump conditions for the gas density, temperature, and pressure given a shock expansion velocity (e.g., Shull \& Draine \citeyear{shull_draine87}, pp. 283--319). The Fermi bubbles' expansion is more complex than this traditional treatment since they do not appear to be spherical, and they are presumably expanding into a medium with varying density. For example, the ratio between $n_{shell}$ (treated as post-shocked material) and the ambient halo gas density along the shell edge (treated as pre-shock material) ranges between $\approx$0.5 near the Galactic center and $\approx$3 at the maximum bubble height. On the other hand, our choice to parameterize the hot gas halo and shell with constant temperatures allows us to use the temperature ratio as a shock strength diagnostic similar to \citetalias{kataoka_etal15}. Assuming a monotonic gas ($\gamma$=5/3), an ambient gas sound speed of $c_s$= 212 km s$^{-1}$, $T_{shell} = 5 \times 10^6$ K, and $T_{halo} = 2 \times 10^6$ K, our temperature ratio implies a fiducial Mach number and corresponding expansion velocity of $\mathcal{M} = 2.3$ and $v_{exp}$ = 490 km s$^{-1}$. The uncertainty in $T_{shell}$ expands these constraints to $\mathcal{M} =$1.9--3.4 and $v_{exp}$ = 413--720 km s$^{-1}$. These shock parameters are broadly consistent with the range of density and pressure ratios we estimate, indicating that these constraints probe the bubbles' current expansion rate into the surrounding medium.
\begin{figure*}
\begin{center}
\includegraphics[width = 1.\textwidth, keepaspectratio=true]{slice_mod_fiducial.pdf}
\caption[Best-fit model profiles as two-dimensional slices]{Our best-fit Fermi bubble model compared to the surrounding hot gas halo profile as two-dimensional slices at the Galactic center. The dashed lines represent the boundaries between the bubble and shell surfaces. The panels represent: density ({a}), temperature ({b}), pressure ({c}), and \ion{O}{8}\ emissivity defined as $n_e^2 \times \epsilon(T)$ ({d}). There is variation with height away from the Galactic plane, but the bubbles are hotter and overpressurized compared to the surrounding halo medium.
}
\label{figure.fit_slice_chap_fb}
\end{center}
\end{figure*}
\subsection{Bubble Energetics and Origin Scenarios}
\label{subsection.origins_chap_fb}
\subsubsection{The Bubbles as a Confined Galactic Wind}
\label{subsubsection.wind_chap_fb}
We treat the bubbles in the framework of a continuous galactic outflow/superbubble with self-similar Sedov-Taylor solutions \citep[e.g., ][]{castor_etal75, weaver_etal77, maclow_mccray88, veilleux_etal05}. The outflow morphology consists of five zones (from closer to farther from the outflow origin): the energy injection zone, a free-flowing outflow, shocked wind material, a shell of shocked ISM/CGM material, and the ambient ISM/CGM. Our model constraints probe the last three zones since we do not model observations in the inner $\approx$1 kpc from the Galactic center. The Sedov-Taylor solutions for this type of outflow relate the ambient density, bubble age, bubble size, expansion velocity, and average energy injection relate to each other. Assuming that the outflow is still in the energy-conserving phase (cooling time greater than the bubble age), the relations between these quantities are as follows:
\begin{equation}
\label{eq.age_chap_fb}
t_{age}
=\ 11.8\ Myr\
\left(\frac{r}{10\ \text{kpc}}\right)
\left(\frac{v}{500\ \text{km s}^{-1}}\right)^{-1},
\end{equation}
\begin{equation}
\label{eq.edot_chap_fb}
\begin{aligned}
&\xi\dot{E}
=\ 3.7 \times 10^{42}\ \text{erg s}^{-1}\
\\
&\times \left(\frac{n_o}{10^{-3}\ \text{cm}^{-3}}\right)
\left(\frac{r}{10\ \text{kpc}}\right)^2
\left(\frac{v_{exp}}{500\ \text{km s}^{-1}}\right)^3,
\end{aligned}
\end{equation}
\noindent where $n_o$ is the ambient density, $r$ is the bubble radius, $v_{exp}$ is the expansion velocity, $t_{age}$ is the bubble age, $\dot{E}$ is the energy injection rate, and $\xi$ is the thermalization efficiency of the mechanical energy. This thermalization efficiency is believed to vary with environment, but is estimated to be $\gtrsim$10\% in average galaxies with a typical assumed value of 0.3 \citep[e.g., ][]{larson74, wada_norman01, melioli_degouvia04}.
Our modeling results constrain three of the five variables in these equations. As discussed above, a hot gas halo model with $r_c$ = 3 kpc and fixed power-law normalization of $1.35 \times 10^{-2}$ cm$^{-3}$ kpc$^{3\beta}$ results in an ambient core density of $n_o = 2.6 \times 10^{-3}$ cm$^{-3}$. The constraints on the bubble and shell temperatures suggest an outflow velocity of 490 km s$^{-1}$. The bubble size is not trivial to estimate in this framework, where the outflow is typically treated as a spherical shell with radius $r$ centered on the injection source. The bubbles' shape is more complex than this since two lobes exist on each side of the Galactic plane. We use a characteristic bubble size defined as the geometric mean of the three ellipsoidal axes, resulting in $r$ = 3.6 kpc.
Given these constrained values, we estimate the bubbles' age and average mechanical energy injection rate. The bubbles' dynamical timescale, $t_{dyn} = d/v_{exp}$, is a crude age estimate that does not incorporate the bubble environment or energy source. For $v_{exp}$ = 490 km s$^{-1}$, the dynamical timescale for the full bubble height is $t_{dyn,\ h}$ = 10 kpc / 490 km s$^{-1}$ = 20.0 Myr, and the dynamical timescale for half the bubble width is $t_{dyn,\ w}$ = 3 kpc / 490 km s$^{-1}$ = 6.0 Myr. The superbubble model calculation (Equation~\ref{eq.age_chap_fb}) is a refined age estimate, where we find $t_{age}$ = 4.3 Myr for $r$ defined as the geometric mean above and $v_{exp}$ = 490 km s$^{-1}$. We also infer a combined energy injection for both bubbles ($2 \times \xi \times \dot{E}$) of $2.3 \times 10^{42}$ erg s$^{-1}$ using Equation~\ref{eq.edot_chap_fb}. Accounting for the uncertainty in $v_{exp}$ leads to a characteristic age range of $\approx$3--5 Myr and energy injection rate of $\approx$1--7$ \times 10^{42}$ erg s$^{-1}$. We compare this characteristic age and energy injection rate with possible bubble formation mechanisms.
\subsubsection{Origin from Sgr A* Accretion}
\label{subsubsection.sgra_chap_fb}
One suggested bubble formation mechanism has been a past accretion event onto Sgr A*, resulting in an AGN episode in the Milky Way. Sgr A* has an estimated mass of $4 \times 10^6\ M_{\odot}$ \citep{schodel_etal02,ghez_etal03,ghez_etal08,gillessen_etal09b,gillessen_etal09a,meyer_etal12}, which is capable of producing significant amounts of energy during an accretion episode. We also know that accretion onto supermassive black holes can produce galactic outflows with significant energy injection rates and morphologies similar to the observed Fermi bubbles \citep[e.g., ][]{mcnamara_nulsen07,yuan_narayan14}. Here, we consider observations of Sgr A* and its possible accretion history, and compare the expected energetics with our modeling constraints.
Sgr A* is currently in a quiescent state with a bolometric luminosity of $L_{bol} \sim 10^{36}$ erg s$^{-1}$ $\sim 2 \times 10^{-9}\ L_{Edd}$ \citep[e.g., ][]{yuan_narayan14}. Our proximity to Sgr A* allows for a combination of techniques to estimate current mass accretion rates. $Chandra's$ resolution is comparable to the Sgr A* Bondi radius, and has constrained the Bondi accretion rate to be $\sim 10^{-5}$ $M_{\odot}$ yr$^{-1}$ \citep{baganoff_etal03}. Polarized radio emission constrains the accretion rate near the event horizon, with limits being between $> 2 \times 10^{-9}$ $M_{\odot}$ yr$^{-1}$ and $< 2 \times 10^{-7}$ $M_{\odot}$ yr$^{-1}$ depending on the magnetic field orientation \citep[e.g., ][]{marrone_etal07}. While this is a significant uncertainty in the current mass accretion rate, the consensus is that Sgr A* is accreting well below its Eddington rate, and has been a well-modeled source for radiatively inefficient accretion flows (RIAFs).
There are a number of observational indications that Sgr A* has been more active in the past \citep{totani06}. \citet{mou_etal14} summarizes these lines of evidence, which include: a higher Sgr A* luminosity is required to produce fluorescent iron emission and reflection nebulae seen in several nearby molecular clouds \citep{koyama_etal96,murakami_etal00,murakami_etal01b,murakami_etal01a}, there exists an ionized halo of material surrounding Sgr A* \citep{maeda_etal02}, there are dynamic features indicating an outflow near the Galactic center in the form of the Galactic Center Lobe \citep{bh_cohen03} and the Expanding Molecular Ring \citep{kaifu_etal72, scoville72}, excess H$\alpha$ emission seen in the Magellanic Stream \citep{bh_etal13}, and possibly the Fermi bubbles themselves. The RIAF modeling from \citet{totani06} argues that Sgr A* should have had an accretion rate $\sim 10^3$--$10^4$ times larger than its current accretion rate over the past $\sim$10 Myr to reproduce these observations. This introduces additional scatter in the inferred past Sgr A* accretion rate, but motivates the assumption that Sgr A* has injected energy into the surrounding medium through an accretion event.
We estimate an energy injection rate due to past Sgr A* accretion and compare with our constrained energy input rate. The mechanical energy injection rate from black hole accretion is tied to the accretion power by the following relation:
\begin{equation}
\label{eq.bh_chap_fb}
\begin{aligned}
\dot{E}_{BH} &= \epsilon \dot{M}_{acc}c^2
\\
&= 5.7 \times 10^{45}\ \text{erg s}^{-1}\
\left( \frac{\epsilon}{0.1} \right) \left(\frac{\dot{M}_{acc}}{M_{\odot}\ \text{yr}^{-1}}\right),
\end{aligned}
\end{equation}
\noindent where $\dot{E}_{BH}$ is the mechanical energy injection rate, $\dot{M}_{acc}$ is the accretion rate near the event horizon, and $\epsilon$ is the mechanical energy injection rate efficiency. If we assume a past accretion rate of $10^{-3}$ $M_{\odot}$ yr$^{-1}$ (near the high end of the values discussed above), we find that $\dot{E}_{BH}$ equals our inferred vale of $2.3 \times 10^{42}$ erg s$^{-1}$ for $\epsilon \approx 0.05$. This efficiency is larger than the typical values inferred from simulations \citep[$10^{-4}$--$10^{-3}$; ][]{yuan_etal15}, but this mechanical efficiency is often treated as a free parameter in simulations. We also point out that the required efficiency is less than one, indicating that this analysis does not violate energy conservation constraints. \textit{Thus, it is plausible that a past accretion episode onto Sgr A* could have produced enough energy to match our energy injection rate constraints.}
The bubble age indicates that this Sgr A* accretion episode had a shorter active period than the typical AGN duty cycle. Studies constrain the AGN duty cycle by either comparing black hole mass functions (inferred from the $M_{BH}$--$\sigma$ relation) to AGN luminosity functions at different redshifts \citep[e.g., ][]{yu_tremaine02,shankar_etal04,hao_etal05,schawinski_etal10} or through analytic models of black hole accretion \citep[e.g., ][]{hopkins_hernquist06,shankar_etal09}. These techniques suggest that black holes with masses of $\sim 10^6$ $M_{\odot}$\ should have active periods of $\sim 10^8$ yr at $z$ = 0, or $\sim$1\% of a Hubble time. Our Fermi bubble age estimate is an upper limit to the active Sgr A* accretion time, and our constraint of 4.3 Myr is much smaller than the inferred duty cycle from AGN populations. One possible explanation for this discrepancy is that the Fermi bubble outburst was one of several past accretion events in the Galactic center. Our constraints imply that the Fermi bubbles are due to a relatively weak AGN event, and it is possible that multiple Sgr A* accretion events of comparable or lower energy have occurred over the past $\sim 10^8$ yr. Our results are also consistent with the overall decrease in AGN activity since $z \sim 2$ \citep[e.g., ][]{hopkins_etal07}, as opposed to a prolonged Sgr A* accretion/growth phase.
\subsubsection{Origin from Nuclear Star Formation}
\label{subsubsection.nsf_chap_fb}
Numerous studies also suggest that the Fermi bubbles formed from a period of enhanced star formation activity near the Galactic center. The Galactic center hosts several young stellar clusters with ages ranging between 5 and 20 Myr and accounting for $\sim 5 \times 10^5$ $M_{\odot}$ of material. The massive stars in these clusters could have generated a galactic-scale outflow due to stellar winds and type-II supernova explosions \citep[e.g., ][]{leitherer_etal99}. Here, we compare the expected energy output from past star formation near the Galactic center to our energy injection rate constraints.
The Galactic center star formation history is complex and difficult to measure, but several studies argue for an average star formation rate (SFR) of $\approx 0.05$ $M_{\odot}$ yr$^{-1}$ over the past $\sim$10 Myr. \citet{crocker12} reviews these studies, most of which utilize \textit{Spitzer} observations of young stellar objects within the inner $\sim$500 pc from the Galactic center. For example, \citet{yusefzadeh_etal09} conducted a census of these objects using the Infrared Array Camera and Multiband Imaging Photometer on board \textit{Spitzer}, and concluded the average SFR has been between 0.04 and 0.08 $M_{\odot}$ yr$^{-1}$ over longer timescales ($\sim$10 Gyr). \citet{immer_etal12} performed a similar analysis using data from the \textit{Spitzer} Infrared Spectrograph, and argue for an average SFR of $\approx 0.08$ $M_{\odot}$ yr$^{-1}$ over the past $\sim$Myr. Others estimate the SFR to be $\approx$0.01--0.02 $M_{\odot}$ yr$^{-1}$ by counting the mass in young star clusters and dividing that by estimates for the period of star formation \citep{figer_etal04, mauerhan_etal10}. Thus, it appears that a characteristic SFR of $\approx 0.05$ $M_{\odot}$ yr$^{-1}$ over the past $\sim$10 Myr is a reasonable assumption.
Similar to our argument concerning the black hole accretion energy above, we estimate the energy injection rate due to star formation in the Galactic center to compare with our constrained energy input rate. Assuming a Kroupa initial mass function \citep{kroupa01}, and $10^{51}$ erg of mechanical energy input from a type-II supernova, the mechanical energy from type-II supernovae is related to the SFR as
\begin{equation}
\label{eq.nsf_chap_fb}
\dot{E}_{nsf} = 1.1\ \times 10^{40}\ \text{erg s}^{-1}\ \left( \frac{\epsilon}{0.3} \right) \left(\frac{SFR}{0.1\ M_{\odot}\ \text{yr}^{-1}}\right),
\end{equation}
\noindent where $\dot{E}_{nsf}$ is the mechanical energy input rate due to nuclear star formation and $\epsilon$ is an efficiency factor typically assumed to be $\approx$0.3 (see Crocker et al. \citeyear{crocker_etal15} or Sarkar et al. \citeyear{sarkar_etal15} for equivalent relations). This implies that an average SFR of 0.05 $M_{\odot}$ yr$^{-1}$ over the past $\sim$10 Myr produces an energy injection rate of $\approx 6 \times 10^{39}$ erg s$^{-1}$. This estimate falls $\approx$400 times lower than our estimated energy input rate of $2.3 \times 10^{42}$ erg s$^{-1}$. It is possible that the SFR has been more variable over the past $\sim$10 Myr, however the upper limits are only $\approx$3 times higher than the average value \citep{yusefzadeh_etal09}. \textit{Thus, star formation in the Galactic center does not produce enough energy to inflate the bubbles based on our energy injection rate constraints.}
\subsection{Thermal Gas Masses}
\label{subsection.mass_chap_fb}
We use our density constraints to estimate the thermal gas mass within the bubble and shell structures. This is a straightforward calculation since we assume that each component has a constant density and fixed volume. Thus, the mass in each component is defined as $M = \mu m_H \times n \times V$, where $\mu = 0.61$ is the average weight per particle, $m_H$ is the mass of hydrogen, $n$ is the inferred density, and $V$ is the volume. Our geometric models imply a bubble volume of $V_{FB} = 2 \times 4/3 \times \pi \times 5 \times 3^2 = 377$ kpc$^3$ (the factor of two is for two ellipsoidal bubbles), and a combined shell volume of $V_{shell} = 411$ kpc$^3$. The densities in Table~\ref{table.inferred_prop_chap_fb} imply thermal gas masses of $M_{FB} = 4.6 \times 10^6$ $M_{\odot}$ and $M_{shell} = 6.1 \times 10^6$ $M_{\odot}$ for the bubble and shell, with a characteristic range between 5 and 10$\times 10^6$ $M_{\odot}$ given the density uncertainties. These masses represent material that has been shock-heated by the bubbles or injected into the bubbles by the energy source.
We first explore whether the bubble and shell plasmas are predominantly shocked/mixed hot halo material by comparing the masses derived above to the inferred hot gas halo mass that would exist within the bubble and shell volumes. If the density inside the bubble+shell volume was defined by our hot gas halo density model with $r_c$ = 3 kpc (instead of the Fermi bubble/shell densities), the halo mass in the volume would be $1.11 \times 10^7$ $M_{\odot}$. The calculations above indicate that the combined thermal gas mass within the bubble+shell volumes is $M_{FB}+M_{shell} = 1.07 \times 10^7$ $M_{\odot}$. The consistency between these values suggests that most of the thermal gas associated with the bubbles is shock-heated ambient material, as opposed to material injected by the energy source.
\begin{figure}[t]
\begin{center}
\includegraphics[width = .5\textwidth, keepaspectratio=true]{epsilon_withratio.pdf}
\caption[Oxygen line emissivities as a function of plasma temperature]{Oxygen line emissivities as a function of plasma temperature. The {top} panel shows the \ion{O}{7}\ (solid green) and \ion{O}{8}\ (dashed blue) line emissivities for different APEC plasma temperatures \citep{foster_etal12}. The {bottom} panel shows the emissivity ratio, which equals the total observed line ratio if the emission consists of a single plasma.
}
\label{figure.epsilon_ratio}
\end{center}
\end{figure}
As a consistency check, we estimate the amount of material injected by the energy source, either AGN or star formation, to compare with the above masses. The mass-loss rate due to nuclear star formation activity (or mass injection rate) is believed to be $\dot{M}_{inj,\ nsf} \approx 0.3 (SFR / M_{\odot}yr^{-1})$ \citep{leitherer_etal99}. Mass-loss rates from black hole accretion events (from either jets or winds) are more uncertain, but simulations of RIAF accretion winds suggest values ranging between 2\% and 20\% of $\dot{M}_{Edd} = 10L_{Edd}/c^2$ \citep{yuan_etal12, yuan_etal15}. Assuming a nuclear SFR of 0.05 $M_{\odot}$ yr$^{-1}$, $\dot{M}_{Edd} \sim 10^{-1}$ $M_{\odot}$ yr$^{-1}$ for Sgr A*, and an active period of for the bubble of 4.3 Myr, we estimate the injected mass is $M_{inj} \lesssim 10^{5}$ $M_{\odot}$ for both origin scenarios. This is significantly less than our constrained mass estimate of $\sim 10^7$ $M_{\odot}$, thus validating our claim that the bubbles contain predominantly shocked halo gas.
\subsection{Comparing with Previous Work}
\label{subsection.compare_chap_fb}
\subsubsection{Analyses at Soft X-Ray Energies}
\label{subsubsection.softxray_chap_fb}
The most direct comparisons to our analysis are the previous soft X-ray spectral analyses \citep{kataoka_etal13,kataoka_etal15,tahara_etal15}. These studies follow a similar methodology and find similar results, with \citetalias{kataoka_etal15} being the most current and comprehensive work of the three. These authors compiled a sample of 29 \textit{Suzaku} observations and 68 \textit{Swift} observations distributed across the Fermi bubbles. They fit the \textit{Suzaku} XIS data and \textit{Swift} X-ray Telescope spectra with a multi-component thermal plasma model, where one component is typically fixed at $kT = 0.1$ keV to represent emission from the LB and residual SWCX, and the other represents the combined emission from the halo and Fermi bubbles. They systematically find a hot gas halo/Fermi bubble plasma temperature of $kT = 0.3$ keV, and this is hotter than the characteristic value found for sight lines away from the bubbles \citep[$kT = 0.2$ keV; ][]{hs13}. From this temperature, they infer a relatively low Mach number of $\mathcal{M}$ = 0.3 keV / 0.2 keV = 1.5 and an expansion velocity of $v_{exp} = 300$ km s$^{-1}$. They also find emission measures that vary by over an order of magnitude, which they note is due to a combination of emission from the Galactic halo and from the Fermi bubbles. The authors model the emission measures in the northern Galactic hemisphere with a hot gas halo density model from \citet{miller_bregman13} and a Fermi bubble shell distribution with an inner radius of 3 kpc, an outer radius of 5 kpc, and a density of $3.4 \times 10^{-3}$ cm$^{-3}$.
While our approach is similar to these studies, there are several differences that can explain why we derive a higher bubble temperature and lower bubble density. The main observational difference is that these studies fit the SXRB spectra for emission measures and temperatures, while we chose to measure oxygen emission line intensities. Emission measures and temperatures are more useful plasma properties to measure, but fitting a full 0.5--2.0 keV SXRB spectrum with a thermal plasma model requires more counts than fitting only the oxygen emission lines. This is why our sample is larger and has better sky coverage than the \citetalias{kataoka_etal15} sample. However, these observables should give consistent results for the inferred bubble temperature and density. The measured plasma temperature is most sensitive to the \ion{O}{8}\ / \ion{O}{7}\ ratio for temperatures between $\approx$0.1 and 0.3 keV, because the lines are strong and rapidly changing in strength in this regime (Figure~\ref{figure.epsilon_ratio}). Thus, any temperature derived from fitting an SXRB spectrum with an APEC model should be consistent with the temperature inferred from fitting the oxygen lines separately \citep{yoshino_etal09}. The analysis becomes more complicated when there are multiple emission components, and the interpretation of multiple SXRB sources is likely the bigger difference between approaches.
The primary difference between our work and those discussed above is the treatment of combined X-ray emission from the hot gas halo and Fermi bubbles. The observed emission includes contributions from the hot halo and Fermi bubbles. Thus, the fitted plasma temperature of 0.3 keV is the emission measure-weighted temperature of the hot halo at 0.2 keV and a Fermi bubble plasma that is likely $>$0.3 keV. The \citetalias{kataoka_etal15} analysis assumes that the Fermi bubbles dominate the observed emission, while our analysis includes the combined emission from the hot halo and Fermi bubbles. This extension leads to a similar result to these previous works, but also explains why we infer a higher temperature for the Fermi bubble plasma than these studies ($kT_{FB,\ shell} \approx$ 0.4--0.5 keV).
A similar interpretation likely explains why the \citetalias{kataoka_etal15} analysis infers bubble densities 3--4 times higher than our constraints. Their Fermi bubble geometric distribution includes only a shell component that is 2 kpc thick, while ours includes both a volume-filled component and a shell component. This implies that our bubble+shell emission model has a longer path length along most sight lines near the bubbles than their model. The emission measure scales with density and path length as $EM \propto n^2 L$, so a longer inferred path length would lead to a lower inferred density. We also point out that their hot gas halo density model extends only to $r$ = 20 kpc. While the hot halo emission is likely dominated by gas within $r \lesssim$25 kpc, failing to account for emission at greater radii can decrease the amount of modeled halo emission. The combined effect is that the \citetalias{kataoka_etal15} analysis assumes shorter bubble path lengths and less halo emission than our emission model, and this results in a higher inferred bubble density required to match the total observed emission.
\subsubsection{Kinematic Estimates from UV Absorption Lines}
\label{subsubsection.uv_chap_fb}
A different approach to constrain the Fermi bubble kinematics involves analysis of UV absorption lines near background quasars. \citet{fox_etal15} observed the quasar PDS 456 ($l,b = 10.4\arcdeg, 11.2\arcdeg$) with the Cosmic Origins Spectrograph on board the $Hubble\ Space\ Telescope$. The spectrum from 1133--1778 \AA\ covers several ionic species indicative of gas with $T \sim 10^4$--$10^5$ K, including \ion{Si}{2}, \ion{Si}{3}, \ion{Si}{4}, \ion{C}{2}, \ion{C}{4}, and \ion{N}{5}. They detected multiple absorption components for each species, but they argue the nearly symmetric components at $v_{LSR}$ = -235 km s$^{-1}$ and +250 km s$^{-1}$ are unlikely to come from absorbers in the disk or farther in the halo. If these absorbers represent gas entrained near the bubble edges, their velocities can be used to constrain the bubble kinematics. Indeed, the authors apply a Galactic wind model from \citet{bordoloi_etal14} to simulate $v_{LSR}$ absorbers and find that an intrinsic outflow velocity of $\geq$900 km s$^{-1}$ is required to reproduce the observed absorption features.
There is tension between these results and our lower inferred expansion rate of $\approx$500 km s$^{-1}$. Although we infer a higher expansion rate than \citet{kataoka_etal15}, they discuss this discrepancy as well. The outflow model used by \citet{fox_etal15} has two important parameters---the outflow velocity and the opening angle. They assume an opening angle of 110$\arcdeg$ to match the hard X-ray arcs seen by \citet{bh_cohen03}. However, this geometry produces a significant correction between the intrinsic outflow velocity and $v_{LSR}$ at low latitudes. Their model implies that most of the bubble velocity at $l,b \approx 10\arcdeg, 10\arcdeg$ is tangential to the line of sight, which may not be the case. If instead the bubbles have a rounder surface at lower $z$ or a stronger outflow velocity vector away from the Galaxy's polar axis, a lower intrinsic velocity could reproduce the observed absorption. Thus, the unknown intrinsic bubble geometry plausibly accounts for the different expansion velocities inferred from these two methodologies.
\subsubsection{Comparing with Simulations}
\label{subsubsection.sims_chap_fb}
The Fermi bubbles have motivated numerous simulations of Galactic outflows since their discovery. Typically, these studies primarily focus on the gamma-ray source, which is tied to the underlying cosmic-ray composition (leptonic or hadronic) and where the cosmic rays are produced (injected from the central source, accelerated in situ, etc.). All of these simulations predict various distributions for the non-thermal and thermal gas within the bubbles, but information on the latter is often not discussed in detail. This limits our comparison to characteristic densities, velocities, and energetics, although our results are initial steps toward constraining these properties.
Simulations are also generally segregated by the assumed energy source, either a black hole accretion event or nuclear star formation. There is much variation with the assumed outflow parameters, but black hole accretion simulations tend to be more energetic on shorter time scales than star formation simulations. For example, simulations producing the bubbles with AGN jets have characteristic total energy injection rates and ages of $\gtrsim 10^{44}$ erg s$^{-1}$ and $\approx$1--3 Myr \citep[e.g., ][]{guo_mathews12a,guo_mathews12b,yang_etal12,yang_etal13}, whereas simulations producing the bubbles from weaker AGN winds suggest values of $10^{41}$--$10^{42}$ erg s$^{-1}$ and 5--10 Myr \citep[e.g., ][]{mou_etal14,mou_etal15}. Alternatively, nuclear star formation simulations can reproduce the bubble morphology with energy injection rates of $\approx$(1--5)$\times 10^{40}$ erg s$^{-1}$ over $\gtrsim$50 Myr timescales \citep[e.g., ][]{crocker_etal14,crocker_etal15,sarkar_etal15}. The AGNs simulations also tend to predict stronger outflow velocities than the star formation simulations ($\gtrsim$1000 km s$^{-1}$ compared to $\lesssim$500 km s$^{-1}$).
Our inferred energy injection rate, bubble age, and expansion velocity are most consistent with the weaker black hole accretion simulations, where the bubbles are inflated by an AGN wind \citep{mou_etal14,mou_etal15}. Simulations of AGN jets predict higher energy input rates than our results, while star formation simulations are typically weaker and over a much longer timescale than our constraints. It is difficult to make stronger claims at this point since these simulations are subject to a number of uncertainties. For example, the energy injection rate required to match the bubble morphology is degenerate with the density of the surrounding medium since it opposes the ram pressure from the galactic wind. Most simulations assume an ambient density comparable to our core density ($\sim 10^{-3}$ cm$^{-3}$), but this is a well-documented degeneracy in the simulations. Regardless of these limitations, our constraints should be used to motivate future simulations designed to analyze the bubbles.
Our thermal pressure constraints also address the bubbles' cosmic-ray composition and whether thermal or non-thermal pressure drives the bubbles' expansion. Simulations can produce the bubbles' gamma-ray and microwave emission by accelerating either leptonic or hadronic cosmic rays, leading to uncertainties in the inferred non-thermal pressure (cosmic rays and magnetic fields). For example, the leptonic AGN jet simulations from \citet{yang_etal13} predict a total pressure inside the bubbles of $\sim 10^{-10}$ dyn cm$^{-2}$ and a cosmic ray pressure of $\sim 10^{-12}$ dyn cm$^{-2}$. This implies either that the bubbles are dominated by a thermal gas pressure much larger than our estimates (magnetic pressure is negligible) or that an additional hadronic cosmic ray source could contribute most of the pressure. The former scenario is consistent with nuclear star formation simulations that accelerate cosmic-ray leptons \citep{sarkar_etal15} or hadrons \citep{crocker_etal15}. Alternatively, limits from hard X-ray spectra near the bubbles imply a cosmic ray electron and magnetic pressure of $\approx 2 \times 10^{-12}$ dyn cm$^{-2}$ \citep{kataoka_etal13}, which is approximately equal to the \citepalias{kataoka_etal15} thermal pressure estimate. Our characteristic thermal gas pressure of (5--20)$\times 10^{-13}$ dyn cm$^{-2}$ should be used in future modeling work to build a more accurate census of the bubbles' energy and pressure budget.
\section{Conclusions}
\label{section.conclusions_chap_fb}
This work is a comprehensive observational analysis of the Fermi bubbles at soft X-ray energies. The \ion{O}{7}\ and \ion{O}{8}\ emission line sample includes data from \textit{XMM-Newton} and \textit{Suzaku}, with 741 sight lines in total and $\sim 100$ sight lines projected near the Fermi bubbles. The new \textit{Suzaku} measurements were processed in a similar way to the \textit{XMM-Newton} measurements, making this the largest emission line sample designed to probe Galactic-scale hot gas distributions.
We used this sample to model the Fermi bubbles' thermal gas emission, resulting in improved constraints on the bubbles' physical properties and their role in the Milky Way's evolution. Our modeling procedure is similar to previous studies at soft X-ray energies \citep{kataoka_etal13,kataoka_etal15,tahara_etal15}, although we model the combined emission from the hot halo and Fermi bubbles simultaneously. This extension confirms the result that the bubbles are hotter than the surrounding medium and expanding supersonically, and leads to a stronger shock than previous works. Thus, these are improved constraints on the Fermi bubbles' thermal gas distribution given the data currently available.
We summarize our primary conclusions and inferred bubble properties:
\begin{enumerate}
\item{The observed \ion{O}{8}\ /\ion{O}{7}\ ratios are systematically larger for sight lines near the bubbles, suggesting the presence of a plasma with $T > 2 \times 10^6$ K. }
\item{Our best-fit parametric model implies $n_{FB} = 8.2 \pm 0.2 \times 10^{-4}$ cm$^{-3}$, $n_{shell} = 10.0 \pm 0.3 \times 10^{-4}$ cm$^{-3}$, log($T_{FB}$) = 6.60--6.65, and log($T_{shell}$) = 6.60--6.95 with an optimal value of 6.70. This involves explicitly fitting the \ion{O}{8}\ line intensities and analyzing the \ion{O}{8}\ /\ion{O}{7}\ ratio distribution near the bubbles. }
\item{These densities imply thermal gas masses within the bubble and shell volumes of $M_{FB}$ = 4.6--5.0$\times 10^6$ $M_{\odot}$ and $M_{shell}$ = 5.2--9.8$\times 10^6$ $M_{\odot}$. We interpret this as predominantly shock-heated hot gas halo material. }
\item{The inferred bubble/shell temperature ($5 \times 10^6$ K) compared to ambient halo gas temperature ($2 \times 10^6$ K) suggests a shock Mach number of $\mathcal{M} = 2.3_{-0.4}^{+1.1}$ and expansion rate of $v_{exp} = 490_{-77}^{+230}$ km s$^{-1}$. These are larger than the values suggested from other soft X-ray modeling analyses \citepalias{kataoka_etal15}, and smaller than the value suggested by the UV absorption line analysis by \citet{fox_etal15}. The differences are likely explained by geometric assumptions for the latter and modeling the hot gas halo emission for the former. }
\item{Treating the bubbles as a galactic outflow with Sedov-Taylor expansion solutions leads to an inferred energy injection rate of $2.3_{-0.9}^{+5.1} \times 10^{42}$ erg s$^{-1}$ and age of $4.3_{-1.4}^{+0.8}$ Myr. These energetics and timescales suggest that the bubbles likely formed from a Sgr A* accretion episode, as opposed to sustained nuclear star formation activity. }
\item{Our results are broadly consistent with predictions from MHD simulations of galactic outflows. The constrained energy injection rate and age are most consistent with simulations that generate the bubbles from a relatively weak AGN wind \citep{mou_etal14,mou_etal15}. }
\end{enumerate}
This analysis is an initial effort to constrain the Fermi bubbles' thermal gas structure using soft X-ray observations, but it should also motivate future observational and theoretical studies. Future analyses using additional spectral data or all-sky maps from MAXI or eROSITA will help probe the bubbles' structure and interaction with the surrounding medium. The results should also motivate future simulations that predict characteristic bubble densities, temperatures, pressures, and expansion rates.
\acknowledgments
We thank Meng Su, H.-Y. Karen Yang, Mateusz Ruszkowski, and the anonymous referee for their insightful comments that improved our manuscript. We also acknowledge the NASA Astrophysics Data Analysis Program grants NNX16AF23G and NNX11AJ55G for funding this work.
\def\apjl{{ApJL}}
\bibliographystyle{apj}
| 4ce703bf7a9ac87dcda1aab718f545b619f6e0d0 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
It is well known that a degree $n$ Blaschke product is an $n$-to-$1$ (counting multiplicity) analytic self-map of the unit disk $\mathbb{D}$, and moreover that every such analytic $n$-to-$1$ self-map of the unit disk is a degree $n$ Blaschke product. We can restate this in a slightly more general setting. Throughout this paper we let $G\subset\mathbb{C}$ denote a simply connected domain, and we let $\mathcal{S}\subset G$ be a smooth Jordan curve, with bounded face $D$.
\begin{theorem}\label{thm: Blaschke prod. version.}
Let $f:G\to\mathbb{C}$ be analytic. The following are equivalent.
\begin{enumerate}
\item $\mathcal{S}$ is a lemniscate of $f$ in $G$.
\item For some positive integer $n$, $f$ maps $D$ precisely $n$-to-$1$ onto $\mathbb{D}$.
\end{enumerate}
If Items $1$ and $2$ hold, the function $f\circ\phi$ is a degree $n$ Blaschke product, where $\phi:\mathbb{D}\to D$ is any Riemann map.
\end{theorem}
The purpose of this note is to generalize the result of Theorem~\ref{thm: Blaschke prod. version.} to the context of meromorphic functions and pseudo-lemniscates. In order to state the theorem, we begin with several definitions. For the remainder of the paper we let $\Gamma$ denote some smooth Jordan curve in $\mathbb{C}$, with bounded face $\Omega_-$ and unbounded face $\Omega_+$ (where $\Omega_+$ includes the point $\infty$).
\begin{definition}\
\begin{itemize}
\item Let $f:G\to\hat{\mathbb{C}}$ be meromorphic. Suppose that $\mathcal{S}$ is a component of $f^{-1}(\Gamma)$. We call $\mathcal{S}$ a $\Gamma$-\textit{pseudo-lemniscate} of $f$.
\item For any $w\in\hat{\mathbb{C}}$, let $\mathcal{N}_f(w)$ denote the number of preimages under $f$ of $w$ in $D$, counted with multiplicity.
\end{itemize}
\end{definition}
In the special case that $\Gamma=\mathbb{T}$, the above definition reduces to the classical definition of a lemniscate. The definition of a pseudo-lemniscate was introduced in~\cite{RY} wherein conformal equivalence of functions on domains bounded by pseudo-lemniscates was studied. We now proceed to the promised generalization of Theorem~\ref{thm: Blaschke prod. version.}.
\begin{theorem}\label{thm: Main.}
Let $f:G\to\hat{\mathbb{C}}$ be meromorphic. The following are equivalent.
\begin{enumerate}
\item $\mathcal{S}$ is a $\Gamma$-pseudo-lemniscate of $f$ in $G$.
\item For some non-negative integers $n_-$ and $n_+$, the following holds.
\begin{itemize}
\item For every $w\in\Omega_-$, $\mathcal{N}_f(w)=n_-$.
\item For every $w\in\Omega_+$, $\mathcal{N}_f(w)=n_+$.
\item For every $w\in\Gamma$, $\mathcal{N}_f(w)=\min(n_-,n_+)$.
\end{itemize}
\end{enumerate}
If Items $1$ and $2$ hold, and $f$ is analytic in $D$, then $n_+=0$, and the function $\psi^{-1}\circ f\circ\phi$ is a degree $n_-$ Blaschke product, where $\phi:\mathbb{D}\to D$ and $\psi:\mathbb{D}\to\Omega_-$ are any Riemann maps.
If Items $1$ and $2$ hold, and $\Gamma=\mathbb{T}$, then $f\circ\phi$ is a ratio of Blaschke products (degree $n_-$ in the numerator and degree $n_+$ in the denominator), where $\phi:\mathbb{D}\to D$ is any Riemann map.
\end{theorem}
Using the fact that any meromorphic function is conformal away from its critical points, we may use Theorem~\ref{thm: Main.} to supply a test for when the image of $\mathcal{S}$ under $f$ will not be a Jordan curve.
\begin{corollary}\label{cor: Non-Jordan curve or crit. point.}
Let $f:G\to\hat{\mathbb{C}}$ be meromorphic. If there are three points $w_1,w_2,w_3\in\hat{\mathbb{C}}$ such that the numbers $\mathcal{N}_f(w_1)$, $\mathcal{N}_f(w_2)$, and $\mathcal{N}_f(w_3)$ are distinct, then either $f$ has a critical point on $\mathcal{S}$ or $f(\mathcal{S})$ is not a Jordan curve.
\end{corollary}
In Section~\ref{sect: Lemmas and proof.} we will provide proofs of Theorem~\ref{thm: Main.} and Corollary~\ref{cor: Non-Jordan curve or crit. point.}. In Section~\ref{sect: Example.} we apply Corollary~\ref{cor: Non-Jordan curve or crit. point.} to show that the image of a certain smooth Jordan curve $\mathcal{S}$ under a certain meromorphic function $f$ is not a Jordan curve, without explicitly finding a point of intersection in $f(\mathcal{S})$.
\section{Lemmas and Proofs}\label{sect: Lemmas and proof.}
We begin with a topological observation on the convergence of preimages under continuous maps. First we introduce a bit of standard notation.
\begin{definition}
For any $x\in\mathbb{C}$, $X\subset\mathbb{C}$, and $\delta>0$, we define $$B(x;\delta)=\{y\in\mathbb{C}:|x-y|<\delta\}$$ and $$B(X;\delta)=\displaystyle\bigcup_{z\in X}B(z;\delta).$$
\end{definition}
\begin{lemma}\label{lem: Convergence of preimages.}
Let $U\subset\mathbb{C}$ be an open bounded set, and let $h:U\to\hat{\mathbb{C}}$ be continuous. Choose some $w_0\in\mathbb{C}$. Then for any $\delta>0$ there exists an $\epsilon>0$ such that if $w\in B(w_0;\epsilon)$, then $h^{-1}(w)\subset B(h^{-1}(w_0)\cup\partial U;\delta)$.
\end{lemma}
\begin{proof}
This is an elementary exercise in compactness and continuity.
\end{proof}
Our second lemma shows that, under the conditions of Item~2 from Theorem~\ref{thm: Main.}, if $E$ is a component of $f^{-1}(\Omega_-)\cap D$, and $z\in\partial E$, then $f(z)\in\partial\Omega_-$.
\begin{lemma}\label{lem: Full copies.}
Let $h:G\to\hat{\mathbb{C}}$ be meromorphic. Let $\Omega\subset\mathbb{C}$ be open, bounded, and connected, and assume that the quantity $\mathcal{N}_h(w)$ is independent of the choice of $w\in\Omega$. Then if $E\subset D$ is a component of the set $h^{-1}(\Omega)\cap D$, and $z_0\in\partial E$, then $h(z_0)\in\partial\Omega$.
\end{lemma}
\begin{proof}
Let $h$, $\Omega$, and $E$ be as in the statement of the lemma. Choose some $z_0\in\partial E$, and set $w_0=h(z_0)$. Suppose by way of contradiction that $w_0\in\Omega$. We proceed by cases.
\begin{case}$z_0\in D$.\end{case}
Since $h$ is continuous and $\Omega$ and $D$ are open, there is a neighborhood of $z_0$ contained in $h^{-1}(\Omega)\cap D$, contradicting our choice of $z_0$.
\begin{case}$z_0\notin D$.\end{case}
Set $m=\mathcal{N}_h(w_0)$, and let $z_1,z_2,\ldots,z_k$ be the distinct preimages of $w_0$ in $D$, having multiplicities $m_1,m_2,\ldots,m_k$ (so that $m=m_1+m_2+\cdots+m_k$). Choose some $\delta>0$ such that $\delta<\min(|z_i-z_0|:1\leq i\leq k)$. Choose disjoint neighborhoods $D_1,D_2,\ldots,D_k\subset D$ such that for each $1\leq i\leq k$, the following holds.
\begin{itemize}
\item $z_i\in D_i$.
\item $D_i\cap B(z_0;\delta)=\emptyset$.
\item $h$ is exactly $m_i$-to-$1$ on $D_i\setminus\{z_i\}$.
\end{itemize}
Choose some point $x\in B(z_0;\delta)\cap E$ sufficiently close to $z_0$ so that $$h(x)\in\displaystyle\bigcap_{i=1}^kh(D_i).$$ Then $h^{-1}(h(x))\cap D$ contains $x$, along with $m_i$ distinct points in $D_i$, for each $1\leq i\leq k$. Thus since $x\notin D_i$ for any $1\leq i\leq k$, we have that $\mathcal{N}_h(h(x))$ is at least $$m_1+m_2+\ldots+m_k+1>m.$$ Since $x\in E$, $h(x)\in\Omega$, so we have a contradiction of the choice of $m$.
We therefore conclude that $w_0\notin\Omega$. Since $h$ is continuous we must have that $w_0\in\overline{\Omega}$, so we conclude that $w_0\in\partial\Omega$.
\end{proof}
With these lemmas under our belt, we proceed to the proof of Theorem~\ref{thm: Main.}.
\begin{proof}[Proof of Theorem~\ref{thm: Main.}.]
We begin by assuming that $\mathcal{S}$ is a $\Gamma$-pseudo-lemniscate of $f$ in $G$.
Fix now some component $E$ of $f^{-1}(\Omega_-)$ in $D$. We wish to show that there is some positive integer $n_E$ such that $f$ maps $E$ exactly $n_E$-to-$1$ onto $\Omega_-$ (counting multiplicity).
Let $\Lambda_1,\Lambda_2,\ldots,\Lambda_k$ denote the components of $\partial E$, with $\Lambda_1$ being the component of $\partial E$ which is in the unbounded component of $E^c$. Since $f$ is continuous and open, $f$ maps each $\Lambda_i$ to the boundary of $\Omega_-$, namely to $\Gamma$. Therefore for each $i$, we define $n_i$ to be the number of times $f(z)$ winds around $\Gamma$ as $z$ traverses $\Lambda_i$ one time with positive orientation.
By the Riemann mapping theorem, for any point $w\in\Omega_-$, we can find some conformal map $\rho:\Omega_-\to\mathbb{D}$ such that $\rho(w)=0$, and since the boundary of $\Omega_-$ is smooth, each such $\rho$ extends conformally to a neighborhood of $\overline{\Omega_-}$. Therefore, replacing $f$ with $\rho\circ f$, the components of $\partial E$ are honest lemniscates of $\rho\circ f$. Since $\rho$ is a conformal map on the closure of $\Omega_-$, if $f(z)$ winds once around $\Gamma$, $\rho\circ f(z)$ winds once around the unit circle, and with the same orientation (since $\Omega_-$ is simply connected and $\rho$ is analytic, $\rho$ cannot reverse the orientation).
Therefore it follows from the argument principle that the number of zeros of $\rho\circ f$ in $E$ (counted with multiplicity) is
$$n_E:=n_1-(n_2+n_3+\cdots+n_k).$$
That is, the number of members of $f^{-1}(w)$ in $E$ (counted with multiplicity) is $n_E$. Since this number $n_E$ is independent of the choice of $w\in\Omega_-$, we conclude that $f$ maps $E$ precisely $n_E$-to-$1$ onto $\Omega_-$. Therefore for any $w\in\Omega_-$, $$\mathcal{N}_f(w)=\displaystyle\sum_E n_E.$$
This sum is taken over all components $E$ of $f^{-1}(\Omega_-)$ in $D$, a quantity which again is independent of the choice of $w\in\Omega_-$. We call this sum $n_-$. Doing the same calculation with $\Omega_-$ replaced by $\Omega_+$, we obtain the corresponding number $n_+$.
We now consider the set $f^{-1}(\Gamma)\cap D$. Since $\mathcal{S}$ is a connected component of $f^{-1}(\Gamma)$ and $\mathcal{S}$ is compact, and $f^{-1}(\Gamma)$ is closed in $G$, it follows that $f^{-1}(\Gamma)\setminus\mathcal{S}$ is isolated from $\mathcal{S}$ by some positive distance. Let $\delta>0$ be chosen to be less than half that distance. That is, $$\displaystyle\inf_{x\in\mathcal{S},y\in f^{-1}(\Gamma)\setminus\mathcal{S}}|x-y|>2\delta.$$
Since $D$ is simply connected, the set $D_\delta:=\left\{z\in D:d(z,\mathcal{S})<\delta\right\}$ is connected (where $d(z,\mathcal{S})$ denotes the distance $\displaystyle\inf_{x\in\mathcal{S}}|z-x|$). Therefore since $f$ is continuous, and $f^{-1}(\Gamma)$ does not intersect $D_\delta$, we conclude that either $f(D_\delta)\subset\Omega_-$ or $f(D_\delta)\subset\Omega_+$. Without loss of generality assume that $f(D_\delta)\subset\Omega_-$.
Choose now some $w_0$ in $\Gamma$. Let $z_1,z_2,\ldots,z_l\in D$ denote the distinct preimages of $w_0$ in $D$, having multiplicities $m_1,m_2,\ldots,m_l$. Choose disjoint domains $D_1,D_2,\ldots,D_l\subset D$ such that the following holds for each $i$.
\begin{itemize}
\item $z_i\in D_i$.
\item $D_i\cap D_\delta=\emptyset$.
\item $f$ has no critical points in $D_i\setminus\{z_i\}$.
\item $f$ is exactly $m_i$-to-$1$ in $D_i\setminus\{z_i\}$.
\end{itemize}
Reduce now $\delta>0$ if necessary so that for each $i$, $B(z_i;\delta)\subset D_i$. By Lemma~\ref{lem: Convergence of preimages.} we can find an $\epsilon>0$ such that if $w\in\Omega_+$ is within $\epsilon$ of $w_0$, then $$f^{-1}(w)\cap D\subset B((f^{-1}(w_0)\cap D)\cup\partial D;\delta)\cap D.$$ Choose $\epsilon>0$ smaller if necessary so that for each $i$, $B(w_0;\epsilon)\subset f(D_i)$.
Fix now some $w\in B(w_0;\epsilon)\cap\Omega_+$. Since $w\in\Omega_+$, and $f(D_\delta)\subset\Omega_-$, no point in $f^{-1}(w)\cap D$ is within $\delta$ of $\partial D$. Therefore by choice of $\epsilon$ we have $$f^{-1}(w)\cap D\subset B(f^{-1}(w_0)\cap D;\delta)\subset\displaystyle\bigcup_{i=1}^lD_i.$$
Now, for each $i$, since $B(w_0;\epsilon)\subset f(D_i)$, $w\in f(D_i)$. Since $f$ is exactly $m_i$-to-$1$ in $D_i\setminus\{z_i\}$, there are exactly $m_i$ distinct points in $f^{-1}(w)\cap D_i$, and since none of these points are critical points, each one contributes exactly $1$ to the total $n_+$ many elements in $f^{-1}(w)\cap D$. We therefore conclude that $\mathcal{N}_f(w_0)=n_+$.
Note that if we had selected the point $w$ to be a point in $\Omega_-$ instead of in $\Omega_+$, we would have obtained the same result, except that some members of $f^{-1}(w)\cap D$ could additionally reside in $D_\delta$. Our conclusion would then be that the $\mathcal{N}_f(w_0)\leq n_-$. Since $\mathcal{N}_f(w_0)=n_+$ and $\mathcal{N}_f(w_0)\leq n_-$, we therefore have that $$\mathcal{N}_f(w_0)=\min(n_-,n_+).$$
If we had adopted the other assumption, that $f(D_\delta)\subset\Omega_+$, then similar work would give the same result. This concludes the first direction of our proof.
Suppose now that the second item holds. That is, that there are some non-negative integers $n_-$ and $n_+$ for which the following holds.
\begin{itemize}
\item For every $w\in\Omega_-$, $\mathcal{N}_f(w)=n_-$.
\item For every $w\in\Omega_+$, $\mathcal{N}_f(w)=n_+$.
\item For every $w\in\Gamma$, $\mathcal{N}_f(w)=\min(n_-,n_+)$.
\end{itemize}
It follows from Lemma~\ref{lem: Full copies.} that if $E$ is any component of $f^{-1}(\Omega_-)\cap D$ or of $f^{-1}(\Omega_+)\cap D$, then $f(\partial E)\subset\Gamma$. We therefore conclude that $f(\mathcal{S})=f(\partial D)\subset \Gamma$. Our goal is to show that $\mathcal{S}$ itself is an entire component of $f^{-1}(\Gamma)$ in $G$.
For any point $z\in\mathcal{S}$, since $f$ is analytic at $z$ and $\Gamma$ is smooth, if $z$ is not a critical point of $f$ then $f^{-1}(\Gamma)$ is a smooth curve close to $z$. On the other hand, if $z$ is a critical point of $f$ with multiplicity $m$ then, close to $z$, $f^{-1}(\Gamma)$ consists of $2(m+1)$ paths meeting at $z$, forming equal angles $\dfrac{2\pi}{2(m+1)}$ around $z$. Since $\mathcal{S}$ is smooth, it therefore suffices to show that our assumption on $f$ implies that $f$ does not have any critical points on $\mathcal{S}$.
Suppose by way of contradiction that $z_0\in\mathcal{S}$ is a critical point of $f$ with multiplicity $m$. Let $\Lambda$ denote the component of $f^{-1}(\Gamma)$ which contains $\mathcal{S}$. Since $m\geq1$, $$\dfrac{2\pi}{2(m+1)}=\dfrac{\pi}{m+1}<\pi.$$ Therefore since $\mathcal{S}$ is smooth, there are edges of $\Lambda$ which extend from $z_0$ into either face of $\mathcal{S}$, and in particular into $D$. Let $\gamma$ denote some such edge of $\Lambda$ emanating from $z_0$ into $D$. Assume without loss of generality that as $z$ traverses $\gamma$, $f(z)$ traverses $\Gamma$ with positive orientation. Then for any point $w\in\Gamma$ taken arbitrarily close to $f(z_0)$ (with positive orientation around $\Gamma$ from $f(z_0)$), there is some point $z\in D$ (on $\gamma$) which is arbitrarily close to $z_0$, such that $f(z)=w$. A contradiction now follows from exactly the same reasoning as was used to prove Lemma~\ref{lem: Full copies.}.
We therefore conclude that $\mathcal{S}$ contains no critical point of $f$, and thus that the component of $f^{-1}(\Gamma)$ which contains $\mathcal{S}$ is just $\mathcal{S}$ itself. That is, $\mathcal{S}$ is a $\Gamma$-pseudo-lemniscate of $f$ in $G$. This concludes the second direction of the proof.
Now suppose that Items~1~and~2 in the theorem do obtain, and that moreover $f$ is analytic in $D$. Since $f^{-1}(\infty)\cap D=\emptyset$, immediately we have that $n_+=0$. If $\phi:\mathbb{D}\to D$ and $\psi:\mathbb{D}\to\Omega_-$ are Riemann maps, then $\psi^{-1}\circ f\circ\phi$ is a $n_-$-to-$1$ analytic self map of $\mathbb{D}$, which by Theorem~\ref{thm: Blaschke prod. version.} is a degree~$n_-$ Blaschke product.
Finally, suppose that Items~1~and~2 in the theorem do obtain, and that $\Gamma=\mathbb{T}$. Let $\phi:\mathbb{D}\to D$ be any Riemann map. Since $\partial D$ is smooth, $\phi$ extends conformally to a neighborhood of the closed unit disk. Since $f$ satisfies Item~2, $f\circ\phi$ has $n_-$ zeros and $n_+$ poles in $\mathbb{D}$, and since $f$ satisfies Item~1, $|f\circ\phi|=1$ on $\mathbb{T}$. Let $A$ denote a degree $n_-$ Blaschke product having the same zeros as $f\circ\phi$, with the same multiplicities, and let $B$ denote a degree $n_+$ Blaschke product, having zeros exactly at the locations of the poles of $f\circ\phi$, with the same multiplicities. Then $(f\circ\phi)\cdot\frac{B}{A}$ has no zeros or poles in $\mathbb{D}$ and on $\partial D$, $$\left|(f\circ\phi)\cdot\frac{B}{A}\right|=|f\circ\phi|\cdot\frac{|B|}{|A|}=1.$$ Therefore by the maximum/minimum modulus theorems, $(f\circ\phi)\cdot\frac{B}{A}$ is a unimodular constant $\lambda$. Replacing $A$ by $\lambda\cdot A$, we have $f\circ\phi=\frac{A}{B}$.
\end{proof}
\begin{proof}[Proof of Corollary~\ref{cor: Non-Jordan curve or crit. point.}.]
Suppose that $f$, $w_1$, $w_2$, and $w_3$ are as in the statement of the lemma. Suppose that $f(\mathcal{S})$ is a Jordan curve. Setting $\Gamma=f(\mathcal{S})$, since $\mathcal{N}_f(w_1)$, $\mathcal{N}_f(w_2)$, and $\mathcal{N}_f(w_3)$ are distinct, $\mathcal{S}$ cannot be a $\Gamma$-pseudo-lemniscate of $f$ by Theorem~\ref{thm: Main.}. Let $\Lambda$ denote the component of $f^{-1}(\Gamma)$ in $G$ which contains $\mathcal{S}$. Since $\Lambda\neq\mathcal{S}$, $\Lambda$ must branch off from $\mathcal{S}$ at some point, and thus $f$ must not be conformal at that point. That is, $f$ must have a critical point on $\mathcal{S}$.
\end{proof}
\section{Application of Corollary~\ref{cor: Non-Jordan curve or crit. point.}}\label{sect: Example.}
\textbf{Example.} Let $\mathcal{S}$ denote some smooth curve in $\mathbb{C}$ which contains the quadrilateral with vertices $0$, $1$, $2i$, and $1+4i$, and which approximates the boundary of this shape closely. Let $D$ denote the bounded face of $\mathcal{S}$. Then setting $f(z)=e^z$, we claim that $f(\mathcal{S})$ is not a Jordan curve.
To see this, observe that for $w_1=e^0i$, $w_2=e^1i$, and $w_3=e^2i$, we have $\mathcal{N}_f(w_1)=1$, $\mathcal{N}_f(w_2)=2$, and $\mathcal{N}_f(w_3)=0$. Therefore Corollary~\ref{cor: Non-Jordan curve or crit. point.} tells us that either $f$ has a critical point on $\mathcal{S}$ or $f(\mathcal{S})$ is not a Jordan curve. Since $f(z)=e^z$ has no critical points, we conclude that $f(\mathcal{S})$ is not a Jordan curve.
\bibliographystyle{amsplain}
| b3e79cce8bd647c4f9a6c1996750b997cae71cc9 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Multi-threading}
We measure the performance of the obtained word vectors on the word similarity task, when the number of threads used for training change.
In Table~\ref{tab:threads} we report the correlation with human judgement for models trained on \texttt{small} for threads ranging from 1 to 32.
In Table~\ref{tab:threads-large} we report the correlation with human judgement for models trained on \texttt{large} for threads ranging from 15 to 30.
\begin{table}
\centering
\begin{tabular}{rcccccc}
\toprule
& 1 & 2 & 4 & 8 & 16 & 32 \\
\midrule
de-gur350 & 68 & 67 & 69 & 67 & 68 & 69 \\
de-gur65 & 70 & 62 & 70 & 69 & 67 & 68 \\
de-zg222 & 49 & 41 & 43 & 44 & 45 & 48 \\
en-rw & 46 & 47 & 47 & 47 & 46 & 46 \\
en-ws353 & 62 & 63 & 63 & 64 & 63 & 64 \\
es-ws353 & 48 & 48 & 50 & 49 & 46 & 49 \\
fr-rg65 & 57 & 57 & 61 & 58 & 55 & 59 \\
\bottomrule
\end{tabular}
\caption{
Correlation between human judgement and our similarity for models trained on small with various number of threads.
}
\label{tab:threads}
\end{table}
\begin{table}
\centering
\begin{tabular}{rcccccc}
\toprule
& 15 & 20 & 25 & 30 \\
\midrule
de-gur350 & 68 & 68 & 67 & 68 \\
de-gur65 & 72 & 77 & 74 & 75 \\
de-zg222 & 41 & 40 & 44 & 43 \\
en-rw & 46 & 46 & 46 & 46 \\
en-ws353 & 65 & 66 & 67 & 66 \\
es-ws353 & 53 & 53 & 51 & 53 \\
fr-rg65 & 56 & 58 & 55 & 55 \\
\bottomrule
\end{tabular}
\caption{
Correlation between human judgement and our similarity for models trained on large with various number of threads.
}
\label{tab:threads-large}
\end{table}
\section{Experimental setup}
\subsection{Baseline}
In most experiments (except in Sec.~\ref{sec:sota}), we compare our model to the C implementation of the \texttt{skipgram} and \texttt{cbow} models from the \texttt{word2vec}\footnote{\smaller\relax\url{https://code.google.com/archive/p/word2vec}} package.
\subsection{Optimization}
We solve our optimization problem by performing stochastic gradient descent on the negative log likelihood presented before.
As in the baseline \texttt{skipgram} model, we use a linear decay of the step size.
Given a training set containing $T$ words and a number of passes over the data equal to $P$, the step size at time $t$ is equal to $\gamma_0 (1 - \frac{t}{TP})$, where $\gamma_0$ is a fixed parameter.
We carry out the optimization in parallel, by resorting to Hogwild~\cite{recht2011hogwild}.
All threads share parameters and update vectors in an asynchronous manner.
\subsection{Implementation details}
For both our model and the baseline experiments, we use the following parameters: the word vectors have dimension $300$.
For each positive example, we sample $5$ negatives at random, with probability proportional to the square root of the uni-gram frequency.
We use a context window of size $c$, and uniformly sample the size $c$ between $1$ and $5$.
In order to subsample the most frequent words, we use a rejection threshold of $10^{-4}$ (for more details, see \cite{mikolov2013distributed}).
When building the word dictionary, we keep the words that appear at least $5$ times in the training set.
The step size $\gamma_0$ is set to $0.025$ for the \texttt{skipgram} baseline and to $0.05$ for both our model and the \texttt{cbow} baseline.
These are the default values in the \texttt{word2vec} package and work well for our model too.
Using this setting on English data, our model with character $n$-grams is approximately $1.5 \times$ slower to train than the \texttt{skipgram} baseline.
Indeed, we process $105$k words/second/thread versus $145$k words/second/thread for the baseline.
Our model is implemented in C++, and is publicly available.\footnote{\smaller\relax\url{https://github.com/facebookresearch/fastText}}
\subsection{Datasets}
Except for the comparison to previous work~(Sec.~\ref{sec:sota}), we train our models on Wikipedia data.\footnote{\smaller\relax\url{https://dumps.wikimedia.org}}
We downloaded Wikipedia dumps in nine languages: Arabic, Czech, German, English, Spanish, French, Italian, Romanian and Russian.
We normalize the raw Wikipedia data using Matt Mahoney's pre-processing perl script.\footnote{\smaller\relax\url{http://mattmahoney.net/dc/textdata}}
All the datasets are shuffled, and we train our models by doing five passes over them.
\section{Results}
We evaluate our model in five experiments: an evaluation of word similarity and word analogies, a comparison to state-of-the-art methods, an analysis of the effect of the size of training data and of the size of character $n$-grams that we consider.
We will describe these experiments in detail in the following sections.
\subsection{Human similarity judgement}
\label{sec:wordsim}
We first evaluate the quality of our representations on the task of word similarity / relatedness.
We do so by computing Spearman's rank correlation coefficient~\cite{spearman04proof} between human judgement and the cosine similarity between the vector representations.
For German, we compare the different models on three datasets: \textsc{Gur65}, \textsc{Gur350} and \textsc{ZG222}~\cite{gurevych2005using,zesch2006automatically}.
For English, we use the \textsc{WS353} dataset introduced by \newcite{finkelstein2001placing} and the rare word dataset (\textsc{RW}), introduced by \newcite{luong2013better}.
We evaluate the French word vectors on the translated dataset \textsc{RG65}~\cite{joubarne2011comparison}.
Spanish, Arabic and Romanian word vectors are evaluated using the datasets described in~\cite{hassan2009cross}.
Russian word vectors are evaluated using the \textsc{HJ} dataset introduced by \newcite{panchenko2016human}.
\begin{table}[t]
\centering
\begin{tabular}{@{}rrcccc@{}}
\toprule
&& \texttt{sg} & \texttt{cbow} & \texttt{sisg-} & \texttt{sisg} \\
\midrule
\textsc{Ar} & \textsc{WS353} & 51 & 52 & 54 & \textbf{55} \\
\midrule
\multirow{ 3}{*}{\textsc{De}} & \textsc{Gur350} & 61 & 62 & 64 & \textbf{70} \\
& \textsc{Gur65} & 78 & 78 & \textbf{81} & \textbf{81} \\
& \textsc{ZG222} & 35 & 38 & 41 & \textbf{44} \\
\midrule
\multirow{ 2}{*}{\textsc{En}} & \textsc{RW} & 43 & 43 & 46 & \textbf{47} \\
& \textsc{WS353} & 72 & \textbf{73} & 71 & 71 \\
\midrule
\textsc{Es} & \textsc{WS353} & 57 & 58 & 58 & \textbf{59} \\
\midrule
\textsc{Fr} & \textsc{RG65} & 70 & 69 & \textbf{75} & \textbf{75} \\
\midrule
\textsc{Ro} & \textsc{WS353} & 48 & 52 & 51 & \textbf{54} \\
\midrule
\textsc{Ru} & \textsc{HJ} & 59 & 60 & 60 & \textbf{66} \\
\bottomrule
\end{tabular}
\caption{
Correlation between human judgement and similarity scores on word similarity datasets.
We train both our model and the \texttt{word2vec} baseline on normalized Wikipedia dumps.
Evaluation datasets contain words that are not part of the training set, so we represent them using null vectors (\texttt{sisg-}).
With our model, we also compute vectors for unseen words by summing the $n$-gram vectors (\texttt{sisg}).
}
\label{tab:wordsim}
\end{table}
We report results for our method and baselines for all datasets in Table~\ref{tab:wordsim}.
Some words from these datasets do not appear in our training data, and thus, we cannot obtain word representation for these words using the \texttt{cbow} and \texttt{skipgram} baselines.
In order to provide comparable results, we propose by default to use null vectors for these words.
Since our model exploits subword information, we can also compute valid representations for out-of-vocabulary words.
We do so by taking the sum of its $n$-gram vectors.
When OOV words are represented using null vectors we refer to our method as \texttt{sisg-} and \texttt{sisg} otherwise (Subword Information Skip Gram).
First, by looking at Table~\ref{tab:wordsim}, we notice that the proposed model (\texttt{sisg}), which uses subword information, outperforms the baselines on all datasets except the English \textsc{WS353} dataset.
Moreover, computing vectors for out-of-vocabulary words (\texttt{sisg}) is always at least as good as not doing so (\texttt{sisg-}).
This proves the advantage of using subword information in the form of character $n$-grams.
Second, we observe that the effect of using character $n$-grams is more important for Arabic, German and Russian than for English, French or Spanish.
German and Russian exhibit grammatical declensions with four cases for German and six for Russian.
Also, many German words are compound words; for instance the nominal phrase ``table tennis'' is written in a single word as ``Tischtennis''.
By exploiting the character-level similarities between ``Tischtennis'' and ``Tennis'', our model does not represent the two words as completely different words.
Finally, we observe that on the English Rare Words dataset (\textsc{RW}), our approach outperforms the baselines while it does not on the English \textsc{WS353} dataset.
This is due to the fact that words in the English \textsc{WS353} dataset are common words for which good vectors can be obtained without exploiting subword information.
When evaluating on less frequent words, we see that using similarities at the character level between words can help learning good word vectors.
\subsection{Word analogy tasks}
\begin{table}[t]
\centering
\begin{tabular}{rrccc}
\toprule
& & \texttt{sg} & \texttt{cbow} & \texttt{sisg} \\
\midrule
\multirow{ 2}{*}{\textsc{Cs}} & Semantic & 25.7 & 27.6 & 27.5 \\
& Syntactic & 52.8 & 55.0 & 77.8 \\
\midrule
\multirow{ 2}{*}{\textsc{De}} & Semantic & 66.5 & 66.8 & 62.3 \\
& Syntactic & 44.5 & 45.0 & 56.4 \\
\midrule
\multirow{ 2}{*}{\textsc{En}} & Semantic & 78.5 & 78.2 & 77.8 \\
& Syntactic & 70.1 & 69.9 & 74.9 \\
\midrule
\multirow{ 2}{*}{\textsc{It}} & Semantic & 52.3 & 54.7 & 52.3 \\
& Syntactic & 51.5 & 51.8 & 62.7 \\
\bottomrule
\end{tabular}
\caption{
Accuracy of our model and baselines on word analogy tasks for Czech, German, English and Italian.
We report results for semantic and syntactic analogies separately.
}
\label{tab:wordanalogy}
\end{table}
We now evaluate our approach on word analogy questions, of the form $A$ is to $B$ as $C$ is to $D$, where $D$ must be predicted by the models.
We use the datasets introduced by \newcite{mikolov2013efficient} for English, by \newcite{svoboda2016new} for Czech, by \newcite{koper2015multilingual} for German and by \newcite{berardi2015word} for Italian.
Some questions contain words that do not appear in our training corpus, and we thus excluded these questions from the evaluation.
\begin{table*}[t]
\centering
\begin{tabular}{rcccccccccc}
\toprule
&& \multicolumn{2}{c}{\textsc{De}} && \multicolumn{2}{c}{\textsc{En}} && \multicolumn{1}{c}{\textsc{Es}} && \multicolumn{1}{c}{\textsc{Fr}} \\
\cmidrule{3-4} \cmidrule{6-7} \cmidrule{9-9} \cmidrule{11-11}
&& \textsc{Gur350} & \textsc{ZG222} && WS353 & RW && WS353 && RG65 \\
\midrule
\newcite{luong2013better} && - & - && 64 & 34 && - && - \\
\newcite{qiu2014colearning} && - & - && 65 & 33 && - && - \\
\newcite{soricut2015unsupervised} && 64 & 22 && 71 & 42 && 47 && 67 \\
\texttt{sisg} && 73 & 43 && 73 & 48 && 54 && 69 \\
\midrule
\newcite{botha2014compositional} && 56 & 25 && 39 & 30 && 28 && 45 \\
\texttt{sisg} && 66 & 34 && 54 & 41 && 49 && 52 \\
\bottomrule
\end{tabular}
\caption{
Spearman's rank correlation coefficient between human judgement and model scores for different methods using morphology to learn word representations.
We keep all the word pairs of the evaluation set and obtain representations for out-of-vocabulary words with our model by summing the vectors of character $n$-grams.
Our model was trained on the same datasets as the methods we are comparing to (hence the two lines of results for our approach).
}
\label{tab:comparison}
\end{table*}
We report accuracy for the different models in Table~\ref{tab:wordanalogy}.
We observe that morphological information significantly improves the syntactic tasks; our approach outperforms the baselines.
In contrast, it does not help for semantic questions, and even degrades the performance for German and Italian.
Note that this is tightly related to the choice of the length of character $n$-grams that we consider.
We show in Sec.~\ref{sec:ngram-size} that when the size of the $n$-grams is chosen optimally, the semantic analogies degrade less.
Another interesting observation is that, as expected, the improvement over the baselines is more important for morphologically rich languages, such as Czech and German.
\subsection{Comparison with morphological representations}
\label{sec:sota}
We also compare our approach to previous work on word vectors incorporating subword information on word similarity tasks.
The methods used are: the recursive neural network of \newcite{luong2013better}, the morpheme \texttt{cbow} of \newcite{qiu2014colearning} and the morphological transformations of \newcite{soricut2015unsupervised}.
In order to make the results comparable, we trained our model on the same datasets as the methods we are comparing to: the English Wikipedia data released by \newcite{shaoul2010westbury}, and the news crawl data from the 2013 WMT shared task for German, Spanish and French.
We also compare our approach to the log-bilinear language model introduced by \newcite{botha2014compositional}, which was trained on the Europarl and news commentary corpora.
Again, we trained our model on the same data to make the results comparable.
Using our model, we obtain representations of out-of-vocabulary words by summing the representations of character $n$-grams.
We report results in Table~\ref{tab:comparison}.
We observe that our simple approach performs well relative to techniques based on subword information obtained from morphological segmentors.
We also observe that our approach outperforms the \newcite{soricut2015unsupervised} method, which is based on prefix and suffix analysis.
The large improvement for German is due to the fact that their approach does not model noun compounding, contrary to ours.
\subsection{Effect of the size of the training data}
\label{sec:exp-data-size}
Since we exploit character-level similarities between words, we are able to better model infrequent words.
Therefore, we should also be more robust to the size of the training data that we use.
In order to assess that, we propose to evaluate the performance of our word vectors on the similarity task as a function of the training data size.
To this end, we train our model and the \texttt{cbow} baseline on portions of Wikipedia of increasing size.
We use the Wikipedia corpus described above and isolate the first $1$, $2$, $5$, $10$, $20$, and $50$ percent of the data.
Since we don't reshuffle the dataset, they are all subsets of each other.
We report results in Fig.~\ref{fig:gulllss}.
\begin{figure*}[t]
\centering
\begin{subfigure}[b]{0.43\textwidth}
\includegraphics[width=\textwidth]{figures/figure-de-gur350.pdf}
\caption{\textsc{De-Gur350}}
\end{subfigure}
\qquad
\begin{subfigure}[b]{0.43\textwidth}
\includegraphics[width=\textwidth]{figures/figure-en-rw.pdf}
\caption{\textsc{En-RW}}
\end{subfigure}
\caption{
Influence of size of the training data on performance.
We compute word vectors following the proposed model using datasets of increasing size.
In this experiment, we train models on a fraction of the full Wikipedia dump.
}
\label{fig:gulllss}
\end{figure*}
As in the experiment presented in Sec.~\ref{sec:wordsim}, not all words from the evaluation set are present in the Wikipedia data.
Again, by default, we use a null vector for these words (\texttt{sisg-}) or compute a vector by summing the $n$-gram representations (\texttt{sisg}).
The out-of-vocabulary rate is growing as the dataset shrinks, and therefore the performance of \texttt{sisg-} and \texttt{cbow} necessarily degrades.
However, the proposed model (\texttt{sisg}) assigns non-trivial vectors to previously unseen words.
First, we notice that for all datasets, and all sizes, the proposed approach (\texttt{sisg}) performs better than the baseline.
However, the performance of the baseline \texttt{cbow} model gets better as more and more data is available.
Our model, on the other hand, seems to quickly saturate and adding more data does not always lead to improved results.
Second, and most importantly, we notice that the proposed approach provides very good word vectors even when using very small training datasets.
For instance, on the German \textsc{Gur350} dataset, our model (\texttt{sisg}) trained on $5\%$ of the data achieves better performance ($66$) than the \texttt{cbow} baseline trained on the full dataset ($62$).
On the other hand, on the English \textsc{RW} dataset, using $1\%$ of the Wikipedia corpus we achieve a correlation coefficient of $45$ which is better than the performance of \texttt{cbow} trained on the full dataset ($43$).
This has a very important practical implication: well performing word vectors can be computed on datasets of a restricted size and still work well on previously unseen words.
In general, when using vectorial word representations in specific applications, it is recommended to retrain the model on textual data relevant for the application.
However, this kind of relevant task-specific data is often very scarce and learning from a reduced amount of training data is a great advantage.
\subsection{Effect of the size of $n$-grams}
\label{sec:ngram-size}
The proposed model relies on the use of character $n$-grams to represent words as vectors.
As mentioned in Sec.~\ref{sec:model-ngrams}, we decided to use $n$-grams ranging from $3$ to $6$ characters.
This choice was arbitrary, motivated by the fact that $n$-grams of these lengths will cover a wide range of information.
They would include short suffixes (corresponding to conjugations and declensions for instance) as well as longer roots.
In this experiment, we empirically check for the influence of the range of $n$-grams that we use on performance.
We report our results in Table~\ref{tab:nsize} for English and German on word similarity and analogy datasets.
\begin{table*}[t]
\centering
\begin{subtable}[b]{0.31\textwidth}
\begin{tabular}{c c c c c c}
\toprule
& 2 & 3 & 4 & 5 & 6 \\
\midrule
2 & 57 & 64 & 67 & 69 & 69 \\
3 & & 65 & 68 & 70 & 70 \\
4 & & & 70 & 70 & \textbf{71} \\
5 & & & & 69 & \textbf{71} \\
6 & & & & & 70 \\
\bottomrule
\end{tabular}
\caption{\textsc{De-Gur350}}
\end{subtable}
\begin{subtable}[b]{0.31\textwidth}
\begin{tabular}{c c c c c c}
\toprule
& 2 & 3 & 4 & 5 & 6 \\
\midrule
2 & 59 & 55 & 56 & 59 & 60 \\
3 & & 60 & 58 & 60 & 62 \\
4 & & & 62 & 62 & 63 \\
5 & & & & 64 & 64 \\
6 & & & & & \textbf{65} \\
\bottomrule
\end{tabular}
\caption{\textsc{De} Semantic}
\end{subtable}
\begin{subtable}[b]{0.31\textwidth}
\centering
\begin{tabular}{c c c c c c}
\toprule
& 2 & 3 & 4 & 5 & 6 \\
\midrule
2 & 45 & 50 & 53 & 54 & 55 \\
3 & & 51 & 55 & 55 & \textbf{56} \\
4 & & & 54 & \textbf{56} & \textbf{56} \\
5 & & & & \textbf{56} & \textbf{56} \\
6 & & & & & 54 \\
\bottomrule
\end{tabular}
\caption{\textsc{De} Syntactic}
\end{subtable}
\vspace{1em}
\begin{subtable}[b]{0.31\textwidth}
\centering
\begin{tabular}{c c c c c c}
\toprule
& 2 & 3 & 4 & 5 & 6 \\
\midrule
2 & 41 & 42 & 46 & 47 & \textbf{48} \\
3 & & 44 & 46 & \textbf{48} & \textbf{48} \\
4 & & & 47 & \textbf{48} & \textbf{48} \\
5 & & & & \textbf{48} & \textbf{48} \\
6 & & & & & \textbf{48} \\
\bottomrule
\end{tabular}
\caption{\textsc{En-RW}}
\end{subtable}
\begin{subtable}[b]{0.31\textwidth}
\centering
\begin{tabular}{c c c c c c}
\toprule
& 2 & 3 & 4 & 5 & 6 \\
\midrule
2 & 78 & 76 & 75 & 76 & 76 \\
3 & & 78 & 77 & 78 & 77 \\
4 & & & 79 & 79 & 79 \\
5 & & & & \textbf{80} & 79 \\
6 & & & & & \textbf{80} \\
\bottomrule
\end{tabular}
\caption{\textsc{En} Semantic}
\end{subtable}
\begin{subtable}[b]{0.31\textwidth}
\centering
\begin{tabular}{c c c c c c}
\toprule
& 2 & 3 & 4 & 5 & 6 \\
\midrule
2 & 70 & 71 & 73 & 74 & 73 \\
3 & & 72 & 74 & \textbf{75} & 74 \\
4 & & & 74 & \textbf{75} & \textbf{75} \\
5 & & & & 74 & 74 \\
6 & & & & & 72 \\
\bottomrule
\end{tabular}
\caption{\textsc{En} Syntactic}
\end{subtable}
\caption{
Study of the effect of sizes of $n$-grams considered on performance.
We compute word vectors by using character $n$-grams with $n$ in $\{i, \dots, j\}$ and report performance for various values of $i$ and $j$.
We evaluate this effect on German and English, and represent out-of-vocabulary words using subword information.
}
\label{tab:nsize}
\end{table*}
We observe that for both English and German, our arbitrary choice of $3$-$6$ was a reasonable decision, as it provides satisfactory performance across languages.
The optimal choice of length ranges depends on the considered task and language and should be tuned appropriately.
However, due to the scarcity of test data, we did not implement any proper validation procedure to automatically select the best parameters.
Nonetheless, taking a large range such as $3-6$ provides a reasonable amount of subword information.
This experiment also shows that it is important to include long $n$-grams, as columns corresponding to $n \leq 5$ and $n \leq 6$ work best.
This is especially true for German, as many nouns are compounds made up from several units that can only be captured by longer character sequences.
On analogy tasks, we observe that using larger $n$-grams helps for semantic analogies.
However, results are always improved by taking $n \geq 3$ rather than $n \geq 2$, which shows that character $2$-grams are not informative for that task.
As described in Sec.~\ref{sec:model-ngrams}, before computing character $n$-grams, we prepend and append special positional characters to represent the beginning and end of word.
Therefore, $2$-grams will not be enough to properly capture suffixes that correspond to conjugations or declensions, since they are composed of a single proper character and a positional one.
\subsection{Language modeling}
In this section, we describe an evaluation of the word vectors obtained with our method on a language modeling task.
We evaluate our language model on five languages (\textsc{Cs}, \textsc{De}, \textsc{Es}, \textsc{Fr}, \textsc{Ru}) using the datasets introduced by \newcite{botha2014compositional}.
Each dataset contains roughly one million training tokens, and we use the same preprocessing and data splits as \newcite{botha2014compositional}.
Our model is a recurrent neural network with $650$ LSTM units, regularized with dropout (with probability of $0.5$) and weight decay (regularization parameter of $10^{-5}$).
We learn the parameters using the Adagrad algorithm with a learning rate of $0.1$, clipping the gradients which have a norm larger than~$1.0$.
We initialize the weight of the network in the range $[-0.05, 0.05]$, and use a batch size of $20$.
Two baselines are considered: we compare our approach to the log-bilinear language model of \newcite{botha2014compositional} and the character aware language model of \newcite{kim2016character}.
We trained word vectors with character $n$-grams on the training set of the language modeling task and use them to initialize the lookup table of our language model.
We report the test perplexity of our model without using pre-trained word vectors (\texttt{LSTM}), with word vectors pre-trained without subword information (\texttt{sg}) and with our vectors (\texttt{sisg}).
The results are presented in Table~\ref{tab:lm}.
\begin{table}[t]
\centering
\begin{tabular}{rccccc}
\toprule
& \textsc{Cs} & \textsc{De} & \textsc{Es} & \textsc{Fr} & \textsc{Ru} \\
\midrule
Vocab. size & 46k & 37k & 27k & 25k & 63k \\
\midrule
\textsc{CLBL} & 465 & 296 & 200 & 225 & 304 \\
\textsc{CANLM} & 371 & 239 & 165 & 184 & 261 \\
\midrule
\texttt{LSTM} & 366 & 222 & 157 & 173 & 262 \\
\texttt{sg} & 339 & 216 & 150 & 162 & 237 \\
\texttt{sisg} & \textbf{312} & \textbf{206} & \textbf{145} & \textbf{159} & \textbf{206} \\
\bottomrule
\end{tabular}
\caption{
Test perplexity on the language modeling task, for 5 different languages.
We compare to two state of the art approaches: \textsc{CLBL} refers to the work of \protect\newcite{botha2014compositional} and \textsc{CANLM} refers to the work of \protect\newcite{kim2016character}.
}
\label{tab:lm}
\end{table}
\begin{table}[t]
\centering
\small
\begin{tabular}{rrrrr}
\toprule
& word & \multicolumn{3}{c}{$n$-grams} \\
\midrule
& autofahrer & fahr & fahrer & auto \\
& freundeskreis & kreis & kreis> & <freun \\
\textsc{De} & grundwort & wort & wort> & grund \\
& sprachschule & schul & hschul & sprach \\
& tageslicht & licht & gesl & tages \\
\midrule
& anarchy & chy & <anar & narchy \\
& monarchy & monarc & chy & <monar \\
& kindness & ness> & ness & kind \\
& politeness & polite & ness> & eness> \\
\textsc{En} & unlucky & <un & cky> & nlucky \\
& lifetime & life & <life & time \\
& starfish & fish & fish> & star \\
& submarine & marine & sub & marin \\
& transform & trans & <trans & form \\
\midrule
& finirais & ais> & nir & fini \\
\textsc{Fr} & finissent & ent> & finiss & <finis \\
& finissions & ions> & finiss & sions> \\
\bottomrule
\end{tabular}
\caption{
Illustration of most important character $n$-grams for selected words in three languages.
For each word, we show the $n$-grams that, when removed, result in the most different representation.
}
\label{tab:morphemes}
\end{table}
We observe that initializing the lookup table of the language model with pre-trained word representations improves the test perplexity over the baseline LSTM.
The most important observation is that using word representations trained with subword information outperforms the plain skipgram model.
We observe that this improvement is most significant for morphologically rich Slavic languages such as Czech (8\% reduction of perplexity over \texttt{sg}) and Russian (13\% reduction).
The improvement is less significant for Roman languages such as Spanish (3\% reduction) or French (2\% reduction).
This shows the importance of subword information on the language modeling task and exhibits the usefulness of the vectors that we propose for morphologically rich languages.
\section{Qualitative analysis}
\subsection{Nearest neighbors.}
We report sample qualitative results in Table~\ref{tab:nn}.
For selected words, we show nearest neighbors according to cosine similarity for vectors trained using the proposed approach and for the \texttt{skipgram} baseline.
As expected, the nearest neighbors for complex, technical and infrequent words using our approach are better than the ones obtained using the baseline model.
\subsection{Character $n$-grams and morphemes}
We want to qualitatively evaluate whether or not the most important $n$-grams in a word correspond to morphemes.
To this end, we take a word vector that we construct as the sum of $n$-grams.
As described in Sec.~\ref{sec:model-ngrams}, each word $w$ is represented as the sum of its $n$-grams: $u_w = \sum_{g \in \mathcal{G}_w} z_g$.
For each $n$-gram~$g$, we propose to compute the restricted representation~$u_{w \backslash g}$ obtained by omitting $g$:
\begin{equation}
u_{w \backslash g} = \sum_{g' \in \mathcal{G} - \{g\}} z_{g'}.\nonumber
\end{equation}
We then rank $n$-grams by increasing value of cosine between $u_w$ and $u_{w \backslash g}$.
We show ranked $n$-grams for selected words in three languages in Table~\ref{tab:morphemes}.
For German, which has a lot of compound nouns, we observe that the most important $n$-grams correspond to valid morphemes.
Good examples include \emph{Autofahrer} (car driver) whose most important $n$-grams are \emph{Auto} (car) and \emph{Fahrer} (driver).
We also observe the separation of compound nouns into morphemes in English, with words such as \emph{lifetime} or \emph{starfish}.
However, for English, we also observe that $n$-grams can correspond to affixes in words such as \emph{kindness} or \emph{unlucky}.
Interestingly, for French we observe the inflections of verbs with endings such as \emph{ais>}, \emph{ent>} or \emph{ions>}.
\subsection{Word similarity for OOV words}
As described in Sec.~\ref{sec:model-ngrams}, our model is capable of building word vectors for words that do not appear in the training set.
For such words, we simply average the vector representation of its $n$-grams.
In order to assess the quality of these representations, we analyze which of the $n$-grams match best for OOV words by selecting a few word pairs from the English RW similarity dataset.
We select pairs such that one of the two words is not in the training vocabulary and is hence only represented by its $n$-grams.
For each pair of words, we display the cosine similarity between each pair of $n$-grams that appear in the words.
In order to simulate a setup with a larger number of OOV words, we use models trained on~$1\%$ of the Wikipedia data as in Sec.~\ref{sec:exp-data-size}.
The results are presented in Fig.~\ref{fig:ngram-match}.
We observe interesting patterns, showing that subwords match correctly.
Indeed, for the word \emph{chip}, we clearly see that there are two groups of $n$-grams in \emph{microcircuit} that match well.
These roughly correspond to \emph{micro} and \emph{circuit}, and $n$-grams in between don't match well.
Another interesting example is the pair \emph{rarity} and \emph{scarceness}.
Indeed, \emph{scarce} roughly matches \emph{rarity} while the suffix \emph{-ness} matches \emph{-ity} very well.
Finally, the word \emph{preadolescent} matches \emph{young} well thanks to the \emph{-adolesc-} subword.
This shows that we build robust word representations where prefixes and suffixes can be ignored if the grammatical form is not found in the dictionary.
\begin{table*}[p]
\centering
\begin{tabular}{lcccccc}
\toprule
query & tiling & tech-rich & english-born & micromanaging & eateries & dendritic \\
\midrule
\texttt{sisg} & tile & tech-dominated & british-born & micromanage & restaurants & dendrite \\
& flooring & tech-heavy & polish-born & micromanaged & eaterie & dendrites \\
\midrule
\texttt{sg} & bookcases & technology-heavy & most-capped & defang & restaurants & epithelial \\
& built-ins & .ixic & ex-scotland & internalise & delis & p53 \\
\bottomrule
\end{tabular}
\caption{
Nearest neighbors of rare words using our representations and \texttt{skipgram}.
These hand picked examples are for illustration.}
\label{tab:nn}
\end{table*}
\begin{figure*}[p]
\begin{minipage}[t]{.45\textwidth}
\vspace{0pt}
\includegraphics[width=\linewidth]{figures/ngram-matching/scarceness-1-rarity-0.pdf}
\vspace{0.5em}
\\
\includegraphics[width=\linewidth]{figures/ngram-matching/asphaltic-1-paving-0.pdf}
\vspace{0.5em}
\\
\includegraphics[width=\linewidth]{figures/ngram-matching/interlink-1-connect-0.pdf}
\end{minipage}
\hspace{2.5em}
\begin{minipage}[t]{.475\textwidth}
\vspace{0pt}
\includegraphics[width=\linewidth]{figures/ngram-matching/preadolescent-1-young-0.pdf}
\vspace{0.5em}
\\
\includegraphics[width=\linewidth]{figures/ngram-matching/microcircuit-1-chip-0.pdf}
\vspace{0.5em}
\\
\includegraphics[width=\linewidth]{figures/ngram-matching/piquancy-1-spiciness-0.pdf}
\end{minipage}
\caption{
Illustration of the similarity between character $n$-grams in out-of-vocabulary words.
For each pair, only one word is OOV, and is shown on the $x$ axis.
Red indicates positive cosine, while blue negative.
}
\label{fig:ngram-match}
\end{figure*}
\section{Model}
In this section, we propose our model to learn word representations while taking into account morphology.
We model morphology by considering subword units, and representing words by a sum of its character $n$-grams.
We will begin by presenting the general framework that we use to train word vectors, then present our subword model and eventually describe how we handle the dictionary of character $n$-grams.
\subsection{General model}
We start by briefly reviewing the continuous skipgram model introduced by \newcite{mikolov2013distributed}, from which our model is derived.
Given a word vocabulary of size $W$, where a word is identified by its index $w~\in~\{1, ..., W\}$, the goal is to learn a vectorial representation for each word $w$.
Inspired by the distributional hypothesis~\cite{harris1954distributional}, word representations are trained to \emph{predict well} words that appear in its context.
More formally, given a large training corpus represented as a sequence of words $w_1, ..., w_T$, the objective of the skipgram model is to maximize the following log-likelihood:
\begin{equation*}
\sum_{t=1}^T \ \sum_{c \in \mathcal{C}_t} \ \log p(w_c \ | \ w_t),
\end{equation*}
where the context $\mathcal{C}_t$ is the set of indices of words surrounding word $w_t$.
The probability of observing a context word $w_c$ given $w_t$ will be parameterized using the aforementioned word vectors.
For now, let us consider that we are given a scoring function~$s$ which maps pairs of (word, context) to scores in~$\mathbb{R}$.
One possible choice to define the probability of a context word is the softmax:
\begin{equation*}
p(w_c \ | \ w_t) = \frac{e^{s(w_t,\ w_c)}}{\sum_{j=1}^W e^{s(w_t,\ j)}}.
\end{equation*}
However, such a model is not adapted to our case as it implies that, given a word $w_t$, we only predict one context word $w_c$.
The problem of predicting context words can instead be framed as a set of independent binary classification tasks.
Then the goal is to independently predict the presence (or absence) of context words.
For the word at position $t$ we consider all context words as positive examples and sample negatives at random from the dictionary.
For a chosen context position $c$, using the binary logistic loss, we obtain the following negative log-likelihood:
\begin{equation*}
\log \left(1 + e^{-s(w_t,\ w_c)} \right) + \sum_{n \in \mathcal{N}_{t, c}} \log \left(1 + e^{s(w_t,\ n)}\right),
\end{equation*}
where $\mathcal{N}_{t,c}$ is a set of negative examples sampled from the vocabulary.
By denoting the logistic loss function $\ell: x \mapsto \log(1 + e^{-x})$, we can re-write the objective as:
\begin{equation*}
\sum_{t=1}^{T} \left [ \sum_{c \in \mathcal{C}_t} \ell(s(w_t,\ w_c)) + \sum_{n \in \mathcal{N}_{t,c}} \ell(-s(w_t,\ n)) \right ].
\end{equation*}
A natural parameterization for the scoring function $s$ between a word $w_t$ and a context word $w_c$ is to use word vectors.
Let us define for each word $w$ in the vocabulary two vectors $u_w$ and $v_w$ in $\mathbb{R}^d$.
These two vectors are sometimes referred to as \emph{input} and \emph{output} vectors in the literature.
In particular, we have vectors $\mathbf{u}_{w_t}$ and $\mathbf{v}_{w_c}$, corresponding, respectively, to words $w_t$ and $w_c$.
Then the score can be computed as the scalar product between word and context vectors as $s(w_t, w_c) = \mathbf{u}_{w_t}^{\top} \mathbf{v}_{w_c}$.
The model described in this section is the skipgram model with negative sampling, introduced by \newcite{mikolov2013distributed}.
\subsection{Subword model}
\label{sec:model-ngrams}
By using a distinct vector representation for each word, the skipgram model ignores the internal structure of words.
In this section, we propose a different scoring function $s$, in order to take into account this information.
Each word $w$ is represented as a bag of character $n$-gram.
We add special boundary symbols \texttt{<} and \texttt{>} at the beginning and end of words, allowing to distinguish prefixes and suffixes from other character sequences.
We also include the word $w$ itself in the set of its $n$-grams, to learn a representation for each word (in addition to character $n$-grams).
Taking the word \emph{where} and $n=3$ as an example, it will be represented by the character $n$-grams:
\begin{center}
\texttt{<wh, whe, her, ere, re>}
\end{center}
and the special sequence
\begin{center}
\texttt{<where>}.
\end{center}
Note that the sequence \texttt{<her>}, corresponding to the word \emph{her} is different from the tri-gram \texttt{her} from the word \emph{where}.
In practice, we extract all the $n$-grams for $n$ greater or equal to 3 and smaller or equal to $6$.
This is a very simple approach, and different sets of $n$-grams could be considered, for example taking all prefixes and suffixes.
Suppose that you are given a dictionary of $n$-grams of size $G$.
Given a word $w$, let us denote by $\mathcal{G}_w \subset \{1, \dots, G \}$ the set of $n$-grams appearing in $w$.
We associate a vector representation $\mathbf{z}_g$ to each $n$-gram $g$.
We represent a word by the sum of the vector representations of its $n$-grams.
We thus obtain the scoring function:
\begin{equation*}
s(w, c) = \sum_{g \in \mathcal{G}_w} \mathbf{z}_g^\top \mathbf{v}_c.
\end{equation*}
This simple model allows sharing the representations across words, thus allowing to learn reliable representation for rare words.
In order to bound the memory requirements of our model, we use a hashing function that maps $n$-grams to integers in 1 to $K$.
We hash character sequences using the Fowler-Noll-Vo hashing function (specifically the \texttt{FNV-1a} variant).\footnote{\smaller\relax\url{http://www.isthe.com/chongo/tech/comp/fnv}}
We set $K = 2.10^6$ below.
Ultimately, a word is represented by its index in the word dictionary and the set of hashed $n$-grams it contains.
\section{Related work}
\paragraph{Morphological word representations.}
In recent years, many methods have been proposed to incorporate morphological information into word representations.
To model rare words better, \newcite{alexandrescu2006factored} introduced factored neural language models, where words are represented as sets of features.
These features might include morphological information, and this technique was succesfully applied to morphologically rich languages, such as Turkish~\cite{sak2010morphology}.
Recently, several works have proposed different composition functions to derive representations of words from morphemes~\cite{lazaridou2013compositional,luong2013better,botha2014compositional,qiu2014colearning}.
These different approaches rely on a morphological decomposition of words, while ours does not.
Similarly, \newcite{chen2015joint} introduced a method to jointly learn embeddings for Chinese words and characters.
\newcite{cui2015knet} proposed to constrain morphologically similar words to have similar representations.
\newcite{soricut2015unsupervised} described a method to learn vector representations of morphological transformations, allowing to obtain representations for unseen words by applying these rules.
Word representations trained on morphologically annotated data were introduced by \newcite{cotterell2015morphological}.
Closest to our approach, \newcite{schutze1993word} learned representations of character four-grams through singular value decomposition, and derived representations for words by summing the four-grams representations.
Very recently, \newcite{wieting2016charagram} also proposed to represent words using character $n$-gram count vectors.
However, the objective function used to learn these representations is based on paraphrase pairs, while our model can be trained on any text corpus.
\paragraph{Character level features for NLP.}
Another area of research closely related to our work are character-level models for natural language processing.
These models discard the segmentation into words and aim at learning language representations directly from characters.
A first class of such models are recurrent neural networks, applied to language modeling~\cite{mikolov2012subword,sutskever2011generating,graves2013generating,bojanowski2015alternative}, text normalization~\cite{chrupala2014normalizing}, part-of-speech tagging~\cite{ling2015finding} and parsing~\cite{ballesteros2015improved}.
Another family of models are convolutional neural networks trained on characters, which were applied to part-of-speech tagging~\cite{santos2014learning}, sentiment analysis~\cite{santos2014deep}, text classification~\cite{zhang2015character} and language modeling~\cite{kim2016character}.
\newcite{sperr2013letter} introduced a language model based on restricted Boltzmann machines, in which words are encoded as a set of character $n$-grams.
Finally, recent works in machine translation have proposed using subword units to obtain representations of rare words~\cite{sennrich2016neural,luong2016hybrid}.
\section{Introduction}
Learning continuous representations of words has a long history in natural language processing~\cite{rumelhart1988learning}.
These representations are typically derived from large unlabeled corpora using co-occurrence statistics~\cite{deerwester1990indexing,schutze1992dimensions,lund1996producing}.
A large body of work, known as distributional semantics, has studied the properties of these methods~\cite{turney2010frequency,baroni2010distributional}.
In the neural network community, \newcite{collobert2008unified} proposed to learn word embeddings using a feedforward neural network, by predicting a word based on the two words on the left and two words on the right.
More recently, \newcite{mikolov2013distributed} proposed simple log-bilinear models to learn continuous representations of words on very large corpora efficiently.
Most of these techniques represent each word of the vocabulary by a distinct vector, without parameter sharing.
In particular, they ignore the internal structure of words, which is an important limitation for morphologically rich languages, such as Turkish or Finnish.
For example, in French or Spanish, most verbs have more than forty different inflected forms, while the Finnish language has fifteen cases for nouns.
These languages contain many word forms that occur rarely (or not at all) in the training corpus, making it difficult to learn good word representations.
Because many word formations follow rules, it is possible to improve vector representations for morphologically rich languages by using character level information.
In this paper, we propose to learn representations for character $n$-grams, and to represent words as the sum of the $n$-gram vectors.
Our main contribution is to introduce an extension of the continuous skipgram model~\cite{mikolov2013distributed}, which takes into account subword information.
We evaluate this model on nine languages exhibiting different morphologies, showing the benefit of our approach.
\input{relatedwork}
\input{model}
\input{experiments}
\section{Conclusion}
In this paper, we investigate a simple method to learn word representations by taking into account subword information.
Our approach, which incorporates character $n$-grams into the skipgram model, is related to an idea that was introduced by \newcite{schutze1993word}.
Because of its simplicity, our model trains fast and does not require any preprocessing or supervision.
We show that our model outperforms baselines that do not take into account subword information, as well as methods relying on morphological analysis.
We will open source the implementation of our model, in order to facilitate comparison of future work on learning subword representations.
\subsection*{Acknowledgements}
We thank Marco Baroni, Hinrich Schütze and the anonymous reviewers for their insightful comments.
| 1bf20cb1f563d18219eecbd565af55b49cec746b | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction and Related Work}
Automatic analysis of urban traffic activities is an urgent need due to essential traffic management and increased vehicle violations. Among many traffic surveillance techniques, computer vision-based methods have attracted a great deal of attention and made great contributions to realistic applications such as vehicle counting, target vehicle retrieval and behavior analysis. In these research areas, efficient and accurate vehicle detection and attributes annotation is the most important component of traffic surveillance systems.
Vehicle detection is a fundamental objective of traffic surveillance. Traditional vehicle detection methods can be categorized into frame-based and motion-based approaches~\cite{sivaraman2013looking,buch2011review}. For motion-based approaches, frames subtraction \cite{park2007video}, adaptive background modeling \cite{stauffer1999adaptive} and optical flow \cite{martinez2008driving,liu2013learning} are often utilized. However, some non-vehicle moving objects will be falsely detected with motion-based approaches since less visual information is exploited. To achieve higher detection performance, recently, the deformable part-based model (DPM) \cite{felzenszwalb2010object} employs a star-structured architecture consisting of root and parts filters with associated deformation models for object detection. DPM can successfully handle deformable object detection even when the target is partially occluded. However, it leads to heavy computational costs due to the use of the sliding window precedure for appearance features extraction and classification.
With the wide success of deep networks on image classification \cite{krizhevsky2012imagenet,karpathy2014large,wu2016deep,wu2014leveraging,dong2015deep}, a Region-based CNN (RCNN) \cite{girshick2014rich} combines object proposals, CNN learned features and an SVM classifier for accurate object detection. To further increase the detection speed and accuracy, Fast RCNN \cite{girshick2015fast} adopts a region of interest (ROI) pooling layer and the multi-task loss to estimate object classes while predicting bounding-box positions. ``Objectness" proposal methods such as Selective Search \cite{uijlings2013selective} and Edgeboxes \cite{zitnick2014edge} can be introduced in RCNN and Fast RCNN to improve the efficiency compared to the traditional sliding window fashion. Furthermore, Faster RCNN \cite{ren2015faster} employs a Region Proposal Network (RPN) with shared convolutional features to enable cost-free effective proposals. All these deep models target general object detection. In our task, we aim for real-time detection of a special object type, vehicle.
Besides, in urban traffic surveillance, another interesting and valuable task is to extract more diverse information from detected vehicles - we call it vehicle attributes annotation. Each individual vehicle on the road has its special attributes: travel direction (i.e., pose), inherent color, type and other more fine-grained information with respect to the headlight, grille and wheel. It is extremely beneficial to annotate a target vehicle's attributes accurately. Lin et al. \cite{yang2014object} presents an auto-masking neural network for vehicle detection and pose estimation. In \cite{li2010vehicle}, an approach by vector matching of template is introduced for vehicle color recognition. In \cite{dong2014vehicle}, unsupervised convolutional neural network is adopted for vehicle type classification from frontal view images. However, independent analysis of different attributes makes visual information not well explored and the process inefficient, and little work has been done for annotating these vehicle attributes simultaneously. Actually, there exist strong correlations between these vehicle attributes learning tasks. For example, vehicle type classification based on visual structures is very dependent on the viewpoint. Therefore, we believe multi-task learning \cite{shao2015deeply,liu2016dap3d} can be helpful since such joint learning schemes can implicitly learn the common features shared by these correlated tasks. Moreover, a unified multi-attributes inference model can also significantly improve the efficiency.
\begin{figure}[t]
\centering
\includegraphics[width=1\textwidth]{system.jpg}\\
\caption{\scriptsize{
\textbf{Illustration of DAVE}. Vehicle detection and corresponding pose, color and type annotation can be simultaneously achieved by DAVE as shown in the right sub-figure.}}
\label{exampless}
\vspace{-3ex}
\end{figure}
Inspired by the advantages and drawbacks of previous work, in this paper, we propose a fast framework DAVE based on convolutional neural networks (CNNs), as shown in Fig. 1, for vehicle detection and attributes annotation in urban traffic surveillance. In particular, DAVE consists of two CNNs: fast vehicle proposal network (FVPN) and attributes learning network (ALN). The FVPN is a shallow\emph{ fully convolutional network} which is able to predict all the bounding-boxes of vehicle-like objects in real-time. The latter ALN configured with a very deep structure can precisely verify each proposal and infer pose, color and type information for positive vehicles, simultaneously. It is noteworthy that informative features learned from the deep ALN can be regarded as latent data-driven knowledge to guide the training of the shallow FVPN, thus we bridge the ALN and FVPN with such knowledge guidance and jointly optimize these two CNNs at the same time in our architecture. In this way, more exhaustive vehicle descriptions learned from the ALN as helpful supervision benefits the FVPN with better performance. Once the joint training is finished, a two-stage inference scheme will be adopted for final vehicle annotation. The main contributions of our work are highlighted as follows:
\textbf{1.} We unify multiple vehicle-related tasks into one deep vehicle annotation framework DAVE which can effectively and efficiently annotate each vehicle's bounding-box, pose, color and type simultaneously.
\textbf{2.} Two CNNs proposed in our method, i.e., the Fast Vehicle Proposal Network (FVPN) and the vehicle Attributes Learning Network (ALN), are optimized in a joint manner by bridging two CNNs with latent data-driven knowledge guidance. In this way, the deeper ALN can benefit the performance of the shallow FVPN.
\textbf{3.} We introduce a new Urban Traffic Surveillance (UTS) vehicle dataset consisting of six $1920\times1080$ (FHD) resolution videos with varying illumination conditions and viewpoints.
\section{Detection and Annotation for Vehicles (DAVE)}
We unify vehicle detection and annotation of pose, color and type into one framework: DAVE. As illustrated in Fig. 2, DAVE consists of two convolutional neural networks called fast vehicle proposal network (FVPN) and attributes learning network (ALN), respectively. FVPN aims to predict all the positions of vehicle-like objects in real-time. Afterwards, these vehicle proposals are passed to the ALN to simultaneously verify all the positive vehicles and infer their corresponding poses, colors and types. In the training phase, FVPN and ALN are optimized jointly, while two-stage inference is used in the test phase. Specifically, training our DAVE is inspired by \cite{hinton2015distilling} that knowledge learned from solid deep networks can be distilled to teach shallower networks. We apply latent data-driven knowledge from the deep ALN to guide the training of the shallow FVPN.
This method proves to be able to enhance the performance of the FVPN to some extent through experiments. The architecture of FVPN and ALN will be described in the following subsections.
\begin{figure}[t]
\centering
\includegraphics[width=0.975\textwidth]{learning.jpg}\\
\caption{
\textbf{Training Architecture of DAVE.} Two CNNs: FVPN and ALN are simultaneously optimized in a joint manner by bridging them with latent data-driven knowledge guidance.}
\label{exampless}
\end{figure}
\subsection{Fast Vehicle Proposal Network (FVPN)}
Searching the whole image to locate potential vehicle positions in a sliding window fashion is prohibitive for real-time applications. Traditional object proposal methods are put forward to alleviate this problem, but thousands of proposals usually contain numerous false alarms and duplicate predictions which heavily lower the efficiency. Particularly for one specific object, we expect very fast and accurate proposal performance can be achieved.
Our proposed fast vehicle proposal network (FVPN) is a shallow \emph{fully convolutional network}, which aims to precisely localize all the vehicle-like objects in real-time. We are interested in exploring whether or not a small scale CNN is enough to handle the single object proposal task. A schematic diagram of the FVPN is depicted in the bottom part of Fig. 2. The first convolutional layer (\emph{conv\_1}) filters the $60\times60$ training images with 32 kernels of size $5\times5$. The second convolutional layer (\emph{conv\_2}) takes as input the feature maps obtained from the previous layer and filters them with 64 kernels of size $5\times5$. Max pooling and Rectified Linear Units (ReLU) layers are configured after the first two convolutional layers. The third convolutional layer (\emph{conv\_3}) with 64 kernels of size $3\times3$ is branched into three sibling $1\times1$ convolutional layers transformed by traditional fully connected layers. In detail, \emph{Conv\_fc\_class} outputs softmax probabilities of positive samples and the background; \emph{Conv\_fc\_bbr} encodes bounding-box coordinates for each positive sample; \emph{Conv\_fc\_knowledge} is configured for learning latent data-driven knowledge distilled from the ALN, which makes the FVPN be trained with more meticulous vehicle features. Inspired by \cite{long2015fully}, these $1\times1$ convolutional layers can successfully lead to differently purposed heatmaps in the inference phase. This property will directly achieve real-time vehicle localization from whole images/frames by our FVPN.
We employ different loss supervision layers for three corresponding tasks in the FVPN. First, discrimination between a vehicle and the background is a simple binary classification problem. A softmax loss layer is applied to predict vehicle confidence, $p^c=\{p_{ve}^c,p_{bg}^c\}$. Besides, each bounding-box is encoded by 4 predictions: $x$, $y$, $w$ and $h$. $x$ and $y$ denote the left-top coordinates of the vehicle position, while $w$ and $h$ represent the width and height of the vehicle size. We normalize all the 4 values relative to the image width and height so that they can be bounded between 0 and 1. Note that all bounding boxes' coordinates are set as \emph{zero} for background samples. Following \cite{girshick2015fast}, a smooth L1 loss layer is used for bounding-box regression to output the refined coordinates vector, $loc=(\hat{x},\hat{y},\hat{w},\hat{h})$. Finally, for guiding with latent data-driven knowledge of an N-dimensional vector distilled from a deeper net, the cross-entropy loss is employed for $p^{know}=\{p_0^{know}\ldots p_{N-1}^{know}\}$.
We adopt a multi-task loss $L_{FVPN}$ on each training batch to jointly optimize binary classification of a vehicle against background, bounding-box regression and learning latent knowledge from a deeper net as the following function:
\begin{equation}
\label{eq:e1}
\small
L_{FVPN}(loc, p^{bic},p^{know})=L_{bic}(p^{bic})+\alpha L_{bbox}(loc)+\beta L_{know}(p^{know}),
\end{equation}
where $L_{bic}$ denotes the softmax loss for binary classification of vehicle and background. $L_{bbox}$ indicates a smooth $\ell_1$ loss defined in \cite{girshick2015fast} as:
\begin{equation}
\label{eq:e1}
\small
L_{bbox}(loc)=f_{L1}(loc-loc_{t}),\ \mathrm{where}\ f_{L1}(x) = \left\{\begin{array}{l}
0.5x^{2},~ \text{if}\ |x|<1 \\
|x|-0.5,~ \text{otherwise}
\end{array}
\right.
\end{equation}
Furthermore, the cross entropy loss $L_{know}$ is to guide the training of the FVPN by a latent N-dimensional feature vector $t^{know}$ learned from a more solid net, which is defined as:
\begin{equation}
\small
L_{know}(p^{know})=-\frac{1}{N}\sum_{i}^{N}t^{know}_{i}\log p^{know}_{i}+(1-t^{know}_{i})\log (1-p^{know}_{i}).
\end{equation}
It is noteworthy that a bounding-box for the background is meaningless in the FVPN back-propagation phase and will cause training to diverge early \cite{redmon2015you}, \emph{thus we set $\alpha=0$ for background samples, otherwise $\alpha=0.5$.} Besides, $\beta$ is fixed to $0.5$.
\subsection{Attributes Learning Network (ALN)}
Modeling vehicle pose estimation, color recognition and type classification separately is less accurate and inefficient. Actually, relationships between these tasks can be explored, so that designing a multi-task network is beneficial for learning shared features which can lead to extra performance gains. The attribute learning network (ALN) is a unified network to verify vehicle candidates and annotate their poses, colors and types. The network architecture of the ALN is mainly inspired by the GoogLeNet \cite{szegedy2015going} model. Specifically, we design the ALN by adding 4 fully connected layers to extend the GoogLeNet into a multi-attribute learning model. The reason to adopt such a very deep structure here is because vehicle annotation belongs to fine-grained categorization problems and a deeper net has more powerful capability to learn representative and discriminative features. Another advantage of the ALN is its high-efficiency inherited from the GoogLeNet which has lower computation and memory costs compared with other deep nets such as the VGGNet \cite{simonyan2014very}.
The ALN is a multi-task network optimized with four softmax loss layers for vehicle annotation tasks. Each training image has four labels in $V$, $P$, $C$ and $T$. $V$ determines whether a sample is a vehicle. If $V$ is a true vehicle, the remaining three attributes $P$, $C$ and $T$ represent its pose, color and type respectively. However, if $V$ is the background or a vehicle with a catch-all\footnote{``Catch-all" indicates other undefined types and colors which are not included in our training model.} type or color, $P$, $C$ and $T$ are set as \emph{zero} denoting attributes are unavailable in the training phase. The first softmax loss layer $L_{verify} (p^V)$ for binary classification (vehicle vs. background) is the same as $L_{bic} (p^c)$ in the FVPN. The softmax loss $L_{pose} (p^P)$, $L_{color} (p^C)$ and $L_{type} (p^T)$ are optimized for pose estimation, color recognition and vehicle type classification respectively, where $p^P=\{p_1^P,\ldots,p_{np}^P\}$, $p^C=\{p_1^C,\ldots,p_{nc}^C\}$ and $p^T=\{p_1^T,\ldots, p_{nt}^T\}$. \{$np$, $nc$, $nt$\} indicate the number of vehicle poses, colors and types respectively. The whole loss function is defined as follows:
\begin{equation}
\label{eq:e1}
\small
L_{ALN}(p^{V},p^{P},p^{C},p^{T})=L_{verify}(p^{V})+\lambda_{1} L_{pose}(p^{P})+\lambda_{2} L_{color}(p^{C})+\lambda_{3} L_{type}(p^{T}),
\end{equation}
where all the four sub loss functions are softmax loss for vehicle verification (\emph{\textbf{``verification" in this paper means confirming whether a proposal is vehicle}}), pose estimation, color recognition and type classification. \emph{Following the similar case of $\alpha$ in Eq. (1), parameters $\{\lambda_1,\lambda_2,\lambda_3\}$ are all fixed as 1 for the positive samples, otherwise as 0 for the background.}
\subsection{Deep Nets Training}
\subsubsection{Training Dataset and Data Augmentation}
We adopt the large-scale CompCars dataset \cite{yang2015large} with more than 100,000 web-nature data as the positive training samples which are annotated with tight bounding-boxes and rich vehicle attributes such as pose, type, make and model. In detail, the web-nature part of the CompCars dataset provides five viewpoints as \emph{front}, \emph{rear}, \emph{side}, \emph{frontside} and \emph{rearside}, twelve vehicle types as \emph{MPV, SUV, sedan, hatchback, minibus, pickup, fastback, estate, hardtop-convertible, sports, crossover} and \emph{convertible}. To achieve an even training distribution, we discard less common vehicle types with few training images and finally select six types with all the five viewpoints illustrated in Fig. 3(a) to train our model. Besides, since color is another important attribute of a vehicle, we additionally annotate colors on more than 10,000 images with five common vehicle colors as \emph{black, white, silver, red} and \emph{blue} to train our final model. Apart from positive samples, about 200,000 negative samples without any vehicles are cropped from Google Street View Images to compose our training data.
For data augmentation, we first triple the training data with increased and decreased image intensities for making our DAVE more robust under different lighting conditions. In addition, image downsampling up to 20\% of the original size and image blurring are introduced to enable that detected vehicles with small sizes can be even annotated as precisely as possible.
\subsubsection{Jointly Training with Latent Knowledge Guidance}
The entire training structure of DAVE is illustrated in Fig. 2. We optimize the FVPN and the ALN jointly but with different sized input training data at the same time. The input resolution of the ALN is $224\times224$ for fine-grained vehicle attributes learning, while it is decreased to $60\times60$ for the FVPN to fit smaller scales of the test image pyramid for efficiency in the inference phase. In fact, the resolution of $60\times60$ can well guarantee the coarse shape and texture of a vehicle is discriminative enough against the background. Besides, another significant difference between the ALN and the FVPN is that input vehicle samples for the ALN are tightly cropped, however, for the FVPN, uncropped vehicles are used for bounding-box (labeled as $loc_t$ in Eq. (2)) regressor training.
\begin{figure*}[t]
\centering
\begin{tabular}{cc}
\includegraphics[width=0.49\textwidth,height=0.22\textheight]{data.jpg} &\includegraphics[width=0.49\textwidth,height=0.22\textheight]{training_loss.JPG}\\
\footnotesize
(a) &\footnotesize (b) \\
\end{tabular}
\caption{
\textbf{}(a) Examples of training data (columns indicate vehicle types, while rows indicate poses and colors), (b) Training loss with/without knowledge learning.}
\label{nasaacc}
\end{figure*}
The pre-trained GoogLeNet model for 1000-class ImageNet classification is used to initialize all the convolutional layers in the ALN, while the FVPN is trained from scratch. A 1024-dimensional feature vector of the \emph{pool5/7$\times$7\_s1} layer in the ALN, which can exhaustively describe a vehicle, is extracted as the latent data-driven knowledge guidance to supervise the same dimensional \emph{Conv\_fc\_knowledge} layer in the FVPN by cross entropy loss.
We first jointly train ALN and FVPN for about 10 epochs on the selected web-nature data that only contains pose and type attributes from the CompCars. In the next 10 epochs, we fine-tune the models by a subset with our complemented color annotations. Throughout the training process, we set the batch size as 64, and the momentum and weight decay are configured as 0.9 and 0.0002, respectively. Learning rate is scheduled as $10^{-3}$ for the first 10 epochs and $10^{-4}$ for the second 10 epochs. To make our method more convincing, we train two models with and without knowledge guidance, respectively. During training, we definitely discover that knowledge guidance can indeed benefit training the shallow FVPN to obtain lower training losses. Training loss curves for the first 10 epochs are depicted in Fig. 3(b).
\begin{figure} [t]
\centering
\includegraphics[width=0.975\textwidth]{inference_phase.jpg}\\
\caption{
\textbf{A two-stage inference phase of DAVE.} Vehicle proposals are first obtained from FVPN in real-time. Afterwards, we use ALN to verify each proposal and annotate each positive one with the vehicle pose, color and type.}
\label{exampless}
\end{figure}
\subsection{Two-stage Deep Nets Inference}
Once the joint training is finished, a two-stage scheme is implemented for inference of DAVE. First, the FVPN takes as input the 10-level test image Gaussian pyramid. For each level, the FVPN is operated over the input frame to infer \emph{Conv\_fc\_class} and \emph{Conv\_fc\_bbr} layers as corresponding heatmaps. All the 10 \emph{Conv\_fc\_class} heatmaps are unified into one map by rescaling all the channels to the largest size among them and keeping the maximum along channels, while the index of each maximum within 10 channels is used to obtain four unified \emph{Conv\_fc\_bbr} heatmaps (10 levels by similar rescaling). After unifying different levels \emph{Conv\_fc\_class} heatmaps into the final vehicle proposal score map, we first filter the score map with threshold $thres$ to discard low hot spots, and then local peaks on the map are detected by a circle scanner with tuneable radius $r$. In all our experiments, $r = 8$ and $thres = 0.5$ are fixed. Thus, these local maximal positions are considered as the central coordinates of proposals, ($\hat{x_i}$ ,$\hat{y_i}$). Coarse width and height of each proposal can be simply predicted based on the bounding-box of its corresponding hot spot centered on each local peak. If one hot spot contains multiple peaks, the width and height will be shared by these peaks (i.e. proposals). For preserving the complete vehicle body, coarse width and height are multiplied by fixed parameter $m = 1.5$ to generate ($\hat{w}_i^{nobbr}$ ,$\hat{h}_i^{nobbr}$). Thus, a preliminary bounding-box can be represented as ($\hat{x_i}$,$\hat{y_i}$,$\hat{w}_i^{nobbr}$ ,$\hat{h}_i^{nobbr}$). Finally, bounding-box regression offset values (within [0,1]) are extracted from four unified heatmaps of \emph{Conv\_fc\_bbr} at those coordinates ($\hat{x_i}$,$\hat{y_i}$) to obtain the refined bounding-box.
Vehicle proposals inferred from the FVPN are taken as inputs into the ALN. Although verifying each proposal and annotation of attributes are at the same stage, we assume that verification has a higher priority. For instance, in the inference phase, if a proposal is predicted as a positive vehicle, it will then be annotated with a bounding-box and inferred pose, color and type. However, a proposal predicted as the background will be neglected in spite of its inferred attributes. Finally, we perform non-maximum suppression as in RCNN \cite{girshick2014rich} to eliminate duplicate detections. The full inference scheme is demonstrated in Fig. 4. At present, it is difficult to train a model that has the capability to annotate all the vehicles with enormously rich vehicle colors and types. During inference, a vehicle with untrained colors and types is always categorized into similar classes or a catch-all ``others" class, which is a limitation of DAVE. In future work, we may expand our training data to include more abundant vehicle classes.
\section{Experiments and Results}
In this section, we evaluate our DAVE for detection and annotation of pose, color and type for each detected vehicle. Experiments are mainly divided as two parts: vehicle detection and attributes learning. DAVE is implemented based on the deep learning framework Caffe \cite{jia2014caffe} and run on a workstation configured with an NVIDIA TITAN X GPU.
\subsection{Evaluation of Vehicle Detection}
To evaluate vehicle detection, we train our models using the large-scale CompCars dataset as mentioned before, and test on three other vehicle datasets. We collect a full high definition ($1920\times1080$) Urban Traffic Surveillance (UTS) vehicle dataset with six videos which were captured from different viewpoints and illumination conditions. Each video sequence contains 600 annotated frames. To be more convincing, we also compare our method on two other public datasets: the PASCAL VOC2007 car dataset \cite{everingham2010pascal} and the LISA 2010 dataset \cite{sivaraman2010general} with four competitive models: DPM \cite{felzenszwalb2010object}, RCNN \cite{ren2015faster}, Fast RCNN \cite{girshick2015fast} and Faster RCNN \cite{ren2015faster}. These four methods obtain state-of-the-art performances on general object detection and the codes are publicly available. We adopt the trained car model in voc-release5 \cite{voc-release5} for DPM, while train models (VGG-16 version) for RCNN, Fast RCNN and Faster RCNN ourselves on the CompCars dataset to implement vehicle detection as our DAVE. The vehicle detection evaluation criterion is the same as PASCAL object detection \cite{everingham2010pascal}. Intersection over Union (IoU) is set as 0.7 to assess correct localization.
\subsubsection{Testing on the UTS dataset}
Since our FVPN can be considered as a high-accuracy proposal model for a single specific object, we test it independently. Then, results from vehicle verification by the deeper ALN (i.e., FVPN+verify in Fig. 5) are shown as our final accuracies. The detection accuracy as average precision (AP) and speed as frames-per-second (FPS) are compared in the left column of Table 1. Our model outperforms all the other methods with obvious improvements. Specifically, our method obtains an increased AP of 3.03\% compared to the best model Faster RCNN, and the shallow proposal network FVPN independently achieves 57.28\% which is only lower than Faster RCNN. The other two deep models, RCNN and Fast RCNN, do not produce satisfactory results mainly due to the low-precision proposals extracted by Selective Search \cite{uijlings2013selective}. Mixture-DPM with bounding-box prediction (MDPM-w-BB \cite{felzenszwalb2010object}) significantly improve the performance compared to MDPM-w/o-BB \cite{felzenszwalb2010object} by 10.77\%. For the evaluation of efficiency, our FVPN with a shallow and thin architecture can achieve real-time performance with 30 fps on FHD video frames. Although the deeper ALN slows the entire detection framework, 2 fps performance still shows competitive speed with Faster RCNN (4 fps).
We also test the FVPN trained without knowledge guidance, the AP decreases by 2.37\%, which explains the significant advancement of knowledge guidance. In addition, experiments are carried out to demonstrate that bounding-box regression can be helpful with the AP increased by 0.96\%.
\begin{table}
\center
\newcommand{\tabincell}[2]{\begin{tabular}{@{}#1@{}}#2\end{tabular}}
\caption{\small
\textbf{Vehicle detection AP (\%) and speed (fps) comparison on the UTS, PASCAL VOC2007 and LISA 2010 datasets}}
\label{table:t1}
\resizebox{1\textwidth}{!}{
\begin{tabular}{|c||c|c||c|c||c|c|}
\cline{1-7}
\multirow{3}{*}{\textbf{\tabincell{c}{Methods}}}&\multicolumn{2}{c||}{\textbf{UTS}} &\multicolumn{2}{c||}{\textbf{PASCAL VOC2007}}&\multicolumn{2}{c|}{\textbf{LISA 2010}} \\
\cline{2-7}
& \textbf{\tabincell{c}{Average\\ Precision (AP)}} & \textbf{\tabincell{c}{Processing\\ Speed (fps)}} & \textbf{\tabincell{c}{Average\\ Precision (AP)}} &\textbf{\tabincell{c}{Processing\\ Speed (fps)}}& \textbf{\tabincell{c}{Average\\ Precision (AP)}} &\textbf{\tabincell{c}{Processing\\ Speed (fps)}}\\
\hline
\hline
MDPM-w/o-BB&41.96\% &0.25 &48.44\% &1.25 &63.61\% &0.7 \\
\hline
MDPM-w-BB&52.73\% &0.2 &57.14\% &1.25 &72.89\% &0.7 \\
\hline
RCNN&44.87\% &0.03 &38.52\% &0.08 &55.37\% &0.06 \\
\hline
FastRCNN&51.58\% &0.4 &52.95\% &0.5 &53.37\% &0.5 \\
\hline
FasterRCNN&59.82\% &\textbf{4} &63.47\% &\textbf{6} &77.09\% &\textbf{6} \\
\hline
\hline
FVPN-w/o-knowledge guide &54.91\% &30 &58.12\% &46 &72.37\% &42 \\
\hline
FVPN-w/o-bbr&56.32\% &30 &58.93\% &46 &71.82\% &42 \\
\hline
\textbf{FVPN}&57.28\% &\textbf{30} &60.05\% &\textbf{46} &73.46\% &\textbf{42} \\
\hline
\textbf{FVPN+Verify}&\textbf{62.85\%} &2 &\textbf{64.44\% }&4 &\textbf{79.41\%} &4 \\
\hline
\end{tabular}
}\tiny
\\``bbr" indicates the bounding-box regressor used in our model, while ``BB" denotes bounding-box prediction used in DPM model. ``w" and ``w/o" are the abbreviations of ``with" and ``without", respectively. ``Verify" denotes the vehicle verification in the ALN.
\vspace{-4ex}
\end{table}
\subsubsection{Testing on the PASCAL VOC2007 car dataset and the LISA 2010 dataset}
To make our methods more convincing, we also evaluate on two public datasets. All the images containing vehicles in the trainval and test sets (totally 1434 images) in the PASCAL VOC 2007 dataset are extracted to be evaluated. In addition, the LISA 2010 dataset contains three video sequences with low image quality captured from a on-board camera.
All the results are shown in the middle and right columns of Table 1. For the PASCAL VOC2007 dataset, our model (FVPN+verify) achieves 64.44\% in AP, which outperforms MDPM-w-BB, RCNN, FastRCNN and Faster RCNN by 7.3\%, 25.92\%, 11.49\% and 0.97\%, respectively. Likewise, our FVPN can even obtain a high AP of 60.05\%. For the LISA 2010 dataset, the highest accuracy of 79.41\% by our model beats all other methods as well. Therefore, it demonstrates that our method is able to stably detect vehicles with different viewpoints, occlusions and image qualities.
Fig. 5 presents the precision-recall curves of all the compared methods on UTS, PASCAL VOC2007 car dataset and the LISA 2010 dataset, respectively. From all these figures, we can further discover that, for all three datasets, FVPN+Verify achieves better performance than other detection methods by comparing Area Under the Curve (AUC). Besides, some qualitative detection results including successful and failure cases are shown in Fig. 6. It can be observed that the FVPN cannot handle highly occluded cases at very small sizes, since local peaks on the FVPN heatmap for vehicles in those cases will be overlapped. The similar situation also exists in most of the deep networks based detection approaches \cite{girshick2014rich,girshick2015fast,redmon2015you,ren2015faster}.
\begin{figure*} [t]
\centering
\begin{tabular}{ccc}
\includegraphics[width=0.33\textwidth,height=0.18\textheight]{UTS.JPG} &\includegraphics[width=0.33\textwidth,height=0.18\textheight]{Pas.JPG}&\includegraphics[width=0.33\textwidth,height=0.18\textheight]{Lisa.JPG}\\
\footnotesize
(a) &\footnotesize (b) &\footnotesize (c)\\
\end{tabular}
\caption{
\textbf{}Precision-recall curves on three vehicle datasets}
\label{nasaacc}
\end{figure*}
\begin{figure} [t]
\centering
\includegraphics[width=0.95\textwidth]{qualitative_results.pdf}\\
\caption{
\textbf{}Examples of successful and failure cases for detection (a green box denotes correct localization, a red box denotes false alarm and a blue box denotes missing detection)}
\label{exampless}
\end{figure}
\subsection{Evaluation of Attributes Learning}
The experiments and analysis of the ALN are mainly based on the CompCars dataset and the UTS dataset. The web-nature data in the CompCars dataset are labeled with five viewpoints and twelve types about 136000 and 97500 images, respectively. We neglect those images without type annotation and randomly split the remaining data into the training and validation subsets as 7:3. In addition to pose estimation and type classification, we complement the annotation of five common vehicle colors on about 10000 images for evaluation of color recognition. Besides, for type classification, we compare the results of both the selected 6 common vehicle types and the total 12 types as mentioned in Section 3. 3. Vehicle verification (i.e., binary classification of vehicle and background) is evaluated in all the experiments as well.
In the following subsections, we implement four different experiments to investigate the gain of the multi-task architecture, the accuracy by inputs with different image qualities, the effect of layer depths and the difficulty of fine-grained classification under different viewpoints.
\begin{table}
\center
\newcommand{\tabincell}[2]{\begin{tabular}{@{}#1@{}}#2\end{tabular}}
\caption{\small
\textbf{Evaluation (\%) of attributes annotation for vehicles on the UTS dataset}}
\label{table:t1}
\resizebox{0.975\textwidth}{!}{
\begin{tabular}{|c|c|c|c|c|c|}
\cline{1-6}
\multirow{2}{*}{\textbf{\tabincell{c}{Tasks}}}&\multirow{2}{*}{\textbf{Vehicle Verification}} &\multirow{2}{*}{\textbf{Pose Estimation}}&\multicolumn{2}{c|}{\textbf{\tabincell{c}{Vehicle Type\\ Classification}}}&\multirow{2}{*}{\textbf{Color Recognition}} \\
\cline{4-5}
& & & \tabincell{c}{\textbf{12 types}} & \tabincell{c}{\textbf{6 types}} &\\
\hline
\hline
\multicolumn{6}{|c|}{\textbf{Comparison of single-task learning (STL) and multi-task learning (MTL) for attributes prediction}}\\
\hline
STL &98.73 &96.94 &60.37 &88.32 &78.33 \\
MTL &\textbf{99.45} &98.03 &69.64 &94.91 &\textbf{79.25 }\\
STL feature+SVM& 99.11&97.12 & 60.86&90.75 & 78.06\\
MTL feature+SVM&99.36 &\textbf{98.10} &\textbf{69.86} &\textbf{95.12} &79.19 \\
\hline
\hline
\multicolumn{6}{|c|}{\textbf{Comparison of Attributes prediction with different sizes of vehicle images}}\\
\hline
$28\times28$&90.45 &83.49 &37.52 &53.66 &49.73 \\
$56\times56$&98.12 &91.33 &52.02 &77.02 &66.14 \\
$112\times112$&99.37 &96.56 &63.41 &90.67 &\textbf{80.31} \\
$224\times224$&\textbf{99.45 }&\textbf{98.03} & \textbf{69.64}&\textbf{94.91} &79.25 \\
\hline
\hline
\multicolumn{6}{|c|}{\textbf{Comparison of Attributes prediction with different deep models}}\\
\hline
ALN based on FVPN ($depth=4$)&95.96 &81.21 &27.26 &43.12 &65.12 \\
ALN based on AlexNet ($depth=8$)&\textbf{99.51} &95.76 &66.01 &89.25 & 77.90\\
ALN based on GoogLeNet ($depth=22$)&99.45 &\textbf{98.03} &\textbf{69.04} &\textbf{94.91} &\textbf{79.25} \\
\hline
\end{tabular}
}
\end{table}
\subsubsection{Single-task learning or multi-task learning? }
We first compare the multi-task ALN with the case of training networks for each attribute separately (i.e., single task). In addition, results by the combination of deep learned features and an SVM classifier are compared as well. All the model architectures are based on the GoogLeNet, and 1024-dimensional features are extracted from layer \emph{pool5/7$\times7$\_s1} to train the corresponding SVM classifier \cite{CC01a}. As shown in the top part of Table 2, the multi-task model consistently achieves higher accuracies on four different tasks, which reveals the benefit of joint training. Although the combination of extracted features and SVM classifiers sometimes can lead to a small increase, we still prefer the proposed end-to-end model because of its elegance and efficiency.
\subsubsection{How small a vehicle size can DAVE annotate?}
Since vehicles within surveillance video frames are usually in different sizes. Visual details of those vehicles far from the camera are significantly unclear. Although they can be selected by the FVPN with coarse requirements, after rescaling to $224\times224$, these vehicle proposals with low image clarity are hard to be annotated with correct attributes by the ALN. To explore this problem, we test vehicle images with original sizes of 224, 112, 56 and 28 using the trained ALN. The middle part of Table 2 illustrates that the higher resolution the original input size is, the better accuracy it can achieve.
\subsubsection{Deep or shallow?}
How deep of the network is necessary for vehicle attributes learning is also worth to be explored. Since our ALN can be established on different deep models, we compare popular deep networks: AlexNet \cite{krizhevsky2012imagenet} and GoogLeNet with 8 layers and 22 layers, respectively. As VGGNet (16 layers version) \cite{simonyan2014very} configured with numerous parameters requires heavy computation and large memory, we do not expect to employ it for our ALN. Besides, our proposed shallow FVPN with 4 layers is also used for attributes learning. From the bottom part of Table 2, we can see that a deeper network does not obtain much better performance on vehicle verification compared to a shallow one. However, for pose estimation, type classification and color recognition, the deepest GoogLeNet consistently outperforms other nets with obvious gaps. Particularly for type classification which belongs to fine-grained categorization, the shallow FVPN gives extremely poor results. It illustrates that a deeper network with powerful discriminative capability is more suitable for fine-grained vehicle classification tasks.
\subsubsection{Fine-grained categorization in different views.}
Finally, since vehicle type classification belongs to fine-grained categorization, we are interested in investigating its difficulty in different views due to its importance for our future work such as vehicle re-identification. As demonstrated in Table 3, for both 12-type and 6-type classification, higher precision is easier to be achieved from side and rearside views, while it is difficult to discriminate vehicle types from the front view. In other words, if we aim to re-identify a target vehicle from two different viewpoints, the type annotation predicted from a side view is more credible than that from a front view.
\begin{table}
\center
\newcommand{\tabincell}[2]{\begin{tabular}{@{}#1@{}}#2\end{tabular}}
\caption{\small
\textbf{Evaluation (\%) of fine-grained vehicle type classification on the UTS dataset}}
\label{table:t1}
\resizebox{0.65\textwidth}{!}{
\begin{tabular}{|c|c|c|c|c|c|}
\hline
Number of vehicle type&Front &Rear &Side &FrontSide &RearSide \\
\hline
12&58.02 &60.37 &66.73 &61.28 &64.90 \\
\hline
6&79.42 &84.60 &92.93 &86.77 &92.53 \\
\hline
\end{tabular}
}
\end{table}
Fig. 7 shows the qualitative results of our DAVE evaluated on the UTS vehicle dataset. It demonstrates that our model is robust to detect vehicles and annotate their poses, colors and types simultaneously for urban traffic surveillance. The failure cases mainly take place on incorrect colors and types.
\begin{figure}
\centering
\includegraphics[width=0.95\textwidth]{annotation_results.pdf}\\
\caption{\footnotesize
\textbf{}Qualitative results of attributes annotation. Red marks denote incorrect annotation, and N/A(C) means a catch-all color.}
\label{exampless}
\end{figure}
\section{Conclusion}
In this paper, we developed a unified framework for fast vehicle detection and annotation: DAVE, which consists of two convolutional neural networks FVPN and ALN. The proposal and attributes learning networks predict bounding-boxes for vehicles and infer their attributes: pose, color and type, simultaneously. Extensive experiments show that our method outperforms state-of-the-art frameworks on vehicle detection and is also effective for vehicle attributes annotation. In our on-going work, we are integrating more vehicle attributes such as make and model into the ALN with high accuracy, and exploiting these attributes to investigate vehicle re-identification tasks.
\bibliographystyle{splncs}
| 4999adac1aa8415d7884d8a26591dfa9ee497e58 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Given a smooth manifold $M$, assume a family of smooth immersions in a semi-Rieman\-nian manifold $(\mathbf{M},\mathbf{g})$, $F_t:M\rightarrow \mathbf{M}$, $t\in [0,\delta)$, $\delta>0$, with mean curvature vector $\vec{H}_t$. The initial immersion $F_0$ is called a solution to the \textit{mean curvature flow} (up to local diffeomorphism) if
\begin{equation}\label{MCF} \left(\frac{d}{dt} F_t\right)^{\perp} = \vec{H}_t,
\end{equation}
where $\perp$ means the orthogonal projection on the normal bundle. In the Euclidean and Minkowski space, there is a famous family of such immersions, namely, translating solitons. A submanifold is called \textit{translating soliton} in the Euclidean Space when its mean curvature $\vec{H}$ satisfies the following equation:
\begin{equation} \label{soliton}
\vec{H} = v^{\perp},
\end{equation}
for some constant unit vector $v\in\mathbb{R}^{n+1}$. Indeed, if a submanifold $F:M\rightarrow \mathbb{R}^{n+1}$ satisfies this condition, then it is possible to define the forever flow $\Gamma:M\times[0,+\infty)\rightarrow \mathbb{R}^{n+1}$, $\Gamma(p,t)=F_t(p)=F(p)+tv$. Clearly,
\[ \left(\frac{d}{dt} F_t\right)^{\perp} = v^{\perp}=\vec{H}.
\]
This justifies our definition. Such solutions have been widely studied in the case where the ambient space is the Euclidean (or the Minkowski) space. Probably, the most famous examples are the Grim Reaper curve in $\mathbb{R}^2$, and the translating paraboloid and translating catenoid, \cite{CSS}. For a good list of other examples, see \cite{MSHS}. Also, in \cite{N} there are some examples with complicated topology. In \cite{ALR}, the authors focus on the weak maximum principle applied to these objetcs.
The starting point of this paper is the fact that the translating paraboloid and translating catenoid are rotationally symmetric, namely, invariant by the Lie group $SO(n)$ acting by isometries. We revise the concept of translating solitons when the ambient manifold is a semi-Riemannian product of a semi-Riemannian manifold and the real line. Needless to say, this includes the Riemannian setting. Second, we are interested in constructing translating solitons having a nice behaviour under the action of Lie groups. We show a technique to construct such examples in different semi-Riemannian manifolds. More precisely, assume that $(M,g)$ is a connected semi-Riemann manifold of dimension $n\geq 2$ and index $0\leq \alpha \leq n-1$. Given $\varepsilon=\pm 1$, we construct the semi-Riemannian product $\bar{M}=M\times\mathbb{R}$ with metric $\produ{,}=g+\varepsilon\mathrm{d}t^2$. The vector field $\partial_t\in\mathfrak{X}(\bar{M})$ is obviously Killing and unit, spacelike when $\varepsilon=+1$ and timelike when $\varepsilon=-1$. Now, let $F:\Gamma\rightarrow \bar{M}$ be a submanifold with mean curvature vector $\vec{H}$. Denote by $\partial_t^{\perp}$ the normal component of $\partial_t$ along $F$.
\begin{definition} With the previous notation, we will call $F$ a {\normalfont (vertical) translating soliton of the mean curvature flow}, or simply, a {\normalfont translating soliton}, if $\vec{H}=\partial_t^{\perp}$.
\end{definition}
To justify this definition we remark that if $F(x)=(F_1(x),F_2(x))$ for any $x\in \Gamma$, we can define the map
\[
F:\Gamma \times \mathbb{R} \rightarrow \bar{M}, \quad F(x,t)=(F_1(x),F_2(x)+t).
\]
Hence each $F(-,t)$ is a submanifold with associated mean curvature vector $\vec{H}_t$ and therefore, up to tangential diffeomorphisms, provides a forever solution to \eqref{MCF} invariant by the Killing vector field $\partial_t$. This clearly generalizes the classical definition of translating solitons in $\mathbb{R}^{n+1}$. Notice that due to the rich group of isometries of $\mathbb{R}^{n+1}$, considering any constant unit vector field $v$ such that $\vec{H}=v^{\perp}$ is equivalent to considering $v=\partial_t$.
In this paper, we will focus on \textit{graphical} translating solitons. Namely, given $u\in C^2(M)$, we construct its graph map $F:M\rightarrow M\times\mathbb{R}$, $F(x)=(x,u(x))$. If $\nu$ is the upward normal vector along $F$ with $\varepsilon'=\mathrm{sign}(\produ{\nu,\nu})=\pm 1$, we characterize in Proposition \ref{basicequation} the graphical translating solitons in $(\bar{M},\produ{,})$ as those satisfying the following PDE:
\begin{equation}
\label{fundamentalequation} \mathrm{div}\left(\frac{ \grad u } { \sqrt{\varepsilon'\big(\varepsilon+\vert \grad u\vert_g^2\big)}} \right)=
\frac{1}{ \sqrt{\varepsilon'\big(\varepsilon+\vert \grad u\vert_g^2\big)}}.
\end{equation}
This equation is in general difficult to solve. The geometric way to reduce the number of variables is to consider on the manifold $M$ a foliation by orbits of a Lie group action. But we can often understand the actions by Lie groups as particular cases of submersions. Thus, we first consider the general situation where $M$ submerses to a base manifold $B$ and prove the following result:
\begin{center}
\parbox{0.88\textwidth}{
\textit{
Let $\pi:(M,g_M)\rightarrow (B,g_B)$ be a harmonic semi-Riemannian submersion. Given $u \in C^2(B)$, let $F:B\rightarrow B\times\mathbb{R}$, $F(x)=(x,u(x))$ be its graph map, and $\widetilde{F}:M\rightarrow M\times\mathbb{R}$ be the map $\widetilde{F}(x)=(x,\pi\circ u(x))$. Then $\widetilde{F}$ is a graphical translating soliton if and only if $F$ is a graphical translating soliton.
}}
\end{center}
In fact, we obtain a more general result in Theorem \ref{bundle2}. Indeed, if we take into account the mean curvature of the fibers in the direction of the normal to the map $\tilde{F}$, we show that the existence of the translating soliton $\widetilde{F}$ in $M$ is equivalent to the existence of what we call an \textit{H-perturbed translating soliton $F$} in $B$.
Moreover, an important particular case arise when $\pi:M\rightarrow B=I$, where $B=I$ is an open interval with metric $\widetilde{\ep} ds^2$ ($\widetilde{\ep} =\pm 1$). Clearly, the fibers of the submersion will be hypersurfaces, and we assume that each fiber $\pi^{-1}(s)$, $s\in I$, has constant mean curvature $h(s)$. Since any function $u$ defined on $M$ which is constant along the fibers, projects to another function $f:I\rightarrow\mathbb{R}$ such that $u=f\circ \pi$, we show in Theorem \ref{equiv_ODE} that equation \eqref{fundamentalequation} reduces to the following ODE,
\begin{equation} \label{keyODE}
f''(s) = \big(\widetilde{\ep} +\varepsilon f'(s)^2\big) \big( 1 - f'(s)h(s) \big).
\end{equation}
Next, we let a Lie group $\Sigma$ act on the manifold in a nice way. In \cite{Mos57}, it is shown that when $\Sigma$ is compact, and there is at least a good orbit with codimension 1, then, the space of orbits is either $\mathbb{S}^1$ or an interval. We will work in the general setting where $\Sigma$ acts by isometries and the projection to the space of orbits $\pi:M\rightarrow M/\Sigma$ is a well-defined smooth map, where $M/\Sigma$ is diffeomorphic to an open interval. We show that if the gradient of the projection is nowhere lightlike, then up to composing with another function, $\pi$ can be assumed to be a semi-Riemannian submersion with constant mean curvature fibers. We state here a more geometric and milder version of Theorem \ref{existence}, which is a consequence of Theorem \ref{equiv_ODE}.
\begin{center}
\parbox{0.88\textwidth}{
\textit{Let $(M,g)$ be a connected semi-Riemannian manifold. Let $\Sigma$ be a Lie group acting by isometries on $M$ and $\pi:(M,g_M)\rightarrow (I,\widetilde{\varepsilon}ds^2)$ be a semi-Riemannian submersion such that the fibers of $\pi$ are orbits of the action, with function $h:I\to\mathbb{R}$ representing the mean curvature of the fibers. Given $u\in C^2(M)$, consider its graph map
$F:M\rightarrow M\times\mathbb{R}$, $F(x)=(x,u(x))$ for any $x\in M$. Then, $F$ is a $\Sigma$-invariant translating soliton if, and only if, there exists a solution $f\in C^2(I,\mathbb{R})$ to \eqref{keyODE} such that $u=f\circ\pi$.
}}
\end{center}
This leads to the existence of translating solitons which are the union of two graphical ones, as stated in Corollary \ref{oide}. We recall that the \textit{translating catenoid} in $\mathbb{R}^3$ is obtained in this way in \cite{CSS}, and it is one of the main classical examples. Thus, we will be able to obtain translating solitons with two ends in some manifolds. Second, we show in Corollary \ref{globalexamples} that, under certain assumptions on the signature and for certain values of the initial value problems, solutions exist on the whole manifold $M$.
Last Section is devoted to obtaining specific examples. All of them share a common technique: We start with a manifold $M$ and a Lie group of isometries $\Sigma$ such that the space of orbits $M/\Sigma$ is an open interval. The associated ODE \eqref{keyODE} deeply relies on the auxiliary function $h:M/\Sigma\rightarrow\mathbb{R}$, which depends on the case. When the ODE becomes a problem with a singularity, we show the existence of local solutions
by using dynamical systems, and we lift them up to the original manifold $M$ to obtain translating solitons.
Inspired by the Grim Reaper Cylinder in $\mathbb{R}^n$, we construct many others by using product manifolds. Needless to say, we recover the already known rotationally symmetric translating solitons in the Euclidean and Minkowski Spaces, but we obtain among others a new \textit{translating catenoid} in Minkowski Space, rotationally symmetric translating solitons in the De Sitter Space and the Hyperbolic Space. By a gluing technique from \cite{BCO}, we construct a $C^{\infty}$, boost invariant, translating soliton in Minkowski Space by gluing (up to) 4 pieces. The goal is not to provide an exhaustive list, but rather to illustrate our technique, and make it useful for future works.
\section{Setup}
The following proposition is well-known in the Euclidean setting (see for example \cite{LTW}). The proof is similar in our context, but we include it for completeness and discuss the necessary modifications.
\begin{proposition}\label{basicequation}
Let $(M,g)$ be a semi-Riemannian manifold, $u:M\rightarrow \mathbb{R}$ a\, $\mathcal{C}^2$ function, and let $F:M\rightarrow M\times\mathbb{R}=:\bar{M}$, $F(x)=(x,u(x))$ be its graph map. Given $\varepsilon=\pm 1$, assume that $F:(M,\gamma=F^*\produ{,})\rightarrow (\bar{M},\produ{,}=g+\varepsilon dt^2)$ is a semi-Riemannian hypersurface with unit upward normal $\nu$ such that $\produ{\nu,\nu}=\varepsilon'=\pm 1$. Then, $F$ is a (vertical) translating soliton if, and only if, function $u$ satisfies \eqref{fundamentalequation}.
\end{proposition}
\begin{proof}
Note that, under the usual identifications, for each $X\in TM$, we have
\[ dF(X) = (X,du(X)) = (X,0)+du(X)\partial_t = (X,g(\grad u,X)) =
(X,0) + g(\grad u,X)\partial_t,
\]
where $\grad u$ is the $g$-gradient of $u$. We consider the metric $\gamma=F^*\produ{,}$ on $M$, but we have to assume that $(M,\gamma)$ is a semi-Riemannian submanifold. In particular, the following function is constant, $\varepsilon'=\mathrm{sign}\big(\varepsilon+\vert\grad u\vert_g^2\big)=\pm 1$.
Therefore, the upward normal vector field is
\begin{equation}\label{normal} \nu = \frac{1}{W}\big(-\varepsilon(\grad u,0)+\partial_t\big), \quad W=+\sqrt{\varepsilon'\big(\varepsilon+\vert\grad u\vert_g^2\big)}.
\end{equation}
Needless to say, $\produ{\nu,\nu}=\varepsilon'$. Then, we consider a local $g$-orthonormal frame $B=(e_1,\ldots,e_n)$ such that $g(e_i,e_j)=\varepsilon_i\delta_{ij}$, for any $i,j=1,\ldots,n$. We denote $u_i=du(e_i)$, $i=1,\ldots,n$. For this frame, we compute the induced metric $\gamma=F^{*}\produ{,}$, so that the coefficients of the Gram matrix are
\[ \gamma_{ij}=\gamma(e_i,e_j) = \varepsilon_i\delta_{ij}+\varepsilon u_iu_j.
\]
Then, the inverse matrix is
\[ \gamma^{ij} = \varepsilon_i\delta_{ij} -\frac{\varepsilon'}{W^2}\varepsilon_i\varepsilon_ju_iu_j.
\]
Now, if $II$ is the second fundamental form of $F$, then the mean curvature vector of $F$ is
\[ \vec{H} = \mathrm{tr}_{\gamma}(II) = \varepsilon' \produ{\vec{H},\nu} \nu
=\varepsilon' \produ{\partial_t^{\perp},\nu}\nu = \varepsilon' \produ{\partial_t,\nu}\nu
=\frac{\varepsilon\ep'}{W}\nu.
\]
But in out setting, what really matters is $ \produ{\nu,\vec{H}} =\frac{\varepsilon}{W}.$ On the other hand, let $\bar\nabla$ the Levi-Civita connection of $(\bar{M},\produ{,})$. We recall O'Neill's book \cite{ON}, and its equations for the Levi-Civita connection of a (warped) product. Thus,
\begin{align*}
&\produ{\nu,\vec{H}} =\produ{\nu,\mathrm{tr}_\gamma(II)}
=\sum_{i,j}\gamma^{ij}\produ{\nu,\bar\nabla_{dF(e_i)}dF(e_j)}
= \sum_{i,j}\gamma^{ij} \frac{\varepsilon}{W} du_j(e_i) \\
& = \frac{\varepsilon}{W}\mathrm{div}(\grad u) -\varepsilon\ep'\sum_{i,j}\frac{\varepsilon_i\varepsilon_ju_iu_j}{W^3}e_i(e_j(u)).
\end{align*}
Now, we compute
\[ \produ{\grad\Big(\frac{1}{W}\Big),\grad u }
=\frac{-1}{W^2}\sum_i \varepsilon_iu_ie_i\Big( \sqrt{\varepsilon'\big(\varepsilon+\vert\grad u\vert_g^2\big)} \Big)
= \frac{-\varepsilon'}{W^3} \sum_{i,j} \varepsilon_i\varepsilon_ju_iu_je_i(e_j(u)).
\]
All together,
\[\produ{\nu,\vec{H}} =\frac{\varepsilon}{W}\mathrm{div}(\grad u) +\varepsilon \produ{\grad\Big(\frac{1}{W}\Big),\grad u } =
\varepsilon \mathrm{div}\Big( \frac{\grad u}{W}\Big).
\]
\end{proof}
\begin{remark}\normalfont
If we call $H=\mathrm{tr}(A)/n$ the mean curvature function of the hypersurface, $A$ being the shape operator, then $\produ{\nu,\vec{H}}=n H \varepsilon'$. In particular,
\[
\mathrm{div}\left(\frac{ \grad u } { \sqrt{\varepsilon'\big(\varepsilon+\vert \grad u\vert_g^2\big)}} \right)=
\frac{1}{ \sqrt{\varepsilon'\big(\varepsilon+\vert \grad u\vert_g^2\big)}} = \varepsilon' n H.
\]
\end{remark}
\begin{corollary} Let $(M,g)$ be a compact, without boundary, orientable Riemannian manifold. Then, $M$ does not admit any globally defined graphical Translating Soliton $F:M\rightarrow (M\times\mathbb{R},g+\varepsilon dt^2)$, for $\varepsilon=\pm 1$.
\end{corollary}
\begin{proof} Assume that there exists a globally defined graphical Translating Soliton on $M$. Then, for some function $u\in C^2(M)$, equation \eqref{fundamentalequation} holds true. By using the volume form $d\mu_g$, we obtain
\[ 0 = \int_M \mathrm{div}\left(\frac{\nabla u}{\sqrt{\varepsilon'(1+\vert \nabla u\vert^2)}}\right) d\mu_g =
\int_M \frac{1}{\sqrt{\varepsilon'(1+\vert \nabla u\vert^2)}} d\mu_g >0.
\]
This is a contradiction.
\end{proof}
\section{Submersions}
We first recall shortly some facts about submersions. For more details we refer to {\cite{Escobales} and the classical O'Neill's book \cite{ON}. If $\pi:(M,g_M)\rightarrow(B,g_B)$ is a semi-Riemannian submersion, we can consider the following decomposition of the tangent bundle $TM$ $$TM=\mathcal{H}\oplus\mathcal{V},$$
where $\mathcal{V}:=\ker{d\pi}$ is the vertical distribution consisting of vectors tangent to the fibers and $\mathcal{H}$ is the horizontal complement with respect to $g_M$. Note that $\mathcal{V}$ is integrable, but it is usually not true for the horizontal distribution.
Every vector field $X$ in $TM$ can then be uniquely
written as $X=\mathcal{H}X+\mathcal{V}X$ where
$\mathcal{H}X$ (resp. $\mathcal{V}X$) is the horizontal (resp.
vertical) component. For completeness and clearness sake, we add the following
\begin{definition}
A vector field $X\in\Gamma(TM)$ is
called horizontal, if for all $x\in M$,
$X_{x}\in\mathcal{H}_{x}$ and vertical,
if for all $x\in M$,
$X_{x}\in\mathcal{V}_{x}$. It is called projectable, if there exists a vector field $\check{X}\in\Gamma(TB)$ such
that for all $x\in M$,
$d\pi(X_{x})=\check{X}_{\pi(x)}$, $X$ and
$\check{X}$ are called $\pi$-related. It is called basic, if it is projectable and horizontal.
\end{definition}
We add the following useful
\begin{remark}\label{remark_submersions} \normalfont
\begin{itemize}
\item[1)] Obviously, from the definition of $\mathcal{V}$, a
vector field $X\in\Gamma(TM)$ is vertical, if and only if
it is $\pi$-related to the null section of $TB$.
\item[2)] For every vector field $\check{X}$ in $TB$ there exists a unique horizontal vector field in $\Gamma(TM)$ which is $\pi$-related to $\check{X}$. It is called the horizontal lift of $\check{X}$. We will denote it by $X^h$. Moreover we will denote by $\mathcal{H}(X)$ the horizontal component of a vector field $X$ of $TM$.
\item[3)] If $X$, $Y$ are basic vector fields on $M$ $\pi$-related to $\check{X}$, $\check{Y}$, then
$g_M(X,Y)=\pi^*g_B(\check{X},\check{Y})$. Moreover $\mathcal{H}\nabla^M_XY$ is the basic vector field corresponding to $\pi^*\nabla^B_{\check{X}}\check{Y}$, where $\nabla^M$ (respectively $\nabla^B$) are the Levi-Civita connection on $M$ (respectively $B$).
\end{itemize}
\end{remark}
Finally let $W$, $Z$ be vector fields in $TM$. We recall the definition of the two O'Neill fundamental tensors.
\begin{eqnarray}
T_WZ&:=&\mathcal{H}\nabla^M_{\mathcal{V}W}(\mathcal{V}Z)+\mathcal{V}\nabla^M_{\mathcal{V}W}(\mathcal{H}Z),\nonumber\\
A_WZ&:=&\mathcal{V}\nabla^M_{\mathcal{H}W}(\mathcal{H}Z)+\mathcal{H}\nabla^M_{\mathcal{H}W}(\mathcal{V}Z)
\end{eqnarray}
Note that if $W$ and $Z$ are vertical, $T$ is the second fundamental form of the fibers. Hence the fibers of the submersion are totally geodesic if and only if $T\equiv 0$. We recall that the submersion is harmonic, if and only if the mean curvature of the fiber $h=0$. Further, let $X$, $Y$ be horizontal vectors, and $U$, $V$ be vertical vectors we recall the following useful formulas
\begin{eqnarray}\label{connection_ONtensors}
\nabla^M_UV&=&T_UV+\hat{\nabla}_UV, \quad \nabla^M_UX=\mathcal{H}\nabla^M_VX+T_UX\\
\nabla^M_XU&=&A_XU+\mathcal{V}\nabla^M_XU,\quad \nabla^M_XY=\mathcal{H}\nabla^M_XY+A_XY
\end{eqnarray}
where $\hat{\nabla}$ is the connection on the fibers.
\begin{theorem}\label{bundle2}
Let $\pi:(M,g_M)\rightarrow (B,g_B)$ be a semi-Riemannian submersion
with constant mean curvature fibers.
Given $u \in C^2(B)$, let $F:B\rightarrow B\times\mathbb{R}$, $F(x)=(x,u(x))$ be its graph map, and $\widetilde{F}:M\rightarrow M\times\mathbb{R}$ be the map $\widetilde{F}(x)=(x,u\circ \pi(x) )$. Let $\nu_F$, $\nu_{\widetilde{F}}$ be the unit upward vector along $F$ and $\widetilde{F}$, respectively, and $H_{\mathrm{fib}}^{\nu_{\widetilde{F}}}$ be the mean curvature of the fibers of $\pi\times 1:M\times\mathbb{R}\rightarrow B\times\mathbb{R}$ in the direction of $\nu_{\widetilde{F}}$.
Then $\widetilde{F}$ is a graphical translating soliton if and only if $F$ satisfies the equation $\vec{H}_{F}=\partial_t^{\perp}+H_{\mathrm{fib}}^{\nu_{\widetilde{F}}}\nu_F$.
\end{theorem}
\begin{proof} Consider the submersion $\pi:(M,g_M)\rightarrow (B,g_B)$.
$\widetilde{F}(x)=(x,v(x))$ for any $x\in M$, where $v:M\rightarrow \mathbb{R}$ is the height function. Similarly, $F(p)=(p,u(p))$ for $p\in B$, $u:B\rightarrow\mathbb{R}$, and by assumption $u\circ\pi = v$.
Now locally for every $x_0\in M$, we take an open neighborhood $U$ of $x_0$ and a collection
of projectable vector fields $\{e_i\}_{i=1}^n$ such that $\{e_i(x)\}_{i=1}^n$ is a basis for $T_xM$,
for all $x$ in $U$. Note that we can assume that $e_1,\ldots,e_k$ are vertical
and $e_{k+1},\ldots,e_{n}$ are horizontal w.r.t. $\pi$. Then by Remark \ref{remark_submersions}
\begin{eqnarray*}
g_M((\grad^Bu)^{h},e_i) =\pi^*g_B(\grad^Bu,d\pi(e_i)) = \pi^*du(d\pi(e_i)) = dv(X_i)=g_M(\grad^Mv,e_i),
\end{eqnarray*}
Hence $(\grad^Bu)^h=\grad^Mv$ and, since the submersion is semi-Riemannian, $\vert\grad^M v\vert ^2 = \vert \grad^B u\vert^2$.
Next, $\pi\times 1:\bar{M}=(M\times \mathbb{R},g_M+\varepsilon dt^2)\rightarrow \bar{B}=(B\times \mathbb{R},g_B+\varepsilon dt^2)$ is another semi-Riemannian submersion, satisfying that the horizontal lift of $\partial_t$ is $\partial_t$. Since $\vert\grad^M v\vert ^2 = \vert \grad^B u\vert^2$, we have $W^2=\varepsilon'(\varepsilon+\vert\grad^M v\vert ^2) =\varepsilon'(\varepsilon+ \vert \grad^B u\vert^2)$.
Notice that obviously by construction $\nu_{_{\widetilde{F}}}$ is horizontal.
Moreover for any $(p,t)\in B\times \mathbb{R}$, $(\pi\times 1)^{-1}(p,t)=\pi^{-1}(p)\times\{t\}$.
By equation \eqref{normal}, the horizontal lift w.r.t. $\pi\times 1$ of the normal $\nu_{F}$ in $B\times\mathbb{R}$ is exactly the normal $\nu_{\widetilde{F}}$ in $M\times \mathbb{R}$.
By extending the basis $\{e_i\}_{i=1}^n$ of $TM$, we can now construct the local orthonormal frame \newline $(e_1,\ldots,e_n,e_{n+1}=\partial_t)$ of $T\bar{M}$, where $\varepsilon_i=g_{\bar{M}}(e_i,e_i)=\pm 1$, depending on the signature of $M$. Note that $\{v_i:=d(\pi\times 1)e_i \vert k+1\leq i\leq n+1\} $ is a local orthonormal frame of $T\bar{B}$ with same signs. Now for the vertical vector fields $e_i$, $1\leq i\leq k$, and the horizontal vector field $\nu_{\widetilde{F}}$ we get using formulas \eqref{connection_ONtensors}
\[ g_{\bar{M}} \big( \nabla_{e_i}^{\bar{M}} \nu_{{\widetilde{F}}},e_i\big) =
g_{\bar{M}}\big( \mathcal{H}\nabla_{e_i}^{\bar{M}} \nu_{\widetilde{F}},e_i\big)+g_{\bar{M}}\big( T_{e_i}\nu_{\widetilde{F}},e_i\big),
\]
Hence $\nabla_{e_i}^{\bar{M}} \nu_{\widetilde{F}}$ is horizontal if and only if the second term of the right hand side vanishes. Since the vector fields $e_i$ are vertical, they are by definition tangent to the leaves of the submersion, and the horizontal vector field $\nu_{\widetilde{F}}$ is normal to the leaves. Hence $\sum_i^kg_{\bar{M}}\big( T_{e_i}\nu_{\widetilde{F}},e_i\big)$ is exactly the mean curvature $H_{\mathrm{fib}}^{\nu_{\widetilde{F}}}$ of the fibers in the direction of $\nu_{\widetilde{F}}$.
Denoting hence respectively by $II_{\widetilde{F}}$ and $II_F$ the second fundamental forms of $\widetilde{F}$ and $F$ with associated mean curvature vectors $\vec{H}_{\widetilde{F}}$ and $\vec{H}_F$ we use Remark \ref{remark_submersions} to compute
\begin{align*}
& \vec{H}_{\widetilde{F}} = \sum_i^n \varepsilon_i II_{\widetilde{F}}(e_i,e_i)=
\sum_i ^n\varepsilon_i g_{\bar{M}}(\nabla^{\bar{M}}_{e_i}\nu_{\widetilde{F}},e_i)\nu_{\widetilde{F}}=
\sum_{i\geq k+1}^n \varepsilon_i g_{\bar{M}}(\nabla^{\bar{M}}_{e_i}\nu_{\widetilde{F}},e_i)\nu_{\widetilde{F}} +H_{\mathrm{fib}}^{\nu_{\widetilde{F}}} \nu_{\widetilde{F}} \\& =
\sum_{i\geq k+1} \varepsilon_i \pi^*g_{\bar{B}}(\pi^*\nabla^{\bar{B}}_{v_i}\nu_{F},v_i)\nu_{F}^h+H_{\mathrm{fib}}^{\nu_{\widetilde{F}}} \nu_{\widetilde{F}}
=\Big( \sum_{i\geq k+1} \varepsilon_i g_{\bar{B}}(\nabla^{\bar{B}}_{v_i}\nu_{F},v_i)\nu_{F}\Big)^{h}+H_{\mathrm{fib}}^{\nu_{\widetilde{F}}} \nu_{\widetilde{F}}\\& = \vec{H}_{F}^{h}+H_{\mathrm{fib}}^{\nu_{\widetilde{F}}} \nu_{\widetilde{F}} .
\end{align*}
Projecting and using the translating soliton equation then directly shows our result. Note that the projection of the mean curvature of the fibers to the base only makes sense if it is constant.
\end{proof}
\begin{remark} \normalfont
We notice that this result especially holds in the case where the submersion has totally geodesic fiber or is harmonic. In fact, the horizontal lift of $\vec{H}_{F}$ is $\vec{H}_{\widetilde{F}}$ if and only if the fibers are minimal with respect to the normal vector $\nu_{\widetilde{F}}$, i.e. $H_{\mathrm{fib}}^{\nu_{\widetilde{F}}}=0$. We point out that only the minimality in the direction of $\nu_{\tilde{F}}$ is needed. If for every point $x \in M$ there is a basis of tangent vectors for $T_{(x,0)} (B\times\mathbb{R})$ and graphical solitons with these as normal vectors then the harmonicity of the submersion $\pi$ is equivalent to the condition that, for every $u$, $F$ is a graphical soliton if and only if $\widetilde{F}$ is so.
\end{remark}
From Theorem \ref{bundle2} we get immediately the following
\begin{corollary}\label{cor_harm_sub}
Let $\pi:(M,g_M)\rightarrow (B,g_B)$ be a harmonic semi-Riemannian submersion. Let $u \in C^2(B)$, $F:B\rightarrow B\times\mathbb{R}$, $F(x)=(x,u(x))$ be its graph map, and $\widetilde{F}:M\rightarrow M\times\mathbb{R}$ be the map $\widetilde{F}(x)=(x,\pi\circ u(x))$.
Then $\widetilde{F}$ is a graphical translating soliton if and only if $F$ is a graphical translating soliton.
\end{corollary}
In view of the preceding theorem, we will call $F$ an $H$-perturbed soliton. Obviously, by equation \eqref{fundamentalequation},
$F$ is an $H$-perturbed soliton if and only if $u$ satisfies
\begin{eqnarray}\label{h_perturbed_eq}
\mathrm{div}\left(\frac{ \grad u } { \sqrt{\varepsilon'\big(\varepsilon+\vert \grad u\vert_g^2\big)}} \right)=
\frac{1}{ \sqrt{\varepsilon'\big(\varepsilon+\vert \grad u\vert_g^2\big)}} -H_{\mathrm{fib}}^{\nu_{\widetilde{F}}}
\end{eqnarray}
We are now going to specialize to the case where the base $B$ is one-dimensional. We will show that in that case equation \eqref{h_perturbed_eq} can be reduced to a particular ODE.
We consider a semi-Riemannian submersion $\pi:(M,g_M)\rightarrow(I,\widetilde{\varepsilon}ds^2)$, where $I$ is an open interval. Needless to say, for each $s\in I$, the fiber $\pi^{-1}(s)$ is a hypersurface in $M$. We assume that each fiber has constant mean curvature (CMC), so that for each $s\in I$, we can call $h(s)$ the value of the mean curvature of $\pi^{-1}(s)$.
Along the paper, we will say that $h$ \textit{represents the mean curvature of the fibers.}
\begin{theorem}\label{equiv_ODE}
Let $I$ be an open interval and $\pi:(M,g_M)\rightarrow(I,\widetilde{\varepsilon}ds^2)$ be a semi-Riemannian submersion with CMC fibers, and function $h$ representing the mean curvature of the fibers. Given $u\in C^2(M)$ which is constant along the fibers of $\pi$, let $\widetilde{F}:M\rightarrow M\times\mathbb{R}$, $\widetilde{F}(x)=(x,u(x))$, be its graph map, and $f:I\rightarrow \mathbb{R}$ be the map such that $u=f\circ\pi$. Then $\widetilde{F}$ is a translating soliton if and only if function $f$ is a solution to \eqref{keyODE}.
\end{theorem}
\begin{proof}
By Theorem \ref{bundle2}, and since the base is one-dimensional, $F$ is a translating soliton if and only if
\begin{eqnarray}\label{h_perturbed_eq_onedim}
\widetilde{\varepsilon}\left(\frac{f' } { \sqrt{\varepsilon'\big(\varepsilon+\widetilde{\varepsilon} f'^2\big)}} \right)'=
\frac{1}{ \sqrt{\varepsilon'\big(\varepsilon+\widetilde{\varepsilon}f'^2\big)}} -H_{\mathrm{fib}}^{\nu_{\widetilde{F}}}
\end{eqnarray}
Since $\pi$ is a semi-Riemannian submersion, $\vert\nabla\pi\vert^2=\widetilde{\varepsilon}$, where $\widetilde{\varepsilon}=\pm 1$ matches the signature of the base $I$. Then we have $(f\circ \pi)'=(f'\circ\pi)\nabla\pi$. After computing out the term on the left-hand side of equation \eqref{h_perturbed_eq_onedim}, we get therefore
\begin{eqnarray*}
\widetilde{\varepsilon}\frac{f''} { \sqrt{\varepsilon'\big(\varepsilon+\widetilde{\varepsilon}f'^2\big)}}\left(1-\frac{\widetilde{\varepsilon}f'^2}{\varepsilon+\widetilde{\varepsilon}f'^2}\right)=
\frac{1}{ \sqrt{\varepsilon'\big(\varepsilon+\widetilde{\varepsilon} f'^2\big)}} -H_{\mathrm{fib}}^{\nu_{\widetilde{F}}}
\end{eqnarray*}
Let us now point out the following fact about mean curvature. Let $H^N$ be the mean curvature of a submanifold $(Z,g_Z)$ with respect to a unit normal vector $N=\sum_ia_i\xi_i$, where $\{\xi_i\}$ are unit normal vectors to the submanifold and $a_i$ are functions on the submanifold. Then if $\{e_i\}$ is an orthonormal basis of the tangent bundle to the submanifold we have by definition
\[H^{N}=\sum_ig_Z(\nabla_{e_i}(\sum_ja_j\xi_j),e_i)=\sum_i\sum_ja_jg_Z(\nabla_{e_i}\xi_i,e_i)+g_Z(\partial_{e_i}(a)\xi_i,e_i)=\sum_ia_iH^{\xi_i}\]
We now compute the mean curvature $H_{\mathrm{fib}}^{\nu_{\widetilde{F}}}$ of the leaves $\pi^{-1}(x)$ in the direction of the normal vector in $M\times\mathbb{R}$. We have that
\[\nu_{\widetilde{F}}=\frac{1}{\sqrt{\varepsilon'\big((\nabla(f\circ\pi))^2+\varepsilon}\big)}(\nabla(f\circ\pi),-1)=\frac{1}{\sqrt{\varepsilon'\big((f'\circ\pi)^2|\nabla\pi|^2+\varepsilon\big)}}((f'\circ\pi)\nabla\pi-\partial_t)\]
Using the preceding computations, and by the fact that the mean curvature vanishes in the direction of $\partial_t$, we get consequently
\[H_{\mathrm{fib}}^{\nu_{\widetilde{F}}}=\frac{(f'\circ\pi)}{\sqrt{(f'\circ\pi)^2\widetilde{\varepsilon}+\varepsilon}}H_{\mathrm{fib}}^{\nabla\pi}.\]
Now the mean curvature of the fibers inside of $M$ is the mean curvature with respect to the unit normal vector $\nabla\pi$. Since the leaves have constant mean curvature, there exists a function $h:I\rightarrow \mathbb{R}$, such that $H_{\mathrm{fib}}^{\nabla\pi}=h\circ\pi$.
Hence equation \eqref{h_perturbed_eq_onedim} becomes after computing out the term on its left-hand side
\begin{eqnarray*}
\widetilde{\varepsilon}\frac{f''} { \sqrt{\varepsilon'\big(\varepsilon+\widetilde{\varepsilon}f'^2\big)}}\left(1-\frac{\widetilde{\varepsilon}f'^2}{\varepsilon+\widetilde{\varepsilon}f'^2}\right)=
\frac{1}{ \sqrt{\varepsilon'\big(\varepsilon+\widetilde{\varepsilon} f'^2\big)}}-\frac{f'}{\sqrt{\varepsilon'\big(\varepsilon+\widetilde{\varepsilon} f'^2\big)}}h.
\end{eqnarray*}
and after simplifying
\begin{eqnarray*}
f''=(\widetilde{\ep}+\varepsilon f'^2)(1-f'h),
\end{eqnarray*}
which proves the result.
\end{proof}
\section{Lie Groups}
A particular example where the condition of the previous theorem are satisfied is the case of a manifold with a Lie group acting by isometries whose orbits give a foliation by codimension one submanifolds. Note that we have CMC fibers, so we can construct the map $h$ representing the mean curvature of the fibers. In this section $I$ will always be an open interval.
\begin{proposition}\label{prop_Liegroup_is_submersionwithCMCfibers}
Let $(M,g)$ be a connected semi-Riemannian manifold. Let $\Sigma$ be a Lie group acting by isometries on $M$ and $\pi:M\rightarrow I$ be a submersion such that the fibers of $\pi$ are orbits of the action. Moreover assume that the gradient of the projection $\nabla\pi$ is nowhere lightlike. Then, there exists $\widetilde{\varepsilon} \in \{\pm 1\}$ and a map $v:I\rightarrow(\mathbb{R},\widetilde{\varepsilon} ds^2)$, such that $v\circ\pi$ is a semi-Riemannian submersion with constant mean curvature fibers.
\end{proposition}
\begin{proof} Since $\grad\pi$ is never zero or lightlike, $\mathrm{sign}(g(\grad\pi,\grad\pi))=\widetilde{\ep}=\pm 1$ is constant.
In addition, since the fibers of $\pi$ are orbits of $\Sigma$, the length of $\grad\pi$ is constant along them. Then, let $v:I\rightarrow(\mathbb{R},\widetilde{\ep} ds^2)$ be a function such that $v'=\big(\sqrt{\widetilde{\varepsilon}|\nabla\pi|^2}\big)^{-1}$. It follows that $v\circ\pi$ satisfies $|\nabla (v \circ \pi)|^2 = \widetilde{\varepsilon}$. Therefore it is a semi-Riemannian submersion. Next, since the fibers of $\pi$ are orbits of the Lie group action, the mean curvature must be constant along them. As $v'$ has constant sign, $v$ must be injective, hence the fibers of $v \circ \pi$ are the same as the fibers of $\pi$. It follows that its fibers have constant mean curvature $H_{\mathrm{fib}}=\mathrm{div}(\nabla (v\circ\pi))$.
\end{proof}
From a technical point of view, the proposition shows that there is no loss of generality if we assume that $\vert\grad\pi\vert^2_g=\widetilde{\ep}$. and $\mathrm{div}(\grad \pi) = h\circ\pi=H^{\nabla{\pi}}_{\mathrm{fib}}$, for a suitable function $h:I\rightarrow\mathbb{R}$. Note that it is necessary for the Lie group to act by isometries. Indeed, consider the standard flat Riemannian metric on $\mathbb{R}^2$ and the action of the Boost Group on $\mathbb{R}^2$. Then, a simple computation shows that we cannot obtain the constancy of the length of $\grad\pi$ (see Example \ref{Magda}) .
We recall that a function $u:M\rightarrow\mathbb{R}$ is called \textit{invariant} by the Lie group $\Sigma$, or also $\Sigma$-invariant, if it satisfies
\begin{equation}
\label{invariance} u:M\rightarrow\mathbb{R}, \quad u(x)=u(\sigma\cdot x), \ \forall x\in M, \ \forall \sigma\in \Sigma.
\end{equation}
Accordingly, we will say that a graphical translating soliton is \textit{invariant} by the Lie group $\Sigma$, or also $\Sigma$-invariant, when its graph map is invariant by $\Sigma$. We are now ready to prove the following
\begin{theorem} \label{existence} Let $(M,g)$ be a connected semi-Riemannian manifold. Let $\Sigma$ be a Lie group acting by isometries on $M$ and $\pi:(M,g_M)\rightarrow (I,\widetilde{\varepsilon}ds^2)$ be a semi-Riemannian submersion such that the fibers of $\pi$ are orbits of the action, with function $h$ representing the mean curvature of the orbits. Take $u\in C^2(M,\mathbb{R})$ and consider its graph map $F:M\rightarrow M\times\mathbb{R}$, $F(x)=(x,u(x))$ for any $x\in M$. Then, $F$ is a $\Sigma$-invariant translating soliton if, and only if, there exists a solution $f\in C^2(I,\mathbb{R})$ to the ODE \eqref{keyODE} such that $u=f\circ \pi$.
\end{theorem}
\begin{proof}
The fact that $u$ is invariant by $\Sigma$ readily shows the existence of a smooth function $f: I\rightarrow \mathbb{R}$ such that $u=f\circ \pi=f(\pi)$. By using Proposition \ref{prop_Liegroup_is_submersionwithCMCfibers}, we can assume that $\pi:M\to I$ is a semi-Riemannian submersion with constant mean curvature fibers. Hence, Theorem \ref{existence} follows directly from Theorem \ref{equiv_ODE}.
\end{proof}
Proposition \ref{prop_Liegroup_is_submersionwithCMCfibers} actually shows that we can refine
Theorem \ref{existence} in the following more general, but also more technical way.
\begin{corollary}
\label{existence2} Let $(M,g)$ be a connected semi-Riemannian manifold. Let $\Sigma$ be a Lie group acting by isometries on $M$ and $\pi:(M,g_M)\rightarrow (I,\widetilde{\varepsilon}ds^2)$ be a submersion such that the fibers of $\pi$ are orbits of the action, with function $h:I\to\mathbb{R}$ representing the mean curvature of the fibers. Assume that the gradient of the projection $\nabla\pi$ is nowhere lightlike. Consider $u\in C^2(M,\mathbb{R})$ and its graph map $F:M\rightarrow M\times\mathbb{R}$, $F(x)=(x,u(x))$ for any $x\in M$. Then, $F$ is a $\Sigma$-invariant translating soliton if and only if there exists a solution $f\in C^2(I,\mathbb{R})$ to the ODE \eqref{keyODE}, such that $u=f\circ v\circ\pi$, with $v:I\rightarrow\mathbb{R}$ a map satisfying $v'=\big(\sqrt{\widetilde{\varepsilon}|\nabla\pi|^2}\big)^{-1}$.
\end{corollary}
Before studying examples, we want to give some results about the behavior of solutions of the ODE \eqref{keyODE} depending on the function $h$ and the signs $\varepsilon$ and $\widetilde{\ep}$.
\begin{corollary} \label{oide} Let $(M,g)$ be a connected semi-Riemannian manifold. Let $\Sigma$ be a Lie group acting by isometries on $M$ and $\pi:M\rightarrow I$ be a submersion such that the fibers of $\pi$ are orbits of the action. Moreover assume that $\vert\grad \pi\vert^2=\widetilde{\ep}=\pm 1$ and $\mathrm{div}(\grad\pi)=h\circ\pi$, for some function $h: I\to\mathbb{R}$. Then for each $y_o\in \mathbb{R}$ and each $s_o\in I$ such that $h(s_o)\neq 0$, there exist a real number $\rho>0$ and a translating soliton $F:(y_0-\rho,y_o+\rho)\times\Sigma \rightarrow\bar{M}$ such that it is the union of two graphical translating solitons.
\end{corollary}
\begin{proof} If $f'(s_o)\neq 0$, then, there is a small interval $J$ around $y_o=f(s_o)\in \mathbb{R}$ such that $\alpha=f^{-1}$ is well-defined on it. Given $y\in J$, we have according to \eqref{keyODE},
\[ -\frac{\alpha''(y)}{\alpha'(y)^3} = \left( \varepsilon +\frac{\widetilde{\ep}}{\alpha'(y)^2}\right)\left(1-\frac{h(\alpha(y))}{\alpha'(y)}\right),
\]
which is
\[ \alpha''(y) = \left(\widetilde{\ep} +\varepsilon \alpha'(y)^2\right) \left( h(\alpha(y)) - \alpha'(y)\right).
\]
From here, we can follow the proof of Lemma 2.3 in \cite{CSS}.
\end{proof}
\begin{corollary} \label{globalexamples}
Under the same conditions, assume that $\varepsilon\widetilde{\ep} =-1$. Then, given $s_o\in I$, $f_1\in (-1,1)$ and $f_o\in \mathbb{R}$, there exists a solution $f:I\rightarrow\mathbb{R}$ to \eqref{keyODE} such that $f(s_o)=f_0$ and $f'(s_o)=f_1$.
\end{corollary}
\begin{proof} We make the change $w=f'$, so that \eqref{keyODE} reduces to
\[w'(s) = \pm (1-w(s)^2) (1 - w(s)h(s)).
\]
Note that constant functions $w(s)=\pm 1$ are solutions to this differential equation, and they do not cross. Then, given initial conditions $s_o\in I$ and $f_1\in (-1,1)$, there exists a well-defined solution $w:I\rightarrow\mathbb{R}$. It remains to compute $f(s)=f_o+\int_{s_o}^s w(x)dx$ for some $f_o\in\mathbb{R}$.
\end{proof}
\section{Examples}
\begin{examp} \label{cero} A generalization of the \textit{Grim Reaper Cylinder} in $\mathbb{R}^{m+1}$. \normalfont Given $(M,g_M)$ a connected semi-Riemannian manifold, let $\Omega$ be an open domain in $M$ such that $\Gamma:\Omega\rightarrow M\times\mathbb{R}$ is a translating soliton in $(M\times\mathbb{R},g_M+\varepsilon dt^2)$, for some $\varepsilon=\pm 1$. Now, we consider another connected semi-Riemannian manifold $(P,g_P)$. The product
\[ \bar{\Gamma}: P\times \Omega\rightarrow (P\times M\times\mathbb{R},g_P+g_M+\varepsilon dt^2) \quad
\bar{\Gamma}(q,x)=(q,\Gamma(x)),
\]
is just another translating soliton. Indeed, if $\vec{H}$ is the mean curvature vector of $\Gamma$, then, $\vec{J}=(0,\vec{H})$ is the mean curvature vector of $\bar{\Gamma}$, since we are dealing with the product spaces. Thus, $\vec{J}=\partial_t^{\perp}.$
$\Box$
\end{examp}
\begin{examp} \label{un} \normalfont We study those translating solitons in $\mathbb{R}^{n+1}$ with standard flat metric which are $SO(n)$-invariant, also known as \textit{rotationally invariant}. If in $\mathbb{R}^n$ with usual coordinates $x=(x_1,\ldots,x_n)$, we take $\pi(x)=\sqrt{\sum_i x_i^2}$, then
\[ \grad \pi (x)= \sum_i \frac{x_i}{\pi(x)}\partial_i\vert_x, \quad \vert \grad\pi(x)\vert^2 = \sum_i\frac{x_i^2}{\pi(x)^2} = 1, \quad \mathrm{div}(\grad \pi)(x) = \frac{n-1}{\pi(x)}.
\]
This means that our manifold has to be $M=\mathbb{R}^n\backslash\{0\}$. However, in order to obtain a $C^2$-class translating soliton, \eqref{keyODE} becomes
\begin{equation} \label{yasesabe}
f''(s)=\big(1+ f'(s)^2\big)\Big(1-\frac{n-1}{s}f'(s)\Big), \quad
f'(0)=0, \ f(0)=a\in\mathbb{R}.
\end{equation}
In \cite{AW}, it was proved that there exist a convex, rotationally symmetric translating soliton over the plane.
This was improved in \cite{CSS} obtaining entire rotationally symmetric translating solitons $F:\mathbb{R}^{n}\times[0,+\infty)\rightarrow\mathbb{R}^{n+1}$, $n\geq 2$. In fact, in both papers \cite{AW} and \cite{CSS}, the same ODE as \eqref{yasesabe} is obtained. This means that this problem has a unique $C^{\infty}[0,+\infty)$ solution. The associated translating soliton is known as \textit{translating paraboloid}. In addition, by Corollary \ref{oide}, we recover the \textit{translating catenoid}.
In the book \cite{Besse}, we can find a list of Lie groups acting transitively and effectively on the sphere $\mathbb{S}^{n-1}$, hence, also on $\mathbb{R}^{n}$, and the space of leaves will be in all cases the interval $[0,+\infty)$. Thus, we can change $O(n)$ by any of them, namely $SO(n)$, $SU(n)$, $Sp(n)Sp(1)$, $Sp(n)U(1)$, $Sp(n)$, $G_2$, $Spin(7)$ and $Spin(9)$, for suitable values of the dimension $n$, depending on the case. This means that we can use all these groups to recover the rotationally symmetric translating solitons in $\mathbb{R}^n$.
$\Box$
\end{examp}
In the following examples, we will need some tools, which can be found in the book \cite{Wiggins}. In $\mathbb{R}^2$, with coordinates $(s,t)\in\mathbb{R}^2$,
an \textit{autonomous vector field} is a map $X:\mathbb{R}^2\rightarrow\mathbb{R}^2$, at least of class $C^1$. Given a point $p\in\mathbb{R}^2$ such that $X(p)=(0,0)$, we compute its \textit{linearlization} at $p$, namely
\[ DX(p) = \begin{pmatrix} \frac{\partial X}{\partial s}(p) \ \frac{\partial X}{\partial x}(p) \end{pmatrix}.\]
We can compute the (complex) eigenvalues of this matrix, $\lambda_1$ and $\lambda_2$, with eigenvectors $v_1$ and $v_2$. When both eigenvalues are real and $\lambda_1\lambda_2<0$, the point $p$ is called a \textit{saddle point}.
In such case, according to Theorem 3.2.1 of \cite{Wiggins}, there is a submanifold (in this case, just a curve) whose tangent space at $p$ is spanned by the eigenvector of negative eigenvalue, called the \textit{unstable submanifold}. The good property for us is that this submanifold can be seen as the graph of map defined on a small neighbourhood of $p$.
\begin{examp}\label{deux} \normalfont Consider the Minkowski space $\mathbb{L}^{n+1}$ with standard flat metric $g=\sum_{i=1}^ndx_i^2-dx_{n+1}^2$. A rotationally invariant space-like soliton $F:\mathbb{R}^n\rightarrow\mathbb{L}^{n+1}$, $F(x)=(x,u(x))$ is determined by the group $O(n)$, in a very similar way as in Example \ref{un}, but \eqref{keyODE} is now
\begin{equation}\label{rotacionalL3}
f''(s)=\big(1 - f'(s)^2\big)\Big(1-\frac{n-1}{s}f'(s)\Big), \quad
f'(0)=0, \ f(0)=a\in\mathbb{R}.
\end{equation}
To study the solution to this problem, first step is to define $w=f'$, obtaining
\begin{equation} \label{wsingular}
w'(s) = (1-w(s)^2)\Big(1-\frac{n-1}{s}w(s)\Big), \quad w(0)=0.
\end{equation}
We consider the following \textit{autonomous vector field}:
\[ X:\mathbb{R}^2\rightarrow\mathbb{R}^2, \quad
X(s,x) = \Big(s, (n-1)x^3-s x^2-(n-1)x+s \Big).
\]
Note that $X(0,0)=(0,0)$. At $(0,0)$, the \textit{linearlization} is
\[ DX(0,0) = \begin{pmatrix} \frac{\partial X}{\partial s}(0,0) \ \frac{\partial X}{\partial x}(0,0) \end{pmatrix}
= \begin{pmatrix} 1& 0 \\ 1 & 1-n \end{pmatrix},
\]
whose eigenvalues are $\lambda_1=1$ and $\lambda_2=1-n$, with corresponding eigenvectors $v_1=(n,1)^t$, $v_2=(0,1)^t$. The point $(0,0)$ is a saddle point. By Theorem 3.2.1 of \cite{Wiggins}, there exist a $1$-dimensional,
local unstable analytic manifold (\textit{of fixed points}), around $(0,0)$,
whose tangent space at $(0,0)$ is spanned by $v_2$, which is a graph in an small interval around $s=0$, namely
\[
W = \{ (s,x)\in \mathbb{R}\times\mathbb{R} : x=w(s), \ \vert s\vert\, small\},
\]
for some analytic function $w$ defined on a small interval $(-\bar{\delta},\bar{\delta})$.
This means that our dynamical system has a solution $\alpha:(-\delta,\delta)\rightarrow W$, $\alpha(t)=(s(t),x(t))$, with $\alpha'(t)=X(\alpha(t))$, which is an analytic diffeomorphish such that $\alpha(0)=(0,0)$, $\alpha'(0)=\lambda (n,1)$ for some $\lambda\in\mathbb{R}$, $\lambda\neq 0$, and $x(t)=w(s(t))$. We compose with the inverse of $s$, so that $t=t(s)$. Thus, $w(s)=x(t(s))$ is analytic near $0$, with $w(0)=0$. In addition, since $X(\alpha(t))=\alpha'(t)$, then for $s>0$,
\begin{gather*}
w'(s) = x'(t(s))t'(s) = \frac{x'(t(s))}{s'(t)} = \frac{(1-x(t(s))^2)(s(t)-(n-1)x(t(s))}{s(t)} \\
= (1-x(t(s))^2) \Big ( 1-\frac{(n-1)x(t(s))}{s(t)}\Big) = (1-w(s))^2) \Big ( 1-\frac{(n-1)w(s)}{s}\Big).
\end{gather*}
From here, recalling $f'=w$, we obtain an analytic solution to \eqref{rotacionalL3}. Then, as in Corollary \ref{globalexamples}, we know that we can extend this solution to $f:[0,\infty)\rightarrow\mathbb{R}$, with $f'(0)=0$.
In a similar fashion to Example \ref{un}, we can construct a spacelike rotationally symmetric translating soliton with two ends, that we can also call the \textit{translating catenoid} in $\mathbb{L}^{n+1}$. $\Box$
\end{examp}
\begin{corollary}\label{notimelikeinLn}
There do not exist time-like $SO(n)$-invariant translating solitons in Minkowski space $\mathbb{L}^{n+1}$.
\end{corollary}
\begin{proof} Let $F:\Omega\subset \mathbb{R}^n\rightarrow \mathbb{L}^{n+1}$, $F(x)=(x,u(x))$ a timelike $SO(n)$-invariant translating soliton. The action of $SO(n)$ in $\mathbb{L}^{n+1}$ is determined by the time-like axis $L=\{(0,x_{n+1}) \in\mathbb{L}^{n+1} : x_{n+1}\in\mathbb{R}\}$. For $F$ be $SO(n)$-invariant and smooth, then the tangent space at the intersection of $F(\Omega)\cap L$ has to be spacelike, which is a contradiction.
\end{proof}
\begin{examp}\label{desitter} \normalfont
We consider the Minkowski space $\mathbb{L}^{n+1}$, $n\geq 3$, with its usual flat metric $g(X,Y)=\sum_{i=1}^n X_iY_i-X_{n+1}Y_{n+1}$. The de Sitter space-time is
\[ dS^n = \{ x=(x_1,\ldots,x_{n+1}) \in \mathbb{L}^{n+1} : g(x,x)=+1 \}.
\]
As usual, we identify the tangent space at $x\in dS^n$,
\[
T_x dS^n = \{ X=(X_1,\ldots,X_{n+1})\in\mathbb{R}^{n+1} : g(X,x)=0\}.
\]
In particular, the position vector is a unit, spacelike, normal vector field $\chi:dS^n\rightarrow\mathbb{L}^{n+1}$.
Now, the Lie group $O(n-1)$ acts by isometries on $dS^n$ as usual:
\[ O(n-1)\times dS^n \rightarrow dS^n, \quad
(A,x)\mapsto A\cdot x = \left(\begin{array}{c|c} A & 0 \\ \hline 0 & 1 \end{array}\right)x = \left( \begin{array}{c} A(x_1,\ldots,x_n)^t \\ x_{n+1}\end{array}\right).
\]
We consider the map
\[ \tau : dS^n\rightarrow \mathbb{R}, \quad \tau(x)=x_{n+1}.
\]
We define also de constant vector
\[ \xi=(0,\ldots,0,-1).\]
Clearly,
\[ \tau = g(\xi,\chi).\]
Given a point $x\in dS^n$ and $X\in T_xdS^n$, we have $g_x(\grad\tau,X)=(d\tau)_xX = g(\xi,X)$. Therefore, $\grad\tau$ is going to be the tangent component of $\xi$, i.~e.,
\[ \grad\tau = \xi - g(\xi,\chi)\chi = \xi -\tau \chi, \quad g(\grad\tau,\grad\tau) = -1-\tau^2.\]
We look for smooth functions $v$ and $\bar{\tau}$ such that $\tau=v\circ\bar\tau$ and $\|\grad \bar{\tau}\|^2= -1$. Then, $\grad\tau=v'(\bar \tau)\grad\bar\tau$. This implies $(v'(\bar\tau))^2(-1)=-1-\tau^2=-1-(v(\bar{\tau}))^2$, so that $v'(s)=\pm\sqrt{1+v(s)^2}$. We choose $v(s)=\sinh(s)$. That means
\[ \bar{\tau} = \sinh^{-1}(\tau) \ \Leftrightarrow \tau=\sinh(\bar{\tau}).
\]
Consider a local orthonormal frame $(e_1,\ldots,e_n)$ of $TdS^n$, with $\varepsilon_i=g(e_i,e_i)=\pm 1$. Let $\nabla$, $\bar{\nabla}$ be respectively the Levi-Civita connection of $dS^n$ and $\mathbb{L}^{n+1}$.
\begin{align*}
\mathrm{div}(\grad\tau) &= \sum_i \varepsilon_i g(\nabla_{e_i}\grad\tau,e_i) = \sum_i \varepsilon_i g(\bar\nabla_{e_i}\grad\tau,e_i) =
\sum_i \varepsilon_i g(\bar\nabla_{e_i}(\xi-\tau\chi),e_i) \\
&= \sum_i \varepsilon_i g(\bar\nabla_{e_i}(-\tau\chi),e_i) = \sum_i (-\varepsilon_i)\tau g(\bar\nabla_{e_i}(\chi),e_i) =
-\tau \sum_{i} \varepsilon_i^2= -n\tau.
\end{align*}
Since $\tau=v\circ\bar{\tau}$, then
\begin{align*}
-n\sinh(\bar{\tau})=-n\tau = \mathrm{div}(\grad\tau) = v'(\bar{\tau}) \mathrm{div}(\grad\bar{\tau}) +v''(\bar\tau)\|\grad\bar{\tau}\|^2 = \cosh(\bar{\tau})\mathrm{div}(\grad\bar{\tau})-\sinh(\bar{\tau}).
\end{align*}
Thus,
\[\mathrm{div}(\bar{\tau}) = -(n-1)\tanh(\bar{\tau}).\]
By taking $h(s)=-(n-1)\tanh(s)$, the initial value problem becomes
\begin{equation}\label{pvi-desitter}
f''(s)=\big(-1+\varepsilon f'(s)^2\big)\left(1+(n-1)\tanh(s)f'(s) \right), \ f'(s_o)=f_0, \ f(s_o)=f_1,
\end{equation}
for any $s_o\in\mathbb{R}$ and any initial values $f_o,f_1\in\mathbb{R}$.
We point out that by Corollary \ref{globalexamples}, when $\varepsilon=+1$, there exist solutions defined on the whole $\mathbb{R}$ whenever $f'(s_o)\in (-1,1)$. They give rise to entire rotationally symmetric translating solitons in $dS^{n}\times \mathbb{R}$. Since $\widetilde{\ep}=-1$, this type of translating solitons will have $\varepsilon' = \mathrm{sign}(\varepsilon +\widetilde{\ep} (f')^2) =+1$, that it to say, all of them will be spacelike. In addition, by Corollary \ref{oide}, it is possible to construct spacelike rotationally symmetric translating solitons with two ends whose topology is $\mathbb{S}^{n-1}\times \mathbb{R}$.
\end{examp}
\begin{examp}\label{hyperbolicspace} \normalfont
We consider the Minkowski space $\mathbb{L}^{n+1}$, $n\geq 2$, with its usual flat metric $g(X,Y)=\sum_{i=1}^n X_iY_i-X_{n+1}Y_{n+1}$. The Weierstra\ss' model of the Hyperbolic Space is
\[ \mathbb{H}^n = \{ x=(x_1,\ldots,x_{n+1}) \in \mathbb{L}^{n+1} : g(x,x)=-1, \ x_{n+1}\geq 1\}.
\]
As usual, we identify the tangent space at $x\in \mathbb{H}^n$,
\[
T_x \mathbb{H}^n = \{ X=(X_1,\ldots,X_{n+1})\in\mathbb{R}^{n+1} : g(X,x)=0\}.
\]
In particular, the position vector is a unit, timelike, normal vector field $\chi:\mathbb{H}^n\rightarrow\mathbb{L}^{n+1}$.
The Lie group $O(n-1)$ acts by isometries on $\mathbb{H}^n$ as usual:
\[ O(n-1)\times \mathbb{H}^n \rightarrow \mathbb{H}^n, \quad
(A,x)\mapsto A\cdot x = \left(\begin{array}{c|c} A & 0 \\ \hline 0 & 1 \end{array}\right)x = \left( \begin{array}{c} A(x_1,\ldots,x_n)^t \\ x_{n+1}\end{array}\right).
\]
We consider the map
\[ \tau : \mathbb{H}^n\rightarrow \mathbb{R}, \quad \tau(x)=x_{n+1}.
\]
We define also the constant vector
\[ \xi=(0,\ldots,0,-1).\]
Clearly,
\[ \tau = g(\xi,\chi).\]
Given a point $x\in \mathbb{H}^n$ and $X\in T_x\mathbb{H}^n$, we have $g_x(\grad\tau,X)=(d\tau)_x X = g(\xi,X)$. Therefore, $\grad\tau$ is going to be the tangent component of $\xi$, i.~e.,
\[ \grad\tau = \xi + g(\xi,\chi)\chi = \xi+\tau \chi, \quad g(\grad\tau,\grad\tau) =\tau^2 -1\geq 0.\]
The only critical point is $p=(0,\ldots,0,1)$, so we remove it and restrict all computations to $\Omega=\mathbb{H}^n\backslash\{p\}$. In addition, $\widetilde{\ep} = \mathrm{sign}(\grad\tau)=+1$.
We look for smooth functions $v$ and $\bar{\tau}$ such that $\tau=v\circ\bar\tau$ and $\|\grad \bar{\tau}\|^2= 1$. Then, $\grad\tau = v'(\bar{\tau})\grad\bar{\tau}$. This implies $\tau^2-1=v(\bar{\tau})^2-1=(v'(\bar\tau))^2$, so that $v'(s)=\pm\sqrt{s^2-1}$. We choose $v(s)=\cosh(s)$. That means
\[ \bar{\tau} = \cosh^{-1}(\tau) \ \Leftrightarrow \tau=\cosh(\bar{\tau}).\]
Consider a local orthonormal frame $(e_1,\ldots,e_n)$ of $T\Omega$. Let $\nabla$, $\bar{\nabla}$ be respectively the Levi-Civita connection of $\Omega$ and $\mathbb{L}^{n+1}$.
\begin{align*}
\mathrm{div}(\grad\tau) &= \sum_i g(\nabla_{e_i}\grad\tau,e_i)
= \sum_i g(\bar\nabla_{e_i}\grad\tau,e_i) = \sum_i g(\bar\nabla_{e_i}(\xi+\tau\chi),e_i) \\
&= \sum_i g(\bar\nabla_{e_i}(\tau\chi),e_i) = \sum_i \tau g(\bar\nabla_{e_i}(\chi),e_i) =
\tau \sum_{i} 1= n\tau.
\end{align*}
Since $\tau=v\circ\bar{\tau}$, then
\begin{align*}
n\cosh(\bar{\tau})=n\tau = \mathrm{div}(\grad\tau) = v'(\bar{\tau}) \mathrm{div}(\grad\bar{\tau}) +v''(\bar\tau)\|\grad\bar{\tau}\|^2
= \sinh(\bar{\tau})\mathrm{div}(\grad\bar{\tau})+\cosh(\bar{\tau}).
\end{align*}
Thus,
\[\mathrm{div}(\bar{\tau}) = (n-1)\coth(\bar{\tau}).\]
By taking $h(s)=(n-1)\coth(s)$, the initial value problem becomes
\begin{equation}\label{pvi-hyperbolicspace}
f''(s)=\big(1+\varepsilon f'(s)^2\big)\left(1-(n-1)\coth(s)f'(s) \right), \ f'(0)=0, \ f(0)=f_1,
\end{equation}
for any initial value $f_1\in\mathbb{R}$.
By taking $w=f'$, we consider the auxiliary problem
\[ w'(s) = \big(1+\varepsilon w(s)^2\big)\big(1-(n-1)\coth(s)w(s) \big), \ w(0)=0.
\]
We want to show that this problem has solutions. Thus, we proceed as in Example \ref{deux}. We consider the dynamical system
\[ X:\mathbb{R}^2\rightarrow\mathbb{R}^2, \quad X(s,x)=\Big(\sinh(s), (1+\varepsilon x^2)\big(\sinh(s)-(n-1)\cosh(s)x\big)\Big).
\]
Note that $X(0,0)=(0,0)$. The \textit{linearization} at $(0,0)$ is
\[ DX(0,0) = \begin{pmatrix} \frac{\partial X}{\partial s}(0,0) \ \frac{\partial X}{\partial x}(0,0) \end{pmatrix}
= \begin{pmatrix} 1& 0 \\ 1 & 1-n \end{pmatrix}.
\]
By repeating the steps in Example \ref{deux}, we obtain an analytic solution $f:[0,T)\rightarrow\mathbb{R}$ to \eqref{pvi-hyperbolicspace} for each $f_1\in\mathbb{R}$, $f(0)=f_1$.
When $\varepsilon=-1$, by Corollary \ref{globalexamples}, $T=+\infty$. This means that we obtain entire rotationally symmetric graphical translating solitons in $\mathbb{H}^n\times\mathbb{R}$ with a Lorentzian metric. In addtion, it is possible to construct a \textit{translation helicoid} in this setting.
Assume now that $\varepsilon=+1$. Firstly, we see that $f'(s)\geq 0$ for any $s\in[0,T)$. Indeed, if there exists $s_o\in (0,T)$ such that $f'(s_o)<0$, by \eqref{pvi-hyperbolicspace}, we obtain $f''(s_o)>0$, which shows that $f'$ is increasing around $s_o$. Since $f'(0)=0 > f'(s_o)$, by continuity, there exists $s_1\in (0,s_o)$ such that $f''(s_1)=0$ and $f'(s_1)\leq f'(s_0)$. But then, by \eqref{pvi-hyperbolicspace}, we obtain $0=\big(1+f'(s_1)^2\big)\big( 1-(n-1)\coth(s_1)f'(s_1)\big)>0$, which is a contradiction. Next, we are going to see that we can extend $f$ from $0$ to $+\infty$. For the sake of simpleness, we call $w=f':[0,T)\rightarrow\mathbb{R}$, and $g, G:[0,T]\times [0,\infty)\rightarrow\mathbb{R}$ given by $g(s,x)=(1+x^2)(1-(n-1)\coth(s)x)$ and $G(s,x)=(1+x^2)(1-(n-1)x/s)$. A simple computation shows
\[G(s,x)-g(s,x)= (n-1)(1+x^2)\left(\coth(s)-\frac1s\right) x \geq 0, \quad \forall (s,x)\in [0,\infty)\times [0,\infty).
\]
Thus, the solution to the problem $w'(s)=G(s,w(s))$, $w(0)=0$ is an upper bound of the solution to the problem $w'(s)=g(s,w(s))$, $w(0)=0$. But we already know that the solution to the problem
\[ w(0)=0, \quad w'(s)= (1+w(s)^2)\left(1 - \frac{n-1}{s}w(s)\right)
\]
is defined on $[0,\infty)$. This readily shows that $f$, solution to \eqref{pvi-hyperbolicspace}, is defined on $[0,\infty)$. $\Box$
\end{examp}
\begin{examp}\normalfont \label{principalfiberbundle} We recall
\[ H_1^3 = \{ (z_1,z_2)\in \mathbb{C}^2 : -\vert z_1\vert^2+\vert z_2\vert^2=-1 \}
\]
is a 3-dimensional quadric in the standard indefinite complex space $\mathbb{C}^2_1$. Also, it is well-known that the unit group of complex numbers $\mathbb{S}^1=\{ a \in \mathbb{C} : a\bar{a}=1 \}$ acts by isometries on $H_1^3$, $(a,(z_1,z_2))\to (az_1,az_2)$, obtaining a principal fiber bundle over the complex hyperbolic line $H_1^3\rightarrow \mathbb{C}H^1$ in the usual way, with timelike, totally geodesic fibers $\mathbb{S}^1$. In addition, $\mathbb{C}H^1$ is isometric to the real hyperbolic plane $\mathbb{R} H^2$. In fact, by denoting $\bar{z}$ the complex conjugate of $z\in\mathbb{C}$,
we can construct the map
\[ \pi: H_1^3 \rightarrow \mathbb{C}\times \mathbb{R}\equiv \mathbb{R}^3, \quad \pi(z_1,z_2) = \big(2z_1\bar{z}_2, \vert z_1\vert^2+\vert z_2\vert^2\big).
\]
The image of this map is the Weierstrass model of $\mathbb{R} H^2$, embedded in the Minkowski 3-space as a quadric, and therefore it is easy to regard $\pi$ as the projection from $H_1^3$ to $\mathbb{C}H^1\equiv\mathbb{R} H^2$. Next, in Example \ref{hyperbolicspace} we obtained rotationally symmetric translating solitons in $\mathbb{R} H^2$, with group $\Sigma=\mathbb{S}^1$. Now, by Corollary \ref{cor_harm_sub}, we obtain a translating soliton $\Gamma$ in $H_1^3$, by lifting the translating soliton from $\mathbb{R} H^2$. Note that $\Gamma$ is invariant by the Lie group $\mathbb{S}^1\times\mathbb{S}^1$. $\Box$
\end{examp}
\begin{examp}\normalfont \label{Magda}
Take the Minkowski plane $\L^2$ with standard flat metric $g=dx^2-dy^2$. The \textit{boost} Lie group is the set of matrices
\[ \Sigma = \left\{ A_{\theta} = \begin{pmatrix} \cosh(\theta) & \sinh(\theta) \\
\sinh(\theta) & \cosh(\theta) \end{pmatrix} : \theta\in\mathbb{R}
\right\},
\]
which acts on $\L^2$ by isometries. However, in order to obtain suitable quotients, we need to split the plane in four regions whose boundaries are made of two light-like geodesics, namely
\begin{align*}
&\Omega_1 = \{ (x,y)\in\L^2 : y^2<x^2, \ 0<x \}, \quad
&\Omega_2 = \{ (x,y)\in\L^2 : y^2>x^2, \ 0<y \} \\
&\Omega_3 = \{ (x,y)\in\L^2 : y^2<x^2, \ 0>x \}, \quad
&\Omega_4 = \{ (x,y)\in\L^2 : y^2>x^2, \ 0>y \}.
\end{align*}
We will use the globally defined orthonormal frame $\{\partial_x,\partial_y\}$.
For $F$ be $\Sigma$-invariant, we have to extend the action of $\Sigma$ to $(\mathbb{R}^3,\produ{,}=g+dt^2)=\L^3$ in the natural way, namely
\[ \Sigma\times \mathbb{R}^3\rightarrow\mathbb{R}^3, \quad (A,(x,y,t))\mapsto \big((x,y)A,t\big).\]
Then, at the intersection point with the axis $\{(0,0)\}\times\mathbb{R}$, the tangent plane has to be orthogonal to the axis. In other words, $(dF)_{(0,0)} T_{(0,0)}\mathbb{R}^2 \perp \partial_t$. This means that $f'(0)=0$. \smallskip
We begin with $\Omega_1$. We define the projection, with its usual properties:
\begin{gather*}\pi:\Omega_1\rightarrow (0,+\infty),\
\pi(x,y)=+\sqrt{x^2-y^2},\\
\grad\pi(x,y)=\frac{x}{\pi(x,y)}\partial_x+\frac{y}{\pi(x,y)}\partial_y, \
\vert \grad\pi \vert^2 =+1=\widetilde{\ep}, \
\mathrm{div}(\grad \pi) =
\frac{1}{\pi}.
\end{gather*}
It is very simple to check that $\pi\big((x,y)A_{\theta}\big) = \pi(x,y)$ for any $(x,y)\in\L^2$ and $\theta\in\mathbb{R}$. This means that $\pi$ can work as the expected projection map.
The product manifold $(\Omega_1\times\mathbb{R},\produ{,}=g+dt^2)$ is an open subset of flat 3-Minkowski space. Thus, we are constructing a time-like translating soliton as a graph over a timelike 2-plane. We transform our differential equation into the following problem.
We recall that we need $W^2=\varepsilon'(\varepsilon+\widetilde{\ep} f'(\pi)^2)\geq 0$. By now, we know $\varepsilon=\widetilde{\ep}=+1$. Then, \eqref{keyODE} becomes
\begin{equation} \label{caseOmega1}
f''(s)=\big(1+f'(s)^2\big)\left(1-\frac{f'(s)}{s}\right), \ f(0)=a, \ f'(0)=0.
\end{equation}
We already know that this problem has a $C^{\infty}[0,+\infty)$ solution by Example \ref{un}.
We can construct our translating soliton $F^1:\Omega_1\rightarrow \L^3$, $F^1(x,y)=\Big(x,y,f\big(\sqrt{x^2-y^2}\big)\Big)$.
Note that we have another similar one
\[F^3:\Omega_3\rightarrow\L^3, \quad F^3(x,y)=\Big(x,y,f\big(\sqrt{x^2-y^2}\big)\Big).\]
In addition, by Corollary \ref{oide}, there exist \textit{boost invariant, spacelike, translating solitons with two ends}. \smallskip
Next, we work on $\Omega_2$. Now, the projection $\pi$ satisfies
\begin{gather*}\pi:\Omega_2\rightarrow (0,+\infty),\ \pi(x,y)=+\sqrt{y^2-x^2},\\
\grad\pi(x,y)=\frac{-x}{\pi(x,y)}\partial_x-\frac{y}{\pi(x,y)}\partial_y, \
\vert \grad\pi \vert^2= -1=\widetilde{\ep}, \
\mathrm{div}(\grad \pi) = \frac{-1}{\pi}.
\end{gather*}
The differential equation \eqref{keyODE} becomes now:
\begin{equation}\label{caseOmega2}
f''(s) = \big(-1+f'(s)^2\big)\left( 1 +\frac{f'(s)}{s}\right), \ f(0)=a, \ f'(0)=0.
\end{equation}
By the easy change $q(s)=-f(s)$, we transform this problem in
\[ q''(s) = \big(1-q'(s)^2\big)\left( 1 -\frac{q'(s)}{s}\right), \ q(0)=-a, \ q'(0)=0.
\]
By Example \ref{deux}, we know that this problem has an analytic solution in $[0,+\infty)$.
For $a=0$, we call $f_1$ the solution to \eqref{caseOmega1} in $[0,+\infty)]$, and $f_2$ the solution to \eqref{caseOmega2} in $[0,+\infty)]$, so that we can define a continous function
\[ u:\mathbb{R}^2\rightarrow \mathbb{R}, \quad u(x,y) = \left\{
\begin{array}{cl} f_1\left(\sqrt{x^2-y^2}\right), & (x,y)\in \Omega_1\cup\Omega_3, \\
0, & (x,y)\in\partial\Omega_i, i=1,2,3,4,\\
f_2\left(\sqrt{y^2-x^2}\right), & (x,y)\in\Omega_2\cup\Omega_4.
\end{array}
\right.
\]
Note that we get immediately $u\in C^0(\mathbb{R}^2)\cap C^{\infty}(\cup_{i=1}^4\Omega_i)$. We want to prove that $u$ is in $C^{\infty}$. Let $g(s):=f_1(is)$ in a small neighborhood of $0$, where $i=\sqrt{-1}$. Then we have
$$[g(s)]'' = - f_1''(is) = - (1 + f_1'(is)^2)(1-\frac{f'_1(is)}{is}) = - (1 - g'(s)^2 )(1 + \frac{g'(s)}{s})$$
Hence $g$ is a solution to equation \eqref{caseOmega2}, and consequently $g=f_2$. But then we get comparing the derivatives of $f_1$ and $f_2$, that
$f_1^{(i)}(0)=f_2^{(i)}(0)=0$, if $i$ is odd, since $f_1$ and $f_2$ are real, and $f_1^{(4i+2)}(0)=-f_2^{(4i+2)}(0)$ and $f_1^{(4i)}(0)=f_2^{(4i)}(0)$ for all $i\geq 0$.
Notice that in the same way if $\widetilde{f_1}(-s):=f_1(s)$ and $\widetilde{f_2}(-s):=f_2(s)$ in a small neighborhood of $0$, similar computations show that $\widetilde{f_1}$ and $\widetilde{f_2}$ are also a solution to \eqref{caseOmega1} and \eqref{caseOmega2} respectively. Therefore, $\widetilde{f_1}=f_1$ and $\widetilde{f_2}=f_2$ and $f_1$ and $f_2$ are even.
We now just need to prove the following
\begin{lemma}
Let $f_1$, $f_2$ be functions in $C^{2n}(\mathbb{R})$, such that $f_1^{(k)}(0)=f_2^{(k)}(0)=0$, if $k$ is odd, $f_1^{(k)}(0)=(-1)^{\frac{k}{2}}f_2^{(k)}(0)$, if $k$ is even. Then the function $u$ defined as above is in $C^{n}(\mathbb{R}^2)$.
\end{lemma}
\begin{proof}
We prove the statement by induction over $n$. The case $n=0$ is trivially satisfied, since $f_1(0)=f_2(0)$. Moreover let $g_1(z)=\frac{f'_1(z)}{z}$, $g_2(z)=\frac{f'_2(z)}{z}$. We have
\begin{eqnarray*}
\partial_xf_1(\sqrt{x^2-y^2})&=&xg_1(\sqrt{x^2-y^2}),\quad \partial_xf_2(\sqrt{y^2-x^2})=-xg_2(\sqrt{y^2-x^2})\\
\partial_yf_1(\sqrt{x^2-y^2})&=&-yg_1(\sqrt{x^2-y^2}),\quad \partial_yf_2(\sqrt{y^2-x^2})=yg_2(\sqrt{y^2-x^2})
\end{eqnarray*}
Obviously $g_1$ and $g_2$ are in $C^{2n-2}$ since $f_1'(0)=f'_2(0)=0$, and $g_1^{(4i+2)}(0)=-g_2^{(4i+2)}(0)$ and $g_1^{(4i)}(0)=g_2^{(4i)}(0)$ for all $i\geq 0$.
Then by the induction hypothesis the function \[ v:\mathbb{R}^2\rightarrow \mathbb{R}, \quad v(x,y) = \left\{
\begin{array}{cl} g_1\left(\sqrt{x^2-y^2}\right), & (x,y)\in \Omega_1\cup\Omega_3, \\
0, & (x,y)\in\partial\Omega_i, i=1,2,3,4,\\
g_2\left(\sqrt{y^2-x^2}\right), & (x,y)\in\Omega_2\cup\Omega_4.
\end{array}
\right.
\]
is in $C^{n-1}(\mathbb{R}^2)$. Hence $\partial_xu$ and $\partial_yu$ extend to $C^{n-1}(\mathbb{R}^2)$, and finally $u$ extends to $C^{n}(\mathbb{R}^2)$.
\end{proof}
This proves that $u$ is in $C^{\infty}$. As a final remark, we point out that the curve joining each two adjacent pieces is a lightlike straight line. Thus, it is possible to consider two or three contiguous pieces, and their gluing straight lines. We cannot consider the other straight half-lines, since the boundary would be lightlike. In other words, we can choose to glue either two, three or four adjacent pieces to obtain solitons.
\end{examp}
\section*{Acknowledgements}
M.~Ortega has been partially financed by the Spanish Ministry of Economy and Competitiveness and European Regional Development Fund (ERDF), project MTM2016-78807-C2-1-P. This project started during a visit of the second author to Imperial College in February 2016: M.~A.~Lawn and M.~Ortega would like to thank the London Mathematical Society and Imperial College London, since this visit was partially supported by a Scheme 4 Grant of the LMS.
| 54b953bfcc6fd80360c9f11474db653b4235b0e6 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Increasing the number of cellular base stations~(BSs) to serve a given area, also known as network densification, is foreseen to be a necessary solution
to address the large data rate
demands of the future fifth-generation~(5G) wireless communication networks~\cite{andrews-etal2014what,bhushan-etal2014network}. Cloud radio access network~(CRAN) provides a cost-effective way to achieve network densification, by replacing the conventional BSs with low-power
distributed remote radio heads~(RRHs) that are deployed close to the users and coordinated by a central processor~(CP)~\cite{andrews-etal2014what}. CRAN significantly improves both the spectral efficiency and energy efficiency compared to conventional cellular networks, due to the centralized resource allocation and joint signal processing over the RRHs at the CP~\cite{park-etal2013joint,zhou-yu2014optimized,liu-zhang2015optimized,jain-etal2016backhaul,zhuang-lau2014backhaul,dai-yu2014sparse,shi-etal2014group,fan-etal2016randomized,luo-etal2015downlink,tao-etal2016content,shi-etal2016smoothed,fan-etal2016dynamic,peng-etal2016recent,liu-etal2015joint,stephen-zhang2016green}. Combining a dense network with centralized joint processing as in the CRAN leads to a powerful new network architecture termed ultra-dense CRAN~(UD-CRAN)~\cite{shi-etal2015large,stephen-zhang2017joint}, in which the number of RRHs can even exceed the number of users being served in a given area to support wireless connectivity of ultra-high throughput.
In a UD-CRAN, the RRHs exchange data and control signals with the CP via high-speed wired or wireless links which are referred to as the fronthaul. With a fully centralized architecture, where the CP performs all the encoding/decoding operations, the RRHs in a UD-CRAN can be simple relay nodes that transmit or receive quantized/compressed baseband signals over the fronthaul links~\cite{park-etal2013joint,zhou-yu2014optimized,liu-etal2015joint,liu-zhang2015optimized,jain-etal2016backhaul}. Although such an architecture provides the maximum joint signal processing gains, the fronthaul links can get saturated easily by the large volume of signals that need to be transmitted over them, especially in UD networks. On the other hand, in traditional cellular networks, the BSs themselves have encoding/decoding capability, and only the user messages are sent over the backhaul, which typically requires much less capacity compared to sending quantized signals over the fronthaul in CRAN. However, in such networks, the performance gains
due to joint signal processing are compromised. This motivates an alternative \emph{hybrid} architecture for UD-CRAN as proposed in~\cite{bi-etal2015wireless}, by which the benefits of both the BS-centric local processing in conventional cellular networks and the centralized processing in CRAN can be achieved by adaptively switching between the two, based on e.g., the wireless channel conditions, the user rate requirements and the fronthaul constraints of the RRHs.
In this paper, we consider the uplink transmission in an orthogonal frequency division multiple access~(OFDMA)-based UD-CRAN with hybrid decoding on each sub-channel~(SC), as shown in~\cref{F:SysModel}. On each SC, the RRHs can choose either to: decode the assigned user's message locally and then forward it to the CP~(namely, decode-and-forward~(DaF)), or simply quantize the signal and forward it to the CP for joint decoding~(termed forward-and-decode~(FaD)), or not process the received signal at all~(to save fronthaul capacity). Most of the existing work on CRAN considers single-channel systems~\cite{park-etal2013joint,zhou-yu2014optimized,liu-zhang2015optimized,jain-etal2016backhaul,zhuang-lau2014backhaul,dai-yu2014sparse,shi-etal2014group,fan-etal2016randomized,luo-etal2015downlink,tao-etal2016content,shi-etal2016smoothed,fan-etal2016dynamic,peng-etal2016recent}, where all the users transmit in the same bandwidth. Previously, a hybrid compression and message-sharing strategy was considered in~\cite{patil-yu2014hybrid} for the downlink transmission in a single-channel CRAN, where the CP shares the uncompressed messages of some users to a subset of RRHs, in addition to transmitting compressed messages of all users to each RRH. In contrast, in this paper we consider the OFDMA-based UD-CRAN with multiple SCs for the uplink transmission with hybrid decoding on each SC. Although allowing multiple users to transmit on the same SC can potentially increase the system throughput, it would require the more complex successive interference cancellation receiver on each SC at the CP and/or the RRHs, compared to the simple single-user decoder considered in this paper. The uplink transmission in an OFDMA-based CRAN,
with only FaD processing on each SC was considered in~\cite{liu-etal2015joint}. In contrast, with the hybrid decoding considered in this paper, each RRH can choose to first decode the message in an SC locally, before forwarding it to the CP, in addition to quantizing and forwarding the signal to the CP as in~\cite{liu-etal2015joint}. For the downlink of a multi-hop fronthaul CRAN, it has been shown in~\cite{liu-yu2017cross} that multicasting user messages with the help of network coding can perform better than unicasting compressed messages with simple routing.
Specifically, we consider a cluster of $M$ RRHs and $K$ users, both with single-antennas, in an OFDMA-based UD-CRAN with $N$ orthogonal SCs,
as shown in~\cref{F:SysModel}. The signal transmitted by the user assigned to each SC can be either quantized and forwarded by a subset of RRHs for joint decoding at the CP, or instead, locally decoded and forwarded by a single RRH. Jointly decoding the message of a user assigned to any SC at the CP provides a receive-combining gain from multiple RRHs over the local decoding
at a single RRH, and hence achieves a higher transmission rate for the user in general. However, to achieve this rate improvement,
the participating RRHs need to forward the quantized signals with sufficiently low error to the CP, which consumes more of their fronthaul capacities compared to a decoded message forwarded by a single RRH. On the other hand, when an RRH forwards a decoded message, the fronthaul capacity is significantly saved, but the receive-combining gain achievable with centralized decoding at the CP is compromised, and only a selection diversity gain over the RRHs can be attained. Besides the selection of DaF or FaD processing modes, the achievable rate for each SC also depends on the user-SC assignment and transmit power allocations by the users. Thus, the RRHs' processing mode~(DaF/FaD/no processing) selections on each SC should be jointly optimized along with the resource allocation in the network.
The main results of this paper are summarized as follows.
\begin{itemize}
\item We study a joint resource allocation problem in UD-CRAN, including the RRHs' processing mode selections, user-SC assignments and the users' transmit power allocations for OFDMA transmission, to maximize the users' weighted-sum-rate in the uplink transmission. To the best of our knowledge, this problem has not been considered yet in the literature. The problem, however, is non-convex, and an exhaustive search over all possible solutions would incur an exponential complexity of $O((K(2^M+M))^N)$, which is evidently not practically affordable in a UD-CRAN with large values of $N$ and/or $M$.
\item Hence, we propose a Lagrange duality based algorithm, which can achieve the optimal solution asymptotically when the number of SCs is large, at a much reduced complexity of $O(NK(2^M+M))$.
\item To further reduce the complexity in UD networks, we propose a greedy algorithm to obtain a suboptimal solution, which has a lower complexity of only $O(NK(M^2+M))$, and is shown to be able to achieve close-to-optimal throughput performance under various practical setups by simulations.
\item Finally, we compare the proposed optimal and suboptimal algorithms with hybrid decoding to both the conventional CRAN that performs FaD processing on all SCs, and a cellular network that performs DaF processing on all SCs by simulations, which show promising throughput gains by the proposed algorithms.
\end{itemize}
{\it Notation}: In this paper, scalars are denoted by lower-case letters, e.g., $x$, while vectors are denoted by bold-face lower-case letters, e.g., $\bm x$. The set of real numbers, complex numbers, and integers are denoted by $\mathbbm{R}$, $\mathbbm{C}$ and $\mathbbm{Z}$, respectively. Similarly, $(\cdot)^{x\times 1}$ denotes the corresponding space of $x$-dimensional column vectors, while $\mathbbm{R}_+$ and $\mathbbm{Z}_{++}$ denote the sets of non-negative real numbers and positive integers respectively. For a real scalar $x\in\mathbbm{R}$, $[x]^+\triangleq\max\{x,0\}$,
and for a complex scalar $x\in\mathbbm{C}$, $|x|\geq 0$ denotes the magnitude of $x$.
For a vector $\bm x$, $\bm x^\mathsf{T}$ denotes its transpose. Vectors with all elements equal to $1$ and $0$ are denoted by $\bm 1$ and $\bm 0$, respectively. For
real vectors $\bm x,\bm y\in\mathbbm{R}^{M\times 1}$,
$\bm x\succeq\bm y$
denotes the
component-wise inequalities $x_i\geq y_i, i=1,\dotsc,M$. Finally, $\mathcal{CN}(\mu,\sigma^2)$ denotes
a circularly symmetric complex Gaussian~(CSCG) random variable with mean $\mu$ and variance $\sigma^2$, and the symbol $\sim$ is used to mean ``distributed as".
\section{System Model}\label{Sec:SysMod}
\begin{figure}[t]
\centering
\includegraphics[width=\linewidth]{Fig1_SysMod}
\caption{OFDMA-based uplink CRAN with hybrid-decoding RRHs.}\label{F:SysModel}
\end{figure}
We study the uplink transmission in a UD-CRAN cluster that consists of $M$ single-antenna RRHs, denoted by the set $\mathcal{M}\triangleq\{1,\dotsc,M\}$, and $K$ single-antenna users, denoted by $\mathcal{K}\triangleq\{1,\dotsc,K\}$, as shown in~\cref{F:SysModel}. The users transmit their signals in the uplink using OFDMA over a total bandwidth of $B$~Hz
which is equally divided into $N$ orthogonal SCs, denoted by the set
$\mathcal{N}\triangleq\{1,\dotsc,N\}$.
These signals are received by the RRHs, and are then forwarded
to the CP via the individual fronthaul links of the RRHs. As the fronthaul link capacity is practically limited for each RRH, one of the following operations are chosen to be performed on each SC $n\in\mathcal{N}$ by each RRH $m\in\mathcal{M}$:
\begin{itemize}
\item The user's message is decoded and forwarded to the CP~(DaF);
\item The user's signal is quantized and forwarded to the CP, which then decodes the message~(FaD); or
\item The received signal is not processed.
\end{itemize}
For example, an illustration is given in~\cref{F:SysModel} with 5~RRHs and 2~users, where the SCs~1 and~3 are assigned to user~1, while SCs~2 and~4 are assigned to user~2. In this case, RRH~2 which is nearest to user~1, decodes the message in SC~3 and forwards it to the CP. Since this message is decoded by RRH~2, the other RRHs do not process the signal on SC~3. On the other hand, user~1's signal in SC~1 is quantized independently by each of the RRHs~2~and~4, and forwarded to the CP, which then decodes the message in SC~1 by combining the quantized signals from these two RRHs. Similarly for user~2, while RRH~1 locally decodes and forwards the message in SC~4, RRHs~1, 4~and~5 quantize and forward their respective received signals in SC~2. Notice that RRH~1, despite its proximity to user~1, does not process user~1's signal on SC~1, since its fronthaul is already fully consumed by user~2's quantized signal in SC~2 and the decoded message in SC~4.
In order to describe the choice of operations outlined above completely, we introduce the following decision variables. First, to indicate whether an RRH $m$ processes the signal on SC $n$ or not, we define the variables $\alpha_{m,n}$
as
\begin{align}
\alpha_{m,n}\triangleq\begin{cases}1&\text{if RRH}~m~\text{processes the signal on SC}~n\\
0&\text{otherwise,}\end{cases
\label{E:ProcSel}
\end{align}
$m\in\mathcal{M}$ and $n\in\mathcal{N}$. The collection of these variables for all RRHs at each SC $n\in\mathcal{N}$ is denoted by the vector $\bm\alpha_n\triangleq\begin{bsmallmatrix}\alpha_{1,n}&\cdots&\alpha_{M,n}\end{bsmallmatrix}^\mathsf{T}\in\{0,1\}^{M\times 1}$.
Next, to indicate the mode of processing chosen for SC $n$, i.e. whether the message in SC $n$ is decoded by an RRH or instead the signal in SC $n$ is quantized and forwarded to the CP, we define the mode selection
variable $\delta_n$ as
\begin{align}
\delta_n\triangleq\begin{cases}1&\text{if the message in SC}~n~\text{is decoded
}\\
0&\text{if the signal in SC}~n~\text{is quantized,
}\end{cases}\label{E:ModeSel
\end{align}
$n\in\mathcal{N}$.
In the DaF mode~$(\delta_n=1)$, at most one RRH in the network processes the
signal in SC $n$, and hence we impose the condition $\bm 1^\mathsf{T}\bm\alpha_n\leq 1$.
On the other hand,
in the FaD mode
$(\delta_n=0)$, a subset $\mathcal{A}_n\subseteq\mathcal{M}$ consisting of one or more RRHs, independently quantize and forward the signals in SC $n$, where $\mathcal{A}_n\triangleq\{m\in\mathcal{M}|\alpha_{m,n}=1\},~n\in\mathcal{N}$.
Finally, we define the variable $\nu_{k,n}$ to indicate whether an SC $n$ is assigned to a user $k$ or not, i.e.,
\begin{align}
\nu_{k,n}\triangleq\begin{cases}1&\text{if SC}~n~\text{is assigned to user}~k\\
0&\text{otherwise},
\end{cases}\label{E:UserSel}
\end{align}
$k\in\mathcal{K},n\in\mathcal{N}$. The vector $\bm\nu_n\triangleq\begin{bsmallmatrix}
\nu_{1,n}&\cdots&\nu_{K,n}
\end{bsmallmatrix}^\mathsf{T}\in\{0,1\}^K$ specifies the user assignment for SC $n$. According to OFDMA, each SC $n\in\mathcal{N}$ is assigned to at most one user for uplink transmission, and thus
$\bm\nu_n$ must satisfy the condition $\bm 1^\mathsf{T}\bm\nu_n\leq 1,\forall n\in\mathcal{N}$. Also, we denote the set of SCs assigned to user $k\in\mathcal{K}$ by $\mathcal{N}_k\triangleq\{n|\nu_{k,n}=1\}\subseteq\mathcal{N}$, so that
$\mathcal{N}_j\cap\mathcal{N}_k=\emptyset,\ \forall j\neq k,\ j,k\in\mathcal{K}$.
Let $h_{m,k,n}\in\mathbbm{C}$ denote the complex wireless channel coefficient from the user $k\in\mathcal{K}$ to RRH $m\in\mathcal{M}$, for SC $n\in\mathcal{N}$. We assume that all the channel coefficients $h_{m,k,n}$'s are known at the CP. Let $p_{k,n}\geq 0$ denote the uplink transmit power allocated by the user $k\in\mathcal{K}$ to SC $n\in\mathcal{N}$ and the vector $\bm p_n\triangleq\begin{bsmallmatrix}
p_{1,n}&\cdots&p_{K,n}
\end{bsmallmatrix}^\mathsf{T}\in\mathbbm{R}_+^K$ denote the transmit powers
of all the users on SC $n\in\mathcal{N}$.
Then, the
signal received at RRH $m$ in SC $n\in\mathcal{N}_k$
is
given by
\begin{align}
y_{m,n}=h_{m,k,n}\sqrt{p_{k,n}}s_{k,n}+z_m
\label{E:RxSig}
\end{align} where $s_{k,n}\sim\mathcal{CN}(0,1)$ denotes user $k$'s information-bearing signal that is assumed to be complex Gaussian
and $z_m\sim\mathcal{CN}(0,\sigma_m^2)$ is the additive-white-Gaussian-noise~(AWGN), with $\sigma_m^2$ denoting the receiver noise power at RRH $m$~(assumed to be equal for all SCs).
\subsection{Forward-and-Decode~(FaD) Processing}\label{SS:JD}
When FaD processing~$(\delta_n=0)$ is performed on an SC $n\in\mathcal{N}_k$, the following operations take place. First, the RRHs $m\in\mathcal{A}_n$, i.e. with $\alpha_{m,n}=1$, quantize the received baseband signals $y_{m,n}$'s in SC $n$ given by~\eqref{E:RxSig}. Then, the RRHs encode these quantized values into digital codewords, and forward them
to the CP. The CP recovers the quantized signals from these digital codewords, and decodes user $k$'s message on SC $n\in\mathcal{N}_k$, by combining the quantized signals from the RRHs. As the received signals at the RRHs in all the SCs are independent of each other and each RRH is assumed to perform signal quantization independently, a simple scalar quantization~(SQ) on the received signals $y_{m,n}$'s
is performed~\cite{liu-etal2015joint}, and the baseband signal after SQ can be expressed as
\begin{align}
\hat{y}_{m,n}=y_{m,n}+e_{m,n},\ m\in\mathcal{M},n\in\mathcal{N},\label{E:QuantRxSig}
\end{align}
where $e_{m,n}$ denotes the complex-valued error induced by the SQ, which is assumed to have zero mean and a variance denoted by $q_{m,n}$. The errors $e_{m,n}$'s are independent over $m$ and $n$ due to the independent SQ in each SC by each RRH. A practical method for SQ is to represent the in-phase~(I) and quadrature~(Q) components of the received complex baseband signal $y_{m,n}$ in~\eqref{E:RxSig} using $\beta_m\in\mathbbm{Z}_{++}$ bits for each.\footnote{It is assumed that the resolution of the SQ, $\beta_m$, is fixed, and is the same on all SCs for a given RRH $m\in\mathcal{M}$.} In other words, each of the I~and~Q components of $y_{m,n}$ are uniformly quantized into one of $2^{\beta_m}$ levels~\cite{liu-etal2015joint}. With such a uniform SQ,
the variance $q_{m,n}$ of the quantization error $e_{m,n}$ in~\eqref{E:QuantRxSig} on SC $n\in\mathcal{N}_k$ at RRH $m\in\mathcal{M}$ can be approximated as~\cite{liu-etal2015joint}
\begin{align}
q_{m,n}=3(|h_{m,k,n}|^2p_{k,n}+\sigma_m^2)2^{-2\beta_m}
.\label{E:QNVar}
\end{align}
Next, each RRH $m\in\mathcal{A}_n$ transmits the digital code-words corresponding to the quantization levels to the CP via its fronthaul link. Since each I and Q component of the signal in an SC is represented by a $\beta_m$-bit codeword, it is not difficult to show that the transmission rate in bits-per-second~(bps) required on the fronthaul link of RRH $m$ to forward the quantized signal in any SC $n$ is given by $2B\beta_m/N$. Upon receiving the digital code-words, the CP first recovers the baseband quantized signals $\hat{y}_{m,n}$'s based on the quantization code-books used by each RRH. Further, to decode the message in SC $n$, the CP applies a linear combining to $\hat{y}_{m,n}$'s. When the optimal combining weights that maximize the signal-to-noise ratio~(SNR) are used, the received SNR for decoding at the CP on SC $n\in\mathcal{N}_k$ can be expressed as~\cite{liu-etal2015joint}
\begin{align}
\gamma^Q_{k,n}(\bm\alpha_n,p_n)&=\sum_{m\in\mathcal{M}}\frac{\alpha_{m,n}|h_{m,k,n}|^2p_n}{\sigma_m^2+q_{m,n}}\notag\\
&=\sum_{m\in\mathcal{M}}\alpha_{m,n}\gamma^Q_{m,k,n}(p_{k,n}),
\label{E:RxSNRJD}
\end{align}
where $q_{m,n}$ is
given by~\eqref{E:QNVar} and $\gamma^Q_{m,k,n}(p_{k,n})$ denotes the contribution of each RRH $m$ to the SNR at the CP in SC $n$. Note that if $\alpha_{m,n}=0$ for some RRH $m\in\mathcal{M}$, it means RRH $m$ does not process the signal on SC $n$ and hence does not contribute to the received SNR at the CP.
Using~\eqref{E:QNVar} in~\eqref{E:RxSNRJD}, $\gamma^Q_{m,k,n}(p_{k,n})$ can be expressed as
\begin{align}
\gamma^Q_{m,k,n}(p_{k,n})=\frac{|h_{m,k,n}|^2p_{k,n}}{\sigma_m^2+3\left(|h_{m,k,n}|^2p_{k,n}+\sigma_m^2\right)2^{-2\beta_m}}
\label{E:PartSNR}
\end{align}
$m\in\mathcal{M},k\in\mathcal{K},n\in\mathcal{N}_k$. If the quantization error $e_{m,n}$ is assumed to be the worst-case Gaussian distributed,
then a lower bound on the achievable rate in bps with FaD processing for SC $n\in\mathcal{N}_k$
is given by
\begin{align}
&r^Q_{k,n}(\bm\alpha_n,p_{k,n})=(B/N)\log_2(1+\gamma^Q_{k,n}(\bm\alpha_n,p_{k,n}))
.\label{E:SCRJ}
\end{align}
Next, we present the following lemma.
\begin{lemma}\label{L:rknQConc}
With given RRH selection $\bm\alpha_n$,
$r^Q_{k,n}(\bm\alpha_n,p_{k,n})$ defined in~\eqref{E:SCRJ} is concave in $p_{k,n}\geq 0$.
\end{lemma}
\begin{proof}
Please refer to~appendix~\ref{App:ProofLrknQConc}.
\end{proof}
\cref{L:rknQConc} indicates that on each SC $n$, the achievable rate with FaD processing is a concave function of the assigned user's transmit power.
\subsection{Decode-and-Forward~(DaF) Processing}
Instead of having the CP perform the decoding by combining the quantized signals from the RRHs, any single selected RRH $m$ can also decode the message in SC $n$ locally, and then forward this message to the CP. The maximum achievable rate in bps for SC $n\in\mathcal{N}_k$ when RRH $m$ locally decodes user $k$'s message is given by
\begin{align}
r^D_{m,k,n}(p_{k,n})=(B/N)\log_2(1+|h_{m,k,n}|^2p_{k,n}/\sigma_m^2)
.\label{E:SCRL}
\end{align}
In this case, RRH $m$ forwards the decoded message on SC $n\in\mathcal{N}_k$ to the CP over its fronthaul link at the rate of at least $r^D_{m,k,n}(p_{k,n})$. Also, notice that if some RRH $m\in\mathcal{M}$ locally decodes the message in SC $n$ and forwards it to the CP, no other RRHs need to process the signal on this SC.
\subsection{Hybrid Decoding}
Here, we present the proposed hybrid decoding scheme, where either a subset of RRHs $\mathcal{A}_n\subseteq\mathcal{M}$ quantize and forward their received signals to be
jointly decoded at the CP as in a conventional CRAN, or a single RRH $m\in\mathcal{M}$ locally decodes a user's message on SC $n$ and forwards it to the CP.
Combining the expressions for the achievable rates in these two cases as given in~\cref{E:SCRJ,E:SCRL} with the indicator variables defined in~\cref{E:ModeSel,E:ProcSel}, the achievable rate with hybrid decoding on an SC $n\in\mathcal{N}_k,\,k\in\mathcal{K},$
is given by
\begin{align
r_{k,n}(\delta_n,\bm\alpha_n,p_{k,n})&=\delta_n\sum_{m\in\mathcal{M}}\alpha_{m,n}r^D_{m,k,n}(p_{k,n})\notag\\
&\quad+(1-\delta_n)r^Q_{k,n}(\bm\alpha_n,p_{k,n})
\label{E:SCR}
\end{align}
Note that if $\bm\alpha_n=\bm 0$, no RRH processes the received signals and the achievable rate on SC $n$
will be zero, irrespective of the value of
$\delta_n$. Also, for an SC $n\in\mathcal{N}_k$, if $p_{k,n}=0$, we assume $\bm\alpha_n=\bm 0$ without loss of generality.
Let $\bar{R}_m$ denote the
fronthaul capacity of the link connecting RRH $m$ to the CP in bps. Then, the total rate at which RRH $m$
forwards the processed signals~(decoded and/or quantized) over all the $N$ SCs from the users to the CP
must satisfy the constraints
\begin{align
&\sum_{n\in\mathcal{N}}\bigg(\delta_n\alpha_{m,n}\sum_{k\in\mathcal{K}}\nu_{k,n}r^D_{m,k,n}(p_{k,n})\notag\\
&\qquad+(1-\delta_n)\frac{2B\beta_m\alpha_{m,n}}{N}\bigg)\leq \bar{R}_m,\,\forall m\in\mathcal{M}.
\label{E:IFHRC}
\end{align}
The left-hand side of~\eqref{E:IFHRC} represents the total rate at which the signals processed by RRH $m$ must be forwarded to the CP,
where $r^D_{m,k,n}(p_{k,n})$ is defined in~\eqref{E:SCRL}, and $2B\beta_m/N$ is the rate in bps corresponding to the resolution $\beta_m$ of the SQ performed by RRH $m$.
In the next section, we formulate the joint uplink resource allocation optimization problem for the OFDMA-based UD-CRAN with hybrid decoding.
\section{Problem Formulation}\label{Sec:ProbForm}
We aim to maximize the weighted-sum-rate of the users in the uplink over all the SCs, by jointly optimizing
the processing mode and RRH selections, which are collectively represented by the variables $\{\delta_n,\bm\alpha_n\}_{n\in\mathcal{N}}$, the user-SC assignments $\{\bm\nu_n\}_{n\in\mathcal{N}}$, and the users' transmit power allocations $\{\bm p_n\}_{n\in\mathcal{N}}$, subject to the fronthaul constraints~\eqref{E:IFHRC}, and the total power constraints at each of the users, which we denote by $\bar{P}_k,\ k\in\mathcal{K}$. Let $\omega_k\geq 0$ denote the rate weight for user $k\in\mathcal{K}$. Then the problem can be formulated as given below.
\begin{subequations}
\label{P:ULHybMain}
\begin{align}
\mathop{\mathrm{maximize}}_{\{\bm p_n,\bm\alpha_n,\bm\nu_n,\delta_n\}_{n\in\mathcal{N}}
}&\sum_{k\in\mathcal{K}}\omega_k\sum_{n\in\mathcal{N}}\nu_{k,n}r_{k,n}(\delta_n,\bm\alpha_n,p_{k,n})\tag{\ref{P:ULHybMain}}\\
\mathrm{subject}~\mathrm{to}\notag\\
&\ \text{\cref{E:IFHRC}}\notag\\
&\ \sum_{n\in\mathcal{N}} p_{k,n}\leq\bar{P}_k\quad\forall k\in\mathcal{K}
\label{C:PCULHybMain}\\
&\ p_{k,n}\geq 0\quad\forall k\in\mathcal{K},\forall n\in\mathcal{N}\label{C:PosPowULHybMain}\\
&\ \delta_n(\bm 1^\mathsf{T}\bm \alpha_n)\leq 1\quad\forall n\in\mathcal{N}\label{C:LDConstULHybMain}\\
&\ \alpha_{m,n}\in\{0,1\}\quad\forall m\in\mathcal{M},\forall n\in\mathcal{N}\label{C:RRHSelULHybMain}\\
&\ \bm 1^\mathsf{T}\bm\nu_n\leq 1\quad\forall n\in\mathcal{N}\label{C:UASUULHybMain}\\
&\ \nu_{k,n}\in\{0,1\}\quad\forall k\in\mathcal{K},\forall n\in\mathcal{N}\label{C:UAVectULHybMain}\\
&\ \delta_n\in\{0,1\}\quad\forall n\in\mathcal{N}.\label{C:ModeSelULHybMain}
\end{align}
\end{subequations}
The constraint~\eqref{C:LDConstULHybMain} ensures that when DaF processing is performed~$\left(\delta_n=1\right)$, at most one RRH processes the signal on SC $n$, i.e. at most one $\alpha_{m,n}=1$,
and all other $\alpha_{m,n}$'s are zero. In the case of FaD processing
$\left(\delta_n=0\right)$, no such restriction is imposed on $\bm\alpha_n$. The above weighted-sum-rate maximization problem is non-convex due to the integer constraints~\cref{C:RRHSelULHybMain,C:ModeSelULHybMain} and the coupled variables in the objective, as well as in constraints~\cref{E:IFHRC,C:LDConstULHybMain}. Notice that if $\delta_n=0\ \forall n\in\mathcal{N}$, then problem~\cref{P:ULHybMain} reduces to the special case where all RRHs quantize and forward their signals to the CP as in an OFDMA-based CRAN with only centralized decoding at the CP~\cite{liu-etal2015joint}. On the other hand, if $\delta_n=1\ \forall n\in\mathcal{N}$,
problem~\cref{P:ULHybMain} reduces to the special case of an OFDMA-based cellular network where the user messages on different SCs can be decoded by different RRHs in general and then forwarded to the CP over the fronthauls. This is unlike the conventional cellular network where each user's signal is decoded by only the single BS to which it is associated. In the general case, the optimal solution to the above problem could have $\delta_n=0$ only on a subset of SCs, and $\delta_n=1$ for other SCs, in order to maximally exploit the hybrid decoding capability.
It is worth noting that in the solution to problem~\eqref{P:ULHybMain}, if $p_{k,n}=0,~\forall n\in\mathcal{N}$ for some user $k$, it means that user $k$ is not served at all in the current scheduling interval. The resulting fairness issue can be overcome in practice by selecting the user rate-weights $\omega_k$'s in problem~\eqref{P:ULHybMain} appropriately. For example, if one user receives very low rate in the current scheduling interval, its priority weight can be increased to ensure that it gets a higher rate in the next interval. The user weights can thus be manipulated to ensure that all the users in the network are fairly served in the long term.
\section{Proposed Solutions}\label{Sec:PropSol}
\subsection{Optimal Solution}\label{SS:OptSol}
Although problem~\eqref{P:ULHybMain} is non-convex,
its duality gap diminishes to zero as the number of SCs $N$ goes to infinity, as under this condition such non-convex problems are shown to satisfy the ``time-sharing" property~\cite{yu-lui2006dual}. Since $N$ is typically large in practice, we propose to apply the Lagrange duality method to obtain an asymptotically optimal solution to problem~\eqref{P:ULHybMain}.\footnote{However, from the simulations in~\cref{Sec:SimResults}, we observe that the duality gap is negligible even for a moderate value of $N=64$.} Let $\lambda_m\geq 0,\ m\in\mathcal{M}$ denote the dual variables associated with the $M$ constraints in~\eqref{E:IFHRC}, and $\mu_k\geq 0,\ k\in\mathcal{K}$, denote the dual variables for the $K$ constraints in~\eqref{C:PCULHybMain}. Also let the vectors $\bm \lambda\triangleq\begin{bsmallmatrix}\lambda_1&\cdots&\lambda_M\end{bsmallmatrix}^\mathsf{T}\in\mathbbm{R}_+^{M\times 1}$ and $\bm\mu\triangleq\begin{bsmallmatrix}
\mu_1&\cdots&\mu_K
\end{bsmallmatrix}^\mathsf{T}\in\mathbbm{R}_+^{K\times 1}$ denote the collections of these dual variables.
Then, the~(partial) Lagrangian of problem~\eqref{P:ULHybMain} is given by
\begin{align
&L\left(\{\delta_n,\bm\nu_n,\bm\alpha_n,\bm p_n\}_{n\in\mathcal{N}},\bm\lambda,\bm\mu\right)\notag\\
&=\sum_{k\in\mathcal{K}}\omega_k\sum_{n\in\mathcal{N}}\nu_{k,n}r_{k,n}(\delta_n,\bm\alpha_n,p_{k,n}
-\sum_{m\in\mathcal{M}}\lambda_m\notag\\
&\quad\cdot\Bigg(\frac{1}{\bar{R}_m}\sum_{n\in\mathcal{N}}\Big(\delta_n\alpha_{m,n}\sum_{k\in\mathcal{K}}\nu_{k,n}r^D_{m,k,n}(p_{k,n})\notag\\
&\quad+(1-\delta_n)\frac{2B\beta_m\alpha_{m,n}}{N}\Big)-1\Bigg
-\sum_{k\in\mathcal{K}}\mu_k\bigg(\sum_{n\in\mathcal{N}} p_{k,n}-\bar{P}_k\bigg)\notag\\
&=\sum_{n\in\mathcal{N}} L_n(\delta_n,\bm\nu_n,\bm\alpha_n,\bm p_n,\bm\lambda,\bm\mu)+\sum_{m\in\mathcal{M}}\lambda_m+\sum_{k\in\mathcal{K}}\mu_k\bar{P}_k,
\label{E:LagW}
\end{align}
where each term in the first summation in~\eqref{E:LagW} is
\begin{align
&L_n\left(\delta_n,\bm\nu_n,\bm\alpha_n,\bm p_n,\bm\lambda,\bm\mu\right)\notag\\
&\triangleq\sum_{k\in\mathcal{K}}\omega_k\nu_{k,n}\bigg(\delta_n\sum_{m\in\mathcal{M}}\alpha_{m,n}r^D_{m,k,n}(p_{k,n})\notag\\
&\quad+(1-\delta_n)r^Q_{k,n}(\bm\alpha_n,p_{k,n})\bigg
-\sum_{m\in\mathcal{M}}\frac{\lambda_m}{\bar{R}_m}\notag\\
&\quad\cdot\bigg(\delta_n\alpha_{m,n}\sum_{k\in\mathcal{K}}\nu_{k,n}r^D_{m,k,n}(p_{k,n})+(1-\delta_n)\frac{2B\beta_m\alpha_{m,n}}{N}\bigg)\notag\\
&\quad-\sum_{k\in\mathcal{K}}\mu_k p_{k,n},\quad n\in\mathcal{N}.\label{E:Lagn}
\end{align}
The Lagrange dual function can thus be expressed as
\begin{subequations}
\label{E:DualFunc}
\begin{align}
g(\bm\lambda,\bm\mu)=\max_{\substack{\{\bm p_n,\bm\alpha_n,\\
\bm\nu_n,\delta_n\}_{n\in\mathcal{N}}}}&\,L\left(\{\delta_n,\bm\nu_n,\bm\alpha_n,\bm p_n,\}_{n\in\mathcal{N}},\bm\lambda,\bm\mu\right)\tag{\ref{E:DualFunc}}\\
\mathrm{s.t.}&\,\text{\cref{C:PosPowULHybMain,C:LDConstULHybMain,C:ModeSelULHybMain}}.\notag
\end{align}
\end{subequations}
The maximization problem in~\eqref{E:DualFunc} can be decomposed into $N$ parallel sub-problems using~\eqref{E:LagW}, where each sub-problem corresponds to a single SC $n$, and has the following structure,
\begin{subequations}
\label{P:DualFuncn}
\begin{align}
\max_{\bm p_n,\bm\alpha_n,\bm\nu_n,\delta_n}&\:L_n(\delta_n,\bm\nu_n,\bm\alpha_n,\bm p_n,\bm\lambda,\bm\mu)\tag{\ref{P:DualFuncn}}\nonumber\\
\mathrm{s.t.}&\:\bm p_n\succeq\bm 0\label{C:PNZDFn}\\
&\ \delta_n(\bm 1^\mathsf{T}\bm\alpha_n)\leq 1\label{C:ModeRRHSelDFn}\\
&\ \bm\alpha_n\in\{0,1\}^{M\times 1}\label{C:RRHSelVectDFn}\\
&\ \bm 1^\mathsf{T}\bm\nu_n\leq 1\label{C:UASUDFn}\\
&\ \bm\nu_n\in\{0,1\}^{K\times 1}\label{C:UAVectDFn}\\
&\ \delta_n\in\{0,1\},\label{C:ModeDFn}
\end{align}
\end{subequations}
with the objective function given by~\eqref{E:Lagn}. Problem~\eqref{P:DualFuncn} is non-convex due to the integer constraints~\cref{C:RRHSelVectDFn,C:UAVectDFn,C:ModeDFn}
and also due to the coupled variables in the objective and the constraint~\eqref{C:ModeRRHSelDFn}. However, since the processing mode selection variable $\delta_n$ takes on only two values 0 or 1, we solve problem~\eqref{P:DualFuncn} for each of these two cases separately, and then choose the
value of $\delta_n$ that gives the maximum objective value for SC $n$. Specifically, let $\bar{L}^Q_n$ denote the optimal value of problem~\eqref{P:DualFuncn} for the FaD mode~$(\delta_n=0)$ and $\bar{L}^D_n$ denote the
corresponding value for the DaF mode~$(\delta_n=1)$. Then, for given dual variables $\bm\lambda$ and $\bm\mu$, the optimal processing mode selection $\bar{\delta}_n$ for problem~\eqref{P:DualFuncn} is given by
\begin{align}
\bar{\delta}_n=\begin{cases}
0&\text{if }\bar{L}^{Q}_n>\bar{L}^{D}_n\\
1&\text{otherwise.}\label{E:OptModeDFn}
\end{cases}
\end{align}
In the following two subsections, we explain in detail how to solve problem~\eqref{P:DualFuncn} for each processing mode, in order to obtain $\bar{L}^{Q}_n$ and $\bar{L}^{D}_n$.
\subsubsection{The Case of FaD Processing~$\left(\delta_n=0\right)$}\label{SS:FD}
With
$\delta_n=0$,
problem~\eqref{P:DualFuncn} reduces to
\begin{subequations}
\label{P:DFnJD}
\begin{align}
\max_{\bm p_n,\bm\alpha_n,\bm\nu_n}&\ L^Q\left(\bm\nu_n,\bm\alpha_n,\bm p_n,\bm\lambda,\bm\mu\right)\tag{\ref{P:DFnJD}}\\
\mathrm{s.t.}&\ \text{\cref{C:PNZDFn,C:RRHSelVectDFn,C:UASUDFn,C:UAVectDFn}},\notag
\end{align}
\end{subequations}
where the objective function is given by
\begin{align
&L^Q(\bm\nu_n,\bm\alpha_n,\bm p_n,\bm\lambda,\bm\mu)\notag\\
&\triangleq\sum_{k\in\mathcal{K}}\nu_{k,n}\omega_kr_{k,n}^Q\left(\bm\alpha_n,p_{k,n}\right)
-\frac{2B}{N}\sum_{m\in\mathcal{M}}\frac{\beta_m\lambda_m\alpha_{m,n}}{\bar{R}_m}\notag\\
&\quad-\sum_{k\in\mathcal{K}}\mu_k p_{k,n}\label{E:ObjDFnJD}
\end{align}
Now, let the user assignment on SC $n$ be fixed as $\bm\nu_n=\hat{\bm\nu}_n$.
If no user is assigned to SC $n$, i.e. $\hat{\bm\nu}_n=\bm 0$, it is evident from~\eqref{E:ObjDFnJD} that the objective of problem~\cref{P:DFnJD}
is maximized by setting $\bm p_n=\bm 0$ and $\bm\alpha_n=\bm 0$,
i.e.
the power allocation is zero and no RRH processes the signal, as expected. Otherwise, if
SC $n$ is assigned to a user $\hat{k}_n\in\mathcal{K}$,
so that $\hat{\nu}_{\hat{k}_n,n}=1$, and $\hat{\nu}_{k,n}=0~\forall k\neq\hat{k}_n$, then problem~\eqref{P:DFnJD} can be written as
\begin{subequations
\label{P:DFnJDFixUA}
\begin{align}
\max_{\substack{p_{\hat{k}_n,n}\geq 0,\\
\bm\alpha_n\in\{0,1\}^M}}&\,\omega_{\hat{k}_n}r_{\hat{k}_n,n}^Q\big(\bm\alpha_n,p_{\hat{k}_n,n}\big)-\frac{2B}{N}\sum_{m\in\mathcal{M}}\frac{\beta_m\lambda_m\alpha_{m,n}}{\bar{R}_m}\notag\\
&\,-\mu_{\hat{k}_n}p_{\hat{k}_n,n}.\tag{\ref{P:DFnJDFixUA}}
\end{align}
\end{subequations}
Although the above problem~\eqref{P:DFnJDFixUA} is non-convex due to the integer variables $\bm\alpha_n$ and the coupled variables in the objective, when the selection of RRHs is also fixed as $\bm\alpha_n=\tilde{\bm\alpha}_n$, it reduces to the problem
\begin{subequations}
\label{P:DFnJDFUARRHSel}
\begin{align}
\max_{p_{\hat{k}_n,n}\geq 0}&\:\omega_{\hat{k}_n}r_{\hat{k}_n,n}^Q\big(\tilde{\bm\alpha}_n,p_{\hat{k}_n,n}\big)-\mu_{\hat{k}_n} p_{\hat{k}_n,n},\tag{\ref{P:DFnJDFUARRHSel}}
\end{align}
\end{subequations}
which is convex due to~\cref{L:rknQConc}. In the special case when only one RRH quantizes and forwards the signal in SC $n$, i.e. when $\bm 1^\mathsf{T}\tilde{\bm\alpha}_n=1$, the optimal user power allocation $\tilde{p}_{\hat{k}_n,n}$ that solves problem~\eqref{P:DFnJDFUARRHSel} can be obtained in a closed-form as given by the following proposition.
\begin{proposition}\label{Prop:OptPASRRHQ}
Let
$\tilde{m}_n\in\mathcal{M}$ be the single RRH that quantizes and forwards the signal in SC $n$, so that $\tilde{\alpha}_{\tilde{m}_n,n}=1$ and $\tilde{\alpha}_{m,n}=0\ \forall m\neq\tilde{m}_n$. Then, the optimal power allocation that solves problem~\eqref{P:DFnJDFUARRHSel} is given by
\begin{align
\tilde{p}_{\hat{k}_n}&=\frac{\left(\theta_{\tilde{m}_n}+2\right)\sigma_{\tilde{m}_n}^2}{2|h_{\tilde{m}_n,\hat{k}_n,n}|^2}\Bigg(\bigg(1+\frac{4}{\left(\theta_{\tilde{m}_n}+2\right)^2}\notag\\
&\quad\cdot\bigg[\frac{\omega_{\hat{k}_n} B|h_{\tilde{m}_n,\hat{k}_n,n}|^2\theta_{\tilde{m}_n}}{\sigma_{\tilde{m}_n}^2\mu_{\hat{k}_n} N\ln 2}-\left(\theta_{\tilde{m}_n}+1\right)\bigg]^+\bigg)^{\frac{1}{2}}-1\Bigg),\label{E:OptPASingleRRH}
\end{align}
where $\theta_{\tilde{m}_n}\triangleq 2^{2\beta_{\tilde{m}_n}}/3>0$.
\end{proposition}
\begin{proof}
Please refer to appendix~\ref{App:ProofOptPASRRHQ}.
\end{proof}
\cref{Prop:OptPASRRHQ} shows that the optimal user power allocation in the case of FaD processing with a single RRH selection has a threshold structure and is non-zero only if $\tfrac{|h_{\tilde{m}_n,\hat{k}_n,n}|^2}{\sigma_{\tilde{m}_n}^2}>\tfrac{\mu_{\hat{k}_n} N\ln 2}{\omega_{\hat{k}_n}B}\big(1+\tfrac{1}{\theta_{\tilde{m}_n}}\big)$.
On the other hand, if more than one RRHs quantize and forward the signal in SC $n$, i.e. if $\bm 1^\mathsf{T}\tilde{\bm\alpha}_n>1$, then
the optimal user power allocation $\tilde{p}_{\hat{k}_n,n}$ for problem~\eqref{P:DFnJDFUARRHSel} can be efficiently obtained by a one-dimensional line search whose complexity does not depend on the number of RRHs $M$. Thus, for a fixed RRH selection $\tilde{\bm\alpha}_n$, $\tilde{p}_{\hat{k}_n,n}$ can be efficiently obtained either using~\eqref{E:OptPASingleRRH} when $\bm 1^\mathsf{T}\tilde{\bm\alpha}_n=1$, or via a one-dimensional line search, when $\bm 1^\mathsf{T}\tilde{\bm\alpha}_n>1$.
The optimal solution to the joint RRH selection and user power allocation problem~\eqref{P:DFnJDFixUA} can thus be obtained by solving problem~\eqref{P:DFnJDFUARRHSel}
for all the $2^M$ possible RRH selections $\tilde{\bm\alpha}_n\in\{0,1\}^M$,
and then choosing the RRH selection with the corresponding optimal user power allocation that gives the largest objective value for~\eqref{P:DFnJDFixUA}. Since the optimal user power allocation can be obtained either in closed-form using~\eqref{E:OptPASingleRRH} when $\bm 1^\mathsf{T}\tilde{\bm\alpha}_n=1$, or using a one-dimensional line search whose complexity does not depend on $M$, when $\bm 1^\mathsf{T}\tilde{\bm\alpha}_n>1$, the overall complexity of optimally solving problem~\eqref{P:DFnJDFixUA} is $O\left(2^M\right)$. Finally, after solving problem~\eqref{P:DFnJDFixUA} for each user assignment $\hat{k}_n\in\mathcal{K}$,
the optimal solution to problem~\eqref{P:DFnJD} for the FaD mode can be obtained by choosing the user assignment and corresponding RRH selection and power allocation that maximizes the objective in~\eqref{P:DFnJD}. Thus,
the overall complexity of solving problem~\eqref{P:DFnJD} optimally is $O\left(K2^M\right)$.
The next section describes how to solve problem~\eqref{P:DualFuncn} for the other DaF processing mode~$(\delta_n=1)$.
\subsubsection{The Case of DaF Processing~$\left(\delta_n=1\right)$}\label{SS:DF}
In the DaF mode, $\delta_n=1$ and problem~\eqref{P:DualFuncn} reduces to
\begin{subequations}
\label{P:DFnLD}
\begin{align}
\max_{\bm p_n\succeq\bm 0,\bm\alpha_n\in\{0,1\}^M}&\,L^D(\bm\nu_n,\bm\alpha_n,\bm p_n,\bm\lambda,\bm\mu)\tag{\ref{P:DFnLD}}\\
\mathrm{s.t.
&\,\bm 1^\mathsf{T}\bm\alpha_n\leq 1,\label{C:RRHSelDFnLD}
\end{align}
\end{subequations}
where the objective function is given by
\begin{align
&L^D(\bm\nu_n,\bm\alpha_n,\bm p_n,\bm\lambda,\bm\mu)\notag\\
&\triangleq\sum_{k\in\mathcal{K}}\omega_k\nu_{k,n}\sum_{m\in\mathcal{M}}\alpha_{m,n}r^D_{m,k,n}(p_{k,n})-\sum_{m\in\mathcal{M}}\frac{\lambda_m\alpha_{m,n}}{\bar{R}_m}\notag\\
&\quad\cdot\sum_{k\in\mathcal{K}}\nu_{k,n}r^D_{m,k,n}(p_{k,n}
-\sum_{k\in\mathcal{K}}\mu_k p_{k,n}.
\end{align}
Similar to the case of FaD processing in~\cref{SS:FD}, if no user is assigned to SC $n$,
it is evident that the optimal power allocation for problem~\eqref{P:DFnLD} is $\bm p_n=\bm 0$, while it can be assumed without loss of generality that no RRH is selected for processing the signal, i.e. $\bm\alpha_n=\bm 0$, in this case. On the other hand, if SC $n$ is assigned to a user $\hat{k}_n\in\mathcal{K}$, then problem~\eqref{P:DFnLD} reduces to
\begin{subequations
\label{P:DFnLDFixUA}
\begin{align}
\max_
p_{\hat{k}_n,n}\geq 0
\bm\alpha_n\in\{0,1\}^
}&\,\sum_{m\in\mathcal{M}}\alpha_{m,n}\Big(\omega_{\hat{k}_n}-\frac{\lambda_m}{\bar{R}_m
\Big)r^D_{m,\hat{k}_n,n}\big(p_{\hat{k}_n,n}\big)\notag\\
&\,-\mu_{\hat{k}_n} p_{\hat{k}_n,n}\tag{\ref{P:DFnLDFixUA}}\\
\mathrm{s.t.}&\,\eqref{C:RRHSelDFnLD}.\notag
\end{align}
\end{subequations}
Problem~\eqref{P:DFnLDFixUA} is non-convex due to the integer constraints on the RRH selection $\bm\alpha_n$ and hence difficult to solve directly to find the optimal power allocation for user $\hat{k}_n$ and the optimal RRH that should locally decode this user's signal. Thus, similar to the procedure in~\cref{SS:FD}, we first assume that an RRH $\tilde{m}_n\in\mathcal{M}$ is chosen to decode user $\hat{k}_n$'s message in SC $n$ i.e., $\alpha_{\tilde{m}_n,n}=1$ and $\alpha_{m,n}=0\ \forall m\neq\tilde{m}_n$, which reduces problem~\eqref{P:DFnLDFixUA} to
\begin{subequations}
\label{P:DFnLDFUARRHSel}
\begin{align}
\max_{p_{\hat{k}_n,n}\geq 0}&\,\Big(\omega_{\hat{k}_n}
\frac{\lambda_{\tilde{m}_n}}{\bar{R}_{\tilde{m}_n}}
\Big)r^D_{\tilde{m}_n,n}\big(p_{\hat{k}_n,n}\big)-\mu_{\hat{k}_n} p_{\hat{k}_n,n}\tag{\ref{P:DFnLDFUARRHSel}}.
\end{align}
\end{subequations}
The optimal solution to the above problem~\eqref{P:DFnLDFUARRHSel} is given by the following proposition.
\begin{proposition}\label{Prop:OptPDFnLDFUARRHSel}
The optimal power allocation $\tilde{p}_{\hat{k}_n,n}$ that solves problem~\eqref{P:DFnLDFUARRHSel} is given by
\begin{align}
\tilde{p}_{\hat{k}_n,n}=\Bigg[\frac{B}{\mu N\ln 2}\Big(\omega_{\hat{k}_n}-\frac{\lambda_{\tilde{m}_n}}{\bar{R}_{\tilde{m}_n}}\Big)-\frac{\sigma_{\tilde{m}_n}^2}{|h_{\tilde{m}_n,\hat{k}_n,n}|^2}\Bigg]^+.\label{E:OptPALD}
\end{align}
\end{proposition}
\begin{proof}
The proof is similar to that for the standard water-filling power allocation for parallel AWGN channels~\cite{goldsmith2005wireless}.
\end{proof}
The optimal power allocation in~\eqref{E:OptPALD} for a fixed user assignment and RRH selection has the same form as the
classic water-filling solution~\cite{goldsmith2005wireless}. However, in general
the water levels are different for different SCs, and are determined by the user $\hat{k}_n$ to which SC $n$ is assigned to, as well as the RRH $\tilde{m}_n$ that is chosen to decode user $\hat{k}_n$'s message, through the parameters $\omega_{\hat{k}_n}$, $\lambda_{\tilde{m}_n}$, $\bar{R}_{\tilde{m}_n}$ and $h_{\tilde{m}_n,\hat{k}_n,n}$. The water-level on SC $n$ can be negative if $\lambda_{\tilde{m}_n}/\bar{R}_{\tilde{m}_n}\geq \omega_{\hat{k}_n}$, which implies that no power should be allocated to user $\hat{k}_n$ on SC $n$ in this case.
The optimal solution to the joint RRH selection and user power allocation problem~\eqref{P:DFnLDFixUA} can thus be obtained by computing $\tilde{p}_{\hat{k}_n,n}$ as given in~\eqref{E:OptPALD} for all the RRHs $\tilde{m}_n\in\mathcal{M}$, and then selecting the RRH that gives the highest objective value for problem~\eqref{P:DFnLDFixUA}. This procedure incurs a complexity of $O\left(M\right)$ and is performed for each user assignment $\hat{k}_n\in\mathcal{K}$. Then the optimal solution to problem~\eqref{P:DFnLD} for the DaF mode can be found by choosing the user that gives the highest objective value in~\eqref{P:DFnLD}.
The overall complexity of optimally solving problem~\eqref{P:DFnLD} on each SC $n$ for the DaF processing mode is thus $O(KM)$.
Finally, after solving the problems~\cref{P:DFnJD,P:DFnLD} for the FaD and DaF processing modes to obtain the maximal objective values $\bar{L}^Q_n$ and $\bar{L}^D_n$, respectively, the optimal solution to problem~\eqref{P:DualFuncn} can be found by selecting the processing mode
that maximizes the objective in~\eqref{E:Lagn}, according to~\eqref{E:OptModeDFn}.
Thus, for given dual variables $\bm\lambda$ and $\bm\mu$, problem~\eqref{P:DualFuncn} can be solved for each SC $n\in\mathcal{N}$ optimally, incurring a worst-case complexity of $O(K(2^M+M))$.
Next, we consider the dual problem corresponding to~\eqref{P:ULHybMain}, given by
\begin{align}
\min_{\bm\lambda\succeq\bm 0,\bm\mu\succeq\bm 0}g(\bm\lambda,\bm\mu).\label{E:DualProb}
\end{align}
The above dual problem~\eqref{E:DualProb} is convex and can be solved efficiently by using e.g., the ellipsoid method~\cite{boyd2014ellipsoid} to find the optimal dual variables $\bm\lambda^\star$ and $\bm\mu^\star$. Then, the optimal solution to problem~\eqref{P:DualFuncn} on each SC $n\in\mathcal{N}$ is given by $(\delta^\star_n,\bm\nu^\star_n,\bm\alpha_n^\star,\bm p_n^\star)$, computed as explained above with the optimal dual variables $\bm\lambda^\star$ and $\bm\mu^\star$. The overall algorithm for solving the joint uplink resource allocation problem~\eqref{P:ULHybMain} is thus summarized in~\cref{A:Overall}. Since the complexity of the ellipsoid method to find the optimal dual variables depends only on the size of the initial ellipsoid and the maximum length of the sub-gradients over the intial ellipsoid~\cite{boyd2014ellipsoid},
it follows from the above discussion that an asymptotically optimal solution to problem~\eqref{P:ULHybMain} for large $N$ can be efficiently computed with an overall complexity of $O(NK(2^M+M))$ using the algorithm in~\cref{A:Overall}.
\begin{table}[h]
\centering
\caption{Algorithm for problem~\eqref{P:ULHybMain}}\label{A:Overall}
\begin{framed}
\begin{algorithmic}[1]
\State Initialization: $\bm\lambda\succeq\bm 0$, $\bm\mu\succeq\bm 0$
\Repeat
\For {each SC $n\in\mathcal{N}$}
\State With $\delta_n=0$, for each user $\hat{k}_n\in\mathcal{K}$ and each RRH selection $\tilde{\bm\alpha}_n\in\{0,1\}^M$, find optimal user power allocation by solving~\eqref{P:DFnJDFUARRHSel}\label{AL:JDOptRRHPA}
\State Choose RRH selection and corresponding user power allocation that maximizes objective in~\eqref{P:DFnJDFixUA}
\State Choose user with optimal RRH selection and power allocation that gives maximum value $\bar{L}^Q_n$ of objective in~\eqref{P:DFnJD}
\State With $\delta_n=1$, for each user $\hat{k}_n\in\mathcal{K}$ and each RRH $\tilde{m}_n\in\mathcal{M}$, find optimal user power allocation using~\cref{E:OptPALD}
\State Choose RRH and corresponding user power allocation that maximizes objective in~\eqref{P:DFnLDFixUA}
\State Choose user with optimal RRH and power allocation that gives maximum value $\bar{L}^D_n$ of objective in~\eqref{P:DFnLD}
\State Find optimal mode selection $\bar{\delta}_n$ using~\eqref{E:OptModeDFn}
\EndFor
\State Update dual variables $\bm\lambda$, $\bm\mu$ using the ellipsoid method
\Until ellipsoid algorithm converges to desired accuracy
\end{algorithmic}
\end{framed}
\end{table}
\subsection{Suboptimal Solution}\label{SS:SubOptSubP}
In~\cref{SS:OptSol}, it is observed that optimally solving the problem~\eqref{P:DFnJDFixUA} for the FaD mode on each SC $n$ for a fixed user assignment requires an exhaustive search over all possible RRH selections, incurring a complexity of $O(2^M)$, which may be unsuitable for dense networks with large values of $M$. In order to reduce this complexity, we propose an alternative way to solve problem~\eqref{P:DFnJDFixUA} suboptimally, via a greedy algorithm, in this subsection.
The objective function of problem~\eqref{P:DFnJDFixUA} can be written in terms of the set of selected RRHs $\mathcal{A}_n$, as
\begin{align
f\big(\mathcal{A}_n,p_{\hat{k}_n,n}\big)&\triangleq\omega_kr_{\hat{k}_n,n}^Q\big(\mathcal{A}_n,p_{\hat{k}_n,n}\big)-\frac{2B}{N}\sum_{m\in\mathcal{A}_n}\frac{\beta_m\lambda_m}{\bar{R}_m}\notag\\
&\quad-\mu_{\hat{k}_n}p_{\hat{k}_n,n}.\label{E:ObjFixUASet}
\end{align}
Let $p_{\hat{k}_n,n}(\mathcal{A}_n)$ denote the optimal power allocation for user $\hat{k}_n$ that solves problem~\eqref{P:DFnJDFUARRHSel} when the selected set of RRHs is $\mathcal{A}_n$, which can be obtained using~\eqref{E:OptPASingleRRH} when $|\mathcal{A}_n|=1$, or by a line-search otherwise. Then, we construct a suboptimal set $\check{\mathcal{A}}_n$ and find the corresponding user power allocation $\check{p}_{\hat{k}_n,n}\triangleq p_{\hat{k}_n,n}\big(\check{\mathcal{A}_n}\big)$ for problem~\eqref{P:DFnJDFixUA} using a greedy algorithm as follows. Let $\mathcal{A}_{n,i}$ denote the set of selected RRHs on SC $n$ at the start of an iteration $i$, and $f_i\triangleq f\big(\mathcal{A}_{n,i},p_{\hat{k}_n,n}(\mathcal{A}_{n,i})\big)$ denote the corresponding maximum objective value of problem~\eqref{P:DFnJDFixUA} as given by~\eqref{E:ObjFixUASet}.
We assume that initially, $\mathcal{A}_{n,1}=\emptyset$, and
$f_1=0$.
At each iteration $i=1,\dotsc,M$, we first find an RRH $j_i\in\mathcal{M}\setminus\mathcal{A}_{n,i}$, which is not currently selected and when added to the current set of RRHs $\mathcal{A}_{n,i}$, maximizes the objective in~\eqref{E:ObjFixUASet} among all the currently un-selected RRHs, i.e.,
\begin{align}
j_i=\mathop{\arg\max}_{\ell\in\mathcal{M}\setminus\mathcal{A}_{n,i}}f\big(\mathcal{A}_{n,i}\cup\{\ell\},p_{\hat{k}_n,n}(\mathcal{A}_{n,i}\cup\{\ell\})\big).\label{E:MaxRRHji}
\end{align}
If RRH $j_i$ improves the current maximum objective value $f_{i-1}$, we add it to the current set of selected RRHs $\mathcal{A}_{n,i}$, i.e., if
\begin{align}
f\big(\mathcal{A}_{n,i}\cup\{j_i\},p_{\hat{k}_n,n}(\mathcal{A}_{n,i}\cup\{j_i\})\big)>f_i
\label{E:NewObjGreater}
\end{align}
holds, the set of selected RRHs is updated as $\mathcal{A}_{n,i+1}=\mathcal{A}_{n,i}\cup\{j_i\}$, and the current maximum objective value is updated as
\begin{align}
f_{i+1}=f\big(\mathcal{A}_{n,i}\cup\{j_i\},p_{\hat{k}_n,n}(\mathcal{A}_{n,i}\cup\{j_i\})\big).\label{E:UpdCObj}
\end{align} This procedure is continued until no RRH $j_i$ can be found which satisfies~\eqref{E:NewObjGreater}, or there is no more remaining RRH to be searched, i.e. $i=M$. If the algorithm stops at iteration $i$, the final suboptimal RRH selection and corresponding user power allocation are given by $\check{\mathcal{A}}_n=\mathcal{A}_{n,i}$ and $\check{p}_{\hat{k}_n,n}=p_{\hat{k}_n,n}(\mathcal{A}_{n,i})$. Notice that the greedy algorithm would recover the optimal solution to problem~\eqref{P:DFnJDFixUA} whenever the optimal RRH selection consists of at most two RRHs. However, in general, for given dual variables $\bm\lambda$ and $\bm\mu$, the greedy algorithm gives only a suboptimal solution $\check{\bm\alpha}_n,\check{\bm p}_n$ to problem~\eqref{P:DFnJDFixUA}. An outline of the algorithm is given in~\cref{A:GreedyRRHSel}.
\begin{table}[h]
\centering
\caption{Greedy algorithm for problem~\eqref{P:DFnJDFixUA}}\label{A:GreedyRRHSel}
\begin{framed}
\begin{algorithmic}[1]
\State Initialization: Iteration $i=1$, set of selected RRHs $\mathcal{A}_{n,1}=\emptyset$, maximum objective value $f_1=0$
\For {each $i=1,\dotsc,M$}
\State Find the RRH $j_{i}\in\mathcal{M}\setminus\mathcal{A}_{n,i}$, according to~\eqref{E:MaxRRHji}
\If {$j_i$ satisfies condition~\eqref{E:NewObjGreater}}
\State Update selected set as $\mathcal{A}_{n,i+1}=\mathcal{A}_{n,i}\cup\left\{j_{i}\right\}$
\State Update current maximum objective value $f_{i+1}$ according to~\cref{E:UpdCObj}
\Else
\State Stop and return RRH set $\check{\mathcal{A}}_n=\mathcal{A}_{n,i}$ and corresponding user power allocation $\check{p}_{\hat{k}_n,n}=p_{\hat{k}_n,n}(\mathcal{A}_{n,i})$
\EndIf
\EndFor
\end{algorithmic}
\end{framed}
\end{table}
From~\eqref{E:MaxRRHji}, it can be seen that in each iteration $i=1,\dotsc,M$, the greedy algorithm searches over the set of RRHs $\mathcal{M}\setminus\mathcal{A}_{n,i}$, which has size $M-(i-1)$ to find the RRH $j_i$ according to~\eqref{E:MaxRRHji}. Since there can be at most $M$ iterations, the worst-case complexity of the greedy algorithm is $\sum_{i=1}^{M}M-i+1=M(M+1)/2$, which
is $O(M^2)$. Since the complexity of finding the user power allocation in each iteration does not depend on $M$, the overall complexity of using the greedy algorithm to solve problem~\eqref{P:DFnJDFixUA} on each SC $n$ is $O(M^2)$, which is only quadratic in $M$, compared to the exponential complexity of the exhaustive search over all possible RRH selections. Thus, in the algorithm in~\cref{A:Overall}, if problem~\eqref{P:DFnJDFixUA} is solved using the suboptimal greedy algorithm in~\cref{A:GreedyRRHSel} instead of the optimal exhaustive search, the worst-case complexity of the algorithm in~\cref{A:Overall} is reduced to $O(NK(M^2+M))$. However, using the greedy algorithm, a convergence to the optimal dual variables $\bm\lambda^\star$ and $\bm\mu^\star$ cannot be guaranteed and the primal solution obtained for problem~\eqref{P:ULHybMain} by this method need not always be feasible. In this case, the power allocations can be made feasible by scaling each of the power constraints in~\eqref{C:PCULHybMain}. Similarly, the constraint in~\eqref{E:IFHRC} can be made feasible by considering each SC in increasing order of the rates achieved. If the DaF mode was selected on an SC, we make the user power allocation zero on this SC. Otherwise, if the FaD mode was selected with more than one RRH processing the signal in the SC, the RRHs are de-selected in increasing order of their contribution to the SNR on each SC, until~\eqref{E:IFHRC} is satisfied. Thus, a feasible solution to problem~\eqref{P:ULHybMain} can always be obtained. However, in~\cref{Sec:SimResults}, it is shown through extensive simulations that there is only a negligible difference between the performance of the greedy algorithm
and the exhaustive search for $\bm\alpha_n$'s in several practical scenarios.
Thus, in practice, the joint resource allocation problem~\eqref{P:ULHybMain} may be solved close to optimally at a worst-case complexity of $O(NK(M^2+M))$ by using the greedy algorithm, which is a large reduction compared to a complexity of $O(NK(2^M+M))$ for the optimal algorithm.
\section{Simulation Results}\label{Sec:SimResults}
For the simulation setup, we first consider a UD-CRAN cluster with $M=5$ RRHs and $K=3$ users. One RRH is located in the center of a square region with side 375 meters~(m), while the others are located on the vertices. The users are randomly located within a larger square region of side 750~m, and whose center coincides with that of the RRH square region. The fronthaul links of all the RRHs are assumed to have the same capacity $\bar{R}_m=\bar{R}~\forall m\in\mathcal{M}$, and all the RRHs use the same resolution of $\beta_m=\beta,~\forall m\in\mathcal{M}$ bits for the uniform SQ in the FaD processing mode. The wireless channel is centered at a frequency of $2$~GHz with a bandwidth $B=20$~MHz, following the Third Generation Partnership Project~(3GPP) Long Term Evolution-Advanced~(LTE-A) standard~\cite{3gpp36211}, and is divided into $N=64$ SCs using OFDMA. The combined path loss and shadowing is modeled as $38+30\log_{10}(d_{m,k})+X$ in dB~\cite{3gpp36931}, where $d_{m,k}$ is the distance between RRH $m$ and user $k$ in meters, and $X$ is the shadowing random variable, which is normally distributed with a standard deviation of $6$~dB.
The multi-path on each wireless
channel is modeled using an exponential power delay profile with $N/4$ taps
and the small-scale fading on each tap is assumed to follow the Rayleigh distribution. The AWGN is assumed to have a power spectral density of $-174$~dBm/Hz with an additional noise figure of $6$~dB at each RRH, while the total transmit power at each user is $\bar{P}_k=23~\text{dBm}\,\forall k\in\mathcal{K}$,
unless mentioned otherwise. The proposed algorithm for hybrid decoding is compared to the following two benchmark schemes:
\begin{itemize}
\item \textbf{FaD processing on all SCs}: In this case,
we solve problem~\eqref{P:ULHybMain} with $\delta_n=0\ \forall n\in\mathcal{N}$ using a similar algorithm as given in~\cref{A:Overall}.
\item \textbf{DaF processing on all SCs}: In this case,
we solve problem~\eqref{P:ULHybMain} with $\delta_n=1\ \forall n\in\mathcal{N}$.
\end{itemize}
For simplicity, we consider maximization of the sum rate in problem~\eqref{P:ULHybMain}, i.e., the user rate weights $\omega_k=1~\forall k\in\mathcal{K}$, and the values averaged over many random user locations and channels are plotted against various system parameters.
\begin{figure}[h]
\centering
\includegraphics[width=\linewidth]{Fig2_SRvbits}
\caption{Users' sum-rate vs.\ number of SQ bits for system with $M=5$, $K=3$, $B=20$~MHz, $N=64$ and $\bar{R}=250$~Mbps.}\label{F:SRvQ}
\end{figure}
\cref{F:SRvQ} plots the sum-rate against the number of bits $\beta$ used for the SQ at each RRH, when the common fronthaul capacity of the RRHs is $\bar{R}=250$~Mbps. From~\cref{F:SRvQ}, it is observed that the hybrid decoding with the proposed optimal and suboptimal algorithms outperforms both the benchmark schemes described above. Moreover, there is a negligible difference between the optimal and suboptimal hybrid decoding schemes, which implies that in practical UD-CRAN systems, the hybrid decoding gains can be achieved with the suboptimal greedy algorithm at a much lower complexity compared to the optimal algorithm. At lower values of $\beta$, the FaD scheme performs worse than the DaF scheme, since in this case the SQ is coarse, which reduces the achievable rate on each SC as given by~\eqref{E:SCRJ}, even though the fronthaul is able to support this rate. On the other hand, when $\beta$ is very high, the higher achievable rates given by~\eqref{E:SCRJ} cannot be supported by the limited fronthaul capacity of the RRHs, which again restricts the performance of the FaD scheme. However, for values of $\beta$ in between these two extremes, the FaD scheme outperforms the DaF scheme, while the proposed hybrid decoding offers the maximum advantage for all values of $\beta$.
\begin{figure}[h]
\centering
\includegraphics[width=\linewidth]{Fig3_SRvPow}
\caption{Users' sum-rate vs.\ per user total transmit power constraint for system with $M=5$, $K=3$, $B=20$~MHz, $N=64$, $\bar{R}=250$~Mbps and $\beta=10$.}\label{F:SRvP}
\end{figure}
\cref{F:SRvP} plots the sum-rate against the maximum transmit power constraint $\bar{P}$ at each user, when the common fronthaul capacity of the RRHs is $\bar{R}=250$~Mbps, and $\beta=10$. Again, from~\cref{F:SRvP}, it is observed that the proposed hybrid decoding outperforms both the benchmark schemes above, while the performance of the proposed optimal and suboptimal algorithms are similar. In addition to the proposed algorithms and the benchmark schemes listed above,~\cref{F:SRvQ,F:SRvP} also plot the dual upper bound to problem~\eqref{P:ULHybMain} given by the optimal value of the dual function $g(\bm\lambda^\star,\bm\mu^\star)$. It can be observed that the difference between the proposed optimal and suboptimal solutions and the dual upper bound is negligible even for $N=64$. Since the duality gap diminishes with $N$, this implies that the proposed algorithms are nearly optimal for practical values of $N$.
Next, we consider a large network with $M=125$ RRHs whose locations are fixed as shown in~\cref{F:MU125M120KL}, and $K=120$ users randomly located within a square region of side $2$~km. Since the users that are far away from an RRH do not contribute to its achievable rate, to reduce the complexity of our proposed algorithms, we assume that the network is divided into $25$ square clusters of $5$ RRHs each, as shown in~\cref{F:MU125M120KL}. We assume that proper frequency reuse has been assigned over adjacent clusters and thus neglect the inter-cluster interference for simplicity. Then, the algorithms in~\cref{A:Overall,A:GreedyRRHSel} can be executed in parallel in each cluster. \cref{F:SRvFHR} plots the sum-rate of the users, averaged over random channel realizations, against the common fronthaul capacity of the RRHs, with
$\beta=10$ bits.
When the fronthaul capacity is low, the DaF scheme performs better than the FaD scheme, and its performance matches that of the proposed hybrid decoding, since in this case the fronthaul cannot support the transmission of accurately quantized signals from multiple RRHs in any SC. On the other hand, when the fronthaul capacity is sufficiently large, FaD outperforms DaF, and its sum-rate
approaches that achieved by hybrid decoding. In this case, most or all the RRHs can participate in the joint decoding on each SC, which provides a joint signal processing gain. In between these two extremes, the proposed hybrid decoding offers significant gains over both the DaF and FaD schemes. In this regime, by performing DaF processing on some of the SCs, the hybrid decoding saves the fronthaul capacities of the respective RRHs, which can in turn be used for carrying quantized signals for FaD processing on other SCs, thus performing better than both the benchmarks. Thus, under practical setups with finite fronthaul capacities, even the suboptimal hybrid decoding algorithm with a low complexity of $O(NK(M^2+M))$ can offer significant throughput gains over the conventional CRAN with FaD processing at all SCs, as well as a cellular network with DaF processing at all SCs.
\begin{figure}[h]
\centering
\includegraphics[width=\linewidth]{Fig4_125M120KLayout}
\caption{Example UD-CRAN layout with $M=125$ and $K=120$.}\label{F:MU125M120KL}
\end{figure}
\begin{figure}[h]
\centering
\includegraphics[width=\linewidth]{Fig5_SRvFHR}
\caption{Users' sum-rate vs.\ per-RRH fronthaul capacity for system with $M=125$, $K=120$, $B=20$~MHz, $N=64$ and $\beta=10$.}\label{F:SRvFHR}
\end{figure}
\section{Conclusion}\label{Sec:Conc}
In this paper, we have studied the uplink transmission in an OFDMA-based UD-CRAN with hybrid decoding at the RRHs. We formulated a joint RRHs' processing mode selection, user-SC assignment, and users' power allocation problem to maximize the weighted-sum-rate of the users over all SCs subject to the given RRHs' individual fronthaul capacity constraints and the individual transmit power constraints at the users. Although the problem is non-convex, we propose two efficient solutions based on the Lagrange duality technique. Through numerical simulations, it is shown that the proposed algorithms for hybrid decoding with optimized resource allocation outperform both state-of-the-art CRAN with FaD processing and conventional cellular network with DaF processing.
\appendices
\section{Proof of~\cref{L:rknQConc}}\label{App:ProofLrknQConc}
In this proof, we drop the user and SC sub-scripts $k$ and $n$ for convenience.
Then, by direct differentiation, it can be shown that the second-order derivative of $\gamma^Q_m(p)$ in~\eqref{E:PartSNR} with respect to~(w.r.t.) $p$ satisfies $\od[2]{\gamma^Q_m(p)}{p}\leq 0,\,\forall p\geq 0$, which implies that $\gamma^Q_m(p)$ in~\eqref{E:PartSNR} is
concave for $p\geq 0$.
For a given $\bm\alpha$, since $\gamma^Q(\bm\alpha,p)$ in~\eqref{E:RxSNRJD} is the non-negative sum of the concave functions $\gamma^Q_m(p),\,m\in\mathcal{A}_n$, it follows that $\gamma^Q(\bm\alpha,p)$ is also concave in $p$~\cite{boyd2004convex}. Now, the logarithm function is concave and its extended value extension on the real line is non-decreasing. Thus, for given $\bm\alpha$, $r^Q(\bm\alpha,p)=(B/N)\log_2(1+\gamma^Q(\bm\alpha,p))$ according to~\eqref{E:SCRJ} is the composition of the concave function $\gamma^Q(\bm\alpha,p)$ with a concave and non-decreasing function, and hence $r^Q\left(\bm\alpha,p\right)$ is also concave for $p\geq 0$~\cite{boyd2004convex}, which completes the proof.
\section{Proof of~\cref{Prop:OptPASRRHQ}}\label{App:ProofOptPASRRHQ}
In this proof, we drop the user and SC sub-scripts
for convenience.
Since the objective of problem~\eqref{P:DFnJDFUARRHSel} is concave in $p$, the user power allocation that maximizes its value can be found by differentiating
the objective w.r.t. $p$ and setting the derivative equal to zero; this leads to the equation,
\begin{align
&\frac{-\mu\theta_{\tilde{m}_n}|h_{\tilde{m}_n}|^2}{\sigma_{\tilde{m}_n}^2}p^2-\frac{\mu(2\theta_{\tilde{m}_n}+1)|h_{\tilde{m}_n}|^2}{\sigma_{\tilde{m}_n}^2}p
-\mu(\theta_{\tilde{m}_n}+1)\notag\\
&+\frac{\omega B|h_{\tilde{m}_n}|^2}{\sigma_{\tilde{m}_n}^2N\ln 2}=0.\label{E:DiffPASRRHQ}
\end{align}
Further, it can be observed that the user power allocation $\tilde{p}$ that maximizes the objective of problem~\eqref{P:DFnJDFUARRHSel} is given by the right-hand root of the quadratic equation~\eqref{E:DiffPASRRHQ}, which can be expressed as
\begin{align
\tilde{p}&=\frac{(\theta_{\tilde{m}_n}+2)\sigma_{\tilde{m}_n}^2}{2|h_{\tilde{m}_n}|^2}\Bigg(\bigg(1+\frac{4}{\left(\theta_{\tilde{m}_n}+2\right)^2}\notag\\
&\quad\cdot\Big(\frac{\omega B|h_{\tilde{m}_n}|^2\theta_{\tilde{m}_n}}{\sigma_{\tilde{m}_n}^2\mu N\ln 2}-(\theta_{\tilde{m}_n}+1)\Big)\bigg)^{\frac{1}{2}}-1\Bigg).\label{E:OptPAJDProof}
\end{align}
Also, since we require $\tilde{p}\geq 0$, from~\eqref{E:OptPAJDProof} we have
\begin{align}
\frac{\omega B|h_{\tilde{m}_n}|^2\theta_{\tilde{m}_n}}{\sigma_{\tilde{m}_n}^2\mu N\ln 2}-(\theta_{\tilde{m}_n}+1)\geq 0.\label{E:CondOptPAJDProof}
\end{align}
Combining~\eqref{E:CondOptPAJDProof} with~\eqref{E:OptPAJDProof} gives the expression in~\eqref{E:OptPASingleRRH}, which completes the proof.
\bibliographystyle{IEEEtran_mod}
| 6880581624ae54db6be805e4420f6d85b197f1f3 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
When dealing with large data sets, it is often crucial to be able to extract a smaller well-diversified subset of the data. This is a classic problem in information retrieval, and appears in many natural settings.
For example, a news website often presents the reader with a small list of highlighted stories that should be as relevant as possible to the reader. However, at the same time, the shown news stories should exhibit a certain diversity.
The analogous problem appears in the context of search engines, showing a small set of hits that are at the same time relevant and diverse.
Similarly, to get a better overview of a large dataset, it is often of
interest to exhibit a small set of entries that reflect the typical
entries found in the data set. Again, a relevant and diverse subsample
is desired. Not surprisingly, diversity maximization became an
important concept in information retrieval, computational geometry and
operations research.
In this paper we focus on one classic diversity measure, namely the \emph{dispersion}, which has been studied in the operations research community~\cite{hassin1997approximation,ravi1994heuristic,birnbaum2009improved,fekete2004maximum} and is currently receiving considerable attention in the information retrieval literature~\cite{gollapudi2009axiomatic,bhattacharya2011consideration,borodin2012max}.
More formally, we are given a finite ground set $X$ of size $n$ and a symmetric nonnegative function $d:X\times X \rightarrow \mathbb{R}_{\geq 0}$ between pairs of $X$ satisfying $d(a,a)=0$ for $a\in X$. Such a function $f$ is called a \emph{distance function}, and we highlight that it does not necessarily have to be metric. The dispersion of a set $A\subseteq X$ is the sum of all pair-wise distances within $A$; in short, we denote the dispersion of $A$ by
\begin{equation*}
d(A) \coloneqq \sum_{\{a,b\}\subseteq A} d(a,b)\enspace.
\end{equation*}
Diversity maximization problems with respect to the dispersion are known under several names, in particular \emph{max-sum diversification} (in short \MSD ), \emph{max dispersion} or \emph{remote clique}. In its most basic form, the task is to maximize the dispersion under a cardinality constraint of rank $k$, i.e.
\begin{equation*}
\max\{d(A) \mid A\subseteq X, |A| \leq k\}\enspace.
\end{equation*}
For brevity, we denote this max-sum diversification problem under a cardinality constraint by $\MSD_k$. A natural generalization of $\MSD_k$ is obtained by replacing the cardinality constraint with the requirement that the set $A$ must be independent in a given matroid $M=(X,\mathcal{I})$ which, as is usual, is assumed to be given by an independence oracle.\footnote{We recall that a matroid $M=(X,\mathcal{I})$ consists of a finite ground set $X$ and a non-empty family $\mathcal{I}\subseteq 2^X$ of subsets called \emph{independent sets}, satisfying: \begin{enumerate*}[label=(\roman*)] \item if $A\in \mathcal{I}, B\subseteq A \Rightarrow B \in \mathcal{I}$, and \item if $A,B\in \mathcal{I}$ with $|A| > |B|$ $\Rightarrow \exists a\in A\setminus B$ such that $B\cup \{a\}\in \mathcal{I}$ \end{enumerate*}. For more information related to matroids we refer to~\cite{schrijver2003combinatorial}.}
Matroid constraints cover natural relevant cases, for instance the items in $X$ can be partitioned into several subgroups, and a certain number of elements have to be chosen in each subgroup.
A wide set of distance functions $d$ have been considered in this context.
A very common distance type is obtained by first representing the elements of the ground set $X$ by vectors in a high-dimensional space $\mathbb{R}^q$, and then selecting a norm in $\mathbb{R}^q$ and using the corresponding induced distances (see, e.g., \cite{manning2008introduction,salton1983introduction}).
Recently, many results have been developed for various combinations of constraints, distances, and objectives. In particular, constant-factor approximations have been obtained for $\MSD_k$ for metric distances $d$. More precisely, Ravi, Rosenkrantz and Tayi~\cite{ravi1994heuristic} showed that a natural greedy procedure is a $4$-approximation, which was later shown by Birnbaum and Goldman~\cite{birnbaum2009improved} to even achieve an approximation factor of $2$, asymptotically. Prior to this improved analysis, Hassin, Rubinstein and Tamir~\cite{hassin1997approximation} presented a different $2$-approximation. An approximation factor of $2$ is tight for this problem assuming that the \emph{planted clique} problem~\cite{alon2011inapproximability} is hard (see~\cite{borodin2012max}). Moreover, Fekete and Meijer~\cite{fekete2004maximum} showed that if the distance $d$ stems from the $\ell_1$-norm in a constant-dimensional space, then a PTAS can be obtained.\footnote{We recall that a polynomial-time approximation scheme (PTAS) is an algorithm that, for any fixed $\epsilon >0$, returns a $(1-\epsilon)$-approximation in polynomial time.}
For \MSD under a matroid constraint, Abbassi, Mirrokni and Thakur~\cite{abbassi2013diversity} and Borodin, Lee, and Ye~\cite{borodin2012max} recently obtained $\frac{1}{2}$-approximations if $d$ is metric, which were the first constant-factor approximations for this setting. The approximation factor of $\frac{1}{2}$ is as well tight here, because this setting captures $\MSD_k$ with metric distances.
Motivated by various applications, interest also arose in generalized objective functions. In particular Bhattacharya, Gollapudi and Munagala~\cite{bhattacharya2011consideration} considered an objective that consists of the dispersion plus an additive linear term, which allowed for also representing \emph{scores of documents}.
Even more generally, Borodin, Lee, and Ye~\cite{borodin2012max} studied the sum of the dispersion function with a monotone submodular function, and showed that even for this generalization a $\frac{1}{2}$-approximation can be achieved for this objective under a matroid constraint via local search.
Very recently, the authors showed~\cite{cevallos_2016_max-sum} that considerably stronger results can be achieved under a matroid constraint for many frequently used distances $d$ that have the property of being of \emph{negative type}, which is defined as follows.
Let $D\in \mathbb{R}_{\geq 0}^{n\times n}$ be the distance matrix corresponding to $d$, i.e., $D_{a,b}=d(a,b)$ for $a,b\in X$. Then, $d$ is of negative type if
\begin{equation*}
x^T D x \leq 0 \qquad \forall x\in \mathbb{R}^n \text{ with } \sum_{i=1}^n x_i=0\enspace.
\end{equation*}
Commonly used distances of negative type include the ones induced by $\ell_1$ and $\ell_2$ norms, the cosine distance or the Jaccard distance~\cite{Pekalska:2005:DRP:1197035}.
For more information on negative type distances, we refer the interested reader to~\cite{schoenberg1938metricI,schoenberg1938metricII,deza1990metric,DezaLaurent97}.
Norms corresponding to negative distance types have as well been used for similarity measures via lower-dimensional bit vectors stemming from sketching techniques~\cite{charikar2002similarity}.
For negative type distances, a PTAS for \MSD can be achieved under a matroid constraint, based on rounding solutions obtained by a convex relaxation of the problem~\cite{cevallos_2016_max-sum}.
Whereas this is essentially optimal in terms of approximation quality, the employed technique requires to solve $n$-dimensional convex optimization problems, which is impractical for large data sets and large data sets are usually to be dealt with in web-search and information retrieval. This motivates the guiding questions of this paper.
\begin{enumerate}
\item Can one profit from the additional structure of negative type distances to obtain \emph{efficient} PTASs that are suitable for \emph{large-scale} problems?
\item Can one obtain such algorithms in other, more general relevant settings, beyond matroid constraints?
\end{enumerate}
\subsection*{Our results}
Our key contribution is the analysis of conceptually simple and
classic local search techniques for \MSD with negative type distances.
These procedures are easy to implement and considerably faster than the convex optimization approach suggested in~\cite{cevallos_2016_max-sum}.
First, this gives us a strong approximation algorithm for matroid constraints, which implies a PTAS.
\begin{theorem}\label{thm:PTASmat}
There is a $(1-\frac{5}{k})$-approximation for \MSD with negative type distances subject to a matroid constraint of rank $k$, running in $O(n k^2 \log k)$ time, when counting distance evaluations and calls to the independence oracle as unit time.
\end{theorem}
The above theorem indeed implies a PTAS: This is clear if $k\geq \frac{5}{\epsilon}$, where $\epsilon>0$ is the error parameter; otherwise, feasible solutions only have constant size and one can enumerate over all possible solutions. This running time is \emph{linear} in $n$. In light of the fact that $k$ is much smaller than $n$ this running time is very desirable for large $n$.
\medskip
Furthermore, our techniques show that local search yields a PTAS for \MSD with negative type distances even for matroid intersection constraints. Whether a PTAS exists in this setting was not known before.
\begin{theorem}\label{thm:PTASmatInt}
There is a PTAS for \MSD with negative type distances subject to a matroid intersection constraint.
\end{theorem}
Our PTAS is a variation of a classic local search technique, which was used by Lee, Sviridenko, and Vondr\'ak~\cite{lee_2010_submodular} in the context of submodular maximization, which considers at each iteration exchanges of constant size. This is the first non-trivial approximation result for matroid intersection constraints in this context, and its purpose is mainly to show that very strong approximation guarantees are possible for this constraint type.
However, due to these larger-size exchanges, the suggested algorithm may have a large (polynomial) running time, depending on the size of the problem and the error parameter.
\medskip
Finally,
by combining our results with recent non-oblivious local search techniques by Filmus and Ward~\cite{filmus2014monotone}, and Sviridenko, Vondr{\'a}k and Ward~\cite{sviridenko2015optimal}, we obtain close-to-optimal approximation guarantees for \MSD with an objective function being a sum of the dispersion for negative type distances and a monotone submodular function, subject to a matroid constraint. Such objectives allow for balancing diversity and \emph{relevance}. Combinations of diversity and a linear function have been studied previously, see~\cite{bhattacharya2011consideration,borodin2012max}.
The approximation factor we obtain depends on the \emph{curvature} of the submodular function, which yields a smooth interpolation between linear functions and submodular functions. We give a formal definition of curvature in Section~\ref{sec:submodObj}.
\begin{theorem}\label{thm:msd+f}
Consider \MSD subject to a matroid constraint of rank $k$, with respect to an objective $g=d+f$, where $d$ is the dispersion for a negative type distance and $f$ is a nonnegative, monotone submodular function of curvature $c$.
Let $\lambda_d=\frac{d(\OPT)}{g(\OPT)}$ and $\lambda_f=\frac{f(\OPT)}{g(\OPT)}$, where $\OPT$ is an optimal solution to the problem of maximizing $g$ subject to the matroid constraint. Then, for any $\epsilon >0$, there is an efficient algorithm returning a solution of approximation guarantee
\begin{equation*}
1-\lambda_d \frac{4}{k}-\lambda_f \frac{c}{e} - \epsilon \geq 1-\max\left\{\frac{4}{k}, \frac{c}{e}\right\} - \epsilon \enspace.
\end{equation*}
\end{theorem}
The above result implies a PTAS for the case of $f$ being linear. This greatly improves upon the known $1/2$-approximation~\cite{bhattacharya2011consideration,borodin2012max} if the distances are of negative type. If $k$ is large enough, the result yields an approximation factor of $1-\frac{c}{e}-\epsilon$, which is known to be optimal even for the special case of maximizing a monotone submodular function with curvature $c$ over a matroid constraint~\cite{sviridenko2015optimal}.
\subsection*{Further related results and implications}
A further approach to deal with diversity maximization in large data sets is the computation of so-called \emph{core-sets}. A core-set is a small subset of the data such that an optimal solution to certain optimization problems, when applied only on the elements of the core-set, is close to the global optimal solution. Core-sets and the generalized notion of \emph{composable core-sets} have recently received considerable attention in the context of diversity maximization, and allow for transforming sequential algorithms into algorithms that run in MapReduce and Streaming models. We refer the interested reader to~\cite{indyk2014composable} and~\cite{ceccarello2016mapreduce} and references therein. We only mention the following direct consequence of the results in~\cite{ceccarello2016mapreduce} and Theorem~\ref{thm:PTASmat}.
\begin{corollary}
Consider $\MSD_k$ on distances of negative type and doubling dimension $q$, such as Euclidean or Manhattan distances in $\mathbb{R}^q$. Then, for any $\epsilon >0$, there is a single-pass streaming algorithm that achieves an approximation guarantee of $1-\frac{5}{k}-\epsilon$, in space $O(\epsilon^{-q}k^2)$.
\end{corollary}
A related geometric measure of dispersion in $\mathbb{R}^q$ is the \emph{volume} of the selected data set. An optimal constant factor approximation algorithm for this measure was recently given by Nikolov~\cite{nikolov2015randomized}.
\subsection*{Organization of the paper}
In Section~\ref{sec:matConstraints}, we present our results for \MSD with respect to negative type distances and subject to a matroid constraint. In particular, this section highlights the main strategy we employ to exploit the property of negative type distances.
Section~\ref{sec:matIntConstraints} presents a PTAS for matroid intersection constraints. Finally, Section~\ref{sec:submodObj} contains our results for generalized objective functions, being a sum of the dispersion and a nonnegative, monotone submodular function.
\section{A local-search based PTAS for matroid constraints}
\label{sec:matConstraints}
Recall that the dispersion of a set $A\in X$ is $d(A):=\sum_{\{a,a'\}\subseteq A} d(a,a')=\frac{1}{2} \sum_{a,a'\in A} d(a,a')$. And we define the following auxiliary function, which represents the total sum of distances between two sets: For sets $A,B\in X$, let
$$d(A,B):= \sum_{a\in A, b\in B} d(a,b).$$
Notice in particular that $d(A,A)=2d(A)$ for any $A\in X$.
Throughout this section, we consider a matroid $M=(X,\mathcal{I})$ of rank $k\in \mathbb{Z}_{\geq 2}$.
The algorithm we analyze in this section, highlighted below as Algorithm~\ref{alg:localSearchMat}, is a well-known canonical local search algorithm that starts with a basis $A$ and iteratively considers exchanges of a single element, always maintaining a basis. For brevity we use the shorthand $A-a+b$ for $(A\setminus \{a\})\cup \{b\}$, where $A$ is a set and $a,b$ are two elements.
\smallskip
{
\begin{algorithm}[H]
\For{$i=1 \ldots \ell$}{
\If{$\exists$ pair $(a,b)\in A\times (X\setminus A)$ such that
$A-a+b\in \mathcal{I}$ and $d(A-a+b)>d(A)$}{
\smallskip
Find such a pair $(a,b)$ maximizing $d(A-a+b)$.\\
Set $A=A-a+b$.
}
}
\textbf{return} $A$.
\smallskip
\caption{Local search for matroids, starting with a basis $A$ and running for $\ell$ iterations.}
\label{alg:localSearchMat}
\end{algorithm}
\medskip
The following lemma provides a key inequality for negative type distances that we exploit. As an intuitive special case, the inequality says that for any two sets $A$ and $B$ of same cardinality, the dispersion within $A$ plus the dispersion within $B$ is no more than the sum of all distances between $A$ and $B$.
\begin{lemma}
\label{lem:negTypeIneq}
Let $d:X\times X \rightarrow \mathbb{R}_{\geq 0}$ be a distance of negative type and $D \in \mathbb{R}_{≥0}^{X \times X}$ be the corresponding matrix, i.e., $D_{a,b}=d(a,b)$ for $a,b\in X$. For any two non-zero vectors $x,y \in \mathbb{R}^X_{≥0}$ one has
\begin{equation}
\label{eq:negTypeIneqVec}
\frac{\|y\|_1}{\|x\|_1} x^T Dx + \frac{\|x\|_1}{\|y\|_1}y^T Dy
\leq 2 x^T D y\enspace,
\end{equation}
and consequently for two non-empty sets $A,B \subseteq X$
\begin{equation}
\label{eq:negTypeIneqSet}
\frac{|B|}{|A|} d(A) + \frac{|A|}{|B|} d(B) \leq d(A,B)\enspace .
\end{equation}
\end{lemma}
\begin{proof}
Let $z=\|y\|_1 x - \|x\|_1 y$. Notice that $\sum_{a\in X}z(a)=0$. Hence, by the fact that $D$ is a distance of negative type we have
\begin{align*}
0 \geq z^T D z = \|y\|_1^2 x^T D x - 2 \|x\|_1 \|y\|_1 x^T D y + \|x\|_1^2 y^T D y.
\end{align*}
Inequality~\eqref{eq:negTypeIneqVec} now follows by dividing both sides of the above inequality by $\|x\|_1\|y\|_1$ and rearranging terms.
Inequality~\eqref{eq:negTypeIneqSet} follows from~\eqref{eq:negTypeIneqVec} by setting $x=\chi^A$ and $y=\chi^B$ to the characteristic vectors of $A$ and $B$, respectively, and dividing both the left-hand side and right-hand side by $2$.
\end{proof}
To analyze Algorithm~\ref{alg:localSearchMat}, we will consider a pairing of the elements of $A$ with the elements of an optimal solution $\OPT$ to the problem. It is well known that for any two bases $A,B$ of a matroid, there exists a pairing that can be used for exchanges in $A$. More precisely, removing any element of $A$ and adding its paired counterpart in $B$ will again lead to a basis. We denote such a pairing, which can be formalized as a bijection $\pi: A \rightarrow B$ and whose properties are stated in the lemma below, a \emph{Brualdi bijection}.
\begin{lemma}[Brualdi~\cite{brualdi1969comments}]\label{lem:brualdi}
For any two independent sets $A,B\in \mathcal{I}$ of equal cardinality, there is a bijection $\pi:A\rightarrow B$ such that for any $a\in A$, we have $A-a+\pi(b)\in \mathcal{I}$. In particular, such a bijection satisfies that it is the identity mapping on $A\cap B$.
\end{lemma}
To show that Algorithm~\ref{alg:localSearchMat} makes sufficient progress as long as $d(A)$ is much smaller than $d(\OPT)$, we consider a Brualdi bijection between $A$ and $\OPT$. As our analysis later shows, the distances between the pairs of a Brualdi bijection will be an error term that we have to bound. This is done by the following lemma.
\begin{lemma}\label{lem:cheapMatching}
For any two sets $A,B\subseteq X$ of equal cardinality $k$, and any bijection $\pi:A \rightarrow B$,
\begin{equation*}
\sum_{a\in A} d(a,\pi(a))\leq \frac{2}{k}d(A,B).
\end{equation*}
\end{lemma}
\begin{proof}
For any $a\in A$ with $a\neq \pi(a)$, Lemma~\ref{lem:negTypeIneq} implies
\begin{equation*}
d(A,a)+d(A,\pi(a))=d(A,\{a,\pi(a)\})\geq \frac{2}{k}d(A)
+\frac{k}{2}d(a,\pi(a)).
\end{equation*}
Notice that the above inequality is also true if $a=\pi(a)$, in which case the inequality reduces to $d(A,a) \geq \frac{1}{k}d(A)$, which is again a direct consequence of Lemma~\ref{lem:negTypeIneq}.
Summing these inequalities over $A$ gives
\begin{equation*}
d(A,A)+d(A,B)\geq 2d(A)+\frac{k}{2}\sum_{a\in A} d(a,\pi(a))\enspace .
\end{equation*}
The terms $d(A,A)$ and $2d(A)$ cancel out, and the claim follows.
\end{proof}
The following lemma shows that a locally optimal solution with respect to the considered exchange steps is a $(1-\frac{4}{k})$-approximation, without going into the details of how fast we will approach a local optimum.
\begin{lemma}\label{lem:matAverageImp}
Let $A$ and $B$ be bases of $M$, and let $\pi:A\rightarrow B$ be a bijection satisfying $\pi(a) = a$ for $a\in A\cap B$. Then
\begin{align}
\sum_{a\in A}\left( d(A-a+\pi(a)) - d(A)\right)
&\geq \left(1-\frac{2}{k}\right) d(B) - \left(1+\frac{2}{k}\right)d(A)\enspace,\label{eq:matAverageImpDetailed}\\
\intertext{which, if $d(B) \geq d(A)$, implies}
\sum_{a\in A}\left( d(A-a+\pi(a)) - d(A)\right)
&\geq \left(1-\frac{4}{k}\right) d(B) - d(A)\enspace.
\label{eq:matAverageImpOpt}
\end{align}
\end{lemma}
\begin{proof}
We first observe that
\begin{equation}\label{eq:expandUpdate}
d(A-a+b) = d(A) + d(A,b) - d(a,b) - d(a,A)
\qquad \forall a\in A, b\in X\setminus (A-a)\enspace.
\end{equation}
The above equation clearly holds if $a=b$, and for $a\neq b$ it can easily be checked by observing that the distance of any pair is counted the same number of times in the right-hand side and left-hand side of the equation (see Figure~\ref{fig:dAmapb}).
\begin{figure}[h]
\begin{center}
\input{dAmapb.tikz}
\end{center}
\caption{Graphical highlighting of the edges counted in the different terms of~\eqref{eq:expandUpdate}, for $a\neq b$. One can easily observe that each edge is counted the same number of times in the left-hand side and right-hand side of~\eqref{eq:expandUpdate}.}\label{fig:dAmapb}
\end{figure}
We finally obtain
\begin{align*}
\sum_{a\in A} \left( d(A-a+\pi(a)) - d(A)\right)
&= \sum_{a\in A} \left( d(A,\pi(a)) - d(a,\pi(a)) - d(a,A) \right)
&& \text{(by~\eqref{eq:expandUpdate})}\\
&= d(A,B) - d(A,A) - \sum_{a\in A}d(a,\pi(a))\\
&\geq \left(1-\frac{2}{k}\right) d(A,B) - d(A,A)
&& \text{(by Lemma~\ref{lem:cheapMatching})}\\
&\geq \left(1-\frac{2}{k}\right) (d(A) + d(B)) - 2d(A)
&& \text{(by~\eqref{eq:negTypeIneqSet})}\\
&= \left(1-\frac{2}{k}\right) d(B) - \left(1+\frac{2}{k}\right) d(A).
\end{align*}
Inequality~\eqref{eq:matAverageImpOpt}, valid for $d(B) \geq d(A)$, now easily follows by replacing in the right-hand side the term $-\frac{2}{k} d(A)$ by $-\frac{2}{k} d(B)$.
\end{proof}
The next theorem is a mostly standard argument showing exponentially fast convergence of the local search algorithm.
\begin{theorem}\label{thm:convLSMat}
Let $A$ be any basis of $M$.
Running Algorithm~\ref{alg:localSearchMat} for $\ell\in \mathbb{Z}_{\geq 0}$ iterations, a basis $A_\ell$ is returned such that
\begin{equation*}
d(A_\ell) \geq \left(
1 -
\left(1-\frac{1}{k}\right)^\ell \right)
\left(1-\frac{4}{k}\right)\cdot d(\OPT) \enspace .
\end{equation*}
\end{theorem}
\begin{proof}
Let $A_0 = A$ be the starting basis and we denote by $A_i$ for $i\in \{0,\ldots, \ell\}$ the basis obtained after $i$ iterations of the local search algorithm. Let $i\in \{0,\ldots, \ell-1\}$, and we consider the improvement from $A_i$ to $A_{i+1}$. Let $\pi: A_i \rightarrow \OPT$ be a Brualdi bijection. By inequality~\eqref{eq:matAverageImpOpt} in Lemma~\ref{lem:matAverageImp}, the average improvement in the dispersion of $A_i$ over the exchanges corresponding to $\pi$ is at least
\begin{equation*}
\frac{1}{k} \sum_{a\in A_i}\left(d(A_i-a+\pi(a)) - d(A_i) \right)
\geq \frac{1}{k}\left(\left(1-\frac{4}{k}\right) d(\OPT) - d(A_i)\right)
\enspace.
\end{equation*}
Since our local search algorithm does an exchange that maximizes $d(A_{i+1}) - d(A_i)$, we have
\begin{equation*}
d(A_{i+1}) - d(A_i)
\geq \frac{1}{k}\left(\left(1-\frac{4}{k}\right) d(\OPT) - d(A_i)\right)
\enspace ,
\end{equation*}
which, by regrouping terms, leads to
\begin{equation*}
\left(1 - \frac{4}{k}\right)d(\OPT) - d(A_{i+1})
\leq \left(1-\frac{1}{k}\right) \left(
\left(1-\frac{4}{k}\right) d(\OPT) - d(A_i)
\right)
\quad \forall i\in \{0,\ldots, \ell-1\}
\enspace .
\end{equation*}
The above family of inequalities implies
\begin{align*}
\left(1-\frac{4}{k}\right) d(\OPT) - d(A_{\ell})
&\leq
\left(1 - \frac{1}{k}\right)^\ell \left(
\left(1-\frac{4}{k}\right) d(\OPT) - d(A_0)
\right)\\
&\leq \left(1-\frac{1}{k}\right)^\ell \left(1-\frac{4}{k}\right) d(\OPT),
\end{align*}
which shows the theorem.
\end{proof}
Putting all ingredients together, we get our main result for \MSD with negative type distances and subject to a matroid constraint, which shows Theorem~\ref{thm:PTASmat}.
\begin{theorem}
It suffices to run Algorithm~\ref{alg:localSearchMat} for $O(k \log k)$ iterations starting with an arbitrary basis to obtain a $(1-\frac{5}{k})$-approximation for \MSD with respect to a negative type distance and subject to a matroid constraint of rank $k$.
Moreover, this algorithm can be implemented to run in $O(n k^2 \log k)$ time, when counting distance evaluations and calls to the independence oracle as unit time.
\end{theorem}
\begin{proof}
By Theorem~\ref{thm:convLSMat}, to obtain a $(1-\frac{5}{k})$-approximation, it suffices to choose the number of iterations $\ell$ of Algorithm~\ref{alg:localSearchMat} such that $(1-\frac{1}{k})^\ell \leq \frac{1}{k}$, which can be achieved by setting $\ell= O(k \log k)$.
To complete the proof, it remains to show that each iteration can be performed in $O(nk)$ time. One way to implement our local search algorithm to get the $O(nk)$ running time per iteration is as follows. At the beginning of each iteration, with current set $A$, we compute $d(A)$, and also $d(A,a)$ for each $a\in A$. For this we need $O(k^2)$ distance evaluations; we recall that $O(k^2)=O(n k)$ because $k\leq n$. We then go through the elements $b\in X\setminus A$, and for each $b\in X\setminus A$, we consider all sets $A-a+b$ for $a\in A, a \neq b$. Since $|X\setminus A| = O(n)$, it suffices to show that for a fixed $b\in X\setminus A$, we can compute the best exchange step involving $b$ in $O(k)$ time. For a fixed $b$, we first compute $d(A,b)$ in $O(k)$ time. Then, for each $a\in A,a\neq b$ we first call the independence oracle to determine whether $A-a+b\in \mathcal{I}$, and if so we compute $d(A-a+b)$ using~\eqref{eq:expandUpdate}, i.e.,
\begin{equation*}
d(A-a+b) = d(A) + d(A,b) - d(a,b) - d(a,A)\enspace .
\end{equation*}
The only quantity on the right-hand side that we did not compute so far is $d(a,b)$, which is obtained by a single distance evaluation. Hence, computing $d(A-a+b)$ for all $a\in A,a\neq b$ for which $A-a+b\in \mathcal{I}$, takes $O(k)$ time as desired, when counting distance evaluations and calls to the independence oracle as unit time.
\end{proof}
\section{A PTAS for matroid intersection}
\label{sec:matIntConstraints}
Throughout this section, $M_1=(X,\mathcal{I}_1)$ and $M_2=(X,\mathcal{I}_2)$ are two matroids on the same ground set $X$, and $k$ is the maximum cardinality of a common independent set.
The algorithm we analyze is a natural local search algorithm considering exchanges up to a certain size. It is almost identical to a procedure suggested by Lee, Sviridenko and Vondr\'{a}k~\cite{lee_2010_submodular}, designed for maximizing a monotone submodular function subject to multiple matroid constraints. The only difference is that after each exchange step, we augment the current set $A$ to a (inclusion-wise) maximal set in $\mathcal{I}_1 \cap \mathcal{I}_2$, i.e., we replace $A$ by a maximal set $A'\in \mathcal{I}_1\cap \mathcal{I}_2$ that contains $A$. It is well-known that any maximal common independent set has cardinality at least $\frac{k}{2}$. This is a property we exploit in our analysis.
For better readability, we did not try to optimize constants and often use rather loose bounds for simplicity.
\medskip
{
\begin{algorithm}[H]
\For{$i=1 \ldots \ell$}{
\If{$\exists$ pair $(S,T)\in 2^A\times 2^{X\setminus A}$ with
\begin{enumerate}[label=(\roman*), nosep, leftmargin=3em]
\item $|S|\leq p, |T|\leq p-1$,
\item $(A\setminus S) \cup T \in \mathcal{I}_1\cap \mathcal{I}_2$, and
\item $d((A\setminus S)\cup T)>d(A)$,
\end{enumerate}
}
{
\smallskip
Find such a pair $(S,T)$ maximizing $d((A\setminus S)\cup T)$.\\
\smallskip
Set $A=(A\setminus S) \cup T$.\\
\smallskip
Augment $A$ to a maximal element in $\mathcal{I}_1\cap \mathcal{I}_2$.
}
}
\textbf{return} $A$.
\smallskip
\caption{Local search for matroid intersection with exchange parameter $p\in \mathbb{Z}_{\geq 2}$, starting with $A\in \mathcal{I}_1 \cap \mathcal{I}_2$ and running for $\ell$ iterations.}
\label{alg:localSearchMatInt}
\end{algorithm}
\medskip
Our analysis heavily relies on recently developed exchange properties for matroid intersection. The following lemma was shown in~\cite{chekuri_2011_multibudgeted} building up on previous work~\cite{lee_2010_submodular,chekuri_2010_dependent}. We state a slightly simplified version of the lemma. The original lemma provided further properties on the sets $P_i$ and also guaranteed that such sets can be found efficiently. These are properties we do not need in our analysis. The set operator $\Delta$ denotes the symmetric difference, i.e., $A\Delta B = (A\setminus B) \cup (B \setminus A)$.
\begin{lemma}[{\cite[Lemma 3.3]{chekuri_2011_multibudgeted}}]
\label{lem:matIntExchange}
Let $p\in \mathbb{Z}_{\geq 2}$, and let $A, B \in \mathcal{I}_1\cap \mathcal{I}_2$ with $|A|=|B|$. Then there exists a family of nonempty sets $P_1, \dots, P_m\subseteq X$ with $|P_i\cap A|\leq p$ and $|P_i\cap B|\leq p-1$ for $i\in [m]:=\{1,\ldots m\}$, and coefficients $\lambda_i > 0$ such that
\begin{enumerate}[label=(\roman*),itemsep=0em]
\item $A\Delta P_i \in \mathcal{I}_1\cap \mathcal{I}_2$ \;$\forall i \in [m]$, and
\item $\sum_{i=1}^m \lambda_i \chi^{P_i} = \frac{p}{p-1} \chi^{A\setminus B} + \chi^{B\setminus A}$.
\end{enumerate}
\end{lemma}
It is not hard to see that the requirement $|A|=|B|$ in Lemma~\ref{lem:matIntExchange} is not necessary. This condition was important in the original (stronger) lemma presented in~\cite{chekuri_2011_multibudgeted} which contained further properties that relied on it. For completeness, we state the lemma without the equal cardinality requirement below and provide a short proof for it.
\begin{lemma}\label{lem:matIntExchangeGen}
Let $p\in \mathbb{Z}_{\geq 2}$, and let $A, B \in \mathcal{I}_1\cap \mathcal{I}_2$. Then there exists a family of nonempty sets $P_1, \dots, P_m\subseteq X$ with $|P_i\cap A|\leq p$ and $|P_i\cap B|\leq p-1$ for $i\in [m]$, and coefficients $\lambda_i > 0$ such that
\begin{enumerate}[label=(\roman*),itemsep=0em]
\item $A\Delta P_i \in \mathcal{I}_1\cap \mathcal{I}_2$ \;$\forall i \in [m]$, and
\item\label{item:exchangeLoad} $\sum_{i=1}^m \lambda_i \chi^{P_i} = \frac{p}{p-1} \chi^{A\setminus B} + \chi^{B\setminus A}$.
\end{enumerate}
\end{lemma}
\begin{proof}
To prove the lemma, we will first ``lift'' the sets $A,B\in \mathcal{I}_1\cap \mathcal{I}_2$, which may have different sizes, to larger sets $A',B'$ of the same size that are common independent sets of two auxiliary matroids $M_1'$ and $M_2'$. We can then apply Lemma~\ref{lem:matIntExchange} to $A',B'$ with respect to these auxiliary matroids, which will imply Lemma~\ref{lem:matIntExchangeGen}.
Let $k$ be the maximum cardinality of a common independent set in $M_1$ and $M_2$, i.e., $k=\max\{|I| \mid I\in \mathcal{I}_1\cap \mathcal{I}_2\}$. Let $\bar{X}$ be a finite set of size $k$ that is disjoint from $X$. We define two auxiliary matroids $M_1'=(X',\mathcal{I}_1')$ and $M_2'=(X',\mathcal{I}_2')$ on ground set $X' = X\cup \bar{X}$. To this end, let $\bar{M}=(\bar{X},\bar{\mathcal{I}})$ be the free matroid over $\bar{X}$, i.e., $\bar{\mathcal{I}} = 2^{\bar{X}}$. For $j\in \{1,2\}$, the matroid $M'_j$ is defined to be the disjoint union of $M_j$ and $\bar{M}$. More formally, for $j\in \{1,2\}$ we have
\begin{equation*}
\mathcal{I}_j' = \{ I_j \cup \bar{I} \mid I_j \in \mathcal{I}_j, \bar{I}\in \bar{\mathcal{I}}\}.
\end{equation*}
The thus defined $M_1'$ and $M_2'$ are indeed matroids because the union of matroids again leads to a matroid (see, e.g.,~\cite[volume B]{schrijver2003combinatorial}).
Given $A,B\in \mathcal{I}_1\cap \mathcal{I}_2$, let $A'=A\cup \bar{S}_A$ and $B'=B\cup \bar{S}_B$, where $\bar{S}_A$ and $\bar{S}_B$ are arbitrary subsets of $\bar{X}$ of size $k-|A|$ and $k-|B|$, respectively. Clearly, we have $A',B'\in \mathcal{I}_1'\cap \mathcal{I}_2'$, and $|A'|=|B'|=k$. We can thus apply Lemma~\ref{lem:matIntExchange} to $A'$ and $B'$ with respect to the matroids $M_1'$ and $M_2'$ to obtain a family $P_1',\dots P_m'\subseteq X'$ with coefficients $\lambda_i > 0$ for $i\in [m]$ satisfying the properties stated in Lemma~\ref{lem:matIntExchange}. It remains to observe that the family defined by $P_i = P'_i \cap X$ for $i\in [m]$ (after removing empty sets), satisfies the properties of Lemma~\ref{lem:matIntExchangeGen}. This immediately follows from the definitions of $M_1'$ and $M_2'$ which imply that for any $S'\in \mathcal{I}'_1\cap \mathcal{I}'_2$, we have $S'\cap X \in \mathcal{I}_1\cap \mathcal{I}_2$.
\end{proof}
As in the previous section, the dispersion within the exchange sets we consider will appear as an error term in our analysis. The following lemma bounds this quantity.
\begin{lemma}\label{lem:boundWithinPi}
Let $p\in \mathbb{Z}_{\geq 2}$, and let $A,B\in \mathcal{I}_1\cap \mathcal{I}_2$. Moreover, let $P_i\subseteq X$ for $i\in [m]$ and $\lambda_i > 0$ for $i\in [m]$ be a family of sets and coefficients satisfying the conditions of Lemma~\ref{lem:matIntExchangeGen}. Then
\begin{equation*}
\frac{|A|}{2p-1}\sum_{i=1}^m \lambda_i \cdot d(P_i)
\leq \frac{p}{p-1} d(A,A) + d(A,B)\enspace.
\end{equation*}
\end{lemma}
\begin{proof}
For $i\in [m]$ we have
\begin{align*}
d(A,P_i \cap A) + d(A, P_i \cap B) &= d(A, P_i)
\geq \frac{|P_i|}{|A|} d(A) + \frac{|A|}{|P_i|} d(P_i),
\end{align*}
where the inequality follows from~\eqref{eq:negTypeIneqSet}.
Multiplying both the left-hand side and right-hand side by $\lambda_i$, and summing the resulting inequality over all $i\in [m]$, we get the following inequality, which follows from property~\ref{item:exchangeLoad} of Lemma~\ref{lem:matIntExchangeGen}:
\begin{align*}
\frac{p}{p-1} d(A,A\setminus B) + d(A, B\setminus A)
&\geq \frac{\frac{p}{p-1} |A\setminus B| + |B\setminus A|}{|A|} d(A)
+ \frac{|A|}{2p-1} \sum_{i=1}^m \lambda_i \cdot d(P_i) \\
&\geq
\frac{|A|}{2p-1} \sum_{i=1}^m \lambda_i \cdot d(P_i).
\end{align*}
The desired statement now follows because $d(A,A) \geq d(A,A\setminus B)$ and $d(A, B) \geq d(A,B\setminus A)$.
\end{proof}
The following lemma shows that the local search algorithm will go towards a set $A$ such that $\sqrt{d(A)}$ approaches
$(1-\frac{2}{p-1}-\frac{24p}{k})\sqrt{d(\OPT)}$, and hence $d(A)$ approaches $(1-\frac{2}{p-1}-\frac{24p}{k})^2 d(\OPT)$.
Notice that for large enough $k$, one can choose a $p$ such that this approximation factor gets arbitrarily small.
\begin{lemma}\label{lem:matIntAverageImp}
Let $p\in \mathbb{Z}_{\geq 2}$ with $8p \leq k$, where $k$ is the cardinality of a maximum cardinality set in $\mathcal{I}_1\cap \mathcal{I}_2$. Let $A\in \mathcal{I}_1\cap \mathcal{I}_2$ with $|A|\geq \sfrac{k}{2}$, and let $P_i\subseteq X$ for $i\in [m]$ and $\lambda_i > 0$ for $i\in [m]$ be a family of sets and coefficients satisfying the conditions of Lemma~\ref{lem:matIntExchangeGen} for the sets $A$ and $\OPT$. Then
\begin{align*}
\frac{1}{\sum_{i=1}^m \lambda_i}&\sum_{i=1}^m \lambda_i \cdot
\left(d(A \Delta P_i) - d(A)\right)\\
&\geq
\begin{cases}
\frac{1}{3k}\sqrt{d(A)} \left( \sqrt{d(\OPT)} \cdot (1-\frac{2}{p-1}-\frac{24p}{k}) - \sqrt{d(A)}\right) &\text{if } d(A) > \frac{1}{50}d(\OPT),\\
\frac{1}{24k} d(\OPT) &\text{if } d(A) \leq \frac{1}{50}d(\OPT)\enspace .
\end{cases}
\end{align*}
\end{lemma}
\begin{proof}
For brevity, let $S_i=P_i\cap A$ and $T_i = P_i \cap \OPT$ for $i\in [m]$. We have for $i\in [m]$,
\begin{align*}
d(A\Delta P_i) &= d((A\setminus S_i) \cup T_i) = d(A\setminus S_i) + d(T_i) + d(A\setminus S_i, T_i)\\
&= d(A) + d(S_i) - d(A,S_i) + d(T_i) + d(A,T_i) - d(S_i,T_i)\\
&\geq d(A) - d(A,S_i) + d(A,T_i) - d(S_i, T_i)\enspace.
\end{align*}
Let $\alpha = \frac{2p-1}{|A|}$ and $\lambda= \sum_{i=1}^m \lambda_i$.
Multiplying the above inequality by $\lambda_i$ and summing it over all $i\in [m]$, we obtain the inequality below, which follows from property~\ref{item:exchangeLoad} of Lemma~\ref{lem:matIntExchangeGen}:
\begin{equation}\label{eq:preBoundSymDiff}
\begin{aligned}
\sum_{i=1}^m \lambda_i \cdot d(A\Delta P_i)
&\geq \lambda\cdot d(A) - \frac{p}{p-1} d(A,A\setminus \OPT)
+ d(A, \OPT\setminus A) - \sum_{i=1}^m \lambda_i \cdot d(S_i, T_i)\\
&\geq \lambda\cdot d(A) - \frac{p}{p-1} d(A,A) + d(A,\OPT) - \sum_{i=1}^m \lambda_i \cdot d(S_i,T_i)\\
&\geq \lambda\cdot d(A) - \frac{p}{p-1} d(A,A) + d(A,\OPT) - \sum_{i=1}^m \lambda_i \cdot d(P_i)\\
&\geq \lambda\cdot d(A) - \frac{p}{p-1} (1+\alpha) d(A,A)
+ (1-\alpha) d(A,\OPT)\\
&\geq \lambda\cdot d(A) - \frac{2 p}{p-1} (1+\alpha) d(A)
+ (1-\alpha) \left(\frac{|\OPT|}{|A|} d(A) + \frac{|A|}{|\OPT|} d(\OPT) \right),
\end{aligned}
\end{equation}
where the second-to-last inequality follows by Lemma~\ref{lem:boundWithinPi}, and the last one by~\eqref{eq:negTypeIneqSet}.
We continue the expansion of inequality~\eqref{eq:preBoundSymDiff} in two different ways, depending on whether we are in the case $d(A) \leq \frac{1}{50}d(\OPT)$ or $d(A) > \frac{1}{50}d(\OPT)$.
\medskip
\noindent\textbf{Case $d(A) \leq \frac{1}{50} d(\OPT)$:}
To lower bound $(\sfrac{|\OPT|}{|A|}) d(A) + (\sfrac{|A|}{|\OPT|}) d(\OPT)$, consider the coefficients in front of $d(A)$ and $d(\OPT)$. For brevity, let $q=\sfrac{|\OPT|}{|A|}$.
Notice that $q + q^{-1} \geq 2$, no matter how large $|A|$ and $|\OPT|$ are, assuming they are both at least one.
Moreover, since $\OPT$ and $A$ are maximal sets in $\mathcal{I}_1\cap \mathcal{I}_2$, we have $\frac{k}{2} \leq |A|,|\OPT| \leq k$, which implies $q,q^{-1} \geq \frac{1}{2}$. Now consider the expression $q_1 \cdot d(A) + q_2 \cdot d(\OPT)$ under the constraints $q_1+q_2 \geq 2$ and $q_2\geq \sfrac{1}{2}$. This expression is minimized for $q_1=\sfrac{3}{2}$ and $q_2=\sfrac{1}{2}$ because $d(\OPT) \geq d(A)$, and hence
\begin{equation*}
\frac{|\OPT|}{|A|} d(A) + \frac{|A|}{|\OPT|} d(\OPT) \geq \frac{3}{2} d(A) + \frac{1}{2} d(\OPT)\enspace.
\end{equation*}
Using the above inequality, we can now further expand~\eqref{eq:preBoundSymDiff} as shown below. Moreover, we use $\alpha = \frac{2p-1}{|A|} \leq \frac{1}{2}$, which follows from $|A|\geq \frac{k}{2}$ and $8p\leq k$.
\begin{align*}
\sum_{i=1}^{m} \lambda_i \left(d(A\Delta P_i) - d(A) \right)
&\geq -\frac{2p}{p-1}(1+\alpha) d(A) + (1-\alpha) \left(\frac{3}{2}d(A) + \frac{1}{2} d(\OPT) \right)\\
&\geq -\frac{3p}{p-1} d(A) + \frac{3}{4} d(A) + \frac{1}{4} d(\OPT)
&& \text{($\alpha \geq \sfrac{1}{2}$)}\\
&\geq -6 d(A) + \frac{3}{4} d(A) + \frac{1}{4}d(\OPT)
&& \text{($p\geq 2$)}\\
&= -\frac{21}{4} d(A) + \frac{1}{4} d(\OPT)\\
&\geq \frac{1}{8} d(\OPT)\enspace.
&&\hspace*{-2cm}\text{(using $d(A) \leq \sfrac{1}{50}\cdot d(\OPT)$)}
\end{align*}
The result for this case now follows by dividing both sides by $\lambda$ and observing that $\lambda \leq 3k$, which holds due to
\begin{align*}
\lambda &\leq \left\|\sum_{i=1}^m \lambda_i \chi^{P_i} \right\|_1
&& \text{(using $P_i\neq\emptyset$ for $i\in [m]$)}\\
&= \left\|\frac{p}{p-1}\chi^{A\setminus B} + \chi^{B\setminus A} \right\|_1
&& \text{(by point~\ref{item:exchangeLoad} of Lemma~\ref{lem:matIntExchangeGen})}\\
&\leq \frac{p}{p-1} k + k \leq 3k.
\end{align*}
\medskip
\noindent\textbf{Case $d(A) > \frac{1}{50} d(\OPT)$:}
The minimum of the function $f(q) = q\cdot d(A) + \sfrac{1}{q}\cdot d(\OPT)$ for $q >0$ is achieved at $q=\sqrt{\sfrac{d(B)}{d(A)}}$, and leads to a value of $2 \sqrt{d(A) d(\OPT)}$. Hence
\begin{equation*}
\frac{|\OPT|}{|A|} d(A) + \frac{|A|}{|\OPT|} d(\OPT) \leq 2 \sqrt{d(A) d(\OPT)}.
\end{equation*}
Using the above inequality to further expand~\eqref{eq:preBoundSymDiff} we obtain
\begin{equation}\label{eq:preBoundSymDiff2}
\begin{aligned}
\sum_{i=1}^m \lambda_i (d(A\Delta P_i) - d(A))
&\geq - \frac{2p}{p-1} (1+\alpha) d(A)
+ 2 (1-\alpha) \sqrt{d(A) d(\OPT)}\\
&= \sqrt{d(A)} \left( \left(1-\frac{2p}{p-1}(1+\alpha)\right)\sqrt{d(A)} + 2(1-\alpha) \sqrt{d(\OPT)} - \sqrt{d(A)} \right)\\
&\geq \sqrt{d(A)} \left( \left(1-\frac{2p}{p-1}(1+\alpha) + 2(1-\alpha) \right)\sqrt{d(\OPT)} - \sqrt{d(A)} \right),
\end{aligned}
\end{equation}
where we used $d(A) \leq d(\OPT)$ and $1-\frac{2p}{p-1}(1+\alpha) < 0$ for the last inequality.
It remains to simplify the coefficient of $\sqrt{d(\OPT)}$ within the parentheses.
\begin{align*}
-\frac{2p}{p-1}(1+\alpha) + 2(1-\alpha) &= -\frac{2}{p-1}(1+\alpha) - 4\alpha\\
&= -\frac{2}{p-1} - \frac{2p-1}{|A|} \cdot \left(\frac{2}{p-1} + 4\right)
&& \text{(using $\alpha = \sfrac{(2p-1)}{|A|}$)} \\
&\geq -\frac{2}{p-1} -\frac{2p}{|A|} \cdot \frac{4p-2}{p-1}\\
&\geq -\frac{2}{p-1} - \frac{12p}{|A|}
&& \text{($\sfrac{(4p-2)}{(p-1)}\leq 6$ for $p\geq 2$)}\\
&\geq -\frac{2}{p-1} - \frac{24p}{k}\enspace.
&& \text{(using $|A|\geq \sfrac{k}{2}$)}
\end{align*}
The result for this case now follows by using the above inequality in~\eqref{eq:preBoundSymDiff2}, dividing both sides by $\lambda$, which satisfies $\lambda\leq 3k$ as shown in the first case.
\end{proof}
The next theorem shows that convergence happens fast. This readily follows by Lemma~\ref{lem:matIntAverageImp}, because as long as $d(A)$ is small, the second case of Lemma~\ref{lem:matIntAverageImp} guarantees an improvement of at least $(\sfrac{1}{24k})d(\OPT)$. Hence, this case can only happen during $O(k)$ iterations. Later, the first case of Lemma~\ref{lem:matIntAverageImp} applies, which guarantees a very fast convergence towards $(1-\frac{2}{p-1}-\frac{24p}{k})^2 d(\OPT)$.
\begin{theorem}\label{thm:convLocalSearchMatInt}
Let $p\in \mathbb{Z}_{\geq 2}$ such that $8p\leq k$, where $k$ is the cardinality of a maximum cardinality set in $\mathcal{I}_1 \cap \mathcal{I}_2$, and $1 - \frac{2}{p-1} - \frac{24p}{k} > 0$.
Let $A\in \mathcal{I}_1\cap \mathcal{I}_2$ with $|A|\geq \sfrac{k}{2}$. Then, letting Algorithm~\ref{alg:localSearchMatInt} run for $\ell \geq k$ iterations with starting set $A$ and parameter $p$, leads to a set $A_\ell \in \mathcal{I}_1\cap \mathcal{I}_2$ satisfying
\begin{equation*}
d(A_{\ell}) \geq
\left(
\left(1-\frac{2}{p-1} - \frac{24p}{k}\right)^2
- 2 \left(1 - \frac{1}{60 k} \right)^{\ell-k} \right)
d(\OPT)\enspace.
\end{equation*}
\end{theorem}
\begin{proof}
For brevity, let $\beta = 1-\frac{2}{p-1} - \frac{24p}{k}$.
Let $A_0 = A$ be the set at the beginning of the algorithm, and for $j\in [\ell]$ we denote by $A_j\in \mathcal{I}_1\cap \mathcal{I}_2$ the common independent set after $j$ iterations of Algorithm~\ref{alg:localSearchMatInt}.
Lemma~\ref{lem:matIntAverageImp} shows a lower bound on the average improvement induced by the exchanges $A\Delta P_i$. Since $A_{j+1}$ corresponds to the best exchange for parameter $p$, the difference $d(A_{j+1}) - d(A_j)$ is at least as high as this lower bound.
Thus, as long as $d(A_j)\leq \frac{1}{50}d(\OPT)$, the second case of Lemma~\ref{lem:matIntAverageImp} implies that we have $d(A_{j+1})) \geq d(A_j) + \frac{1}{24k} d(\OPT)$.
Hence after at most $k$ steps, we have a set of value strictly more than~$\frac{1}{50} d(\OPT)$, i.e.,
\begin{equation}\label{eq:lateAjsLarge}
d(A_j) > \frac{1}{50} d(\OPT) \qquad \forall j\in \{k,\ldots \ell\}\enspace.
\end{equation}
hus, for the iterations $k+1, k+2 ,\ldots, \ell$, the first case of Lemma~\ref{lem:matIntAverageImp} applies, which implies for $j\in \{k,\ldots, \ell-1\}$:
\begin{align*}
d(A_{j+1}) - d(A_j) &\geq \frac{1}{3k} \sqrt{d(A_j)}
\left( \beta \sqrt{d(\OPT)} - \sqrt{d(A_j)} \right)\\
&\geq \frac{1}{30 k} \sqrt{d(\OPT)}
\left( \beta \sqrt{d(\OPT)} - \sqrt{d(A_j)} \right),
\end{align*}
where the second inequality follows from~\eqref{eq:lateAjsLarge}.
Dividing both sides by $\sqrt{d(A_{j+1})} + \sqrt{d(A_j)}$ we get
\begin{align*}
\sqrt{d(A_{j+1})} - \sqrt{d(A_j)} &\geq
\frac{1}{30k} \frac{\sqrt{d(\OPT)}}{\sqrt{d(A_{j+1})} + \sqrt{d(A_j)}}
\left( \beta \sqrt{d(\OPT)} - \sqrt{d(A_j)} \right)\\
&\geq \frac{1}{60k}
\left( \beta \sqrt{d(\OPT)} - \sqrt{d(A_j)} \right)\enspace,
\end{align*}
where the second inequality follows from $d(A_{j+1}), d(A_j)\leq d(\OPT)$.
The above inequality can be rewritten as follows
\begin{align*}
\beta \sqrt{d(\OPT)} - \sqrt{d(A_{j+1})}
&\leq \left( 1- \frac{1}{60k} \right)
\left(\beta \sqrt{d(\OPT)} - \sqrt{d(A_j)} \right)
\qquad \forall j\in \{k,\ldots, \ell-1\}\enspace.
\end{align*}
Hence,
\begin{align*}
\beta \sqrt{d(\OPT)} - \sqrt{d(A_{\ell})}
&\leq \left( 1- \frac{1}{60k} \right)^{\ell-k}
\left(\beta \sqrt{d(\OPT)} - \sqrt{d(A_k)} \right)\\
&\leq \left( 1- \frac{1}{60k} \right)^{\ell-k}
\beta \sqrt{d(\OPT)}\enspace.
\end{align*}
Multiplying both sides with $\beta\sqrt{d(\OPT)} + \sqrt{d(A_\ell)}$ we get
\begin{align*}
\beta^2 d(\OPT) - d(A_{\ell})
&\leq \left( 1- \frac{1}{60k} \right)^{\ell-k}
\beta \sqrt{d(\OPT)} \left(\beta \sqrt{d(\OPT)} + \sqrt{d(A_\ell)}\right)\\
&\leq \left( 1- \frac{1}{60k} \right)^{\ell-k} d(\OPT) \beta(\beta+1)\\
&\leq 2 \left( 1- \frac{1}{60k} \right)^{\ell-k} d(\OPT)\enspace ,
&& \text{(using $\beta\leq 1$)}
\end{align*}
which implies the desired result.
\end{proof}
To obtain a PTAS, and therefore prove Theorem~\ref{thm:PTASmatInt}, it remains to choose the right value for $p$ in Theorem~\ref{thm:convLocalSearchMatInt} if $k$ is large enough. For small $k$, one can simply enumerate over all (polynomially many) feasible solutions.
This leads to Theorem~\ref{thm:PTASmatInt}, which we repeat below for completeness.
\begin{reptheorem}{thm:PTASmatInt}
There is a PTAS for \MSD with negative type distances subject to a matroid intersection constraint.
\end{reptheorem}
\begin{proof}
Let $M_j=(X,\mathcal{I}_j)$ for $j\in \{1,2\}$ be the two matroids imposing the matroid intersection constraint to the \MSD problem.
As usual, let $k$ be the cardinality of a maximum cardinality set in $\mathcal{I}_1\cap \mathcal{I}_2$.
Let $\epsilon > 0$ be our error tolerance, i.e., we want to obtain a $(1-\epsilon)$-approximation.
If $k < \frac{144}{\epsilon}
\left( \lceil \frac{12}{\epsilon} \rceil +1 \right)$, then we can enumerate over all sets in $\mathcal{I}_1\cap \mathcal{I}_2$, since there are only polynomially many such sets.
In what follows, we therefore assume $k \geq \frac{144}{\epsilon} \left( \lceil \frac{12}{\epsilon} \rceil +1 \right)$. We set $p=\lceil\frac{12}{\epsilon}\rceil + 1$ and let $A$ be any set in $\mathcal{I}_1\cap \mathcal{I}_2$ of cardinality $|A|\geq \sfrac{k}{2}$. Now we let Algorithm~\ref{alg:localSearchMatInt} run with starting set $A$, parameter $p$, and for a number $\ell \geq k$ of iterations satisfying
\begin{equation}\label{eq:downToE3}
2 \left( 1- \frac{1}{60 k} \right)^{\ell - k} \leq \frac{\epsilon}{3}\enspace.
\end{equation}
Notice that the above inequality is satisfied for a value $\ell=O(k \log \sfrac{1}{\epsilon})$, which is polynomially bounded. Since moreover $p = O(1)$, Algorithm~\ref{alg:localSearchMatInt} runs in polynomial time returning some set $A_\ell \in \mathcal{I}_1\cap \mathcal{I}_2$.
By Theorem~\ref{thm:convLocalSearchMatInt}, the dispersion of $A_\ell$ satisfies
\begin{align*}
d(A_{\ell}) &\geq \left(
\left( 1- \frac{2}{p-1} - \frac{24p}{k} \right)^2
- 2 \left( 1- \frac{1}{60k} \right)^{\ell - k} \right)
d(\OPT)\\
&\geq \left(
\left( 1- \frac{2}{p-1} - \frac{24p}{k} \right)^2
-\frac{\epsilon}{3} \right)
d(\OPT)
&& \text{(by~\eqref{eq:downToE3})}\\
&\geq \left(
\left( 1- \frac{\epsilon}{6} - \frac{24(\lceil\frac{12}{\epsilon}\rceil+1)}{k} \right)^2
- \frac{\epsilon}{3}
\right)
d(\OPT)
&& \text{(using~$p=\lceil\sfrac{12}{\epsilon}\rceil + 1$)}\\
&\geq \left(
\left(1-\frac{\epsilon}{3}\right)^2 - \frac{\epsilon}{3}
\right) d(\OPT)
&& \text{(using $k\geq \sfrac{144}{\epsilon} (\lceil\sfrac{12}{\epsilon}\rceil + 1)$)}\\
&\geq \left( 1 - \epsilon \right) d(\OPT)\enspace,
\end{align*}
as desired.
\end{proof}
\section{Combinations with submodular objectives}
\label{sec:submodObj}
In this section, we are interested in generalized diversity measures, which extend the dispersion function considered previously, by adding a monotone submodular function $f: 2^X \rightarrow \mathbb{Q}_{\geq 0}$ to it. Our goal is to maximize the objective $g(A) = d(A) + f(A)$ subject to a matroid constraint $M=(X,\mathcal{I})$. For brevity, we denote this problem by \MSDf. Throughout this section, we assume that $f$ is a nonnegative monotone submodular function on $X$ that is normalized, i.e., $f(\emptyset) = 0$. Notice that $f$ being normalized is an assumption without loss of generality, because any (non-normalized) function $f$ can be replaced by the normalized submodular function $h(S) = f(S) - f(\emptyset)$.
Submodular functions capture many natural diversity functions. One classic example is to count how many different ``types'' are covered by a set of elements. Here, one can define an arbitrary family $T_1,\dots,T_p \subseteq X$ of subsets of the ground set, where each $T_i$ can be thought of items sharing a particular property, and elements in $T_i$ are said to be of type $T_i$. Notice that the same element can be part of many different types. The function that assigns to each set $A\subseteq X$, the number of different types that are contained in $A$ is a well-known example of a submodular function, which is called a coverage function.
The problem of maximizing the sum of the dispersion and a monotone submodular function has previously been considered by Borodin et al.~\cite{borodin2012max}, who considered metric distance spaces. We recall that for metric distance spaces, a locally optimal solution is (only) a $\sfrac{1}{2}$-approximation. As we briefly explain below, our stronger analysis for negative type distances is compatible with their analysis and leads to stronger results for the case of negative type distances. Moreover, we show that these results can be combined with recent results on local search algorithms for maximizing submodular functions, leading to approximation guarantees that are asymptotically optimal for a wide range of problems.
For simplicity, we will focus on the approximation guarantee of solutions that are locally optimal with respect to our local search algorithm, without going into details on how to efficiently compute a solution whose objective value is no more than a factor $1+\epsilon$ larger than the objective value of a locally optimal solution, for a fixed error $\epsilon >0$. The steps to show fast convergence of the local search algorithm are standard, and can also be done analogously to our analysis in Section~\ref{sec:matConstraints}.
We start by giving a very brief overview of some recent results on which we build up. More precisely, we first give a quick overview of local search results by Borodin et al.~\cite{borodin2012max}, which we strengthen in the following. We then recap a non-oblivious local search algorithm by Filmus and Ward~\cite{filmus2014monotone} to obtain an optimal approximation guarantee for submodular maximization under a matroid constraint, and an extension of it by Sviridenko et al.~\cite{sviridenko2015optimal} to submodular functions with bounded curvature.
Our strengthened results then readily follow by combining these previous approaches together with our stronger analysis for distances of negative type.
Throughout this section, $A,B\subseteq X$ are bases of $M$, and $\pi: A \rightarrow B$ is a Brualdi bijection between $A$ and $B$, if not stated otherwise.
\subsection*{Local search results by Borodin et al.~\cite{borodin2012max}}
Borodin et al.~\cite{borodin2012max} study the \MSDf problem for metric distance spaces. They show that the local search algorithm achieves an approximation guarantee of $\frac{1}{2}$ for each objective function individually, and they prove that the same result carries over to the combined objective function. More specifically, in \cite[Lemmas 5 and 7]{borodin2012max} it is proven that,\footnote{Some results cited in this section are originally stated in less generality, but their original proofs carry on directly to the more general versions.} if $d$ is metric, and $f$ is submodular, monotone and normalized, then
\begin{align}
d(A)&\geq \frac{1}{2}d(B)+\frac{1}{2}\sum_{a\in A}\left(d(A)-d(A-a+\pi(a))\right), \ \text{ and} \label{ineq:metric} \\
f(A)&\geq \frac{1}{2}f(B)+\frac{1}{2}\sum_{a\in A}\left(f(A)-f(A-a+\pi(a))\right). \label{ineq:submod}
\end{align}
Inequality~\eqref{ineq:metric} only holds if the rank $k$ of the underlying matroid is at least $3$. For simplicity, we assume $k\geq 3$ in what follows.
Summing up~\eqref{ineq:metric} and~\eqref{ineq:submod}, one immediately obtains
$$g(A)\geq \frac{1}{2}g(B)+\frac{1}{2}\sum_{a\in A}\left(g(A)-g(A-a+\pi(a))\right).$$
From each of these lines, one can easily derive that any local optimum has an approximation guarantee of $\frac{1}{2}$ for the corresponding objective function. This approximation is tight for function $d$, as we mentioned in the introduction, assuming that the planted clique problem is hard; and for function $f$ this approximation is known to match the locality gap of this local search algorithm.
\subsection*{Non-oblivious local search by Filmus and Ward~\cite{filmus2014monotone}}
When trying to maximize a function $f$, sometimes it is convenient to define an auxiliary potential function $F:2^X \rightarrow \mathbb{R}_{\geq 0}$, and run a local search algorithm over $F$ instead of $f$. More precisely, an exchange step is done if it leads to a strict improvement of the function value of $F$ instead of $f$. Such an algorithm is called \emph{non-oblivious}.
Filmus and Ward \cite[Theorem 5.1]{filmus2014monotone} prove that there is a potential function $F: 2^X \rightarrow \mathbb{R}_{\geq 0}$ for $f$ such that
\begin{equation}\label{ineq:filmus}
f(A)\geq \left(1-\frac{1}{e}\right)f(B)+\left(1-\frac{1}{e}\right)\sum_{a\in A}\left(F(A) - F(A-a+\pi(a))\right)\enspace.
\end{equation}
From this, they conclude that the non-oblivious algorithm associated to $F$ offers an approximation guarantee of $1-\frac{1}{e}-\epsilon$ for the matroid-constrained maximization of a monotone, normalized submodular function. This is best possible, as it is proven by Feige \cite{feige1998threshold} that improving the bound $1-\frac{1}{e}$ is NP-hard even if $f$ is an explicitly given coverage function. And Nemhauser and Wolsey \cite{nemhauser1978best} show that improving upon this bound requires an exponential number of queries in the value oracle model.
We remark that an exact evaluation of the potential function $F$ defined in~\cite{filmus2014monotone} requires a superpolynomial number of evaluations of $f$. However, the authors prove that all the necessary evaluations of $F$ during the execution of the algorithm can be approximated efficiently, in such a way that with high probability, the ensuing loss in the approximation ratio is arbitrarily small. For clarity of exposition, we assume in this paper that we have access to a value oracle for $F$. It is also proven in~\cite{filmus2014monotone} that $F$ has further properties which are sufficient for the non-oblivious algorithm to converge fast.
\subsection*{Submodular maximization with bounded curvature by Sviridenko et al.~\cite{sviridenko2015optimal}}
Conforti and Cornu\'ejols \cite{conforti1984submodular} define the \emph{curvature} of a monotone submodular function $f$ as
$$c=1-\min_{x\in X}\frac{f(X)-f(X-x)}{f(x)-f(\emptyset)}.$$
The curvature is a coefficient between 0 and 1 that measures how close the function is to being linear, where $c=0$ corresponds to a linear function and $c=1$ to an arbitrary submodular function. For a monotone submodular function $f$ of curvature $c$, Sviridenko et al.~\cite{sviridenko2015optimal} consider the decomposition $f=l+f'$, where $l(A)=f(\emptyset)+\sum_{a\in A} (f(X)-f(X-a))$, and $f'(A)=f(A)-l(A)$, for each $A\subseteq X$. They prove that $l$ is linear, $f'$ is submodular, monotone and normalized, and $f'(A)\leq c \cdot f(A)$ for each $A\subseteq X$. It is easy to see that for any linear function $l$,
\begin{equation}\label{eq:linear}
l(A)= l(B)+\sum_{a\in A}\left(l(A)-l(A-a+\pi(a))\right).
\end{equation}
Hence, if we define a potential function $F'$ for $f'$ as in inequality (\ref{ineq:filmus}) (see \cite{filmus2014monotone}), and use it to define the potential function $F=l+\left(1-\frac{1}{e}\right)F'$ for $f$, the sum of inequality (\ref{ineq:filmus}) and equation (\ref{eq:linear}) gives for any $A,B\in \mathcal{I}$
\begin{align}\label{ineq:sviridenko}
f(A) &\geq f(B) -\frac{1}{e}f'(B)+\sum_{a\in A}\left(F(A) - F(A-a+\pi(a))\right)\nonumber\\
& \geq \left(1-\frac{c}{e}\right)f(B)+\sum_{a\in A}\left(F(A)-F(A-a+\pi(a))\right).
\end{align}
Thus, in \cite{sviridenko2015optimal} they conclude that, for a monotone submodular function $f$ of curvature $c$, the local search algorithm associated to the potential function $F$ offers an approximation guarantee of $1-\frac{c}{e} - \epsilon$. Moreover, they extend the negative result of \cite{nemhauser1978best} to prove that this bound is best possible; namely they prove that, for each $c\in [0,1]$, improving upon the bound of $1-\frac{c}{e}$ requires an exponential number of queries in the value oracle model.
\subsection*{Putting things together}
By combining our stronger analysis for local search subject to negative distance spaces with the results highlighted above, we obtain our main result which leads to Theorem~\ref{thm:msd+f}.
\begin{theorem}\label{thm:d+f_NT}
Consider \MSDf over a rank $k$ matroid constraint, where $d$ is of negative type and $f$ has curvature $c$. Let $\lambda_d=\frac{d(\OPT)}{g(\OPT)}$ and $\lambda_f=\frac{f(\OPT)}{g(\OPT)}$. Then, there is a non-oblivious local search algorithm such that any locally optimal solution has an approximation guarantee of
$$1-\lambda_d \frac{4}{k}-\lambda_f \frac{c}{e}\geq 1-\max\left\{\frac{4}{k}, \frac{c}{e}\right\}.$$
\end{theorem}
\begin{proof}
For the function $f$, consider the potential function $F$ as defined in inequality (\ref{ineq:sviridenko}) (see \cite{sviridenko2015optimal}); and for $g = d + f$, define the potential function $G=\left(1-\frac{2}{k}\right)d + F$.
Notice that inequality~\eqref{eq:matAverageImpDetailed} implies
\begin{equation*}
d(A) \geq \left(1-\frac{4}{k}\right) d(\OPT) +
\left(1-\frac{2}{k}\right)\sum_{a\in A}\left( d(A) - d(A-a+\pi(a))\right)\enspace,
\end{equation*}
which follows by multiplying~\eqref{eq:matAverageImpDetailed} by $1-\frac{2}{k}$, rearranging terms, and using $(1-\sfrac{2}{k})^2 \geq 1-\sfrac{4}{k}$, and $(1+\sfrac{2}{k})\cdot (1-\sfrac{2}{k}) \leq 1$.
Combining the above inequality with \eqref{ineq:sviridenko}, we obtain for any basis $A$ of $M$
\begin{equation*}
g(A)\geq g(\OPT) - \frac{4}{k} d(\OPT)-\frac{c}{e} f(\OPT) +\sum_{a\in A} \left(G(A)-G(A-a+\pi(a))\right)\enspace,
\end{equation*}
where $\pi:A\rightarrow \OPT$ is a Brualdi bijection between $A$ and $\OPT$.
We run the non-oblivious local search algorithm with respect to the potential $G$. Let $A$ be a local optimum with respect to $G$. The local optimality of $A$ implies $G(A)-G(A-a+\pi(a))\geq 0$, which, by the above inequality, implies
\begin{equation*}
g(A)\geq \left(1-\lambda_d \frac{4}{k}-\lambda_f \frac{c}{e}\right)g(\OPT)\enspace.
\end{equation*}
To complete the proof, notice that this approximation ratio is a convex combination of $1-\frac{4}{k}$ and $1-\frac{c}{e}$, and hence it is larger than the smaller of the two values.
\end{proof}
For completeness, we highlight that even for metric distances $d$, the recent non-oblivious local search procedures presented in \cite{filmus2014monotone,sviridenko2015optimal} lead to a strengthening of the result of Borodin et al.~\cite{borodin2012max}.
\begin{theorem}\label{thm:d+f_metric}
Consider \MSDf over a matroid constraint, where $d$ is metric, and $f$ has curvature $c$. Let $\lambda_d\geq 0$ such that $\frac{d(O)}{g(O)}\leq \lambda_d$. Then, there is a non-oblivious local search algorithm such that any locally optimal solution has an approximation guarantee of
\begin{equation*}
1-\lambda_d \frac{1}{2}-(1-\lambda_d)\frac{c}{e}\enspace.
\end{equation*}
\end{theorem}
\begin{proof}
The proof is analogous to the previous one, except that we add up inequalities (\ref{ineq:metric}) and (\ref{ineq:sviridenko}) instead of the inequality provided by Lemma~\ref{lem:matAverageImp} and~\eqref{ineq:sviridenko}; and consequently for $g=d+f$ we define the potential function $G=\frac{1}{2}d + F.$
\end{proof}
\bibliographystyle{plain}
| 8d6bee068377f2e1d4f6472d820e71d44453e0cc | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Entanglement is a valuable resource in precision measurement, as measurements using entangled probe systems have fundamentally higher optimal sensitivity than those using unentangled states \cite{Bollinger1996}. A generic measurement using $N$ unentangled probes will have a standard deviation from the true value asymptotically proportional to $1/ \sqrt{N}$. By using $N$ maximally entangled probes, a single parameter coupled independently to each probe system can be measured with an uncertainty proportional to $1/N$. This is the best possible scaling consistent with the Heisenberg uncertainty principle and is known as the Heisenberg limit \cite{Wineland1992,Bollinger1996}.
The procedure can also be reversed--enhanced sensitivity to disturbances can provide experimental evidence of entanglement \cite{Pezze2009,Strobel2014,Toth2012}.
Measurements making use of entanglement usually couple one parameter to $N$ different systems \cite{Bollinger1996,Dinani2016,Kessler2014}. However, the emerging potential of long-range quantum information opens new avenues for metrology \cite{Komar2014,Komar2016} and entanglement distribution \cite{Valivarthi2016}. The ability to distribute entanglement across spatially separated regions has already been used for recent loophole-free tests of Bell's inequality \cite{Giustina2015b,Shalm2015,Hensen2015}.
In this work, we are interested in coupling $N$ parameters to $N$ different systems, which may be spatially separated, and measuring a linear function of all of them (see Fig.~\ref{fig:setup}a) such as a single mode of a spatially varying field. Such measurements may be of interest in geodesy, geophysics, or medical imaging \cite{Nabighian2005,Ding2013,Wright2004,Stacey1964,Hamalainen1993}, but in this paper we focus on potential application to nanoscale nuclear magnetic resonance (NMR) imaging. Later in this paper we will discuss precisely how our method might apply in this setting.
\begin{figure}[tb]
\includegraphics[width=8.6cm]{protein}
\includegraphics[width=8.6cm]{partialtime}
\caption{(a) An illustration of the network setup in a nanoscale NMR setting. Nodes, located at different points relative to a large molecule, share an entangled state; at each node there is both an unknown parameter $\theta_i$ and a known relative weight $\alpha_i$. We are concerned with estimating $\bm{\alpha} \cdot \bm{\theta}$. (b) Illustration of the partial time evolution protocol for three qubits. Solid green segments of the timeline represent periods when a qubit is evolving due to coupling to the local parameter $\theta_i$, while dashed red segments represent periods after the qubit stops evolving. The switches occur at times corresponding to the qubits' weights in the final linear combination. The weight of the last qubit is $\alpha_3 = 1$.}
\label{fig:setup}
\end{figure}
The function $q$ we wish to measure is a weighted sum of the deterministic individual parameters $\theta_i$, where $i$ indexes the individual systems and each weight is denoted by a known real number $\alpha_i$,
\begin{equation}
q = \sum_{i=1}^N \alpha_i \theta_i = \bm{\alpha} \cdot \bm{\theta}.
\label{eqn:qdef}
\end{equation}
In this paper, we characterize the advantage entanglement provides in this setting and construct an optimal strategy equivalent to turning some qubits' evolution ``on'' and ``off'' for time proportional to the weight with which their parameter contributes to the function $q$ (see Fig.~\ref{fig:setup}b). With this scheme of ``partial time evolution,'' we can measure a linear function with the minimum variance permitted by quantum mechanics, which can be viewed as an extension of the Heisenberg limit to linear combinations. We will also show that our method can protect the secrecy of the result, allowing the network as a whole to perform a measurement without eavesdroppers learning any details of $\bm{\alpha} \cdot \bm{\theta}$.
\section{Setup}
We consider a system in which there are $N$ sensor nodes. Each sensor node $i$ possesses a single qubit coupled to an unknown parameter $\theta_i$ unique to each node. We suppose that the state evolves unitarily under the Hamiltonian
\begin{equation}
\label{eqn:hamiltonian}
\hat{H} = \hat{H}_c (t) + \sum_{i=1}^N \frac{1}{2} \theta_i \hat{\sigma}^z_i.
\end{equation}
Here, $\hat{H}_c (t)$ is a time-dependent control Hamiltonian chosen by us, which may include coupling to additional ancilla qubits and $\hat{\sigma}^{x,y,z}_i$ are the Pauli operators acting on qubit $i$.
We wish to measure the quantity $q$ defined in Eq.~\eqref{eqn:qdef}.
We assume that $\forall i : \abs{\alpha_i} \leq 1$ and additionally that there is at least one $\alpha_i$ such that $\alpha_i = 1$. These conditions simply set a scale for the function, and for an arbitrary $\bm{\alpha}$ all that is needed is division by the largest $\alpha_i$ to meet this requirement. As an example, a network with two nodes interested in measuring the contrast between those nodes would set $\bm{\alpha} = \left( 1, -1\right)$ to measure $\theta_1 - \theta_2$. We would like to establish how well an arbitrary measurement of $\bm{\alpha} \cdot \bm{\theta}$ can be made and what the best measurement protocol is for doing so. By ``protocol'' we mean three different choices: (1) which input state we begin with, (2) what auxiliary control Hamiltonian $\hat{H}_c(t)$ we implement, and (3) how the final measurement is made.
We define the quality of measurement in terms of an estimator, $Q$, constructed from experimental data. (Throughout this paper, we denote operators with hats, vectors by boldface, quantities to be estimated by lowercase, and corresponding estimators by uppercase.) We assume that the estimator is unbiased, so that its expectation value is the true value $\mathbf{E} \left[Q \right] = q$. Then our metric for the quality of the measurement is the average squared error, or variance, of the estimator,
\begin{equation}
\Var Q = \mathbf{E} \left[ \left( Q - q \right)^2 \right].
\end{equation}
If measurements of $\theta_i$ can be made locally with accuracy $\Var \Theta_i$ for an estimator $\Theta_i$, then we could compute the linear combination by local measurements and classical computation. In this case, the variance is given by classical statistical theory as $\Var Q = \| \bm{\alpha} \|^2 \Var \Theta_0$ assuming that $\Var \Theta_i$ is identical at each site and equal to $\Var \Theta_0$. A measurement of an individual $\theta_i$ in Eq.~\eqref{eqn:hamiltonian} can be made in time $t$ with a variance of $1/t^2$ \cite{Wineland1992}. Therefore, our entanglement-free figure of merit is
\begin{equation}
\Var Q \geq \frac{\| \bm{\alpha} \|^2}{t^2}.
\label{eqn:sql}
\end{equation}
We consider this the standard quantum limit for networks. To compare to the typical case where $N$ independent qubits measure a single parameter, consider the average $\bar{\theta}$, which is equivalent to setting all $\alpha_i = 1$ and then using $\bar{\Theta} = Q/N$ to obtain $\Var \bar{\Theta} = 1/ N t^2$. It is our goal in this paper to present a means to improve on the limit in Eq.~\eqref{eqn:sql}.
\section{Heisenberg Limit for Sensor Networks }
\subsection{Using Fisher Information Matrix}
Our task is to perform parameter estimation on a quantum system evolving under some set of parameters $\{\theta_i\}$ linearly coupled to sensor qubits as in Eq.~\eqref{eqn:hamiltonian} \cite{Braunstein1996,Paris2008,Boixo2007,Pang2014}. Although we are only interested in measuring a single number, we still need to treat a system that has many parameters in the evolution, necessitating the use of a multi-parameter theory as in Refs.~\cite{Baumgratz2015,Crowley2014,Gao2014,Genoni2013,Humphreys2013,Kok2017,Yue2014,Zhang2014, Proctor2017}. It is known from classical estimation theory that, given a probability distribution $p(z)$ over a set of outcomes $z$ that depends on a number of parameters, all estimators of the parameters obey the Cram\'{e}r-Rao inequality \cite{Helstrom1968,Helstrom1969},
\begin{equation}
\Sigma \geq \frac{F^{-1}}{M}.
\label{eqn:crb}
\end{equation}
Here, $M$ is the number of experiments performed, $F$ is the Fisher information matrix (see below), and $\Sigma$ is the covariance matrix, where $\Sigma_{ij} = \mathbf{E} \left[ \left( \Theta_i - \theta_i \right) \left( \Theta_j - \theta_j \right) \right]$. The inequality is a matrix inequality, meaning that $M \Sigma - F^{-1}$ is positive semidefinite. We will concern ourselves with the single-shot Fisher information, and set $M = 1$ from now on. The Fisher information matrix captures how each parameter changes the probability distribution of outcomes,
\begin{equation}
F_{ij} = \int p(z) \left( \pdpd{ \ln p(z)}{\theta_i} \right) \left( \pdpd{ \ln p(z)}{\theta_j} \right) \dd{z}.
\label{eqn:fishermat}
\end{equation}
This bound is a purely classical statement about probability distributions, and is saturated asymptotically using a maximum-likelihood estimator \cite{Braunstein1994}. Note that although we have presented the formulas for the Fisher information matrix, in the case of a single parameter the Fisher information will be a scalar which can be obtained by setting $i = j$ in Eq.~\eqref{eqn:fishermat}.
Quantum theory bounds the probability distributions that can result from a state evolved under a parameter-dependent unitary operation \cite{Braunstein1996}. We thus define the quantum Fisher information $F_Q$ for a process with a given initial state as the maximization of the Fisher information over all possible measurement schemes. This gives rise to the quantum Cram\'{e}r-Rao bound (QCRB), which simply replaces $F$ with $F_Q$ in Eq.~\eqref{eqn:crb}. A matrix element of $F_Q$ for a pure state evolving under a Hamiltonian $\hat{H}$ is given by
\begin{equation}
\left( F_{Q} \right)_{ij} = 4 t^2 \left[ \expect{\hat{g}_i \hat{g}_j} - \expect{\hat{g}_i} \expect{\hat{g}_j} \right],
\label{eqn:qfishmat}
\end{equation}
where $\hat{g}_i = \left( \partial \hat{H} / \partial \theta_i \right)$ is the generator corresponding to parameter $i$. For instance, in Eq.~\eqref{eqn:hamiltonian} the generator $\hat{g}_i$ is the operator $\frac{1}{2} \hat{\sigma}^z_i$. Unlike the Cram\'{e}r-Rao bound, the QCRB cannot always be satisfied, even asymptotically. However, in the setting of this paper, where all generators commute, it can be \cite{Baumgratz2015}. Equation~\eqref{eqn:crb} then takes the form:
\begin{equation}
\Sigma \geq F^{-1} \geq F_Q^{-1}.
\label{eqn:crbchain}
\end{equation}
To formulate the appropriate Cram\'{e}r-Rao bound in the case where the quantity we wish to estimate is a linear combination of the $\theta_i$, we simply use the fact that the variance of a linear combination $\bm{\alpha} \cdot \bm{\theta}$ can be written as $\bm{\alpha}^T \Sigma \bm{\alpha}$. It follows immediately from Eq.~\eqref{eqn:crbchain} that
\begin{equation}
\Var Q \geq \bm{\alpha}^T F_Q^{-1} \bm{\alpha}.
\label{eqn:lincombcrb}
\end{equation}
Note that although we began by considering the full covariance matrix, we now focus on just a \emph{single} scalar $\bm{\alpha}^T F_Q^{-1} \bm{\alpha}$ because our quantity of interest is a \emph{single} linear transformation of the original parameters.
In order to properly define the Cram\'{e}r-Rao bound, it is necessary to consider the fact that $F$ and $F_Q$ are only positive semi-definite and not necessarily invertible. For instance, if a parameter has no effect on probabilities at all, then it cannot be estimated from experimental results and the bound on the variance of its estimator is undefined. To sidestep this issue, we can instead look at $\tilde{F}_Q$, the quantum Fisher information projected onto its own image \cite{Proctor2017}, assuming that $\bm{\alpha}$ has no overlap with the kernel of $F_Q$. This matrix (and its inverse) are now both positive definite, meaning they can always be inverted. Equation~\eqref{eqn:lincombcrb} is therefore always well-defined if $\tilde{F}_Q$ is used.
Since $\tilde{F}_Q$ is Hermitian and positive definite, $\sqrt{\tilde{F}_Q}$ is Hermitian. We can then write the following for an arbitrary real $\bm{b}$ by invoking the Cauchy-Schwarz inequality:
\begin{align}
\bm{\alpha}^T \tilde{F}_Q^{-1} \bm{\alpha} &= \frac{\| \sqrt{\tilde{F}_Q^{-1}} \bm{\alpha} \|^2 \| \sqrt{\tilde{F}_Q} \bm{b} \|^2}{ \bm{b}^T \tilde{F}_Q \bm{b} } \\ &\geq \frac{\| \bm{\alpha}^T \sqrt{\tilde{F}_Q^{-1}} \sqrt{\tilde{F}_Q} \bm{b} \|^2}{ \bm{b}^T \tilde{F}_Q \bm{b} } \\
& \geq \frac{\| \bm{\alpha}^T \bm{b} \|^2}{ \bm{b}^T \tilde{F}_Q \bm{b} }.
\end{align}
Taking $\bm{b}$ to be the $b$th element of the standard basis gives
\begin{equation}
\Var{Q} \geq \bm{\alpha}^T \tilde{F}_Q^{-1} \bm{\alpha} \geq \frac{ \alpha_b^2 }{ \left(\tilde{F}_Q \right)_{bb}}.
\label{eqn:multiparamqcrb}
\end{equation}
Here, $\left( \tilde{F}_Q \right)_{bb}$ is the quantum Fisher information for a single parameter, as defined by Eq.~\eqref{eqn:qfishmat}. In Ref.~\cite{Boixo2007}, it was shown that for any time-dependent control Hamiltonian $\hat{H}_c(t)$, including those with ancilla qubits,
\begin{equation}
\left(\tilde{F}_Q \right)_{bb} \leq t^2 \| \hat{g}_b \|_s^2.
\end{equation}
Here $\| \hat{g}_b \|_s$ is the operator seminorm (difference between the largest and smallest eigenvalues) of the generator corresponding to parameter $\theta_b$. Our final bound comes from applying this condition and recognizing that the formula must hold for all $b$:
\begin{equation}
\Var Q \geq \max_b \frac{ \alpha_b^2}{t^2 \| \hat{g}_b \|_s^2}.
\label{eqn:generalbound}
\end{equation}
We emphasize that Eq.~\eqref{eqn:generalbound} remains true no matter what time-dependent control $\hat{H}_c(t)$ is applied.
In Eq.~\eqref{eqn:hamiltonian}, all $\hat{g}_b = \frac{1}{2} \hat{\sigma}^z_b$, $\| \hat{g}_b \|_s = 1$, and we find a bound,
\begin{equation}
\Var Q \geq \max_i \frac{\alpha_i^2}{t^2} = \frac{1}{t^2}.
\label{eqn:ptebound}
\end{equation}
Here we have used the fact that the largest $\alpha_i = 1$. If we want to estimate the average of the $\theta_i$, then all qubits are equally weighted and the desired quantity is $\bar{\theta} = q/N$, so $\Var{\bar{\Theta}} \geq 1 / N^2 t^2$ and we reproduce the desired Heisenberg scaling which is more precise than the $1/N$ in Eq.~\eqref{eqn:sql}. However, note that if we wanted to estimate only a single $\theta_i$, then we would not benefit from the entanglement. In general, we can, for some situations, greatly improve the precision of parameter esitmation with nonlocal techniques if the parameter itself is also non-local. Our bound allows us to explore the full range of possible $\bm{\alpha}$ between these two extremes. Compared to the bound on unentangled states [Eq.~\eqref{eqn:sql}], Eq.~\eqref{eqn:generalbound} simply picks out the largest contribution due to uncertainty from a single site. Equation~\eqref{eqn:generalbound} can be viewed as an extension of the usual Heisenberg bound to linear combinations of parameters.
We can illustrate the above argument by optimizing over the space of all control Hamiltonians $\hat{H}_c(t)$. As this is computationally expensive, we limit ourselves to a two-qubit sensor network with no ancillas. The Hamiltonians we optimize over include enough operators to provide universal control on two qubits, meaning we can effectively modify the input state as well as the final measurement basis in order to optimize the Fisher information. In order to test the form of our bound, Eq.~\eqref{eqn:generalbound}, which depends both on relative weights of each parameter and the underlying generator, we couple $\theta_1$ to a generator $\hat{\sigma}^z_1$ which has $\| \hat{\sigma}^z_1 \|_s = 2$. We leave the second qubit coupled to a generator $\frac{1}{2} \hat{\sigma}^z_2$ as in Eq.~\eqref{eqn:hamiltonian}. The bound corresponding to the first qubit from Eq.~\eqref{eqn:generalbound} is $\alpha_1^2/4 t^2$ and that of the second qubit is $\alpha_2^2 / t^2$. In our numerics, we set $\alpha_1 = t = 1$, meaning the two bounds are $1/4$ and $\alpha_2^2$. Our analytic result leads us to believe therefore that if $\alpha_2^2 > 1/4$, the minimum possible variance should be $\alpha_2^2$. However, if $\alpha_2^2 < 1/4$, then the lower bound should be $1/4$. That behavior is precisely what we find through the numerical optimization shown in Fig.~\ref{fig:graddescent}, confirming Eq.~\eqref{eqn:generalbound}.
\begin{figure}[tb]
\includegraphics[width=8.6cm]{optimization}
\caption{Numerical optimization of $\bm{\alpha}^T F^{-1}_Q \bm{\alpha}$ for two qubits with $\alpha_1 = 1$ compared to the bound predicted by our analytic result. Each point is generated by running a gradient descent algorithm until convergence; the control parameters begin at small random values. The dashed (dotted) line is the analytic bound derived from the first (second) qubit. As $\alpha_2$ increases, the second qubit becomes the source of the relevant bound.}
\label{fig:graddescent}
\end{figure}
\subsection{Using Single-Parameter Bounds}
\label{sec:spbound}
It is tempting to dismiss the above argument as unnecessarily complicated, as the ultimate quantity of interest is only a single parameter. Why not simply apply the Cram\'{e}r-Rao bound directly to $\bm{\alpha} \cdot \bm{\theta}$ instead of using the matrix approach? We will now show that the single-parameter bound that arises from naive application of the Cram\'{e}r-Rao bound is looser than Eq.~\eqref{eqn:generalbound}. This gap occurs because the single-parameter bound can only be applied if there is only one unknown parameter controlling the evolution of the input state, which implicitly places a constraint on the other components of the field. Later, we will discuss how the single-parameter approach can be amended to take this into account and agree with Eq.~\eqref{eqn:generalbound}.
To apply the single-parameter Cram\'{e}r-Rao bound to our evolution Hamiltonian Eq.~\eqref{eqn:hamiltonian}, we consider the Hamiltonian as $\frac{1}{2}\bm{\theta} \cdot \bm{\hat{\sigma}}$ where $\bm{\hat{\sigma}}$ is simply a vector of operators whose $i$th element is $\hat{\sigma}^z_i$. We then rewrite $\bm{\theta}$ in a new basis,
\begin{equation}
\bm{\theta} = \sum_{i=0}^{N - 1} \left(\bm{\alpha}_i \cdot \bm{\theta} \right) \bm{\beta}_i.
\label{eqn:rebasis}
\end{equation}
We assume that $\bm{\alpha}_0 = \bm{\alpha}$ and that the other $\bm{\alpha}_{i>0}$ make up a basis. The set of vectors $\bm{\beta}_i$ is then a dual basis such that $\bm{\alpha}_i \cdot \bm{\beta}_j = \delta_{ij}$. (For this basis as well, we will drop the subscript 0 to indicate that this particular vector corresponds to the parameter of interest.) The advantage of rewriting $\bm{\theta}$ this way is that we can now identify the term in the Hamiltonian which is proportional only to $\bm{\alpha} \cdot \bm{\theta}$. The generator corresponding to the quantity $\bm{\alpha} \cdot \bm{\theta}$ is:
\begin{equation}
\hat{g} = \pdpd{\hat{H}}{\left( \bm{\alpha} \cdot \bm{\theta} \right)} = \frac{\bm{\beta} \cdot \bm{\hat{\sigma}}}{ 2}.
\end{equation}
To obtain the quantum Fisher information corresponding to this generator, we consider the variance of the operator $\hat{g}$. The maximum variance of this generator is given by the operator seminorm \cite{Boixo2007}. Using this fact, we can write:
\begin{equation}
F_Q \leq t^2 \| \hat{g} \|_s^2 = t^2 \left( \sum_i \abs{\beta_i} \right)^2.
\label{eqn:badbound}
\end{equation}
In general the bound on $\Var Q$ derived from Eq.~\eqref{eqn:badbound} is a looser lower bound than Eq.~\eqref{eqn:ptebound}. For example, with $\bm{\alpha} = (1, \frac{1}{2})$ and $\bm{\alpha}_1 = (\frac{1}{2}, -1)$, this implies that $\bm{\beta} = (\frac{4}{5}, \frac{2}{5})$. Equation~\eqref{eqn:badbound} would suggest that
\begin{equation}
\Var Q \geq \frac{25}{36 t^2},
\end{equation}
which is looser than the $1/t^2$ given by Eq.~\eqref{eqn:generalbound}. This discrepancy can be addressed by thinking more closely about the process of choosing a new basis. We will use the seminorm condition again to bound the maximum possible Fisher information. To start calculating the seminorm, we express it in terms of the elements of $\bm{\beta}$:
\begin{equation}
\| \hat{g} \|_s = \| \sum_{j} \beta_{j} \frac{1}{2} \hat{\sigma}^z_j \|_s = \sum_j \abs{\beta_{j}}.
\label{eqn:genbound}
\end{equation}
We will now show that it is possible to choose a basis such that the seminorm in Eq.~\eqref{eqn:genbound} goes to infinity. This shows that the approach which led us to Eq.~\eqref{eqn:badbound} should not be applied blindly, and we will then discuss how to control for this issue. First, an illustration of the bound diverging. Suppose that in a two-parameter problem, the basis vectors we choose are $\bm{\alpha}$ and $\bm{\alpha}'$. It can then be shown by direct computation of the matrix inverse that yields the dual basis that the implied maximum Fisher information from Eq.~\eqref{eqn:genbound} is:
\begin{equation}
F \leq \| \hat{g} \|_s = \frac{\alpha'_2 + \alpha'_1}{\abs{ \alpha_1 \alpha'_2 - \alpha'_1 \alpha_2}}.
\end{equation}
If we then choose $\bm{\alpha}' = \left(\alpha_1/\alpha_2 + \varepsilon, 1 \right)$, it follows that:
\begin{equation}
F \leq \frac{1 + \varepsilon + \frac{\alpha_1}{\alpha_2}}{\varepsilon \alpha_2}.
\end{equation}
As $\varepsilon \to 0$, this becomes arbitrarily large. From this we conclude that our previous approach was ill-advised as it can yield arbitrarily small lower bounds on the estimator variance -- using this basis, we would conclude that the right-hand side of Eq.~\eqref{eqn:badbound} could be $\infty$.
In order to produce a useful bound from Eq.~\eqref{eqn:genbound}, we recognize that any possible choice of basis must yield a valid bound. Therefore, rather than look at one particular basis (as we did in deriving Eq.~\eqref{eqn:badbound}) we instead need to optimize for the highest lower bound over all possible choices of basis. Finding the tightest bound on Fisher information will then produce the highest lower bound on parameter uncertainty. To do this, we first write the following chain of inequalities using the relationship of $\bm{\alpha}$ and $\bm{\beta}$:
\begin{align}
1 &= \sum_j \alpha_{j} \beta_j \leq
\sum_j \abs{ \alpha_j \beta_j} \leq
\sum_j \abs{ \beta_j},
\label{eqn:matrixbound}
\end{align}
where the last line follows due to the fact that $\abs{\alpha_j} \leq 1$. Note that we can achieve equality,
\begin{equation}
\sum_j \abs{\beta_j} = 1,
\end{equation}
by taking the other $N-1$ basis vector $\bm{\alpha}_j$ to be unit vectors $\bm{e}_j$ in the standard basis, making sure that the $j$ that does not appear has $\alpha_j \neq 0$ to ensure the entire space is spanned. Now we look to the minimum possible value of $\|\hat{g}\|_s$. The minimum possible value is interesting to us because the minimum $\|\hat{g} \|_s$ will be the choice of basis for which the bound on Fisher information is tightest.
It follows from Eq.~\eqref{eqn:genbound} and Eq.~\eqref{eqn:matrixbound} that the minimum seminorm $\| \hat{g}\|_s$ is equal to 1, implying that the maximum value for $\Var \hat{g}$ is $1/4$ \cite{Boixo2007}. Using this to optimize the bound in Eq.~\eqref{eqn:badbound} over all possible choices of re-parameterization implies that $\Var \bm{\alpha} \cdot \bm{\theta} \geq 1/t^2$, just as we found in Eq.~\eqref{eqn:ptebound}.
The single-parameter bound \textit{is} applicable in our situation, but it requires careful accounting of the influence of other parameters in the problem. The reason that our previous results such as Eq.~\eqref{eqn:generalbound} do not hold in this case is that Cram\'{e}r-Rao bound does not apply if we can take advantage of constraints on the signal field $\bm{\theta}$ to improve our estimation strategy.
These naive single-parameter bounds can be applied and saturated if the field structure is known before the measurement takes place. To demonstrate, suppose that for a set of fields $\bm{\theta}$ where we wish to learn $\bm{\alpha} \cdot \bm{\theta}$, we know that the fields are proportional to $\alpha_i$. Then we can write the total field as:
\begin{equation}
\bm{\theta} = q \frac{\bm{\alpha}}{\| \bm{\alpha} \|^2},
\end{equation}
and our goal is to estimate $q = \bm{\alpha} \cdot \bm{\theta}$. This is now a truly one-parameter problem, enabling a new strategy which saturates Eq.~\eqref{eqn:badbound}.
By defining $\bm{w}$ as a new vector such that $w_i = \sgn{\alpha_i}$, we can measure the quantity
\begin{equation}
\bm{w} \cdot \bm{\theta} = q \frac{ \sum \abs{\alpha_i}}{\| \bm{\alpha} \|^2}.
\end{equation}
Since $\bm{w}$ is a linear combination which satisfies the condition $\abs{w_i} \leq 1$, we can estimate $q' = \bm{w} \cdot \bm{\theta}$ with accuracy bounded by $1/t^2$ as shown in Sec.~\ref{sec:protocols}. Then:
\begin{align}
\Var \left( Q' \right) &= \Var \left( Q \frac{ \sum \abs{\alpha_i}}{\| \bm{\alpha} \|^2} \right) \geq \frac{1}{t^2}, \\
\implies \Var Q &\geq \frac{\| \bm{\alpha} \|^4}{ t^2 \left(\sum \abs{\alpha_i} \right)^2}.
\end{align}
This saturates the bound in Eq.~\eqref{eqn:badbound}.
The reason we are able to outperform Eq.~\eqref{eqn:generalbound} is that we have assumed something about the structure of the field which reduces it to a lower-dimensional problem. This is only possible by using knowledge about components of $\bm{\theta}$ not parallel to $\bm{\alpha}$. Otherwise, there is no guarantee that $\bm{\theta}$ will be proportional to $\bm{\alpha}$. In general cases $\bm{w} \cdot \bm{\theta}$ will contain noise from ``undesired'' components.
In many situations where the field structure is known, new strategies can be introduced which may outperform our previous results, even asymptotically. Consider as an example a case with a field:
\begin{equation}
\bm{\theta} = \theta \frac{\bm{\alpha}}{\| \bm{\alpha} \|^2} + \theta_\gamma \bm{\gamma},
\end{equation}
where $\bm{\alpha} \cdot \bm{\gamma} = 0$ and $\theta_\gamma$ is a nuisance parameter describing the field magnitude orthogonal to $\bm{\alpha}$. Any field can be written in this way to separate out the $\bm{\alpha}$ component. Suppose we measure $\bm{w} \cdot \bm{\theta}$. We know that:
\begin{equation}
\Var Q' \geq \frac{1}{t^2}
\end{equation}
is an achievable bound. By writing $\bm{w} = c_\alpha \bm{\alpha} + c_\gamma \bm{\gamma}$, decomposing $\bm{w}$ into its only relevant components, we can obtain the following bound on $\Var Q$:
\begin{equation}
\Var Q \geq \frac{1}{t^2 c_\alpha} + c_\gamma^2 \| \bm{\gamma} \|^2.
\end{equation}
Therefore, the optimal strategy is to pick a $\bm{w}$ which maximizes $\bm{w} \cdot \bm{\alpha}$ while minimizing (preferably to zero) $\bm{w} \cdot \bm{\gamma}$. However, in general, learning the structure of the field perpendicular to $\bm{\alpha}$ is just as difficult as learning the component parallel to $\bm{\alpha}$, so beginning from a state of ignorance, it is still optimal to measure $\bm{\alpha} \cdot \bm{\theta}$ rather than a different linear combination.
The bound in Eq.~\eqref{eqn:badbound} can actually be found by other statistical methods which fully treat the initial multi-parameter structure, for instance, the constrained Cram\'{e}r-Rao bound of Ref.~\cite{Stoica1998}. It can also be derived from the Van Trees inequality \cite{VanTrees2013} by assuming that we have pre-existing knowledge that the components of $\bm{\theta}$ perpendicular to $\bm{\alpha}$ have a normal distribution of width $\varepsilon$ and then taking the limit $\varepsilon \to 0$.
If rather than a constraint we simply have some initial information in the form of a prior distribution, the Van Trees inequality (which takes into account that prior information) will reduce to the Cram\'{e}r-Rao bound in the limit of many measurements. This is because the information gained from measurements scales linearly with the number of measurements while the prior information is static.
\section{Protocols}
\label{sec:protocols}
We now present two protocols that saturate the bound of Eq.~\eqref{eqn:ptebound} and are therefore optimal. The first begins from the conceptually simple Greenberger-Horne-Zeilinger (GHZ) state or a spin-squeezed state and uses time-dependent control during phase accumulation to produce an output state sensitive to the desired $\bm{\alpha} \cdot \bm{\theta}$, while the second method uses a more complicated initial state but requires no control during the phase accumulation.
\subsection{Protocols Involving Time-Dependent Control}
\label{sec:tdprotocols}
\subsubsection{Using GHZ Input State}
\label{sec:pte}
We start by considering an $N$-qubit GHZ state:
\begin{equation}
\frac{1}{\sqrt{2}} \left( \ket{0}^{\otimes N} + \ket{1}^{\otimes N} \right).
\end{equation}
Under $\hat{\sigma}^z$ evolution, each $\ket{1}$ accumulates a phase relative to $\ket{0}$. By allowing qubits to accumulate phase proportional to the desired weight $\alpha_i$, we obtain a final state in which $\ket{1}^{\otimes N}$ has accumulated a total phase of $\bm{\alpha} \cdot \bm{\theta} t$ relative to $\ket{0}^{\otimes N}$.
We refer to our protocol as ``partial time evolution'' because it relies on a qubit undergoing evolution for a fraction of the total measurement time (see Fig.~\ref{fig:setup}). We can realize this by applying $\hat{\sigma}^x_i$ to a qubit at time $t_i = t \left(1 + \alpha_i \right)/2$ so that the qubit evolution will be identical to evolving it for a time $\alpha_i t$. Note that if there is a fixed experimental time $t$, this scheme can realize values of $\alpha_i \in \left[ -1, 1 \right]$, which motivates our restrictions on the values of individual $\alpha_i$. Specifying this sequence of gates identifies the $\hat{H}_c(t)$ which defines the protocol.
The result of this protocol is an effective evolution according to the unitary operator
\begin{equation}
\hat{U}(t) = e^{ - i \frac{t}{2} \sum_{i = 1}^N \alpha_i \theta_i \hat{\sigma}^z_i}.
\label{eqn:pteunitary}
\end{equation}
Under this evolution, the final state is:
\begin{align}
\frac{1}{\sqrt{2}} \left( e^{-i \frac{t}{2} q} \ket{0}^{\otimes N} + e^{i \frac{t}{2} q } \ket{1}^{\otimes N} \right).
\label{eqn:postptestate}
\end{align}
Now we make a measurement of the overall parity of the state, $ \hat{P} = \bigotimes_{i=1}^N \hat{\sigma}^x_i$. The details of this measurement and calculation of $\langle \hat{P} \rangle$ are given in Ref.~\cite{Bollinger1996}; notably, the measurement can be performed locally at each site.
Measurement of the time-dependent expectation value $\langle \hat{P} \rangle (t)$ allows for the estimation of $Q$ with accuracy \cite{Wineland1994}
\begin{equation}
\Var Q = \frac{\Var \hat{P}(t)}{\left(\partial{\expect{\hat{P}}} / \partial{q} \right)^2} = \frac{ \sin^2 q t }{t^2 \sin^2 q t } = \frac{1}{t^2},
\label{eqn:ptesensitivity}
\end{equation}
saturating the bound in Eq.~\eqref{eqn:ptebound} and Fig.~\ref{fig:graddescent}.
We can also directly evaluate $F_Q$ and $F$ for this protocol. $F_Q$ can be found by noting that this protocol is identical to evolution under the Hamiltonian $\frac{1}{2} \sum \alpha_i \theta_i \hat{\sigma}_i^z$. Therefore the quantum Fisher information matrix $F_Q$ is simply
\begin{equation}
\left( F_Q \right)_{ij} = t^2 \left[ \expect{ \alpha_i \hat{\sigma}_i^z \alpha_j \hat{\sigma}_j^z } - \expect{ \alpha_i \hat{\sigma}_i^z } \expect{ \alpha_j \hat{\sigma}_j^z } \right] = \alpha_i \alpha_j t^2.
\end{equation}
Furthermore, we can show that this $F_Q$ satisfies the second inequality in Eq.~\eqref{eqn:multiparamqcrb}. The inverse of $\tilde{F}_Q$ can be easily written, as $F_Q$ simply projects onto $\bm{\alpha}$. In order to get $\tilde{F}_Q^{-1} \tilde{F}_Q = \bm{\alpha}^T \bm{\alpha} / \| \bm{\alpha} \|^2$ (identity on the image of $F_Q$), we must have
\begin{equation}
\tilde{F}_Q^{-1} = \frac{ \alpha_i \alpha_j }{t^2 \| \bm{\alpha} \|^4}.
\end{equation}
$\bm{\alpha}^T \tilde{F}_Q^{-1} \bm{\alpha}$ is then equal to $1/t^2$, saturating the second inequality in Eq.~\eqref{eqn:multiparamqcrb} for the basis vector $\bm{b}$ corresponding to the largest $\bm{\alpha}$ component, $\alpha_b$= 1.
To evaluate the classical Fisher information in this case, we note that the final measurement \cite{Wineland1994} projects onto one of two outcomes with probability $\sin^2 \left( \bm{\alpha} \cdot \bm{\theta} t/2 \right)$ and $\cos^2 \left( \bm{\alpha} \cdot \bm{\theta} t /2 \right)$. Therefore the classical Fisher information is simply:
\begin{align}
F &= \frac{\left(\pdpd{\left( \sin^2 \frac{\bm{\alpha} \cdot \bm{\theta} t}{2} \right) }{\bm{\alpha} \cdot \bm{\theta}} \right)^2}{\sin^2 \bm{\alpha} \cdot \bm{\theta} t/2} + \frac{\left(\pdpd{\left( \cos^2 \frac{\bm{\alpha} \cdot \bm{\theta} t}{2} \right) }{\bm{\alpha} \cdot \bm{\theta}} \right)^2}{\cos^2 \bm{\alpha} \cdot \bm{\theta} t/2} \\ &= t^2.
\end{align}
This Fisher information also implies the variance bound in Eq.~\eqref{eqn:ptebound}.
It may seem surprising that an optimal measurement can be one in which most qubits spend some of the measurement time idle. Since more time yields more signal, intuition suggests that the most effective strategy would make better measurements on the less-weighted qubits rather than keep them off for much of the measurement time. For example, by disentangling a qubit from the larger state halfway through the protocol, a separate measurement could be made on $\theta_1 + \frac{1}{2} \theta_2$ and $\frac{1}{2} \theta_2$, which appears to yield more information than just measuring the quantity of interest $\theta_1 + \frac{1}{2} \theta_2$.
This reasoning fails because there is no way to use information about $\theta_2$ to improve an estimate of $\theta_1 + \frac{1}{2} \theta_2$ without also knowing about $\theta_1$. Because we do not know about the individual parameters, only a measurement of the entire function is usable and our scheme is optimal in this case.
However, once we account for pre-existing knowledge about the parameter values (drawn from physically-motived estimates or less-precise previous measurements) our bound will instead apply in the regime of asymptotically many measurements $(M \gg 1)$ and in that setting our scheme will also saturate it \cite{Gill1995}. This is because the value of prior knowledge becomes increasingly low as we accumulate measurement data.
One advantage of this protocol is that an eavesdropper cannot learn the result of the network measurement by capturing a subset of the nodes' $\hat{\sigma}^x$ measurement results. This privacy can be shown by tracing out the first qubit in Eq.~\eqref{eqn:postptestate}, which leaves no phase information in the resulting mixed state.
The central node can receive the measurement outcomes from all other nodes but keep its own secret, and no eavesdropper is able to extract information from the broadcasted results. This is true even if the central node's qubit is unweighted (i.e., $\alpha_i = 0$), which follows simply from the properties of the GHZ state.
\subsubsection{Using Spin Squeezed States}
\label{sec:tdsss}
The perfect security of the GHZ state arises because obtaining the measurement result requires every qubit, but this also implies an extreme sensitivity to noise. This noise can be a serious problem for metrological applications \cite{Escher2011,Vidrighin2014}. Because the GHZ state decoheres faster than an individual qubit, the advantage provided by entanglement is nullified if the interrogation time of the qubit is limited by its coherence time \cite{Huelga1997}. However, in many settings the time spent on a single measurement will be much shorter than the decoherence time, for instance, to gather data on short timescales. In these cases, GHZ states still provide a metrological advantage. Note that dynamical decoupling \cite{Viola1999} or quantum error correction \cite{Kessler2014a,Lu2015} could be used to lengthen the effective decoherence time in some cases.
In other situations, however, it may be that decoherence is the dominant concern. In these situations, the best strategy uses a highly-symmetric entangled state which is more robust to noise than the GHZ state \cite{Huelga1997}. Under dephasing, these states can still offer a constant factor improvement over unentangled metrology. In this section, we show that spin-squeezed states can also function as inputs to the partial time evolution protocol, and so may be good candidates for a sensor network operating in a situation where decoherence limits the interrogation time. Squeezed states are collective spin states which, due to entanglement, have reduced variance along one axis of the collective Bloch sphere at the cost of increased variance along an orthogonal axis \cite{Kitagawa1993,Wineland1992}. Recently, it has been shown that these states may allow Heisenberg-scaling measurements even without single-particle detection, which makes them very attractive for experimental implementations \cite{Davis2016a}.
We consider a state whose overall spin vector is aligned along $+x$, such that $\expect{\hat{\sigma}^x_i} \approx 1$. We assume that the other spin components have zero expectation value, but that the variance of the collective spin projection $\hat{J}_y = \frac{1}{2} \sum_i \hat{\sigma}^y_i$ is decreased while the variance of $\hat{J}_z$ is increased. We quantify this effect through the spin-squeezing parameter $\xi$ \cite{Wineland1992},
\begin{equation}
\xi = \sqrt{ \frac{ \Var \hat{J}_y}{N / 4}}.
\end{equation}
Suppose that we perform Ramsey interferometry on such a state \cite{Wineland1992,Wineland1994}. The protocol includes both partial time evolution $\hat{U}(t)$ and a final rotation pulse $\hat{R}_x \left( \frac{\pi}{2} \right) = \exp \left( - i \frac{\pi}{4} \sum_i \hat{\sigma}^x_i \right)$.
A final measurement is made of the total spin projection $\hat{J}_z$ after applying these operations:
\begin{align}
\expect{\hat{J}_z (t)} &= \expect{ \hat{U}^\dag (t) \hat{R}_x^\dag \left( \frac{\pi}{2} \right) \hat{J}_z (0) \hat{R}_x \left( \frac{\pi}{2} \right) \hat{U} (t) }, \\
&= \frac{1}{2} \expect{ \sum_{i = 1}^N \hat{\sigma}_i^x \sin \alpha_i \theta_i t + \hat{\sigma}^y_i \cos \alpha_i \theta_i t }.
\label{eqn:jzpostevolve}
\end{align}
If we specify that this expectation is to be taken over a squeezed state with $\expect{\hat{\sigma}_i^x} \approx 1$ and $\expect{\hat{\sigma}_i^y} = 0$, then our signal will be sensitive only to $\bm{\alpha} \cdot \bm{\theta}$ if each individual phase is small:
\begin{equation}
\expect{\hat{J}_z (t)}_\mathrm{squeezed} \approx \frac{1}{2} \sum_{i = 1}^N \sin \alpha_i \theta_i t \approx \frac{t}{2} \sum_{i=1}^N \alpha_i \theta_i.
\end{equation}
This shows that a squeezed state can be used for measurements of linear functions. The sensitivity can then be calculated just as in Eq.~\eqref{eqn:ptesensitivity},
\begin{equation}
\Var Q = \left. \frac{ \Var \hat{J}_z(t)}{ \left( \partial \expect{\hat{J}_z (t)} / \partial q \right)^2} \right|_{q = 0} = \frac{ \Var \hat{J}_y }{t^2/4} = \frac{N \xi^2}{t^2} .
\label{eqn:ssssensitivity}
\end{equation}
We evaluate the sensitivity at $q = 0$ because we are interested in small signals. Partial time evolution with spin-squeezed input beats the standard quantum limit if $\xi \leq \| \bm{\alpha} \| / \sqrt{N}$. Note that there are $N$ components of $\bm{\alpha}$ and therefore $\| \bm{\alpha} \|$ will generally be of order $\sqrt{N}$ assuming that the moments of the field being measured are well distributed. Squeezed states can achieve squeezing proportional to $N^{-1/2}$ \cite{Wineland1992,Kitagawa1993}, which approaches the bound in Eq.~\eqref{eqn:ptebound} up to numerical prefactors not scaling with $N$.
Other highly-entangled states such as Dicke states also have metrological value in the presence of noise and could also serve as input states to partial time evolution with similarly favorable scaling \cite{Dicke1954,Holland1993a,Zhang2014,Campbell2009,Lucke2014}.
\subsection{Time-Independent Protocols}
In this section, we present two other possible measurement schemes for linear combinations of parameters. Both of these differ from the protocols of Sec.~\ref{sec:tdprotocols} because they prepare a particular state and then allow for free evolution during phase accumulation, rather than using pulses to evolve for an effective time of $\alpha_i t$ on qubit $i$. We will present time-independent schemes that begin with both a GHZ-like state and the spin-squeezed state. Note that these protocols rely on assumptions about the size of signals $\theta_i$ or the evolution time $t$.
\subsubsection{Using GHZ-like Input State}
We begin by defining a single-qubit state $\ket{\bm{\tau}}$, where $\bm{\tau}$ is a vector whose elements are $\tau_j = -1,0,1$:
\begin{equation}
\ket{\bm{\tau}} = \bigotimes_{j=1}^N \begin{cases} \ket{0} & \tau_j \neq -1 \\ \ket{1} & \tau_j = - 1 \end{cases}.
\end{equation}
We then define the entangled state $\ket{\psi(\bm{\tau})}$ as
\begin{equation}
\ket{\psi(\bm{\tau})} = \frac{1}{\sqrt{2}} \left( \ket{\bm{\tau}} + \ket{\bm{-\tau}} \right).
\end{equation}
This state can be understood as a general class that includes the GHZ state as the case $\tau_j = 1$ for all $j$. For every $\tau_j = -1$, spin $j$ is flipped relative to the GHZ state, while for every $\tau_j = 0$, spin $j$ is entirely disentangled.
In order to measure $\bm{\alpha} \cdot \bm{\theta}$, we will evolve $\ket{\psi(\bm{\tau})}$ under the Hamiltonian in Eq.~\eqref{eqn:hamiltonian} and then measure the following observable $\hat{\Pi}(\bm{\tau})$:
\begin{equation}
\hat{\Pi} ( \bm{\tau} ) = \bigotimes_{j} \left(\hat{\sigma}_x^j\right)^{\tau_j}.
\end{equation}
That is, we multiply the outcomes of the individual projective $\sigma^x$ measurements for each qubit which was originally entangled with the others ($\tau_j \neq 0$). It can be shown that probability distribution of this observable is
\begin{equation}
P\left( \hat{\Pi} = \pm 1 \middle| \bm{\tau}, \bm{\theta} \right) = \begin{cases} \cos^2 \left( \bm{\theta} \cdot \bm{\tau} t/2 \right) & 1, \\ \sin^2 \left( \bm{\theta} \cdot \bm{\tau} t/2 \right) & -1 \end{cases}.
\end{equation}
To create a final protocol, we will now randomize the choice of $\bm{\tau}$, which in turn means we will randomly select both the initial state and the final measurement. An overall sensitivity to $\bm{\alpha} \cdot \bm{\theta}$ can be realized if the probability distribution for every individual spin to be $\tau_j$ is given by:
\begin{equation}
P(\tau_j) = \begin{cases} \frac{ \alpha_j ( \alpha_j \pm 1) }{2} & \tau_j = \pm1 \\ 1 - \alpha_j^2 & \tau_j = 0 \end{cases}.
\end{equation}
By then summing over $P(\bm{\tau})$, we find that $\expect{\hat{\Pi}(\bm{\theta}, t)} = 1 - t^2 \left( \bm{\alpha} \cdot \bm{\theta} \right)^2$ to lowest order in $t$. Since $\hat{\Pi}^2 = 1$, we can use the same approach as Eq.~\eqref{eqn:ptesensitivity} to find that the sensitivity for this measurement is $\Var Q = 1/t^2$, leading to the same sensitivity as the time-independent protocol.
\subsubsection{Using Spin-Squeezed States}
To implement a time-independent protocol that makes use of a spin-squeezed input state, we will actually use a two-part measurement protocol. First we will derive a general expression that applies to both parts, and then show how they can be combined.
Much as in Sec.~\ref{sec:tdsss}, we will use the Heisenberg evolution of the total angular momentum along one axis to evaluate the final observable. We can begin with the result of Eq.~\eqref{eqn:jzpostevolve}, but with two alterations. First, rather than $\hat{U}$ representing a partial time evolution on each qubit, instead it will be the full time evolution operator $\hat{U} = \exp \left( -i t \sum \theta_i \hat{\sigma}^z_i \right)$. Second, we will add an additional operator at the beginning of the protocol, which we write as $\hat{Q}(\bm{\eta})$:
\begin{equation}
\hat{Q} \left( \bm{\eta} \right) = \bigotimes_{i=1}^N \hat{r}^i_z \left( \eta_i \right).
\end{equation}
Here, $\hat{r}^i_z$ is the single-qubit rotation about the $z$ axis. That is, we apply a qubit-dependent rotation about the $z$ axis before we begin the evolution. The final operator $\hat{J}_z (t)$ will be:
\begin{equation}
\hat{J}_z(t) = \hat{Q}^\dag(\bm{\eta}) \hat{U}^\dag (t) \hat{R}_x^\dag \left( \frac{\pi}{2} \right) \hat{J}_z (0) \hat{R}_x \left( \frac{\pi}{2} \right) \hat{U} (t) \hat{Q}(\bm{\eta}).
\end{equation}
The effect of $\hat{Q}$ is to add an additional phase to the evolution, meaning the final value for $\expect{\hat{J}_z(t)}$ can be found by substituting the angles $\theta_i t + \eta_i$ for $\alpha_i \theta_i t$ in Eq.~\eqref{eqn:jzpostevolve}. As a result, we find that the final expectation value is:
\begin{equation}
\expect{\hat{J}_z(t)} = \frac{1}{2} \expect{ \sum_{i=1}^N \hat{\sigma}^x_i \sin \left( \theta_i t + \eta_i \right) + \hat{\sigma}_i^y \cos \left( \theta_i t + \eta_i \right) }.
\end{equation}
By using the conditions that $\expect{\hat{\sigma}^x_i} \approx 1$ and $\expect{\hat{\sigma}^y_i} \approx 0$, we find that:
\begin{equation}
\expect{\hat{J}_z(t)} \approx \frac{1}{2} \sin \left( \theta_i t + \eta_i \right).
\end{equation}
Now we introduce a two-step protocol. In the first step, we perform this sequence (prepare a spin-squeezed state, add qubit-dependent rotations, evolve, measure $\hat{J}_z$) with $\eta_i = \phi_i$, where $\cos \phi_i = \alpha_i$. We will call the quantity measured $\hat{J}^{+}_z$. Then, we repeat the process with $\eta_i = - \phi_i$, and call the resulting quantity $\hat{J}^{-}_z$. The expectation value of the sum of these quantities is:
\begin{align}
\expect{\hat{J}^{+}_z + \hat{J}^{-}_z} &\approx \frac{1}{2} \sum_{i=1}^N \sin \left( \theta_i t + \phi_i \right) + \sin \left( \theta_i t - \phi_i \right) \\
&= \sum_{i=1}^N \cos \phi_i \sin \theta_i t \approx \sum_{i=1}^N \alpha_i \theta_i t.
\end{align}
Here, as in Sec.~\ref{sec:tdsss}, we have assumed that the phases to be detected, $\theta_i t$, are small enough to make the small-angle approximation.
In order to evaluate the sensitivity of this measurement, we look at the point of zero signal as in Eq.~\eqref{eqn:ssssensitivity}. At zero signal, $J^+_z + J^-_z$ gives $\sum \alpha_i \sigma_i^y$. It can be shown that $\Var \sum \alpha_i \sigma_i^y \leq 4 \Var J_y$, and so, by the same calculations used in Eq.~\eqref{eqn:ssssensitivity}, the variance is no more than $4 N \xi^2/ t^2$. Note, however, that this assumes that both $\hat{J}_+$ and $\hat{J}_-$ are measured for time $t$. For a fairer comparison, we can replace $t$ with $t/2$ so the time required for the two-step protocol is the same as for one time-dependent round. In this case, the sensitivity is no worse than $16 N \xi^2/ t^2$.
Interestingly, this two-step protocol requires only single-qubit operations once the initial squeezed state is created. This may make it a more tractable scheme for experimental realizations of quantum enhancements in measurements of linear combinations of parameters.
\section{Entanglement-enhanced molecular NMR}
Many applications of entangled sensor networks may emerge as distributed entanglement becomes easier to achieve. In this section we focus on an application which may be viable in the near future: nanoscale nuclear magnetic resonance (NMR) as a form of molecular microscopy. NMR has long been used to investigate the chemical composition of molecular structures and perform medical imaging \cite{Mansfield2004}. The spatial resolution of NMR had been limited to a few micrometers until the recent advent of nitrogen-vacancy (NV) center magnetometers \cite{Mamin2013,Rondin2014,DeVience2015}. These magnetometers are sensitive to nanotesla magnetic fields with spatial resolution on the nanometer scale and can be used to image molecules or single proteins deposited on a diamond layer with embedded NV centers \cite{Lovchinsky2016,Lazariev2015,Perunicic2016}.
Nanoscale NMR applications are a promising setting for entanglement-enhanced sensor networks.
The electronic spin associated with an NV center in diamond can be operated as a two-level system whose free evolution results in the accumulation of phase dependent on the local magnetic field \cite{DeVience2015}. Because NV centers are useful platforms for quantum information processing, entangling protocols already exist and have been demonstrated experimentally \cite{Dolde2013,Dolde2014,Zu2014,Bernien2013}. Our protocol is particularly useful for studies of chemical or magnetic dynamics, such as Ref.~\cite{Wolfe2016}, because the measurement timescale may be much shorter than the decoherence time of the GHZ state, making our noise-free treatment applicable.
Linear combinations of spatially separated field values are interesting measurement quantities in nanoscale NMR. Reference~\cite{Arai2015} describes an imaging protocol which combines many different Fourier spatial modes, and Ref.~\cite{Lazariev2015} similarly combines many signals to perform molecular microscopy. These measurements could be performed more accurately using entangled NV sensors. In addition, our entanglement scheme can perform simple subtraction of the signal between two qubits. This allows common mode noise subtraction between a sensor qubit and another qubit exposed only to environmental noise. In general, even if a full GHZ state of all sensors is not feasible, smaller clusters of entangled sensors can still enhance sensitivity.
Entanglement-enhanced imaging of objects larger than single molecules may also be a fruitful area of research.
An experiment detecting the firing of a single animal neuron with accuracy near the standard quantum limit has already been performed \cite{Jensen2016}, making exploration of techniques surpassing the limit a natural next step. Similar experiments could demonstrate an enhancement due to distributed entanglement in the near future.
\section{Outlook}
We have presented measurement protocols for quantum networks which are useful for measuring linear combinations of parameters and developed a Heisenberg limit for the optimal estimation of linear combinations. Our protocol can be considered a generalization of entanglement-enhanced Ramsey spectroscopy, as in Ref.~\cite{Bollinger1996}, to the measurement of spatially varying quantities. In the future, we hope to search for further protocols and to remove the requirements of small signal or evolution time where we have imposed them. We identified magnetometry in general and nanoscale NMR in particular as candidate applications of our protocol, but we wish to stress our protocol's significantly broader scope. In particular, we expect that our protocol will be useful for measuring spatially varying quantities in contexts such as gravimetry \cite{DeAngelis2008,Debs2011}, spectroscopy \cite{Dinani2016}, and rotation sensing \cite{Ragole2016,Gustavson1997,Kolkowitz2016}. Note there is also no requirement that the parameters measured in a linear combination be of the same physical source. For instance, a sensor network could measure a linear combination of both electric and magnetic fields.
In general, our protocol can be applied in any setting where Ramsey spectroscopy can be applied if the quantity of interest is nonlocal. In addition, recent work \cite{Layden2017} indicates that spatial correlations in measurements may be a useful tool for noise-filtering and error correction in quantum sensors.
Many schemes for quantum sensing rely on coherence in photonic, rather than atomic, degrees of freedom, such as spectroscopic microscopy \cite{Kee2004}. A recent manuscript, Ref.~\cite{Proctor2017} provides a general framework for treatment of sensor networks which is applicable to photonic systems and others.
We would like to thank P. Barberis Blostein, J. Borregaard, T. Brun, C. Caves, M. Cicerone, Z.-X. Gong, M. Hafezi, M. Oberthaler, J. Simon, V. Lekic, E. Polzik, K. Qian, and J. Ye for discussions. This work was supported by ARL CDQI, ARO MURI, NSF QIS, ARO, NSF PFC at JQI, and AFOSR. Z. E. is supported in part by the ARCS Foundation.
| c9465747b2e13e5d849a976f0225d33fe8d5a386 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
\label{sec-1}
``The devil is in the tails" is the title of the paper by Donnelly and
Embrechts (2010) who refute the harsh criticism of mathematics (Salmon,
2012) in general, and of the Gaussian copula-based credit risk model
of Li (2000) in particular. As a member of the unholy trinity (Kousky and
Cooke, 2009), the notion of tail dependence is in the very center of this
controversy. Speaking plainly, tail dependence is about the clustering of
extreme events, and it is a rather
daunting phenomenon in the context of risk management. For example, it implies
that devastating losses
within portfolios of risks as well as defaults of financial enterprises in
credit risk portfolios occur together (R\"{u}schendorf, 2013; Wang et al.,
2013; Puccetti and R\"{u}schendorf, 2014).
Mathematically, there exist a variety of ways to quantify the extent of tail dependence
in bivariate random vectors with dependence structures given by copula functions
$C:[0,\ 1]^2\rightarrow [0,\ 1]$ (see, e.g., Nelsen, 2006; Durante and Sempi, 2015, for reviews of the
theory of copulas).
Arguably the most popular
measure of lower tail dependence is nowadays attributed to Joe (1993) (also
Sibuya, 1959) and given by
\begin{eqnarray}
\label{lambda}
\lambda_L:=\lambda_L(C)= \lim_{u \downarrow 0} {C(u,u)\over u}.
\end{eqnarray}
Non-zero (more precisely $(0,\ 1]$) values of index (\ref{lambda})
suggest lower tail dependence in $C$. Just like other synthetic measures, $\lambda_L$ is not always
reliable because it sometimes underestimates the extent of lower tail dependence in copulas
as the next example demonstrates.
\begin{example}
\rm
\label{ExC1}
Consider the following copula (Nelsen, 2006, Example 3.3)
\[
C_{\theta}(u,\ v)=\left\{
\begin{array}{ll}
u, & 0\leq u\leq \theta v\leq \theta, \\
\theta v, & 0\leq \theta v< u < 1- (1-\theta)v, \\
u+v - 1, & \theta \leq 1- (1-\theta) v\leq u\leq 1,
\end{array}
\right.
\]
parametrized by $\theta \in [0,1]$. This copula has two singularities, and it is fully {co-monotonic} (fully {counter-monotonic})
for $\theta=1$ ($\theta=0$, respectively). It is easy to see that
\[
\lambda_L(C_\theta)=\lim_{u\downarrow 0}\frac{\theta u}{u}=\theta.
\]
Let $\lambda_L^\ast(C_{\theta})$ be the measure
as in (\ref{lambda}) but now along the path $\left(u\sqrt{\theta},\ u/\sqrt{\theta}\right)_{0\leq u\leq 1}$ rather than along the diagonal $(u,u)$. Clearly in this case
\[
\lambda_L^\ast(C_\theta)=\lim_{u \downarrow 0}\frac{C(u\sqrt{\theta},u/\sqrt{\theta})}{u}=
\sqrt{\theta}>\theta=\lambda_L(C_\theta),
\]
for every $\theta\in(0,\ 1)$.
\end{example}
On a different note,
when limit (\ref{lambda}) is zero, it is often useful to rely on the somewhat more
delicate index of weak tail dependence $\chi_L\in[-1,\ 1]$ (Coles et al., 1999; Fischer and Klein, 2007)
that is given by
\begin{equation}\label{ind-cl}
\chi_{L}:=\chi_L(C)=\lim_{u\downarrow 0}\frac{2\log u }{\log C(u,u)}-1 ,
\end{equation}
and/or to the index of tail dependence $\kappa_L:=\kappa_L(C)\in[1,\ 2]$ (Ledford and Tawn, 1996) that solves the equation
\begin{equation}\label{ind-kl}
C(u,u)= \ell(u) u^{\kappa_L} \quad \textrm{when} \quad u\downarrow 0,
\end{equation}
assuming that we can find a slowly varying at $0$ function $\ell(u)$.
The following example demonstrates that $\kappa_L$ can also be misleading in a similar way to that of Example \ref{ExC1}.
\begin{example}
\label{ExMO1}
\rm
Recall that the Marshall-Olkin copula is given by
\begin{equation}\label{copula-mo}
C_{a,b}(u,v)=\min(u^{1-a}v,uv^{1-b})
\quad \textrm{for} \quad 0\le u,v\le 1,
\end{equation}
where $a,b\in [0,1]$ are parameters (Cherubini et al., 2013). Denote by $\kappa_L^\ast(C_{a,b})$ a measure
{mimicking} (\ref{ind-kl}) that verifies the tail dependence of $C_{a,b}$ along the path
$\left(u^{2a/(a+b)},u^{2b/(a+b}\right)_{0\leq u\leq 1}$. We readily check that
\[
\kappa_L^\ast(C_{a,b})=2-\frac{2ba}{a+b}\leq 2-\min(a,b)=\kappa_L(C_{a,b}),
\]
where the equality holds only if $a=b$.
\end{example}
Speaking generally, indices (\ref{lambda}), (\ref{ind-cl}) and (\ref{ind-kl}) may underestimate the amount of tail dependence even
in copulas that are symmetric and do not have singularities (Furman et al., 2015).
The reason is that all the aforementioned indices of lower tail dependence rely entirely on the
behavior of copulas along their main diagonal
$(u,\ u)_{0\leq u\leq 1}$. However,
the tail dependence of copulas can be substantially stronger along the paths other than
the main diagonal. This can be a serious disadvantage, as reported by, e.g., Schmid and Schmidt (2007), Zhang (2008), Li et al. (2014), and Furman et al. (2015).
When it comes to the bivariate Gaussian copula, henceforth denoted by $C_{\rho}$, which has become a synonym of the recent
subprime mortgage crisis, it can be shown that indices (\ref{lambda}), (\ref{ind-cl}) and
(\ref{ind-kl})
are equal to $0,\ \rho$ and $2/(1+\rho)$, respectively, where $\rho\in(0,\ 1)$ is the
correlation coefficient.
In the light of discussion hitherto, the following most natural problem
arises:
\begin{problem}
\label{pr}
Let $\psi,\varphi:[0,\ 1]\rightarrow [0,\ 1]$ be functions yielding an admissible path
$(\psi(u),\ \varphi(u))_{0\leq u\leq 1}$ in $[0,\ 1]^2$, and let $\lambda_L^\ast$,
$\chi_L^\ast$ and $\kappa_L^\ast$ be the counterparts of (\ref{lambda}), (\ref{ind-cl}) and
(\ref{ind-kl}), respectively, calculated along the noted path. Is there an admissible path such that
any of the following
bounds holds
\begin{equation}
\label{pr-lim}
\lambda_L^\ast(C_\rho)>\lambda_L(C_\rho),\textnormal{ }
\chi_L^\ast(C_\rho)>\chi_L(C_\rho) \textnormal{ and/or }
\kappa_L^\ast(C_\rho)<\kappa_L(C_\rho)?
\end{equation}
\end{problem}
\noindent
A positive answer to this question would reinstate to an extent the Gaussian copula in
public favor, whereas a negative answer would mean that index (\ref{lambda}) is
maximal in the Gaussian case, which of course does not imply the same conclusion for
other copulas.
In this paper we investigate the aforementioned problem. To this end, in Section \ref{sec-2} we set out to formally define the class of `admissible' path functions as well as the collection of
`admissible' paths mentioned in Problem \ref{pr}.
In Section \ref{sec-3}, we then provide a complete solution to Problem \ref{pr}. Our
proofs rely on subtle geometric arguments involving intersections of convex curves with
their rotations. Section \ref{sec-4} concludes the
paper.
\section{Paths and indices of maximal tail dependence}
\label{sec-2}
Our main goal in this section is to describe admissible paths $(\psi(u),\ \varphi(u))_{0\leq u\leq 1}$ formally.
We borrow heavily from Furman et al. (2015).
\begin{definition}
\label{paths}
A function $\varphi: [0,1] \to [0,1]$ is called {\it admissible} if it satisfies the following conditions:
\begin{enumerate}[\rm (1)]
\item [(C1)]
$\varphi(u) \in [u^2,1] $ for every $u\in [0,1] $;
\item [(C2)]
$\varphi(u)$ and $u^2/\varphi(u) $ converge to $0$ when $u\downarrow 0$.
\end{enumerate}
Then the path $(\varphi(u),u^2/\varphi(u))_{0\le u \le 1}$ is admissible whenever the function $\varphi $ is admissible. Also, we denote by $\mathcal{A}$ the set of all admissible
functions $\varphi$.
\end{definition}
A number of observations are instrumental to clarify the definition. First,
condition (C1) makes sure that $\varphi(u)\in[0,\ 1]$ and
$u^2/\varphi(u) \in [0,\ 1]$, whereas condition (C2) is motivated by the fact that we are
interested in the behavior of the copula $C$
near the lower-left vertex of its domain of definition. Second, the function $\varphi_0(u)=u,\ u\in[0,\ 1]$, is admissible and yields the main
diagonal $(u,\ u)_{0\leq u\leq 1}$. Third, for the independence
copula {$C^{\perp}$}, it holds that $C^{\perp}(\varphi(u),u^2/\varphi(u))=u^2,\ u\in[0,\ 1]^2$,
which is path-independent as expected, thus warranting the choice
$\psi(u)=u^2/\varphi(u),\ u\in [0,\ 1]$.
In order to determine the strongest extreme co-movements of risks
for any copula $C$, we search for functions $\varphi\in\mathcal{A}$ that maximize the probability
\[
\Pi_{\varphi }(u) =C\big (\varphi(u),u^2/\varphi(u)\big ), \;\;\; u \in (0,1),
\]
or, equivalently, the function
\[
d_\varphi(C,C^\perp)(u)=C\big (\varphi(u),u^2/\varphi(u)\big )-
C^\perp (\varphi(u),u^2/\varphi(u)\big ), \;\;\; u \in (0,1),
\]
which is non-negative for positively quadrant dependent (PQD) (Lehmann, 1966)
copulas $C$.
Then an admissible function $\varphi^* \in \cal{A} $ is called {\it a function of maximal dependence} if
\begin{equation}
\label{Pi-m}
\Pi_{\varphi^* }(u)=\max_{\varphi \in \cal{A} }\Pi_{\varphi }(u)
\end{equation}
for all $u\in (0,1)$. The corresponding admissible path
$(\varphi^*(u),u^2/\varphi^*(u))_{0\leq u\leq 1}$ is called
{\it a path of maximal dependence}.
Generally speaking, the path $\varphi^\ast$ is not unique, but for each such path the value of $\Pi_{\varphi^\ast}$ is the same. In what follows, we use the notation $\Pi^*(u)$ instead of $\Pi_{\varphi^* }(u)$.
Given the new paradigm of prudence that has taken the world of quantitative risk management by storm (OSFI, 2015), it is sensible to introduce conservative variants of indices
(\ref{lambda}), (\ref{ind-cl}) and (\ref{ind-kl}) that would rely on path of maximal dependence (\ref{Pi-m}), rather than on the main
diagonal path of the copula $C$. Namely, we suggest
\begin{equation}
\lambda_L^*:=\lambda_L^*(C)= \lim_{u \downarrow 0} {\Pi^*(u)\over u}
\textnormal{ instead of }
\lambda_{L}(C)=\lim_{u \downarrow 0} \frac{C(u,u)}{u},
\end{equation}
and
\begin{equation*}
\label{chai}
\chi_L^*:=\chi_L^*(C)=\lim_{u\downarrow 0}\frac{2\log u }{\log \Pi^*(u)}-1
\textnormal{ instead of }
\chi_L=\lim_{u\downarrow 0}\frac{2\log u }{\log C(u,\ u)}-1,
\end{equation*}
subject to the existence of the limits, and also
\begin{equation*}
\label{mo-1b}
\Pi^*(u)=\ell^*(u) u^{\kappa_L^*},\ u\downarrow 0
\textnormal{ instead of }
\Pi(u)=\ell(u) u^{\kappa_L},\ \quad u\downarrow 0,
\end{equation*}
assuming that there exist slowly varying at zero functions $\ell^\ast(u)$ and $\ell(u)$
(Ledford and Tawn, 1996).
These new indices of tail dependence provide a more prudent
estimation of the extent of tail dependence in copulas and, in conjunction with tail-based
risk measures, are capable of distinguishing
between risky positions in situations where the classical indices of tail
dependence fail to do so
(Furman et al., 2015, Section 3).
A useful technique for deriving function(s) of maximal dependence, and thus in turn of the corresponding
indices, consists of three
steps:
\begin{enumerate}
\item [(S1)] searching for critical points of the function $x \mapsto C(x,u^2/x)$ over the interval $[u^2,1]$ and for each $u\in [0,\ 1]$;
\item [(S2)] checking which solution(s) is/are global maximum/maxima;
\item [(S3)] verifying that the function $u\mapsto \varphi^*(u)$ is in $\mathcal{A}$.
\end{enumerate}
Accomplishing these tasks sometimes results in explicit formulas for maximal dependence functions, while in
some other cases obtaining closed-form solutions may not be possible. {For example, as we see from Furman et al. (2015), for the Farlie-Gumbel-Morgenstern (FGM) copula this task is doable, whereas for the generalized Clayton copula there is no closed-form solution.
We refer the reader to, respectively, Sections 4 and 6 in Furman et al. (2015) for more technical discussions.}
Sometimes, especially when formulas for conditional copulas
are readily available, it is useful to recall that partial derivatives of copulas are
conditional copulas, and thus the task of determining the set of critical points becomes
equivalent to
finding all the solutions in $x\in[u^2,\ 1]$ to the equation
\begin{equation}
\label{cond_c}
xC_{2|1}\left(\frac{u^2}{x}|x\right)= \frac{u^2}{x}C_{1|2}\left(x|\frac{u^2}{x}\right).
\end{equation}
{Interestingly, as has been also pointed out by one of the referees, there
are symmetric copulas whose paths of maximal dependence are diagonal:
e.g., the FGM and Clayton copulas, as well as the symmetric subclass of
the Marshall-Olkin copulas (see, Furman et al., 2015). Hence, a
natural question is whether or not the path of maximal dependence always
coincides with the diagonal when the copula function is symmetric?
Unfortunately, the answer to this question is not always positive, as we
illustrate next.
First, we recall from Furman et al. (2015) that there are symmetric copulas whose paths of maximal dependence are not diagonal, such as the $0.5/0.5$ mixture of two `mirrored' Marshall-Olkin copulas. Next we present an example showing that this argument also holds for the absolutely continuous subclass of symmetric copulas.
For this, we recall the bivariate extreme value copula (Pickands, 1981; Guillotte and Perron, 2016)
\[
C_A(u,v)=\exp\left\{\ln(uv){A\left(\frac{\ln v}{\ln uv} \right)}\right\},
\]
where $A:[0,1]\rightarrow [0.5,1]$ is the Pickands dependence function, which is convex and satisfies the bounds $(1-t)\vee t \leq A(t) \leq 1$ for $t\in [0,1]$. Define the $0.5/0.5$ mixture of two `mirrored' extreme value copulas by
\begin{eqnarray}
\label{mix-extreme-c}
C_{A_1,A_2}(u,v)=\frac{1}{2}\big(C_{A_1}(u,v)+C_{A_2}(u,v) \big),
\end{eqnarray}
where $A_1$ and $A_2$ are two Pickands dependence functions such that $A_1(t)=A_2(1-t)$ for $t\in [0,1]$. It is not difficult to see that copula (\ref{mix-extreme-c}) is PQD, symmetric around the diagonal, and absolutely continuous when the Pickands dependence functions $A_1$ and $A_2$ are differentiable.
To prove that the path of maximal dependence for the just defined copula may not be diagonal, it is sufficient to show that there exists $A_1$ such that
\begin{eqnarray}
\label{ineqn-extreme-c}
\frac{\partial^2}{\partial x^2}C_{A_1,A_2}(x,u^2/x)\Big|_{x=u} > 0.
\end{eqnarray}
Equivalently, we need to verify that for $\psi(t,u)=u^{A_1(1-t)}+u^{A_1(t)}$ and $u\in [0,\ 1]$ we have
\begin{eqnarray}
\frac{\partial^2}{\partial t^2} \psi(t,u)\big|_{t=0.5}
=2 u^{A_1(1/2)}\ln (u)\left(
(A_1^{(1)}(1/2))^2\ln u+A_1^{(2)}(1/2)
\right)
>0,
\label{deriv-3}
\end{eqnarray}
where $A_1^{(k)}$ is the $k$-th derivative of $A_1$. Hence, unless the function $A_1$ attains its minimum at $t=0.5$, there exists $u^\ast\in[0,\ 1]$ such that statement (\ref{deriv-3}) holds for all $u\in[0,\ u^\ast]$.
This suggests that $x\mapsto C(x,\ u^2/x)$ is convex at $x=u$ for $u\in[0,\ u^\ast]$, and so the path of maximal dependence
cannot coincide with the diagonal on the aforementioned
interval, that is, $\varphi^\ast(u)\neq u$ for $u\in[0,\ u^\ast]$.
We conclude this section by noting that other scholars have also
considered other than the diagonal paths when measuring tail dependence. For
example, Asimit et al. (2016) use the conditional Kendall's tau to measure
tail dependence. Joe et al. (2010) introduce the tail dependence function $b(w_1,w_2;C)=\lim_{u\downarrow 0}C(uw_1,uw_2)/u$ for $w_1,w_2>0$ to measure tail dependence via different directions. Hua and Joe (2014) use the excess-of-loss economic
pricing functional to study tail dependence. All of these measures as well
as the notion of maximal tail dependent discussed in the present paper
provide complementary ways for understanding tail dependence.
}
\section{Main results}
\label{sec-3}
The bivariate Gaussian copula arises from the bivariate normal distribution. As such,
it is arguably the most popular and well-studied copula, extensively used in financial and
insurance mathematics (MacKenzie and Spears, 2014).
We recall that the Gaussian copula $C_{\rho}(u,v)$ is defined, for $0\leq u,v\leq 1$,
as follows
\begin{equation}\label{def_CG}
C_{\rho}(u,v)=\Phi_2(\Phi^{-1}(u),\Phi^{-1}(v);\rho),
\end{equation}
where $\Phi(u)$ and $\Phi^{-1}(u)$ are the standard-normal distribution function and its inverse, and
\begin{equation}\label{def_Phi2}
\Phi_2(s,t;\rho)=\int_{-\infty}^{s} \int_{-\infty}^{t} \frac{1}{{2\pi}\sqrt{1-\rho^2}}\exp \left \{-\frac{x^2-2\rho x y +y^2}{2(1-\rho^2)}\right \}dy dx
\end{equation}
is the distribution function of the bivariate normal distribution with correlation parameter $\rho\in (-1,1)$, defined for all $s,t\in\mathbb{R}$.
\begin{theorem}\label{th}
For the Gaussian copula $C_{\rho}$,
\begin{enumerate}[\rm (I)]
\item \label{th-a}
when $\rho \in (-1,0)$, there is no admissible path of maximal dependence;
\item \label{th-b}
when $\rho=0$, every admissible path is a path of maximal dependence;
\item \label{th-c}
when $\rho \in (0,1)$, the only path of maximal dependence is the diagonal $(u,u)_{0\leq u \leq 1}$.
\end{enumerate}
\end{theorem}
\begin{corollary}\label{cor}
For the Gaussian copula $C_{\rho}$, when $\rho \in [0,1)$ we have
\begin{enumerate}[\rm (A)]
\item $\lambda_L^*(C_{\rho})=\lambda_L(C_{\rho})=0$;
\item $\chi_L^*(C_{\rho})=\chi_L(C_{\rho})=\rho$;
\item $\kappa_L^*(C_{\rho})=\kappa_L(C_{\rho})=2/(1+\rho)$.
\end{enumerate}
\end{corollary}
\begin{proof}[Proof of Theorem \ref{th}]
The proof of parts (\ref{th-a}) and (\ref{th-b}) of Theorem \ref{th} is simple. When $\rho \in (-1,0)$ we have $C_{\rho}(u,v)< uv$ for all $u,v\in (0,1)$. Therefore, $\Pi_{\varphi }(u)$ achieves its maximum at either $\varphi(u)=u^2$ or $\varphi(u)=1$. However, the two paths $(u^2,1)_{0\leq u \leq 1}$ and $(1,u^2)_{0 \leq u \leq 1}$ are not admissible, which establishes statement (\ref{th-a}). Statement (\ref{th-b}) follows from the fact that in the case $\rho=0$ the Gaussian copula $C_{\rho} $ reduces to the independence copula $C^\perp(u,\ v)=uv,\ 0\leq u,v,\leq 1$.
The proof of part (\ref{th-c}) of Theorem \ref{th} is much more involved, and it requires several auxiliary results.
Our first goal is to rephrase the statement about the location of paths of maximal dependence as a geometric statement about certain curves. To this end, for $\alpha \in (0,1)$, we define
\begin{equation}\label{def_C_alpha}
{\cal{C}}_{\alpha}:=\left \{(w,z) \;:\; \Phi(w)\Phi(z)=\alpha \right \} \subset {\mathbb R}^2.
\end{equation}
These are the level sets of the function of two variables $(w,z) \mapsto \Phi(w) \Phi(z)$, and these sets play a pivotal role in our proof.
Given a point $(x_1,x_2) \in {\mathbb R}^2$, we denote by ${}^{\beta}(x_1,x_2)$ the point $(y_1,y_2)$ obtained from $(x_1,x_2)$ by rotation by angle $\beta$ counter-clockwise, that is
\begin{align*}
y_1&=\cos(\beta)x_1 - \sin(\beta) x_2, \\
y_2&=\sin(\beta)x_1 +\cos(\beta) x_2.
\end{align*}
Similarly, for any set $\gamma \subset {\mathbb R}^2$ we denote by ${}^{\beta} \gamma$ the result of rotating the set $\gamma$ counter-clockwise by angle $\beta$.
Consider the following statement:
\begin{align}\label{condition}
&{\textnormal{ \it For any $\alpha \in (0,1)$ and $\beta \in (0,\pi/2)$, the intersection
${\cal{C}}_{\alpha}\cap {^{\beta}}{\cal{C}}_{\alpha}$}}\\ \nonumber
&\qquad \qquad\qquad \;\;{\textnormal{\it consists of a unique point.} }
\end{align}
We next prove that the statement above is in fact stronger than statement (\ref{th-c}) of Theorem
\ref{th}. Thus proving \eqref{condition} automatically completes the proof of Theorem \ref{th}(\ref{th-c}).
\begin{lemma}\label{lemma1}
Statement \eqref{condition} implies part (\ref{th-c}) of Theorem \ref{th}.
\end{lemma}
\begin{proof}
Finding a path of maximal dependence is equivalent to solving the following optimization problem: For every (fixed) $u \in (0,1)$ we want to find the maximum of the function
$[u^2,1]\ni x \mapsto C_{\rho}(x,u^2/x)$. First of all, note that the restriction
$\rho \in (0,1)$ implies $C_{\rho}(u,v)> uv$ for all $u,v\in (0,1)$. From this result we see that
$C_{\rho}(x,u^2/x)>u^2$
for all $x\in (u^2,1)$, and since $C_{\rho}(x,u^2/x)=u^2$ for $x=u^2$ or $x=1$,
we conclude that the function $x \mapsto C_{\rho}(x,u^2/x)$ achieves the
global maximum for some $\tilde x$ in the open interval $(u^2,1)$.
Since the function $x \mapsto C_{\rho}(x,u^2/x)$ is smooth
in the interval $(u^2,1)$, the global maximum must be one of its critical points, so that
$\frac{\d}{\d x} C_{\rho}(x,u^2/x)=0$ at the maximum point $x=\tilde x$.
To find the critical points, we solve the following equation:
\begin{equation}\label{eqn_hprime_1}
x\Phi\left(\frac{1}{\sqrt{1-\rho^2}}(\Phi^{-1}(u^2/x)-\rho\Phi^{-1}(x) ) \right)
=\frac{u^2}{x}\Phi\left(\frac{1}{\sqrt{1-\rho^2}}
(\Phi^{-1}(x)-\rho\Phi^{-1}(u^2/x)) \right)
\end{equation}
in $x\in(u^2,\ 1)$ for $u\in (0,\ 1)$, which was derived by using equation (3.1) and (3.2) in Meyer (2013) (also Fung and Seneta, 2011; McNeil et al., 2005) as well as \eqref{cond_c}, \eqref{def_CG} and \eqref{def_Phi2} in the current paper.
Note that $x=u$ is clearly a solution of equation (\ref{eqn_hprime_1}). Thus to establish
our main result, it is enough to prove that for every fixed $u \in (0,1)$ there are no other
solutions to \eqref{eqn_hprime_1} except for $x=u$.
Also, we introduce the following change of variables: let $\Lambda$ be the map that
sends any point $(u,x)$ from the domain
\begin{equation*}
D=\{ 0 < u < 1, \; \; u^2 < x < 1\} \subset {\mathbb R}^2
\end{equation*}
into another point $\Lambda (u,x)=(w,z) \in {\mathbb R}^2$ according to the rule:
\begin{equation}\label{def_w_z}
w = \Phi^{-1}(x) \;\; \textnormal{ and } \;\; z = \frac{1}{\sqrt{1-\rho^2}}\Phi^{-1}(u^2/x)-\frac{\rho}{\sqrt{1-\rho^2}}\Phi^{-1}(x).
\end{equation}
It is easy to see that $\Lambda$ is a diffeomorphism of $D$ onto ${\mathbb R}^2$, with the inverse map
$(u,x)=\Lambda^{-1} (w,z)$ given by
\begin{equation}\label{inverse_Lambda}
u=\left[\Phi(w) \Phi(\rho w + \sqrt{1-\rho^2} z )\right]^{1/2}, \;\;\; x=\Phi(w).
\end{equation}
Using (\ref{def_w_z}) it is easy to check that the diagonal $(u,u)$ of the set $D$ is mapped onto the straight line
$$
l:=\left\{ (z,w) \in {\mathbb R}^2 \; : \; z = \frac{1-\rho}{\sqrt{1-\rho^2}} w \right\}.
$$
>From \eqref{inverse_Lambda} we find
$$
u^2/x=\Phi(\rho w + \sqrt{1-\rho^2} z)
$$
and
$$
\frac{1}{\sqrt{1-\rho^2}}
(\Phi^{-1}(x)-\rho\Phi^{-1}(u^2/x))=\sqrt{1-\rho^2} w - \rho z.
$$
Combining the above formulas we conclude that equation \eqref{eqn_hprime_1} is equivalent to
$$
\Phi(w) \Phi(z)=\Phi(\sqrt{1-\rho^2} w - \rho z) \Phi(\rho w + \sqrt{1-\rho^2} z).
$$
Finally, we denote $\beta=\arcsin(\rho)$, so that $\rho=\sin(\beta)$ and $\sqrt{1-\rho^2}=\cos(\beta)$,
and rewrite the above equation as
\begin{equation}\label{eqn_phiz_phiw}
\Phi(w)\Phi(z)=\Phi(w') \Phi(z'),
\end{equation}
where $w'=\cos(\beta)w-\sin(\beta) z$ and $z'=\sin(\beta) w + \cos(\beta) z$. Note that
$\beta \in (0,\pi/2)$ and
the point $(w',z')$ is obtained from the point $(w,z)$ by rotation by the angle $\beta$ counter-clockwise. Using our previous notation we can write $(w',z')={}^{\beta} (w,z)$.
To summarize, we have shown that after a change of variables $\Lambda: D \mapsto {\mathbb R}^2$ equation \eqref{eqn_hprime_1} is equivalently transformed into equation \eqref{eqn_phiz_phiw}. The latter equation has a simple geometric interpretation: a point $(w,z)$ satisfies \eqref{eqn_phiz_phiw} if and only if $(w,z) \in {\cal C}_{\alpha} \cap {}^{\beta}{\cal C}_{\alpha}$, where $\alpha=\Phi(z)\Phi(w)$ and
${\cal C}_{\alpha}$ is the level set defined in \eqref{def_C_alpha}. Next, all points on the straight line $l$ defined above are the solutions
of \eqref{eqn_phiz_phiw}; this is easy to check directly, and it also follows at once from the fact that the points on $l$ are the images $(w,z)=\Lambda(u,u)$ of points on the diagonal
of $D$, which do satisfy the equivalent equation \eqref{eqn_hprime_1}. Therefore, we have shown that for every
$\alpha \in (0,1)$ there is {\it at least one} point in the intersection ${\cal C}_{\alpha} \cap {}^{\beta}{\cal C}_{\alpha}$. If we assume
that for every $\alpha \in (0,1)$ there exists a {\it unique} point in the intersection ${\cal C}_{\alpha} \cap {}^{\beta}{\cal C}_{\alpha}$,
this would imply that there are no other solutions to \eqref{eqn_phiz_phiw} except for those on the line $l$,
which in turn implies that the diagonal points $(u,u) \in D$ are the only solutions to \eqref{eqn_hprime_1}. Recall that the solutions to
equation \eqref{eqn_hprime_1} give us the critical points of the function $x \mapsto C_{\rho}(x,u^2/x)$. Therefore, if there are no other critical points except for the diagonal ones, then the path of maximal dependence must be diagonal.
\end{proof}
As we demonstrate next, each level set ${\cal C}_{\alpha},\ \alpha\in(0,\ 1)$ is in fact a smooth curve in ${\mathbb R}^2$ and a boundary of a convex set. Some of these curves are shown in Figure \ref{fig1}.
\begin{figure}
\centering
\includegraphics[height =7cm]{p1}
\caption{The curve ${\cal{C}}_{\alpha}$ for $\alpha = 1/10$ (blue), 1/4 (red), 1/3 (black), and 3/5 (green).}
\label{fig1}
\end{figure}
\begin{lemma}\label{lemma_C_properties} For every $\alpha \in (0,\ 1)$, the following properties hold:
\begin{enumerate}
\item [\textnormal{(P1)}] \label{prop-a}
${\cal{C}}_{\alpha}$ is a smooth curve in ${\mathbb R}^2$;
\item [\textnormal{(P2)}] \label{prop-c}
${\cal{C}}_{\alpha}$ is symmetric with respect to the diagonal line $\{(w,z) : w=z\}\subset {\mathbb R}^2$;
\item [\textnormal{(P3)}] \label{prop-b}
${\cal{C}}_{\alpha}$ is the boundary of a convex set which lies in
$\{w > \Phi^{-1}(\alpha)\} \cap \{z > \Phi^{-1}(\alpha)\}$. The lines
$z=\Phi^{-1}(\alpha)$ and $w=\Phi^{-1}(\alpha)$ are the asymptotes of ${\cal C}_{\alpha}$.
\end{enumerate}
\end{lemma}
\begin{proof}
First of all, we note that the equation $\Phi(w)\Phi(z)=\alpha$ can be solved for $z$ in terms of $w$ as follows
\begin{equation}\label{fn_z_w}
z(w)=\Phi^{-1}(\alpha/\Phi(w)).
\end{equation}
The level set ${\cal C}_{\alpha}$ is simply the graph of $z(w)$, which is clearly a
smooth function defined for $w>\Phi^{-1}(\alpha)$. This proves property (P1). The symmetry with respect to interchanging $w \leftrightarrow z$ follows at once from
definition \eqref{def_C_alpha}. This proves property (P2).
Since
$$
z(w) \to \Phi^{-1} (\alpha),\ {\text{for}}\ w\to +\infty,
$$
the horizontal line $z=\Phi^{-1}(\alpha)$ is an asymptote, and the vertical asymptote $w=\Phi^{-1}(\alpha)$ follows from the above-mentioned symmetry with respect to $z\leftrightarrow w$.
The function $z(w)$ given by \eqref{fn_z_w} is convex. An easy way to prove this is via the fact that
$\Phi(w)$ is log-concave, which implies that the function of two variables $(w,z) \mapsto \ln(\Phi(z)\Phi(w))$ is concave. Therefore, its upper set
\begin{equation}\label{def_U_alpha}
{\cal U}_{\alpha}:=\{ (w,z) \in {\mathbb R}^2 \; : \; \ln(\Phi(z)\Phi(w))>\ln(\alpha)\}
\end{equation}
must be convex. It is clear that ${\cal C}_{\alpha}$ is the boundary of ${\cal U}_{\alpha}$. This proves property (P3) and completes the proof of Lemma \ref{lemma_C_properties}.
\end{proof}
We recall that for any set $\gamma \subset {\mathbb R}^2$ we denote by ${}^{\beta} \gamma$ the rotation
of $\gamma$ by angle $\beta$ counter-clockwise. In particular, if $\gamma$ is a curve which can be written in polar coordinates $(r, \theta )$ as
\begin{equation}\label{curve-0}
\gamma=\left \{ (r(\theta)\cos(\theta), r(\theta) \sin(\theta)) \; : \; \theta \in (\theta_1, \theta_2) \right \} ,
\end{equation}
then ${^{\beta}}\gamma$ is also a curve whose expression in polar coordinates is given by
\begin{equation}\label{curve-1}
{^{\beta}}\gamma=\left \{(r(\theta-\beta)\cos(\theta), r(\theta-\beta) \sin(\theta)) \; : \; \theta \in (\theta_1+\beta, \theta_2+\beta) \right \}.
\end{equation}
\begin{figure}
\centering
\includegraphics[height =7cm]{p3}
\caption{Illustration to the proof of Lemma \ref{lemma_inter-point}(i):
the curves $\gamma$ (red) and ${}^{\beta}\gamma$ (blue) in polar coordinates ($\theta$ is on x-axis, and $r$ is on y-axis).}
\label{fig2}
\end{figure}
\begin{lemma}\label{lemma_inter-point}
${}$
\begin{itemize}
\item[\textnormal{(i)}]
Assume that a curve $\gamma $ is written in polar coordinates in form (\ref{curve-0}). If the function $r(\theta)$ is strictly decreasing on $(\theta_1,\tilde \theta)$ and strictly increasing on $(\tilde \theta, \theta_2)$ for some $\tilde \theta \in (\theta_1, \theta_2)$ then for every $\beta \in (0,2\pi-\theta_2+\theta_1)$ there is at most one point of intersection of $\gamma$ and ${}^{\beta}\gamma$.
\item[\textnormal{(ii)}] Assume that a curve $\gamma$ is given in the parametric form $(w(t),z(t))$, $t\in I$, where $I\subset {\mathbb R}$ is an interval. If
$r(t)=\sqrt{w(t)^2+z(t)^2}$ is nonzero and strictly monotone for $t\in I$, then for any $\beta \in (0,2\pi)$ the curves
$\gamma$ and ${}^{\beta} \gamma$ do not intersect.
\end{itemize}
\end{lemma}
\begin{proof}
While the proof of part (i) is quite obvious from Figure \ref{fig2}, we present the details of the proof for mathematical rigour. When $\theta_1+\beta>\theta_2$, then it is clear that ${^{\beta}}\gamma \cap \gamma=\emptyset$ because these curves lie in non-intersecting sectors.
We are left with the case when $\theta_1+\beta<\theta_2$. We further restrict ourselves to the case $\theta_1+\beta<\tilde{\theta}$; the argument in the case $\theta_1+\beta \in [\tilde \theta, \theta_2)$ is identical. Denote $r_1(\theta)=r(\theta)$ and $r_2(\theta)=r(\theta-\beta)$.
Since $r_1(\theta)$ is decreasing on $(\theta_1,\tilde \theta)$ we have
$r_2(\theta)>r_1(\theta)$ for $\theta \in (\theta_1+\beta,\tilde \theta)$. Therefore, the curves $\gamma$ and ${^{\beta}}\gamma$
do not intersect when $\theta \in (\theta_1+\beta,\tilde \theta)$.
When $\theta \in (\tilde \theta, \min(\theta_2, \tilde\theta+\beta))$, the function $r_1(\theta)$ is strictly increasing
while $r_2(\theta)$ is strictly decreasing. By considering the values of these two functions at the endpoints of the interval we conclude that there exists a unique number $\theta^*$ for which $r_1(\theta^*)=r_2(\theta^*)$. The point with polar coordinates
$(r_1(\theta^*),\theta^*)$ then gives us the unique point of intersection of $\gamma$ and ${^{\beta}}\gamma$. On the interval $\theta \in [\tilde \theta+\beta,\theta_2)$ we have the inequality $r_1(\theta)>r_2(\theta)$, and so the curves $\gamma$ and ${^{\beta}}\gamma$ do not intersect in this sector. Hence, we have shown that there exists a unique point of intersection of $\gamma$ and ${^{\beta}}\gamma$
when $\theta_1+\beta<\theta_2$.
To establish part (ii), let us assume that $(\tilde w, \tilde z) \in \gamma \cap {}^{\beta} \gamma$. This condition implies
that a circle $B_R$ with radius $R=\sqrt{\tilde w^2+ \tilde z^2}$ and center at the origin must intersect the curve $\gamma$ at two distinct points. However, this contradicts the condition that
the radius $r(t)$ is strictly monotone along the curve $\gamma$. Thus we have arrived at a contradiction. Therefore, the intersection of $\gamma$ and ${}^{\beta} \gamma$ must be empty.
\end{proof}
\begin{lemma}\label{lemma3}
Let $z=f(w)$ be a function defined on $0<w<w_0$ (where $w_0$ can be $+\infty$). Assume that $f(w)$ is smooth, decreasing, convex and its graph $\gamma=\{(w,f(w)) \; : \; 0<w<w_0\}$ is symmetric with respect to the line $z=w$. Then the curve $\gamma$ can be written in polar coordinates in form \eqref{curve-0}, and $r(\theta)$ is strictly decreasing on the interval $(0,\pi/4)$ and strictly increasing
on the interval $(\pi/4,\pi/2)$.
\end{lemma}
\begin{proof}
Note that the condition that $\gamma$ is symmetric with respect to the line $z=w$ implies $f(w_0)=0$. This result and the fact that
$f(w)/w$ is strictly decreasing allows us to represent $\gamma$ in polar coordinates in form \eqref{curve-0} with $\theta_1=0$ and
$\theta_2=\pi/2$. Let $(w_0,z_0)$ be the point of intersection of $\gamma$ and the line $z=w$, so that $z_0=f(w_0)$. The part of the graph
with $w>w_0$ corresponds to the polar coordinate representation with $\theta \in (0,\pi/4)$.
The symmetry of $\gamma$ with respect to $z=w$ implies $f'(w_0)=-1$, and since the function $f(w)$ is convex we see that
$f'(w)>-1$ for $w>w_0$. The radius in polar coordinates is given by $r=\sqrt{w^2+f(w)^2}$. Thus for $w>w_0$,
$$
\frac{\d r}{\d w}=r^{-1} \left(w+f(w)f'(w)\right)>r^{-1} (w-f(w))>r^{-1} (w-f(w_0))=r^{-1}(w-w_0)>0,
$$
where we have used the fact that $f(w)$ is strictly decreasing. This result combined with the fact that
$\d w/\d \theta<0$ for $\theta \in (0,\pi/2)$
shows that $\d r/\d \theta<0$ for $\theta \in (0,\pi/4)$. The fact that $\d r/\d \theta>0$ for $\theta \in (\pi/4,\pi/2)$ follows
by symmetry.
\end{proof}
We are now ready to complete the proof of part (\ref{th-c}) of Theorem \ref{th}. According to
Lemma \ref{lemma1}, it is enough to establish the validity of statement \eqref{condition}. We do this in four steps, depending on the value of $\alpha\in(0,\ 1)$.
\vspace{0.35cm}
\noindent
{\bf Proof of statement \eqref{condition} for $\alpha \in (0,1/4)$:}
\begin{figure}
\centering
\includegraphics[height =7cm]{pic1}
\caption{Illustration to the proof of statement \eqref{condition} for $\alpha \in (0,1/4)$.}
\label{pic1}
\end{figure}
\noindent
We begin by noting that if $\alpha \in (0,1/4)$ then the origin belongs to the convex set ${\cal U}_{\alpha}$ defined above in
\eqref{def_U_alpha}. Given this fact and the properties of ${\cal C}_{\alpha}$ which were described in Lemma \ref{lemma_C_properties}, it is clear that the curve ${\cal C}_{\alpha}$ must lie in the second, third and fourth quadrants; see Figure \ref{pic1} or the blue curve in Figure \ref{fig1}). We can express this curve in polar coordinates as follows
\[
{\cal{C}}_{\alpha}=\left \{(r(\theta)\cos(\theta),r(\theta) \sin(\theta)) \; : \; \pi/2 < \theta <2\pi \right \},
\]
and then divide it into three parts $\gamma_1$, $\gamma_2$ and $\gamma_3$, which lie respectively in the fourth, third and second quadrants (see
Figure \ref{pic1}).
First we consider the curve $\gamma_1$. We claim that the radius (the distance from the origin) strictly increases as we move along this curve to the right. To see this, we parametrize points on $\gamma_1$ by $(w,z(w))$ for $w>0$, where $z(w)$ is given by \eqref{fn_z_w}. For $w>0$ the function $z(w)$ is negative and strictly decreasing, which shows that $r(w)=\sqrt{w^2+z(w)^2}$ is strictly increasing.
Next, we consider the curve $\gamma_2$ parametrized by polar coordinates $(r,\theta)$, $\theta \in (\pi,3\pi/2)$. Our goal is to prove that
$r(\theta)$ is strictly decreasing on the interval $(\pi,5\pi/4)$ and strictly increasing on the interval $(5\pi/4,3\pi/2)$.
Note that the function $r(\theta)$ satisfies the equation
$$
\Phi(r(\theta) \cos(\theta))\Phi(r(\theta) \sin(\theta))=\alpha, \;\;\; \pi/2 < \theta <2\pi.
$$
Differentiating both sides of this equation with respect to $\theta$ we obtain
\begin{equation}\label{r_prime_over_r}
\frac{1}{r}\frac{\d r}{\d \theta}= \frac{\Phi(w)w e^{-z^2/2}-\Phi(z)ze^{-w^2/2}}{\Phi(w)z e^{-z^2/2}+\Phi(z)we^{-w^2/2}},
\end{equation}
where we denoted $w=r \cos(\theta)$ and $z=r\sin(\theta)$.
The denominator on the right-hand side of \eqref{r_prime_over_r} is strictly negative in the third quadrant. Thus to prove our claim about the increase/decrease of $r(\theta)$ it is enough to demonstrate that the numerator on the right-hand side of \eqref{r_prime_over_r} satisfies
\begin{align}\label{numerator}
&\Phi(w)w e^{-z^2/2}-\Phi(z)ze^{-w^2/2} >0 \quad \textrm{if} \quad w>z ,\\ \nonumber
&\Phi(w)w e^{-z^2/2}-\Phi(z)ze^{-w^2/2} <0 \quad \textrm{if} \quad w<z .
\end{align}
This is indeed true because the function $h(w)=w e^{w^2/2} \Phi(w)$ is strictly increasing
for all $w\in\mathbb{R}$. The monotonicity of $h(w)$ is obvious for $w>0$ and follows
from Pinelis (2002), in which the monotonicity of
$-h(-w)=w e^{w^2/2} {(1-{\Phi}(w))}$ was studied.
Let us summarize what we have established so far about the curve ${\cal C}_{\alpha}=\gamma_1 \cup \gamma_2 \cup \gamma_3$. As we move along this curve, starting in its upper part, the radius strictly decreases until it reaches its global minimum at the point of intersection of
${\cal C}_{\alpha}$ and the line $z=w$, and afterwards the radius strictly increases.
Now we are ready to prove that there exists a unique point of intersection between ${\cal C}_{\alpha}$ and ${}^{\beta}{\cal C}_{\alpha}$.
Let us denote $R=-\Phi^{-1}(2\alpha)$, so that $(0,-R)$ is the point of intersection of ${\cal C}_{\alpha}$ and the $z$-axis; this follows from \eqref{fn_z_w}. The monotonicity properties of the radius imply that the curves $\gamma_1$ and $\gamma_3$ lie outside of the circle $B_R$ with the center at the origin and the radius $R$, while the curve $\gamma_2$ lies completely inside this circle (of course the boundaries of these curves meet at the circle). Let us see what happens when we rotate the curve
${\cal C}_{\alpha}=\gamma_1 \cup \gamma_2 \cup \gamma_3$ by angle $\beta \in (0, \pi/2)$ counter-clockwise. The intersection of $\gamma_1 \cap {}^{\beta} \gamma_1$ and $\gamma_3 \cap {}^{\beta} \gamma_3$ is empty due to
Lemma \ref{lemma_inter-point}(ii). The curves ${}^{\beta}\gamma_1$ and $\gamma_3$ (and, similarly, $\gamma_1$ and ${}^{\beta} \gamma_3$) do not intersect since $\gamma_1$ and $\gamma_3$ lie in the fourth and second quadrants and the angle of rotation $\beta$ is strictly less
than $\pi/2$. The intersection $\gamma_1 \cap {}^{\beta} \gamma_2$ and ${\gamma_3} \cap {}^{\beta} \gamma_2$ is empty since these curves lie in different regions separated by the circle $B_R$ (one is inside and the other one is outside of this circle). And finally,
the intersection $\gamma_2 \cap {}^{\beta} \gamma_2$ consists of at most one point due to Lemma \ref{lemma_inter-point}(i). In fact, such a point of intersection must exist since we know that the curves ${\cal C}_{\alpha}$ and ${}^{\beta} {\cal C}_{\alpha}$ do intersect; see the last paragraph of the proof of Lemma \ref{lemma1}. Thus we have proved that for any $\alpha \in (0,1/4)$ and any $\beta \in (0,\pi/2)$ there exists a
unique point of intersection ${\cal C}_{\alpha} \cap {}^{\beta} {\cal C}_{\alpha}$.
\qed
\vspace{0.25cm}
\noindent
{\bf Proof of statement \eqref{condition} for $\alpha=1/4$:}
\noindent
The curve ${\cal{C}}_{1/4}$ contains the origin; see the red curve in Figure \ref{fig1}.
The proof of statement \eqref{condition} is the same as the proof above in the case $\alpha \in (0,1/4)$, except that now the curve $\gamma_2$ degenerates to a single point $(0,0)$ so that ${\cal C}_{1/4} \cap {}^{\beta} {\cal C}_{1/4}=\{(0,0)\}$.
\qed
\vspace{0.25cm}
\noindent
{\bf Proof of statement \eqref{condition} for $\alpha \in (1/4,1/2)$:}
\begin{figure}
\centering
\includegraphics[height =7cm]{pic2}
\caption{Illustration to the proof of statement \eqref{condition} for $\alpha \in (1/4,1/2)$.}
\label{pic2}
\end{figure}
\noindent
When $\alpha \in (1/4,1/2)$ the curve ${\cal C}_{\alpha}$ lies in the first, second and fourth quadrants; see
Figure \ref{pic2} or the black curve
in Figure \ref{fig1}. The proof
of statement \eqref{condition} is the same as in the case $\alpha \in (0,1/4)$, except that we now use Lemma \ref{lemma3} to prove that
the radius $r(\theta)$ is strictly decreasing for $\theta \in (0,\pi/4)$ and strictly increasing for
$\theta \in (\pi/4,\pi/2)$.
\qed
\vspace{0.25cm}
\noindent
{\bf Proof of statement \eqref{condition} for $\alpha \in [1/2,1)$:}
\noindent
When $\alpha \in [1/2,1)$ the curve ${\cal C}_{\alpha}$ lies entirely in the first quadrant;
see the green curve in Figure \ref{fig1}.
In this case the proof of statement \eqref{condition} follows from Lemma \ref{lemma_inter-point}(i) and Lemma \ref{lemma3}.
\qed
This ends the proof of statement (\ref{condition}), and according to Lemma \ref{lemma1}, the proof of Theorem \ref{th} is now complete.
\end{proof}
\section{Concluding comments}
\label{sec-4}
Traditional methods in insurance and finance work well for symmetric and light-tailed risks. It is however
a well-known empirical fact that in reality risks are skewed (Graham and Harvey, 2001), and it is exactly the severe
tail risks that drive economic capital allocations in portfolios of risks. It is not surprising
therefore that the notions of asymmetry and `fat tails' have been
gaining unprecedented popularity among theoreticians and practitioners (e.g., Staudt, 2010).
The phenomenon of dependent tail risks is equally subtle. In this respect, the not-too-distant
financial crisis doubtlessly demonstrated that, e.g., dependent defaults may be disastrous for economies of entire countries. However, the quantification of tail dependence is not a simple problem. In fact, the
classical approaches that are nowadays commonly employed seem to often underestimate the amount of tail dependence, as they rely solely
on the main
diagonal of the copula whereas the copula's (tail) behavior can be very different otherwise.
For this reason, the aforementioned approaches can miss the so-called
maximal tail dependence
even in some symmetric dependence structures.
Many, if not the majority, of the models in financial theory have been built with the Gaussian
distribution in mind, for which in this paper we have established that all of the classical
indices of tail dependence (Joe, 1993; Ledford and Tawn, 1996; Coles et al., 1999; Fischer and Klein, 2007) are maximal and
thus conform to the prudence-oriented character of current regulations (e.g., OSFI, 2015).
As the Gaussian copula has been very popular, and it will likely
remain such in the foreseeable future (e.g., MacKenzie and Spears, 2014), our findings are reassuring news for practitioners.
\section*{Acknowledgments}
{
We thank the anonymous referees for valuable comments
and suggestions that improved the work and resulted in a better
presentation of the material.
We are grateful to Prof. Dr. Paul Embrechts and all participants
of the ETH Series of Talks in Financial and Insurance Mathematics for feedback and
insights.
Our research has been supported by the Natural Sciences and Engineering Research Council (NSERC) of Canada. Jianxi Su also acknowledges the financial support of the Government of Ontario and MITACS Canada via, respectively, the Ontario Graduate Scholarship program
and the Elevate Postdoctoral fellowship.}
\section*{References}
\def\hang{\hangindent=\parindent\noindent}
\hang
{Asimit, V., Gerrard, R., Yanxi, H., Peng, L., 2016. Tail dependence measure for examining financial extreme co-movements. Journal of Econometrics, forthcoming.}
\hang
Cherubini, U., Durante, F., Mulinacci, S. (Eds.), 2013. Marshall--Olkin Distributions - Advances in Theory and Applications. Springer, {Switzerland}.
\hang
Coles, S., Heffernan, J., Tawn, J., 1999. Dependence measures for extreme value analyses. Extremes 2 (4), 339--365.
\hang
Donnelly, C., Embrechts, P., 2010. The devil is in the tails: Actuarial mathematics and the subprime mortgage crisis. ASTIN Bulletin 40 (1), 1--33.
\hang
Durante, F., Sempi, C., 2015. Principles of Copula Theory. Chapman and Hall/CRC, London.
\hang
Fischer, M.J., Klein, I., 2007. Some results on weak and strong tail dependence coefficients for means of copulas.
Diskussionspapiere No. 78/2007 // Friedrich-Alexander-Universit\"{a}t Erlangen-N\"{u}rnberg, Lehrstuhl f\"{u}r
Statistik und \"{O}konometrie, {available at http://econstor.eu/bitstream/10419/29623/1/614058171.pdf, accessed on March 3, 2016}.
\hang
Fung, T. and Seneta, E., 2011. The bivariate normal copula is regularly varying. Statistics and Probability Letters 81 (11), 1670--1676.
\hang
Furman, E., Su, J., Zitikis, R., 2015. Paths and indices of maximal tail dependence. ASTIN Bulletin, forthcoming.
\hang
Graham, J.R., Harvey, C.R., 2001. Expectations of equity risk premia, volatility and asymmetry from a corporate finance perspective. Working Paper No. 8678, National Bureau of Economic Research.
\hang
{Guillotte, S., Perron, F., 2016. Polynomial Pickands functions. Bernoulli 22 (1), 213-241.}
\hang
{ Hua, L., Joe, H., 2014. Strength of tail dependence based on conditional tail expectation. Journal of Multivariate Analysis 123, 143--159.}
\hang
Joe, H., 1993. Parametric families of multivariate distributions with given margins. Journal of Multivariate Analysis 46 (2), 262--282.
\hang
{Joe, H., Li, H., Nikoloulopoulos, A.K., 2010. Tail dependence functions and vine copulas. Journal of Multivariate Analysis 101, 252--270.}
\hang
Kousky, C., Cooke, R.M., 2009. The unholy trinity: Fat tails, tail dependence, and micro-correlations. Discussion Paper, Resources for the Future, Washington DC.
\hang
Ledford, A.W., Tawn, J.A., 1996. Statistics for near independence in multivariate extreme values. Biometrika 83 (1), 169--187.
\hang
Lehmann, E., 1966. Some concepts of dependence. Annals of Mathematical Statistics 37 (5), 1137--1153.
\hang
Li, D.X., 2000. On default correlation: a copula function approach. Journal of Fixed Income 9 (4), 43--54.
\hang
Li, L., Yuen, K.C., Yang, J., 2014. Distorted mix method for constructing copulas with tail dependence. Insurance: Mathematics and Economics 57, 77--89.
\hang
MacKenzie, D., Spears, T., 2014. `A device for being able to book P\&L': The organizational embedding of the Gaussian copula. Social Studies of Science 44 (3), 418--440.
\hang
McNeil, A.J., Frey, R., Embrechts, P., 2005. Quantitative Risk Management. Princeton University Press, Princeton.
\hang
Meyer, C., 2013. The bivariate normal copula. Communications in Statistics - Theory and Methods 42 (13), 2402--2422.
\hang
Nelsen, R.B., 2006. An Introduction to Copulas, second edition. Springer, New York.
\hang
OSFI, 2015. Own risk and solvency assessment (ORSA). Office of the Superintendent of Financial Institutions, Government of Canada, Ottawa, available at http://www.osfi-bsif.gc.ca/eng/fi-if/rg-ro/gdn-ort/gl-ld/Pages/e19.aspx, accessed on June 17, 2015.
\hang
{Pickands, J., 1981. Multivariate extreme value distributions. Bulletin of the International Statistical Institute 49, 859--878.}
\hang
Pinelis, I., 2002. Monotonicity properties of the relative error of a pad\'{e} approximation for {Mills'} ratio. Journal of Inequalities in Pure and Applied Mathematics 3 (2), 1--8.
\hang
Puccetti, G., R\"{u}schendorf, L., 2014. Asymptotic equivalence of conservative value-at-risk and expected shortfall-based capital charges. Journal of Risk 16 (3), 3--22.
\hang
R\"{u}schendorf, L., 2013. Mathematical Risk Analysis. Springer, Berlin.
\hang
Salmon, F., 2012. The formula that killed Wall Street. Significance 9 (1), 16--20.
\hang
Schmid, F., Schmidt, R., 2007. Multivariate conditional versions of Spearman's rho and related measures of tail dependence. Journal of Multivariate Analysis 98 (6), 1123--1140.
\hang
Sibuya, M., 1959. Bivariate extreme statistics. Annals of the Institute of Statistical {Mathematics} 11 (2), 195--210.
\hang
Staudt, A., 2010. Tail risk, systemic risk and copulas. Casualty Actuarial Society E-Forum 2, 1--23.
\hang
Wang, R., Peng, L., Yang, J., 2013. Bounds for the sum of dependent risks and worst value-at-risk with monotone marginal densities. Finance and Stochastics 17 (2), 395--417.
\hang
Zhang, M-H., 2008. Modelling total tail dependence along diagonals. Insurance: Mathematics and Economics 42 (1), 73--80.
\end{document}
| 4c40d362c47894bd9f83d0de11c13cc1928c3631 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction} \label{intro}
The goal of this work is to further explore heterotic target space duality, with the hope of moving towards a more fully developed notion of dualities involving the $(0,2)$ heterotic string. At present, many heterotic string dualities -- including any complete notion of $(0,2)$ mirror symmetry \cite{Blumenhagen:1996vu,Adams:2003zy,Melnikov:2010sa,Melnikov:2012hk} -- remain relatively unexplored compared to their counterparts in $(2,2)$ string theories. Heterotic mirror symmetry, for instance, must necessarily exhibit an even more intricate structure than $(2,2)$ (Calabi-Yau) Mirror symmetry which interchanges a pair of manifolds (and has sparked a rich interchange of ideas with mathematics, including the field of homological mirror symmetry\footnote{See \cite{homological_mirror} for a review.}.). In the case of heterotic $(0,2)$ theories it has been established that duality between $(0,2)$ sigma models and their supergravity limits can involve more complicated relationships between the geometry of Calabi-Yau manifolds and vector bundles over them. In many cases, rather than mirror pairs, heterotic dualities could involve whole chains of compactification geometries leading to the same effective theory. One concrete step towards understanding such correspondences has been taken in the direction of ``Target Space Duality" originating in $(0,2)$ Gauged Linear Sigma Models in \cite{Distler:1995bc} and further studied in \cite{Blumenhagen:1997vt,Blumenhagen:1997cn,Blumenhagen:2011sq,Rahn:2011jw}.
Within $(0,2)$ GLSMs, the target space duality of \cite{Distler:1995bc} is manifested by the existence of a shared sub-locus in the moduli space of the paired theories. More precisely each of the ``dual" $(0,2)$ GLSMs shares a non-geometric phase (commonly a Landau-Ginzburg phase). In this web of theories, the geometric phases -- consisting of a Calabi-Yau threefold, $X$, and a holomorphic bundle $\pi: V \to X$ over it -- are connected through a wealth of geometric correspondences, including geometric (i.e. conifold) transitions between the base CY geometries $(X, \widetilde{X})$. However, the relationship between the vector bundles $(V, \widetilde{V})$ is more subtle and no systematic geometric characterization of the full paired geometries, $(X,V), (\widetilde{X},\widetilde{V})$, has yet been provided. However, one essential feature of this target space duality is that the correspondence between two different geometric phases preserves the net number of moduli of the theory:
\begin{equation}\label{mod_count}
h^{1,1}(X)+h^{2,1}(X)+h^1(X, End_0(V))=h^{1,1}(\widetilde{X})+h^{2,1}(\widetilde{X})+h^1(X, End_0(\widetilde{V)})~.
\end{equation}
Unlike the simple interchange of $h^{1,1} \leftrightarrow h^{2,1}$ of CY mirror symmetry, the distribution of the total degrees of freedom given above between the moduli of the base CY manifold and those of the vector bundle can change dramatically across a $(0,2)$ dual pair.
In recent work \cite{Blumenhagen:2011sq,Rahn:2011jw}, systematic landscape studies of $(0,2)$ target space duals were undertaken. In particular, the connected geometric phases were explored in detail and it was found that the gauge symmetry, moduli and charged matter spectrum were generically identical in the geometric phases. This last was explored in the context of vector bundles with $SU(n)$ structure group, for which the charged matter is counted by
\begin{equation}\label{chargmat}
h^i(X, \wedge^k V)=h^i(\widetilde{X}, \wedge^k \widetilde{V})
\end{equation}
for $k=0, \ldots , n-1$. In the extensive studies carried out in \cite{Blumenhagen:2011sq,Rahn:2011jw}, the complete charged and singlet particle spectra were found to correspond exactly in almost all $(0,2)$ target space dual pairs\footnote{In a handful of cases ($< 2\%$ of roughly $83,000$ models analyzed) in the landscape scan, these spectra did not fully agree. The authors conjecture that this is due to the fact that several technical simplifications/assumptions had been made in the scan. See \cite{Blumenhagen:2011sq} for details.}.
Since the first observation of target space duality, it has been an open question whether or not the relationship described above corresponds to a true isomorphism of the target space theories/GLSMs (or indeed full non-linear sigma models) or rather represents a transition -- i.e. a shared locus in moduli space for two distinct theories (perhaps similar to conifold transitions in type II theories \cite{Strominger:1995cz})? To establish which of these possibilities is correct requires a more a detailed study of the ``dual" theories and a clear map between them -- which could determine whether they can be identified everywhere in their moduli spaces or only at some sublocus.
In the present work, our goal is to move towards this goal and test the duality further. Thus far, the study of such pairs has focused primarily on the degrees of freedom (i.e. particle spectra) of the theories as described above. In the present work, we hope to extend this further to explore the \emph{vacuum spaces} of target space dual pairs. This is in general a non-trivial task, since for $(0,2)$ theories the full form of the matter field K\"ahler potential and hence, the full scalar potential is largely unknown. However, recent work has made progress in explicitly connecting the underlying geometry of $(X, V)$ with the form of the D-term and F-term contributions to the $4$-dimensional ${\cal N}=1$ potential \cite{Anderson:2009nt, Anderson:2009sw,Anderson:2010ty,Anderson:2013qca, Anderson:2011ty}. Briefly, explicit obstructions to the slope-stability of the bundle correspond to D-terms (associated to Green-Schwarz anomalous $U(1)$ symmetries \cite{Lukas:1999nh,Blumenhagen:2005ga,Sharpe:1998zu,Anderson:2009sw,Anderson:2009nt}) and holomorphy conditions on $V$ arise as F-terms in the ${\cal N}=1$ effective theory \cite{Anderson:2011ty,Anderson:2013qca}.
With this link between bundle geometry and the effective potential in mind, the goals of the present work are:
\begin{itemize}
\item Generate examples of target space dual pairs in which the vector bundles lead to non-trivial D-terms and/or F-terms in their effective potentials (i.e. bundles that are not everywhere slope stable or holomorphic in the moduli space of the base CY geometry).
\item To build examples of such bundles, we must extend the class of monad bundles (see \cite{horrocks, beilinson}) that have been the primary focus of $(0,2)$ GLSMs to date. Thus far, the bundles appearing in the majority of the literature have been of the class of so-called ``positive" monad bundles. These consist of bundles defined via a complex built of \emph{ample} line bundles
\begin{equation}
0 \to \bigoplus_{k} {\cal O}({\bf a}_k) \stackrel{E}{\longrightarrow} \bigoplus_i {\cal O}({\bf b}_i) \stackrel{F}{\longrightarrow} \bigoplus_j {\cal O}({\bf c}_j) \to 0\label{first_mon}
\end{equation}
where $f \cdot g=0$ and $a_k,b_i, c_j \geq 0$ for all $i, j, k$. One motivation for focusing on this simple class of bundles is that they are frequently (though not always) stable with respect to the entire K\"ahler cone of $X$ \cite{Anderson:2008ex}. In this work, we will extend this class to involve some negative degree entries in \eref{first_mon} and demonstrate that although the bundles are not everywhere stable in K\"ahler moduli space, none-the-less the associated monad bundles correspond to good $(0,2)$ GLSMs. These more general monad bundles have already been considered in a variety of contexts in heterotic vacua and the extension of the class may be an important step towards applying sigma model tools to bundles of interest in string phenomenology \cite{Anderson:2009mh,Anderson:2008uw,Anderson:2007nc,Anderson:2008ex}.
\item At special points in the vacuum space of the theories, enhanced symmetries can arise. These include enhancements of what would in the full sigma model be worldsheet $(0,2)$ to $(2,2)$ supersymmetry (when $V$ is deformable into the holomorphic tangent bundle $TX$) and the loci in moduli space corresponding to ``stability walls" in which the bundle $V$ becomes reducible ($V \to V_1 \oplus V_2 \oplus \ldots$) and enhanced Green-Schwarz massive $U(1)$ symmetries are present in the effective theory. In the latter case, such loci can lead to complicated branch structure in the vacuum space of the theory in which many different bundles over the same $X$ are connected \cite{Anderson:2010ty}.
It is our goal in this context to explore the existence of such loci on each side of a target space dual pair and to explore whether a) such enhancement points always correspond across the target space dual pair? and b) whether the vaccum branch structure is identical in both halves of the pair?
\item Finally, to build example geometries $(X, V)$ with non-trivial D-term and F-term potentials that fit well into GLSMs a number of new technical tools will be needed. In particular, it will be necessary to describe such bundles and their possible deformations via monads. A systematic study of such bundles involves generalizations of the target space duality procedure and developing multiple ways of describing the same bundle to make branch structure in the effective theory evident.
\end{itemize}
In the following sections we will approach the problems/goals outlined above. It should be noted that this list raises a number of interesting questions in the GLSMs themselves -- most prominently how the large volume constraints of bundle holomorphy and stability are realized in the GLSM, and whether the corresponding D- and F-term lifting of moduli in the ${\cal N}=1$ effective theory extends to other non-geometric phases of the theory? For now, we will leave these important questions unanswered and focus primarily on the structure of the geometric phases and target space theories themselves. It is our goal to test target space duality itself via the geometry and potentials described above.
In the first two sections, the two primary starting points of this work are reviewed. In Section \ref{Sec:Review} this is the $(0,2)$ target space duality procedure as developed in \cite{Distler:1995bc,Blumenhagen:1997vt,Blumenhagen:1997cn,Blumenhagen:2011sq,Rahn:2011jw} and in Section \ref{potential_hym}, the correspondence between bundle stability/holomorphy and non-trivial D-terms and F-terms in the effective theory as studied in \cite{Sharpe:1998zu,Anderson:2009nt, Anderson:2009sw,Anderson:2010ty,Anderson:2013qca, Anderson:2011ty}. In Section \ref{dterms} we provide an example of a bundle with non-trivial D-term constraints and track those conditions through the possible target space dual chains of geometries. In particular, this analysis allows for the exploration of the branch structure arising in the vacuum space in the neighborhood of a stability wall on both sides of a target space dual pair. In addition, examples are given of possible perturbative/non-perturbative dual pairs and examples in which disconnected components of the moduli space are reproduced in the dual theory. In Section \ref{Fterm_sec} this procedure is repeated for a bundle with a non-trivial F-term constraint. To probe the target space dual correspondence at yet another point of enhanced symmetry, in Section \ref{tan_defs} dual pairs are constructed in which the starting bundle is constructed as a deformation of the holomorphic tangent bundle (i.e. the $(0,2)$ theory contains a $(2,2)$ locus). Finally, our results are summarized and future directions outlined in Section \ref{conclusions}. Section \ref{AppendixA} contains details of bundle stability, while Appendix \ref{appendixB} provides a comprehensive list of target space dual chains for several examples.
To begin, in the following two sections, we review briefly the necessary ingredients for this work - namely a) The basic ideas of target space duality in heterotic $(0,2)$ GLSMs and b) the relationship between the Hermitian Yang-Mills equations and non-trivial D-/F-terms in the effective theory.
\section{A Brief Review of Target Space Duality}\label{Sec:Review}
The notion of $(0,2)$ target space duality is well-studied in the literature \cite{Distler:1995bc,Blumenhagen:1997vt,Blumenhagen:1997cn,Blumenhagen:2011sq,Rahn:2011jw} (see \cite{Blumenhagen:2011sq} for a thorough review) and in this section we briefly summarize the basic ingredients as well as clarify the simple class of geometries we will consider in order to explore the vacuum structure of the ${\cal N}=1$ effective theories.
\subsection{${\bf (0,2)}$ Gauged Linear Sigma Models}\label{sec:GLSM}
Gauged linear sigma models (GLSMs) are massive $2$-dimensional theories which can flow in the infrared (under suitable conditions) to true superconformal theories (i.e. string sigma models). The famous ``phase" structure of GLSMs \cite{Witten:1993yc} links the vacuum structure of the $2$-dimensional theory to the geometric data of toric geometry and string compactifications. The freedom to vary a Fayet-Illiopolos parameter leads to a variety of distinct solutions including non-linear sigma models (and their associated target space geometries -- include CY manifolds), Landau-Ginzburg orbifolds, and a rich variety of hybrid models.
In this work we will restrict our consideration to the simplest, Abelian class of GLSMs. The field content will consist of multiple $U(1)$ gauge fields $A^{(\alpha)}$ with $\alpha = 1,...,r$. Following standard notation in the literature, we will label two sets of chiral superfields as $\{X_i|i=1,...,d\}$ with $U(1)$ charges $Q_i^{(\alpha)}$, and $\{P_l|l=1,...,\gamma\}$ with $U(1)$ charges $-M_l^{(\alpha)}$. The essential field content will also include two sets of Fermi superfields: $\{\Lambda^a|a=1,...,\delta\}$ with charges $N_a^{(\alpha)}$, and $\{\Gamma^j|j=1,...,c\}$ with charges $-S_j^{(\alpha)}$.
The $U(1)$ charges appearing in the GLSM are frequently taken to satisfy positivity conditions that are compatible with realizing Calabi-Yau manifolds and stable, holomorphic vector bundles over them in some geometric phase. In this work, we will take the charges $Q_i^{(\alpha)} \geq 0$ and require that for each $i$, there exists at least one $r$ such that $Q_i^{(\alpha)}>0$. This assumption of (semi-) positivity will hold in addition for the charges $S_j^{(\alpha)}$ and $M_l^{(\alpha)}$. However, in some cases we will consider solutions in which charges $N_a^{(\alpha)}$ may be negative (the suitability of such special cases will be further discussed in Sections \ref{dterms} and \ref{Fterm_sec}).
The key field content and charges of the GLSM can be concisely summarized by a ``charge matrix" of the form:
\begin{equation}
\begin{aligned}
&\begin{array}{|c||c|}
\hline
x_i & \Gamma^j \\
\noalign{\hrule height 1pt}
\begin{array}{cccc}
Q^{(1)}_1 & Q^{(1)}_2 & \ldots & Q^{(1)}_d\\
Q^{(2)}_1 & Q^{(2)}_2 & \ldots & Q^{(2)}_d\\
\vdots & \vdots & \ddots & \vdots \\
Q^{(r)}_1 & Q^{(r)}_2 & \ldots & Q^{(r)}_d
\end{array}
&
\begin{array}{cccc}
-S^{(1)}_1 & -S^{(1)}_2 & \ldots & S^{(1)}_c\\
-S^{(2)}_1 & -S^{(2)}_2 & \ldots & S^{(2)}_c\\
\vdots & \vdots & \ddots & \vdots \\
-S^{(r)}_1 & -S^{(r)}_2 & \ldots & S^{(r)}_c\\
\end{array}\\
\hline
\end{array}
\\[0.1cm]
&\begin{array}{|c||c|}
\hline
\Lambda^a & p_l \\
\noalign{\hrule height 1pt}
\begin{array}{cccc}
N^{(1)}_1 & N^{(1)}_2 & \ldots & N^{(1)}_\delta\\
N^{(2)}_1 & N^{(2)}_2 & \ldots & N^{(2)}_\delta\\
\vdots & \vdots & \ddots & \vdots \\
N^{(r)}_1 & N^{(r)}_2 & \ldots & N^{(r)}_\delta
\end{array}
&
\begin{array}{cccc}
-M^{(1)}_1 & -M^{(1)}_2 & \ldots & -M^{(1)}_\gamma\\
-M^{(2)}_1 & -M^{(2)}_2 & \ldots & -M^{(2)}_\gamma\\
\vdots & \vdots & \ddots & \vdots \\
-M^{(r)}_1 & -M^{(r)}_2 & \ldots & -M^{(r)}_\gamma\\
\end{array}\\
\hline
\end{array}
\end{aligned}\label{charge_matrix}
\end{equation}
Familiar GLSM anomaly cancellation conditions (including gauge and gravitational anomalies) impose the following linear and quadratic constraints on the $U(1)$ charges to hold for all $\alpha, \beta = 1, ..., r$:
\begin{eqnarray}\nonumber \label{anomalies}
\sum_{a=1}^\delta N_a^{(\alpha)} = \sum_{l=1}^\gamma M_l^{(\alpha)} \qquad&& \sum_{i=1}^d Q_i^{(\alpha)} = \sum_{j=1}^c S_j^{(\alpha)}\\
\sum_{l=1}^\gamma M_l^{(\alpha)} M_l^{(\beta)} - \sum_{a=1}^\delta N_a^{(\alpha)} N_a^{(\beta)} &=& \sum_{j=1}^c S_j^{(\alpha)} S_j^{(\beta)} - \sum_{i=1}^d Q_i^{(\alpha)} Q_i^{(\beta)}
\end{eqnarray}
The structure of the GLSM is further determined by a non-trivial scalar potential and superpotential. The latter is given by
\begin{equation} \label{superpotential}
S = \int d^2z d\theta \bigg[\sum_j \Gamma^j G_j (x_i) + \sum_{l,a} P_l \Lambda^a F_a^l (x_i) \bigg]
\end{equation}
where the functions $G_j$ and $F_a^l$ are quasi-homogeneous polynomials whose multi-degrees are determined by $U(1)$ gauge invariance to be:
\begin{equation}
\begin{aligned}
&\begin{array}{|c|}
\hline
G^j \\
\noalign{\hrule height 1pt}
\begin{array}{cccc}
S_1 & S_2 & \ldots & S_c
\end{array}\\
\hline
\end{array}
\\[0.1cm]
&\begin{array}{|c|}
\hline
F_a{}^l \\
\noalign{\hrule height 1pt}
\begin{array}{cccc}
M_1-N_1 & M_1-N_2 & \ldots & M_1-N_\delta\\
M_2-N_1 & M_2-N_2 & \ldots & M_2-N_\delta\\
\vdots & \vdots & \ddots & \vdots \\
M_\gamma-N_1 & M_\gamma-N_2 & \ldots & M_\gamma-N_\delta
\end{array}\\
\hline
\end{array}
\end{aligned}
\end{equation}
Furthermore, the function $F_a^l$ will be chosen to satisfy a \emph{transversality condition} such that $F_a^l(x)=0$ only when $x_i=0$
The scalar potential includes contributions from F-terms:
\begin{equation}
V_F = \sum_j \big |G_j(x_i) \big|^2 + \sum_a\big |\sum_l p_lF_a^l(x_i) \big|^2
\end{equation}
and D-terms
\begin{equation}
V_D = \sum_{\alpha=1}^r \bigg( \sum_{i=1}^d Q_i^{(\alpha)} |x_i|^2 - \sum_{l=1}^\gamma M_l^{(\alpha)}|p_l|^2 - \xi^{(\alpha)} \bigg)^2
\end{equation}
In this latter expression the $\xi^{(\alpha)} \in \mathbb R$ is the Fayet-Iliopoulos (FI) parameter which determines the structure of the vacuum. In particular the sign of each of the FI terms (for $\alpha=1, \ldots r)$, determine whether the solution of theory is a geometric, Landau-Ginzburg, or hybrid phase.
This can be easily illustrated in the case of a single $U(1)$ symmetry: For $\xi > 0$ the D-term implies that not all $X_i$ are zero simultaneously, thus not all $F_a$ are zero, and the F-term implies $G_j(x_i) = 0$ and $<p> = 0$. This ``geometric" phase contains the data of a $(0,2)$ non-linear sigma-model on a generally singular complete intersection, $X$, in a weighted projective space $\mathbb P_{Q_1,...,Q_d}[S_1,...,S_c]$. Moreover, the superpotential in \eref{superpotential} leads to a mass term of the form $\sum_a \pi \lambda^a F_a$, which makes massive one linear combination of the $\lambda^a$. The details of fermionic gauge symmetries, etc can be found in \cite{Distler:1995mi,Distler:1996tj,Blumenhagen:2011sq}. In brief summary, the remaining massless combinations of the left-moving fermions $\lambda_a$ couple to a vector bundle over it defined by the monad
\begin{equation}\label{full_monad}
0 \to \mathcal O_{X}^{\oplus r_{\mathcal V}} \xrightarrow{E_i^a} \bigoplus_{a=1}^\delta \mathcal O_{X}(N_a) \xrightarrow{F_a^l} \bigoplus_{l=1}^\gamma \mathcal O_{X}(M_l) \to 0
\end{equation}
on the complete intersection CY manifold $X=\cap_{j=1}^{c} G_j$. The rank $(\delta - \gamma - r_{\cal V})$ vector bundle $V \to X$ is defined as
\begin{equation}
V = \frac{ker(F_a^l)}{im(E_i^a)}
\end{equation}
where $E_i^{a}$ arises from $r_{{\cal V}}$ additional fermionic gauge symmetries that have been introduced leading to a deviation in chirality of ${\overline {\cal D}}\Lambda^a=\sqrt{2} \Sigma^i E_{i}^{a}$. Note that although the new, neutral, chiral superfields, $\Sigma^i$, contribute to the scalar potential, they will play no role in the analysis undertaken here and will be omitted. In general, $E_i^{a}$ obeys a composition rule with $F_a^l$: $E \circ F=0$ on $X$. Furthermore, in this work, we will consider the simple case that $E_i^{a}=0$ and $V$ is defined
as a kernel: $V = ker(F_a^l)$ defined by a short exact sequence
\begin{equation}
0 \to V \to \bigoplus_{a=1}^\delta \mathcal O_{\mathcal M}(N_a) \xrightarrow{F_a^l} \bigoplus_{l=1}^\gamma \mathcal O_{\mathcal M}(M_l) \to 0
\end{equation}
The simplest possible ``non-geometric" phase can be realized with $\xi < 0$. In this case, the D-term implies that $<p> \neq 0$ thus all $<x_i> = 0$. This vacuum corresponds to a Landau-Ginzburg orbifold with a superpotential:
\begin{equation}\label{LG_superpotential}
\mathcal W(x_i, \Lambda^a, \Gamma^i) = \sum_j\Gamma^jG_j(x_i) + \sum_a \Lambda^aF_a(x_i)
\end{equation}
Finally, when $r>1$, the possibility of hybrid phases -- with some $x^{(\alpha)}$ less than and some great than zero -- naturally arises.
The simple form of \eref{LG_superpotential} first inspired the observation in \cite{Distler:1995bc} of target space duality. By inspection of \eref{LG_superpotential} it is clear that an exchange/relabeling of the functions $G_j$ and $F_a$ will not affect the Landau-Ginzburg model, as long as anomaly cancellation conditions are satisfied. The natural observation is that two distinct GLSMs could ``share" a non-geometric phase in which the original (large volume) role of $G_j$ and $F_a$ is obscured. This notion of duality has been extended in a variety of ways including by allowing for the resolution of singularities in the geometry \cite{Chiang:1997kt} and extending the duality with a trivial re-writing which allows for a change in the number of $U(1)$ symmetries in the dual pair \cite{Distler:1996tj,Blumenhagen:2011sq}.
In the following section we review and illustrate the target space duality algorithm as laid out clearly in \cite{Blumenhagen:2011sq,Rahn:2011jw} and extend it still further by applying new redundant descriptions of the bundle geometry that make new target space duals manifest.
\subsection{Target Space Duality}\label{the_algorithm}
Given a starting point of a $(0,2)$ GLSM, the procedure for generating a target space dual model was laid out succinctly in \cite{Blumenhagen:2011sq,Rahn:2011jw}. We follow the conventions of \cite{Blumenhagen:2011sq} by taking $||X||$ to indicate the charge of the field $X$ and $||G||$ to denote the multi-degree of a homogeneous function $G(X)$. For completeness we summarize the algorithm here: \\
{\bf The Algorithm}:
\begin{enumerate}
\item Find all phases of a smooth $(0,2)$ GLSM, including a geometric phase with geometry $(X,V)$.
\item Construct a phase in which at least one field $p_l$ (for illustration, take $p_1$) is required to have a non-zero vacuum, $\langle p_1 \rangle \neq 0$.
\item Rescale $k$ Fermi superfields by the constant vev, $\langle p_1 \rangle$, and exchange the role of some $\lambda^a$ and $\Gamma^j$:
\begin{equation}
\tilde{\Lambda}^{a_i}:=\frac{\Gamma^{j_i}}{\langle p_1 \rangle} ~,~~~~\tilde{\Gamma}^{j_i}=\langle p_1 \rangle \Lambda^{a_i}~,~~~ \forall ~i=1, \ldots k,
\end{equation}
For consistency, this relabeling/rescaling can only be done when $\sum_i ||G_{j_1} || = \sum_i || F_{a_i}^{1} ||$ for anomaly cancellation.
\item Tune the bundle moduli (i.e. morphisms $F_a^l$) so that $\Lambda^{a_i}$ appear only in terms with $P_1$ for all $i$. i.e. choose
\begin{equation}
F_{a_i}^{l}=0 ~,~~~~ \forall~ l \neq 1,~~i=1, \ldots k
\end{equation}
\item Move away from this shared non-geometric phase and define the Fermi superfields of the new GLSM such that each term in the superpotential is gauge invariant:
\begin{equation}
|| \tilde{\Lambda}^{a_i} ||= || \Gamma^{j_i} || - ||P_1 ||~~~~,~~~|| \tilde{\Gamma}^{j_i} ||= || \Lambda^{a_i} || + || P_1 ||
\end{equation}
\item Allow the moduli to move to a generic point and define a new (dual) $(0,2)$ GLSM with a distinct geometric phase corresponding to a new Calabi-Yau and vector bundle, $(\widetilde{X}, \widetilde{V})$.
\end{enumerate}
It was proved in \cite{Blumenhagen:2011sq} that for the algorithm as described above if the initial $(0,2)$ GLSM satisfies the anomaly cancellation conditions in \eref{anomalies} then so does the constructed dual. Moreover, as mentioned in the Introduction, it was observed in the landscape scan of \cite{Blumenhagen:2011sq} that throughout this process, in the target space theories the net multiplicities of charged matter, and the total number of massless gauge singlets are preserved for almost all of the examples, where the individual number of complex, K\"ahler and bundle moduli are interchanged as
\begin{eqnarray}\nonumber
h^*(X,\wedge^{k}V) &=& h^*(\widetilde X, \wedge^{k} \widetilde V) \\
h^{2,1}(X) + h^{1,1}(X) + h^1_X(End_0(V)) &=& h^{2,1}(\widetilde X) + h^{1,1}(\widetilde X) + h^1_X(End_0(\widetilde V))
\end{eqnarray}
On a technical note, the dimensions of the cohomology groups above can be computed using standard tools such as \cite{Blumenhagen:2010pv,cicy_package}. This procedure is illustrated explicitly below.
\subsubsection{Illustration of the duality procedure}
For concreteness, it is easiest to understand the target space duality algorithm in the context of an example. In principle an arbitrary number of $G$'s and $F$'s can be interchanged in the non-geometric phase. Below we will consider the simplest case in which two such maps play a role (note that in this case the quadratic anomaly cancellation conditions are known to be satisfied in the dual theory \cite{Blumenhagen:2011sq}).
In the non-geometric phase, we will make explicit the two terms to be exchanged in the corresponding superpotential (here the first two terms with $a=1,2$):
\begin{equation}
\mathcal W = \sum_{j=1}^c\Gamma^jG_j + \sum_{a=1}^2 P_1 \Lambda^aF_a^1 + \sum_{a=3}^\delta P_1 \Lambda^aF_a^1+ \sum_{l=2}^\gamma \sum_{a=1}^\delta P_l \Lambda^aF_a^l
\end{equation}
In the GLSM, not all the $<p>$'s are allowed to vanish simultaneously. Choosing $p_1$ to obtain a vev, the superpotential can be written as
\begin{equation}\label{eg_switch1}
\mathcal W = \Gamma^1G_1 + \Gamma^2G_2 + \sum_{j=3}^c\Gamma^jG_j
+ <p_1> \Lambda^1F_1^1 + <p_1> \Lambda^2F_2^1 + \sum_{a=3}^\delta <p_1> \Lambda^aF_a^1+ \sum_{l=2}^\gamma \sum_{a=1}^\delta P_l \Lambda^aF_a^l
\end{equation}
Performing the interchange of the first two $G_j$'s and $F_a$'s as prescribed above, the equivalent superpotential can be expressed by $\tilde G_j$'s and $\tilde F_a$'s
\begin{equation}\label{eg_switch2}
\mathcal W = \tilde \Gamma^1\tilde G_1 + \tilde \Gamma^2\tilde G_2 + \sum_{j=3}^c\Gamma^jG_j \\
+ <p_1> \tilde \Lambda^1\tilde F_1^1 + <p_1> \tilde \Lambda^2\tilde F_2^1 + \sum_{a=3}^\delta <p_1> \Lambda^aF_a^1+ \sum_{l=2}^\gamma \sum_{a=1}^\delta P_l \Lambda^aF_a^l ~.
\end{equation}
In this case, the rescalings are
\begin{eqnarray}\nonumber
&\tilde{\Gamma}^1 := <p_1>\Lambda^1, \tilde{\Gamma}^2 := <p_1>\Lambda^2, \tilde{\Lambda}^1 := \frac{\Gamma^1}{<p_1>}, \tilde{\Lambda}^2 := \frac{\Gamma^2}{<p_1>}&\\
&\tilde G_1 :=\ F_1^1,\ \tilde G_2 :=\ F_2^1,\ \tilde F_1^1 :=\ G_1,\ \tilde F_2^1 :=\ G_2 &
\end{eqnarray}
and $||G_1|| + ||G_2|| = ||F_1^1|| + ||F_2^1||$. From this point, a return to a geometric phase would yield a new configuration corresponding to a geometric pair $(\tilde X, \tilde V)$.
\subsubsection{Target space duals with an additional $U(1)$}\label{tsd_extrau1}
As outlined above, it is clear that the ``relabeling" of fields at the shared Landau-Ginzburg point can mix the degrees of freedom in $h^{2,1}(X)$ and $h^1(X, End_0(V))$ in the target space dual. More general possibilities are possible however that can \emph{also} change the dimension of $h^{1,1}$. As one particular example of such a change, we will be interested here in geometries $X, \widetilde{X}$ which are related by geometric (i.e. conifold-type) transitions.
To change the dimension of $h^{1,1}$ across the target space dual it is necessary to re-write the initial GLSM in an equivalent/redundant way. Stated briefly, it is always possible to introduce into the GLSM a new coordinate (i.e. a new Fermi superfield) $y_1$ with multi-degree ${\mathcal B}$ and a new hypersurface (i.e. a Chiral superfield with opposite charge to the new Fermi superfield) $G^{\cal B}$ corresponding to a homogeneous polynomial of multi-degree ${\cal B}$. Denoting the initial bundle/manifold in terms of its GLSM charges as
\begin{equation}\label{starting_point}
V_{N_1, \ldots N_{\delta}}[M_1, \ldots, M_{\gamma}] \longrightarrow \mathbb{P}_{Q_1, \ldots Q_d}[S_1, \ldots, S_c]
\end{equation}
the above addition can be written
\begin{equation}\label{same_base}
V_{N_1, \ldots N_{\delta}}[M_1, \ldots, M_{\gamma}] \longrightarrow \mathbb{P}_{Q_1, \ldots Q_d,{\cal B}}[S_1, \ldots, S_c,{\cal B}]
\end{equation}
As argued in \cite{Blumenhagen:2011sq}, this addition leaves the GLSM and the corresponding target space geometry completely unchanged. However, as we will demonstrate, this redundant description can play a non-trivial role in the target space duality re-labeling.
Following \cite{Blumenhagen:2011sq}, suppose that in an example as in the previous Subsection, there are two chosen map elements $F_1^1$ and $F_2^1$ that have been chosen to be interchanged with a defining relation $S_1$. In this case we can choose the redundant new coordinate, $y_1$, to have charge
\begin{equation}
||{\cal B}|| = ||F_1^1|| + ||F_2^1|| - S_1
\end{equation}
and under the re-labelings required in the algorithm of Section \ref{the_algorithm} it is possible to choose, for example,
\begin{equation}\label{u1_field_defs}
\tilde{N}_1=M_1-S_1=0~,~\tilde{N}_2=M_2 - {\cal B}~,~\tilde{S}_1 = ||F_1^1||~,~\tilde{{\cal B}}=||F_2^1 ||
\end{equation}
That is, a Fermi superfield has now been generated with apparently vanishing charge. This is misleading however because as we will see, this field may be thought of as carrying an additional $U(1)$ charge. To see the origin of this new $U(1)$, it should be noted that the redundancy/relabeling described above can, unfortunately, lead to singular geometries. As pointed out in \cite{Distler:1996tj}, to avoid such singularities in the new model and have a valid, smooth large volume limit $(\tilde X, \tilde V)$, it is possible to add an additional $U(1)$ gauge symmetry under which the formally uncharged Fermi-superfield carries a non-vanishing charge. Geometrically this corresponds with introducing one further coordinate $y_2$ and identifying the two new coordinates as the homogeneous coordinates on a $\mathbb{P}^1$ (i.e. a blow-up of the original geometry). So the charges now read
\begin{tablehere}
\begin{center}
\begin{tabular}{|ccccc|cccc||cccc|cccc| }
\hline
$x_1$ & \ldots & $x_d$ & $y_1$ & $y_2$ &$\Gamma^1$ & \dots & $\Gamma^c$ & $\Gamma^B$ & $\Lambda^1$ & $\Lambda^1$ & \ldots & $\Lambda^\delta$ & $p_1$ & $p_2$ & \ldots & $p_\gamma$\\
\noalign{\hrule height 1pt}
0 & \ldots & 0 & 1 & 1 & 0 & \ldots & 0 & $-1$ & 0 & 0 & \ldots & 0 & $-1$ & 0 & \ldots & 0 \\
$Q_1$ & \ldots & $Q_d$ & ${\cal B}$ & 0 & $-S_1$ & \ldots & $-S_c$ & $-{\cal B}$ & $N_1$ & $N_2$ & \ldots & $N_\delta $ & $-M_1$ & $-M_2$ & \ldots & $-M_\gamma$ \\
\hline
\end{tabular}
\hfill
\end{center}
\end{tablehere}
where once again, this configuration is equivalent to the initial one in \eref{starting_point}. This equivalence can be verified by noting that the constraint $G^{\cal B} = y_1 = 0$ can be used to eliminate the coordinate $y_1$. Further, $y_2$ can be fixed by the additional $U(1)$ gauge symmetry and associated D-term. Thus, this configuration is simply the original space $\times$ a single point. Use the same method discussed before, we can perform rescalings to this new configuration, resulting in a new target space dual with an additional $U(1)$.
Applying the field redefinitions in \eref{u1_field_defs} we arrive at last to the new configuration
{\scriptsize
\begin{tablehere}
\begin{center}
\begin{tabular}{|ccccc|cccc||cccc|cccc| }
\hline
$x_1$ & \ldots & $x_d$ & $y_1$ & $y_2$ &$\tilde{\Gamma}^1$ & \dots & $\Gamma^c$ & $\tilde{\Gamma}^B$ & $\tilde{\Lambda}^1$ & $\tilde{\Lambda}^1$ & \ldots & $\Lambda^\delta$ & $p_1$ & $p_2$ & \ldots & $p_\gamma$\\
\noalign{\hrule height 1pt}
0 & \ldots & 0 & 1 & 1 & $-1$ & \ldots & 0 & $-1$ & 1 & 0 & \ldots & 0 & $-1$ & 0 & \ldots & 0 \\
$Q_1$ & \ldots & $Q_d$ & ${\cal B}$ & 0 & $-(M_1-N_1)$ & \ldots & $-S_c$ & $-(M_1-N_2)$ & $0$ & $M_2-{\cal B}$ & \ldots & $N_\delta $ & $-M_1$ & $-M_2$ & \ldots & $-M_\gamma$ \\
\hline
\end{tabular}
\hfill
\end{center}
\end{tablehere}}
For simplicity, in this work all examples will be chosen so that the target space duals satisfy ${\cal B}=0$ which will guarantee that the theories stay within the class of simple complete intersection Calabi-Yau manifolds in products of projective spaces \cite{Candelas:1987kf,Hubsch:1986ny}.
\subsubsection{Redundant bundle geometry and more general target space duals}\label{repeated_line_bundles}
It is useful to observe here that the procedure of inducing a redundant description of the manifold, $X$, illustrated above can also be applied to the vector bundle, $V$. At the level of the GLSM, once again we introduce a new pair consisting of a chiral/fermi superfields of opposite charge, but this time to the GLSM data corresponding to the bundle $V$, rather than to $X$ as in the previous section. Once again denoting the bundle in terms of its GLSM charges as in \eref{starting_point}
this time the addition of fields takes the form
\begin{equation}\label{repeated_ent}
V_{N_1, \ldots N_\delta, {\cal B}}[M_1,\ldots M_{\gamma},{\cal B}] \to \mathbb{P}_{Q_1,\ldots Q_d}[S_1, \ldots, S_c]
\end{equation}
Note that this is exactly the redundancy utilized in \eref{same_base}, only here it is applied to $V$ rather than to $X$.
At the level of the bundle geometry, the procedure involves a simple modification of the monad defining $V$. Beginning with the short exact sequence
\begin{equation}\label{monadfirst}
0 \to V \to B \stackrel{F}{\longrightarrow} C \to 0
\end{equation}
with $B,C$ sums of line bundles, we can realize the addition in \eref{repeated_ent} by adding the same line bundle $L$ to both $B$ and $C$:
\begin{equation}\label{add_line_bundle}
0 \to V' \to B \oplus L \stackrel{F'}{\longrightarrow} C \oplus L \to 0
\end{equation}
It is clear that the bundles $V'$ and $V$ share a locus in moduli space for which the map $F'$ takes the form
\begin{equation}\label{simple_map}
F'=\left(\begin{array}{cc}
F & \bf{0} \\
\bf{0} & \mathbb{C}
\end{array}\right)
\end{equation}
where $F$ is the original fiberwise morphism in \eref{monadfirst} and $\mathbb{C}$ represents the constant map from $L \to L$. For this choice of $F'$ the bundles $V$ and $V'$ are isomorphic. It is also straightforward to verify that the topology of $V$ and $V'$ are identical. Their agreement at a general point in moduli space can be probed via Lemmas \ref{OSS_lemma_thm} and \ref{OSS_corollary_thm} in Appendix \ref{AppendixA}.
For more general maps of the form
\begin{equation}\label{complicated_map}
F'=\left(\begin{array}{cc}
F & \bf{\alpha} \\
\bf{\beta} & \mathbb{C}
\end{array}\right)
\end{equation}
where $\alpha \in H^0(X, L^{\vee} \otimes C)$ and $\beta \in H^0(X, B^{\vee} \otimes L)$ are non-trivial, it is possible to probe a broader region of the shared moduli space. As we will see below, adding $L$ such that $\alpha, \beta \neq 0$ can \emph{lead to new target space dual theories} and is a generalization of the procedure systematically explored in \cite{Blumenhagen:2011sq}. Moreover, demanding that there exist generic non-trivial maps, $\alpha, \beta \neq 0$, bounds the set of possible line bundles $L$ that can be added to produce new target space duals to a finite set. Roughly, these conditions amount to the constraint that the entries $O(a,b,\ldots)$ in $L$ are ``small enough" to map into $C$ but ``big enough" that some entries in $B$ can be mapped non-trivially into $L$.
For example, starting with the geometry
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccc}
1& 1& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1
\end{array}&\begin{array}{c}
-2 \\
-4
\end{array}&\begin{array}{cccccccc}
0& 0& 1& 1& 1& -1& 2& 2 \\
1& 1& 0& 0& 3& 1& 1& 2
\end{array}&\begin{array}{ccc}
-3& -1& -2 \\
-2& -4& -3
\end{array}\\
\hline
\end{array} \label{eq:bundle3a}\end{aligned}\end{equation}
where $V$ is defined via the monad
\begin{equation}
\begin{split}
0 \to V & \to \mathcal O(0,1)^{\oplus2} \oplus \mathcal O(1,0)^{\oplus2} \oplus \mathcal O(1,3) \oplus \mathcal O(-1,1) \oplus \mathcal O(2,1) \oplus \mathcal O(2,2) \\
&\stackrel{F}{\longrightarrow} \mathcal O(3,2) \oplus \mathcal O(1,4) \oplus \mathcal O(2,3) \to 0
\end{split}
\end{equation}
For a component $B_i = \mathcal O(b_1, b_2)$ mapping into $C_j = \mathcal O(c_1, c_2)$, $h^0(B_i^*\otimes C_j) = h^0(\mathcal O(c_1-b_1, c_2-b_2)) \neq 0$. Which means $c_1 \geq b_1, c_2 \geq b_2$, or $c_1 - b_1 = -1, c_2 - b_2 \geq 4$.
In this case, $L$ can only be one of the set
\begin{align}\label{eg_set}
&\{\mathcal O(-1,2),~\mathcal O(-1,3),~\mathcal O(-1,4), ~\mathcal O(0,2),
~\mathcal O(0,3), ~\mathcal O(0,4), ~\mathcal O(1,1), ~\mathcal O(1,2), \nonumber \\
& ~\mathcal O(1,3), ~\mathcal O(2,0), ~\mathcal O(2,1), ~\mathcal O(2,2), ~\mathcal O(3,0), ~\mathcal O(3,1) \}
\end{align}
It should be noted finally that another concern is whether each of the line bundles appearing in \eref{eg_set} can appear in \eref{add_line_bundle} an arbitrary number of times? Having identified the range of possible line bundles in \eref{add_line_bundle} such that the maps are non-trivial, we may inquire whether they could each be added an arbitrary (i.e. infinite) number of times to the monad? In principle, such an infinite repetition of a line bundle leads to an equivalent monad, however as we will see in future sections, such additions do not increase the number of non-trivial target space dual theories. For the examples in consideration in this work, we begin with at most two defining relations in the geometric realization of the CY manifold (i.e. the index on the fields, $\Gamma^j$, runs only over $j=1,2$ in the present examples). Thus, at most $3$ line bundles can play a non-trivial role in the algorithm (i.e. $\Lambda^1$, $\Lambda^2$ and $P_1$ as in \eref{eg_switch1} and \eref{eg_switch2}). So there nothing to be gained by adding more than $3$ additional line bundles, $\mathcal L_i$. The utility of adding ``repeated entires" to the monad bundle will be further illustrated in Appendix \ref{appendixB}.
\subsubsection{Summary of choices for the target space dual geometries considered here}
It is worth briefly summarizing the classes of target space duals we will consider in this work. We will require that
\begin{itemize}
\item All target space duals constructed will increase the number of $U(1)$ symmetries in the GLSM by $1$ and take ${\cal B}=0$ in the notation of Section \ref{tsd_extrau1} to remain within the class of CICYs \cite{Candelas:1987kf}. This simple choice furthermore ensures that $X$ and $\widetilde{X}$ are related by geometric transitions (in general, so-called ``effective splits" for CICY configuration matrices \cite{Hubsch:1992nu}).
\item For the initial geometric pairs, $(X,V)$, we take $c_2(V)=c_2(TX)$ and it was established in \cite{Blumenhagen:2011sq} that this will guarantee that $c_2(\widetilde{V})=c_2(T\widetilde{X})$ in the target space dual geometry. The structure group of $V$ is embedded into a single $E_8$ factor of the $E_8 \times E_8$ theory. The second, unbroken $E_8$ factor to the gauge symmetry will not play a role in our analysis.
\item We choose bundles, $V$, defined as kernels: $V=Ker(f)$ and will allow for repeated line bundle entries in the monad, subject to the constraints described in the previous Section.
\end{itemize}
We turn now to a brief review of the structure of the ${\cal N}=1$ effective potential in the $4$-dimensional heterotic theory.
\section{The Target Space Potential: Supersymmetry Conditions and Vector Bundle Geometry}\label{potential_hym}
To preserve $\mathcal N = 1$ supersymmetry in the 4-dimensional effective theory, the geometry of the pair $(X,V)$ is constrained. The simplest class of solutions are found when $X$ is a Calabi-Yau threefold and the vector bundle $\pi: V \to X$ satisfies the Hermitian Yang-Mills equations \cite{Green:1987mn}:
\bea \label{HYM}
g^{a \bar{b}} F_{a \bar{b}}=0 \;,\; F_{ab}=0 \;,\;
F_{\bar{a}\bar{b}}=0 \ .
\eea
where $F$ is the gauge field strength
associated with the vector bundle $V$, and $a$ and $\bar{b}$ are holomorphic and anti-holomorphic indices on the
Calabi-Yau manifold. These equations involve all three types of singlet ``moduli" described in the previous section ($h^{1,1}(X)$,$h^{2,1}(X)$ and $h^1(X, End_0(V))$) and in general not all possible values of the CY and bundle moduli will correspond to a solution of \eref{HYM}.
It is straightforward to see that a solution to \eref{HYM} is directly linked to the minimum of the potential in the $4$-dimensional effective theory \cite{Witten:1985bz}. The argument below was given in \cite{Anderson:2011ty} and we briefly summarize it here:
The structure of the $4$-dimensional effective potential can be seen by focusing on several terms in the ten-dimensional effective action,
\begin{eqnarray}
\label{spartial}
S_{\textnormal{partial}} = -\frac{1}{2 \kappa_{10}^2} \frac{\alpha'}{4}\int_{{\cal M}_{10}} \sqrt{-g} \left\{ \textnormal{tr} (F^{(1)})^2+\textnormal{tr} (F^{(2)})^2 - \textnormal{tr} R^2 + \ldots \right\}
\end{eqnarray}
The notation here is standard \cite{Green:1987mn} with the field
strengths $F^{(1)}$ and $F^{(2)}$ being associated with the two $E_8$
factors in the gauge group. In addition to the action, the ten-dimensional
Bianchi Identity holds,
\begin{eqnarray}
dH = -\frac{3 \alpha'}{\sqrt{2}} \left( \textnormal{tr} F^{(1)} \wedge F^{(1)} +\textnormal{tr} F^{(2)} \wedge F^{(2)} - \textnormal{tr} R \wedge R \right) .
\end{eqnarray}
The associated integrability condition can be phrased as
\begin{eqnarray}
\label{intcond}
\int_{M_{6}} \omega \wedge \left( \textnormal{tr} \;F^{(1)} \wedge F^{(1)} + \textnormal{tr} \; F^{(2)} \wedge F^{(2)} - \textnormal{tr}\; R \wedge R \right) = 0 \;,
\end{eqnarray}
where $\omega$ is the K\"ahler form. Working
to lowest order, with a Ricci flat metric on a manifold of $SU(3)$
holonomy, equation \eref{intcond} then leads to
\begin{eqnarray}
\int_{M_{10}} \sqrt{-g} \left( \textnormal{tr} (F^{(1)})^2+\textnormal{tr} (F^{(2)})^2 - \textnormal{tr} R^2 +2 \,\textnormal{tr} (F^{(1)}_{a \bar{b}} g^{a \bar{b}})^2 +2\, \textnormal{tr} (F^{(2)}_{a \bar{b}} g^{a \bar{b}})^2 \right. \\ \nonumber \left.-4 \, \textnormal{tr} (g^{a \bar{a}} g^{b \bar{b}} F^{(1)}_{ab} F^{(1)}_{\bar{a}\bar{b}}) -4 \, \textnormal{tr} (g^{a \bar{a}} g^{b \bar{b}} F^{(2)}_{ab} F^{(2)}_{\bar{a}\bar{b}})\right)= 0~.
\end{eqnarray}
Using this relation in \eref{spartial}, the result is
\begin{eqnarray}
\label{finalintro}
S_{\textnormal{partial}}= - \frac{1}{2 \kappa_{10}^2} \alpha' \int_{{\cal M}_{10}} \sqrt{-g} \left\{ -\frac{1}{2} \textnormal{tr} (F^{(1)}_{a \bar{b}} g^{a \bar{b}})^2 - \frac{1}{2} \textnormal{tr} (F^{(2)}_{a \bar{b}} g^{a \bar{b}})^2 \right. \\ \nonumber \left. + \textnormal{tr} (g^{a \bar{a}} g^{b \bar{b}} F^{(1)}_{ab} F^{(1)}_{\bar{a}\bar{b}}) + \textnormal{tr} (g^{a \bar{a}} g^{b \bar{b}} F^{(2)}_{ab} F^{(2)}_{\bar{a}\bar{b}}) \right\}~.
\end{eqnarray}
By inspection, the terms included in \eref{finalintro} are a part of the ten-dimensional
theory which does not contain any four-dimensional derivatives. Thus, under dimensional reduction, they will contribute to the potential energy of the four-dimensional theory. In the case of a supersymmetry preserving field configuration, the terms in the integrand of \eref{finalintro}
vanish, corresponding exactly to a geometric moduli choice which solves \eref{HYM}. In this case, no potential is generated.
The detailed form of the potential generated in the ${\cal N}=1$, $4$-dimensional effective theory has been developed in recent work \cite{Anderson:2009sw,Anderson:2009nt,Anderson:2013qca,Anderson:2011ty}. By a theorem of Donaldson-Uhlenbeck and Yau \cite{duy1,duy2}, the first condition in \eref{HYM} is equivalent to the condition of Mumford (slope) poly-stability of $V$. The characterization of slope stability as well as tools to test whether a given bundle is stable are provided in Appendix \ref{AppendixA}. The condition of stability (equivalently \eref{HYM}) manifestly depends on the K\"ahler moduli through the presence of the metric. The K\"ahler class can in principle be smoothly varied so that the bundle transitions from being fully stable in one region of moduli space to poly-stable along a higher co-dimensional boundary (a so-called ``stability wall" \cite{Anderson:2009sw}) to full unstable (and hence supersymmetry breaking) for some region of moduli space. In the effective theory, this is associated with \emph{$D$-term breaking }of supersymmetry at stability walls. Along the stability walls the condition of poly-stability forces the bundle $V$ to become reducible (i.e. $V \to {\cal F} \oplus V/{\cal F}$), leading to the presence of enhanced, Green-Schwarz massive $U(1)$ symmetries in the effective theory. See \cite{Sharpe:1998zu,Anderson:2009nt,Anderson:2009sw,Anderson:2010ty}. In the neighborhood of a stability wall, the D-term can be written in the effective theory as
\begin{equation}\label{dterm_first}
D^{U(1)} \sim FI(t) - \frac{1}{2} \sum_{i} Q_iG_{P\bar{Q}} {C_i}^P{\bar{C}_i}^{\bar Q}
\end{equation}
where the first term is a moduli-dependent Fayet-Iliopolous parameter (depending on the K\"ahler moduli, $t^r$, $r=1,\ldots h^{1,1}(X)$ of $X$) and the second arises from matter fields, $C_i$, charged under the anomalous $U(1)$. Briefly, the form of the FI term to leading order is\footnote{See \cite{Anderson:2009nt} for a discussion of the one-loop correction to this FI term and its geometric interpretation.}
\begin{equation}\label{FI_slope}
FI(t) \sim \frac{\mu({\cal F})}{Vol(X)}
\end{equation}
where $Vol(X)$ is the volume of the CY 3-fold, $Vol(X)= \int_X J \wedge J \wedge J$, with $J$ is the K\"ahler form on $X$. Here $\mu({\cal F})=\frac{1}{rk({\cal F})}\int_X c_1({\cal F}) \wedge J \wedge J$ is a quasi-topological number called the ``slope" \cite{huybrechts_book}, associated to a crucial (geometrically de-stabilizing \cite{Anderson:2009nt}) sub-sheaf, ${\cal F}$. This last object, ${\cal F}$, is defined by the fact that near the stability wall, $V$ must be decomposed as $V \to {\cal F} \oplus \frac{V}{{\cal F}}$. See \cite{Anderson:2009nt} for a review and Appendix \ref{AppendixA} for further details.
\begin{figure}
\centering \includegraphics[width=0.6\textwidth]{Plot3Dpotentiala} \caption{\emph{An illustration of the D-term effective potential corresponding to \eref{dterm_first} (and a part of \eref{finalintro}) and geometrically realized by the bundle failing to be slope-stable for a sub-cone of K\"ahler moduli space. The white boundary line is referred to as a ``stability wall" in the K\"ahler cone. }} \label{fig:wall}
\end{figure}
Likewise, the second conditions in \eref{HYM} correspond geometrically to the condition that the bundle is holomorphic. It is clear that the conditions $F_{ab}=F_{\bar{a}\bar{b}}=0$ depend on the definition of holomorphic and anti-holomorphic coordinates on $X$. This choice clearly depends on the complex structure. For a fixed topology of the gauge fields, specific changes in the complex structure of the Calabi-Yau threefold may be such that these holomorphy equations no longer have a solution \cite{Anderson:2010mh,Anderson:2011cza}. In the effective theory, this obstruction is associated with the \emph{$F$-term breaking} of supersymmetry by the complex structure moduli. Geometrically, the flat directions in moduli space corresponding to holomorphic deformations of the complete geometry (manifold $+$ bundle) are described by the ``Atiyah class" of $V$ \cite{atiyah}. More precisely, the infinitesimal complex moduli of the heterotic theory are not the independent bundle moduli ($h^1(X, End_0(V))$) and complex structure moduli ($h^{2,1}(X)$) but rather the elements of the cohomology group
\begin{equation}\label{h1q}
H^1(X, {\cal Q})
\end{equation}
which describes the tangent space to \emph{the infinitesimal deformation space of the pair $(X,V)$}. Here ${\cal Q}$ is defined by \cite{atiyah}:
\begin{equation}
0 \to End_0(V) \to {\cal Q} \to TX \to 0
\end{equation}
and
\begin{equation}\label{mod_def}
h^1(X, {\cal Q})=h^1(X, End_0(V)) \oplus ker\{H^1(TX) \stackrel{[F^{1,1}]}{\longrightarrow} H^2(End_0(V))\}
\end{equation}
In general, the moduli in \eref{mod_def} contain the bundle moduli, but only a subset of the complex structure moduli of $X$. The scale at which the moduli are ``lifted" is generally the compactification scale (and thus, they should be excluded from the start in the zero-mode count given in \eref{mod_count}), however in some special cases this scale is lowered so that the moduli fixing due to holomorphy of the pair $(X,V)$ can be realized in the effective theory by the Gukov-Vafa-Witten superpotential \cite{Gukov:1999ya}:
\begin{equation}
W = \int_X H \wedge \Omega
\end{equation}
and its associated F-terms
\begin{equation}
F_{C_i}= \frac{\partial W}{\partial C_i} \sim \int_X \frac{\partial \omega^{3YM}}{\partial C_i}
\end{equation}
where $\omega^{3YM}= tr(F \wedge A - \frac{1}{3} A \wedge A \wedge A)$.
See \cite{Anderson:2010mh,Anderson:2011ty,Anderson:2011cza,Anderson:2013qca} for further details and computational tools. These moduli considerations have recently been extended to include heterotic non-K\"ahler vacua \cite{Anderson:2014xha,delaOssa:2014cia,delaOssa:2015maa,delaOssa:2015rhu} and a more detailed understanding of moduli space metrics \cite{Candelas:2016usb,McOrist:2016cfl}. The geometry and field theory of these F-terms and holomorphy are explored further in Section \ref{Fterm_sec}.
\section{D-terms, Bundle Stability, and Duality}\label{dterms}
In this section we will begin to explore the role of a non-trivial D-term condition in target space dual chains. One complication in this process is that we do not have a concrete map (i.e. an explicit isomorphism) between the two moduli spaces of the $(0,2)$ theories. In general, the presence of a non-trivial D-term potential does not lift entire flat directions (i.e. change the dimension of the vacuum space) but merely change its shape (see Figure \ref{fig:wall}). At the level of the effective theory, this is a realization of the fact that the vacuum space can be written as a solution to the F-terms, modulo D-term constraints \cite{Luty:1995sd}. In the bundle geometry, this same notion plays a crucial role in the definition of bundle moduli spaces, geometric invariant theory (GIT), quotients, etc \cite{huybrechts_book}.
To simplify our analysis then, we will take as a first example an extreme case in which the bundle is only stable for a sub-cone of the K\"ahler moduli space of \emph{strictly smaller dimension}. This will make it possible to simply count and observe whether or not there is a drop in the total degrees of freedom on both sides of the duality. For the details of how to analyze the slope-stability of a vector bundle and how to engineer interesting monad examples, see Appendix \ref{AppendixA}. Using these tools, we can select a pair, $(X,V)$, of CY manifold and vector bundle $\pi: X \to V$ with a D-term which ``lifts" whole directions of K\"ahler moduli space. Below, we will start with a simple example with $h^{1,1}=2$ and a bundle over it which contains \emph{two} destabilizing sub-sheaves, ${\cal F}_1, {\cal F}_2$. These sheaves, as described in the previous Section, appear in the associated D-terms as in \eref{FI_slope} and in the slope-stability analysis of $V$ as described in Appendix \ref{AppendixA}. Here, the sub-sheaves are chosen such that
\begin{equation}\label{twosheaves}
c_1({\cal F}_1)= - c_{1}({\cal F}_2)
\end{equation}
Since a bundle is only stable if $\mu(V)> \mu({\cal F})$ for all ${\cal F} \subset V$ (see Appendix \ref{AppendixA}), it is clear that with two de-stabilizing sub-sheaves as in \eref{twosheaves}, and $c_1(V)=0$, $V$ cannot be properly stable at all. Indeed it can only solve the Hermitian-Yang-Mills equations, \eref{HYM}, on the locus in K\"ahler moduli space for which
\begin{equation}\label{slopey_cond}
\mu({\cal F}_1)= - \mu({\cal F}_2)=\mu(V)=0
\end{equation}
where $V$ itself must become reducible as a poly-stable bundle $V \to V_1 \oplus V_2\oplus \ldots$.
For the CY manifold, $X$, consider the simple degree $\{2,4\}$ hypersurface in $\mathbb P^1 \times \mathbb P^3$ with Hodge numbers $h^{1,1}(X)=2$ and $h^{2,1}(X)=86$. Expanding the second Chern class of the manifold in a basis of $\{1,1\}$ forms as $c_2(TX)_r=d_{rst}{c_2(TX)}^{st}$ (where $d_{rst}$ are the triple intersection numbers and $r=1, \ldots h^{1,1}$) we have $c_2(TX)^r=\{24,44\}$ (see Appendix \ref{bundle_man_top}). Finally, to complete the geometry, a rank $5$ monad bundle, $V$, can be chosen over $X$, written as a kernel:
\begin{equation}\label{moneg}
0 \to V \to B \stackrel{F}{\longrightarrow} C \to 0
\end{equation}
where $V=ker(F)$ and $B$ and $C$ are sums of line bundles over $X$ as in \eref{full_monad}. In GLSM language, the manifold and bundle are given by the following data:
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccc}
1& 1& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1
\end{array}&\begin{array}{c}
-2 \\
-4
\end{array}&\begin{array}{cccccccc}
1& -1& 0& 0& 2& 1& 1& 2 \\
-1& 1& 1& 1& 1& 2& 2& 2
\end{array}&\begin{array}{ccc}
-3& -1& -2 \\
-2& -4& -3
\end{array}\\
\hline
\end{array} \label{bundle2a}\end{aligned}\end{equation}
By inspection, $c_1(V)=0$ and using the tools of \cite{Anderson:2008ex} it can be verified that
\begin{equation}
c_2(V)=c_2(TX)=\{24,44\}
\end{equation}
and the third Chern class is $c_3(V)=-80$. Finally, a straightforward, but lengthy calculation yields the following naive total for the uncharged (geometric) moduli of the theory:
\begin{equation}\label{naive_count}
dim({\cal M}_{0})=h^{1,1}(X)+h^{2,1}(X)+h^1(X,End_0(V))=2+86+340=428
\end{equation}
Thus far we are in familiar territory. As described in Section \ref{potential_hym}, it is now possible to go further and explore the theory in more depth by considering the structure of its effective potential. As will be shown in the next subsections, the bundle in \eref{bundle2a} is exactly of the form required to generate a non-trivial D-term constraint and a corresponding reduction in the number of flat directions. We will begin by analyzing the effective theory arising from this geometry along the stable ray in the K\"ahler cone (the ``stability wall" \cite{Sharpe:1998zu,Anderson:2009sw,Anderson:2009nt}).
The non-vanishing triple intersection numbers of the CY 3-fold are $d_{122}=4$ and $d_{222}=2$. Expanding the K\"ahler form in a basis of $\{1,1\}$ forms as $J=t^r J_r$ (with $t^r$ the K\"ahler moduli), it is clear that the condition in \eref{slopey_cond} restricts the theory to the locus in K\"ahler moduli space for which $-8t^1 t^2+2 (t^2)^2=0$. Choosing perturbative solutions in the interior of the K\"ahler cone requires $t^r >0$ for all $r$, and thus the ``stability wall" in this case is the one-dimensional ray in K\"ahler moduli space where
\begin{equation}\label{constrained_t}
t^2=4 t^1~~~\text{equivalently}~~~ s_2=s_1
\end{equation}
in the dual K\"ahler coordinates with $s_r=d_{rst}t^st^t$. The stable subcone for $V$ is shown in Figure \ref{fig:firstwall}.
\begin{figure}
\centering \includegraphics[width=0.4\textwidth]{Slide1} \caption{\emph{The stable ray in K\"ahler moduli space shown for the monad bundle in \eref{bundle2a}. Along this co-dimension 1 subcone the bundle takes the poly-stable form shown in \eref{decomp_on_wall}.}} \label{fig:firstwall}
\end{figure}
This bundle is an example of one where the naive degree-of-freedom counting given in \eref{mod_count} fails to capture important structure. In fact, the dimension of the singlet moduli space is not the sum given in \eref{naive_count}. Instead, with the stability constraint in \eref{constrained_t} it is reduced to
\begin{equation}\label{real_count}
dim({\cal M}_1)=dim({\cal M}_{0})-1=427
\end{equation}
This type of example provides a direct new probe of target space duality. Will the dual theories also reflect this reduction in degrees of freedom? What is the structure of the potential in the dual theory and will the moduli lifting happen through D-terms or F-terms? Before answering these questions however, we will see below that in fact the detailed structure of the theory on the stability wall provides many new features that can be compared.
\subsection{Effective theory near the stability wall}
As described in Appendix \ref{AppendixA}, the presence of each of the line bundles with negative entries in $B$ (the middle term of the monad) in \eref{moneg} implies the existence of a particular sub-sheaf of $V$. In this case, following \eref{mon_subsheaf} we see that $V$ can be written in terms of its de-stabilizing sub-sheaves as
\begin{align}
&0 \to Q_4 \to V \to {\cal O}(1,-1) \to 0 \\
&0 \to \tilde{Q}_4 \to V \to {\cal O}(-1,1) \to 0
\end{align}
where the rank $4$ sheaves, $Q_4$ and $\tilde{Q}_4$ are described via the monads
\begin{align}\label{qdefs}
& 0 \to Q_4 \to {\cal O}(-1,1)\oplus {\cal O}(0,1)^{\oplus 2}\oplus {\cal O}(2,1) \oplus {\cal O}(1,2)^{\oplus 2} \oplus {\cal O}(2,2) \to {\cal O}(3,2) \oplus {\cal O}(1,4) \oplus {\cal O}(2,3) \to 0 \nonumber \\
& 0 \to \tilde{Q}_4 \to {\cal O}(1,-1)\oplus {\cal O}(0,1)^{\oplus 2}\oplus {\cal O}(2,1) \oplus {\cal O}(1,2)^{\oplus 2} \oplus {\cal O}(2,2) \to {\cal O}(3,2) \oplus {\cal O}(1,4) \oplus {\cal O}(2,3) \to 0
\end{align}
Since $Q_4$ and $\tilde{Q}_4$ are sub-sheaves of $V$, with non-trivial first Chern classes, it is clear that supersymmetry will only be preserved if the bundle is poly-stable. This implies a decomposition of $V$ into each de-stabilizing sub-sheaf as $V \to {\cal F} \oplus V/{\cal F}$. In this case, for the two subsheaves of interest:
\begin{align}
& V \to Q_4 \oplus {\cal O}(1,-1) \\
&V \to \tilde{Q}_4 \oplus {\cal O}(-1,1)
\end{align}
Moreover, these descriptions are only compatible if $Q_4$ and $\tilde{Q}_4$ decompose still further as
\begin{align}
Q_4 \to U_3 \oplus {\cal O}(-1,1) \\
\tilde{Q}_4 \to U_3 \oplus {\cal O}(1,-1)
\end{align}
for \emph{the same} rank $3$ sheaf $U_3$ with $c_1(U_3)=0$. By inspection of \eref{qdefs} it is straightforward to verify that such a reducible locus in the moduli space of $Q_4$ and $\tilde{Q}_4$ does indeed exist and here $U_3$ is in fact locally free (i.e. an $SU(3)$ bundle) itself described as a monad:
\begin{equation}
0 \to U_3 \to {\cal O}(0,1)^{\oplus 2}\oplus {\cal O}(2,1) \oplus {\cal O}(1,2)^{\oplus 2} \oplus {\cal O}(2,2) \to {\cal O}(3,2) \oplus {\cal O}(1,4) \oplus {\cal O}(2,3) \to 0
\end{equation}
One less obvious observation can be made using the tools of Appendix \ref{AppendixA}: the bundle $U_3$ is properly stable for the stability wall, $t^2=4t^1$. Thus, in summary, the only locus in the moduli space of the monad bundle $V$, given in \eref{bundle2a} for which the bundle is poly-stable and supersymmetry is preserved is when the map $f$ is chosen so that the bundle is reducible:
\begin{equation}\label{decomp_on_wall}
V \to U_3 \oplus L \oplus L^{\vee}
\end{equation}
with $L={\cal O}(1,-1)$.
The consequences of this decomposition are immediate for the effective target space theory. Since we began with a seemingly indecomposable $SU(5)$ bundle the natural expectation would have been to find a $4$-dimensional $SU(5)$ effective theory (since $SU(5) \times SU(5)$ is a maximal subgroup of $E_8$). However, once the D-term is taken into account we see that the structure group of the bundle on the stability wall has decreased to $S[U(1) \times U(1)] \times SU(3) \simeq U(1) \times SU(3)$ which has a commutant within $E_8$ of $SU(6) \times U(1)$. The $U(1)$ symmetry is in general Green-Schwarz massive \cite{Lukas:1999nh} and as described in Section \ref{potential_hym}, leads to a non-trivial D-term constraint. Thus \emph{the presence of a stability wall in the theory has led not only to a reduction in K\"ahler moduli, but also to a non-Abelian enhancement of the $4$-dimensional gauge symmetry}.
The charged matter spectrum can be determined by beginning with the decomposition \cite{Slansky:1981yr}
\begin{equation}
{\bf 248}_{E_8}\rightarrow \left[({\bf 3},{\bf 1},{\bf 1})\oplus ({\bf 1},{\bf 8},{\bf 1})\oplus ({\bf 1},{\bf 1},{\bf 35})\oplus ({\bf 1},{\bf 3},\overline{\bf 15})
\oplus ({\bf 1},\overline{\bf 3},{\bf 15})
\oplus ({\bf 2},{\bf 3},{\bf 6})\oplus ({\bf 2},\overline{\bf 3},\overline{\bf 6})\oplus ({\bf 2},{\bf 1},{\bf 20})\right]\label{248dec}
\end{equation}
of the adjoint of $E_8$ into $SU(2) \times SU(3) \times SU(6)$. Then the further reduction of $SU(2) \to S[U(1) \times U(1)]$ allows us to read off the final charged matter content of the $SU(6) \times U(1)$ theory\footnote{Note that locally $S[U(n) \times U(1)] \simeq SU(n) \times U(1)$. Globally, however, there is a different normalization on the $U(1)$ which is comprises part of the commutant of this symmetry within $E_8$. Because it will not affect the vacuum analysis in this section, for simplicity we will follow the local charge normalization conventions of \cite{Slansky:1981yr} for all $U(1)$ charges in this work.}. The result is given in Table \ref{table1}.
\begin{table}
\begin{centering}
\begin{tabular}{|c|c|c|c|c|c|}\hline
Field & Cohom. & Multiplicity & Field & Cohom. & Multiplicity \\ \hline
${\bf 1}_{+2}$ & $H^1(L \otimes L)$ & 0 &${\bf 1}_{-2}$ & $H^1(L^{\vee} \otimes L^{\vee})$ & $10$ \\ \hline
${\bf 15}_0$ & $H^1({U_3}^{\vee})$ & 0 & $\overline{ {\bf 15}}_0$ & $H^1(U_3)$ & 80 \\ \hline
${\bf 20}_{+1}$ & $H^1(L)$ & 0 & ${\bf 20}_{-1}$ & $H^1(L^{\vee})$ & 0 \\ \hline
${\bf 6}_{+1}$ & $H^1(L \otimes U_3)$ & 72 & ${\bf 6}_{-1}$ & $H^1(L^{\vee} \otimes U_3)$ & 90 \\ \hline
${\bf \overline{6}}_{+1}$ & $H^1(L \otimes {U_{3}}^{\vee})$ & 0 & $\overline{\bf{6}}_{-1}$ & $H^1(L^{\vee} \otimes {U_3}^{\vee})$ & 2 \\ \hline
${\bf 1}_0$ & $H^1(U_3 \otimes {U_3}^{\vee})$ & 166 & & & \\
\hline
\end{tabular}
\caption{\it Particle content of the $SU(6) \times U(1)$ theory associated to the bundle \eref{bundle2a} along its reducible and poly-stable locus $V= {\cal O}(-1,1)\oplus {\cal O}(1,-1) \oplus U_3$ (i.e. on the stability wall given by \eref{constrained_t} and shown in Figure \ref{fig:firstwall}).}\label{table1}
\end{centering}
\end{table}
In addition to the D-terms associated to the non-Abelian $SU(6)$ gauge symmetry there is a D-term associated to the Green-Schwarz massive $U(1)$ which is self commuting in the $S[U(1) \times U(1)]$ factor in $E_8$. Written in the expansion parameters of heterotic M-theory, this takes the schematic form (see \cite{Anderson:2009nt} for details):
\begin{equation}
D^{U(1)} \sim \frac{3}{16}\frac{\epsilon_S {\epsilon_R}^2 \mu(\cal{F})}{{\kappa_4}^2 Vol(X)}- \frac{1}{2} \sum_{i} Q_iG_{P\bar{Q}} {C_i}^P{\bar{C}_i}^{\bar Q}
\end{equation}
where $Vol(X)$ is the volume of the CY threefold $X$, $G_{P\bar{Q}}$ is the field space metric, and $C_i$ denotes any of the matter fields charged under the Green-Schwarz massive $U(1)$ symmetry (with charge $Q_i$) appearing in Table \ref{table1}. Once again, we note the key feature of this example: because of the non-trivial D-term above, when $\langle C \rangle =0$ (corresponding to the monad in \eref{bundle2a}), the D-term non-trivially constrains the K\"ahler moduli through the condition that $\mu({\cal F})=0$ and the moduli of the theory are not those counted in \eref{naive_count} but rather the $dim({\cal M})=427$ moduli in \eref{real_count}.
We will explore the vacuum branch structure of this theory in more detail in Section \ref{branch_sec}, but for now we turn to the target space dual theories.
\subsection{Target Space Duals of Theories with D-terms}
Following the target space duality algorithm as described in Section \ref{the_algorithm} (including repeated line bundles as in Section \ref{repeated_line_bundles}) it is possible to generate 16 target space duals for the pair $(X,V)$ given in \eref{bundle2a} in which the new manifolds, $\widetilde{X}_i$, are still described as CICYs. The complete chain of target space dual theories is given in Appendix \ref{appendixB}. To investigate these results, a simple starting point is given below -- a dual for which the Calabi-Yau manifold, $X$, is related to the original threefold by a geometric transition.
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt} \begin{array}{cccccccc}
0& 0& 0& 0& 0& 0& 1& 1 \\
1& 1& 0& 0& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1& 0& 0
\end{array}&\begin{array}{cc}
-1& -1 \\
-2& 0 \\
-2& -2
\end{array}&\begin{array}{cccccccc}
0& 0& 1& 0& 0& 0& 0& 0 \\
1& -1& 0& 0& 2& 1& 1& 2 \\
-1& 1& -1& 1& 3& 2& 2& 2
\end{array}&\begin{array}{ccc}
0& 0& -1 \\
-3& -1& -2 \\
-2& -4& -3
\end{array}\\
\hline
\end{array}\end{aligned}\label{bundle2a.1}\end{equation}
In view of the results of the previous subsection, obvious questions include:
\begin{enumerate}
\item Does $(\widetilde{X},{\widetilde V})$ give rise to a stability wall?
\item Is $dim(\widetilde{\cal{M}}_{1})$ also reduced in dimension compared to $dim(\widetilde{\cal{M}}_{0})$? Are the number of lifted moduli the same?
\item Does the structure group of ${\widetilde V}$ also reduce, leading to a $4$-dimensional non-Abelian enhancement of symmetry (i.e. $SU(6)$)?
\item Do the charged matter spectra of the two theories match?
\end{enumerate}
We will address each of these questions in turn.
First, the complete intersection CY (CICY) manifold and rank $5$ vector bundle given in \eref{bundle2a.1} satisfies $h^{1,1}(\tilde{X})=3$, $h^{2,1}(\tilde{X})=55$ and
\begin{equation}
c_2(\tilde{V})=c_2(T\tilde{X})=\{24,24,44\}
\end{equation}
and the third Chern class is $c_3(\tilde{V})=-80$ as expected. The comparison to \eref{bundle2a} can begin with the uncharged (geometric) moduli of the theory:
\begin{equation}\label{naive_count_mirror}
dim({\cal \widetilde{M}}_{0})=h^{1,1}(\widetilde{X})+h^{2,1}(\widetilde{X})+h^1(X,End_0(\widetilde{V}))=3+55+370=428
\end{equation}
Once again, it is important to observe that this naive moduli count may not be the correct zero mode count for the theory. As illustrated in the previous Section, this count is only preliminary since we have not yet taken into account any non-trivial D-term potential associated to the geometry above.
By inspection, the negative charges in \eref{bundle2a.1} indicate that the bundle will possess stability walls (see Appendix \ref{AppendixA}). For the bundle in \eref{bundle2a.1} it can be verified that there are now \emph{three} de-stabilizing subsheaves rather than the two of \eref{bundle2a}. Explicitly, these are given by
\begin{align}
& 0 \to {\cal \tilde{F}}_1 \to \tilde{V} \to {\cal O}(0,1,-1) \to 0 & c_1({\cal \tilde{F}}_1)=(0,-1,1) \\
& 0 \to {\cal \tilde{F}}_2 \to \tilde{V} \to {\cal O}(0,-1,1) \to 0 & c_1({\cal \tilde{F}}_2)=(0,1,-1) \\
& 0 \to {\cal \tilde{F}}_3 \to \tilde{V} \to {\cal O}(1,0,-1) \to 0 & c_1({\cal \tilde{F}}_3)=(-1,0,1)
\end{align}
To proceed, it can be noted that the non-vanishing triple intersection numbers for $\widetilde{X}$ are
\begin{equation}
d_{123}=d_{133}=d_{233}=4~~~,~~~d_{333}=2
\end{equation}
\begin{figure}
\centering \includegraphics[width=0.4\textwidth]{Slide2} \caption{\emph{The $2$-dimensional stable sub-cone in K\"ahler moduli space shown for the monad bundle in \eref{bundle2a.1}. Along this co-dimension 1 sub-cone the bundle takes the poly-stable form shown in \eref{mirror_walltheory}.}} \label{fig:secondwall}
\end{figure}
Combining this with the stability conditions that $\mu({\cal \tilde{F}}_i) \leq 0$ for $i=1,2,3$ leads to the constraint that (for $t^i >0$ in the interior of the K\"ahler cone) that $t^2>t^1$ and $t^3= 2t^2+2\sqrt{t^1 t^2 + (t^2)^2}$. In the dual K\"ahler variables this takes the form of a simple plane the 3-dimensional K\"ahler cone of $\widetilde{X}$:
\begin{equation}\label{mirror_wall}
s_3 < s_1~~~\text{and}~~~s_2=s_3
\end{equation}
The 2-dimensional stable plane is shown in Figure \ref{fig:secondwall}.
Although the number of de-stabilizing subsheaves differs in the target space dual theory, the total number of moduli has been precisely matched! That is, in this case we have
\begin{equation}
dim({\cal \widetilde{M}}_1)=dim({\cal \widetilde{M}}_0)-1=427
\end{equation}
exactly as required for a correspondence of the target space theories. Thus, the first two questions posed in the previous section are answered: \emph{The presence of the stability wall (i.e. the locus of enhanced symmetry) and the reduction in moduli due to the D-term constraint are both exactly mirrored in the target space dual theory.}
To see if the enhancement of non-Abelian symmetry is the same as that for the bundle given in \eref{bundle2a} it should be noted that the bundle in \eref{bundle2a.1} has a poly-stable decomposition along the $2$-dimensional stability wall given by \eref{mirror_wall} as
\begin{equation}\label{mirror_walltheory}
\tilde{V} \to (\tilde{L} \oplus {\tilde{L}}^{\vee})\oplus \widetilde{U}_3
\end{equation}
with $\tilde{L}={\cal O}(0,1,-1)$ and the rank $3$ bundle $\tilde{U}_3$ is given by the monad
{\small \begin{equation}
0 \to \tilde{U}_3 \to {\cal O}(1,0,-1)\oplus {\cal O}(0,0,1) \oplus {\cal O}(0,2,3)\oplus {\cal O}(0,1,2)^{\oplus 2} \oplus {\cal O}(0,2,2) \to {\cal O}(0,3,2) \oplus {\cal O}(0,1,4) \oplus {\cal O}(1,2,3) \to 0
\end{equation}}
The presence of the negative line bundle, ${\cal O}(1,0,-1)$ in $\widetilde{U}_3$ indicates that there is a sub-locus in moduli space where $\widetilde{V}$ can be decomposed still further (namely, where $\widetilde{U}_3 \to {\cal O}(1,0,-1) \oplus \widetilde{U}_2$). However, unlike the decomposition in \eref{mirror_walltheory} which is \emph{required} by stability of $\widetilde{V}$, this further decomposition is not generic and would correspond to some further tuning in bundle moduli of \eref{bundle2a} in the dual theory.
We are now in a position to answer yet another of our questions. From the form of \eref{mirror_walltheory} it is clear that the \emph{non-Abelian symmetry enhancement is identical}. Once again, on the wall, the $4$-dimensional gauge symmetry enhances from $SU(5) \to SU(6)$. Furthermore, in this case the number of Green-Schwarz massive $U(1)$ symmetries appears to to be identical\footnote{Although the Green-Schwarz massive $U(1)$ symmetry is identical in this example, in general one would not expect that a discrepancy in the number of such symmetries should signal a breakdown of duality. The mass scale of the massive $U(1)$ symmetries is only just below the compactification scale \cite{Anderson:2009sw,Anderson:2009nt} and care should be taken with how they are included in the effective theories. In general, while the presence of the Green-Schwarz massive $U(1)$ can affect the structure of the theory (for example in Yukawa textures and the structure of the matter field K\"ahler potential \cite{Anderson:2010tc,Anderson:2011ns,Anderson:2013xka,Anderson:2012yf,Anderson:2014hia}), it should not be expected that the massive $U(1)$ gauge bosons themselves should be visible in the low-energy effective theory and the difference noted above does not indicate a failure of duality.}.
\begin{table}
\begin{centering}
\begin{tabular}{|c|c|c|c|c|c|}\hline
Field & Cohom. & Multiplicity & Field & Cohom. & Multiplicity \\ \hline
${\bf 1}_{+2}$ & $H^1(\tilde{L} \otimes \tilde{L})$ & 0 &${\bf 1}_{-2}$ & $H^1(\tilde{L}^{\vee} \otimes \tilde{L}^{\vee})$ & $10$ \\ \hline
${\bf 15}_0$ & $H^1({\tilde{U}_3}^{\vee})$ & 0 & $\overline{ {\bf 15}}_0$ & $H^1(\tilde{U}_3)$ & 80 \\ \hline
${\bf 20}_{+1}$ & $H^1(\tilde{L})$ & 0 & ${\bf 20}_{-1}$ & $H^1(\tilde{L}^{\vee})$ & 0 \\ \hline
${\bf 6}_{+1}$ & $H^1(\tilde{L} \otimes \tilde{U}_3)$ & 72 & ${\bf 6}_{-1}$ & $H^1(\tilde{L}^{\vee} \otimes \tilde{U}_3)$ & 90 \\ \hline
${\bf \overline{6}}_{+1}$ & $H^1(\tilde{L} \otimes {\tilde{U}_{3}}^{\vee})$ & 0 & $\overline{\bf{6}}_{-1}$ & $H^1(\tilde{L}^{\vee} \otimes {\tilde{U}_3}^{\vee})$ & 2 \\ \hline
${\bf 1}_0$ & $H^1(\tilde{U}_3 \otimes {\tilde{U}_3}^{\vee})$ & 196 & & & \\
\hline
\end{tabular}
\caption{\it Particle content of the $SU(6) \times U(1)$ theory associated to the bundle \eref{bundle2a.1} along its reducible and poly-stable locus $V=\tilde{L} \oplus {\tilde{L}}^{\vee} \oplus \tilde{U}_3$ (i.e. on the \emph{stability wall} given by \eref{mirror_wall} and shown in Figure \ref{fig:secondwall}).}\label{table2}
\end{centering}
\end{table}
To address the final question, that of whether the $SU(6)$-charged matter spectra of the theories agree, the cohomology of the reducible bundle in \eref{bundle2a.1} (equivalently \eref{mirror_walltheory}) can be computed. The results are given in Table \ref{table2}. By inspection, it can be verified that the number of fields in the ${\bf 15}, {\bf 20}, {\bf 6}$, etc representations is identical to the particle content given in Table \ref{table1}. Moreover, the $U(1)$-charges under the GS massive $U(1)$ symmetries are identical and would appear to lead to identical constraints from gauge invariance on the ${\cal N}=1$ $4$-dimensional theory!\footnote{It would be illuminating to probe the effect of higher-codimensional loci such as that determined by $\widetilde{U}_3 \to {\cal O}(1,0,-1) \oplus \widetilde{U}_2$ and its effect on Yukawa couplings and compare this with special loci in the moduli space of the target space dual. At the moment however, this is difficult to accomplish without an explicit moduli map between $(X,V)$ and $(\widetilde{X},\widetilde{V})$.}.
Beyond the geometry listed in \eref{bundle2a.1}, there are another $15$ target space dual theories in the chain related to \eref{bundle2a}. These are listed explicitly in Appendix \ref{appendixB} and share many of the features of the dual theory described above. In each case we can verify the existence of a stability wall in the new theory and in each case
\begin{equation}
dim({\cal \widetilde{M}}_1)=427
\end{equation}
However, in 6 out of 16 we find a surprising feature which is that the stability condition forces the theory to the boundary of the K\"ahler cone. Such a point is a metric singularity and corresponds to a perturbative/non-perturbative duality and will be explored in Section \ref{non_pert_duals}.
\subsection{Branch Structure in the vacuum space of the dual theories}\label{branch_sec}
An intriguing aspect of theories with stability walls is the rich branch structure in the vacuum space of the theories. As described in \cite{Anderson:2009nt,Anderson:2009sw,Anderson:2010ty} the distinct branches of the vacuum moduli space near the wall correspond to distinct infinitesimal deformations at the level of the bundle geometry. These walls also provide non-trivial information about discrete symmetries and Yukawa textures in the effective target space theories away from the wall \cite{Kuriyama:2008pv,Anderson:2010tc}.
As a result of this, heterotic theories with stability walls provide a further detailed check of the target space duality correspondence. We have seen in the previous sections that a stability wall in one theory, $(X,V)$ is reproduced in its target space dual $(\widetilde{X},\widetilde{V})$ and that the charged and singlet matter of the two wall theories matches exactly. In the next sections we briefly explore the branch structures in these two moduli spaces away from the stability walls (i.e. away from the locus of enhanced symmetry. To begin, consider the bundle given in \eref{bundle2a} and its associated $SU(6) \times U(1)$ theory with the particle spectrum given in Table \ref{table1}.
We will consider deformations away from the stability wall that lead to indecomposable, properly stable, rank $5$ vector bundles. Such a deformation of bundle geometry would correspond in the $4$-dimensional theory to a Higgsing $SU(6) \to SU(5)$ of the effective theory along a flat direction. Several such directions can be readily identified. First, there are two non-trivial D-term constraints on the theory corresponding to the Green-Schwarz massive $U(1)$, and to the $U(1)$ associated with $SU(5) \times U(1) \subset SU(6)$ (${\bf 6} \to {\bf 1}_{-5} + {\bf 5}_{+1}$). Note we will take the Vevs of all fields charged under $SU(5) \subset SU(6)$ to be vanishing.
\subsubsection{Branches for $(X,V)$}\label{branch_in_v}
We begin with the geometry in \eref{bundle2a}. Taking into account the spectrum of Table \ref{table1} and adding a second $U(1)$ charge subscript for the $U(1) \subset SU(6)$ described above, the D-terms take the schematic form:
\begin{align}
&D^{U(1)}_{GS} \sim \frac{3}{16}\frac{\epsilon_S {\epsilon_R}^2 \mu(L^{\vee})}{{\kappa_4}^2 {\cal V}}- \frac{1}{2}\left((-2)|C_{-2,0}|^2+(+1)|C_{+1,-5}|^2+(-1)|C_{-1,-5}|^2+(-1)|C_{-1,+5}|^2\right)\label{du1} \\
&D^{U(1)}_{SU(6)}\sim \frac{1}{2}\left((-5)|C_{+1,-5}|^2+(-5)|C_{-1,-5}|^2+(+5)|C_{-1,+5}|^2\right) \label{dsu6}
\end{align}
In the above, sums over multiplicities of each type of term are suppressed and for each category of charged matter field in Table \ref{table1} we abbreviate $\sum_{i} Q_iG_{P\bar{Q}} {C_i}^P{\bar{C}_i}^{\bar Q}\simeq Q|C|^2$. In considering the vacuum structure, note that \eref{dsu6} leads to a condition of the form
\begin{equation}\label{vevsbranch1}
\langle C_{-1,+5} \rangle^2 \sim \langle C_{-1,-5} \rangle^2+ \langle C_{+1,-5} \rangle^2
\end{equation}
However, it is also clear that substituting this constraint into to \eref{du1} we find that D-flatness requires that
\begin{equation}\label{vevsbranch2}
\frac{3}{16}\frac{\epsilon_S {\epsilon_R}^2 \mu(L^{\vee})}{{\kappa_4}^2 Vol(X)}+\frac{1}{2} \left(\langle C_{-2,0}\rangle^2 +\langle C_{-1,-5} \rangle^2\right) \sim 0
\end{equation}
Since this involves only matter fields with negative charge under the first (GS massive) $U(1)$ symmetry, it is clear that \emph{all vacuum solutions breaking $SU(6) \to SU(5)$ with $\langle C \rangle \neq 0$ must involve a non-trivial FI term with $\mu(L^{\vee})<0$}.
The indecomposable, $SU(5)$ bundle corresponding to this deformation will be properly slope-stable only in the region of K\"ahler moduli space for which $\mu(L^{\vee}) <0$. This chamber structure is depicted in Figure \ref{fig:branchchamber1}.
\begin{figure}
\centering \includegraphics[width=0.4\textwidth]{Slide3} \caption{\emph{The stable sub-cone in K\"ahler moduli space shown for the $SU(5)$ monad bundle in \eref{branch2a}. This corresponds to the branch of vacuum space described by \eref{vevsbranch1} and \eref{vevsbranch2}.}} \label{fig:branchchamber1}
\end{figure}
To describe this bundle as a monad (and hence, build the associated GLSM) we need two facts: 1) the new monad bundle corresponding to this branch of the vacuum space must admit a point in its moduli space for which it can be decomposed as in \eref{decomp_on_wall} and 2) By the results of Appendix \ref{AppendixA}, if both ${\cal O}(1,-1)$ and ${\cal O}(-1,1)$ appear in the monad, it cannot be stable in the region shown in Figure \ref{fig:branchchamber1}.
A useful technical observation is that to accommodate both these facts, it is clear that one of the line bundles $L$ or $L^{\vee}$ must be ``replaced" in the monad with an alternate description. To that end, we observe that ${\cal O}(-1,1)$ itself can be written as a monad:
\begin{equation}\label{lnew}
0 \to L_{new} \to \mathcal O(0,1)^{\oplus 2} \stackrel{g}{\longrightarrow} \mathcal O(1,1) \to 0
\end{equation}
This monad describes $L_{new}$ as a rank 1 sheaf given by the kernel of the map $g$, with $c_1(L_{new})=-J_1 + J_2$. Since line bundles on CY 3-folds are classified by their first Chern classes, it remains to show only that \eref{lnew} describes a locally free sheaf. To this end, recall that the kernel of a monad sequence defines a locally free sheaf only if the map satisfies the transversality condition described in Section \ref{sec:GLSM} (i.e. the condition that the rank of $g$ does not drop anywhere on $X$). In this case, the map, $g$, can be written as a vector of two linear functions of the ambient $\mathbb{P}^1$ coordinates which manifestly do not vanish simultaneously on $X$, thus $L_{new}\simeq {\cal O}(-1,1)$.
As a result, the new branch, corresponding to deforming off of the stability wall is given by the following monad:
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccc}
1& 1& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1
\end{array}&\begin{array}{c}
-2 \\
-4
\end{array}&\begin{array}{ccccccccc}
1& 0& 0& 0& 0& 2& 1& 1& 2 \\
-1& 1& 1& 1& 1& 1& 2& 2& 2
\end{array}&\begin{array}{cccc}
-1& -3& -1& -2 \\
-1& -2& -4& -3
\end{array}\\
\hline
\end{array} \label{branch2a}\end{aligned}\end{equation}
The $SU(5)$ charged matter spectrum associated to this bundle can be computed directly and is shown in Table \ref{table3}. Note that here
\begin{equation}\label{branchy_total}
h^{1,1}(X)+h^{2,1}(X)+h^1(End_0(V))=2+86+338=426
\end{equation}
In this case \emph{there are no obstructions in moduli} and $dim({\cal M}_1)=dim ({\cal M}_0)=426$ since despite the presence of the stability wall forming a sub-chamber of K\"ahler moduli space, there are still two free degrees of freedom associated to the K\"ahler moduli. More importantly, the count in Table \ref{table3} matches exactly with what is expected from the Higgsing in the vacuum analysis above with precisely two degrees of freedom are removed by the D-terms above. This agrees with comparison to \eref{real_count} on the branch in moduli space corresponding to $\langle C \rangle \neq 0$ (i.e. ``gluing" the reducible bundle in \eref{decomp_on_wall} back into the irreducible $SU(5)$ bundle given in \eref{branch2a}). In addition, in the Higgsing $SU(6) \to SU(5)$ a number of ${\bf 5},{\bf \bar{5}}$ pairs can become massive\footnote{We thank J. Gray for useful suggestions on computer algebra methods \cite{Gray:2008zs} used in the computation of the particle spectrum in Table \ref{table3}.}. The exact number depends on the detailed form of the effective potential, for example, the number of non-vanishing couplings of the form
\begin{equation}
\langle 1_{+1,-5} \rangle {\bf \bar{5}}_{0,+4} {\bf 5}_{-1,+1}~~\text{and}~~ \langle 1_{-1,-5} \rangle {\bf \bar{5}}_{0,+4} {\bf 5}_{+1,+1}
\end{equation}
induced from the ${\bf \overline{15}}_0{\bf 6}_{-1}{\bf 6}_{+1}$ couplings of the original $SU(6) \times U(1)$ theory. These moduli-dependent mass terms are realized geometrically in Table \ref{table3} by the dependence of the ${\bf 5},{\bf \bar{5}}$ spectrum on the monad map.
The fact that we can ``see" the complete branch structure of the theory on the stability wall (and off of it) both in terms of the stability and D-term analysis and as monad bundles leads to a complete and clear picture. It only remains to compare these results to those in the target space dual theories.
\begin{table}
\begin{centering}
\begin{tabular}{|c|c|c|c|c|c|}\hline
Field & Cohom. & Multiplicity & Field & Cohom. & Multiplicity \\ \hline
${\bf 1}$ & $H^1(V \otimes V^{\vee})$ & 338 & & & \\ \hline
${\bf 10}$ & $H^1(V^{\vee})$ & 0 & $\overline{ {\bf 10}}$ & $H^1(V)$ & $80$ \\ \hline
${\bf 5}$ & $H^1(\wedge^2 V)$ & $81+x$ & $\overline{{\bf 5}}$ & $H^1(\wedge^2 V^{\vee})$ & $1+x$ \\ \hline
\end{tabular}
\caption{\it Particle content of the $SU(5)$ theory associated to the indecomposable $SU(5)$ bundle \eref{branch2a} -- the branch of the theory in the stable chamber of K\"ahler moduli space. See Figure \ref{fig:branchchamber1}. The ambiguity (denoted by $x$) in the ${\bf 5},{\bf \bar{5}}$ spectrum arises from its dependence on the monad map in \eref{branch2a}.}\label{table3}
\end{centering}
\end{table}
\subsubsection{Branches for $(\widetilde{X},\widetilde{V})$}\label{branch_in_dual}
In the case of the dual geometry in \eref{bundle2a.1}, the symmetry $SU(6) \times U(1)$ and the charged matter content is precisely identical (see Table \ref{table2}). The only difference arises in the count of the uncharged singlets $h^1(\widetilde{X},\tilde{U}_3 \otimes \tilde{U}_3^{\vee})=196$. Thus, the vacuum structure of the theory under the Higgsing $SU(6) \to SU(5)$ is also \emph{manifestly the same}.
As in the previous Section, the vacuum equations, \eref{vevsbranch1} and \eref{vevsbranch2} indicate that there will be a solution when $\mu(\tilde{L}^{\vee}) <0$. It only remains to understand how to geometrically realize this deformation as a monad.
From here, the analysis exactly mirrors that of the previous theory. There is only one branch to the theory leaving the stability wall and breaking $SU(6) \to SU(5)$. As before we can replace one negative entry line bundle in the monad (in this case, ${\cal O}(0,-1,1)$) as
\begin{equation}
0 \to \tilde{L}_{new} \to \mathcal O(0,0,1)^{\oplus 2} \stackrel{\tilde{g}}{\longrightarrow} \mathcal O(0,1,1) \to 0
\end{equation}
This leads at last to the bundle
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt} \begin{array}{cccccccc}
0& 0& 0& 0& 0& 0& 1& 1 \\
1& 1& 0& 0& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1& 0& 0
\end{array}&\begin{array}{cc}
-1& -1 \\
-2& 0 \\
-2& -2
\end{array}&\begin{array}{ccccccccc}
0& 0& 0& 1& 0& 0& 0& 0& 0 \\
1& 0& 0& 0& 0& 2& 1& 1& 2 \\
-1& 1&1& -1& 1& 3& 2& 2& 2
\end{array}&\begin{array}{cccc}
0& 0& 0& -1 \\
-1& -3& -1& -2 \\
-1&-2& -4& -3
\end{array}\\
\hline
\end{array}\end{aligned}\label{branch2a.1}\end{equation}
The charged matter of this $SU(5)$ theory matches exactly to that given in Table \ref{table3}. Once again, the only difference arises in the number of uncharged moduli. In this case we find
\begin{equation}
h^{1,1}(\widetilde{X})+h^{2,1}(\widetilde{X})+h^1(\widetilde{X},End_0(\widetilde{V}))=3+55+368=426
\end{equation}
In agreement with \eref{branchy_total}.
This bundle is stable in a 3-dimensional chamber of the K\"ahler cone of $\widetilde{X}$ determined by
\begin{equation}
\mu({\cal O}(0,-1,1))<0~~\text{and}~~\mu({\cal O}(-1,0,1)<0~~\Leftrightarrow~~ s_3<s_2~~\text{and}~~s_3<s_1
\end{equation}
This region is shown in Figure \ref{fig:branchchamber2}.
\begin{figure}
\centering \includegraphics[width=0.4\textwidth]{Slide4} \caption{\emph{The stable sub-cone in K\"ahler moduli space shown for the $SU(5)$ monad bundle in \eref{branch2a.1}. This corresponds to the Higgsed branch of the target space dual theory in \eref{bundle2a.1} and Figure \ref{fig:secondwall}.}} \label{fig:branchchamber2}
\end{figure}
With this monad bundle in hand we have formed a complete picture of the local deformation space near the stability wall in both the original theory and its target space dual. It is clear from the above analysis that not only do the loci of enhanced symmetry correspond in the two theories, but their vacuum branch structure is also identical. In the end we have a commutative diagram of geometric deformations and target space duals (i.e. ``transgressed" \cite{Candelas:2007ac} geometries):
\begin{equation}
\begin{array}{lllll}
&V_1&\stackrel{dual}{\longrightarrow}&\widetilde{V}_1&\\
\langle C \rangle &\downarrow&&\downarrow&\langle \tilde{C} \rangle \\
&V_2 &\stackrel{dual}{\longrightarrow}&\widetilde{V}_2&
\end{array}
\label{first_def}
\end{equation}
It can be verified that the deformed manifolds/bundles (labeled as $V_2$ and $\widetilde{V}_2$ above) given in \eref{branch2a} and \eref{branch2a.1} also form a target space dual pair, as expected. Thus, the correspondence given in \eref{first_def} is complete.
\subsection{Perturbative/Non-perturbative duality}\label{non_pert_duals}
Of the $17$ bundles in the target space dual chain associated to \eref{bundle2a} (see Appendix \ref{appendixB}), the majority of them exhibit the same essential features seen in the previous example, including: a) new stability walls in the bundle geometry, b) a matching reduction of dimension arising from the associated D-terms, c) Agreement in the degrees of freedom (both charged and uncharged) in the target space effective theory and d) perfect correspondence between the vacuum branch loci near the stability wall as described in \eref{first_def}.
There are however, several interesting exceptions that arise and we turn to one such now. The following target space bundle
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccccc}
0& 0& 0& 0& 0& 0& 1& 1 \\
1& 1& 0& 0& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1& 0& 0
\end{array}&\begin{array}{cc}
-1& -1 \\
-2& 0 \\
-3& -1
\end{array}&\begin{array}{ccccccccc}
1& 0& 0& 0& 0& 0& 0& 0& 0 \\
1& -1& 0& 0& 2& 1& 1& 2& 3 \\
-2& 1& 1& 1& 1& 2& 2& 2& 2
\end{array}&\begin{array}{cccc}
-1& 0& 0& 0 \\
-3& -1& -2& -3 \\
-2& -4& -3& -1
\end{array}\\
\hline
\end{array}\end{aligned}\label{non_pert_dual}\end{equation}
Once again exhibits a stability wall and shows the exact reduction in Kahler moduli expected by the D-term condition. However, the new feature in this case is that the stability wall is \emph{forced to the boundary of the K\"ahler cone.} That is, the target space dual leads to a theory that is necessarily at a singular point in moduli space. For the CY 3-fold given in \eref{non_pert_dual}, the K\"ahler cone is given by the positive octant $t^a >0$, $a=1,2,3$ and the D-term condition on the stability wall constrains the Ka\"ahler moduli of $X$ to
\begin{equation}\label{weird_wall_cond}
t^1=0~,~~t^2 \geq 0~,~~t^3=2t^2-2\sqrt{(t^2)^2}~~~\text{or}~~~t^1>0~,~~t^2 = 0~,~~t^3=\frac{3}{2}\left(t^1 \pm \sqrt{(t^1)^2} \right)
\end{equation}
Although the overall volume is still finite when \eref{weird_wall_cond} holds, the presence of small cycles leads us to expect the appearance of non-perturbative physics. Indeed, a closer inspection of \eref{non_pert_dual} suggests that the metric singularities in $X$ must play an important role in the effective physics. On the stability wall, the bundle in \eref{non_pert_dual} splits as
\begin{equation}
V_5 \to L_1 \oplus L_2 \oplus U_3
\end{equation}
with $L_1={\cal O}(1,1,-2)$, $L_2={\cal O}(0,-1,1)$ and $c_1(U_3)=(-1,0,1)$. The structure group of the bundle on the stability wall is $S[U(1)\times U(1) \times U(3)]$ which has commutant $SU(5) \times U(1) \times U(1)$ within $E_8$. Thus, this bundle naively does not lead to the same non-Abelian enhancement of symmetry (i.e. $SU(6)$) seen in the initial bundle \eref{bundle2a}. This is not wholly unexpected however, since it is well-known that singularities in the base manifold, $X$, can force bundles (and more generally small instantons) to exhibit non-Abelian enhancements of symmetry (see for example \cite{Berglund:1998ej}). We leave a full exploration of this interesting class of examples to further work.
\subsection{Isomorphic geometry in target space duality}\label{isomorphic_CYs}
An interesting possibility within the target space duality algorithm includes cases in which the base manifold $X$ is mapped to an identical or related geometry. As an example, consider the following target space dual to the geometry given in \eref{bundle2a}:
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccccc}
0& 0& 0& 0& 0& 0& 1& 1 \\
1& 1& 0& 0& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1& 0& 0
\end{array}&\begin{array}{cc}
-1& -1 \\
-1& -1 \\
-4& \ 0
\end{array}&\begin{array}{ccccccccc}
1& 0& 0& 0& 0& 0& 0& 0& 0 \\
0& -1& 0& 0& 2& 1& 1& 2& 2 \\
-1& 1& 1& 1& 1& 2& 2& 2& 3
\end{array}&\begin{array}{cccc}
0& 0& -1& 0 \\
-3& -1& -2& -1 \\
-2& -4& -3& -3
\end{array}\\
\hline
\end{array}\end{aligned}\label{bundle2a.11m_study}\end{equation}
It is well known that many CICY threefolds can have redundant descriptions \cite{Candelas:1987kf} and it is straightforward to verify that the CY threefold given in \eref{bundle2a.11m_study} is in fact \emph{the same manifold} as the $\{2,4\}$ hypersurface in $\mathbb{P}^1 \times \mathbb{P}^3$ given in \eref{bundle2a} (this is an example of a so-called ``ineffective split" \cite{Hubsch:1992nu}). Solving the linear constraint in the defining relation of \eref{bundle2a.11m_study} leads to $X \times pt$ with $X$ defined in \eref{bundle2a}. Interestingly, even when $X = \widetilde{X}$ the action on the associated monad bundle is not manifestly trivial. To test whether or not the bundles are also isomorphic, it is possible to employ a useful Lemma (Lemma \ref{OSS_corollary_thm}, Appendix \ref{AppendixA}) for isomorphisms of stable bundles.
Finally, another quantitatively different possibility can arise which $X$ and $\widetilde{X}$ are presented as distinct complete intersection manifolds in different ambient products of projective spaces but are ``redundant" manifolds from the point of view of their topological data. Wall's Theorem \cite{wallthm} says that two (simply connected) CY 3-folds which share the same total Chern classes and triple intersection numbers are of the same homotopy type and from the point of view the effective $4$-dimensional physics, equivalent. For example, the following pair of CY 3-folds have equivalent data in the sense of Wall's theorem:
\begin{equation}
X_1=\left[\begin{array}{c|cc}
\mathbb{P}^1 & 1& 1 \\
\mathbb{P}^1& 0 & 2 \\
\mathbb{P}^3 & 2& 2
\end{array}\right]~~~~~~~~~~,~~~~~~~~X_2=\left[\begin{array}{c|cc}
\mathbb{P}^1 & 1& 1 \\
\mathbb{P}^1& 1 & 1 \\
\mathbb{P}^3 & 2& 2
\end{array}\right]
\end{equation}
Both of these manifolds appear in the list of target space duals to \eref{bundle2a} given in Appendix \ref{appendixB}. Interestingly, the number of bundles appearing over these manifolds in the target space duality procedure can apparently differ (starting with the bundle in \eref{bundle2a} the target space duality algorithm generates 3 bundles over $X_1$ and 5 over $X_2$). This indicates that the target space duality procedure probes some properties of the manifolds beyond their topology.
\subsection{On S-Equivalence Classes}\label{sequivalence_sec}
Given the strong role played by stability walls in the chain of examples in the previous Sections, it is natural to wonder whether all bundles that ``share" a wall must somehow be connected in moduli space? The mathematics associated with this question includes the notion of a so-called ``S-Equivalence Class" \cite{huybrechts_book} this is the unique decomposition of a semi-stable bundle (on the stability wall) into a poly-stable sum
\begin{equation}
V \to V_1 \oplus V_2 \oplus \ldots
\end{equation}
This decomposition on the wall characterizes the physical theory and as we will see below bundles with different S-equivalence classes are generally unconnected in the vacuum space of the ${\cal N}=1$, $4$-dimensional theory. The mathematics of S-equivalence classes and their relationship to stability walls and the decomposition into poly-stable bundles (such as \eref{decomp_on_wall} and \eref{mirror_walltheory}) can be found in Appendix \ref{AppendixA}.
In this section we provide an example to demonstrate that even for bundles with the same topology and a shared stability wall, there can be disconnected components to the moduli space, ${\cal M}_{bundle}(rk, c_1,c_2,c_3)$ of (semi-) stable bundles with fixed rank and total Chern class.
To illustrate this possibility, consider the following pair of bundles on the bi-cubic hypersurface, $\{3,3\}$ in $\mathbb P^2 \times \mathbb P^2$:
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccc}
1& 1& 1& 0& 0& 0 \\
0& 0& 0& 1& 1& 1
\end{array}&\begin{array}{c}
-3 \\
-3
\end{array}&\begin{array}{cccccccc}
-2& 1& 1& 2& 2& 2& 2& 0 \\
1& 0& 1& 2& 0& 0& 0& 2
\end{array}&\begin{array}{ccc}
-4& -3& -1 \\
-1& -2& -3
\end{array}\\
\hline
\end{array}\end{aligned}\label{bundle5a}\end{equation}
and
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccc}
1& 1& 1& 0& 0& 0 \\
0& 0& 0& 1& 1& 1
\end{array}&\begin{array}{c}
-3 \\
-3
\end{array}&\begin{array}{cccccccc}
2& 4& 1& 0& 0& 0& 0& 1 \\
-1& 1& 1& 1& 1& 1& 2& 0
\end{array}&\begin{array}{ccc}
-4& -3& -1 \\
-1& -2& -3
\end{array}\\
\hline
\end{array}\end{aligned}\label{bundle5b}\end{equation}
The topology for each bundle is identical, both satisfying $c_2(V)=c_2(TX)$ and $Ch_3(V)=-63$. In both cases, the charged matter spectrum is $h^*(V) = (0, 63, 0, 0)$. Labeling the bundle in \eref{bundle5a} as $V_a$ and that in \eref{bundle5b} as $V_b$, we see that they are each stable in complementary chambers of K\"ahler moduli space (determined by whether $\mu({\cal O}(-2,1))$ is $>0$ or $<0$). Along the shared stability wall they decompose as
\begin{align}
& V_a \to {\cal O}(-2,1) \oplus U_a \\
& V_b \to {\cal O}(2,-1) \oplus U_b
\end{align}
with
\begin{align}
& 0 \to U_a \to {\cal O}(1,0)\oplus {\cal O}(1,1) \oplus {\cal O}(2,2) \oplus {\cal O}(2,0)^{\oplus 3}\oplus {\cal O}(0,2) \to {\cal O}(4,1) \oplus {\cal O}(3,2) \oplus {\cal O}(1,3) \to 0 \\
&0 \to U_b \to {\cal O}(4,1)\oplus {\cal O}(1,1)\oplus {\cal O}(0,1)^{\oplus 3}\oplus {\cal O}(0,2) \oplus {\cal O}(1,0) \to {\cal O}(4,1) \oplus {\cal O}(3,2) \oplus {\cal O}(1,3) \to 0
\end{align}
Direct calculation yields that for $V_a$, $H^*(X, {\cal O}(-2,1) \otimes V_a)=(0,99,0,0)$. Thus, only one possible ``direction" of deformation exists to move the bundle away from its reducible locus and off of the stability wall (corresponding to only one sign of charged bundle moduli in \eref{dterm_first}, \cite{Anderson:2009sw,Anderson:2009nt}). Furthermore, on the reducible locus, $Hom(U_a, U_b)=0$. Thus, there is no deformation of $V_a$ can lead to a stable bundle in the region determined by $V_b$ and the bundles live in different S-equivalence classes.
Just as we saw in Section \ref{branch_in_dual}, where connected branches of moduli space were carried faithfully through the target space duality (see \eref{first_def}), it can also be verified that disconnected components of bundle moduli space on $X$ lead to disconnected components on $\widetilde{X}$.
For each bundle, there are many possible target space dual theories, but to illustrate that the ``disconnected" bundles persist in the dual theories, the following pair is sufficient to illustrate the phenomenon described above.
A target space dual to \eref{bundle5a} is
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccccc}
0& 0& 0& 0& 0& 0& 1& 1 \\
1& 1& 1& 0& 0& 0& 0& 0 \\
0& 0& 0& 1& 1& 1& 0& 0
\end{array}&\begin{array}{cc}
-1& -1 \\
-2& -1 \\
-1& -2
\end{array}&\begin{array}{cccccccc}
0& 0& 1& 0& 0& 0& 0& 0 \\
-2& 1& 0& 2& 2& 2& 3& 0 \\
1& 0& -1& 2& 0& 0& 2& 2
\end{array}&\begin{array}{ccc}
0& -1& 0 \\
-4& -3& -1 \\
-1& -2& -3
\end{array}\\
\hline
\end{array}\end{aligned}\label{bundle5a.1m}\end{equation}
and the following target space dual to \eref{bundle5b}
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccccc}
0& 0& 0& 0& 0& 0& 1& 1 \\
1& 1& 1& 0& 0& 0& 0& 0 \\
0& 0& 0& 1& 1& 1& 0& 0
\end{array}&\begin{array}{cc}
-1& -1 \\
-2& -1 \\
-1& -2
\end{array}&\begin{array}{ccccccccc}
0& 0& 1& 0& 0& 0& 0& 0& 0 \\
2& 4& 0& 0& 0& 0& 0& 1& 3 \\
-1& 1& -1& 1& 1& 1& 2& 0& 2
\end{array}&\begin{array}{cccc}
0& -1& 0& 0 \\
-4& -3& -1& -2 \\
-1& -2& -3& \ 0
\end{array}\\
\hline
\end{array}\end{aligned}\label{bundle5b.1m}\end{equation}
Once again, we see that the stability wall structure persists into the target space dual theory and here a similar analysis to that shown above demonstrates that $\widetilde{V}_a$ and $\widetilde{V}_b$ again form a disjoint pair of bundles that share a stability wall.
Finally, a note of caution should be raised that it is possible that seemingly distinct monads can in fact correspond to the same elements of the S-equivalence class:
To see this, consider the following rank $5$ bundle defined on the $\{2, 4 \}$ hypersurface in $\mathbb P^1 \times \mathbb P^3$:
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccc}
1& 1& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1 \\
\end{array}&\begin{array}{c}
-2 \\
-4
\end{array}&\begin{array}{ccccccc}
0& 0& 0& 0& 1& 1& 1 \\
1& 1& 1& 2& -1& 1& 2
\end{array}&\begin{array}{cc}
-1& -2 \\
-4& -3
\end{array}\\
\hline
\end{array}\end{aligned}\label{bundle2b_again}\end{equation}
This bundle shares \emph{identical topology} with the bundle given in \eref{branch2a} in Section \ref{branch_in_v}. Moreover it shares a stability wall (given by the condition that $\mu({\cal O}(1,-1)=0$) and is stable in the same sub-cone in K\"ahler moduli space. However, on the stability wall it decomposes as
\begin{equation}
V_5 \to {\cal O}(1,-1) \oplus {\cal Q}_4
\end{equation}
where
\begin{equation}
0 \to {\cal Q}_4 \to {\cal O}(0,1)^{\oplus 3} \oplus {\cal O}(0,2) \oplus {\cal O}(1,1) \oplus {\cal O}(1,2) \to {\cal O}(1,4) \oplus {\cal O}(2,3) \to 0
\end{equation}
We can now ask: is this decomposition the same as the one arising on the wall for the $SU(5)$ bundle given in \eref{branch2a}?
There, the bundle data of \eref{branch2a} makes it clear that the bundle decomposes as $V_5 \to {\cal O}(1,-1) \oplus U_4$, with
\begin{equation}
0 \to U_4 \to {\cal O}(0,1)^{\oplus 4} \oplus {\cal O}(2,1) \oplus {\cal O}(1,2)^{\oplus 2} \oplus {\cal O}(2,2) \to {\cal O}(1,1) \oplus {\cal O}(3,2) \oplus {\cal O}(1,4) \oplus {\cal O}(2,3) \to 0
\end{equation}
At first pass, ${\cal Q}_4$ and $U_4$ appear to be different rank $4$ monad bundles. However, a direct calculation yields that
\begin{equation}
dim (Hom({\cal Q}_4, U_4))=1
\end{equation}
and once again it is possible to apply Lemma \ref{OSS_corollary_thm} in Appendix \ref{AppendixA} to determine that these are, in fact, isomorphic stable bundles. Interestingly, the different monad realizations of isomorphic bundles seem in general to lead to new target space dual theories. This effect and the corresponding list of target space dual theories is given in Appendix \ref{appendixB}.
\section{F-Terms, Bundle Holomorphy, and Duality}\label{Fterm_sec}
\subsection{F-terms in the effective theory}
In this section we will explore examples of $(0,2)$ GLSMs with non-trivial F-term contributions to their $4$-dimensional, ${\cal N}=1$ potential. At the level of geometry, this corresponds to obstructions to geometric deformations arising from the constraint of bundle holomorphy.
In general, it is exceedinly challenging computationally to calculate directly the Atiyah obstruction (largely due to the difficulty in determining the the class of the Atiyah map ($[F^{1,1}]$) in \eref{mod_def}. To find tractable/computable examples a series of approaches were developed in \cite{Anderson:2009sw,Anderson:2009nt,Anderson:2011ty,Anderson:2013qca}. In particular, one simple way to generate examples employs methods of constructing bundles that \emph{manifestly depend on the complex structure moduli of $X$}. One such tool is the well-known ``jumping" phenomena of bundle-valued cohomology on a CY manifold, $X$. This was applied to build examples of monad and extension bundles that presented non-trivial holomorphy obstructions. For a thorough treatment of the deformation theory associated to this idea, see \cite{Anderson:2011ty}. Here we will briefly review the relevant techniques for monads.
In the case of monad bundles, a simple class of holomorphy obstructions can be realized in the dependence on $F_a^l(x^i)$ on the complex structure of $X$. In general, elements of $F_a^l(x^i)$ may be generically zero and only non-vanishing for certain higher co-dimensional sub-loci of complex structure moduli space. Suppose we consider a rank $n$ monad of the form
\begin{equation}
0 \to V \to {\cal O}({\bf b}_1)\oplus \ldots {\cal O}({\bf b}_{n+1}) \stackrel{F}{\longrightarrow} {\cal O}({\bf c}) \to 0
\end{equation}
with the property that a given map element, say $Hom({\cal O}({\bf b}_1),{\cal O}({\bf c}))$ is generically vanishing. That is,
\begin{equation}
h^0(X, {\cal O}({\bf c- b_1}))=0
\end{equation}
Such a monad does not define a good, slope stable bundle since in general it will be reducible (since ${\cal O}({\bf b}_1) \in ker(F)$ generically):
\begin{equation}
{\cal O}({\bf b_1}) \oplus V'
\end{equation}
Note that in general if ${\cal O}({\bf b_1})$ is of purely positive or negative degree, then the sum above \emph{cannot} be poly-stable within the K\"ahler cone. However, on special loci in complex structure moduli space, line bundle cohomology can ``jump" to a non-zero value. We will denote this locus as ${\cal CS}_{jump}$. On that locus, $h^0(X, {\cal O}({\bf c- b_1}))\neq 0$ and we find new global holomorphic sections to ${\cal O}({\bf c- b_1})$ that can be used to define a stable, holomorphic bundle, $V=ker(F)$. As an example, consider the following line bundle on the $\{2,4\}$ hypersurface in $\mathbb{P}^1 \times \mathbb{P}^3$:
\begin{equation}\label{nocohom_gen}
h^0(X,{\cal O}(-2,4))=0~~\text{for generic values of complex structure}
\end{equation}
However, as computed in \cite{Anderson:2011ty} on a $53$-dimensional sub-locus of the $86$-dimensional complex structure moduli space, this cohomology can ``jump" to
\begin{equation}\label{cohom_jump}
\text{On}~{\cal CS}_{jump},~~~h^0(X, {\cal O}(-2,4))=1
\end{equation}
In the following section, we will make use of this calculation to build a monad bundle with a non-trivial holomorphy obstruction and construct its target space dual theory. Before we begin however, two notes of caution are in order.
First, it should be noted that the inclusion of map elements such as the one above $F \in H^0(X, {\cal O}(-2,4)$ introduce more dramatic negative charges into the definition of the GLSM then those we previously considered (only negative charges for some $\Lambda^a$ and all maps, $F_a^l$ positive in the D-term examples of the previous Sections). This can change the vacuum structure of the GLSM and care must be taken interpreting these both in the geometric regime and for other phases. For this section, we will consider only examples that we can confirm correspond to good geometric phases (i.e. large volume, perturbative, etc). We leave the important and intriguing question of how many negative degrees can be included\footnote{Note this is closely related to the new ``Generalized Complete Intersection" (gCICY) method of constructing Calabi-Yau manifolds \cite{Anderson:2015yzz,Anderson:2015yzz,Berglund:2016yqo}.} in the GLSM to a future publication \cite{to_appear} and focus in this Section only on the well-defined target space theories.
Second, the calculations of holomorphy obstructions in the following sections will be done at the level of the compactification geometry. Caution must be used in describing this reduction in moduli from the naive count in \eref{naive_count} in the effective theory. In the examples that follow below, the mass scale of the holomorphy obstructions is too high to be included in the effective theory. Instead, the geometric obstructions are directly calculated at the compactification scale (as is done to compute \eref{chargmat} in the charged matter) and the resulting massless modes (corresponding to the solution of \eref{mod_def}) are presented. For some steps towards realizing holomorphy obstructions at the level of sigma models, see \cite{Melnikov:2011ez}.
As a final comment, it should be noted that compared to the D-term computations of Section \ref{dterms}, the holomorphy computations required here are significantly more computationally intensive and introduce a number of difficult problems in geometry/deformation theory. As a result, we focus on one example geometry and leave a broader scan for examples to future work.
We turn now to a simple example of a monad bundle inducing an obstruction in the complex structure of $X$ and its target space dual.
\subsection{A simple example}
In this section we will consider a simple vector bundle which was inspired by an example in \cite{Anderson:2011ty}. This $SU(3)$ monad bundle:
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccc}
1& 1& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1
\end{array}&\begin{array}{c}
-2 \\
-4
\end{array}&\begin{array}{ccccc}
2& -1& -1 &1 &0 \\
0& 2& 2 & 0 & 2
\end{array}&\begin{array}{cc}
0 & -1 \\
-4 & -2
\end{array}\\
\hline
\end{array}\end{aligned}\label{eq:Fbundle2}\end{equation}
cannot be defined for generic choices of the complex structure of $X$ given in \eref{eq:Fbundle2}. It is clear that by \eref{nocohom_gen}, for generic values of the complex structure the map, $F$ takes the form
\begin{equation}\label{two_row_map}
F_a^l=\left(
\begin{array}{ccccc}
f_{(-2,4)} & f_{(1,2)} & f'_{(1,2)} & f_{(-1,4)} &f_{(0,2)}) \\
0 & f_{(1,0)} & f'_{(1,0)} & f_{(0,2)} & f_{(1,0)}
\end{array}
\right)
\end{equation}
where $f_{(a,b)}$ is generic polynomials on $X$ of multi-degree $\{a,b\}$. Of note are the maps with ``negative" degree. It is straightforward to verify that on $X$, $h^0(X, {\cal O}(-1,4))=2$ and thus, despite the negative degree this map is an ordinary holomorphic function on $X$. In \cite{Anderson:2015iia} tools were developed to describe such ``non-polynomial" functions explicitly. Here we will not require the explicit description of $f'$ and merely need to know that it is non-trivial.
As noted above however, the map $f_{(-2,4)}$ is on a different footing since \emph{for generic values of the complex structure} $h^0(X, {\cal O}(-2,4))=0$. When $f_{(-2,4)}$ is not available as a non-trivial map, it is clear that this $F_a^l(x)$ cannot define a good bundle. Here $V$ is reducible as the sum of a line bundle and rank $2$ sheaf:
\begin{equation}
V \sim {\cal O}(2,0) \oplus {\cal V}
\end{equation}
and \emph{cannot be made poly-stable} in the K\"ahler cone of $X$. Thus, in general, this monad cannot provide a good definition of the background gauge fields in the heterotic theory. However, as shown in \cite{Anderson:2011ty} and using \eref{cohom_jump}, on the 53-dimensional sub-locus of complex structure moduli space, there is a new map element available given by $f_{(-2,4)} \in H^0(X, {\cal O}(-2,4))$. Once again, despite the negative degree in the line bundle, for the chosen complex structure this is an ordinary, global, holomorphic section to ${\cal O}(-2,4)$ on $X$. Thus on the locus ${\cal CS}_{jump}$, it is possible to define a stable, \emph{indecomposable} monad, $V$. We will begin there and use this example to construct a target space dual.
We have chosen this example because of its simplicity and the fact that the difficult computational exercise of computing ${\cal CS}_{jump}$ and the associated Atiyah calculation was already completed in \cite{Anderson:2011ty}. However, it should be noted that the monad given above does not satisfy the anomaly cancellation condition alone since $c_2(V)=\{16,32\}$. This can be remedied by choosing to ``complete" this example with another monad bundle. We will embed this second bundle into the second $E_8$ factor. One such good $SU(3)$ bundle is
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccc}
1& 1& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1
\end{array}&\begin{array}{c}
-2 \\
-4
\end{array}&\begin{array}{ccccc}
1& 0& 0 &0 & 0 \\
0& 1& 1 &1 &1
\end{array}&\begin{array}{cc}
-1 & 0 \\
-2 &-2
\end{array}\\
\hline
\end{array}\end{aligned}\label{eq:bundle2_hidden}\end{equation}
with $c_2(V)=\{8,12 \}$ as required so that
\begin{equation}
c_{2}(V_{total})=c_2(V_1) + c_2(V_2)=c_2(TX)
\end{equation}
With this addition, the full $4$-dimensional theory has gauge symmetry $E_6 \times E_6$.
In what follows the bundle in \eref{eq:bundle2_hidden} plays a trivial, spectating role and thus, we will recall that it must be included, but it will not be central to our further analysis. Focusing first on the rank $3$ bundle in \eref{eq:Fbundle2}, naive moduli count associated to the first $SU(3)$ bundle and the CY manifold would have produced
\begin{equation}
dim({\cal M}_0)=h^{1,1}+h^{2,1}+h^1(X,End_0(V))=2+86+92=180
\end{equation}
However, taking into account the constraint that the complex structure lies within the ``jumping locus" in order for the bundle to be well-defined we have
\begin{equation}\label{lifted_goodstuff}
dim({\cal M}_1)=dim({\cal M}_0)-33=147
\end{equation}
See \cite{Anderson:2011ty} for a discussion of the relationship between this jumping cohomology calculation and the Atiyah obstruction. To complete the degree of freedom count, we note that for $V$ in \eref{eq:Fbundle2} the charged matter is
\begin{align}\label{charged_e6}
& h^1(X,V)=41~~~(\text{no. of }{\bf 27}) \\
& h^1(X, V^\vee)=1~~~(\text{no. of }{\bf \overline{27}})
\end{align}
For completeness, the spectrum of the second $E_6$ factor consists of $12$ ${\bf 27}$s (and no ${\bf \overline{27}}$s) and $38$ singlets.
Finally, one last interesting note about this geometry: despite the negative line bundle entry (${\cal O}(-1,2)$) in \eref{eq:Fbundle2}, the $SU(3)$ bundle presented there is stable throughout the \emph{entire} K\"ahler cone since $\mu({\cal O}(-1,2))<0$ for all K\"ahler moduli, $t^r$, on $X$. Thus, there is no stability wall and no non-trivial D-term anywhere in the moduli space of this theory.
With these calculations in hand, we turn finally to the target space dual theory. Once again, we can ask the relevant questions of the duality:
\begin{itemize}
\item Is a holomorphy obstruction in $(X,V)$ reflected in some (deformation theoretic) obstruction in $(\widetilde{X},\widetilde{V})$?
\item Since in general the dimensions of $h^{2,1}(X)$ and $h^{2,1}(\widetilde{X})$ (and any Atiyah obstructions within these spaces) are different, is the final set of zero modes the same in each theory?
\end{itemize}
We will address these questions in turn in the following Section.
\subsection{The target space dual}
The target space dual is easy to construct for the rank $3$ bundle given in \eref{eq:Fbundle2}. We obtain
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccccc}
0& 0& 0& 0& 0& 0& 1& 1 \\
1& 1& 0& 0& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1& 0& 0
\end{array}&\begin{array}{cc}
-1& -1 \\
-1& -1 \\
-2& -2
\end{array}&\begin{array}{ccccc}
0& 1& 0 & 0 & 0 \\
2& -2& 0 & 1 & 0 \\
0& 0& 4 & 0 & 2
\end{array}&\begin{array}{cc}
-1 & 0 \\
0 & -1 \\
-4 & -2
\end{array}\\
\hline
\end{array}\end{aligned}\label{fterm_mirror2}\end{equation}
The new CY manifold, $\widetilde{X}$ has Hodge numbers $h^{1,1}=3$ and $h^{2,1}=55$.
Now, to answer the first question in our list, we can ask whether the bundle in \eref{fterm_mirror2} provides any obvious obstruction to holomorphy? By inspection, the answer is immediately yes! We see that in fact, the map $\widetilde{F}_a^l(x)$ associated to $\widetilde{V}$ takes the form
\begin{equation}\label{two_row_map_mirror}
\widetilde{F}_a^l=\left(
\begin{array}{ccccc}
f_{(1,-2,4)} & f_{(0,2,4)} & f_{(1,0,0)} & f_{(-1,1,4)} & 0 \\
0 & f_{(-1,3,2)} & 0 & f_{(0,0,2)} & f_{(0,1,0)}
\end{array}
\right)
\end{equation}
All the map entries above are generically non-vanishing polynomials of the given degree with the exception of $\widetilde{F}_1^1$ of degree $(1,-2,4)$. As expected, the line bundle cohomology associated to this
\begin{equation}
h^0(\widetilde{X},{\cal O}(1,-2,4))=0~~~\text{for generic values of the complex structure}
\end{equation}
but once again, \emph{on a higher co-dimensional locus in complex structure moduli space, which can be labeled} ,$\widetilde{CS}_{jump}$, $h^0(\widetilde{X} ,{\cal O}(1,-2,4))=1$. Once again, it is possible to define a good, indecomposable rank $3$ monad bundle, but only by tuning the complex structure moduli of the CY manifold. In this case, the locus has dimension
\begin{equation}
dim(\widetilde{CS}_{jump})=37
\end{equation}
That is, the jumping of the relevant monad map fixes $18$ of the $55$ complex structure moduli of $\widetilde{X}$. The initial moduli count for this bundle is then
\begin{equation}\label{silly_count}
dim({\cal \widetilde{M}}_0)=3+55+122=180
\end{equation}
With the obvious constraint above this reduces the total to $dim({\cal \widetilde{M}}_0)-18=162$. Moreover, the charged matter of the $E_6$ theory is identical to that in \eref{charged_e6}:
\begin{align}\label{charged_e6_mirror}
& h^1(\widetilde{X},\widetilde{V})=41~~~(\text{no. of }{\bf 27}) \\
& h^1(\widetilde{X}, \widetilde{V}^\vee)=1~~~(\text{no. of }{\bf \overline{27}})
\end{align}
This is almost in agreement with the final singlet count of $147$ in the original geometry in \eref{lifted_goodstuff}. We are left however, with an apparent discrepancy of $15$ moduli. Could these too be lifted from the count for $(\widetilde{X}, \widetilde{V})$? There are two hints that this is indeed the case.
The first is that the lengthy calculation of $h^1(\widetilde{X}, End_0(\widetilde{V}))=122$ in \eref{silly_count} was performed by taking all induced monad maps in $\widetilde{V} \otimes \widetilde{V}^{\vee}$ to be generic (a necessary computational simplification). However, it is clear that for the special bundle at hand a more detailed calculation is necessary. Two hints at this structure are significant.
First, the line bundle ${\cal O}(0,-2,4)$ appears in the calculation of $h^1(\widetilde{X}, End_0(\widetilde{V}))$ (corresponding to morphisms of the form $Hom(\widetilde{B},\widetilde{B})$. As for the similar line bundle on $X$, generically this has no global sections. However, if we further tune the complex structure of $\widetilde{X}$ to make $h^0(\widetilde{X},{\cal O}(0,-2,4))$ jump to a non-zero value, this fixes exactly $15$ additional moduli\footnote{This can be verified using the tools of \cite{Anderson:2011ty,Anderson:2013qca}.}. Such a choice might be expected in comparison with the original pair $(X, V)$ since it is exactly these additional moduli that ``pass through" the conifold transition connecting $X$ and $\widetilde{X}$ and are in some sense ``shared" between the CY manifolds. See \cite{Anderson:2013qca} for results on how jumping loci (${\cal CS}_{jump}$) are fixed across conifold transitions.
Finally, as another hint for where the moduli fixing may arise, it should be noted that unlike the starting geometry, the bundle in \eref{fterm_mirror2} \emph{does display a stability wall in K\"ahler moduli space}. In general, the bundle is stable for the $3$ dimensional sub-cone of K\"ahler moduli space determined by
\begin{equation}
0<t^1~~,~~2 t^1<t^2 ~~,~~0< t^3 < 2t^2-4t^1
\end{equation}
On the 2-dimensional stability wall given by $t^3=2t^2-4t^1$ and $2 t^1<t^2$ the bundle decomposes as
\begin{equation}
{\cal O}(1,-2,0)\oplus \tilde{U}_2
\end{equation}
with
\begin{equation}
0 \to \tilde{U}_2 \to {\cal O}(0,2,0)\oplus {\cal O}(0,0,4)\oplus {\cal O}(0,1,0) \oplus {\cal O}(0,0,2) \to {\cal O}(1,0,4) \oplus {\cal O}(0,1,2) \to 0
\end{equation}
At this locus the bundle decomposes as $SU(3) \to S[U(1)\times U(2)]$ and the full matter spectrum is charged under the Green-Schwarz anomalous $U(1)$ symmetry as in Section \ref{dterms}. Of interest to us are the $123$ singlets corresponding to bundle moduli:
\begin{align}
&C_{+3}~~~~h^1(\widetilde{X}, \tilde{L}^{\vee} \otimes \tilde{U}_2)=68 \\
&C_{-3}~~~~h^1(\widetilde{X}, \tilde{L} \otimes {\tilde{U}^{\vee}}_2)=10 \\
&C_{0}~~~~~h^1(\widetilde{X}, \tilde{U}_2 \otimes {\tilde{U}^{\vee}}_2)=45
\end{align}
with $\tilde{L}={\cal O}(1,-2,0)$. In this description it is clear that the form of $U_2$, itself and the number of fields $C_0$ depends on the choice complex structure of $\widetilde{X}$. Moreover, this description makes it clear that additional bundle moduli can acquire a mass. Near the stability wall, one origin for such mass terms is evident from the fact that there are allowed, gauge-invariant trilinear couplings in the superpotential of the form
\begin{equation}\label{singlet_trilinear}
W \sim C_0 C_{+3} C_{-3}
\end{equation}
By leaving the stability wall with $\langle C_{\pm 3} \rangle \neq 0$, it is clear that mass terms can be generated for singlets which could lift degrees of freedom, including the significant $15$ moduli above. We leave a detailed analysis of the bundle moduli of \eref{fterm_mirror2} and the computation of trilinear couplings such as \eref{singlet_trilinear} for future work.
To complete this example, it should be noted that as before, another bundle in the second $E_8$ factor is necessary to satisfy anomaly cancellation and the bundle in \eref{eq:bundle2_hidden} trivially carries through the target space duality procedure.
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccccc}
0& 0& 0& 0& 0& 0& 1& 1 \\
1& 1& 0& 0& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1& 0& 0
\end{array}&\begin{array}{cc}
-1& -1 \\
-1& -1 \\
-2& -2
\end{array}&\begin{array}{ccccc}
0 & 0 & 0 & 0 &0 \\
1& 0& 0 &0 & 0 \\
0& 1& 1 &1 &1
\end{array}&\begin{array}{cc}
0 & 0 \\
-1 & 0 \\
-2 &-2
\end{array}\\
\hline
\end{array}\end{aligned}\label{eq:bundle2_hidden_mirror}\end{equation}
The charged and singlet matter spectrum for this bundle is identical to that associated to \eref{eq:bundle2_hidden}.
In summary then, we have found that F-term obstructions are non-trivially preserved across this target space dual pair and correspond to a sensitive inter-mixing of complex structure and bundle moduli which are non-trivially redefined across the duality.
\section{Target space duality, ${\bf (2,2)}$ loci, and tangent bundle deformations}\label{tan_defs}
Having seen that loci of enhanced symmetry (i.e. stability walls, etc) were reflected in the target space duals in previous sections, it is natural to investigate whether or not another special locus is preserved -- namely, the $(2,2)$ locus of a $(0,2)$ theory. Such a locus --realized in the geometric regime by GLSMs in which $V$ is a deformation of the holomorphic tangent bundle of the CY base manifold, $X$ -- play a crucial role in the study of $(0,2)$ GLSMs. In this section, we will provide examples of starting GLSMs with a $(2,2)$ locus and explore the structure of their target space duals.
To begin, we consider perhaps the simplest case available: the rank-changing deformation of the holomorphic tangent bundle to the quintic hypersurface in $\mathbb{P}^4$. The following $SU(4)$ bundle is a non-trivial deformation of ${\cal O}_X \oplus TX$ on the quintic:
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{ccccc}
1& 1& 1& 1& 1 \\
\end{array}&\begin{array}{c}
-5
\end{array}&\begin{array}{ccccc}
1& 1& 1& 1& 1 \\
\end{array}&\begin{array}{c}
-5
\end{array}\\
\hline
\end{array}\end{aligned}\label{tandef}\end{equation}
Indeed, this $V_4$ is in fact the most general slope-stable, rank $4$ deformation of this type and the entire local moduli space of deformations can be captured in the moduli of the monad (see \cite{Li:2004hx}). Writing the bundle, $V_4$, as
\begin{equation}
0 \to V \to {\cal O}(1)^{\oplus 5} \stackrel{f}{\longrightarrow} {\cal O}(5) \to 0
\end{equation}
the locus in bundle moduli space for which it decomposes as ${\cal O}_X \oplus TX$ is given by
\begin{equation}
f= dP_5
\end{equation}
where $P_5$ is the homogeneous, degree $5$ defining relation of the quintic in $\mathbb{P}^4$.
An application of the tangent space duality procedure can produce the following geometry, including a new manifold $\widetilde{X}$ which is related to the quintic via a conifold transition:
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{ccccccc}
1& 1& 0& 0& 0& 0 & 0 \\
0& 0& 1& 1& 1& 1 & 1
\end{array}&\begin{array}{cc}
-1 & -1 \\
-4 & -1
\end{array}&\begin{array}{cccccc}
1& 0& 0& 0& 0& 0 \\
0& 5& 1& 1& 1& 1
\end{array}&\begin{array}{cc}
-1 & 0\\
-5 & -4
\end{array}\\
\hline
\end{array}\end{aligned}\label{quintic_dual}\end{equation}
As required by the target space duality algorithm, this bundle leads to the same effective theory as the original geometry in \eref{tandef}. In particular, it is clear that the Chiral Index (the difference between ${\bf 16}$ and ${\bf \overline{16}}$ representations in the target space $SO(10)$ theory) is given by
\begin{equation}
Ch_3(\widetilde{V})=-100
\end{equation}
as required by our starting point of $V=TX$ on the quintic.
However, this fact leads immediately to the observation that this bundle \emph{cannot be a standard deformation of $T\widetilde{X}$}, since for the new manifold\footnote{This difference in index is precisely arising from the classic conifold transition between $X$ and $\widetilde{X}$ -- the difference corresponding to resolving the $16$ singular points on the nodal quintic \cite{Candelas:1989js}.} in \eref{quintic_dual}
\begin{equation}
Ch_3(T\widetilde{X})=-84~.
\end{equation}
This result is clearly a general feature of the target space duality procedure. Starting from a bundle near the $(2,2)$ locus on $X$, the algorithm produces a bundle with $ch_2(\widetilde{V})=ch_2(T\widetilde{X})$ but with $ch_3(\widetilde{V}) \neq ch_3(T\widetilde{X})$, as required by the ${\cal N}=1$, $4$-dimensional effective theories being identical.
If however, the target space dual is really isomorphic in some deeper sense (as a $(0,2)$ sigma model), it is natural to expect that the $(2,2)$ locus should still somehow be present. One possibility is that the only deformations leading to the $(2,2)$ locus in the dual geometry of \eref{quintic_dual} force the theory to a non-perturbative regime. A candidate for such a possibility might be the singular shared locus in the moduli space of $X$ and $\widetilde{X}$. Since the two CY 3-folds are related by a conifold transition \cite{Candelas:1989js}, their common locus in moduli space is given by a nodal quintic of the form
\begin{equation}
P_5= l_1(x)q_2(x)-l_2(x)q_1(x)=0
\end{equation}
where $l_i(x)$ and $q_i(x)$ are linear and quartic polynomials, respectively, in the homogeneous coordinates of $\mathbb{P}^4$ (and the CY 3-fold in \eref{quintic_dual} is the resolution of this singular variety).
A second possibility is that even at general points in complex structure moduli space, the bundle given in \eref{quintic_dual} is deformable to the holomorphic tangent bundle through some more general, chirality changing transition. Such possibilities have been previously conjectured in the heterotic literature \cite{Douglas:2004yv}, though explicit geometric realizations have yet to be fully understood (see \cite{Anderson:2015cqy} for some recent, related results). It would be intriguing to explore the geometry of such transitions further in this context.
To conclude, we provide one interesting hint towards the behavior of $(2,2)$ loci under the duality. A useful observation is that it is possible for some target space duals to lead to a new vector bundle on \emph{an isomorphic CY threefold}. To illustrate this, consider the following rank $5$ deformation of the tangent bundle $TX$ of the $\{2,4\}$ hypersurface in $ \mathbb P^1 \times \mathbb P^3$:
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccc}
1& 1& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1
\end{array}&\begin{array}{c}
-2 \\
-4
\end{array}&\begin{array}{cccccc}
1& 1& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1
\end{array}&\begin{array}{c}
-2 \\
-4
\end{array}\\
\hline
\end{array}\end{aligned}\label{tan_def24}\end{equation}
The bundle above is a stable deformation of $TX \oplus {\cal O}_X \oplus {\cal O}_X$. This starting point admits a target space dual in which the original manifold appears in a redundant CICY description (achieved by adding the repeated entry ${\cal O}(1,4)$):
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccccc}
0& 0& 0& 0& 0& 0& 1& 1 \\
1& 1& 0& 0& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1& 0& 0
\end{array}&\begin{array}{cc}
-1& -1 \\
-1& -1 \\
-4& 0
\end{array}&\begin{array}{ccccccc}
1& 0& 0& 0& 0& 0& 0 \\
0& 1& 0& 0& 0& 0& 2 \\
0& 0& 1& 1& 1& 1& 4
\end{array}&\begin{array}{cc}
-1& 0 \\
-2& -1 \\
-4& -4
\end{array}\\
\hline
\end{array}\end{aligned}\label{cy_rewrite}\end{equation}\\ \\
As described in Section \ref{isomorphic_CYs}, this target space dual exhibits a particularly interesting structure because it has left the CY $3$-fold unchanged. As argued in Section \ref{isomorphic_CYs}, this threefold is equivalent to $X$ in \eref{tan_def24} times a point. Despite the fact that the manifold has not changed, this example is not trivial however, since the vector bundles given in \eref{tan_def24} and \eref{cy_rewrite} are not obviously the same, nor is their apparent dependence on the complex structure of $X =\widetilde{X}$.
To clarify things, it can be noted that although there are three divisors restricted from the ambient hyperplanes in \eref{cy_rewrite}, these divisors are linearly dependent and $h^{1,1}(\widetilde{X})=2$. Furthermore, a simple exploration of linear equivalence of divisors shows that ${\cal O}(a,b,c)$ on $\tilde{X}$ in \eref{cy_rewrite} is the same line bundle as ${\cal O}(a+b,c)$ on the threefold in \eref{tan_def24} for any integers, $a,b,c \in \mathbb{Z}$. With this observation it is possible to re-write the dual theory in \eref{cy_rewrite} as
\begin{equation}\begin{aligned}&\begin{array}{|c|c| |c|c|}
\hline
x_i & \Gamma^j & \Lambda^a & p_l \\ \noalign{\hrule height 1pt}\begin{array}{cccccc}
1& 1& 0& 0& 0& 0 \\
0& 0& 1& 1& 1& 1
\end{array}&\begin{array}{c}
-2 \\
-4
\end{array}&\begin{array}{ccccccc}
1& 1& 0 & 0 & 0& 0 & 2 \\
0& 0& 1 &1 & 1& 1 &4
\end{array}&\begin{array}{cc}
-3& -1 \\
-4& -4
\end{array}\\
\hline
\end{array}\end{aligned}\label{mirrorbun_again}\end{equation}
which is the same manifold $X$ with an apparently distinct monad bundle $\widetilde{V}$. Now we can ask the question: are the bundles on $X$ given in \eref{tan_def24} and \eref{mirrorbun_again} equivalent? To answer this puzzle it is possible to appeal to a useful lemma which establishes when two stable bundles on $X$ can be identified (See Lemmas \ref{OSS_lemma_thm} and \ref{OSS_corollary_thm} in Appendix \ref{AppendixA}). Briefly, the lemma states if there exists a non-trivial map (in $Hom(V, \widetilde{V})$) between two stable bundles on $X$ with the same slope, $\mu(V)$, then they are isomorphic. In this case, a direct calculation leads to
\begin{equation}
dim(Hom(V, \widetilde{V}))=h^0(X, V \otimes \widetilde{V}^\vee)=1
\end{equation}
Thus, the naively distinct monads given in \eref{tan_def24} and \eref{mirrorbun_again} are in fact equivalent stable, holomorphic vector bundles, complete with the same $(2,2)$ locus (though the locus is clearly most easily realized in the first description).
Thus in this simple case, the $(2,2)$ locus is preserved in the target space dual, though in a subtle way.
\section{Conclusions and Future Directions}\label{conclusions}
In this work we have explored the notion of $(0,2)$ target space duality by studying in detail the $4$-dimensional effective potential in dual theories and the associated vacuum spaces. It has been an open question whether or not $(0,2)$ target space duality may correspond to a true string duality or merely a shared sub-locus in the moduli space of two distinct theories. While a full correspondence of $(0,2)$ heterotic string theories has yet to be established, our results appear to provide evidence for the former possibility. As in \cite{Blumenhagen:2011sq}, we find that the full massless spectrum of the dual theories (in the geometric phase) is preserved
\begin{eqnarray}\nonumber
h^*(X,\wedge^{k}V) &=& h^*(\widetilde X, \wedge^{k} \widetilde V) \\
h^{2,1}(X) + h^{1,1}(X) + h^1(X,End_0(V)) &=& h^{2,1}(\widetilde X) + h^{1,1}(\widetilde X) + h^1(X,End_0(\widetilde V))
\end{eqnarray}
with $k=1 \ldots rk(V)$ and that the anomaly cancellation conditions and equations of motion are automatically satisfied in the dual theories. Moreover, we make the novel observation that non-trivial geometric obstructions --corresponding to D- and F-term potentials -- lead to the same vacuum space in the low energy theories. The central results of this work include the following
\begin{enumerate}
\item D-term and F-term constraints associated to stability and holomorphy constraints in the pair $(X,V)$ appear to be faithfully reflected in the dual geometry $(\widetilde{X},\widetilde{V})$ for all the examples studied in this work.
\item The structure of the ${\cal N}=1$ vacuum space of the dual theories is identical. That is, beginning at given points in moduli space infinitesimal fluctuations are preserved, leading to commutative diagrams of bundle deformations as in \eref{first_def}.
\item We find that loci of enhanced symmetry -- including stability walls (with enhanced Green-Schwarz massive $U(1)$ symmetries) and $(2,2)$ loci are preserved across the duality (though the identification of such enhancement points can be subtle and not always in the perturbative regime of the theory).
\end{enumerate}
To make the observations above we have developed a number of new technical tools including a generalization of the target space duality procedure itself (in Section \ref{repeated_line_bundles}) and the ability to describe complete infinitesimal bundle deformation spaces as monads (see Sections \ref{branch_in_v} and \ref{branch_in_dual}).
In the present work, our primary goal has been to compare the structure of the target space effective theories. A natural next step is to explore the examples considered here more fully at the level of GLSMS (or indeed full sigma models when possible). In this work, we have not considered the possible contributions from world sheet instantons to the potential (though in many cases for simple CICY GLSMs such as considered here, they may be expected to vanish \cite{Beasley:2003fx,Beasley:2005iu,Bertolini:2014dna}) or indeed the full structure of even the perturbative Yukawa couplings (see \cite{Anderson:2009ge,Anderson:2010tc,Blesneag:2015pvz,Buchbinder:2016jqr} for example). Both should be studied to more fully understand the validity of the effective theories considered here and their possible duality. Other interesting future comparisons might include a calculation of spectra of the target space dual GLSMS in non-geometric phases (for example the Landau-Ginsburg spectra \cite{Kachru:1993pg}), a study of the quantum cohomology rings in some simple cases \cite{Donagi:2012sca} and perhaps even applying tools from localization \cite{Closset:2015ohf} to investigate the detailed structure of the GLSMS themselves in more detail. We hope to address some of these questions in future work.
Finally, another interesting area of investigation, first mentioned in \cite{Blumenhagen:2011sq} would be to understand the consequences of $(0,2)$ target space duality in Heterotic/F-theory duality \cite{Friedman:1997yq}. In particular, if both $X$ and $\widetilde{X}$ admit an elliptic fibration: $\pi: X \to B_2$ and $\tilde{\pi}: \widetilde{X} \to \widetilde{B}_2$ (with distinct base manifolds, $B_2$ and $\widetilde{B}_2$), then it is expected that there will exist dual $4$-dimensional theories arising from F-theory compactificed on $K3$-fibered CY 4-folds. It was speculated in \cite{Blumenhagen:2011sq} that perhaps the dual F-theory geometries may be realizable as multiple elliptic/$K3$ fibrations of the same CY $4$-fold, $Y_4$:
\begin{equation}
\xymatrix{
& Y_4 \ar[ld]^{\pi_{1}}_{\mathbb{E}} \ar[rd]^{\mathbb{E}}_{\pi_{2}} &\\
{\cal B}_{3}& & \widetilde{{\cal B}}_{3}}
\end{equation}
where the $3$-fold base manifolds ${\cal B}_3$ and $\widetilde{{\cal B}}_3$ are both $\mathbb{P}^1$ fibered as required by compatibility of the elliptic and $K3$ fibrations. That is, there exist maps, $\rho_1: {\cal B}_{3}\stackrel{\mathbb{P}^1}{\longrightarrow}B_2$ and $\rho_2: \widetilde{{\cal B}}_{3} \stackrel{\mathbb{P}^1}{\longrightarrow} \widetilde{B}_2$. This too, would be in intriguing topic for further research and could shed further light on recent progress in heterotic/F-theory duality \cite{Anderson:2014gla,Anderson:2016kuf,Cvetic:2016ner,Cvetic:2015uwu}.
\section*{Acknowledgements}
The authors would like to thank Per Berglund, Ralph Blumenhagen, James Gray, Sheldon Katz, Ilarion Melnikov, and Eric Sharpe for useful discussions. LA also thanks the UI Chicago and the UI Urbana-Champaign for hospitality during the conclusion of this work. The work of LA and HF is supported by NSF grant PHY-1417337. This project is part of the working group activities of the 4-VA initiative ``A Synthesis of Two Approaches to String Phenomenology".
| f4df125c0d644e7f17ff4ef06bd26dc048b280de | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section*{Introduction}
M. F. Atiyah \cite{MR0420729}
introduced the notion of the $\Gamma$-index $\mathrm{index}_{\Gamma}(\widetilde{D})$
for a lifted elliptic differential operator $\widetilde{D}$
on a Galois $\Gamma$-covering over a closed manifold
and proved that the $\Gamma$-index $\mathrm{index}_{\Gamma}(\widetilde{D})$
of a lifted operator $\widetilde{D}$
equals the Fredholm index $\mathrm{index} (D)$ of the elliptic differential operator $D$
on the base manifold.
On the other hand, Atiyah \cite{MR0420729} also
investigate properties of a $\Gamma$-trace $\mathrm{tr}_{\Gamma}$ at the same time.
The $\Gamma$-trace is a trace of the $\Gamma$-trace operators,
so it induces a homomorphism $(\mathrm{tr}_{\Gamma})_{\ast}$
from
$K_{0}$-group of the $\Gamma$-compact operators to the real numbers.
Out of a lifted elliptic differential operator $\widetilde{D}$,
we can define the $\Gamma$-index class $\mathrm{Ind}_{\Gamma}(\widetilde{D})$
by using the Connes-Skandalis idempotent \cite[II.9.$\alpha$ (p.131)]{MR1303779}
and send it by the induced homomorphism $(\mathrm{tr}_{\Gamma})_{\ast}$,
then the image $(\mathrm{tr}_{\Gamma})_{\ast}(\mathrm{Ind}_{\Gamma}(\widetilde{D}))$
equals the $\Gamma$-index $\mathrm{index}_{\Gamma}(\widetilde{D})$
and thus the Fredholm index $\mathrm{index} (D)$ of
the elliptic differential operator $D$ on the base manifold.
On the other hand,
there is another pseudo-differential calculus
on Heisenberg manifolds which is called
the Heisenberg calculus; see, for instance \cite{MR2417549}.
Roughly speaking, Heisenberg calculus is
``weighted'' calculus and the
product of the ``Heisenberg principal symbols'' are defined by convolution product.
When the Heisenberg principal symbol of $P$
is invertible, we call $P$ a Heisenberg elliptic operator.
Note that any Heisenberg elliptic operator is not elliptic.
For a Heisenberg elliptic operator $P$,
we can construct a parametrix by using its inverse,
so $P$ is a Fredholm operator if the base manifold is closed.
Thus the Fredholm index of $P$ on a closed manifold
is well defined, but
a solution of an index problem of $P$
does not obtained in general.
However,
index problems for Heisenberg elliptic operators
on contact manifolds or foliated manifolds
are solved by E. van Erp and P. F. Baum;
see \cite{MR3261009}, \cite{MR2680395}, \cite{MR2680396}, \cite{MR2746652}.
In this note,
we study that we can define
the $\Gamma$-index $\mathrm{index}_{\Gamma}(\widetilde{P})$
and the $\Gamma$-index class $\mathrm{Ind}_{\Gamma}(\widetilde{P})$
for a lifted Heisenberg elliptic differential operator $\widetilde{P}$
by using a parametrix.
Once these ingredients are defined,
the proof of the matching of three ingredients
$\mathrm{index}_{\Gamma}(\widetilde{P})$,
$(\mathrm{tr}_{\Gamma})_{\ast}(\mathrm{Ind}_{\Gamma}(\widetilde{P}))$
and $\mathrm{index} (P)$,
is straightforward; see subsection \ref{sub:main}.
We also investigate an example
of Heisenberg differential operators
on a contact manifold
with non-trivial $\Gamma$-index
by using the index formula in \cite{MR3261009}; see subsection \ref{sub:exm}.
\section{Short review of Atiyah's $\Gamma$-index theorem}
\label{sec:atiyah}
In this section,
we recall Atiyah's $\Gamma$-index theorem in
ordinary pseudo-differential calculus.
The main reference of this section is
Atiyah's paper \cite{MR0420729}.
Let $\widetilde{M} \to M$ be a Galois covering
with a deck transformation group $\Gamma$
over a closed manifold $M$ with a smooth measure $\mu$
and $D : C^{\infty}(E) \to C^{\infty}(F)$
an elliptic differential operator
on Hermitian vector bundles $E,F \to M$.
We lift these ingredients on $\widetilde{M}$
and denote by
$\widetilde{D} : C^{\infty}(\widetilde{E}) \to C^{\infty}(\widetilde{F})$
and $\tilde{\mu}$.
Let $\mathrm{Ker}_{L^{2}}(\widetilde{D})$ (resp. $\mathrm{Ker}_{L^{2}}(\widetilde{D}^{\ast})$)
be the $L^{2}$-solutions of $\widetilde{D}u = 0$ (resp. $\widetilde{D}^{\ast}u = 0$)
and
denote by $\Pi_{0}$ (resp. $\Pi_{1}$)
the orthogonal projection on a
closed subspace
$\mathrm{Ker}_{L^{2}}(\widetilde{D})$ (resp. $\mathrm{Ker}_{L^{2}}(\widetilde{D}^{\ast})$)
of the $L^{2}$-sections.
A $\Gamma$-invariant bounded operator $T$ on the
$L^{2}$-sections $L^{2}(\widetilde{E})$
of $\widetilde{E}$
is of \textit{$\Gamma$-trace class}
if $\phi T \psi \in L^{2}(\widetilde{E})$
is of trace class for any
compactly suppoted smooth functions $\phi , \psi$
on $\widetilde{M}$.
Denote by $\mathcal{L}^{1}_{\Gamma}$
the set of $\Gamma$-trace class operators
and $\mathrm{tr}_{\Gamma}(T)$ the \textit{$\Gamma$-trace}
of a $\Gamma$-trace class operator $T$ defined by
\[
\mathrm{tr}_{\Gamma}(T) = \textrm{Tr}(\phi T \psi) \in \mathbb{C}.
\]
Here, the right hand side is the trace of
a trace class operator $\phi T \psi$
and this quantity does not depend on
the choice of functions $\phi , \psi$.
By using ellipticity of $\widetilde{D}$,
operators $\phi \Pi_{0} \psi$ and $\phi \Pi_{1} \psi$
are smoothing operators on compact sets.
Thus $\Pi_{0}$ and $\Pi_{1}$ are of $\Gamma$-trace class
and thus one obtains the $\Gamma$-index of $\widetilde{D}$:
\[
\mathrm{index}_{\Gamma}(\widetilde{D})
= \mathrm{tr}_{\Gamma}(\Pi_{0}) - \mathrm{tr}_{\Gamma}(\Pi_{0}) \in \mathbb{R}.
\]
In the context of the $\Gamma$-index theorem,
the most important class of $\Gamma$-trace class operators
is the lifts of almost local smoothing operators on $M$.
Let $S$ be an almost local smoothing operator with
a smooth kernel $k_{S}$
and $\widetilde{S}$ a lift of $S$.
Then $\widetilde{S}$ is of $\Gamma$-trace class
and its $\Gamma$-trace is calculated by the following:
\begin{equation}
\label{eq:gammatr}
\mathrm{tr}_{\Gamma}(\widetilde{S})
= \int_{M}\mathrm{tr}(k_{S}(x,x)) d\mu
= \mathrm{Tr}(S). \tag{$\ast$}
\end{equation}
Denote by $\mathcal{K}_{\Gamma}$
the $C^{\ast}$-closure of $\mathcal{L}^{1}_{\Gamma}$
and $K_{0}(\mathcal{K}_{\Gamma})$
the analytic $K_{0}$-group.
Then $\mathrm{tr}_{\Gamma}$ induces
a homomorphism of abelian groups by substitution:
\[
(\mathrm{tr}_{\Gamma})_{\ast}
: K_{0}(\mathcal{K}_{\Gamma}) \to \mathbb{R}.
\]
On the other hand,
since $D$ is an elliptic differential operator,
there exist an almost local parametrix $Q$ and
almost local smoothing operators $S_{0},S_{1}$
such that one has
$QD = 1 - S_{0}$ and $DQ = 1 - S_{1}$.
Denote by
$\widetilde{Q}$, $\widetilde{S_{0}}$ and $\widetilde{S_{1}}$
lifts of these operators
and then one has same relations
$\widetilde{Q}\widetilde{D} = 1 - \widetilde{S_{0}}$
and $\widetilde{D}\widetilde{Q} = 1 - \widetilde{S_{1}}$.
Set
\[
e_{\widetilde{D}} =
\begin{bmatrix}
\widetilde{S_{0}}^{2} &
\widetilde{S_{0}}(1 + \widetilde{S_{0}}) \widetilde{Q} \\
\widetilde{S_{1}}\widetilde{D} & 1-\widetilde{S_{1}}^{2}
\end{bmatrix} \quad \text{and} \quad
e_{1} =
\begin{bmatrix}
0 & 0 \\
0 & 1
\end{bmatrix}.
\]
By $\widetilde{Q}\widetilde{S_{1}} = \widetilde{S_{0}}\widetilde{Q}$
and $\widetilde{S_{1}}\widetilde{D} = \widetilde{D}\widetilde{S_{0}}$,
one has $e_{\widetilde{D}}^{2} = e_{\widetilde{D}}$, that is,
$e_{\widetilde{D}}$ is an idempotent.
Note that this idempotent $e_{\widetilde{D}}$
is called the Connes-Skandalis idempotent;
see, for instance \cite[II.9.$\alpha$ (p.131)]{MR1303779}.
Moreover, a difference $e_{\widetilde{D}} - e_{1}$
is of $\Gamma$-trace class.
Hence
we can define a $\Gamma$-index class
\[
\mathrm{Ind}_{\Gamma}(\widetilde{D})
= [e_{\widetilde{D}}] - [e_{1}] \in K_{0}(\mathcal{K}_{\Gamma}).
\]
By the definition of a map $(\mathrm{tr}_{\Gamma})_{\ast}$
and Atiyah's paper, one has the following:
\begin{thm}[Atiyah's $\Gamma$-index theorem {\cite[Theorem 3.8]{MR0420729}}]
In the above settings, we have the following equality:
\[
\mathrm{index}_{\Gamma}(\widetilde{D})
= (\mathrm{tr}_{\Gamma})_{\ast}(\mathrm{Ind}_{\Gamma}(\widetilde{D}))
= \mathrm{index} (D) \in \mathbb{Z}.
\]
\end{thm}
As described in subsection \ref{sub:main},
Atiyah's proof of matching of these ingredients in the above equality
does not essentially use ellipticity.
Note that
ellipticity of $D$ and $\widetilde{D}$ is only used in
the definition of these ingredients.
\section{Atiya's $\Gamma$-index theorem in Heisenberg calculus}
Let $(M,H)$ be a closed Heisenberg manifold,
that is, $M$ is a closed manifold and $H \subset TM$ is a hyperplane bundle.
Let
$P : C^{\infty}(E) \to C^{\infty}(F)$
be a Heisenberg elliptic differential operator on
Hermitian vector bundles $E,F \to (M,H)$,
that is,
the Heisenberg principal symbol $\sigma_{H}(P)$ of $P$
is an invertible element.
In this section,
we prove the $\Gamma$-index theorem for $P$,
which is analogous to
Atiyah's $\Gamma$-index theorem.
Note that $P$ is not an elliptic operator
in the sense of ordinary pseudo-differential calculus.
\subsection{Statement and proof}
\label{sub:main}
By \cite[Proposition 3.3.1]{MR2417549},
there exist parametrix $Q$ and
smoothing operators $S_{0},S_{1}$
such that one has
$QP = 1 - S_{0}$ and $PQ = 1 - S_{1}$.
Thus
$P$ is a Fredholm operator
and one has the Fredholm index $\mathrm{index} (P) \in \mathbb{Z}$ of $P$
by compactness of $M$.
Moreover, since a integral kernel of $Q$
is smooth off the diagonal,
we can choose $Q$ as
an almost local operator
and then $S_{0}$ and $S_{1}$ are
also almost local operators.
Let $\widetilde{M} \to M$ be a Galois covering
with a deck transformation group $\Gamma$
over a closed manifold $M$ with a smooth measure $\mu$.
We lift all structures on $M$ to $\widetilde{M}$.
Then $(\widetilde{M} , \widetilde{H})$
is a Heisenberg manifold,
$\widetilde{P} : C^{\infty}(\widetilde{E}) \to C^{\infty}(\widetilde{F})$ is a
Heisenberg elliptic differential operator
and one has $\widetilde{Q}\widetilde{P} = 1 - \widetilde{S}_{0}$
and $\widetilde{P}\widetilde{Q} = 1 - \widetilde{S}_{1}$.
Since $P$ is a differential operator (in particular, $P$ is local),
there exists a constant $C = C(\widetilde{P} , \phi ) > 0$ such that
we have an inequality
\[
\| \widetilde{P}(\phi f) \|_{L^{2}}
\leq C( \|\chi \widetilde{P}f \|_{L^{2}} + \| \chi f \|_{L^{2}} )
\]
for any $f \in C^{\infty}(\widetilde{E})$; see \cite[Proposition 3.3.2]{MR2417549}.
Here,
$\phi , \chi \in C_{c}^{\infty}(\widetilde{M})$ are compactly supported
smooth functions and
one assumes $\chi = 1$ on the support of $\phi$.
Thus by using Atiyah's technique of the proof of \cite[Proposition 3.1]{MR0420729},
we have the following:
\begin{lem}
\label{lem:closure}
The minimal domain of $\widetilde{P}$ equals the maximal domain of $\widetilde{P}$.
\end{lem}
By Lemma \ref{lem:closure},
$\widetilde{P}$ has the unique
closed extension denoted by the same letter $\widetilde{P}$
and thus the closure of the formal adjoint of $\widetilde{P}$
(the formal adjoint is also Heisenberg elliptic)
equals the Hilbert space adjoint $\widetilde{P}^{\ast}$.
On the other hand,
any $L^{2}$-solutions of $\widetilde{P}u = 0$ and $\widetilde{P}^{\ast}u = 0$
are smooth by
existence of a parametrix.
Thus the orthogonal projection $\Pi_{0}$ (resp. $\Pi_{1}$)
onto a closed subspace
$\mathrm{Ker}_{L^{2}}(\widetilde{P})$ (resp. $\mathrm{Ker}_{L^{2}}(\widetilde{P}^{\ast})$)
of the $L^{2}$-sections
is of $\Gamma$-trace class
since operators $\phi \Pi_{0} \psi$ and $\phi \Pi_{1} \psi$
are smoothing operators on compact sets.
Thus one obtains the well-defined $\Gamma$-index of $\widetilde{P}$.
\begin{dfn}
The $\Gamma$-index of $\widetilde{P}$ is defined to be
\[
\mathrm{index}_{\Gamma}(\widetilde{P})
= \mathrm{tr}_{\Gamma}(\Pi_{0}) - \mathrm{tr}_{\Gamma}(\Pi_{0}) \in \mathbb{R}.
\]
\end{dfn}
By using operators $\widetilde{P}, \widetilde{Q}, \widetilde{S}_{0}$
and $\widetilde{S}_{1}$, we define
\[
e_{\widetilde{P}} =
\begin{bmatrix}
\widetilde{S_{0}}^{2} &
\widetilde{S_{0}}(1 + \widetilde{S_{0}}) \widetilde{Q} \\
\widetilde{S_{1}}\widetilde{P} & 1-\widetilde{S_{1}}^{2}
\end{bmatrix} \quad \text{and} \quad
e_{1} =
\begin{bmatrix}
0 & 0 \\
0 & 1
\end{bmatrix}.
\]
Since a difference $e_{\widetilde{P}} - e_{1}$ is of $\Gamma$-trace class,
one can define a $\Gamma$-index class of $\widetilde{P}$.
\begin{dfn}
We define $\Gamma$-index class $\mathrm{Ind}_{\Gamma}(\widetilde{P})$ of $\widetilde{P}$ by
\[
\mathrm{Ind}_{\Gamma}(\widetilde{P})
= [e_{\widetilde{P}}] - [e_{1}] \in K_{0}(\mathcal{K}_{\Gamma}).
\]
\end{dfn}
By using a $\Gamma$-trace,
we have the $\Gamma$-index theorem
in Heisenberg calculus.
\begin{thm}
\label{thm:GammaHeisen}
Let $P$ be a Heisenberg elliptic differential operator
on a closed Heisenberg manifold $(M,H)$
and $\widetilde{P}$ its lift as previously.
Then one has
\[
\mathrm{index}_{\Gamma}(\widetilde{P})
= (\mathrm{tr}_{\Gamma})_{\ast}(\mathrm{Ind}_{\Gamma}(\widetilde{P}))
= \mathrm{index} (P) \in \mathbb{Z}.
\]
\end{thm}
\begin{proof}
First, note that equalities
\begin{align*}
1 - S_{0}^{2} &= 1 - (1 - QP)^{2} = (2Q - QPQ)P \quad \text{and} \\
1 - S_{1}^{2} &= 1 - (1 - PQ)^{2} = P(2Q - QPQ),
\end{align*}
and note that operators $2Q - QPQ$ , $S_{0}^{2}$ and $S_{1}^{2}$
are almost local operators.
Thus
by Atiyah's technique in \cite[Section 5]{MR0420729},
one has
\[
\mathrm{index}_{\Gamma}(\widetilde{P})
= \mathrm{tr}_{\Gamma}(\Pi_{0}) - \mathrm{tr}_{\Gamma}(\Pi_{1})
= \mathrm{tr}_{\Gamma}(\widetilde{S_{0}}^{2}) - \mathrm{tr}_{\Gamma}(\widetilde{S_{1}}^{2}).
\]
Next, the definition of a map $(\mathrm{tr}_{\Gamma})_{\ast}$,
one has
\[
(\mathrm{tr}_{\Gamma})_{\ast}(\mathrm{Ind}_{\Gamma}(\widetilde{P}))
= \mathrm{tr}_{\Gamma}
\begin{bmatrix}
\widetilde{S_{0}}^{2} &
\widetilde{S_{0}}(1 + \widetilde{S_{0}}) \widetilde{Q} \\
\widetilde{S_{1}}\widetilde{P} & -\widetilde{S_{1}}^{2}
\end{bmatrix}
= \mathrm{tr}_{\Gamma}(\widetilde{S_{0}}^{2}) - \mathrm{tr}_{\Gamma}(\widetilde{S_{1}}^{2}).
\]
Since operators $\widetilde{S}_{0}^{2}$ and $\widetilde{S}_{1}^{2}$ are
lifts of almost local smoothing operators,
one has
\[
\mathrm{index} (P)
= \mathrm{Tr}(S_{0}^{2}) - \mathrm{Tr}(S_{1}^{2})
= \mathrm{tr}_{\Gamma}(\widetilde{S_{0}}^{2}) - \mathrm{tr}_{\Gamma}(\widetilde{S_{1}}^{2})
\]
by using (\ref{eq:gammatr}) in Section \ref{sec:atiyah}.
This proves the equality in the theorem.
\end{proof}
\begin{rmk}
As pointed out in \cite[Section 4]{Erppolycontant},
the results in \cite[Section 3]{MR2417549} hold
verbatim for arbitrary codimension.
That is,
we do not need to assume that
a distribution $H$ is of codimension $1$.
\end{rmk}
\subsection{Example}
\label{sub:exm}
Index problems for Heisenberg elliptic operators
on arbitrary closed Heisenberg manifolds
are not solved yet.
However, van Erp \cite{MR2680395,MR2680396} and
Baum and van Erp \cite{MR3261009} solved
the index problem on contact manifolds,
which are
good examples of Heisenberg manifolds.
In this subsection,
we investigate an example of Heisenberg elliptic differential operators
with non-trivial $\Gamma$-index
on a Galois covering over a closed contact manifold
by using the index formula for subLaplacians.
Let $T^{2} = S^{1} \times S^{1} = \{ (e^{ix},e^{iy}) \}$ be a $2$-torus
and set
\[
e(x,y) =
\begin{bmatrix}
f(x) & g(x) + h(x)e^{iy} \\
g(x) + h(x)e^{-iy} & 1-f(x)
\end{bmatrix}.
\]
Here, let $f$ be a $[0,1]$-valued
$2\pi \mathbb{Z}$-periodic function on $\mathbb{R}$
such that $f(0) = 1$ and $f(\pi) = 0$
and
set $g(x) = \chi_{[0, \pi]}(x)\sqrt{f(x)-f(x)^{2}}$ and
$h(x) = \chi_{[\pi, 2\pi]}(x)\sqrt{f(x)-f(x)^{2}}$; see \cite[Section I. 2]{Loringthesis}.
Moreover, we assume $f$, $g$ and $h$ are smooth functions.
Then $e$ defines an $M_{2}(\mathbb{C})$-valued
smooth function on $T^{2}$.
Since $e$ is an idempotent of rank $1$,
$e$ defines an complex line bundle $E$ on $T^{2}$.
As well known,
the first Chern class $c_{1}(E)$ of $E$ is given by
a $2$-form
\[
\frac{-1}{2\pi i}\mathrm{tr}(e(de)^{2})
= \frac{-1}{\pi}(hh' + 2f'h^{2} -2fhh')dx \wedge dy.
\]
Thus, by using an equality $h^{2} = f-f^{2}$ on $[\pi , 2\pi]$,
we can calculate the first Chern number of $E$:
\[
\int_{T^{2}}c_{1}(E) = -\int_{\pi}^{2\pi}f'dx = -1.
\]
Let $T^{3} = T^{2} \times S^{1} = \{ (e^{ix}, e^{iy} , e^{iz}) \}$ be a $3$-torus and
$q : T^{3} \to T^{2}$ the projection onto $T^{2}$ of the first component.
Set $\theta_{k} = \cos (kz)dx - \sin (kz)dy$ for a positive integer $k$,
$H_{k} = \mathrm{ker}(\theta_{k})$,
$f_{l}(x,y,z) = e^{ilz} + 1$ for a integer $l$
and $F = q^{\ast}E$.
Then $(T^{3},H_{k})$ is a contact manifold
and $H_{k}$ is a flat vector bundle.
Denote by $T_{k}$ the Reeb vector field for $\theta_{k}$ and
$\Delta_{H_{k}}^{F}
= -\nabla_{X_{k}}^{F}\nabla_{X_{k}}^{F}
- \nabla_{Y_{k}}^{F}\nabla_{Y_{k}}^{F}$
the sum of squares on $F$,
where $\{ X_{k} , Y_{k} \}$ is a local frame of $H_{k}$.
Set
\[
P_{k,l} = \Delta_{H_{k}}^{F} + if_{l}\nabla_{T_{k}}^{F}.
\]
Since the values of $f_{l} - n$ contained in $\mathbb{C}^{\times}$ for any odd integer $n$,
an operator $P_{k,l} : C^{\infty}(F) \to C^{\infty}(F)$
is a Heisenberg elliptic differential operator of Heisenberg order $2$.
By the index formula for $P_{k,l}$ in \cite[Example 6.5.3]{MR3261009}, one has
\[
\mathrm{index} (P_{k,l}) = \int_{T^{3}}\frac{-1}{2\pi i}e^{-ilz}de^{ilz} \wedge c_{1}(F)
= \frac{-1}{2\pi i}\int_{S^{1}}e^{-ilz}de^{ilz} \int_{T^{2}}c_{1}(E)
= l.
\]
Note that
a contact structure $H_{k}$ is a lift of $H_{1}$ by
a $k$-fold cover $p_{k} : T^{3} \to T^{3}$;
$(e^{ix},e^{iy},e^{iz}) \mapsto (e^{ix},e^{iy},e^{ikz})$.
Since the lift $\widetilde{P_{1,l}}$ of a subLaplacian $P_{1,l}$
by $p_{k}$ equals $P_{k,kl}$,
we have the $\Gamma (= \mathbb{Z}/k\mathbb{Z})$-index of $\widetilde{P_{1,l}}$:
\[
\mathrm{index}_{\Gamma}(\widetilde{P_{1,l}}) = \frac{1}{k}\mathrm{index} (\widetilde{P_{1,l}})
= \frac{1}{k}\mathrm{index} (P_{k,kl}) = l = \mathrm{index} (P_{1,l}).
\]
Next, we consider a general Galois covering of $T^{3}$.
Let $X \to T^{3}$ be a Galois covering
with a deck transformation group $\Gamma$,
which is a quotient of $\pi_{1}(T^{3}) = \mathbb{Z}^{3}$,
for example, $X = \mathbb{R}^{3}$ and $\Gamma = \mathbb{Z}^{3}$
the universal covering.
By Theorem \ref{thm:GammaHeisen},
we have a non-trivial $\Gamma$-index as follows:
\[
\mathrm{index}_{\Gamma} (\widetilde{P_{k,l}}) = \mathrm{index} (P_{k,l}) = l.
\]
\bibliographystyle{plain}
| f7096ff14ed9ecb1a085ff141f2af1e3c16b97c2 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section*{Appendix}
\subsection{Proof of Lemma \ref{lem:sigmak}}
\begin{IEEEproof}
We first state the following result that will be used in the proof.
\begin{lemma}[Ger\v{s}hgorin circle theorem \cite{HJ12}]
Let A be a complex $n \times n$ matrix, with entries $a_{ij}$. Then, every eigenvalue of A lies within at least one of the Ger\v{s}hgorin discs $D_i(A) (i=1,...,n)$, where
$D_i(A):=\{ z \in \C: |z-a_{ii}| \leq \sum_{j\neq{i}} \left|a_{ij}\right|\}$.
\end{lemma}
For any given $\I$ with $|\I| \leq k$, since $W$ has unit-norm columns, and $|W^\dag_iW_j|\leq \mu$ for all $i \neq j$, from Ger\v{s}hgorin circle theorem, we have $\|I- W^\dag_{\I}W_{\I}\|\leq (k-1)\mu <1$, where the last inequality follows from $k\mu<1$. Then,
\begin{align*}
\|(W^\dag_{\I}W_{\I})^{-1}\| = &\|\sum_{i=0}^\infty (I- W^\dag_{\I}W_{\I})^i\| \leq \sum_{i=0}^\infty \|(I- W^\dag_{\I}W_{\I})^i\|\nonumber \\
\leq & 1/(1-(k-1)\mu)\nonumber.
\end{align*}
The lemma follows from the definition of $\sigma_k$.
\end{IEEEproof}
\subsection{Proof of Lemma \ref{lem:cond}}
\begin{IEEEproof}
For any $\Delta \in \C^{t\times n}$, $\langle L'+\Delta W^T, C'-\Delta \rangle$ is feasible to (\ref{eqn:opt}). Let $G$ be such that $\|G\|=1$, $\left\langle G, \P_{T'^{\perp}}(\Delta W^T) \right\rangle = \|\P_{T'^{\perp}}(\Delta W^T)\|_{*}$ and $\P_{T'}(G)=0$. Then $\P_{T'}(Q)+G$ is a subgradient of $\|L'\|_{*}$. Let $F$ be such that $F_i=-\Delta_i/\|\Delta_i\|_2$ if $i \in \bar{\I}^c$ and $\Delta_i \neq \bm0$, and $F_i=\bm0$ otherwise. Then $\P_{\bar{\I}} (QW^{\ddagger})/\lambda +F$ is a subgradient of $\|C'\|_{1,2}$. Then
\begin{equation}\label{eqn:worse}
\begin{aligned}
&\|L'+ \Delta W^T\|_{*}+\lambda \|C'-\Delta\|_{1,2}-\|L'\|_{*}-\lambda\|C'\|_{1,2} \\
\geq & \langle \P_{T'}(Q)+G, \Delta W^T \rangle -\lambda \langle \P_{\bar{\I}} (QW^{\ddagger})/\lambda +F, \Delta\rangle\\
=& \|\P_{T'^{\perp}}(\Delta W^T)\|_{*} + \lambda \|\P_{\bar{\I}^c}(\Delta)\|_{1,2} + \langle Q-\P_{T'^\perp}(Q), \Delta W^T \rangle \\
&-\langle QW^{\ddagger}-\P_{\bar{\I}^c} (QW^{\ddagger}),
\Delta\rangle \\
\geq & (1-\|\P_{T'^\perp}(Q)\|)\|\P_{T'^{\perp}}(\Delta W^T)\|_{*} \\ &+(\lambda-\|(QW^{\ddagger})_{\bar{\I}^c}\|_{\infty,2} )\|\P_{\bar{\I}^c}(\Delta)\|_{1,2}\\
\geq & 0
\end{aligned}
\end{equation}
From (\ref{eqn:worse}), $\langle L', C'\rangle$ is an optimal solution to (\ref{eqn:opt}). If (\ref{eqn:cond}) holds with strict inequality, the last inequality of (\ref{eqn:worse}) is strict unless
\begin{equation}\label{eqn:temp1}
\|\P_{T'^{\perp}}(\Delta W^T)\|_{*}= \|\P_{\bar{\I}^c}(\Delta)\|_{1,2}=0.
\end{equation}
(\ref{eqn:temp1}) implies that $ \Delta W^T \in \P_{T'}$ and $\Delta \in \P_{\bar{\I}}$. Note that $\Delta \in \P_{\bar{\I}}$ implies that $\Delta W^T \in \P_{\bar{\J}}$. Then
\begin{align}\label{eqn:med}
\P_{\bar{\J}}(\Delta W^T)&=\Delta W^T =\P_{T'}(\Delta W^T) \nonumber\\
&=\P_{U'}(\Delta W^T)+\P_{V'}\P_{U'^\perp}(\Delta W^T)\nonumber\\
&= \P_{\bar{\J}}\P_{U'}(\Delta W^T)+\P_{V'}\P_{U'^\perp}(\Delta W^T),
\end{align}
where the last equality holds since $\P_{\bar{\J}}(\Delta W^T)=\Delta W^T$.
Thus, from (\ref{eqn:med}) we have $\P_{\bar{\J}}\P_{U'^{\perp}}(\Delta W^T)=\P_{V'}\P_{U'^\perp}(\Delta W^T)$,
which means $\P_{U'^\perp}(\Delta W^T) \in \P_{\bar{\J}} \cap \P_{V'}$. Then $\P_{U'^\perp}(\Delta W^T)$ is $0$ from the assumption. Then, $\P_{\bar{U}}(\Delta W^T)=\P_{U'}(\Delta W^T) =\Delta W^T$, where the first equality holds from (\ref{eqn:UU'}). Therefore, for any optimal solution $\langle L'+\Delta W^T, C'-\Delta \rangle$ for some $\Delta \neq 0$ to (\ref{eqn:opt}), $\Delta W^T \in \P_{\bar{U}}$, and $\Delta \in \P_{\bar{\I}}$. The claim follows.
\end{IEEEproof}
\subsection{Construction of $Q$}
Here we demonstrate that $Q$ in (\ref{eqn:Q}) is well defined. The key is to show (a) there exists $\hat{H} \in \G(C')$ such that (\ref{eqn:hath}) holds, and (b) the infinite sum in (\ref{eqn:delta2}) converges. We prove these two properties through the following lemmas.
\begin{lemma}\label{lem:convex}
There exists $\hat{H} \in \G(C')$ such that (\ref{eqn:hath}) holds.
\end{lemma}
\begin{proof}
Since $\langle L', C' \rangle$ is an optimal solution to the Oracle problem (\ref{eqn:oracle}), there exists $G',A' \in \C ^{t \times p}$, $B',Z \in \C^{t \times n}$, and some $\hat{H} \in \G (C)$ such that
\begin{equation}\label{eqn:seq1}
(\bar{U}\hat{V}^\dag + G'+ \P_{\bar{U}^\perp} (A'))W^{\ddagger}= \lambda (\hat{H}+Z)+\P_{\I^c}(B'),
\end{equation}
where $\P_{T'^\perp}(G')=0$ and $\P_{\I}(Z)=0$. Then
\begin{equation}\label{eqn:seq2}
\P_{\bar{U}}\P_{\bar{\I}}(((\bar{U}\hat{V}^\dag + G'+ \P_{\bar{U}^\perp} (A'))W^{\ddagger})
=\bar{U}\hat{V}^\dag W^{\ddagger}_{\bar{\I}},
\end{equation}
\begin{equation}\label{eqn:seq3}
\P_{\bar{U}}\P_{\bar{\I}}( \lambda (\hat{H}+Z)+\P_{\I^c}(B'))= \lambda \P_{\bar{U}}\P_{\bar{\I}}(\hat{H}) =\lambda\bar{U}\bar{U}^\dag \hat{H} \end{equation}
Combining (\ref{eqn:seq1})-(\ref{eqn:seq3}), we have
\begin{equation}\label{eqn:multiu}
\bar{U}\hat{V}^\dag W^{\ddagger}_{\bar{\I}} = \lambda\bar{U}\bar{U}^\dag \hat{H}.
\end{equation}
By multiplying $\bar{U}^\dag$ to both sides of (\ref{eqn:multiu}), we obtain Lemma \ref{lem:convex}.
\end{proof}
\begin{lemma}\label{lem:psi}
\begin{equation} \nonumber
\psi:=\|\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}}\| \leq \tilde{\psi}<1
\end{equation}
\end{lemma}
\begin{proof}
\begin{align}\nonumber
&\|\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}}(X)\| \\\nonumber
=&\|X\hat{V}\hat{V}^\dag W^{\ddagger}_{\bar{\I}}(W_{\bar{\I}}^T W^{\ddagger}_{\bar{\I}})^{-1}W_{\bar{\I}}^T \hat{V}\hat{V}^\dag\| \\\nonumber
\stackrel{(\rm{a})}{=}&\|X\hat{V}(\lambda\bar{U}^\dag \hat{H})(W_{\bar{\I}}^T W^{\ddagger}_{\bar{\I}})^{-1}(\lambda\bar{U}^\dag \hat{H})^\dag\hat{V}^\dag\| \\\nonumber
\leq & \|X\|\|\hat{V}\bar{U}^\dag\| \|\lambda \hat{H}\|\|(W_{\bar{\I}}^\dag W_{\bar{\I}})^{-1}\|\|\lambda\hat{H}^\dag \|\|\bar{U}\hat{V}^\dag\|\\\nonumber
\stackrel{(\rm{b})}{\leq} & \|X\| \cdot 1 \cdot\lambda \sqrt{k} \cdot \sigma_k \cdot \lambda \sqrt{k}\cdot 1\\\nonumber
\stackrel{(\rm{c})}{\leq} & \|X\| \lambda_{\max} \tilde{k} \sigma_{\tilde{k}} \stackrel{(\rm{d})}{=} \|X\| \tilde{\psi},
\end{align}
where (a) follows from Lemma \ref{lem:convex}, (b) follows from the fact that $\hat{H}$ has at most $k$ nonzero columns with unit-norm, (c) follows from the property that $\lambda \leq \lambda_{\max}$, $k \leq \tilde{k}$ and $\sigma_{k} \leq \sigma_{\tilde{k}}$, and (d) follows from the definition of $\tilde{\psi}$. Then Lemma \ref{lem:psi} follows.
\end{proof}
\begin{lemma}\label{lem:inverse}
$\P_{\hat{V}} (I-\P_{W_{\bar{\I}}})\P_{\hat{V}}$ is an injection from $\P_{\hat{V}}$ to $\P_{\hat{V}}$, and its inverse operation is $(I+\sum_{i=1}^\infty (\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}})^i)$.
\end{lemma}
\begin{proof}
Since $\|\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}}\| <1$ from Lemma \ref{lem:psi}, then $(I+\sum_{i=1}^\infty (\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}})^i)$ is well defined. For any $X \in \P_{\hat{V}}$, we have
\begin{align}
&\P_{\hat{V}}(I-\P_{W_{\bar{\I}}})\P_{\hat{V}}(I+\sum_{i=1}^\infty (\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}})^i)(X)\nonumber\\=&\P_{\hat{V}}(I-\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}})(I+\sum_{i=1}^\infty (\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}})^i)(X)\nonumber\\
=& \P_{\hat{V}}(X)=X.
\end{align}
Then the lemma follows.
\end{proof}
\subsection{Proof of Lemma \ref{lem:certificate}}
\begin{proof}
We need to show that $Q$ defined in (\ref{eqn:Q}) satisfies all the conditions in (\ref{eqn:cond}). We first summarize some properties that will be used in the proof. Since $W$ has unit-norm columns, $|W^\dag _i W_j| \leq \mu$ for all $i \neq j$, and $|\bar{\I}|\leq k$, we have
\begin{equation}\label{eqn:WI}
\|W_{\bar{\I}}\|=\sqrt{\lambda_{\max}(W_{\bar{\I}}^\dag W_{\bar{\I}})} \leq \sqrt{1+(k-1)\mu},
\end{equation}
where the inequality follows from the Ger\v{s}hgorin circle theorem.
From $|\bar{\I}|\leq k$ and $|W^\dag _i W_j| \leq \mu$ for all $i \neq j$, we have $\|(W_{\bar{\I}}^\dag W_{\bar{\I}^c})\|_{\infty,2} \leq \sqrt{k} \mu$. Since $\hat{H}$ has at most $k$ unit-norm columns while other columns are zero, we have
\begin{equation}\label{eqn:lambdaH}
\|\lambda \hat{H} \| \leq \lambda \sqrt{k}.
\end{equation}
Step 1: verification of (a) of (\ref{eqn:cond}).
\begin{equation}\label{eqn:U}
\P_{U'}(Q)\stackrel{(\rm{a})}{=}\P_{\bar{U}}(Q)= \bar{U}\hat{V}^\dagger+\P_{\bar{U}}(\Phi)-\P_{\bar{U}}(\Phi)-0= \bar{U}\hat{V}^\dagger,
\end{equation}
where (a) follows from (\ref{eqn:space}). From (\ref{eqn:UU'}), we have
\begin{equation} \nonumber
\hat{V}\hat{V}^\dag =V' U'^\dag \bar{U} \bar{U}^\dag U' V'^\dagger\stackrel{(\rm{b})}{=}V' U'^\dag U'U'^\dag U' V'^\dagger=V'V'^\dagger,
\end{equation}
where (b) follows from (\ref{eqn:UU'}). Thus, $\P_{V'}(\cdot)=\P_{\hat{V}}(\cdot)$. Then
\begin{align}\label{eqn:V}
\P_{V'}(Q)&=\P_{\hat{V}}(Q)\stackrel{(\rm{c})}{=} \bar{U}\hat{V}^\dagger+\P_{\hat{V}}(\Phi)-\P_{\hat{V}}\P_{\bar{U}}(\Phi) \nonumber\\&- \P_{\hat{V}} (I-\P_{W_{\bar{\I}}})\P_{\hat{V}}(I+\sum_{i=1}^\infty (\P_{\hat{V}}\P_{W_{\bar{\I}}}\nonumber\\
& \P_{\hat{V}})^i)\P_{\hat{V}} \P_{\bar{U}^\perp}(\Phi)\nonumber\\
&\stackrel{(\rm{d})}{=} \bar{U}\hat{V}^\dagger+\P_{\hat{V}}(\Phi)-\P_{\hat{V}}\P_{\bar{U}}(\Phi)-\P_{\hat{V}} \P_{\bar{U}^\perp}(\Phi) \nonumber\\
&=\bar{U}\hat{V}^\dagger.
\end{align}
(c) follows since $\P_{W_{\bar{\I}}}$, $\P_{\hat{V}}$, and $\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}}$ are all given by right matrix multiplication, while $\P_{\bar{U}^\perp}$
is given by left matrix
multiplication. (d) follows from Lemma \ref{lem:inverse}. Combining (\ref{eqn:U}) and (\ref{eqn:V}), we obtain that (a) of (\ref{eqn:cond}) holds.
Step 2: verification of (b) of (\ref{eqn:cond}).
\begin{align}\label{eqn:stp2}
&\|\P_{T'^{\perp}}(Q)\|=\|\P_{\hat{V}^{\perp}}\P_{\bar{U}^{\perp}}(\Phi)-\nonumber \\
& \quad \P_{\bar{U}^{\perp}}\P_{\hat{V}^\perp}(I-\P_{W_{\bar{\I}}})\P_{\hat{V}}(I+\sum_{i=1}^\infty (\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}})^i)\P_{\hat{V}}(\Phi)\|\nonumber \\
\leq & \|\Phi\|+ (1+\sum_{i=1}^{\infty}\psi^i)\|\Phi\| = \frac{2-\psi}{1-\psi} \|\Phi\|\nonumber\\
\stackrel{(\rm{e})}{\leq} & \frac{2-\psi}{1-\psi} \|\lambda\hat{H} \|\|(W_{\bar{\I}}^\dag W_{\bar{\I}})^{-1}\| \|W_{\bar{\I}}^T\| \nonumber\\
\stackrel{(\rm{f})}{\leq} & \frac{2-\psi}{1-\psi}\lambda \sqrt{k} \sigma_k \sqrt{1+(k-1)\mu}\\
\stackrel{(\rm{g})}{\leq} & \frac{2-\tilde{\psi}}{1-\tilde{\psi}} \sqrt{\frac{\tilde{\psi}}{\tilde{k}\sigma_{\tilde{k}}}} \sqrt{\tilde{k}} \sigma_{\tilde{k}} \sqrt{1+(\tilde{k}-1)\mu}\\
\stackrel{(\rm{h})}{\leq}& \frac{2-\tilde{\psi}}{1-\tilde{\psi}} \sqrt{\tilde{\psi}} \sqrt{\frac{1+(\tilde{k}-1)\mu}{1-(\tilde{k}-1)\mu}}\\
\stackrel{(\rm{i})}{\leq} & \frac{2-\tilde{\psi}}{1-\tilde{\psi}} \sqrt{\tilde{\psi}} \sqrt{\frac{1+c}{1-c}} \stackrel{(\rm{j})}{\leq} 1 \nonumber.
\end{align}
where (e) follows from the definition of $\Phi$, and (f) follows from (\ref{eqn:WI}) and (\ref{eqn:lambdaH}). (g) follows from the property that $\psi \leq \tilde{\psi}$, $1 \leq k \leq \tilde{k}$, $\lambda \leq \lambda_{\max,\tilde{k}}$, and $\sigma_k \leq \sigma_{\tilde{k}}$. (h) follows from Lemma \ref{lem:sigmak}. (i) follow from $\tilde{k}\mu \leq c$, and (j) follows from (\ref{eqn:psistar}). Then (b) of (\ref{eqn:cond}) holds.
Step 3: verification of (c) of (\ref{eqn:cond}). First consider
\begin{align}
&(\Delta_2W^{\ddagger})_{\bar{\I}}\nonumber\\
=& (\P_{\bar{U}^\perp}(I-\P_{W_{\bar{\I}}})\P_{\hat{V}}(I+\sum_{i=1}^\infty (\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}})^i)\P_{\hat{V}}(\Phi) W^{\ddagger})_{\bar{\I}}\nonumber\\
\stackrel{(\rm{k})}{=}& (\P_{\bar{U}^\perp}\P_{\hat{V}}(I+\sum_{i=1}^\infty (\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}})^i)\P_{\hat{V}}(\Phi))(I-\nonumber\\ & W^{\ddagger}_{\bar{\I}}(W_{\bar{\I}}^T W^{\ddagger}_{\bar{\I}})^{-1}W_{\bar{\I}}^T)W^{\ddagger}_{\bar{\I}} =0 \nonumber
\end{align}
where (k) holds since $\P_{W_{\bar{\I}}}$, $\P_{\hat{V}}$, and $\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}}$ are all given by right matrix multiplication, while $\P_{\bar{U}^\perp}$ is given by left matrix multiplication. Then
\begin{align}
(QW^{\ddagger})_{\bar{\I}}&=(\bar{U}\hat{V}W^{\ddagger}+\Phi W^{\ddagger} -\P_{\bar{U}}(\Phi)W^{\ddagger})_{\bar{\I}}-(\Delta_2W^{\ddagger})_{\bar{\I}}\nonumber \\
&= \bar{U}\hat{V} W^{\ddagger}_{\bar{\I}}+\Phi W^{\ddagger}_{\bar{\I}}-\P_{\bar{U}}(\Phi)W^{\ddagger}_{\bar{\I}}-0\nonumber\\
&\stackrel{(\rm{l})}{=} \lambda \bar{U}\bar{U}^\dag \hat{H}+\lambda \hat{H}- \lambda \bar{U}\bar{U}^\dag \hat{H}\nonumber\\
&=\lambda \hat{H} \in \lambda \G(C'),
\end{align}
where (l) follows from Lemma \ref{lem:convex} and the definition of $\Phi$ in (\ref{eqn:phi}). Then (c) of (\ref{eqn:cond}) holds.
Step 4: verification of (d) of (\ref{eqn:cond}). First consider
\begin{align}
&\|(\Delta_2W^{\ddagger})_{\bar{\I}^c}\|_{\infty,2}\nonumber\\
=& \|\P_{\bar{U}^\perp}\P_{\hat{V}}(I+\sum_{i=1}^\infty (\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}})^i)\nonumber \\
& \cdot \Phi\hat{V}\hat{V}^\dag(I- W^{\ddagger}_{\bar{\I}}(W_{\bar{\I}}^T W^{\ddagger}_{\bar{\I}})^{-1}W_{\bar{\I}}^T )W^{\ddagger}_{\bar{\I}^c}\|_{\infty,2}\nonumber\\
=& \|\P_{\bar{U}^\perp}\P_{\hat{V}}(I+\sum_{i=1}^\infty (\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}})^i)\Phi(\hat{V}\hat{V}^\dag W^{\ddagger}_{\bar{\I}^c}- \nonumber\\& \hat{V}\hat{V}^\dag W^{\ddagger}_{\bar{\I}}(W_{\bar{\I}}^T W^{\ddagger}_{\bar{\I}})^{-1}W_{\bar{\I}}^T )W^{\ddagger}_{\bar{\I}^c}\|_{\infty,2}\nonumber\\
\leq & \|I+\sum_{i=1}^\infty (\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}})^i\|\|\Phi\|\Big(\|\hat{V}\| \|\hat{V} W^{\ddagger}_{\bar{\I}^c}\|_{\infty,2}\nonumber\\
&+ \| \hat{V}\|\hat{V}^\dag W^{\ddagger}_{\bar{\I}}\|\|(W_{\bar{\I}}^\dag W_{\bar{\I}})^{-1}\|\|W_{\bar{\I}}^T W^{\ddagger}_{\bar{\I}^c}\|_{\infty,2}\Big)\nonumber\\
\leq & \frac{\|\Phi\|(\epsilon + \lambda \sqrt{k}\sigma_{k} \sqrt{k}\mu)}{1-\psi}\leq \frac{\epsilon + \lambda k \sigma_{k} \mu}{2-\psi} \leq \frac{\epsilon + \lambda k\sigma_{k}\mu}{2-\tilde{\psi}}, \nonumber
\end{align}
where the second to last inequality follows from (e) to (j) in step 2.
\begin{align}
&\|(QW^{\ddagger})_{\bar{\I}^c}\|_{\infty,2}\nonumber\\
=&\|(\bar{U}\hat{V}W^{\ddagger}+\Phi W^{\ddagger} -\P_{\bar{U}}(\Phi)W^{\ddagger}-\Delta_2W^{\ddagger})_{\bar{\I}^c}\|_{\infty,2} \nonumber\\
=& \|\bar{U}\hat{V} W^{\ddagger}_{\bar{\I}^c}+\P_{\bar{U}^\perp}(\Phi)W^{\ddagger}_{\bar{\I}^c}-(\Delta_2W^{\ddagger})_{\bar{\I}^c}\|_{\infty,2}\nonumber\\
\leq &\|\bar{U}\hat{V} W^{\ddagger}_{\bar{\I}^c}\|_{\infty,2}+ \| (I-\bar{U}\bar{U})^\dag \lambda \hat{H} (W_{\bar{\I}}^T W^{\ddagger}_{\bar{\I}})^{-1}W_{\bar{\I}}^T W^{\ddagger}_{\bar{\I}^c} \|_{\infty,2}\nonumber\\
& +\|(\Delta_2W^{\ddagger})_{\bar{\I}^c}\|_{\infty,2}\nonumber\\
\leq &\|\bar{U}\| \|\hat{V} W^{\ddagger}_{\bar{\I}^c}\|_{\infty,2} + \nonumber\\
& \|(I- \bar{U}\bar{U})^\dag \|\|\lambda \hat{H}\|\|(W_{\bar{\I}}^\dag W_{\bar{\I}})^{-1}\|\| W_{\bar{\I}}^T W^{\ddagger}_{\bar{\I}^c} \|_{\infty,2} +\nonumber\\
& \|(\Delta_2W^{\ddagger})_{\bar{\I}^c}\|_{\infty,2}\nonumber\\
\leq & \epsilon+ \lambda \sqrt{k} \sigma_k \sqrt{k}\mu +\nonumber \\
& \frac{\lambda \sigma_{k} \sqrt{k+(k^2-k)\mu} (\epsilon + \sigma_{k} \mu\sqrt{k+(k^2-k)\mu})}{1-\psi}\nonumber\\
\leq & (1+\frac{1}{2-\tilde{\psi}})(\epsilon+ \lambda k \sigma_k \mu) , \nonumber\\
\leq & (1+\frac{1}{2-\tilde{\psi}})(\epsilon+ \lambda \tilde{k} \sigma_{\tilde{k}} \mu), \label{eqn:stp4}\\
\leq & \lambda\nonumber,
\end{align}
where the last inequality follows from $\lambda \geq \lambda_{\min,\tilde{k}}$. Then (d) of (\ref{eqn:cond}) holds.
\end{proof}
\subsection{Proof of Lemma \ref{lem:cond1}}
\begin{proof}
We define
\begin{equation*}
\tilde{C} = \bar{C}+\P_{\bar{\I}}\P_{\bar{U}}(C^*-\bar{C}) \text{ and } \tilde{L} = \bar{L}-\P_{\bar{\I}}\P_{\bar{U}}(C^*-\bar{C})W^T.
\end{equation*}
Note that $\P_{\bar{U}}(\tilde{L})=\tilde{L}$, $\P_{\bar{\I}}(\tilde{C})=\tilde{C}$ and $\bar{L}+\bar{C}W^T=\tilde{L}+\tilde{C}W^T$. We further define $N_L=L^*-\bar{L}$, $N_C=C^*-\bar{C}$, and $N^+_C = C^*-\tilde{C}$. Note that $\P_{\bar{\I}^c}(N_C^+)=\P_{\bar{\I}^c}(N_C)$ from the definition of $N^+_C$. Let $E=N_L+N_CW^T$. We have
\begin{align}\label{eqn:E}
&\|E\|_F = \|L^*+C^*W^T-(\bar{L}+\bar{C}W^T)\|_F \nonumber\\
\le & \|L^*+C^*W^T-M\|_F+\|N\|_F \le 2\eta,
\end{align}
where the last inequality holds since ($L^*$, $C^*$) is the solution to (\ref{eqn:opt}) and $\|N\|_F \le \eta$. Let $G$ be such that $\|G\|=1$, $\left\langle G, \P_{T^{*{\perp}}}(\Delta W^T) \right\rangle = \|\P_{T^{*{\perp}}}(\Delta W^T)\|_{*}$ and $\P_{T^*}(G)=0$. Let $F$ be such that $F_i=\Delta_i/\|\Delta_i\|_2$ if $i \in \bar{\I}$ and $\Delta_i \neq 0$, and $F_i=0$ otherwise. Then
\begin{equation}\label{eqn:worse1}
\begin{aligned}
&\|\bar{L}\|_*+\lambda\|\bar{C}\|_{1,2} \stackrel{(\rm{m})}{\ge} \|L^*\|_*+\lambda\|C^*\|_{1,2} \\
\stackrel{(\rm{n})}{\ge} & \|\bar{L}\|_*+\lambda\|\bar{C}\|_{1,2}+\langle \P_{\bar{T}}(Q)+G,N_L\rangle+\lambda\langle \P_{\bar{\I}}(QW^{\ddagger})/\lambda \\
&+F,N_C\rangle \\
= & \|\bar{L}\|_*+\lambda\|\bar{C}\|_{1,2}+\|\P_{\bar{T}^{\perp}}(N_L)\|_*+\langle \P_{\bar{T}}(Q),N_L \rangle\\
&+\lambda\|\P_{\bar{\I}^c}(N_C)\|_{1,2}+\langle \P_{\bar{\I}}(QW^{\ddagger}),N_C \rangle \\
= & \|\bar{L}\|_*+\lambda\|\bar{C}\|_{1,2}+\|\P_{\bar{T}^{\perp}}(N_L)\|_*+\lambda\|\P_{\bar{\I}^c}(N_C)\|_{1,2} \\
&-\langle \P_{\bar{T}^{\perp}}(Q),N_L \rangle-\langle \P_{\bar{\I}^c}(QW^{\ddagger}),N_C \rangle+ \langle Q,N_L+N_CW^T \rangle \\
\ge & \|\bar{L}\|_*+\lambda\|\bar{C}\|_{1,2}+(1-\|\P_{\bar{T}^{\perp}}(Q)\|)\|\P_{\bar{T}^{\perp}}(N_L)\|_* \\
&+(\lambda-\|\P_{\bar{\I}^c}(QW^{\ddagger})\|_{\infty,2})\|\P_{\bar{\I}^c}(N_C)\|_{1,2}+\langle Q,E \rangle \\
\ge & \|\bar{L}\|_*+\lambda\|\bar{C}\|_{1,2}+\frac{1}{2}\|\P_{\bar{T}^{\perp}}(N_L)\|_*+\frac{\lambda}{2}\|\P_{\bar{\I}^c}(N_C)\|_{1,2}\\
&-2\eta\|Q\|_F,
\end{aligned}
\end{equation}
where (m) holds because of the optimality of ($L^*$, $C^*$) and (n) holds because of the convexity of the objective function of (\ref{eqn:opt}). We can see that the last inequality of (\ref{eqn:worse1}) follows from (b) and (d) of (\ref{eqn:cond1}). Then we have
\begin{equation}\label{eqn:inequa1}
\frac{1}{2}\|\P_{\bar{T}^{\perp}}(N_L)\|_*+\frac{\lambda}{2}\|\P_{\bar{\I}^c}(N_C)\|_{1,2}-2\eta\|Q\|_F \le 0.
\end{equation}
Note that
\begin{align}
& \|Q\|_F = \|\P_{\bar{T}}(Q)+\P_{\bar{T}^{\perp}}(Q)\|_F \nonumber \\
= & \sqrt{\|\P_{\bar{T}}(Q)\|^2_F+\|\P_{\bar{T}^{\perp}}(Q)\|^2_F} \nonumber\\
= & \sqrt{\|\bar{U} \bar{V}^{\dagger}\|^2_F+\|\P_{\bar{T}^{\perp}}(Q)\|^2_F} \stackrel{(\rm{o})}{\leq} \frac{1}{2}\sqrt{\min(t,p)+3r}, \label{eqn:QF}
\end{align}
where the last equality follows from (a) of (\ref{eqn:cond1}).
The inequality (o) holds from $\|\bar{U} \bar{V}^{\dagger}\|_F = \sqrt{\rm{trace}(\bar{V}\bar{U}^{\dagger}\bar{U}\bar{V}^{\dagger})}=\sqrt{r}$,
and
\begin{equation}\nonumber
\|\P_{\bar{T}^{\perp}}(Q)\|_F \le \textrm{rank}(\P_{\bar{T}^{\perp}}(Q))\cdot\|\P_{\bar{T}^{\perp}}(Q)\|\leq \frac{\sqrt{\min(t,p)-r}}{2}.
\end{equation}
Since $\theta = \min(t,p)$, combining (\ref{eqn:inequa1}) and (\ref{eqn:QF}), we have
\begin{equation}\label{eqn:PNL}
\|\P_{\bar{T}^{\perp}}(N_L)\|_F \le \|\P_{\bar{T}^{\perp}}(N_L)\|_* \le 2\eta\sqrt{\theta+3r},
\end{equation}
\begin{equation}\label{eqn:PNC}
\|\P_{\bar{\I}^c}(N_C)\|_F \le \|\P_{\bar{\I}^c}(N_C)\|_{1,2} \le \frac{2}{\lambda}\eta\sqrt{\theta+3r}.
\end{equation}
From the definition of $\P_{W_{\bar{\I}}}$ in (\ref{eqn:PWI}), one can check that
\begin{equation}\label{eqn:PWIWI}
\P_{W_{\bar{\I}}}(\P_{\bar{\I}}(W)^T)= \P_{\bar{\I}}(W)^T.
\end{equation}
Then we have
\begin{equation}\label{eqmain}
\begin{aligned}
& \P_{\bar{\I}}(N_C^+)W^T = \P_{\bar{\I}}(N_C^+)\P_{\bar{\I}}(W)^T\\
= & \P_{\bar{\I}}(N_C^+)\P_{W_{\bar{\I}}}(\P_{\bar{\I}}(W)^T)=\P_{\bar{\I}}(N_C^+)\P_{W_{\bar{\I}}}(W^T)\\
= & \P_{W_{\bar{\I}}}(N_C^+W^T-\P_{\bar{\I}^c}(N_C^+)W^T)\\
\stackrel{(\rm{p})}{=} & \P_{W_{\bar{\I}}}(E-\P_{\bar{T}^{\perp}}(N_L)-\P_{\bar{T}}(N_L)-\P_{\bar{\I}}\P_{\bar{U}}(N_C)W^T \\
& -\P_{\bar{\I}^c}(N_C^+)W^T)\\
\stackrel{(\rm{q})}{=} & \P_{W_{\bar{\I}}}(E-\P_{\bar{T}^{\perp}}(N_L)-\P_{\bar{T}}(E)+\P_{\bar{T}}(N_CW^T)\\
& -\P_{\bar{\I}}\P_{\bar{U}}(N_C)W^T-\P_{\bar{\I}^c}(N_C^+)W^T) \\
= & \P_{W_{\bar{\I}}}(\P_{\bar{T}^{\perp}}(E)-\P_{\bar{T}^{\perp}}(N_L)-\P_{\bar{\I}^c}(N_C)W^T\\
& +\P_{\bar{T}}(\P_{\bar{\I}}(N_C)W^T)+\P_{\bar{T}}(\P_{\bar{\I}^c}(N_C)W^T)\\ &-\P_{\bar{\I}}\P_{\bar{U}}(N_C)W^T)\\
\stackrel{(\rm{r})}{=} &
\P_{W_{\bar{\I}}}(\P_{\bar{T}^{\perp}}(E)-\P_{\bar{T}^{\perp}}(N_L)-\P_{\bar{\I}^c}(N_C)W^T+\\
& \P_{\bar{T}}(\P_{\bar{\I}^c}(N_C)W^T)+\P_{\bar{U}}(\P_{\bar{\I}}(N_C)W^T)+\\
&\P_{\bar{V}}(\P_{\bar{\I}}(N_C)\P_{\bar{\I}}(W)^T)-\P_{\bar{U}}\P_{\bar{V}}(\P_{\bar{\I}}(N_C)W^T)\\
&-\P_{\bar{\I}}\P_{\bar{U}}(N_C)W^T)\\
\stackrel{(\rm{s})}{=} &
\P_{W_{\bar{\I}}}(\P_{\bar{T}^{\perp}}(E)-\P_{\bar{T}^{\perp}}(N_L)-\P_{\bar{\I}^c}(N_C)W^T+\\
& \P_{\bar{T}}(\P_{\bar{\I}^c}(N_C)W^T)+\P_{\bar{V}}(N_C\P_{\bar{\I}}(W)^T)-\P_{\bar{V}}(\P_{\bar{\I}^c}(N_C)\\
& \P_{\bar{\I}}(W)^T) - \P_{\bar{U}}\P_{\bar{V}}(\P_{\bar{\I}}(N_C)W^T))\\
\stackrel{(\rm{t})}{=} & \P_{W_{\bar{\I}}}(\P_{\bar{T}^{\perp}}(E)-\P_{\bar{T}^{\perp}}(N_L)-\P_{\bar{\I}^c}(N_C)W^T+\\
& \P_{\bar{T}}(\P_{\bar{\I}^c}(N_C)W^T)+\P_{\bar{V}}(N_C^+\P_{\bar{\I}}(W)^T)).
\end{aligned}
\end{equation}
where (p) and (q) follow from the definition $E=N_L+N_CW^T$ and $N^+_C = N_C - \P_{\bar{\I}}\P_{\bar{U}}(N_C)$. (r) follows the definition of $\P_{\bar{T}}$. (s) holds because $\P_{\bar{U}}(\P_{\bar{\I}}(N_C)W^T)=\P_{\bar{\I}}\P_{\bar{U}}(N_C)W^T$. (t) holds because of the equality (\ref{eqn:sec2}) shown as follows:
\begin{equation}\label{eqn:sec2}
\begin{aligned}
& \P_{\bar{V}}(N_C\P_{\bar{\I}}(W)^T)-\P_{\bar{U}}\P_{\bar{V}}(\P_{\bar{\I}}(N_C)\P_{\bar{\I}}(W)^T) \\
= & \P_{\bar{V}}(N_C\P_{\bar{\I}}(W)^T-\P_{\bar{\I}}\P_{\bar{U}}(N_C)\P_{\bar{\I}}(W)^T) \\
= & \P_{\bar{V}}(N_C^+\P_{\bar{\I}}(W)^T)
\end{aligned}
\end{equation}
Note that
\begin{align}
& \|\P_{W_{\bar{\I}}}\P_{\bar{V}}(N_C^+\P_{\bar{\I}}(W)^T)\|_F\nonumber\\
= & \|\P_{W_{\bar{\I}}}\P_{\bar{V}}(N_C^+\P_{W_{\bar{\I}}}\P_{\bar{\I}}(W)^T)\|_F\nonumber\\
= & \|N_C^+\P_{\bar{\I}}(W)^TW^{\ddagger}_{\bar{\I}}(W_{\bar{\I}}^{T}W^{\ddagger}_{\bar{\I}})^{-1}W_{\bar{\I}}^{T}\bar{V}\bar{V}^{\dagger}W^{\ddagger}_{\bar{\I}}(W_{\bar{\I}}^{T}W^{\ddagger}_{\bar{\I}})^{-1}W_{\bar{\I}}^{T}\|_F\nonumber\\
\stackrel{(\rm{u})}{\leq} & \|N_C^+\P_{\bar{\I}}(W)^T\|_F\| \bar{V}^{\dagger}W^{\ddagger}_{\bar{\I}}(W_{\bar{\I}}^{T}W^{\ddagger}_{\bar{\I}})^{-1}W_{\bar{\I}}^{T}\bar{V}\|\nonumber\\
= & \|N_C^+\P_{\bar{\I}}(W)^T\|_F\|\bar{V}\bar{V}^{\dagger}W^{\ddagger}_{\bar{\I}}(W_{\bar{\I}}^{T}W^{\ddagger}_{\bar{\I}})^{-1}W_{\bar{\I}}^{T}\bar{V}\bar{V}^{\dagger}\|\nonumber\\
= & \psi\|\P_{\bar{\I}}(N_C^+)W^T\|_F \leq \tilde{\psi}\|\P_{\bar{\I}}(N_C^+)W^T\|_F, \nonumber
\end{align}
where the first equality holds from (\ref{eqn:PWIWI}), and (u) holds because $\|AB\|_F \leq \|A\|_F\|B\|$ and $\|A^\dag A\|=\|AA^\dag\|$ for matrices $A$ and $B$. From (\ref{eqmain}), we have
\begin{align}
& \|\P_{\bar{\I}}(N_C^+)W^T\|_F \nonumber \\
\le & (\|\P_{\bar{T}^{\perp}}(E)\|_F+\|\P_{\bar{T}^{\perp}}(N_L)\|_F+\|\P_{\bar{T}^{\perp}}(\P_{\bar{\I}^c}(N_C)W^T)\|_F) \nonumber\\
& \|W^{\ddagger}_{\bar{\I}}(W_{\bar{\I}}^{T}W^{\ddagger}_{\bar{\I}})^{-1}W_{\bar{\I}}^{T}\|+\tilde{\psi}\|\P_{\bar{\I}}(N_C^+)W^T\|_F \nonumber\\
\le & \|E\|_F+\|\P_{\bar{T}^{\perp}}(N_L)\|_F+\|\P_{\bar{\I}^c}(N_C)\|_F\|W\| +\nonumber\\
& \tilde{\psi}\|\P_{\bar{\I}}(N_C^+)W^T\|_F, \label{eqn:PINCW}
\end{align}
where the last inequality uses the property that
$\|W^{\ddagger}_{\bar{\I}}(W_{\bar{\I}}^{T}W^{\ddagger}_{\bar{\I}})^{-1}W_{\bar{\I}}^{T}\|=1$. From similar arguments as in (\ref{eqn:WI}), we have $\|W\| \le \sqrt{1+(n-1)\mu}$. Then combining (\ref{eqn:E}), (\ref{eqn:PNL}), (\ref{eqn:PNC}), and (\ref{eqn:PINCW}), we obtain
\begin{align}\label{eqn:PNCW}
& \|\P_{\bar{\I}}(N_C^+)W^T\|_F \le (1+\frac{\lambda+\sqrt{1+(n-1)\mu}}{\lambda}\sqrt{\theta+3r})\frac{2\eta}{1-\tilde{\psi}}.
\end{align}
Furthermore,
\begin{equation}\label{eqn:ineq1}\nonumber
\begin{aligned}
& \|\P_{\bar{\I}}(N^+_C)\|_F = \|\P_{\bar{\I}}(N^+_C)W^TW^{\ddagger}_{\bar{\I}}(W^T_{\bar{\I}}W^{\ddagger}_{\bar{\I}})^{-1}\|_F\\
\le & \|\P_{\bar{\I}}(N^+_C)W^T\|_F\|W^{\ddagger}_{\bar{\I}}\|\|(W^T_{\bar{\I}}W^{\ddagger}_{\bar{\I}})^{-1}\|\\
\le & (1+\frac{\lambda+\sqrt{1+(n-1)\mu}}{\lambda}\sqrt{\theta+3r})\frac{2\eta\sigma_k\sqrt{1+(k-1)\mu}}{1-\tilde{\psi}},
\end{aligned}
\end{equation}
where the last inequality follows from (\ref{eqn:PNCW}), (\ref{eqn:WI}), and (\ref{eqn:sigmak}).
We also have
\begin{equation}\label{eqn:ineq2}\nonumber
\begin{aligned}
& \|N^+_CW^T\|_F = \|\P_{\bar{\I}^c}(N_C)W^T+\P_{\bar{\I}}(N^+_C)W^T\|_F\\
\le & \|\P_{\bar{\I}^c}(N_C)W^T\|_F+\|\P_{\bar{\I}}(N^+_C)W^T\|_F\\
\le & \|\P_{\bar{\I}^c}(N_C)\|_F\|W\|+\|\P_{\bar{\I}}(N^+_C)W^T\|_F\\
\le & (1+\frac{\lambda+(2-\tilde{\psi})\sqrt{1+(n-1)\mu}}{\lambda}\sqrt{\theta+3r})\frac{2\eta}{1-\tilde{\psi}}.
\end{aligned}
\end{equation}
Finally, we have
\begin{align}
& \|C^*-\tilde{C}\|_F = \|\P_{\bar{\I}^c}(N_C)+\P_{\bar{\I}}(N^+_C)\|_F\nonumber\\
\le & \|\P_{\bar{\I}^c}(N_C)\|_F+\|\P_{\bar{\I}}(N^+_C)\|_F\nonumber\\
\le & (1+(\frac{\lambda+\sqrt{1+(n-1)\mu}}{\lambda}+\frac{1-\tilde{\psi}}{\lambda\sigma_k\sqrt{1+(k-1)\mu}})\sqrt{\theta+3r})\nonumber\\
& \frac{2\eta\sigma_k\sqrt{1+(k-1)\mu}}{1-\tilde{\psi}},\nonumber
\end{align}
and
\begin{align}
& \|L^*-\tilde{L}\|_F
= \|L^*-\bar{L}+\tilde{C}W^T-\bar{C}W^T\|_F \nonumber\\
= & \|L^*-\bar{L}+C^*W^T-\bar{C}W^T+\tilde{C}W^T-C^*W^T\|_F\nonumber\\
= & \|E-N^+_CW^T\|_F\nonumber
\le \|E\|_F+\|N^+_CW^T\|_F\nonumber\\
\le & (2-\tilde{\psi}+\frac{\lambda+(2-\tilde{\psi})\sqrt{1+(n-1)\mu}}{\lambda}\sqrt{\theta+3r})\frac{2\eta}{1-\tilde{\psi}}.\nonumber
\end{align}
\end{proof}
\subsection{Proof of Lemma \ref{lem:certificate1}}
\begin{proof}
Since equalities (a) and (c) of (\ref{eqn:cond1}) are the same as those in (\ref{eqn:cond}) and the construction of $Q$ remains the same, then (a) and (c) of (\ref{eqn:cond1}) have been proved in step 1 and 3 of the proof of Lemma \ref{lem:certificate}. We only need to show that (b) and (d) hold when $\lambda$ belongs to [$\lambda'_{\text{min},\tilde{k}}$, $\lambda'_{\text{max},\tilde{k}}$]. From (\ref{eqn:stp2}), that is proved in the proof of Lemma \ref{lem:certificate}, and $\lambda \leq \lambda'_{\text{max},\tilde{k}}$, we have
\begin{equation}\nonumber
\|\P_{T'^{\perp}}(Q)\| \leq \frac{2-\tilde{\psi}}{1-\tilde{\psi}}\lambda \sigma_{\tilde{k}} \sqrt{\tilde{k}+(\tilde{k}^2-\tilde{k})\mu} \leq \frac{1}{2}.
\end{equation}
From (\ref{eqn:stp4}) and $\lambda \geq \lambda'_{\text{min},\tilde{k}}$, we have
\begin{equation}\nonumber
\|(QW^{\ddagger})_{\bar{\I}^c}\|_{\infty,2} \leq
(1+\frac{1}{2-\tilde{\psi}})(\epsilon+ \lambda \tilde{k} \sigma_{\tilde{k}} \mu) \leq \frac{\lambda}{2}.
\end{equation}
\end{proof}
\section{Conclusion}\label{sec:con}
We address the problem of detecting successive unobservable cyber data attacks to PMU measurements. We formulate the identification problem as a matrix decomposition problem of a low-rank matrix and a transformed column-sparse matrix. We propose a convex-optimization-based method and provide its theoretical guarantee. Although motivated by power system monitoring, our results on matrix decomposition can be applied to other scenarios. One future direction is the analysis of the detection performance when some of the measurements are lost during the communication to the central operator.
\section*{Acknowledgement}
We thank New York Power Authority for providing PMU data for the Central NY Power System. This research is supported in part by the ERC Program of NSF and DoE under the supplement to NSF Award EEC-1041877 and the CURENT Industry Partnership Program, and in part by NSF Grant 1508875, NYSERDA Grants \#36653 and \#28815.
\section{Introduction}\label{sec:intro}
\IEEEPARstart{T}{he} integration of cyber infrastructures into future smart grids greatly enhances the monitoring, dispatch, and scheduling of power systems. Such integration, however, makes the power systems more susceptible to cyber attacks. It is reported that cyber spies have penetrated U.S. electrical grid \cite{meserve07}. Researchers have also launched an experimental cyber attack that caused a generator to self-destruct \cite{HLEO09}.
State estimation \cite{AE04} is a critical component of power system monitoring. System state is estimated based on the obtained measurements across the system. Bad data can affect the state estimation and mislead the system operator. Many efforts have been devoted to develop methods that can identify bad data, see e.g., \cite{CA06,HSKF75,MG83,MS71,XWCT13}.
Cyber data attacks (firstly studied in \cite{LNR11}) can be viewed as ``the worst interacting bad data injected by an adversary''\cite{KJTT10}. Malicious intruders with system configuration information can simultaneously manipulate multiple measurements so that these attacks cannot be detected by any bad data detector. Because the removal of affected measurements would make the system unobservable, these attacks are termed as ``unobservable attacks''\footnote{The term ``unobservable'' is used in this sense throughout the paper.} in \cite{KJTT10}.
State estimation in the presence of cyber data attacks has attracted much research attention recently \cite{BRWKNO10,DS10,KJTT10,LEDEH14,LNR11,SJ13,STJ10,TKPC11}. Existing approaches include protecting a small number of key measurement units such that the intruders cannot inject unobservable attacks without hacking protected units \cite{BRWKNO10,DS10,KP11}, as well as detectors designed for attacks in the observable regime \cite{KJTT10}. The research on the detection of unobservable attacks is still limited. Refs. \cite{SJ13,LEDEH14} proposed different methods to detect unobservable attacks in Supervisory Control and Data Acquisition (SCADA) system. The method in \cite{SJ13} relies critically on the assumption that the measurements at different time instants are i.i.d. samples of random variables. This assumption might not hold when the system is under disturbance. Ref. \cite{LEDEH14} focused on the scenarios that an intruder attacks a different set of measurements at each time instant, and no theoretical analysis of the detection performance is provided in \cite{LEDEH14}.
This paper considers cyber data attacks to PMU measurements. It focuses on the case when an intruder injects unobservable data attacks to the same set of PMUs constantly. Because PMUs under attack do not provide any accurate measurement at any time instant to the operator, the attack identification in this case is very challenging and has not been addressed before. We propose a method that can identify the successive unobservable cyber data attacks and provide the theoretical guarantee even when the system is under disturbance. The intuition is that even though an intruder can constantly inject data attacks that are consistent with each other at each time instant, as long as the intruder does not know the system dynamics, one can identify the attacks by comparing time series of different PMUs and locating the PMUs that exhibit abnormal dynamics.
Because PMU measurements are synchronized and correlated, the high-dimensional PMU data matrix exhibits low-rank property \cite{CXK13,DKM12,GWGC14,WGGCFSR14}. We formulate the identification problem as a matrix decomposition problem of a low-rank matrix plus a transformed column-sparse matrix. The matrix decomposition problem has attracted much research attention recently, see e.g., \cite{CLMW11,CSPW11,RFP10,XCS12}, and have wide applications in areas like Internet monitoring \cite{LCD04,MMG13,TJ03}, medical imaging \cite{FNDRL06,GCSZ11}, and image processing \cite{BJ03}. The situation that one component is a transformed column-sparse matrix, however, has not been addressed before.
The contributions of this paper are threefold. (1) We propose the idea of exploiting spatial-temporal correlations in PMU measurements to identify unobservable data attacks. (2) We formulate the identification problem into a matrix decomposition problem and propose a computationally efficient method that does not require the modeling of power system dynamics. (3) We provide theoretical guarantees of attack detection, as well as the general matrix decomposition problem.
The rest of the paper is organized as follows. We formulate our problem and point out its connection to other applications in Section \ref{sec:model}. We describe our detection method and analyze its theoretical guarantee with both noiseless (Section \ref{sec:method}) and noisy measurements (Section \ref{sec:method1}). Section \ref{sec:simu} records our numerical experiments. We conclude the paper in Section \ref{sec:con}.
\section{Attack Identification without Noise}\label{sec:method}
\subsection{Identification method and guarantee}\label{sec:analysis}
\floatname{algorithm}{Method}
\begin{algorithm}
\begin{algorithmic}
\REQUIRE PMU measurements $M$ in $t$ instants; coefficient $\eta$; the set $\Omega$ of the locations $(i,j)$ of the observed entries.
\STATE Find ($L^*$, $C^*$), the optimum solution to the following optimization problem
\begin{equation} \label{eqn:opt1}
{\hspace{-0.2in}\min\limits_{L \in \C^{t\times p}, C\in \C^{t \times n}} \| L\|_{*} + \lambda \|C\|_{1,2} }
\end{equation}
\begin{equation}\label{eqn:opt}
\textrm{s.t.} \quad \sum_{i,j \in \Omega}|M_{ij}-L_{ij}-(CW^T)_{ij}|^2 \le \frac{|\Omega|}{tp}\eta^2
\end{equation}
\STATE Compute the SVD of $L^*=U^* \Sigma^* V^{* \dagger}$.
\STATE Find column support of $D^*=C^*W^T$, denoted by $\J^*$.
\RETURN $L^*$, $C^*$, $L^*_{\J^{*c}}$, $U^*$ and $\J^*$.
\end{algorithmic}
\caption{Unobservable cyber attack identification method}
\end{algorithm}
We first consider noiseless measurements ($\eta=0$). We assume a complete set of measurements for analysis, but our method can be extended to cases when measurements are partially lost. Moreover, although we consider attack patterns in (\ref{eqn:state}), our method can be generalized to detect combined attacks. In this case, $\bar{D}$ is generalized to
\begin{equation}\label{eqn:Dg}
\bar{D}=\bar{C}W^T+\bar{S},
\end{equation}
where a sparse matrix $\bar{S}$ represents attacks (observable and/or unobservable) that have different locations across time. Then (\ref{eqn:opt1})-(\ref{eqn:opt}) are generalized to
\begin{equation} \label{eqn:opt3}
{\hspace{-0.2in}\min\limits_{L \in \C^{t\times p}, C\in \C^{t \times n}, S \in \C^{t\times p}} \| L\|_{*} + \lambda_1 \|C\|_{1,2} } + \lambda_2 \sum_{ij} |S_{ij} |
\end{equation}
\begin{equation}\label{eqn:opt4}
\textrm{s.t.} \quad \sum_{i,j \in \Omega}|M_{ij}-L_{ij}-(CW^T)_{ij}-S_{ij}|^2 \le \frac{|\Omega|}{tp}\eta^2,
\end{equation}
with given positive constants $\lambda_1$, $\lambda_2$. We study this extension numerically in Section \ref{sec:compare}.
To formally present the theoretical result, we need the following definitions.
Given $\bar{L}=\bar{U}\bar{\Sigma}\bar{V}^\dagger$ and $W$, we define
\begin{equation}
\epsilon:=\|\bar{V}^\dag W^{\ddagger}\|_{\infty,2}, \text{ } \mu:=\max_{i \neq j}\|W_i^{\dagger} W_j\|,
\end{equation}
\begin{equation} \label{eqn:sigmak}
\text{ and } \sigma_k:=\max_{\I: |\I| \leq k}\| (W^{\dagger}_\I W_\I)^{-1}\|.
\end{equation}
Note that $\sigma_1=1$ as $W$ has unit-norm columns, and $\epsilon$ depends on the rank $r$ of $\bar{L}$, since $\|\bar{V}\|_{F}^2=r$.
Pick any constants $\tilde{\psi}$ and $c$ in $(0,1)$ such that
\begin{equation}\label{eqn:psistar}
(2-\tilde{\psi})\sqrt{\tilde{\psi}}/(1-\tilde{\psi}) \leq \sqrt{(1+c)/(1-c)}.
\end{equation}
For any integer $k$, define
\begin{equation}\label{eqn:lambdamin}
\lambda_{\min,k}=\frac{(1+(2-\tilde{\psi})^{-1})\epsilon}{1-(1+(2-\tilde{\psi})^{-1})k\sigma_k\mu}
\end{equation}
\begin{equation}\label{eqn:lambdamax}
\text{ and } \lambda_{\max,k}=\sqrt{\tilde{\psi}/(k\sigma_k)}.
\end{equation}
Our detection method is summarized in Method 1. (\ref{eqn:opt}) is a convex program and can be solved efficiently by generic solvers such as CVX\cite{CVX}. Its recovery guarantee is as follows.
\begin{theorem}\label{thm:recover}
Suppose there exists nonzero $\tilde{k}$ such that
\begin{equation}
\tilde{k} \mu \leq c, \textrm{ and } \lambda_{\min,\tilde{k}} \leq \lambda_{\max,\tilde{k}},
\end{equation}
with $c$, $\lambda_{\min,\tilde{k}}$, and $\lambda_{\max,\tilde{k}}$ defined in (\ref{eqn:psistar})-(\ref{eqn:lambdamax}).
Then as long as the column support of $\bar{C}$ has size at most $\tilde{k}$, for any $\lambda \in [\lambda_{\min,\tilde{k}}, \lambda_{\max,\tilde{k}}]$, the output of Method 1 satisfies
\begin{equation} \label{eqn:space}
U^*U^{*\dag}=\bar{U} \bar{U}^\dag,
\end{equation}
\begin{equation}
\J^*=\bar{\J}\label{eqn:index} \text{ and } L^*_{\J^{*c}}=\bar{L}_{\bar{\J}^{c}}. \nonumber
\end{equation}
\end{theorem}
Theorem \ref{thm:recover} guarantees that the affected PMUs can be correctly located and thus, the ``clean'' PMU measurements could be identified. Furthermore, the subspace spanned by the actual phasors can be recovered. Since we do not obtain any actual measurements from PMUs that are under attack, it is impossible to recover the exact measurements in the affected PMUs without further regularization. Under the conditions of Theorem \ref{thm:recover}, the recovery is also successful when the column support of $\bar{C}$ is zero. Thus, the false alarm rate is zero.
Method 1 is motivated by \cite{XCS12}. In fact, after post-multiplying $W^{\ddagger} (W^TW^{\ddagger})^{-1}$ to both sides of (\ref{eqn:M}), we have
\begin{equation} \nonumber
MW^{\ddagger} (W^TW^{\ddagger})^{-1}=\bar{L}W^{\ddagger} (W^TW^{\ddagger})^{-1} + \bar{C} + NW^{\ddagger} (W^TW^{\ddagger})^{-1}
\end{equation}
where the right-hand side is the sum of a low-rank matrix plus a column-sparse matrix and noise.
Then, the results of \cite{XCS12} can be directly applied to our problem. We do not follow this path due to two reasons. First, $MW^{\ddagger}(W^TW^{\ddagger})^{-1}$ cannot be computed if some entries of $M$ are missing, while Method 1 can be easily extended to scenarios with missing data by restricting the constraints in (\ref{eqn:opt}) to the observed measurements. Second, $(W^TW^{\ddagger})^{-1}$ does not exist when $W$ is a flat matrix, i.e., $p < n$, while Method 1 and Theorem \ref{thm:recover} can be applied to an arbitrary $W$.
\subsection{Discussion of $\lambda$ and $\tilde{k}$}\label{sec:lambda}
We remark that due to the slackness in the proof, $\lambda \in [\lambda_{\min,\tilde{k}}, \lambda_{\max,\tilde{k}}]$ in Theorem \ref{thm:recover} is sufficient but not necessary\footnote{Specially, the requirements on dual certificate in Lemma \ref{lem:certificate} are sufficient but not necessary. Furthermore, we use loose bounds in the proofs to simplify analysis. $\epsilon$, $\mu$, and $\sigma_k$ are in turn defined based on worst-case scenarios.}. There may exist $\lambda$ outside $[\lambda_{\min,\tilde{k}}, \lambda_{\max,\tilde{k}}]$ that can still lead to correct recovery. We observe from numerical experiments that recovery performance is generally much better than the bound in Theorem \ref{thm:recover}. Furthermore, when $L$ is fixed, as $\tilde{k}$ decreases, $\lambda_{\min,\tilde{k}}$ decreases, and $\lambda_{\max,\tilde{k}}$ increases. Thus, intuitively, if the number of affected PMUs decreases, a wider range of $\lambda$ is proper for Method 1. For a detailed discussion, we state the following lemma and defer its proof to the Appendix.
\begin{lemma}\label{lem:sigmak}
Suppose $k\mu < 1$, then $\sigma_k \leq (1-(k-1)\mu)^{-1}$.
\end{lemma}
Since $\sigma_k$ increases in $k$, $\sigma_1 \geq 1$, and $k \mu \leq c<1$,
together with Lemma \ref{lem:sigmak}, we know $\sigma_{\tilde{k}} = \Theta(1)$\footnote{We use the notations $g(n)\in O(h(n))$, $g(n) \in \Omega(h(n))$, or $g(n)=\Theta(h(n))$ if as $n$ goes to infinity, $g(n) \leq c \cdot h(n)$, $g(n) \geq c \cdot h(n)$ or $c_1 \cdot h(n) \leq g(n) \leq c_2 \cdot h(n)$ eventually holds for some positive constants $c$, $c_1$ and $c_2$ respectively.}. Since $\tilde{\psi}$ is a constant, one can check that $\lambda_{\min,\tilde{k}}= \Theta(\epsilon)$, and $\lambda_{\max,\tilde{k}}= \Theta(\sqrt{1/\tilde{k}})$.
Note that $\|\bar{V}^ {\dagger}\|_F^2 =r$. We assume that $\|\bar{V}^ {\dagger}\|$ is column-incoherent \cite{XCS12} with some positive constant $\rho>1$, i.e.,
\begin{equation}\label{eqn:rho}
\|\bar{V}^ {\dagger}\|_{\infty, 2} \le \sqrt{\rho r/p}.
\end{equation}
We assume the number of PMU channels incident to each bus is in the range of $[d, Cd]$ for some $d>0$ and some constant $C$. This is also the number of nonzero entries in each column of $W$ with unit column-norm. Then $p=\Theta(dn)$, and we have
\begin{align}
\epsilon = \|\bar{V}^\dag W^{\ddagger}\|_{\infty,2}
\leq & \sqrt{\frac{\rho r}{p}} \max_{i} \sum_{j} |W_{ij}| = O(\sqrt{\frac{r}{n}}). \label{eqn:ep}
\end{align}
Therefore, as long as $\tilde{k} =O(n/r)$, when $n$ is sufficiently large, $\lambda_{\min,\tilde{k}} \leq \lambda_{\max,\tilde{k}}$. $\tilde{k} \mu \leq c$ requires that $\tilde{k} =O(1/\mu)$. Note that $\mu= \Theta(\frac{1}{d})$. Thus, if both $\tilde{k} =O(n/r)$ and $\tilde{k}=O(d)$ hold, then a proper $\lambda$ exists, and Theorem \ref{thm:recover} holds.
In the case that $d= \Theta(n)$, $\tilde{k}$ could be $\Theta(n/r)$. If $r$ is a constant, our method succeeds even when a constant fraction of bus voltages are corrupted. Also consider the case that $\tilde{k}=1$. We pick $\tilde{\psi}$ and $c$ in (\ref{eqn:psistar}) arbitrarily close to one, then $\lambda=1$ is a proper choice (see Fig. \ref{fig:case2} for results on actual PMU data) provided that $\epsilon +\mu \leq 0.5$. Since $\epsilon$ scales as $1/\sqrt{n}$ and $\mu$ scales as $1/d$, the condition will be met in large systems that are tightly connected.
Intuitively, $\mu$ is small if the bus degree is high, and the line impedances are in the same range.
\begin{figure}[h]
\begin{center}
\includegraphics[height=0.4\linewidth]{Mesh_ring.eps}
\caption{$n$-bus ring network} \label{fig:mesh}
\end{center}
\end{figure}
We next use an example to illustrate the existence of proper $\lambda$. Consider an $n$-bus ($n$ is even) ring network in Fig.~\ref{fig:mesh}. Each odd-numbered bus is connected to all even-numbered buses. There is no connection among odd buses and no connection among even-numbered buses. A PMU is installed on each odd bus and measures the corresponding voltage phasor and all incident line current phasors. For the simplicity of analysis, we assume $Z^{ij}=1$ and $Y^{ij}=0$ in this ring network. $W$ is a $(\frac{n^2}{4}+\frac{n}{2}) \times n$ matrix with unit norm columns. Specifically, for every integer $k$,
$W_{ij} =
\begin{cases}
\sqrt{2/(n+2)}, & \mbox{if }i\in \mathcal{I}_{k1}\mbox{ and }j=2k-1 \\
-\sqrt{2/n}, & \mbox{if }i\in \mathcal{I}_{k2}\mbox{ and }j=2k \\
0, & \mbox{otherwise}
\end{cases}$,
\noindent where
\begin{equation*}
\mathcal{I}_{k1}:=\left\{ k+(k-1)\frac{n}{2}+k' ~|~ \textrm{interger } k'=0,1,2,...,\frac{n}{2} \right\},
\end{equation*}
\begin{equation*}
\mathcal{I}_{k2}:=\left\{ k+1+(\frac{n}{2}+1)k' ~|~ \textrm{interger }k'=0,1,2,...,\frac{n}{2}-1 \right\}.
\end{equation*}
Note that $|\I_{k1}|=\frac{n}{2}+1$, $|\I_{k2}|=\frac{n}{2}$ for all $k$. Here, $\mu=2/\sqrt{n^2+2n}$. Then we have
\begin{equation}\label{eqn:column}
(V^{\dagger}W^{\ddagger})_j =
\begin{cases}
\sqrt{2/(n+2)}\sum_{i\in \mathcal{I}_{k1}}(V^{\dagger})_i, & \mbox{if }j=2k-1 \\
-\sqrt{2/n}\sum_{i\in \mathcal{I}_{k2}}(V^{\dagger})_i, & \mbox{if }j=2k
\end{cases},
\end{equation}
where $V \in \C^{(\frac{n^2}{4}+\frac{n}{2}) \times r}$ contains the right singular vectors of the rank-$r$ measurement matrix $\bar{L} \in \C^{t \times (\frac{n^2}{4}+\frac{n}{2})}$. If $\|\bar{V}^ {\dagger}\|$ is column-incoherent \cite{XCS12} with some positive constant $\rho$, then
\begin{align}
\epsilon = \|\bar{V}^\dag W^{\ddagger}\|_{\infty,2} \leq & \max \big(\sqrt{\frac{2}{n+2}} |\I_{k1}|, \sqrt{\frac{2}{n}} |\I_{k2}|\big) \cdot\|\bar{V}^ {\dagger}\|_{\infty, 2} \nonumber \\
\leq & \sqrt{\frac{n+2}{2} }\cdot \sqrt{\frac{\rho r}{(\frac{n}{2}+1)\frac{n}{2}}} \leq \sqrt{\frac{2\rho r}{n}}, \label{eqn:ep1}
\end{align}
where the first inequality follows from (\ref{eqn:column}), and the second inequality follows from (\ref{eqn:rho}).
To find $\lambda$, we pick $c=1/4$ and $\tilde{\psi}=1/8$. We choose $\tilde{k} = \frac{n}{48\rho r}$. One can check that (\ref{eqn:psistar}) follows. Then
\begin{equation}
\tilde{k} \mu = \frac{n}{48\rho r} \times \frac{2}{\sqrt{n^2+2n}} \leq \frac{1}{24 \rho r} \leq \frac{1}{24} \leq \frac{1}{4}=c,
\end{equation}
where the last inequality follows since $\rho>1$ and $r \geq 1$.
Then from Lemma \ref{lem:sigmak}, we have
\begin{equation}
\sigma_{\tilde{k}} \leq (1-(\tilde{k}-1)\mu)^{-1} \leq (1-\tilde{k} \mu)^{-1} \leq 24/23.
\end{equation}
From (\ref{eqn:lambdamin}) and (\ref{eqn:lambdamax}),
\begin{equation}
\lambda_{\min,\tilde{k}}\leq \frac{(1+(2-\tilde{\psi})^{-1})\epsilon}{1-(1+(2-\tilde{\psi})^{-1})\tilde{k}\mu \sigma_{\tilde{k}}}
\leq \frac{23\epsilon}{14} \leq \frac{23}{14}\sqrt{\frac{2\rho r}{n}} .
\end{equation}
\begin{equation}
\lambda_{\max,\tilde{k}}= \sqrt{\frac{1/8}{\frac{n}{48\rho r} \sigma_{\tilde{k}}}} \geq \frac{1}{2} \sqrt{\frac{23\rho r}{n}}.
\end{equation}
Since $\frac{23}{14}\sqrt{\frac{2\rho r}{n}} <\frac{1}{2}\sqrt{\frac{23\rho r}{n}}$, then $\lambda_{\min,\tilde{k}}<\lambda_{\max,\tilde{k}}$. Then there exists $\lambda$ such that Method 1 correctly identifies the corruptions in up to $\tilde{k} = \frac{n}{48\rho r}$ bus voltages. In fact, any $\lambda \in [\frac{23}{14}\sqrt{\frac{2\rho r}{n}},\frac{1}{2}\sqrt{\frac{23\rho r}{n}}]$ suffices. Note that for a constant $r$, $\tilde{k}$ is linear in $n$, the total number of buses.
\subsection{Proof sketch of Theorem \ref{thm:recover}}
The proof of Theorem \ref{thm:recover} follows the same line as the proof of Theorem 1 in \cite{XCS12}. With the additional projection matrix $W$, our proof is more involved than the one in \cite{XCS12}.
Like \cite{XCS12}, we design the following Oracle Problem (\ref{eqn:oracle}) by adding explicit constraints that the solution pair should have the correct column space of $\bar{L}$ and the correct column support of $\bar{C}$. The major step is to show that an optimal solution ($L^*$,$C^*$) to (\ref{eqn:opt}) is also an solution to the Oracle problem (\ref{eqn:oracle}). Note that Oracle problem is only designed for analysis, since $\bar{U}$ and $\bar{\I}$ are unknown to the operator.
\begin{equation}\label{eqn:oracle}
\begin{tabular}{p{0.9in}p{0.1 in}p{1.6 in}}
Oracle Problem & $\min\limits_{L, C}$
& $\| L\|_{*} + \lambda \|C\|_{1,2}$ \\
& s.t. & $M = L+ C W^T $\\
&& $\P_{\bar{U}}(L)=L$, $\P_{\bar{\I}}(C)=C$.
\end{tabular}
\end{equation}
Let $(L', C')$ be an optimal solution to the Oracle problem (\ref{eqn:oracle}). We define $\P_{T'} := \P_{U'} +\P_{V'}-\P_{U'}\P_{V'}$, where the SVD of $L'= U' \Sigma' V'^{\dag}$. Define
\begin{align}\nonumber
\G(C'):=\{ H \in \C^{t \times k} ~|~ \forall i\in \I' : H_i= C_i'/\|C_i'\| ; \\ \forall i \in \bar{\I} \cap (\I')^c: \|H_i\|_2 \leq 1\},\nonumber
\end{align}
where $\I'$ is the column support of $C'$. We have
\begin{lemma}[Lemma 4 and Lemma 5 of \cite{XCS12}]
\begin{equation} \nonumber
U'U'^\dag=\bar{U}\bar{U}^\dag.
\end{equation}
There exists an orthonormal matrix $\hat{V} \in \C ^{t \times p}$ such that
\begin{equation}\label{eqn:UU'}
U' V'^\dag =\bar{U} \hat{V}^\dag.
\end{equation}
Also, we have
\begin{equation}\nonumber
\P_{T'}:= \P_{U'}+\P_{V'}-\P_{U'}\P_{V'}=\P_{\bar{U}}+\P_{\hat{V}}-\P_{\bar{U}}\P_{\hat{V}}.
\end{equation}
\end{lemma}
The following lemma establishes that the solution to the Oracle problem (\ref{eqn:oracle}) is also a solution to (\ref{eqn:opt}),
\begin{lemma}\label{lem:cond}
An optimal solution $(L', C')$ to (\ref{eqn:oracle}) is an optimal solution to (\ref{eqn:opt}) if there exists $Q \in \C^{t\times p}$ that satisfies
\begin{equation}\label{eqn:cond}
\begin{aligned}
&(a) \P_{T'} (Q) = U' V'^{\dagger},
\quad\quad (b) \| \P_{T'^{\perp}} (Q)\| \leq 1 ,\\
&(c) (QW^{\ddagger})_{\bar{\I}}/\lambda \in \G(C'),
\quad \textrm{and }(d) \|(QW^{\ddagger})_{\bar{\I}^c} \|_{\infty,2} \leq \lambda.
\end{aligned}
\end{equation}
If both (b) and (d) are strict, and $\P_{\bar{\J}} \cap \P_{V'}=\{0\}$, then any optimal solution $(L^*,C^*)$ to (\ref{eqn:opt}) satisfies $\P_{\bar{U}}(L^*)=L^*$, $\P_{\bar{\I}}(C^*)=C^*$.
\end{lemma}
The major technical step is to construct $Q$, called the \textit{dual certificate}, that satisfies (\ref{eqn:cond}). Our construction method is as follows. Pick $\hat{H} \in \G(C')$ that satisfies
\begin{equation}\label{eqn:hath}
\hat{V}^\dag W^{\ddagger}_{\bar{\I}} = \lambda\bar{U}^\dag \hat{H}.
\end{equation}
Define
\begin{equation}\label{eqn:phi}
\Phi := \lambda \hat{H} (W_{\bar{\I}}^{T}W^{\ddagger}_{\bar{\I}})^{-1}W_{\bar{\I}}^{T}, \text{ } \Delta_1:=\P_{\bar{U}}(\Phi),
\end{equation}
\begin{equation} \label{eqn:delta2}
\Delta_2:=\P_{\bar{U}^\perp}(I-\P_{W_{\bar{\I}}})\P_{\hat{V}}(I+\sum_{i=1}^\infty (\P_{\hat{V}}\P_{W_{\bar{\I}}}\P_{\hat{V}})^i)\P_{\hat{V}}(\Phi),
\end{equation}
\begin{equation} \label{eqn:PWI}
\text{ where } \P_{W_{\bar{\I}}}(X):= X W^{\ddagger}_{\bar{\I}}(W_{\bar{\I}}^{T}W^{\ddagger}_{\bar{\I}})^{-1}W_{\bar{\I}}^{T}.
\end{equation}
\begin{equation}\label{eqn:Q}
\quad \quad Q:= \bar{U} \hat{V}^\dag +\Phi-\Delta_1-\Delta_2.
\end{equation}
We show that $Q$ in (\ref{eqn:Q}) is well defined in Appendix-B. Lemma \ref{lem:certificate} shows that $Q$ in (\ref{eqn:Q}) is the desired dual certificate.
\begin{lemma}\label{lem:certificate}
Suppose there exists nonzero $\tilde{k}$ such that $\tilde{k} \mu \leq c$ for $c$ in (\ref{eqn:psistar}), and $\lambda_{\min,\tilde{k}} \leq \lambda_{\max,\tilde{k}}$ with $\lambda_{\min,\tilde{k}}$ and $\lambda_{\max,\tilde{k}}$ defined in (\ref{eqn:lambdamin}) and (\ref{eqn:lambdamax}). Then as long as the column support of $\bar{C}$ has size at most $\tilde{k}$, for any $\lambda \in [\lambda_{\min,\tilde{k}}, \lambda_{\max,\tilde{k}}]$, $Q$ defined in (\ref{eqn:Q}) satisfies (\ref{eqn:cond}).
\end{lemma}
Theorem \ref{thm:recover} follows when we combine Lemmas \ref{lem:cond} and \ref{lem:certificate}. Please refer to the Appendix for the proofs.
\section{Problem Formulation and Related Work} \label{sec:model}
\subsection{Low-rankness of PMU measurements}\label{sec:pmu}
Consider a $n$-bus power grid with PMUs installed on some buses. Let $p$ denote the total number of PMU channels that measure bus voltage and line current phasors\footnote{In three phase AC systems, a phasor is defined as a complex number that represents both the magnitude and phase angle of the sinusoidal waveforms.}. Phasors are expressed in Cartesian coordinates throughout the paper. Matrix $M \in \C^{t\times p}$ contains the collected phasor measurements in $t$ synchronized time instants. $\bar{\J} \in \lsem p \rsem$ denotes the set of PMU channels that are under data attacks. The observed measurement matrix can be presented as
\begin{equation}\label{eqn:M}
M=\bar{L}+\bar{D}+N,
\end{equation}
where $\bar{L} \in \C^{t\times p}$ represents the actual phasors without data attacks, $\bar{D}\in \C^{t\times p}$ represents the additive errors introduced by an intruder, and $N$ represents the measurement noise.
High-dimensional PMU data matrices exhibit low-rank property \cite{CXK13,DKM12,GWGC14,WGGCFSR14}. We analyzed actual PMU data from six multi-channel PMUs deployed in the Central New York (NY) Power System (Fig.~\ref{fig:NY}). Six PMUs measure twenty-three voltage and current phasors, and the data rate is thirty samples per second per channel.
Fig.~\ref{fig:current} shows the current magnitudes of PMU data in twenty seconds. An event occurs around $2.5$s. The obtained data are collected into a $600 \times 23$ matrix. Fig.~\ref{fig:svd} plots the singular values of the
matrix with the ten largest ones being 832.8, 194.8, 35.1, 18.1, 4.3, 2.5, 2.1, 1.3, 1.2, 0.5. Therefore, we can approximate the $600 \times 23$ matrix by a low-rank matrix with little approximation error.
\begin{figure}[h]
\begin{center}
\includegraphics[height=0.43 \linewidth]{CNYPS.eps}
\caption{PMUs in the Central NY Power System. (Circles and lines represent buses and transmission lines. A PMU measures the voltage phasor and the incident current phasors of the bus where it is located.)} \label{fig:NY}
\end{center}
\end{figure}
\begin{figure}[h]
\begin{center}
\psfrag{Current Magnitude}[][][0.7]{Current magnitude}
\psfrag{Time(s)}[][][0.7]{Time(s)}
\vspace{-5mm}
\includegraphics[height=0.28\linewidth]{Mag_and_angle.eps}
\caption{Visualization of Partial PMU data (Magnitude of nine current phasors)} \label{fig:current}\vspace{-5mm}
\end{center}
\end{figure}
\begin{figure}[t]
\begin{center}
\psfrag{Singular value index}[][][0.7]{Singular value index}
\psfrag{Singular values}[][][0.7]{Singular values}
\includegraphics[height=0.28\linewidth]{Low_Rank_new.eps}
\caption{Singular values of PMU data matrix in decreasing order} \label{fig:svd}
\end{center}
\end{figure}
The Singular Value Decomposition (SVD) of $\bar{L}$ is
\begin{equation}
\bar{L}=\bar{U}\bar{\Sigma}\bar{V}^ {\dagger},
\end{equation}
where $\bar{U}\in \C^{t\times r}$, $\bar{\Sigma}\in \C^{r\times r}$, $\bar{V}\in \C^{p \times r}$ ($r \ll t, p$). We assume throughout the paper that nonzero columns of $\bar{D}$ do not lie in the column space of $\bar{L}$ ($\bar{D} \neq \bar{U} \bar{U}^\dag \bar{D}$). It is a legitimate assumption when the intruders do not have full information about the system dynamics.
The notations are summarized in Table \ref{tbl:notation}. Matrix $A$ is \textit{column-sparse} if it contains a small fraction of non-zero columns. We call the set of indices of nonzero columns the \textit{column support} of $A$.
\begin{table}[h]
\caption{Notations}
\begin{tabular}{|p{0.55in} p{2.71in}|}
\hline $A_i$, $A_{i,:}$ & the $i$th column and the $i$th row of matrix $A$, respectively. \\
\hline $A_{\I}$ & the submatrix of $A$ with column indices in set $\I$. \\
\hline $A^{\ddagger}$, $A^{\dagger}$ & the conjugate and conjugate transpose matrix of $A$.\\
\hline $\P_{\I}(A)$ & matrix obtained from $A$ by setting $A_i$ to zero for all $i \notin \I$. \\
\hline $A \in \P_{\I}$ & if and only if $\P_{\I}(A)=A$.\\
\hline $\|A\|$, $\|A\|_{F}$ & the spectral and Frobenius norm of $A$, respectively. \\
\hline $\|A\|_{*}$ & the nuclear norm of $A$, which is the sum of singular values. \\
\hline $\|A\|_{1,2}$ & the sum of $\ell_2$ norms of the columns of $A$. \\
\hline $\|A\|_{\infty, 2}$ & the largest $\ell_2$ norm of the columns. \\
\hline $\P_{U} (A)$ & $:= U U^\dag A$, the projection of $A$ onto the column space of $L$. \\
\hline $\P_{V} (A)$ & $:= A V V^\dag$, the projection of $A$ onto the row space. \\
\hline $\P_{T} (\cdot)$ & $:=\P_{U} (\cdot)+\P_{V} (\cdot) - \P_{U}\P_{V}(\cdot)$. \\
\hline $\P_{U^\perp} (A)$ & $:= (I-U U^\dag)A$. \\
\hline $\P_{V^\perp} (A)$ & $:= A(I- V V^\dag)$. \\
\hline $\P_{T^\perp} (A)$ & $:= \P_{U^\perp}\P_{V^\perp} (A)$.\\
\hline $A \in \P_{T}$ & if and only if $\P_{T}(A)=A$.\\
\hline $\I^c$ & the complimentary set of set $\I$. \\
\hline
\end{tabular}
\label{tbl:notation}
\end{table}
\subsection{Unobservable cyber data attacks and problem formulation}
We use bus voltage phasors as state variables, and let $X \in \C^{t\times n}$ contain the state variables at $t$ instants. We use the $\pi$ equivalent model to represent a transmission line (Fig.~\ref{fig:pi}). $Z^{ij}$ and $Y^{ij}$ denote the impedance and admittance of the transmission line between bus $i$ and bus $j$. Current $I^{ij}$ from bus $i$ to bus $j$ is related to bus voltage $V^i$ and $V^j$ by
\begin{equation}
I^{ij}=\frac{V^i- V^j}{Z^{ij}}+V^i \frac{Y^{ij}}{2}.
\end{equation}
\begin{figure}[h]
\begin{center}
\includegraphics[height=0.2\linewidth]{Pi_model.eps}
\caption{$\pi$ model of a transmission line} \label{fig:pi} \vspace{-3 mm}
\end{center}
\end{figure}
We define $\bar{W} \in \C^{p \times n}$ as follows. If the $k$th PMU channel measures the voltage phasor of bus $j$, $\bar{W}_{kj}=1$; if it measures the current phasor from bus $i$ to bus $j$, then $\bar{W}_{ki}=1/Z^{ij}+Y^{ij}/2$, $\bar{W}_{kj}=-1/Z^{ij}$; $\bar{W}_{kj}=0$ otherwise. The PMU measurements and the state variables are related by
\begin{equation}\label{eqn:linear}
\bar{L}= X \bar{W}^T.
\end{equation}
The attack at time $t$, denoted by data injection $\bar{D}_{t,:}$, is called \textit{unobservable}\footnote{\cite{LNR11} focuses on DC model where power measurements and state variables are approximately related by linear equations. Here PMU measurements and state variables are accurately related by linear equation (\ref{eqn:linear}).} if and only if
\begin{align}
& \bar{D}_{t,:}=c^t\bar{W}^T
\end{align}
holds for some nonzero row vector $c^t \in \C^{1\times n}$. In this case,
\begin{align}
&\bar{L}_{t,:}+\bar{D}_{t,:}=(X_{t,:}+ c^t) \bar{W}^T,
\end{align}
and the operator would have the wrong impression that the state is $X_{t,:}+ c^t$. We focus on the cases that the attacks from time 1 to $t$ are all unobservable\footnote{ Our detection method can be extended to cases that both unobservable and observable attacks exist. See the beginning of Section \ref{sec:analysis}}, then we have
\begin{equation} \label{eqn:state}
\bar{D}=\left[\begin{array}{c}
c^1\\
\vdots\\
c^t
\end{array}\right]\bar{W}^T:= \bar{C} W^T,
\end{equation}
where $W_j=\bar{W}_j/\|\bar{W}_j\|$. $\bar{C}$ represents the additive error (up to a scaling factor) to bus voltages due to data attacks, i.e., $\|\bar{W}_j\|\bar{C}_j$ is the error to bus voltage $V^j$. Let $\bar{\I} \in \lsem n \rsem$ denote the column support of $\bar{C}$. We assume $\bar{C}$ is column-sparse because intruders might only alter some of the state variables due to resource constraints. With increasing installation of PMUs, we anticipate that the total number of PMU channels $p$ will be larger than the number of buses $n$. The transform in (\ref{eqn:state}) reduces the degree of freedom in $\bar{D}$. Combining (\ref{eqn:M}) and (\ref{eqn:state}), the obtained measurements under attack can be written as
\begin{equation}\label{eqn:problem}
M=\bar{L}+\bar{C}W^T+N.
\end{equation}
The attack identification problem is formulated as follows. Given $M$ and $W$, is it possible to separate $\bar{L}$ and $\bar{C}$? We assume noise level is bounded and given, i.e., $\|N\|_{F}\leq \eta$. We say a method can \textit{identify} an unobservable attack if it successfully determines the set of PMU channels that are under attack and recovers measurements that are not attacked.
Although cannot be detected at a given time instant, the unobservable attacks can be detected if the time series in the affected PMU channels exhibit dynamics different from those of unaffected PMUs. Mathematically, the matrix decomposition is possible if columns in $\bar{D}$ do not belong to the column space of $\bar{L}$.
\begin{figure}[h]
\begin{center}
\includegraphics[height=0.45\linewidth]{PMU_model.eps}
\caption{Three-bus example. PMUs are installed at bus 1 and bus 2 measuring the corresponding voltage phasors and incident line current phasors. } \label{fig:example} \vspace{-5mm}
\end{center}
\end{figure}
We use a three-bus network (Fig.~\ref{fig:example}) to illustrate the notations. Let $\V^i$ and $\bfI^{ij}$ ($i,j\in\{1,2,3\}$) in $\C^{t \times 1}$ denote the bus voltages and line currents in $t$ instants. Then
\begin{align}
\bar{L} &= [\bfV^1 \ \bfI^{12} \ \bfI^{13} \ \bfV^2 \ \bfI^{21} \ \bfI^{23}]= [\V^1 \ \V^2 \ \V^3] \bar{W}^T
\end{align}
where $\bar{W}^T$ is
\begin{equation}\nonumber
\notag \left[ \begin{array}{ccccccc}
1 & \frac{1}{Z^{12}}+\frac{Y^{12}}{2} & \frac{1}{Z^{13}}+\frac{Y^{13}}{2} & 0 & -\frac{1}{Z^{12}} & 0\\
0 & -\frac{1}{Z^{12}} & 0 & 1 & \frac{1}{Z^{12}}+\frac{Y^{12}}{2}& \frac{1}{Z^{23}}+\frac{Y^{23}}{2}\\
0 & 0 & -\frac{1}{Z^{13}} & 0 & 0 & -\frac{1}{Z^{23}}
\end{array}\right].
\end{equation}
Suppose the intruder manipulates measurements in all channels of PMU 1 and the channel of PMU 2 that measures $\bfI^{21}$ and $\bfI^{23}$ so that the system operator would have the wrong impression that the system states are $[\V^1 +\bm\beta^1 \ \V^2 \ \V^3+\bm\beta^2]$ for any nonzero $\bm\beta^1$, $\bm\beta^2 \in \C^{t\times 1}$. In this case, the observed measurements under attacks when there is no noise are
\begin{align}
&M = [\V^1 +\bm\beta^1 \ \ \V^2 \ \ \V^3+\bm\beta^2] \bar{W}^T \nonumber \\
& = [\V^1 +\bm\beta^1 \ \ \bfI^{12}+\frac{\bm\beta^1}{Z^{12}}+\frac{\bm\beta^1Y^{12}}{2} \ \ \bfI^{13}+\frac{\bm\beta^1-\bm\beta^2}{Z^{13}}+\frac{\bm\beta^1Y^{13}}{2}\nonumber \\ & \quad \quad \V^2 \ \ \bfI^{21}-\bm\beta^1/Z^{12} \ \ \bfI^{23}-\bm\beta^2/Z^{23}].
\end{align}
The additive errors due to attacks are
\begin{align}
\bar{D}
=[ \bm\beta^1 \ \ \bm 0 \ \ \bm\beta^2]\bar{W}^T= [\|\bar{W}_1\|\bm\beta^1 \ \ \bm 0 \ \ \|\bar{W}_3\|\bm\beta^2]W^T.
\end{align}
\subsection{Connections to existing work}\label{sec:app}
The detection of unobservable cyber data attacks has not been much addressed. \cite{SJ13} and \cite{LEDEH14} considered the detection of unobservable attacks to SCADA data and provided numerical results. \cite{SJ13} assumes the measurements across time are i.i.d. distributed and detects the attacks based on statistical learning. \cite{LEDEH14} assumes the SCADA measurements under DC power flow model are low-rank and proposes to detect the attacks by decomposing a low-rank matrix and a sparse matrix from their sum. Our work differs from \cite{LEDEH14} in that we assume the intruder constantly injects data attacks to the same set of PMUs, while \cite{LEDEH14} assumes the intruder attacks different PMUs at different time instants. Furthermore, we provide the theoretical guarantee of our detection method.
Our problem formulation of matrix decomposition is closely related to those in \cite{XCS12} and \cite{MMG13}. When $W$ is an identity matrix, our problem reduces to the one in \cite{XCS12}. The difference between our model and the one in \cite{MMG13} is that the sparse matrix $\bar{C}$ in \cite{MMG13} has nonzero entries located independent of each other, while $\bar{C}$ here is a column-sparse matrix. Our method and analysis are built up those in \cite{XCS12}, but we consider a more general framework of matrix decomposition through the introduction of the transform matrix $W$.
The significance of our work is twofold. First, we for the first time consider the case that the additive error matrix $\bar{D}$ can be dense (i.e., $W$ is a dense matrix), while the error matrices in \cite{XCS12} and \cite{MMG13} are sparse. We show through both theoretical analysis and numerical experiments that it is possible to achieve matrix decomposition with dense $\bar{D}$. Second, when $\bar{D}$ is a column-sparse matrix itself (i.e., $W$ is sparse), our decomposition method outperforms those in \cite{XCS12} and \cite{MMG13} (see Section \ref{sec:compare} and \ref{sec:PMU}) in the sense that our recovery method can tolerate a higher level of corruption (i.e., large support size of $\bar{D}$). This advance results from exploiting (\ref{eqn:state}), which reduces the degree of freedom of $\bar{D}$.
Note that our method and analysis hold for an arbitrary $W$ and can be applied to other domains that involve decomposing a matrix as in (\ref{eqn:problem}). As discussed in \cite{MMG13}, applications include unveiling network traffic anomalies \cite{LCD04,TJ03}, dynamic magnetic resonance imaging \cite{FNDRL06,GCSZ11}, face recognition \cite{BJ03}, and music analysis \cite{LGWWCM09,LW07}.
\section{Attack Identification with Noise}\label{sec:method1}
We now analyze the detection performance when $M$ contains noise ($N \neq 0$) with $\|N\|_F \le \eta$. Given $k$, define
\begin{equation}\nonumber
\lambda'_{\min,k}=\frac{(1+(2-\tilde{\psi})^{-1})\epsilon}{1/2-(1+(2-\tilde{\psi})^{-1})k\sigma\mu}, \text{ and } \lambda'_{\max,k}= \frac{1}{2}\sqrt{\frac{\tilde{\psi}}{k\sigma_k}}.
\end{equation}
\begin{theorem}\label{thm:recover1}
Suppose there exists nonzero $\tilde{k}$ such that $\tilde{k} \mu \leq c$ for $c$ in (\ref{eqn:psistar}), and $\lambda'_{\min,\tilde{k}} \leq \lambda'_{\max,\tilde{k}}$.
Then if column support size of $\bar{C}$ is at most $\tilde{k}$, for any $\lambda \in [\lambda'_{\min,\tilde{k}}, \lambda'_{\max,\tilde{k}}]$, there exists a pair $(\tilde{L}$, $\tilde{C})$, where $\tilde{L}+\tilde{C}W^T = \bar{L}+\bar{C}W^T$, $\P_{\bar{U}}(\tilde{L})=\tilde{L}$ and $\P_{\bar{\I}}(\tilde{C})=\tilde{C}$, such that the output of Method 1 satisfies
\begin{align}\label{eqn:L}\nonumber
& \|L^*-\tilde{L}\|_F \\
\le & (2-\tilde{\psi}+\frac{\lambda+(2-\tilde{\psi})\sqrt{1+(n-1)\mu}}{\lambda}\sqrt{\theta+3r})\frac{2\eta}{1-\tilde{\psi}},
\end{align}
\vspace{-5mm}
\begin{align}\label{eqn:C}\nonumber
&\text{and }\|C^*-\tilde{C}\|_F \\ \nonumber
\le & (1+(\frac{\lambda+\sqrt{1+(n-1)\mu}}{\lambda}+\frac{1-\tilde{\psi}}{\lambda\sigma_k\sqrt{1+(k-1)\mu}})\sqrt{\theta+3r})\\
& \frac{2\eta\sigma_k\sqrt{1+(k-1)\mu}}{1-\tilde{\psi}},
\end{align}
where $\theta := \min(t, p)$.
\end{theorem}
The discussion of the existence of $\lambda$ is very similar to the discussion for Theorem \ref{thm:recover}, so we skip it. If $\tilde{k} \mu \leq c$ and $\tilde{k}=O(n/r)$ hold, then a proper $\lambda$ exists. Theorem \ref{thm:recover1} guarantees that ($L^*$, $C^*$) returned by Method 1 is ``close'' to a pair that has the correct column space and column support, and the distance measured by Frobenius norm is proportional to the noise level $\eta$. The proof of Theorem \ref{thm:recover1} follows the same line as the proof of Theorem 2 in \cite{XCS12} mostly with modifications to address the projection matrix $W$. We establish Lemma \ref{lem:cond1}, a counterpart in the noisy case of Lemma \ref{lem:cond}, that demonstrates that Method 1 succeeds if there exists a dual certificate $Q$ with tighter requirements than that in the noiseless case.
\begin{lemma}\label{lem:cond1}
There exists $(\tilde{L}$, $\tilde{C})$ where $\tilde{L}+\tilde{C}W^T = \bar{L}+\bar{C}W^T$, $\P_{\bar{U}}(\tilde{L})=\tilde{L}$, $\P_{\bar{\I}}(\tilde{C})=\tilde{C}$, such that the output of Method 1 satisfies (\ref{eqn:L}) and (\ref{eqn:C}),
if there exists $Q \in \C^{t\times p}$ that satisfies
\begin{equation}\label{eqn:cond1}
\begin{aligned}
&(a) \P_{\bar{T}} (Q) = \bar{U} \bar{V}^{\dagger},
\quad\quad (b) \| \P_{\bar{T}^{\perp}} (Q)\| \leq 1/2 ,\\
&(c) (QW^{\ddagger})_{\bar{\I}}/\lambda \in \G(\bar{C}),
\quad \textrm{and }(d) \|(QW^{\ddagger})_{\bar{\I}^c} \|_{\infty,2} \leq \lambda/2.
\end{aligned}
\end{equation}
\end{lemma}
The construction of $Q$ is the same as that in Section \ref{sec:method} (equations (\ref{eqn:hath}) to (\ref{eqn:Q})). We show that $Q$ is the desire dual certificate if $\lambda$ belongs to $[\lambda'_{\min}, \lambda'_{\max}]$ in Lemma \ref{lem:certificate1}.
\begin{lemma}\label{lem:certificate1}
If the column support size of $\bar{C}$ is at most $\tilde{k}$, then for any $\lambda \in [\lambda'_{\min}, \lambda'_{\max}]$, $Q$ defined in (\ref{eqn:Q}) satisfies (\ref{eqn:cond1}).
\end{lemma}
Theorem \ref{thm:recover1} follows when we combine Lemmas \ref{lem:cond1} and \ref{lem:certificate1}. Please refer to the Appendix for the proofs.
\section{Simulation}\label{sec:simu}
We explore the performance of data attack identification methods on both synthetic data and actual PMU data from the Central NY power system. We use CVX \cite{CVX} to solve (\ref{eqn:opt}). We identify a column of $C^*$ to be nonzero if its $\ell_2$ norm exceeds the predefined threshold $\epsilon_1$. Method 1 succeeds if $\|U^*U^{*\dagger}-\bar{U}\bar{U}^{\dagger}\|\leq \epsilon_2$ for some small positive $\epsilon_2$, and the column supports of $\bar{C}$ and $C^*$ are the same.
\subsection{Performance on synthetic data}
Fix $t=p=50$. Given rank $r$, we generate matrices $A$ $\in$ $\R^{t \times r}$ and $B$ $\in$ $\R^{p \times r}$ with each entry independently drawn from Gaussian $\mathcal{N}(0,1)$ and set $\bar{L} := A B^{T}$. We generate matrix $W$ $\in$ $\R^{p \times n}$ with independent $\mathcal{N}(0,1)$ entries. To generate a column-sparse matrix $\bar{C} \in \R^{t \times n}$, we randomly select the column support and set the nonzero entries to be independent $\mathcal{N}(0,1)$. We vary $r$ and the number of corrupted columns, and take 100 runs for each case. $\lambda$ is set to be 0.95.
\subsubsection{Noiseless formulation}
We simulate the observed measurement matrix $M$ according to (\ref{eqn:problem}) with $N=0$. We apply Method 1 to obtain the estimation $(L^*,C^*)$. We set $\epsilon_1$ and $\epsilon_2$ to be 0.002 and 0.01, respectively. Fig.~\ref{fig:gray} shows the transition property of Method 1 in gray scale. White stands for 100\% success while black denotes 100\% failure. When $n$ is 25, $W$ is a tall matrix ($p>n$). When $n$ is 100, $W$ is a flat matrix ($p<n$). For both simulations, the identification is successful even when rank $r$ is six, and $\bar{C}$ has two nonzero columns.
\begin{figure}[h]
\begin{center}
\includegraphics[height=0.43\linewidth]{Gray.eps}
\caption{Matrix decomposition performance for different $n$} \label{fig:gray} \vspace{-3 mm}
\end{center}
\end{figure}
We further assume some of the observations are missing. We generate $M$ as before and then delete some randomly selected entries. Fig. \ref{fig:graynew} shows the decomposition performance of Method 1 for partial observation. We can see that the successful decomposition rate is close to the complete observation case even only 80\% of the entries are observed.
\begin{figure}[h]
\begin{center}
\includegraphics[height=0.43\linewidth]{Gray_new.eps}
\caption{Matrix decomposition performance for different $n$ with 80\% observed entries} \label{fig:graynew} \vspace{-7mm}
\end{center}
\end{figure}
\subsubsection{Noisy formulation}
We generate matrix $N \in \R^{t \times p}$ with independent Gaussian $\mathcal{N}(0,\sigma^2)$ entries. We fix the matrix rank $r$ to be 3 and the number of corrupted columns to be 3. We simulate the observed measurement matrix $M$ according to (\ref{eqn:problem}). We set $\eta$ to be $\|N\|_F$ and apply Method 1 to obtain the estimation $(L^*,C^*)$. $\epsilon_1$ is set to be 0.001.
\begin{figure}[h]
\begin{center}
\psfrag{Noise level sigma}[][][0.65]{Noise level $\sigma$}
\psfrag{noise level sigma}[][][0.65]{Noise level $\sigma$}
\psfrag{Succeed rate}[][][0.65]{Succeed rate}
\psfrag{Difference of the column space}[][][0.48]{Difference of the column space}
\psfrag{(a) Column space}[][][0.65]{(a) Column space}
\psfrag{(b) Set of corrupted columns}[][][0.65]{(b) Set of corrupted columns}
\psfrag{difference}[][][0.6]{$\|U^*U^{*\dagger}-\bar{U}\bar{U}^{\dagger}\|$}
\includegraphics[height=0.3\linewidth]{U_diff.eps}
\caption{Performance of Method 1 for different noise level $\sigma$} \label{fig:udiff} \vspace{-2mm}
\end{center}
\end{figure}
Fig. \ref{fig:udiff} shows the difference between the original and reconstructed column space ($\|U^*U^{*\dagger}-\bar{U}\bar{U}^{\dagger}\|$) and the succeed rate for determining the set of corrupted columns according to different noise level $\sigma$. We can see that Method 1 can successfully identify the corrupted columns when the noise level $\sigma$ is below 0.25. Method 1 can recover the column space with small errors when $\sigma$ is smaller than 0.1.
\subsection{Comparison with other methods on synthetic data}\label{sec:compare}
\subsubsection{{\small $\bar{D}=\bar{C}W^T$} is column-sparse}
Refs. \cite{XCS12} \cite{LEDEH14} considered matrix decomposition problem when $\bar{D}$ is column-sparse and scattered-sparse, respectively. We compare our method with them in the special case that $\bar{D}=\bar{C}W^T$ is column-sparse. Fix $t=p=50$, $n=20$, and $r=2$. We generate $\bar{L}$ and $\bar{C}$ with the same rules as in Section V-A. We generate a binary matrix $W$ $\in$ $\R^{p \times n}$ with two `1's each row and five `1's each column. Then the ratio of support sizes of $\bar{D}$ and $\bar{C}$ is about five. $\bar{D}$ is column-sparse when $\bar{C}$ is column-sparse. We simulate the measurement matrix $M$ according to (\ref{eqn:problem}) with $N=0$.
$\lambda$ in our method is set to be 0.9. $\lambda$'s in methods of \cite{XCS12} and \cite{LEDEH14} are set to be 0.5 and 0.1, respectively.
\begin{figure}[h]
\begin{center}
\psfrag{Number of outliers}[][][0.7]{Number of corrupted columns in $\bar{C}$}
\psfrag{success rate}[][][0.7]{Success rate}
\includegraphics[height=0.35\linewidth]{Compare_new.eps}
\caption{Success rates when $\bar{D}=\bar{C}W^T$ is column-sparse.} \label{fig:compare} \vspace{-5mm}
\end{center}
\end{figure}
Fig.~\ref{fig:compare} shows the success rates of three methods with different support sizes of $\bar{C}$. Our method performs the best since we exploit the structure $\bar{D}=\bar{C}W^T$ besides sparsity. The false alarm rate of our method is zero.
\subsubsection{Combination of attack patterns.} We consider the general case that the attacks satisfy (\ref{eqn:Dg}). We use the generalized version in (\ref{eqn:opt3})-(\ref{eqn:opt4}) to detect combined attacks. $\lambda_1$ and $\lambda_2$ in (\ref{eqn:opt3}) are set to be 1 and 0.1, respectively. $\lambda$'s in methods of \cite{XCS12} and \cite{LEDEH14} are set to be 0.5 and 0.1, respectively. $\bar{L}$, $\bar{C}$, and $W$ are generated the same as above. $\bar{S}$ is a sparse matrix with nonzero entries independently drawn from $\mathcal{N}(0,1)$. We define the correct estimation of the column space of $\bar{L}$ as a successful recovery. Fig.~\ref{fig:compare2} compares the methods when $\bar{C}$ is a zero matrix. The attacks are scattered-sparse, and our method performs as well as that in \cite{LEDEH14}. Fig.~\ref{fig:compare3} compares the methods when both column-sparse and scattered-sparse attacks exist. Besides a sparse $\bar{S}$, we randomly select two columns in $\bar{C}$ and select their entries independently from $\mathcal{N}(0,1)$. Only our method succeeds when both attacks exist.
\begin{figure}[h]
\begin{center}
\psfrag{Average corruption rate}[][][0.7]{Percentage of corrupted entries in $\bar{S}$}
\psfrag{success rate}[][][0.7]{Success rate}
\includegraphics[height=0.35\linewidth]{Compare2_new.eps}
\caption{Success rates when $\bar{D}=\bar{S}$ is scattered-sparse.} \label{fig:compare2}\vspace{-7mm}
\end{center}
\end{figure}
\begin{figure}[h]
\begin{center}
\psfrag{Average corruption rate}[][][0.7]{Percentage of corrupted entries in $\bar{S}$ when $\bar{C}$ has two nonzero columns}
\psfrag{success rate}[][][0.7]{Success rate}
\includegraphics[height=0.35\linewidth]{Compare3.eps}
\caption{Success rates when $\bar{D}=\bar{C}W^T+\bar{S}$.} \label{fig:compare3}\vspace{-7mm}
\end{center}
\end{figure}
\subsection{Performance comparison on actual PMU data}\label{sec:PMU}
We consider the PMU data shown in Section \ref{sec:pmu}. Two two-second PMU datasets are tested. One contains ambient data, and the other contains an abnormal event ($t = 17-19s$ and $t = 2-4s$ in Fig. \ref{fig:current}, respectively). We first inject data attacks as an intruder and then use Method 1 to detect the attacks.
We consider the scenario that an intruder alters the PMU channels that measure $I^{12}$,$I^{52}$,$I^{13}$ and $I^{43}$ in order to corrupt voltage estimations of Buses 2 and 3. Fig.~\ref{fig:attack} visualizes the actual PMU data and the data after the injection of attacks for two 2-second datasets. $\eta$ and $\lambda$ are set to be 5 and 1 respectively in Method 1. Fig.~\ref{fig:PMUfig2} shows the $\ell_2$ norm of each column of the resulting $\bar{D}$ matrix. The columns with significant $\ell_2$ norm correspond to channels that measure $I^{12}$,$I^{52}$,$I^{13}$ and $I^{43}$. Therefore, our method successfully identifies the four PMU channels under attack. We repeat the same experiment when an intruder alters the channels that measure $V^{5}$, $I^{52}$, $I^{54}$, $I^{59}$, and $I^{45}$ to corrupt voltage estimation of Buses 5. Fig.~\ref{fig:PMUfig3} shows the $\ell_2$ norm of each column of the resulting $\bar{C}$ matrix in this case. The column with significant $\ell_2$ norm corresponds Bus 5. Thus the recovery is also successful.
\begin{figure}[h]
\begin{center}
\includegraphics[height=0.33\linewidth]{PMU_attack.eps}
\caption{The actual PMU data and PMU data under attack} \label{fig:attack} \vspace{-5mm}
\end{center}
\end{figure}
\begin{figure}[h]
\begin{center}
\begin{psfrags}
\psfrag{l2 norm of the column}[][][0.6]{$\ell_2$ norm of the column}
\includegraphics[height=0.33\linewidth]{PMU_fig_3_new.eps}
\caption{$\ell_2$ norm of each column of $\bar{D}$} \label{fig:PMUfig2} \vspace{-5 mm}
\end{psfrags}
\end{center}
\end{figure}
\begin{figure}[h]
\begin{center}
\psfrag{l2 norm of the column}[][][0.6]{$\ell_2$ norm of the column}
\includegraphics[height=0.33\linewidth]{PMU_fig_5_new.eps}
\caption{$\ell_2$ norm of each column of $\bar{C}$} \label{fig:PMUfig3} \vspace{-5 mm}
\end{center}
\end{figure}
{Fig. \ref{fig:case2} compares our method and that in \cite{XCS12} on the ambient PMU data. Given support size of $\bar{C}$, the result is averaged over all possible attack locations. Our method outperforms \cite{XCS12} because we exploit (\ref{eqn:state}) to reduce the degree of freedom in $\bar{D}$. For example, 7 out of 23 channels needs to be attacked to change the state of Bus 1. That means 30\% of the columns of $\bar{D}$ are nonzero. This high percentage of corruption in $\bar{D}$ cannot be handedly by \cite{XCS12}.
\begin{figure}[h]
\begin{center}
\begin{psfrags}
\psfrag{Number of buses under attack}[][][0.7]{Number of buses under attack}
\psfrag{Success rate}[][][0.7]{Success rate}
\includegraphics[height=0.33\linewidth]{case2.eps}
\caption{Success rates with varying support size of $\bar{C}$, or equivalently, the number of affected system states. } \label{fig:case2}
\end{psfrags}
\end{center}
\end{figure}
| 3d6e6c152249f5d4aa15855efba15437cf4fff95 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction} \label{Sec_Introduction}
Modification of hadron properties in extreme environments such as finite temperature and density is one of the important problems in hadron physics, which is useful to clarify the relation between hadrons and the vacuum.
Similarly, an external magnetic field comparable to the typical scale of quantum chromodynamics (QCD) may drastically change hadron properties.
Such an intense field can be induced in noncentral heavy-ion collisions between two charged nuclei, where the strength is estimated to reach $|eB| \sim m_\pi^2 \sim 0.02 \, \mathrm{GeV^2}$ at the Relativistic Heavy Ion Collider (RHIC) and $|eB| \sim 15 m_\pi^2 \sim 0.3 \, \mathrm{GeV^2}$ at the Large Hadron Collider (LHC) \cite{Kharzeev:2007jp,Skokov:2009qp,Voronyuk:2011jd,Ou:2011fm,Bzdak:2011yy,Deng:2012pc,Bloczynski:2012en,Bloczynski:2013mca,Deng:2014uja,Huang:2015oca}.
Moreover, lattice QCD simulations are powerful tools to numerically confirm hadron properties in strong magnetic fields.
For light mesons (with only up, down and strange quarks) in a magnetic field, many studies have been performed using SU(2) \cite{Braguta:2011hq,Luschevskaya:2012xd} and SU(3) \cite{Bali:2011qj, Hidaka:2012mz, Luschevskaya:2014lga} lattice gauge theory, AdS/QCD \cite{Callebaut:2011ab,Ammon:2011je,Callebaut:2013wba} as well as other phenomenological models \cite{Chernodub:2010qx,Chernodub:2011mc,Chernodub:2012zx,Simonov:2012if,Kojo:2012js,Li:2013aa,Andreichikov:2013zba,Orlovsky:2013wjd,Frasca:2013kka,Chernodub:2013uja,Liu:2014uwa,Taya:2014nha,Kawaguchi:2015gpt,Hattori:2015aki,Liu:2016vuw,Zhang:2016qrl}.
Some of them were devoted to discussions on $\rho$ meson condensation \cite{Chernodub:2010qx,Chernodub:2011mc}.
There have been a few more studies for light baryons
\cite{Andreichikov:2013pga,Haber:2014ula,Taya:2014nha,He:2015zca}.
It is known that the spontaneous chiral symmetry breaking ($\langle \bar{q} q \rangle$ condensate) in the QCD vacuum is enhanced by a magnetic field, which is the so-called magnetic catalysis of $\langle \bar{q} q \rangle$.
This effect has been studied using various QCD-like models \cite{Klevansky:1989vi,Suganuma:1990nn,Klimenko:1992ch,Gusynin:1994re,Gusynin:1994va,Gusynin:1994xp,Gusynin:1995nb,Hong:1996pv,Elias:1996zu,Shushpanov:1997sf,Ebert:1999ht,Agasian:1999sx,Kabat:2002er,Miransky:2002rp,Cohen:2007bt,Werbos:2007ym,Scherer:2012nn,Watson:2013ghq,Mueller:2014tea} as well as lattice QCD simulations \cite{Buividovich:2008wf, D'Elia:2010nq, Braguta:2010ej, Ilgenfritz:2012fw, Bali:2011qj, D'Elia:2011zu, Bali:2012zg, Bali:2013esa, Bruckmann:2013oba,Ilgenfritz:2013ara,Bali:2014kia,Endrodi:2015oba} (see Refs.~\cite{Shovkovy:2012zn,Miransky:2015ava} for recent reviews).
As temperature or density dependence of the $\langle \bar{q} q \rangle$ condensate can modify hadron properties, the contribution of the magnetic field dependence of the condensate to hadron properties is also one of the more attractive topics in hadron physics.
Properties of hadrons with heavy quarks (charm and bottom) in a magnetic field were also investigated by various approaches.
They include potential models \cite{Machado:2013rta,Alford:2013jva,Bonati:2015dka,Suzuki:2016kcs}, QCD sum rules \cite{Machado:2013yaa,Cho:2014exa,Cho:2014loa,Gubler:2015qok} and AdS/QCD \cite{Dudal:2014jfa}.
Among them, the constituent quark model is a suitable tool to investigate properties of heavy hadrons in an external field.
In such a model, there are mainly two physical effects which lead to the change of hadronic properties: (i) {\it mixing between different spin states} by the $-\bm{\mu}_i \cdot \bm{B}$ term and (ii) {\it modification of quark kinetic energy} by the $\bm{B} \times \bm{r}$ term.
In addition, for heavy-light meson systems such as $D$ and $B$ mesons, we expect two additional effects (iii) {\it Zeeman splitting} by the $-\bm{\mu}_i \cdot \bm{B}$ term and (iv) {\it magnetic field dependence of constituent light quark mass} induced by the magnetic catalysis.
Furthermore, (v) {\it the anisotropy of the linear and Coulomb potential} between a quark and an antiquark manifests itself in a magnetic field \cite{Miransky:2002rp,Andreichikov:2012xe,Chernodub:2014uua,Bonati:2014ksa,Rougemont:2014efa,Simonov:2015yka}.
The latter effect was evaluated by recent lattice QCD simulations \cite{Bonati:2014ksa} and its correction in the potential model for quarkonia was investigated in Ref.~\cite{Bonati:2015dka}.
In this work, we will discuss the effects of (i)-(iv) by focusing on the neutral heavy meson systems.
The main purpose of this paper is a systematic understanding of heavy-meson spectra in a magnetic field,
while our previous work \cite{Suzuki:2016kcs} is devoted to the findings of the level structure and wave function deformation in charmonia.
To this end, we emphasize the importance of the comparison between charmonia and bottomonia,
which is characterized by their different masses and electric charge.
Furthermore, comparing quarkonia and heavy-light mesons is helpful to clarify the role of a light quark in heavy-meson systems under magnetic field.
It is not technically easy to solve the two-body Schr\"{o}dinger equation with a confining potential under strong magnetic field and to extract the anisotropic wave functions for ground and excited states simultaneously.
In Refs.~\cite{Alford:2013jva,Bonati:2015dka}, the finite differential time domain (FDTD) method is applied to obtain the eigenvalues and eigenstates for the ground states of quarkonia.
In our previous work \cite{Suzuki:2016kcs}, we proposed an alternative approach based on the variational method, which we call the cylindrical Gaussian expansion method (CGEM).
This method is an extension from the conventional Gaussian expansion method (GEM) \cite{Kamimura:1988zz,Hiyama:2003cu} mainly used for nuclear and atomic few-body systems.
The CGEM has the following nice properties: (i) it respects the symmetry of the Hamiltonian under constant magnetic field, (ii) it can deal fully with higher excited states and (iii) it reduces the computational cost substantially.
This paper is organized as follows.
In Sec. \ref{Sec_Formalism}, we construct a two-body Hamiltonian for heavy mesons in a magnetic field and develop the CGEM to solve the Schr\"{o}dinger equation in the cylindrical coordinate.
Numerical results obtained from this approach are shown in Sec. \ref{Sec_Results} and their physical pictures are discussed.
Section \ref{Sec_Conclusion and outlook} is devoted to our conclusion and outlook.
\section{Formalism} \label{Sec_Formalism}
\subsection{Quark model in a magnetic field}
Nonrelativistic two-body Hamiltonian of the constituent quark model in a magnetic field is written as \cite{Alford:2013jva}
\begin{equation}
H = \sum_{i=1}^2 \left[ \frac{1}{2m_i} \left( \bm{p}_i - q_i \bm{A} \right)^2 - \bm{\mu}_i \cdot \bm{B} +m_i \right] +V(r),
\end{equation}
where $m_i$, $\bm{p_i}$, $q_i$ and $\bm{\mu}_i$ are the constituent quark mass, the momentum of quarks, the quark electric charge and the quark magnetic moment, respectively.
For the vector potential, we choose the symmetric gauge $\bm{A}(\bm{r}_i) = \frac{1}{2} \bm{B} \times \bm{r}_i$.
We introduce the center of mass and relative coordinates, $\bm{R}=(m_1 \bm{r}_1 + m_2 \bm{r}_2)/M $ and $\bm{r} = \bm{r}_1 - \bm{r}_2$, where $M=m_1+m_2$ is the total mass of the two particles.
As a new conserved quantity, we define the pseudomomentum \cite{Avron:1978}:
\begin{equation}
\hat{\bm{K}} = \sum_{i=1}^2 \left[ \bm{p}_i +\frac{1}{2} q_i \bm{B} \times \bm{r}_i \right]
\end{equation}
whose commutation relation is given by $[\hat{K}_i, \hat{K}_j] = -i(q_1+q_2) \epsilon_{ijk} B_k$.
When the system is charge neutral, the components $\hat{K}_i$ commute with each other.
By using the pseudomomentum, the total wave function for the Hamiltonian can be factorized as follows:
\begin{equation}
\Phi(\bm{R},\bm{r}) = \exp \left[ i (\bm{K} -\frac{1}{2} q\bm{B} \times \bm{r}) \cdot \bm{R} \right] \Psi(\bm{r}).
\end{equation}
For a neutral system with $q_1=-q_2=q$, the total Hamiltonian can be reduced into
\begin{eqnarray}
H_\mathrm{rel} &=& \frac{\bm{K}^2}{2M} - \frac{q}{M}(\bm{K} \times \bm{B}) \cdot \bm{r} \nonumber\\
&& + \frac{q}{2} \left( \frac{1}{m_1} - \frac{1}{m_2} \right) \bm{B} \cdot( \bm{r} \times \bm{p}) \\
&& + \frac{\bm{p}^2}{2\mu} + \frac{q^2}{8\mu}( \bm{B} \times \bm{r})^2 + V(r) + \sum_{i=1}^2 [-\bm{\mu}_i \cdot \bm{B} + m_i], \nonumber
\end{eqnarray}
where $\bm{p}=\frac{m_2\bm{p_1}-m_1\bm{p_2}}{m_1+m_2}$, $\mu = m_1 m_2/M$ are the relative momentum and the reduced mass, respectively.
If the orbital angular momentum along $\bm{B}$ is zero, we have $\bm{B} \cdot (\bm{r} \times \bm{p}) = 0$.
When we focus on a magnetic field along the $z$-axis, $\bm{B} =(0,0,B) $, $\bm{B} \times \bm{r} = B(-y,x,0)$ and $\bm{K} =(K_x,K_y,0) $, the Hamiltonian is reduced to, by using $p=-i \nabla$,
\begin{eqnarray}
H_\mathrm{rel} &=& \frac{\bm{K}^2}{2M} -\frac{\nabla^2}{2\mu} + \frac{q^2 B^2}{8\mu} \rho^2 + \frac{qB}{4\mu} K_x y - \frac{qB}{4\mu} K_y x \nonumber\\
&& + V(r) + \sum_{i=1}^2 [-\bm{\mu}_i \cdot \bm{B} + m_i], \label{H_rel2}
\end{eqnarray}
where $\rho=\sqrt{x^2+y^2}$.
In this work, we investigate only the case of $\bm{K}=0$. In this situation, we can calculate the eigenvalues and eigenstates for mesons at rest. Then, the Hamiltonian maintains the rotational symmetry on the transverse $x$-$y$ plane and the reflection symmetry along the $z$-axis.
For the potential term in the model, we adopt the Cornell potential \cite{Eichten:1974af} with a spin-spin interaction:
\begin{eqnarray}
V(r) &=& \sigma r - \frac{A}{r} + \alpha (\bm{S}_1 \cdot \bm{S}_2) e^{- \Lambda r^2} + C \nonumber\\
&=& \sigma \sqrt{\rho^2 + z^2} -\frac{A}{\sqrt{\rho^2 + z^2}} + \alpha (\bm{S}_1 \cdot \bm{S}_2) e^{- \Lambda (\rho^2 + z^2)} \nonumber\\
&& + C,
\end{eqnarray}
where $\bm{S}_1 \cdot \bm{S}_2 = -3/4$ and $1/4$ for the spin singlet and triplet, respectively.
Here, choosing the conventional Gaussian form for the spin-spin interaction leads to the analytic expressions of the matrix elements.
Let us now consider the contribution of the effects induced by the term $-\bm{\mu}_i \cdot \bm{B}$ \cite{Machado:2013rta,Alford:2013jva}, where the quark magnetic moment is given by $\bm{\mu}_i = g q_i \bm{S}_i/2m_i$ with the Land$\acute{\mathrm{e}}$ $g$-factor. We have assumed $g=2$.
As a result, for a neutral meson system, we obtain
\begin{eqnarray}
-(\bm{\mu}_1 + \bm{\mu}_2) \cdot \bm{B} &=& -\frac{gq}{2} \left( \frac{\bm{S}_1}{m_1} - \frac{\bm{S}_2}{m_2} \right) \cdot \bm{B} \nonumber\\
&=& -\frac{gq}{4} \left( \frac{\bm{\sigma}^1}{m_1} - \frac{\bm{\sigma}^2}{m_2} \right) \cdot \bm{B},
\end{eqnarray}
where $\bm{\sigma}^i$ is the Pauli matrix.
Then the eigenstates with the different spin quantum numbers, the singlet $| 00 \rangle =(| \uparrow \downarrow \rangle - | \downarrow \uparrow \rangle) / \sqrt{2}$ and ``longitudinal" component $| 10 \rangle =(| \uparrow \downarrow \rangle + | \downarrow \uparrow \rangle) / \sqrt{2}$ of the triplet, mix with each other through the off-diagonal matrix element:
\begin{equation}
\langle 10 | -(\bm{\mu}_1 + \bm{\mu}_2) \cdot \bm{B} |00 \rangle = -\frac{gqB}{4} \left( \frac{1}{m_1} + \frac{1}{m_2} \right). \label{mat_ele_1}
\end{equation}
To take into account the off-diagonal components of the matrix elements, Eq.~(\ref{mat_ele_1}), we have to solve a {\it coupled-channel} Schr\"{o}dinger equation.
On the other hand, ``transverse" components $| 1 \pm1 \rangle $ ($| \uparrow \uparrow \rangle$ and $| \downarrow \downarrow \rangle$) of the triplet cannot mix with other states and the off-diagonal components are zero.
For neutral systems with $m_1 \neq m_2$ (e.g. heavy-light mesons), the expectation values of these diagonal components are given by
\begin{equation}
\langle 1 \pm1 | -(\bm{\mu}_1 + \bm{\mu}_2) \cdot \bm{B} | 1 \pm1 \rangle = \mp \frac{gqB}{4} \left( \frac{1}{m_1} - \frac{1}{m_2} \right). \label{mat_ele_2}
\end{equation}
For systems with $m_1=m_2$ (quarkonia), these values become zero.
Then, the couplings between the quark magnetic moments and the magnetic field are completely canceled between the quark and the antiquark.
\subsection{Cylindrical Gaussian expansion method}
In this section, we review the cylindrical Gaussian expansion method.
After that, we introduce a coupled-channel basis to solve the coupled-channel Schr\"{o}dinger equation for the mixing between the spin singlet $| 00 \rangle$ and longitudinal component $| 10 \rangle$ of the spin triplet.
\subsubsection{Cylindrical Gaussian basis}
We have no analytic solution for the two-body Hamiltonian with the confinement potential and a magnetic field.
To solve the two-body Schr\"{o}dinger equation, we use the GEM based on the Rayleigh-Ritz variational method, which is a powerful tool in nuclear and atomic physics \cite{Kamimura:1988zz,Hiyama:2003cu}.
In a magnetic field, a wave function should be expanded individually by the Gaussian basis on the transverse plane, $e^{- \beta_n \rho^2}$, and the basis along the longitudinal axis, $e^{- \gamma_n z^2}$, in the cylindrical coordinate $(\rho,z,\phi)$.
Such a cylindrical basis was first applied to the quark model in our previous work \cite{Suzuki:2016kcs} and a similar form was also successfully used for atomic systems such as hydrogen and helium in a magnetic field (e.g. Refs.~\cite{Aldrich1979,Becken:2001}).
The spatial part of the trial wave function for $l_z=0$ is given as follows \cite{Suzuki:2016kcs}:
\begin{eqnarray}
&& \Psi (\rho,z,\phi) = \sum_{n=1}^N C_n \Phi_n (\rho,z,\phi), \label{basis1} \\
&& \Phi_n (\rho,z,\phi) = N_n e^{- \beta_n \rho^2} e^{- \gamma_n z^2}, \label{basis2}
\end{eqnarray}
where $N$ and $C_n$ are the number of basis functions and the expansion coefficients.
$N_n$ is the normalization constant defined by $ \langle \Phi_n| \Phi_n \rangle =1$.
$\beta_n$ and $\gamma_n$ are Gaussian range parameters for the $\rho$ and $z$ directions, respectively.
It is empirically known that the best set of the range parameters, $\beta_1 \cdots \beta_N$ and $\gamma_1 \cdots \gamma_N$, are the geometric progressions:
\begin{eqnarray}
&& \beta_n = \frac{1}{\rho_n^2}, \ \ \ \rho_n = \rho_1 b^{n-1}, \label{geo1} \\
&& \gamma_n = \frac{1}{z_n^2}, \ \ \ z_n = z_1 c^{n-1}. \label{geo2}
\end{eqnarray}
Here, the four parameters, $\rho_1$, $\rho_N$, $z_1$ and $z_N$, will be optimized as the energy eigenvalue is minimized.
The analytic form of the matrix elements from the Hamiltonian (\ref{H_rel2}) and the wave function basis (\ref{basis1}) are summarized in Appendix \ref{App_mat}.
We checked the applicability of our numerical code by comparing our variational result with the analytic solution of the three-dimensional anisotropic harmonic oscillator.
\subsubsection{Coupled-channel basis for spin mixing}
Since the spin singlet $| 00 \rangle$ and longitudinal component $| 10 \rangle$ of the spin triplet mix with each other in a magnetic field as seen in Eq.~(\ref{mat_ele_1}), the wave function should take into account the coupled channel.
Then the basis function is given as a linear combination of the spin-0 and spin-1 components:
\begin{eqnarray}
&& \Psi (\rho,z,\phi) = \sum_{n=1}^N C^{S=0}_n \Phi^{S=0}_n (\rho,z,\phi) \nonumber \\
&& \hspace{43pt} + \sum_{n=N+1}^{2N} C^{S=1}_n \Phi^{S=1}_n (\rho,z,\phi), \label{basis3} \\
&& \Phi^{S=0}_n (\rho,z,\phi) = N^A_n e^{- \beta^A_n \rho^2} e^{- \gamma^A_n z^2}, \\
&& \Phi^{S=1}_n (\rho,z,\phi) = N^B_n e^{- \beta^B_n \rho^2} e^{- \gamma^B_n z^2},
\end{eqnarray}
where we have the four range parameters, $\beta^A_n$, $\gamma^A_n$, $\beta^B_n$ and $\gamma^B_n$.
These parameters are also expressed by the geometrical progression, so that we finally optimize the eight parameters.
A brief review of the generalized eigenvalue problem for the coupled-channel basis is shown in Appendix \ref{App_GEP}.
\subsection{Numerical setup} \label{subsec_parameter}
In this section, we set the parameters of our constituent quark model.
For parameters in charmonium systems, the charm-quark kinetic mass $m_c=1.7840 \, \mathrm{GeV}$, the Coulomb parameter $A=0.713$ and the string tension $\sqrt{\sigma} = 0.402 \, \mathrm{GeV}$ are determined from the equal-time $Q\bar{Q}$ Bethe-Salpeter amplitude in lattice QCD \cite{Kawanai:2015tga}.
For the spin-dependent potential given as $ \alpha \exp{(-\Lambda r^2)}$, we adopt $\Lambda=1.020 \, \mathrm{GeV}^2$ obtained from lattice QCD \cite{Kawanai:2011jt} by fitting a charmonium potential.
The remaining parameters, $\alpha=0.4778 \, \rm{GeV}$ and $C=-0.5693 \, \rm{GeV}$, are chosen to reproduce the experimental values of the masses of the ground states of $\eta_c$ and $J/\psi$.
We checked that these parameters can reproduce the masses of the ground and excited states of the charmonium well.
For bottomonium system, we have no information from lattice QCD.
Therefore, we fix $\sqrt{\sigma}$ on the same value as that of charmonium.
Furthermore, we put $C=0$ and tune the bottom-quark kinetic mass $m_b$, $A$ and $\alpha$ to reproduce the experimental masses of $\eta_b$, $\eta_b(2S)$ and $\Upsilon(1S)$.
For $D_s$ meson systems, we know the values of $m_s=0.656 \, \mathrm{GeV}$, $A=1.30 \, \mathrm{GeV}$ and $\sqrt{\sigma}=0.324 \, \mathrm{GeV}$ from the lattice QCD \cite{Kawanai:2015tga}.
Then, we tune only $\alpha$ and $C$ by matching our results with the experimental masses of $D_s$ and $D^\ast_s$ mesons.
For $D$, $B$ and $B_s$ meson systems, we fix $A$ and $\sqrt{\sigma}$ on the same values as those of $D_s$ meson.
The light-quark kinetic mass is fixed as $m_q=0.3 \, \mathrm{GeV}$ and only $\alpha$ and $C$ are tuned by the experimental masses of $D$--$D^\ast$, and $B$--$B^\ast$ and $B_s$--$B_s^\ast$ mesons.
Under these criteria, the resulting parameters are summarized in Tables \ref{paramter_list_M} and \ref{paramter_list_P}.
A list of the predicted meson masses is shown in Table \ref{paramter_list_com}.
\begin{table}[t!]
\begin{center}
\caption{Parameters for kinetic (constituent) quark masses.
$m_c$ for charmonium and $m_s$ for the $D_s$ meson were obtained from lattice QCD in Ref.~\cite{Kawanai:2015tga}.}
\label{paramter_list_M}
\begin{tabular}{c|c|c|c}
\hline \hline
$m_c \ [\mathrm{GeV}]$ & $m_b \ [\mathrm{GeV}]$ & $m_s \ [\mathrm{GeV}]$ & $m_q \ [\mathrm{GeV}]$ \\
\hline
$1.784$ & $4.8078$ & $0.656$ & $0.3$ \\
\hline \hline
\end{tabular}
\end{center}
\end{table}
\begin{table}[t!]
\begin{center}
\caption{Parameters for potential terms.
$A$ and $\sqrt{\sigma}$ for charmonium and the $D_s$ meson were obtained from lattice QCD in Ref.~\cite{Kawanai:2015tga}.
$\Lambda$ for charmonium was estimated from the lattice QCD in Ref.~\cite{Kawanai:2011jt}. }
\label{paramter_list_P}
\begin{tabular}{l|c|c|c|c|c}
\hline \hline
Meson & $A$ & $\sqrt{\sigma} \ [\mathrm{GeV}]$ & $\alpha \ [\mathrm{GeV}]$ & $\Lambda \ [\mathrm{GeV}^2]$ & $C \ [\mathrm{GeV}]$ \\
\hline
$c\bar{c}$ & $0.713$ & $0.402$ & $0.4778$ & $1.020$ & $-0.5693$ \\
$b\bar{b}$ & $0.531$ & $0.402$ & $0.1322$ & $1.020$ & $0$ \\
$D_s \, (c\bar{s})$ & $1.30$ & $0.324$ & $0.7954$ & $1.020$ & $-0.1765$ \\
$B_s \, (b\bar{s})$ & $1.30$ & $0.324$ & $0.2210$ & $1.020$ & $+0.2376$ \\
$D \, (c\bar{q})$ & $1.30$ & $0.324$ & $1.7812$ & $1.020$ & $-0.2419$ \\
$B \, (b\bar{q})$ & $1.30$ & $0.324$ & $0.5589$ & $1.020$ & $+0.1119$ \\
\hline \hline
\end{tabular}
\end{center}
\end{table}
\begin{table}[t!]
\begin{center}
\caption{Comparison between meson masses predicted from parameters in Tables \ref{paramter_list_M} and \ref{paramter_list_P} and corresponding experimental values.}
\label{paramter_list_com}
\begin{tabular}{l|c|c|c}
\hline \hline
State & Meson & Our result $[\mathrm{GeV}]$ & Experiment $[\mathrm{GeV}]$ \cite{Agashe:2014kda} \\
\hline
$c\bar{c}$ $1^1S_0$ & $\eta_c(1S)$ & 2.984 & 2.984 \\
$c\bar{c}$ $1^3S_1$ & $J/\psi$ & 3.097 & 3.097 \\
$c\bar{c}$ $2^1S_0$ & $\eta_c(2S)$ & 3.669 & 3.639 \\
$c\bar{c}$ $2^3S_1$ & $\psi(2S)$ & 3.707 & 3.686 \\
\hline
$b\bar{b}$ $1^1S_0$ & $\eta_b(1S)$ & 9.398 & 9.398 \\
$b\bar{b}$ $1^3S_1$ & $\Upsilon(1S)$ & 9.460 & 9.460 \\
$b\bar{b}$ $2^1S_0$ & $\eta_b(2S)$ & 9.999 & 9.999 \\
$b\bar{b}$ $2^3S_1$ & $\Upsilon(2S)$ & 10.013 & 10.002 \\
$b\bar{b}$ $3^1S_0$ & $\eta_b(3S)$ & 10.330 & ... \\
$b\bar{b}$ $3^3S_1$ & $\Upsilon(2S)$ & 10.339 & 10.355 \\
\hline
$c\bar{s}$ $1^1S_0$ & $D_s$ & 1.968 & 1.968 \\
$c\bar{s}$ $1^3S_1$ & $D_s^\ast$ & 2.112 & 2.112 \\
\hline
$b\bar{s}$ $1^1S_0$ & $B_s$ & 5.367 & 5.367 \\
$b\bar{s}$ $1^3S_1$ & $B_s^\ast$ & 5.415 & 5.415 \\
\hline
$c\bar{q}$ $1^1S_0$ & $D$ & 1.870 & 1.870 \\
$c\bar{q}$ $1^3S_1$ & $D^\ast$ & 2.010 & 2.010 \\
\hline
$b\bar{q}$ $1^1S_0$ & $B$ & 5.279 & 5.279 \\
$b\bar{q}$ $1^3S_1$ & $B^\ast$ & 5.325 & 5.325 \\
\hline \hline
\end{tabular}
\end{center}
\end{table}
Here we comment on the possible magnetic-field dependence of those parameters.
As discussed in Refs.~\cite{Miransky:2002rp,Andreichikov:2012xe,Chernodub:2014uua,Bonati:2014ksa,Rougemont:2014efa,Simonov:2015yka}, the linear and Coulomb potential can have an anisotropy in a magnetic field, which leads to different $A$ and $\sqrt{\sigma}$ between the $\rho$ and $z$ directions.
Similarly, one can expect to appear $B$-dependences of the spin-spin interaction and constant term $C$.
In this work, although we do not consider those effects, such improvements would be interesting in the future.
Furthermore, the constituent light quark mass, $m_q$ and $m_s$, can be related to the chiral symmetry breaking in vacuum, so that the $B$-dependence of the chiral condensate can modify these masses.
Such effects in a heavy-light meson will be investigated in Sec. \ref{SubSec_MC}.
\section{Numerical results} \label{Sec_Results}
\begin{figure}[t!]
\centering
\includegraphics[clip, width=1.0\columnwidth]{charmT_m}
\caption{Masses of transverse $J/\psi, \psi(2S)$ in a magnetic field.
Densities of wave functions, defined by $|\Psi(\rho,z,\phi)|^2$, are plotted on the horizontal $x$ (or $y$)-vertical $z$ plane. }
\label{Results_charmT_m}
\end{figure}
\begin{figure}[t!]
\centering
\includegraphics[clip, width=1.0\columnwidth]{charmT_r}
\caption{RMS radii of transverse $J/\psi, \psi(2S)$ in a magnetic field, defined by Eqs.~(\ref{RMS_rho}) and (\ref{RMS_z}).}
\label{Results_charmT_r}
\end{figure}
\begin{figure*}[tb!]
\centering
\includegraphics[clip, width=17cm]{charmL_m}
\caption{Masses of $\eta_c (1S,2S)$ and longitudinal $J/\psi$, $\psi(2S)$ in a magnetic field.}
\label{Results_charmL_m}
\end{figure*}
\subsection{Charmonia}
The transverse vector charmonia [$J/\psi_{|1 \pm1 \rangle}$ and $\psi(2S)_{|1 \pm1 \rangle}$] in a magnetic field do not show mixing effects.
The mass spectra and the shapes of the wave functions are shown in Fig.~\ref{Results_charmT_m}.
In a weak magnetic field, both the masses of the ground and excited states increase gradually.
We note that the mass shifts of the ground state estimated in weak magnetic fields agree with those obtained by using the same Hamiltonian and the FDTD method in Refs.~\cite{Alford:2013jva,Bonati:2015dka}.
We find that the excited state $\psi(2S)$ is more sensitive to a magnetic field than the ground state $J/\psi$.
For instance, the mass shifts at $eB=0.3 \, \mathrm{GeV}^2$ are $+12.4$ and $+50.8 \, \mathrm{MeV}$ for $J/\psi$ and $\psi(2S)$, respectively.
The reason is that the size of the wave function of $\psi(2S)$ is twice as large as that of the corresponding ground state, so that the expectation value of the $\rho^2$ term in Eq.~(\ref{H_rel2}) also becomes larger.
In a strong magnetic field, we see that the masses are linearly raised.
Such a mass shift can be quantitatively understood as follows.
The energy of a single constituent quark in a magnetic field is shifted by the {\it nonrelativistic} quark Landau level, $(n + 1/2)|q|B/m_c$, where $n$ is the discretized quantum number.
The linear increase of the mass of a charmonium, as observed in our result, is consistent with the sum of such energy shifts for the two quarks inside the hadron.
Thus, in the strong fields, {\it the single-particle picture for the mass shifts is considered as a good approximation, even though it is not the rigorous two-body problem}.
We mention that, in Ref.~\cite{Taya:2014nha}, the masses of light mesons and baryons in a strong magnetic field were discussed from relativistic Landau levels of single particle quarks.
\begin{figure*}[tb!]
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{charmL_r}
\end{minipage}%
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{charm_mix}
\end{minipage}
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{charmL_r2}
\end{minipage}%
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{charm_mix2}
\end{minipage}
\caption{RMS radii and mixing rates of $\eta_c (1S,2S)$ and longitudinal $J/\psi, \psi(2S)$ in a magnetic field.}
\label{Results_charmL_r}
\end{figure*}
\begin{figure*}[t!]
\centering
\includegraphics[clip, width=17cm]{SPfunc}
\caption{A schematic picture for an expected spectral function of longitudinal vector quarkonia in a magnetic field.
In zero field, there are only vector states corresponding to $J/\psi$ and $\psi(2S)$ in the spectral function.
In finite $eB$, the additional peaks such as the first and third states are newly induced.
Simultaneously the residues and positions of peaks can be modified.
By the level repulsion with the mixing partner, the second and third states approach each other.
$eB_\mathrm{cross}$ means the ''crossing magnetic field'' where the two peaks are located in the nearest positions.
After $eB_\mathrm{cross}$, the two peak positions start to separate.}
\label{SPfunc}
\end{figure*}
In Fig.~\ref{Results_charmT_r}, we show the root-mean square (RMS) radii of the wave function of $J/\psi_{|1 \pm1 \rangle}$ and $\psi(2S)_{|1 \pm1 \rangle}$, calculated from the matrix elements in Appendix \ref{App_RMS}.
Here, we defined the RMS radii by
\begin{eqnarray}
&& \sqrt{\langle \rho^2 \rangle} \equiv \sqrt{(3/2) \langle \Psi | \rho^2 |\Psi \rangle}, \label{RMS_rho} \\
&& \sqrt{\langle z^2 \rangle} \equiv \sqrt{ 3 \langle \Psi | z^2 |\Psi \rangle}, \label{RMS_z}
\end{eqnarray}
in the $\rho$- and $z$-directions, respectively.
These values in vacuum are the same as the usual spherical RMS radius $\sqrt{\langle r^2 \rangle} $: $\sqrt{\langle r^2 \rangle}_{B=0} = \sqrt{\langle \rho^2 \rangle}_{B=0} = \sqrt{\langle z^2 \rangle}_{B=0}$.
From this figure, we observe that both $\sqrt{\langle \rho^2 \rangle}$ and $\sqrt{\langle z^2 \rangle}$ of $J/\psi_{|1 \pm1 \rangle}$ decrease gradually.
The shrinkage on the $\rho$-plane is induced by the $\rho^2$ term in Eq.~(\ref{H_rel2}) as expected naively.
In the case of $\psi(2S)_{|1 \pm1 \rangle}$, $\sqrt{\langle \rho^2 \rangle}$ decreases gradually in all the regions of the magnetic field while $\sqrt{\langle z^2 \rangle}$ increases in weak magnetic fields.
At about $eB=1.0 \, \mathrm{GeV}$, the enlargement of the wave function in the $z$-direction saturates, and, after that, it begins to decrease little by little.
Then, in the wave function of a $2S$ state, only the excitation in the $z$-direction survives, which means that it behaves as a quasi-one-dimensional object.
The results for pseudoscalar [$\eta_c(1S)$ and $\eta_c(2S)$] and longitudinal vector [$J/\psi_{|1 0 \rangle}$ and $\psi(2S)_{|1 0 \rangle}$] charmonia are shown in Figs.~\ref{Results_charmL_m} and \ref{Results_charmL_r}.
In these hadrons in a magnetic field, spin-0 ($\eta_c$) and spin-1 ($\psi$) components are mixed with each other by the $-\bm{\mu}_i \cdot \bm{B}$ term in Eq.~(\ref{H_rel2}), so that we could no longer call these states $\eta_c$ or $J/\psi$ .
Therefore, we will label these states as {\it first}, {\it second}, {\it third}, and {\it fourth states}.
For the {\it first} state starting from $\eta_c(1S)$ in vacuum, the mixing partners include all spin-1 excited states [$\psi(2S)$, $\psi(3S)$ ...] as well as the ground state $J/\psi$.
$\eta_c(1S)$ is expected to be predominantly mixed with $J/\psi$ and its mass decreases gradually by the level repulsion with increasing magnetic field.
Such predominance of the mixing partners is also shown as the agreement between the first and second states in the figure of the mixing rates as shown in the upper right of Fig.~\ref{Results_charmL_r}.
One of the interesting points is that the RMS radii in both $\rho$- and $z$-directions, for the first state in weak magnetic fields up to $eB=0.2 \, \mathrm{GeV}^2$, seem to increase slightly, as shown in the upper left of Fig.~\ref{Results_charmL_r}.
The behavior might come from the contamination of the $J/\psi$ components with the larger RMS radius.
The radii of the second state are in contrast decreasing as $B$ grows.
The {\it second} state starting from $J/\psi$ in vacuum is also expected to be mixed with $\eta_c(1S)$ in a weak magnetic field.
Its mass approaches the third state like $\eta_c(2S)$ at a ``crossing magnetic field'' located in about $eB=1.0$-$1.1 \, \mathrm{GeV}^2$.
Such a crossing point (strictly, avoided crossing) can also be seen as a jump in the figure of the RMS radii and the mixing rates.
In this region, the dominance of the $\psi$ component, characterized by the mixing rate more than $50\%$, is reversed, and the rate approaches almost $50\%$.
After the crossing, the wave function is converted to the $2S$-like shape with one node, and the mass of the second state begins to decrease slowly.
Note that, as with the case of transverse $J/\psi$, the mass shifts of the first and second states in the weak magnetic fields obtained by the previous works in Refs.~\cite{Alford:2013jva,Bonati:2015dka} are consistent with our results.
We also checked the comparison with the result from a hadron effective Lagrangian suggested in Refs.~\cite{Cho:2014exa,Cho:2014loa} in Appendix~\ref{App_EFT}.
Furthermore, the {\it third} state starting from $\eta_c(2S)$ can be expected to mix with the $\psi(2S)$ state in a weak magnetic field.
An important point is that the mass shifts of the excited states in the weak magnetic field are more sensitive to $eB$ than that of the ground state.
For instance, the obtained mass shifts of the first, second, third and fourth states at $eB=0.1 \, \mathrm{GeV}^2$ are $-9.9$, $+12.4$, $-16.6$ and $+28.8 \, \mathrm{MeV}$, respectively.
One of the reasons is that the hyperfine splitting in vacuum is smaller so that their mixing in a magnetic field becomes stronger: $\Delta m_{J/\psi-\eta_c}=113\mathrm{MeV}$ for the ground state and $\Delta m_{\psi(2S)-\eta_c(2S)}=47\mathrm{MeV}$ for the excited state \cite{Agashe:2014kda}.
Such a tendency can be easily inferred from a simplified two-level model as discussed in Refs.~\cite{Alford:2013jva,Cho:2014exa,Cho:2014loa} (also see Appendix.~\ref{App_EFT}).
Thus the mixing in the weak magnetic fields quantitatively agrees with the behavior in the two-level system while the behavior in intermediate magnetic fields becomes more complicated because of the mixing between various states.
Then the qualitative behavior of RMS radii is similar to the case of transverse $J/\psi$ as shown in Fig.~\ref{Results_charmT_r}.
At $eB=1.0$-$1.1 \, \mathrm{GeV}^2$, the third state approaches the second one like $J/\psi$.
After the crossing, the mixing rate of the state shows a $\psi$-component dominance, which means that the state is predominantly mixed with the first state like $\eta_c(1S)$.
After that, the state approaches the fourth one like $\eta_c(3S)$ at $eB=1.8$-$1.9 \, \mathrm{GeV}^2$ and the wave function changes to the $3S$-like shape with two nodes.
Finally, we also find that the {\it fourth} state starting from $\psi(2S)$ shows similar behavior, which approaches the fifth state like $\eta_c(3S)$ at $eB=0.6$-$0.7 \, \mathrm{GeV}^2$, the third state like $J/\psi$ and the fifth state like $\eta_c(4S)$ at $eB=2.4$-$2.5 \, \mathrm{GeV}^2$.
\begin{figure}[t!]
\centering
\includegraphics[clip, width=1.0\columnwidth]{bottomT_m}
\caption{Masses of transverse $\Upsilon (1S,2S,3S)$ in a magnetic field.}
\label{Results_bottomT_m}
\end{figure}
\begin{figure}[t!]
\centering
\includegraphics[clip, width=1.0\columnwidth]{bottomT_r}
\caption{RMS radii of transverse $\Upsilon (1S,2S,3S)$ in a magnetic field.}
\label{Results_bottomT_r}
\end{figure}
\begin{figure*}[t!]
\centering
\includegraphics[clip,width=17cm]{bottomL_m}
\caption{Masses of $\eta_b (1S,2S,3S)$ and longitudinal $\Upsilon (1S,2S,3S)$ in a magnetic field.}
\label{Results_bottomL_m}
\end{figure*}
Here, we emphasize that the quantitative determination of crossing magnetic field would be important to compare our prediction with observables from other theoretical approaches or experiments.
When we obtain hadron masses from QCD sum rules and lattice QCD simulations, we may evaluate a spectral function from the corresponding hadronic current-current correlator.
For example, in Fig.~\ref{SPfunc}, we show expected spectral functions of the longitudinal vector quarkonia.
Since $J/\psi$ mixes with $\eta_c$ components in a magnetic field, the spectral function obtained from the vector-vector current correlator in a magnetic field can have not only the ``original" $J/\psi$ and $\psi(2S)$ peaks (and continuum) but also ``additional" peaks such as $\eta_c$ and $\eta_c(2S)$ \cite{Cho:2014exa,Cho:2014loa}.
Then the level crossing in a magnetic field means that the positions of the two peaks in the spectral functions approach each other.
After that, the two peaks are expected not to slip thorough but rather to begin to repel each other by the avoided crossing.
Although it might be usually difficult to extract the higher-energy structures except for the ground state from QCD sum rules and lattice QCD simulations, phenomenological estimation as constructed by Refs.~\cite{Cho:2014exa,Cho:2014loa} and some numerical techniques such as the maximum entropy method \cite{Asakawa:2000tr,Gubler:2010cf,Araki:2014qya} could be useful to investigate excited states and crossing structures in hadronic spectral functions under finite magnetic field.
Furthermore, the spectral function for vector channels can be related to the experimental dilepton spectra (e.g. $J/\psi \to e^+ e^-$ in vacuum) and the modifications of such observables by mixing (crossing) effect under magnetic field would be more interesting.
\subsection{Bottomonia}
\begin{figure*}[t!]
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{bottomL_r}
\end{minipage}%
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{bottomL_mix}
\end{minipage}
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{bottomL_r2}
\end{minipage}%
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{bottomL_mix2}
\end{minipage}
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{bottomL_r3}
\end{minipage}%
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{bottomL_mix3}
\end{minipage}
\caption{RMS radii and mixing rates of $\eta_b (1S,2S,3S)$ and longitudinal $\Upsilon (1S,2S,3S)$ in a magnetic field.}
\label{Results_bottomL_r}
\end{figure*}
Next, we move to the discussions on bottomonia.
For pseudoscalar and vector bottomonia, there are six states below the $B\bar{B}$ threshold located at about $10.56 \, \mathrm{GeV}$: $\eta_b(1S)$, $\Upsilon(1S)$, $\eta_b(2S)$, $\Upsilon(2S)$, $\eta_b(3S)$, and $\Upsilon(3S)$.
The main differences between charmonia and bottomonia are (i) quark mass and (ii) quark electric charge.
For (i), since bottom quarks are heavier than charm quarks, the kinetic energies of the heavy quarks moving inside a quarkonia are suppressed and the sizes of the wave functions tend to be smaller than those of charmonia.
Furthermore, for (ii), the electric charge of bottom quarks is $|q|=(1/3)|e|$ while that of charm quarks is $|q|=(2/3)|e|$.
This means that bottomonia are less sensitive to a magnetic field than charmonia, as seen in the $(q^2 B^2/8\mu) \rho^2$ term in Eq.~(\ref{H_rel2}).
From the definition of the magnetic moment, $\bm{\mu}_i = g q_i \bm{S}_i/2m_i$, we expect that the mixing effect characterized by Eq.~(\ref{mat_ele_1}) also becomes smaller.
Results for the transverse vector bottomonia [$\Upsilon(1S)_{|1 \pm 1 \rangle}$, $\Upsilon(2S)_{|1 \pm 1 \rangle}$ and $\Upsilon(3S)_{|1 \pm 1 \rangle}$] are shown in Figs.~\ref{Results_bottomT_m} and \ref{Results_bottomT_r}.
We find that the mass shifts and the wave function deformation of bottomonia are less sensitive to a magnetic field than those of charmonia.
For instance, the mass shifts of $\Upsilon(1S)$, $\Upsilon(2S)$ and $\Upsilon(3S)$ at $eB=0.3\mathrm{GeV}^2$ are $+0.4$, $+2.3$ and $+5.0\mathrm{MeV}$.
These smaller mass shifts could be experimentally irrelevant in heavy-ion collisions.
Also, the shapes of the wave functions are almost spherical even at $eB=1.0 \ \mathrm{GeV}^2$.
We find that first the mass of the ground state increases loosely and finally changes to the linear increase by the quark Landau levels.
Such linear behavior can also be shown for the excited state in weaker magnetic fields.
For the RMS radii shown in Fig.~\ref{Results_bottomT_r}, while the radii of the ground state in both the $\rho$- and $z$-directions decease, for the excited states $2S$ and $3S$, we find the shrinkage in the $\rho$-plane and the expansion in the $z$-direction.
In particular, the radii of $\Upsilon(2S)$ and $\Upsilon(3S)$ in the $\rho$-plane approach to the same value as that of $\Upsilon(1S)$ because the excitation in the $\rho$-plane is removed by a magnetic field and only that in the $z$-direction survives.
From the result, in the strong field limit, we expect that the $\rho$-radii of all the states reach to the same saturation value for a magnetic field.
On the other hand, the excitations in the $z$-direction behave as a quasi-one-dimensional object with a confinement potential.
The results of the pseudoscalar [$\eta_b(1S)$, $\eta_b(2S)$ and $\eta_b(3S)$] and longitudinal vector [$\Upsilon(1S)_{|1 0 \rangle} $, $\Upsilon(2S)_{|1 0 \rangle}$ and $\Upsilon(3S)_{|1 0 \rangle}$] bottomonia are shown in Figs.~\ref{Results_bottomL_m} and \ref{Results_bottomL_r}.
For the {\it first} state starting from $\eta_b(1S)$ in vacuum, the modification in a magnetic field is less sensitive than that of the corresponding state in charmonium.
For instance, the mass shift and the mixing rate at $eB=0.3 \, \mathrm{GeV}^2$ are $-5.9 \, \mathrm{MeV}$ and $8\%$, respectively.
Such behavior is quantitatively consistent with the previous studies based on the same Hamiltonian in Refs.~\cite{Alford:2013jva,Bonati:2015dka}.
The mass of the {\it second} state starting from $\Upsilon(1S)$ increases gradually by the level repulsion and approaches to the third state at about $eB=5.5 \, \mathrm{GeV}^2$.
After the crossing, the wave function like the ground state whose $\sqrt{\langle z^2 \rangle} \simeq 0.2 \, \mathrm{fm}$ converts to the $2S$-excitation in the $z$-direction, of which $\sqrt{\langle z^2 \rangle} = 0.7 \, \mathrm{fm}$.
Also, the mixing rate having been $45\%$ changes to almost $50\%$.
The mass of the {\it third} state does not change almost up to $eB=5.5 \, \mathrm{GeV}^2$.
In this region, the wave function in the $z$-direction is gradually expanded.
After that, it approaches the second state like $\Upsilon (1S)$ and the third state like $\eta_c(3S)$ at $eB=8.5 \, \mathrm{GeV}^2$.
These crossing magnetic fields seem to be quite larger than the case of the charmonia, where the corresponding fields are $eB=1.0$-$1.1 \, \mathrm{GeV}^2$ and $eB=1.8$-$1.9 \, \mathrm{GeV}^2$.
The upward mass shifts of the {\it fourth} state starting from $\Upsilon(2S)$ in weak magnetic fields are larger than those of the first state.
The state approaches the fifth one like $\eta_c(3S)$, and, after the crossing at $eB=3.0 \, \mathrm{GeV}^2$, the mass is almost unchanged.
Moreover, after the crossing to the third state, the mass increases and will approach to the fifth state again.
The behavior of the {\it fifth} ({\it sixth}) state starting from $\eta_b(3S)$ [$\Upsilon(3S)$] is similar to the third (fourth) state.
Then we stress that the $B\bar{B}$ threshold in vacuum is located at about $10.56 \, \mathrm{GeV}$, so that the modifications of bottomonia up to $\Upsilon(3S)$ in the weak magnetic field can be estimated without the threshold effect.
In a stronger magnetic field, we need to take into account the competition with threshold effects related to the magnetic field dependences of $B$ mesons as shown in the next section.
\subsection{Heavy-light mesons}
In this section, we investigate the neutral heavy-light mesons: $D$, $B$ and $B_s$ mesons.
As with quarkonia, heavy-light meson systems exhibit the mixing between $| 00 \rangle$ and $| 10 \rangle$ components \cite{Machado:2013rta,Gubler:2015qok}.
In addition, these mesons consist of one heavy quark and one light antiquark and have a nonzero magnetic moment, which induces the Zeeman splitting between the transverse components of the spin triplet ($| 1 +1 \rangle$ and $| 1 -1 \rangle$) in a magnetic field.
As a result, the three states of the spin triplet, which degenerate in vacuum, are completely separated in a magnetic field.
It is important that these mesons include one (constituent) light quark so that one of the origins of such a constituent quark mass should come from the chiral symmetry breaking.
Since a magnetic field enhances the chiral symmetry breaking by the magnetic catalysis, we may actually take into account the effect in our analyses for mesons with a light quark.
In this section, we show meson systems from the model {\it without} the magnetic catalysis.
Analyses with the magnetic catalysis will be discussed in the next section.
\begin{figure}[t!]
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{Dmeson_m}
\end{minipage}
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{Dmeson_r}
\end{minipage}
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{Dmeson_mix}
\end{minipage}
\caption{Masses, RMS radii and mixing rates of neutral $D$ and $D^\ast$ mesons in a magnetic field.}
\label{Results_D}
\end{figure}
\begin{figure}[t!]
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{Bmeson_m}
\end{minipage}
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{Bmeson_r}
\end{minipage}
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{Bmeson_mix}
\end{minipage}
\caption{Masses, RMS radii and mixing rates of neutral $B$ and $B^\ast$ mesons in a magnetic field.}
\label{Results_B}
\end{figure}
\begin{figure}[t!]
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{Bsmeson_m}
\end{minipage}
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{Bsmeson_r}
\end{minipage}
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{Bsmeson_mix}
\end{minipage}
\caption{Masses, RMS radii and mixing rates of neutral $B_s$ and $B_s^\ast$ mesons in a magnetic field.}
\label{Results_Bs}
\end{figure}
Results for $D$, $B$ and $B_s$ mesons are shown in Figs.~\ref{Results_D}, \ref{Results_B}, and \ref{Results_Bs}, respectively.
In a weak magnetic field, the transverse vector heavy-light mesons ($D^\ast_{|1 \pm1 \rangle}$, $B^\ast_{|1 \pm1 \rangle}$, and $B^\ast_{s |1 \pm1 \rangle}$) are affected by the Zeeman splitting induced by the matrix element Eq.~(\ref{mat_ele_2}) and $s_z=\pm1$ states which degenerate in vacuum are split.
Namely, the masses of upper states (symbolized as green points) increase and those of lower states (symbolized as orange points) decrease.
In stronger magnetic fields, the contribution from quark Landau levels becomes larger and comparable with that from the Zeeman splitting.
The masses of the upper states continue to increase.
On the other hand, in the masses of the lower states, the quark Landau levels balance with the Zeeman splitting and the masses begin to increase at a ``critical'' magnetic field, at $eB=0.6$, $5.0$, $ 6.0 \, \mathrm{GeV}^2$ for $D^\ast$, $B^\ast$, and $B^\ast$, respectively.
$s_z=\pm1$ states have the same RMS radii because the operator $-\bm{\mu}_i \cdot \bm{B}$ only contributes to spin wave functions and does not affect to spatial wave functions.
From the middle figures in Figs.~\ref{Results_D}, \ref{Results_B}, and \ref{Results_Bs}, we find that the RMS radii of the ground states of transverse vector heavy-light mesons continue to decrease.
Pseudoscalar ($D$, $B$, $B_s$) and longitudinal vector ($D^\ast_{|1 0 \rangle}$, $B^\ast_{|1 0 \rangle}$, $B^\ast_{s|1 0 \rangle}$) mesons under magnetic field are mixed with each other, which is similar to quarkonia.
Also, we can see the similar spectroscopic structures including the level crossing.
Furthermore, the mass shifts of the upper states of the transverse vector mesons (green points) and the longitudinal vector mesons (blue points) almost agree with each other in weak fields.
This behavior can be simply understood by considering the form of
the matrix elements of Eqs.~(\ref{mat_ele_1}) and (\ref{mat_ele_2}) near the heavy quark limit ($m_2 \to \infty$).
In the limit, the diagonal component by the $-\mu\cdot B$ term in transverse vector mesons agrees with
the off-diagonal components in the longitudinal vector mesons.
Therefore, as shown by the simplified two-level model, the mass increase by the mixing is the same as
that by the Zeeman splitting.
For all the heavy meson channels, the evaluated ``level crossing magnetic fields'' are summarized in Table \ref{crossing}.
Here, it is interesting that we compare $D$ ($B$ and $B_s$) mesons with charmonia (bottomonia), where the quark electric charges are the same and the only difference is one light quark mass.
The crossing field in the $D$ ($B$ and $B_s$) meson channel is about $3$ ($8$ and $3$) times smaller than in the charmonium (bottomonium) channel.
These ratios are almost consistent with the strength of the matrix element Eq.~(\ref{mat_ele_1}) which is proportional to the first order of $|eB|$ [$\propto |eB|(1/m_1 +1/m_2)$].
\begin{table*}[t!]
\begin{center}
\caption{List of level crossing magnetic fields for pseudoscalar-vector mixing states in heavy mesons in a magnetic field, which is evaluated by the constituent quark model, Eq.~(\ref{H_rel2}), and CGEM.
For heavy-light mesons, contributions from the magnetic catalysis are {\it not} taken into account.}
\label{crossing}
\begin{tabular}{l|cccccc}
\hline \hline
Charmonium & $J/\psi$-$\eta_c(2S)$ & $\eta_c(2S)$-$\psi(2S)$ & $\psi(2S)$-$\eta_c(3S)$ & & &\\
\hline
Crossing $B$-field [GeV$^2$] & $1.0$-$1.1$ & $1.8$-$1.9$ & $0.6$-$0.7$, $2.4$-$2.5$ & & \\
\hline \hline
Bottomonium & $\Upsilon(1S)$-$\eta_b(2S)$ & $\eta_b(2S)$-$\Upsilon(2S)$ & $\Upsilon(2S)$-$\eta_b(3S)$ & $\eta_b(3S)$-$\Upsilon(3S)$ & $\Upsilon(3S)$-$\eta_b(4S)$ \\
\hline
Crossing $B$-field [GeV$^2$] & $5.5$ & $8.5$ & $3.0$ & $5.0$ & $2.5$, $6.5$ \\
\hline \hline
Heavy-light meson & $D^\ast(1S)$-$D(2S)$ & $B^\ast(1S)$-$B(2S)$ & $B_s^\ast(1S)$-$B_s(2S)$& &\\
\hline
Crossing $B$-field [GeV$^2$] & $0.3$-$0.4$ & $0.7$-$0.8$ & $1.7$-$1.8$ & & \\
\hline \hline
\end{tabular}
\end{center}
\end{table*}
This can be checked by a replacement of one quark mass, $m_c \to m_q$ ($m_b \to m_q$ or $m_s$).
Furthermore, RMS radii and mixing rates of $D$ ($B$) mesons are more sensitive than the corresponding charmonia (bottomonia).
As with the RMS radii of quarkonia, we also found the increases of radii of the first states in smaller magnetic fields.
For the second states, we do not see such behavior.
We obtained the mixing rates of $39\%$, $43\%$ and $36\%$ at $eB=0.3 \, \mathrm{GeV}^2$ for $D$, $B$, and $B_s$ mesons, respectively.
\subsection{Magnetic catalysis in heavy-light mesons} \label{SubSec_MC}
As shown in Ref.~\cite{Park:2016xrw}, in the constituent quark model, the chiral symmetry restoration (or the reduction of $\langle \bar{q}q \rangle$) in the nuclear medium corresponds to smaller values of the constituent quark mass, which leads to an {\it increase} of the mass of a heavy-light meson.
Such a picture for $D$ meson masses at finite density can be understood \cite{Park:2016xrw} by a balance between a decrease of the constituent quark mass, $m_q$, and an increase of the kinetic and potential energies, $E_K+E_V$, which is consistent with the results from QCD sum rules \cite{Hilger:2008jg,Suzuki:2015est} and some phenomenological approaches \cite{Blaschke:2011yv,Sasaki:2014asa,Suenaga:2014sga}.
In contrast, in a magnetic field, the chiral symmetry breaking is enhanced by the magnetic catalysis and the constituent light quark mass should increase.
Therefore, we may naively expect that the mass of a heavy-light meson {\it decreases} with increasing magnetic field.
In this section, we investigate the light quark mass dependence of the analyses of the last section.
To introduce the $B$-field dependence of a constituent quark mass, we adopt the chiral condensate estimated from the two-flavor NJL model \cite{Boomsma:2009yk} as shown in Appendix \ref{App_NJL}, where we input the differences from the value in vacuum, $\langle \bar{q} q \rangle_{eB} - \langle \bar{q} q \rangle_{\mathrm{vac}}$. Here, the values of $\langle \bar{u} u \rangle$ and $\langle \bar{d} d \rangle$ are distinguished from each other and the instanton interaction is neglected.
Therefore, for analyses of neutral $D$ mesons composed of charm and up quarks, we use the values of $\langle \bar{u} u \rangle$.
The $D$-meson masses with the $B$-field dependence of the light quark mass are shown in Fig.~\ref{Results_D_MC}.
In this figure, the dashed lines correspond to the masses taking into account the magnetic catalysis.
The light quark mass can contribute to the two components: (i) the kinetic and potential energies $E_K+E_V$, and (ii) the magnetic-moment given by $\bm{\mu}_q \equiv g q_q \bm{S}_q/2m_q$.
As explained above, for (i), $E_K+E_V$ in a heavy-light meson decreases with increasing $m_q$.
On the other hand, for (ii), the magnetic moment of the light quark becomes smaller as $m_q$ increases, so that the strength of the spin mixing (or the Zeeman splitting) is weakened. [See also the form of the matrix elements, Eqs.~(\ref{mat_ele_1}) and (\ref{mat_ele_2})].
As a result, we find that the pseudoscalar $D$-meson mass is almost unchanged because of the cancellation between the mass decrease by (i) and the mass increase from the weaker level repulsion based on (ii).
The effect of (i) can also be applied to $D^\ast$ mesons.
The mass of the $| 10 \rangle$ component of $D^\ast$ decreases by the magnetic catalysis because of both the mass decreases by (i) and the smaller level repulsion.
Similarly, for the $| 1 \pm1 \rangle$ components of $D^\ast$, we can find the mass shifts which can be understood by the mass decrease by (i) and the reduction of the Zeeman splitting.
\begin{figure}[t!]
\begin{minipage}[h]{1.0\columnwidth}
\centering
\includegraphics[clip, width=1.0\columnwidth]{Dmeson_m_withMC}
\end{minipage}
\caption{Light quark mass dependences of masses of $D$ and $D^\ast$ mesons in a magnetic field up to $eB=0.3 \rm{GeV}^2$.
Dashed (solid) lines correspond to mass with (without) the magnetic catalysis (MC).
The solid lines are the same as the lines in Fig.~\ref{Results_D}.}
\label{Results_D_MC}
\end{figure}
We note that the results in this section may theoretically be important in the sense that, in some previous works, a hadron effective model for $D$ mesons was used \cite{Gubler:2015qok} and such a model does not include contributions from the magnetic catalysis.
If we would like to modify this model to include contributions from the magnetic catalysis, our current results imply necessities of quantitative corrections for $D^\ast$ mesons.
Furthermore, the previous analyses by QCD sum rules in Refs.~\cite{Machado:2013yaa, Gubler:2015qok} have taken into account the magnetic catalysis of the $\langle \bar{q} q \rangle$ condensate.
We additionally mention that hadron masses from QCD sum rules in a magnetic field can be contributed from not only $\langle \bar{q} q \rangle$ but also other condensates (e.g. $\langle \bar{q} \sigma_{\mu \nu} q \rangle$ as included in Ref.~\cite{Gubler:2015qok}) more or less, so the relation of such condensates to the quark model would be one of the future issues.
\section{Conclusion and outlook} \label{Sec_Conclusion and outlook}
In this paper, we investigated properties of heavy meson (charmonia, bottomonia, neutral $D$, $B$, and $B_s$ mesons) spectra in a magnetic field.
We developed the method to solve the two-body Schr\"{o}dinger equation for a quark model in the cylindrical coordinate.
By using this approach, our results can include full contributions from higher excited states.
As a result, we found mass shifts and deformation of wave functions in heavy meson systems in a magnetic field.
For transverse vector quarkonia [$J/\psi_{|1 \pm1 \rangle}$, $\psi(2S)_{|1 \pm1 \rangle}$, $\Upsilon(1S)_{|1 \pm1 \rangle}$, $\Upsilon(2S)_{|1 \pm1 \rangle}$, and $\Upsilon(3S)_{|1 \pm1 \rangle}$], we found that there are mass shifts although these hadrons have no charge.
On the other hand, for pseudoscalar and longitudinal vector charmonia [$\eta_c(1S)$, $J/\psi_{|1 0 \rangle}$, $\eta_c(2S)$, and $\psi(2S)_{|1 0 \rangle}$] and bottomonia [$\eta_b(1S)$, $\Upsilon_{|1 0 \rangle}(1S)$, $\eta_b(2S)$, $\Upsilon(2S)_{|1 0 \rangle}$, $\eta_b(3S)$, and $\Upsilon(3S)_{|1 0 \rangle}$], the spin partners (spin-0 and spin-1) mix with each other, so that we obtained the level structures with the level repulsion.
For the transverse vector heavy-light mesons ($D^\ast_{|1 \pm1 \rangle}$, $B^\ast_{|1 \pm1 \rangle}$, and $B^\ast_{s |1 \pm1 \rangle}$), we discussed mass shifts by Zeeman splitting.
For the pseudoscalar and longitudinal vector heavy-light mesons ($D$, $D^\ast_{|1 0 \rangle}$, $B$, $B^\ast_{|1 0 \rangle}$, $B_s$, $B^\ast_{s|1 0 \rangle}$), we examined the level structures induced by the mixing.
Moreover, for $D$ meson masses, the modification of the constituent quark mass based on the magnetic catalysis is discussed.
Here, we stress that heavy-light mesons under magnetic field can be useful for a probe of the magnetic catalysis.
In association with the magnetic catalysis, we can also expect hadron modifications from the gluonic magnetic catalysis \cite{Bali:2013esa,Ozaki:2013sfa} because properties of quarkonia are partially attributed to the gluon condensate instead of the chiral condensate.
It is worthwhile to note possible corrections in our approach: (i) {\it $B$-dependences of the model parameters}, (ii) {\it relativistic corrections}, and (iii) {\it threshold and continuum effects}.
(i) As stated in Sec. \ref{subsec_parameter}, in this work, only the $B$-dependences of the constituent light quark mass $m_q$ for $D$-meson channels were investigated.
The anisotropic linear and Coulomb potentials were evaluated from Wilson loops in a recent lattice QCD simulation \cite{Bonati:2014ksa}.
The applications of such potentials to the quark model are interesting, as performed in Ref.~\cite{Bonati:2015dka}.
For example, because, in Ref.~\cite{Bonati:2014ksa}, the linear potential becomes larger (smaller) in the $\rho$-direction ($z$-direction), we can naively expect the squeezing of wave functions in the $\rho$-direction to become more drastic than the current results.
In addition, the flavor dependence of the anisotropic potential and the $B$-dependences of the spin-spin interaction will also be discussed in future works.
(ii) Our approach leads to the physics in a nonrelativistic limit and its validity should be discussed.
The applicable region of the approach could be naively estimated from a ratio of the constituent quark mass $m$ and the strength of magnetic field (e.g. $eB/m^2 \ll 1$).
In strong fields beyond the applicable region, one has to consider the connection with the relativistic approach.
(iii) Some of the excited states investigated in this work are located near the continuum threshold, so that we need the treatments of threshold (or continuum) effects in a magnetic field.
Some possibilities of charmonium continuum in a magnetic field were discussed in Ref.~\cite{Cho:2014loa}.
As more realistic situations in heavy-ion collisions, we can investigate hadron properties at finite temperature (and/or density) in a magnetic field.
In the constituent quark model, a temperature (density) dependence can be introduced as changes of model parameters and functional forms.
A representative example is the potential model approach for $J/\psi$ suppression near the critical temperature \cite{Matsui:1986dk}.
The approach can be simply extended to thermal quarkonia in a magnetic field.
In addition, as discussed in this work, $D$ meson mass shifts are expected to be a probe of the magnetic catalysis so that studying the temperature (or density) dependence of the $\langle \bar{q} q \rangle$ condensate would be interesting \cite{Klimenko:1991he,Agasian:2001ym,Ebert:2003yk,Inagaki:2003yi,Agasian:2008tb,Fraga:2008qn,Menezes:2008qt,Boomsma:2009yk,Menezes:2009uc,Fukushima:2010fe,Mizher:2010zb,Gatto:2010qs,Fayazbakhsh:2010bh,Gatto:2010pt,Preis:2010cq,Chatterjee:2011ry,Frasca:2011zn,Rabhi:2011mj,Kashiwa:2011js,Andersen:2011ip,Skokov:2011ib,Andersen:2012dz,Fukushima:2012xw,Andersen:2012bq,Fayazbakhsh:2012vr,Fayazbakhsh:2013cha,Ballon-Bayona:2013cta,Andersen:2013swa,Fraga:2013ova,Tawfik:2014hwa,Andersen:2014xxa,Tawfik:2015tga,Mamo:2015dea,Hattori:2015aki} and would also be related to the experimental explorations of the inverse magnetic catalysis scenarios (e.g. see Refs. \cite{Fukushima:2012kc,Kojo:2012js,Bruckmann:2013oba,Chao:2013qpa,Ferreira:2013tba,Kamikado:2013pya,Farias:2014eca,Ferreira:2014kpa,Yu:2014sla,Ayala:2014iba,Ayala:2014gwa,Ferrer:2014qka,Andersen:2014oaa,Yu:2014xoa,Mueller:2015fka}) observed by lattice QCD simulations \cite{Bali:2011qj,Bali:2012zg,Bali:2013esa,Bruckmann:2013oba,Bornyakov:2013eya,Bali:2014kia}.
Our results in the weak magnetic fields are useful to predict the observables in a magnetic field induced by heavy-ion collisions, such as the $\eta_c + \gamma \to J/\psi$ transition rate \cite{Yang:2011cz}, production cross section of quarkonia \cite{Machado:2013rta}, anisotropic production \cite{Guo:2015nsa} and quarkonium melting \cite{Marasinghe:2011bt,Dudal:2014jfa}.
In particular, the dilepton spectra constituting decay modes such as $J/\psi, \psi^\prime, \Upsilon \to e^+ e^-$ can shift by a magnetic field and can exhibit a splitting between the transverse and longitudinal components \cite{Alford:2013jva,Filip:2013xza}.
Simultaneously, ``anomalous'' channels such as $\eta_c, \eta_b \to e^+ e^-$ can be opened due to the mixing with $J/\psi$ components, and these would be the signal of a magnetic field.
Recently, it was suggested that the isospin (QCD) Kondo effect in heavy-light mesons (heavy quarks) at finite density with a Fermi surface can occur and may modify transport properties of the systems \cite{Yasui:2013xr,Hattori:2015hka,Yasui:2016ngy,Yasui:2016svc} (see a part of Ref.~\cite{Hosaka:2016ypm} for a review).
Under the existence of a magnetic field, the magnetically induced Kondo effect can also be predicted \cite{Ozaki:2015sya}.
Properties of heavy quark systems with such effects are one of the more promising topics in heavy hadron physics.
\begin{acknowledgments}
The authors gratefully thank Makoto Oka, Sho Ozaki, Aaron Park and Shigehiro Yasui for fruitful discussions. K.S. was partially supported by a Grant-in-Aid for JSPS Fellows from the Japan Society for the Promotion of Science (JSPS) (Grant No. 26-8288). T.Y. acknowledges the Junior Research Associate scholarship
at RIKEN.
\end{acknowledgments}
| 93c7a32c05cd3f3efec74522e15aebb538758ce3 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}\label{sec:introduction}
Quantitative investigations in logic, where properties and behaviour of typical objects are studied, form a rich and well-established branch of mathematics on the border of logic, combinatorics and theoretical computer science. From a combinatorial point of view, logical formulae are objects with finite representations and, sometimes several, naturally associated notions of size. In cases when the assumed size notion imposes finitely many objects of any size, we can consider uniformly random formulae within such a family of objects. Analysing the asymptotic behaviour of the probability that a uniformly random object of size $n$ satisfies a certain property~$P$ as $n$ tends to infinity, yields the notion of asymptotic density of $P$ and, in consequence, leads to investigations how certain natural properties, such as satisfiability etc., behave in the case of typical formulae.
There is a long history of using this kind of asymptotic approach applied to logic (see, e.g.~\cite{mtz00, kos-zaionc03, FGGZ07, GK2012}) and recently its computational aspects. In~\cite{dgkrtz}, David et al. initiated quantitative investigations in lambda calculus and combinatory logic. Considering the set of closed $\lambda$-terms in a canonical representation where variables do not contribute to the term size, David et al. showed that typical $\lambda$-terms are strongly normalising whereas in the case of $\S \mathbf{K}$-combinators the situation is precisely opposite -- asymptotically almost no $\S \mathbf{K}$-combinator is strongly normalising. Somewhat contrary to their result, Bendkowski et al. in~\cite{Bendkowski16} considered a different representation of $\lambda$-terms with de~Bruijn indices, showing that similarly to the case of $\S \mathbf{K}$-combinators, asymptotically almost no $\lambda$-term is strongly normalising.
Despite many efforts, the associated counting problem for closed $\lambda$-terms is still one of the remaining major open problems. Throughout the years, different variants of lambda calculus have been considered. In~\cite{BGGJ2013}, Bodini et al. studied the enumeration of BCI $\lambda$-terms. John Tromp in~\cite{T2014}, as well as Grygiel and Lescanne in~\cite{JFP:10090684}, considered the counting problem in the so called binary lambda calculus. Recently, in~\cite{gittenberger_et_al:LIPIcs:2016:5741} Gittenberger and Gołębiewski considered $\lambda$-terms in the de~Bruijn notation with various size notions, giving tight lower and upper asymptotic bounds on the number of closed $\lambda$-terms. Due to the lack of bound variables, combinatory logic circumvents the intrinsic issues present in lambda calculus, serving as a minimalistic formalism capable of expressing all lambda-definable functions. Its simple syntax yields a natural representation using plane binary trees with labelled leaves, which greatly facilitates the analysis of asymptotic properties.
The paper is organised as follows. In~\autoref{sec:combinatory-logic} we recall preliminary definitions and notation regarding combinatory logic. In the following sections~\ref{sec:combinatorial-classes} and~\ref{sec:generating-functions-and-analytic-tools}, we state basic notions of combinatorial classes and generating functions, listing main tools used to study asymptotic properties of combinators, in particular, the analytic methods of singularity analysis. In~\autoref{sec:basis-independent-results} we study the class of plane binary trees with labelled leaves, deriving basis-independent combinatory logic results as easy corollaries. In~\autoref{sec:sk-combinators} we focus on the $\S \mathbf{K}$-basis proving, inter alia, that for each positive $n$, the set of combinators reducing in $n$ normal-order reduction steps has positive asymptotic density. Finally, in~\autoref{sec:experimental-results} we discuss some experimental results.
\section{Combinatory logic}\label{sec:combinatory-logic}
Let $\mathcal{B}$ be a finite \emph{basis} of primitive combinators. The set $\class{\mathcal{B}}$ of \emph{$\mathcal{B}$-combinators} is defined inductively as follows. If $\comb{X} \in \mathcal{B}$, then $\comb{X} \in \class{\mathcal{B}}$. If $N, M \in \class{\mathcal{B}}$, then $(N M) \in \class{\mathcal{B}}$. In the latter case, we say that $(N M)$ is an \emph{application} of $N$ to $M$. If the underlying basis is clear from the context, we simply write \emph{combinators} instead of $\mathcal{B}$-combinators. Following standard notational conventions (see, e.g.~\cite{BAR84}), we omit outermost parentheses and drop parentheses from left-associated combinators, e.g.~instead of writing $((M N) (P Q))$ we write $M N (P Q)$.
Each primitive combinator $\comb{X} \in \mathcal{B}$ contributes a \emph{reduction rule} of the form
\begin{equation*}
\comb{X} N_1 \ldots N_m \to M
\end{equation*}
where $m \geq 1$, $M \in \class{\set{N_1, \ldots, N_m}}$ and $N_1 \ldots N_m$ are arbitrary combinators. In other words, $\comb{X} N_1 \ldots N_m$ \emph{reduces} to some determined combinator $M$ built from $N_1,\ldots,N_m$ and term application.
The reduction relation $\to$ is then extended onto all combinators such that if $P \to Q$, then $PR \to QR$ and $RP \to RQ$ for each combinator $R$.
Let $P$ be a combinator. If there exists no combinator $Q$ such that $P \to Q$, then $P$ is said to be in \emph{normal form}. If there exists a finite sequence of combinators $P_0,P_1,\ldots,P_k$ such that $P = P_0 \to P_1 \to \ldots \to P_k$ and $P_k$ is in normal form, then $P$ is \emph{weakly normalising}, or simply \emph{normalising}. If there does not exist an infinite sequence of combinators $P_0,P_1,\ldots$ such that $P=P_0 \to P_1 \to \ldots$, then we say that $P$ is \emph{strongly normalising}. Naturally, strong normalisation implies weak normalisation.
A major part of combinatory logic is devoted to its simple type theory corresponding to the implicational fragment of minimal logic (see, e.g.~\cite{BAR84}) and hence also to simply typed $\lambda$-calculus (see \cite{Hindley}). In its most common basis $\mathcal{B} = \set{\comb{S},\comb{K}}$, the type-assignment deduction system $TA_{\comb{S} \comb{K}}$ is given by the following two axiom schemes for $\comb{S}$ and $\comb{K}$, with a single \emph{modus ponens} inference rule.
\begin{equation}\tag{Axiom $\comb{S}$}
\begin{mathprooftree}
\AxiomC{}
\UnaryInf$\comb{S} : (\alpha \to \beta \to \gamma) \to (\alpha \to \beta) \to \alpha \to \gamma \fCenter$
\end{mathprooftree}
\end{equation}
\begin{equation}\tag{Axiom $\comb{K}$}
\begin{mathprooftree}
\AxiomC{}
\UnaryInf$\comb{K} : \alpha \to \beta \to \alpha\fCenter$
\end{mathprooftree}
\end{equation}
\begin{equation}\tag{Modus ponens}
\begin{mathprooftree}
\AxiomC{$N : \alpha \to \beta$}
\AxiomC{$M : \alpha$}
\BinaryInf$(N M) : \beta\fCenter$
\end{mathprooftree}
\end{equation}
The primitive combinators $\comb{S}$ and $\comb{K}$ can be assigned any types fitting to their axiom schemes. On the other hand, an application $(N M)$ can be assigned a type $\beta$, denoted $(N M) : \beta$, if and only if $N : \alpha \to \beta$, whereas $M : \alpha$. If $P : \alpha$ for some type $\alpha$, we say that $P$ is \emph{typeable}. Naturally, not every combinator is typeable, e.g.~$\omega := \comb{S} \comb{I} \comb{I}$, where $\comb{I} := \comb{S} \comb{K} \comb{K}$. Though not all strongly normalising combinators are typeable, the converse implication holds, i.e.~each typeable combinator is strongly normalising.
In this paper, we focus mostly on bases which are capable of expressing all computable functions. A sufficient and necessary condition for a basis $\mathcal{B}$ to be \emph{Turing-complete} is to span $\mathcal{B}$-combinators extensionally-equivalent to $\comb{S}$ and $\comb{K}$. In such a case, we make the natural assumption that $\mathcal{B}$ defines a set of axiom schemes, one for each primitive combinator, in such a way that when enriched with the modus ponens inference rule, $TA_{\mathcal{B}}$ constitutes a sound typing system -- if $N$ is extensionally-equivalent to either $\comb{S}$ or $\comb{K}$, then the types of $N$ in $TA_{\mathcal{B}}$ are the same as the types of, respectively, $\comb{S}$ or $\comb{K}$ in $TA_{\comb{S} \comb{K}}$. Throughout the paper, we use the over line notation $\overline{\comb{X}}$ to denote arbitrary terms extensionally-equivalent to $\comb{X}$.
We refer the reader to~\cite{curry-feys} or~\cite{BAR84} for a more detailed exposition of combinatory logic.
\section{Combinatorial classes}\label{sec:combinatorial-classes}
Let $B$ be a countable set of objects with an associated \emph{size} function $f \colon B \to \mathbb{N}$. If for each $n \in \mathbb{N}$ the set of $B$'s objects of size $n$ is finite, then $B$ together with $f$ forms a \emph{combinatorial class} (see, e.g.~\cite{FlajoletSedgewick2009}). In such a case, we can associate a \emph{counting sequence} $\set{b_n}_{n \in \mathbb{N}}$ of natural numbers $b_n$ capturing the number of objects in $B$ of size $n$. Naturally, if $A \subseteq B$, then $A$ is a combinatorial class as well with $a_n \leq b_n$ for each $n \in \mathbb{N}$. Assuming that $b_n > 0$ for each $n\in \mathbb{N}$, we can then define the \emph{asymptotic density} $\density{A}{B}$ of $A$ in $B$ as
\begin{equation*}
\density{A}{B} = \lim_{n\to\infty} \frac{a_n}{b_n}.
\end{equation*}
Note that if it exists, we can interpret $\density{A}{B}$ as the asymptotic probability of finding an object of $A$ in the class of objects $B$. In other words, the likelihood that $A$ represents `typical' objects in $B$.
Unfortunately, sometimes we do not know whether the asymptotic density of $A$ in $B$ exists, however we can use the \emph{lower} and \emph{upper limits} defined as
\begin{equation*}
\densityminus{A}{B} = \liminf_{n\to\infty} \frac{a_n}{b_n} \quad \text{and} \quad \densityplus{A}{B} = \limsup_{n\to\infty} \frac{a_n}{b_n}.
\end{equation*}
As $0 \leq \frac{a_n}{b_n} \leq 1$, these two numbers are well defined for any set $A$, even when the limiting ratio $\density{A}{B}$ is not known to exist.
Henceforth, given a combinatorial class $A$, we use $a_n$ to denote the number of objects in $A$ of size $n$.
\section{Generating functions and analytic tools}\label{sec:generating-functions-and-analytic-tools}
Let $A$ be a combinatorial class. The formal power series $A(z) = \sum_{n \geq 0} a_n z^n$ with $A$'s counting sequence $\set{a_n}_{n\in \mathbb{N}}$ as coefficients, is called the \emph{ordinary generating function} of $A$. Using the powerful theory of \emph{Analytic Combinatorics} developed by Flajolet and Sedgewick~\cite{FlajoletSedgewick2009}, many questions concerning the asymptotic behaviour of $\set{a_n}_{n \in \mathbb{N}}$ can be efficiently resolved by analysing the behaviour of $A(z)$ viewed as an analytic function in some neighbourhood around the complex plane origin. This is the approach we take to study the asymptotic fractions of interesting combinatory logic terms.
Throughout the paper we use $A(z)$ to denote the ordinary generating function associated with the combinatorial class $A$. We write $[z^n]A(z)$ to denote the coefficient standing by $z^n$ in the Taylor series expansion of $A(z)$ around $z = 0$. We say that two sequences $\set{a_n}_{n \in \mathbb{N}}$ and $\set{b_n}_{n \in \mathbb{N}}$ are \emph{asymptotically equivalent} if $\lim_{n \to \infty} \frac{a_n}{b_n} = 1$. In such a case we write $a_n \sim b_n$.
\subsection{Main tools}\label{subsec:main-tools}
In our endeavour to study the asymptotic behaviour of `typical' classes of combinatory logic terms, we use the method of \emph{singularity analysis}~\cite{FlajoletSedgewick2009}. Starting with a particular class $A$ of combinators, we find its corresponding generating function $A(z)$. The location of $A(z)$'s dominant singularities determines the exponential growth rate of $\set{a_n}_{n \in \mathbb{N}}$ as dictated by the following theorem.
\begin{theorem}[Exponential Growth Formula, see {\cite[Theorem IV.7]{FlajoletSedgewick2009}}]\label{thm:exponential-growth-formula}
If $A(z)$ is analytic at $0$ and $R$ is the modulus of a singularity nearest to the origin in the sense that
\[ R = \sup \{ r \geq 0 ~:~ A(z) \text{ is analytic in } |z| < r \} ,\]
then the coefficient $a_n = [z^n] A(z)$ satisfies
\[ a_n = R^{-n} \theta(n) \quad \text{with} \quad \limsup |\theta(n)|^{\frac{1}{n}} = 1 .\]
\end{theorem}
In the case of analytic functions derived from combinatorial classes, the location of dominant singularities is significantly simplified as it suffices to look for singularities on the real line.
\begin{theorem}[Pringsheim, see {\cite[Theorem~IV.6]{FlajoletSedgewick2009}}]
\label{th:pringsheim}
If $A(z)$ is representable at the origin by a series expansion that has non-negative
coefficients and radius of convergence $R$, then the point $z = R$ is a singularity of $A(z)$.
\end{theorem}
The sub-exponential factors determining the asymptotic growth rate of $\set{a_n}_{n \in \mathbb{N}}$ can be then further established using, in our case, the standard function scale for algebraic singularities of square-root type.
\begin{theorem}[Standard function scale, see~{\cite[Theorem~VI.1]{FlajoletSedgewick2009}}]\label{th:standard-func-scale}
Let $\alpha \in \mathbb{C} \setminus \mathbb{Z}_{\leq 0}$. Then $f(z) = {(1 - z)}^{-\alpha}$ admits for large $n$ a complete asymptotic expansion in form of
\begin{equation*}
[z^n]f(z) = \frac{n^{\alpha-1}}{\Gamma(\alpha)} \left( 1 + \frac{\alpha(\alpha-1)}{2n} + \frac{\alpha(\alpha-1)(\alpha-2)(3\alpha-1)}{24n^2} + O(\frac{1}{n^3}) \right)
\end{equation*}
where $\Gamma$ is the Euler Gamma function.
\end{theorem}
\begin{theorem}[Newton-Puiseux, see~{\cite[Theorem~VII.7]{FlajoletSedgewick2009}}]
Let $f(z)$ be a branch of an algebraic function $P(z, f(z)) = 0$. Then in a circular neighbourhood of a singularity $\zeta$ slit along a ray emanating from $\zeta$, $f(z)$ admits a fractional series expansion that is locally convergent and of the form
\begin{equation*}
f(z) = \sum_{k \geq k_0} c_k {\left( z - \zeta \right)}^{\nicefrac{k}{\kappa}}
\end{equation*}
where $k_0 \in \mathbb{Z}$ and $\kappa \geq 1$.
\end{theorem}
Finally, combining the scaling rule for Taylor expansions and the Newton-Puiseux expansion of algebraic functions with unique dominating singularities, we obtain the following corollary theorem.
\begin{theorem}[Algebraic singularity analysis]\label{th:singularity-analysis}
Let $f(z) = {(1 - \zeta^{-1} z)}^{\nicefrac{1}{2}} g(z) + h(z)$ be an algebraic function, analytic at $0$, having a unique dominant singularity $z = \zeta$. Assume that $g(z)$ and $h(z)$ are analytic in the disk $|z| < \zeta + \eta$ for some $\eta > 0$. Then the coefficient $[z^n]f(z)$ satisfies the following asymptotic approximation
\begin{equation*}
[z^n]f(z) \sim \zeta^{-n} \frac{\overline{C} n^{-\nicefrac{3}{2}}}{\Gamma(-\frac{1}{2})}
\end{equation*}
where $\overline{C}$ is the coefficient standing by $\sqrt{1 - \zeta^{-1} z}$ in the Newton-Puiseux expansion of $f(z)$, i.e.~$g(\zeta)$.
\end{theorem}
\begin{proof}
See~\cite{FlajoletSedgewick2009}, Theorem VII.8.
\end{proof}
In order to simplify the reasoning about the type and location of singularities of generating functions given without explicit closed-form solutions, we use the following technical lemma guaranteeing certain natural closure properties of analytic functions with a single square-root type dominating singularity.
\begin{lemma}\label{lem:sqrt-commutative-ring}
Let $\Omega$ be the open disk $|z| < \zeta + \eta$ for some $0 < \zeta < 1$ and $\eta > 0$. Let $F$ denote the set of functions $f \colon (0,\zeta) \to \mathbb{C}$ in form of $f(z) = \sqrt{1-\zeta^{-1} z}\, P(z) + Q(z)$ for arbitrary $P(z)$ and $Q(z)$ analytic in $\Omega \setminus \set{0}$. Then $F$ with natural function addition and multiplication forms a commutative ring.
\end{lemma}
\begin{proof}
Note that it suffices to show that $F$ is closed under addition and multiplication, as the commutative ring laws are clearly preserved. Let $(U, +, \times)$ be the commutative ring of functions analytic in $\Omega \setminus \set{0}$. Consider arbitrary $f,g \in F$ given by $f(z) = \sqrt{1-\zeta^{-1} z}\, P_f(z) + Q_f(z)$ and $g(z) = \sqrt{1-\zeta^{-1} z}\, P_g(z) + Q_g(z)$.
Let us start with $f(z) + g(z)$. Note that
\begin{equation*}
f(z) + g(z) = \sqrt{1-\zeta^{-1} z}\, \big(P_f(z) + P_g(z)\big) + Q_f(z) + Q_g(z).
\end{equation*}
Clearly, $f(z) + g(z) = \sqrt{1-\zeta^{-1} z}\, \overline{P}(z) + \overline{Q}(z)$ where both $\overline{P}(z) \in U$ and $\overline{Q}(z) \in U$. Hence, $f(z) + g(z) \in F$.
Now, let us consider $f(z)\cdot g(z)$. By rewriting, we obtain
\begin{eqnarray}\label{eq:sqrt-mult-eq}
f(z)\cdot g(z) &=& \sqrt{1-\zeta^{-1} z} \big(P_g(z) Q_f(z)+P_f(z) Q_g(z)\big)\\
\nonumber && +(1-\zeta^{-1}z)P_f(z) P_g(z) +Q_f(z) Q_g(z).
\end{eqnarray}
Clearly, $f(z)\cdot g(z) \in F$.
\end{proof}
\subsection{Removable singularities}\label{subsec:removable-singularities}
In order to apply~\autoref{th:singularity-analysis} to the analysis of a generating function $A(z)$, we have to guarantee that $A(z)$ is analytic at $z = 0$. If it is not the case, yet $A(z)$ has a \emph{removable pole singularity} at $z = 0$, we can consider its analytic extension $\widetilde{A}(z)$, instead of $A(z)$. The following theorem due to Bernhard Riemann provides a sufficient and necessary condition to determine whether $A(z)$'s pole singularity at $z = 0$ can be removed.
\begin{theorem}[Riemann's Removable Singularities Theorem, see e.g.~\cite{Krantz1999}]\label{th:riemann-rem-sing}
Let $f$ be analytic on the punctured disk $\Omega \setminus \set{z_0}$ of the complex plane. Then $f$ has an analytic extension on $\Omega$ if and only if $\lim_{z \to z_0} (z - z_0) f(z) = 0$.
\end{theorem}
As a direct consequence, we obtain the following technical lemma.
\begin{lemma}\label{lem:sing-powers}
Let $\Omega \setminus \set{z_0}$ be a punctured disk on the complex plane. Suppose that $f$ is analytic on $\Omega \setminus \set{z_0}$ and has an analytic continuation at $z = z_0$. Then for each $n \geq 2$, the function ${f(z)}^n$ has an analytic extension on $\Omega$.
\end{lemma}
\section{Basis-independent results}\label{sec:basis-independent-results}
In this section we are interested in universal basis-independent asymptotic properties of combinatory logic. We prove certain general results about labelled plane binary trees, deriving the combinatory logic results as immediate corollaries.
\begin{definition}
Suppose that $L$ is a finite set of $d$ distinct \emph{labels}. Then, the set of $L$-trees consists of plane binary trees where each leaf has a corresponding label in the set $L$. We use $\T{L}$ to denote the set of $L$-trees.
\end{definition}
Let us notice that the asymptotic growth rate of $L$-trees greatly depends on the asymptotic approximation of Catalan numbers $\catalan{n}$ counting the number of plane binary trees with $n$ inner nodes. It is well known that
\begin{equation*}
\catalan{n} = \frac{1}{n+1}{2n \choose n} \qquad \text{and} \qquad \catalan{n} = 4^n \frac{n^{-\nicefrac{3}{2}}}{\sqrt{\pi}}.
\end{equation*}
\begin{prop}
Let $\T{L}$ be the set of $L$-trees over a set $L$ of size $d$. Suppose that $|\cdot| \colon \T{L} \to \mathbb{N}$ is a function assigning each $L$-tree $t$ the number of binary nodes in $t$. Then $(\T{L}, |\cdot|)$ forms a combinatorial class.
\end{prop}
\begin{proof}
Let us start with noticing that $t \in \T{L}$ has $|t| + 1$ leaves. Fix $n \in \mathbb{N}$. The number of plane binary trees with $n$ inner nodes is counted by the $n$th Catalan number $\catalan{n}$. Taking into account all possible $L$-labellings of $n+1$ leaves and using the closed-form expression for $\catalan{n}$, we derive the following formula counting the number $\T{L,n}$ of $L$-terms of size $n$.
\begin{equation*}
\T{L,n} = d^{n+1} \catalan{n} = \frac{d^{n+1}}{n+1}\binom{2n}{n}.
\end{equation*}
\end{proof}
\subsection{Counting $L$-trees containing fixed $L$-trees as subtrees}\label{subsec:counting-l-trees}
Suppose that $t \in \T{L}$. Let $\overline{\T{L}}(z)$ denote the generating function counting the cardinalities of $L$-trees containing $t$ as a subtree. In the following series of propositions, we derive the closed-form solution for $\overline{\T{L}}(z)$ and check the conditions of~\autoref{th:singularity-analysis} used subsequently to show that in fact $[z^n]\overline{\T{L}}(z) \sim [z^n]\T{L}(z)$, independently of $L$.
\begin{prop}
Let $\T{L}$ be the set of $L$-trees where $|L| = d$. Then its counting sequence ${\set{\T{L,n}}}_{n \in \mathbb{N}}$ has a corresponding generating function $\T{L}(z)$ given by
\begin{equation}\label{eq:TL(z)-closed-form-solution}
\T{L}(z) = \frac{1-\sqrt{1-4 d z}}{2 z}.
\end{equation}
\end{prop}
\begin{proof}
Note that $\T{L}$ can be defined as $\T{L} = L + \T{L} \times \T{L}$,
which translates into the following function equation defining $\T{L}(z)$:
\begin{equation}\label{eq:TL(z)-fun-eq}
\T{L}(z) = d + z {\T{L}(z)}^2.
\end{equation}
Solving~\eqref{eq:TL(z)-fun-eq} for $\T{L}(z)$, we obtain two possible solutions:
\begin{equation*}
\T{L}(z) = \frac{1 \pm \sqrt{1-4 d z}}{2 z}.
\end{equation*}
Since the number of $L$-trees of size $0$ is equal to $d$, the limit $\lim_{z \to 0} \T{L}(z) = d$. It follows that~\eqref{eq:TL(z)-closed-form-solution} is indeed the desired solution.
\end{proof}
\begin{prop}
Let $L$ be a set of $d$ distinct labels. Assume that $t \in \T{L}$ is an $L$-tree of size $p \geq 1$. Then the set of $L$-trees containing $t$ as a subtree, denoted as $\overline{\T{L}}$, has the following generating function:
\begin{equation}\label{eq:overline-TL(z)-closed-form-solution}
\overline{\T{L}}(z) = \frac{-\sqrt{1-4 d z} + \sqrt{1 - 4 d z+4 z^{p+1}}}{2 z}.
\end{equation}
\end{prop}
\begin{proof}
Let us start with noticing that any $L$-tree containing $t$ as a subtree is either equal to $t$, or one of its left or right subtrees contains $t$ whereas the other one is a tree in $\T{L}$. However, since trees in $\T{L}$ may contain $t$ as a subtree, we have to subtract trees containing $t$ in both branches to avoid double counting. This specification allows us to write down the following functional equation defining $\overline{\T{L}}(z)$:
\begin{equation}\label{eq:TL(z)-fixed-term-fun-eq}
\overline{\T{L}}(z) = z^{p} + 2 z \T{L}(z) \overline{\T{L}}(z) - z {\overline{\T{L}}(z)}^{2}.
\end{equation}
Solving~\eqref{eq:TL(z)-fixed-term-fun-eq} for $\overline{\T{L}}(z)$ we obtain two possible solutions:
\begin{equation*}
\frac{-\sqrt{1-4 d z} \pm \sqrt{1 - 4 d z+4 z^{p+1}}}{2 z}.
\end{equation*}
Note that $p \geq 1$ and hence there are no $L$-trees of size $0$ containing $t$ as a subterm. It follows that $\lim_{z \to 0} \overline{\T{L}}(z) = 0$, yielding the desired solution.
\end{proof}
\begin{prop}\label{prop:T(z)-rho-only-sing}
Let $\zeta = \frac{1}{4d}$. Then $\zeta$ is the only singularity on both $\T{L}(z)$ and $\overline{\T{L}}(z)$'s circle of convergence.
\end{prop}
\begin{proof}
From~\eqref{eq:TL(z)-closed-form-solution} it is clear that $\zeta$ is the only singularity of $\T{L}(z)$ on the circle $|z| < \zeta$. Moreover, since $\sqrt{1 - 4d}$ is a part of $\overline{\T{L}}(z)$'s closed-form expression~\eqref{eq:overline-TL(z)-closed-form-solution}, it suffices to check that $F(z) = 1 - 4 d z+4 z^{p+1}$ has no complex roots of modulus $\zeta$. Note that we can rewrite~\eqref{eq:overline-TL(z)-closed-form-solution} as
\begin{eqnarray*}
\overline{\T{L}}(z) &=& \frac{1 -\sqrt{1-4 d z} - \left( 1 - \sqrt{1 - 4 d z+4 z^{p+1}} \right)}{2 z}\\
&=& \T{L}(z) - \frac{1 - \sqrt{1 - 4 d z+4 z^{p+1}}}{2 z}.
\end{eqnarray*}
Both $\T{L}(z)$ and $\overline{\T{L}}(z)$ are generating functions counting sequences of non-negative integers, hence the coefficients in the Maclaurin series of $\frac{1 - \sqrt{1 - 4 d z+4 z^{p+1}}}{2 z}$ are non-negative integers as well. To finish the proof we notice that
\begin{equation*}
F(\zeta) = 4^{-p} \left(\frac{1}{d}\right)^{p+1} > 0
\end{equation*}
and hence due to~\hyperref[th:pringsheim]{Pringsheim's Theorem}, $F(z)$ cannot have complex roots of modulus $\zeta$.
\end{proof}
The generating functions $\T{L}(z)$ and $\overline{\T{L}}(z)$ are not defined at $z = 0$, however due to~\autoref{th:riemann-rem-sing}, both have analytic extensions to functions analytic in the origin and we can consider them instead of $\T{L}(z)$ and $\overline{\T{L}}(z)$ in the subsequent theorem.
\begin{theorem}\label{th:TL(z)-asymptotic-approx}
Both $[z^n]\T{L}(z)$ and $[z^n]\overline{\T{L}}(z)$ admit for large $n$ the following asymptotic approximation:
\begin{equation}\label{eq:TL(z)-asymptotic-approx}
[z^n]\T{L}(z) \sim [z^n]\overline{\T{L}}(z) \sim {(4 d)}^n \frac{-2d n^{-\nicefrac{3}{2}}}{\Gamma(-\frac{1}{2})}.
\end{equation}
\end{theorem}
\begin{proof}
Let us rewrite the closed-form solutions of $\T{L}(z)$~\eqref{eq:TL(z)-closed-form-solution} and $\overline{\T{L}}(z)$~\eqref{eq:overline-TL(z)-closed-form-solution} as
\begin{eqnarray*}
\T{L}(z) &=& \sqrt{1-4 d}\left( -\frac{1}{2z} \right) + \frac{1}{2z}, \quad \text{and}\\
\overline{\T{L}}(z) &=& \sqrt{1-4 d}\left( -\frac{1}{2z} \right) + \frac{\sqrt{1 - 4 d z+4 z^{p+1}}}{2 z}.
\end{eqnarray*}
As all the assumptions hold, the result follows now easily by applying~\autoref{th:singularity-analysis}.
\end{proof}
Immediately, we obtain the following corollary theorem.
\begin{theorem}
Let $t \in T_L$. Then asymptotically almost all $L$-trees contain $t$ as a subtree.
\end{theorem}
\begin{proof}
In the case of $|t| \geq 1$, our claim follows directly from~\autoref{th:TL(z)-asymptotic-approx}. Suppose that $|t| = 0$, i.e.~$t$ is a primitive combinator. We can assume that $|L| > 1$, as otherwise our claim is trivial. Let us consider the set $\T{L \setminus \set{t}}$ of $L$-trees avoiding $t$. Note that from~\eqref{eq:TL(z)-closed-form-solution}, we have $[z^n]\T{L \setminus \set{t}}(z) \sim c_1 4^n(d-1)^n n^{-3/2}$, whereas $[z^n]\T{L}(z) \sim c_2 (4d)^n n^{-3/2}$ for some constants $c_1$ and $c_2$. It follows that asymptotically almost no $L$-tree avoids the primitive combinator $t$, finishing the proof.
\end{proof}
The above theorem provides a general result showing that `local' properties of $L$-trees propagating to supertrees, span asymptotically almost the whole set of $L$-trees. Using the natural bijection between $\mathcal{B}$-combinators and $\mathcal{B}$-trees, we can reinterpret this observation in the language of combinatory logic and state that each `local' property of $\mathcal{B}$-combinators closed under taking superterms is typical, i.e.~has asymptotic probability $1$ in the set of all $\mathcal{B}$-combinators. In particular, we obtain the following corollaries generalising the results in~\cite{dgkrtz}.
\begin{cor}\label{cor:subterm-prop}
For each Turing-complete basis of primitive combinators $\mathcal{B}$, asymptotically almost no $\mathcal{B}$-combinator is in normal form, simply typeable nor strongly normalising.
\end{cor}
\begin{proof}
Fix $t := \overline{\comb{S}} \comb{I} \comb{I} (\overline{\comb{S}} \comb{I} \comb{I})$ where $\comb{I} := \overline{\comb{S} \comb{K} \comb{K}}$.
\end{proof}
\subsection{Counting normalising combinators}\label{subsec:counting-normalizing-combinators}
Let $\mathcal{B}$ be a Turing-complete set of primitive combinators. Let us start with the classical observation is that the set of normalising $\mathcal{B}$-combinators is undecidable. It follows that its corresponding generating function has no computable closed-form solution. For that reason we take the following approach. We find feasible subclasses of normalising and non-normalising $\mathcal{B}$-combinators and use them to bound the density of normalising combinators $\mathcal{WN}_\mathcal{B}$. Let us start with recalling the famous standardisation theorem.
\begin{theorem}[Standardisation theorem, see e.g.~\cite{curry-feys}]\label{th:standardization-theorem}
If $M$ is normalising, then the leftmost outermost reduction always leads to $M$'s normal form.
\end{theorem}
Although usually stated in the $\S \mathbf{K}$-basis, the standardisation theorem easily generalises to every Turing-complete basis of combinators, e.g.~through the classical translation to $\lambda$-calculus (see, e.g.~\cite{BAR84}). Hence, we obtain the following result.
\begin{theorem}
Let $\mathcal{B}$ be a Turing-complete basis of primitive combinators. Then
\begin{equation*}
0 < \densityminus{\mathcal{WN}_\mathcal{B}}{\class{B}} \quad \text{and} \quad \densityplus{\mathcal{WN}_\mathcal{B}}{\class{B}} < 1.
\end{equation*}
\end{theorem}
\begin{proof}
Let us start with the lower bound. Since $\mathcal{B}$ is Turing-complete, there exists a combinator $\overline{\comb{K}} \in \class{\mathcal{B}}$ extensionally equivalent to $\comb{K}$. Let us consider the set $L$ of combinators in form of $\overline{\comb{K}} \comb{X} M$ where $\comb{X} \in \mathcal{B}$ is a primitive combinator and $M \in \class{\mathcal{B}}$. Notice that if $t \in L$, then $t$ has a normal form. Hence $L \subset \mathcal{WN}_\mathcal{B}$. Let us fix $p := |\overline{\comb{K}}|$. Then
\begin{eqnarray*}
\density{L}{\class{\mathcal{B}}} &=& \lim_{n \to \infty} \frac{|L_n|}{|\class{\mathcal{B},n}|} = \lim_{n \to \infty} \frac{d \cdot |\class{\mathcal{B},n-p-2}|}{|\class{\mathcal{B},n}|}\\
&=& \lim_{n \to \infty} \frac{d^{n-p} \cdot C_{n-p-2}}{d^{n+1} \cdot C_n} =
\frac{1}{d^{p+1} \cdot 4^{p+2}} > 0.
\end{eqnarray*}
Naturally we have
\begin{equation*}
\density{L}{\class{\mathcal{B}}} \leq \densityminus{\mathcal{WN}_\mathcal{B}}{\class{\mathcal{B}}}
\end{equation*}
hence indeed, the lower bound holds.
Now, let us consider the upper bound. Let $\omega = \overline{\comb{S} \comb{I} \comb{I}}$. Since $\comb{S} \comb{I} \comb{I} x \to_w x x$, we know that $\comb{S} \comb{I} \comb{I} (\comb{S} \comb{I} \comb{I})$ has no normal form. Immediately, nor does $\omega \omega$. Consider the transformation $\Phi \colon \class{\mathcal{B}} \to \class{\mathcal{B}}$ which for a given $\mathcal{B}$-combinator substitutes $\omega \omega$ for its leftmost primitive combinator $\comb{X}$ (see~\autoref{fig:U-transform}).
\begin{figure}[th!]
\resizebox{.75\linewidth}{!}{
\begin{subfigure}{.3\textwidth}
\centering
\begin{tikzpicture}[triangle/.style = {regular polygon, regular polygon sides=3 },
level1/.style ={level distance=1cm},]
\node [circle,draw,fill=black,scale=0.5] (a) {}
child[level1] {node [circle,draw,fill=black,scale=0.5] (b) {}
child[level1] {node {$\vdots$}
child[level1] {node [label={[yshift=-8mm]$\comb{X}$},circle,draw,fill=black,scale=0.5] (d) {}}
child[level1] {
node {$\vdots$}
}
}
child[level1] {
node {$\vdots$}
}
}
child[level1] {
node {$\vdots$}
};
\end{tikzpicture}
\end{subfigure}
\begin{subfigure}{.2\textwidth}
\[ \mathrel{\overset{\makebox[0pt]{\mbox{$\Phi$}}}{\longmapsto}} \]
\end{subfigure}
\begin{subfigure}{.3\textwidth}
\centering
\tikzset{
itria/.style={
draw,shape border uses incircle, scale=0.75,
isosceles triangle,shape border rotate=90,yshift=-1.45cm}
}
\begin{tikzpicture}[triangle/.style = {regular polygon, regular polygon sides=3 },
level1/.style ={level distance=1cm},]
\node [circle,draw,fill=black,scale=0.5] (a) {}
child[level1] {node [circle,draw,fill=black,scale=0.5] (b) {}
child[level1] {node {$\vdots$}
child[level1] {
node [circle,draw,fill=black,scale=0.5] (q) {}
node[itria] {$\comb{\omega \omega}$}}
child[level1] {
node {$\vdots$}
}
}
child[level1] {
node {$\vdots$}
}
}
child[level1] {
node {$\vdots$}
};
\end{tikzpicture}
\end{subfigure}
}
\caption{Transformation $\Phi$}
\label{fig:U-transform}
\end{figure}
Let $U$ be the image of $\class{\mathcal{B}}$ through $\Phi$. By~\autoref{th:standardization-theorem}, we know that $U$ is a set of non-normalising combinators. In other words, we have $\mathcal{WN}_\mathcal{B} \subset \class{\mathcal{B}} \setminus U$. Let $M$ be an arbitrary combinator in $U$. Note that since there are $d$ primitive combinators, the map $\Phi$ sends exactly $d$ distinct combinators to $M$. For convenience, let us set $p := |\omega \omega|$. Then we obtain
\begin{eqnarray*}
\density{U}{\class{\mathcal{B}}} &=& \lim_{n \to \infty} \frac{|U_n|}{|\class{\mathcal{B},n}|} = \lim_{n \to \infty} \frac{d \cdot |\class{\mathcal{B},n-p}|}{|\class{\mathcal{B},n}|}\\
&=& \lim_{n \to \infty} \frac{d^{n-p+2} \cdot C_{n-p}}{d^{n+1} \cdot C_n} =
\frac{d}{{(4 d)}^{p}} > 0.
\end{eqnarray*}
Since $\mathcal{WN}_\mathcal{B} \subset \class{\mathcal{B}} \setminus U$, we have
\begin{equation*}
\densityplus{\mathcal{WN}_\mathcal{B}}{\class{\mathcal{B}}} \leq 1 - \density{U}{\class{\mathcal{B}}},
\end{equation*}
and thus the upper bound holds as well, finishing the proof.
\end{proof}
Using the fact that asymptotically no $\mathcal{B}$-combinator is strongly normalising (see~\autoref{cor:subterm-prop}), we obtain the following corollary generalising the result in~\cite{Bendkowski2015}.
\begin{cor}
For each Turing-complete basis $\mathcal{B}$ of primitive combinators, asymptotically every weakly normalising $\mathcal{B}$-combinator is not strongly normalising.
\end{cor}
\section{SK-combinators}\label{sec:sk-combinators}
In this section we address the problem of estimating the asymptotic density of normalising $\comb{S}\comb{K}$-combinators in the set of all $\comb{S}\comb{K}$-combinators. In~\cite{Bendkowski2015} authors provided the following bounds:
\begin{equation*}
\frac{1}{32} \leq \densityminus{\mathcal{WN}_{\comb{S}\comb{K}}}{\class{\comb{S}\comb{K}}} \quad \text{and} \quad \densityplus{\mathcal{WN}_{\comb{S}\comb{K}}}{\class{\comb{S}\comb{K}}} \leq 1 - \frac{1}{2^{18}}.
\end{equation*}
Here, we prove that $\comb{S}\comb{K}$-combinators reducing in exactly $n>0$ normal-order reduction steps (leftmost outermost redex first, see~\autoref{th:standardization-theorem}) have positive asymptotic density in the set of all $\comb{S}\comb{K}$-combinators. We provide a constructive method of finding their asymptotic approximations and related densities, yielding a systematic approach to improving the above lower bound. For simplicity, we use $\class{\comb{S}\comb{K}}$ and $\rg{0}$ to denote the set of $\comb{S}\comb{K}$-combinators and the set of normal forms, respectively. Let us start with a few technical propositions regarding the generating functions $C(z)$ and $\rgf{0}$.
\begin{prop}[see, e.g.~\cite{Bendkowski2015}]
The generating function $C(z)$ enumerating $\S \mathbf{K}$-terms and its corresponding dominating singularity $\domsing{C}$ are given by
\begin{equation}\label{eq:c(z)}
C(z) = \frac{1-\sqrt{1-8z}}{2z} \quad \text{and} \quad \domsing{C} = \frac{1}{8}.
\end{equation}
\end{prop}
\begin{prop}\label{prop:c(z)-properties}
Let $n \geq 1$. Then ${C(z)}^n = \sqrt{1-8z}\, P(z) + Q(z)$ for some rational functions $P(z)$ and $Q(z)$ analytic in $\mathbb{C} \setminus \set{0}$. Moreover, ${C(z)}^n$ has a single removable singularity at $z = 0$ in the disk $|z| < \domsing{C}$.
\end{prop}
\begin{proof}
From equation~\eqref{eq:c(z)}, $C(z)$ can be rewritten as \[ C(z) = \sqrt{1-8z}\, P(z) + Q(z) \quad \text{where} \quad P(z) = -\frac{1}{2z} \quad \text{and} \quad Q(z) = \frac{1}{2z}. \]
Both $P(z)$ and $Q(z)$ are rational and hence also analytic in the complex plane except the origin. From~\autoref{lem:sqrt-commutative-ring}, ${C(z)}^n$ can be expressed as
\[ {C(z)}^n = \sqrt{1-8z}\, \overline{P}(z) + \overline{Q}(z) \]
for some $\overline{P}(z)$ and $\overline{Q}(z)$ analytic in $\mathbb{C} \setminus \set{0}$. Moreover, following~\eqref{eq:sqrt-mult-eq} and the closure properties of rational functions, it is clear that $\overline{P}(z)$ and $\overline{Q}(z)$ are also rational. As $\lim_{z \to 0} z C(z) = 0$, \autoref{th:riemann-rem-sing} guarantees that $C(z)$ has an analytic extension at $z = 0$ and, in consequence of~\autoref{lem:sing-powers}, so does ${C(z)}^n$, finishing the proof.
\end{proof}
\begin{prop}[see~\cite{Bendkowski2015}]
The generating function $\rgf{0}$ enumerating $\S \mathbf{K}$-terms in normal form and its corresponding dominating singularity $\domsing{0}$ are given by
\begin{equation}\label{eq:r0(z)}
\rgf{0} = \frac{1 - 2z - \sqrt{1-4z-4z^2}}{2z^2} \quad \text{and} \quad \domsing{0} = \frac{1}{2}\left(\sqrt{2}-1\right) \approx 0.207107.
\end{equation}
\end{prop}
\begin{prop}\label{prop:r0(z)-properties}
Let $n \geq 1$. Then ${R_0(z)}^n = \sqrt{1-4z-4z^2}\, P(z) + Q(z)$ for some rational functions $P(z)$ and $Q(z)$ analytic in $\mathbb{C} \setminus \set{0}$. Moreover, ${R_0(z)}^n$ has a single removable singularity at $z = 0$ in the disk $|z| < \domsing{0}$.
\end{prop}
\begin{proof}
From the shape of equation~\eqref{eq:r0(z)} we can write \[ R_0(z) = \sqrt{1-4z-4z^2}\, P(z) + Q(z) \quad \text{where} \quad P(z) = -\frac{1}{2z^2} \quad \text{and} \quad Q(z) = \frac{1-2z}{2z^2}. \]
Clearly, both $P(z)$ and $Q(z)$ are rational and analytic in $\mathbb{C} \setminus \set{0}$. The result follows now from the same arguments as in~\autoref{prop:c(z)-properties}.
\end{proof}
Our method relies on the effective computation and asymptotic properties of \emph{normal-order reduction grammars} $\set{\rg{n}}_{n\in\mathbb{N}}$. In~\cite{Bendkowski16} the author provided a recursive algorithm, which for given $n \geq 1$, constructs the $n$th normal-order reduction grammar $\rg{n}$
defining the set of $\S \mathbf{K}$-combinators reducing in exactly $n$ normal-order reduction steps. Applying the Symbolic Method of Flajolet and Sedgewick~\cite{FlajoletSedgewick2009}, $\rg{n}$ is then translated into a functional equation
\begin{equation}\label{eq:rgf-gen-fun-eq-psi}
\rgf{n} = \Psi(\rgf{n},z)
\end{equation}
involving the generating function $\rgf{n}$ and its formal parameter $z$. Due to the specific structure of $\Phi(\rg{n})$ -- the set of productions $\alpha \in \rg{n}$ not referencing $\rg{n}$ --~\eqref{eq:rgf-gen-fun-eq-psi} turns out to be linear in $\rgf{n}$, yielding a unique closed-form solution.
\begin{theorem}[see~\cite{Bendkowski16}]\label{th:reduction-grammars}
Let $n \geq 0$. Then there exists a computable unambiguous regular tree grammar $\rg{n}$ defining the set of $\S \mathbf{K}$-combinators reducing in exactly $n$ normal-order reduction steps. Moreover, $\rg{n}$ has a computable generating function $\rgf{n}$ of the following closed-form solution:
\begin{equation}\label{eq:rn(z)-fun-eq-simplified}
\rgf{n} = \frac{1}{\sqrt{1-4 z - 4 z^2}} \sum_{\alpha \in \Phi(\rg{n})} R_{\alpha}(z),
\end{equation}
where
\begin{equation}\label{eq:ralpha-fun-eq}
R_{\alpha}(z) = z^{k(\alpha)} {C(z)}^{c(\alpha)} \prod_{i=0}^{n-1} {R_i(z)}^{r_i(\alpha)}
\end{equation}
and $k(\alpha)$, $c(\alpha)$, $r_i(\alpha)$ are some non-negative integers depending on $\alpha$.
\end{theorem}
In the reminder of this section, we exploit the structure of the normal-order reduction grammars, showing the following main result.
\begin{theorem}\label{th:main}
Let $k \geq 1$. Then the asymptotic growth rate of $[z^n]\rgf{k}$ is given by
\[ [z^n]\rgf{k} \sim 8^n \frac{\overline{C}_k n^{-\nicefrac{3}{2}}}{\Gamma(-\nicefrac{1}{2})}, \]
where $\overline{C}_k$ is a constant depending on $k$.
\end{theorem}
Before we provide a proof, let us present two propositions preparing the background for~\autoref{th:singularity-analysis}.
\begin{prop}\label{prop:sing-at-0}
Let $n \geq 0$. Then each $\rgf{n}$ has a removable singularity at $z = 0$.
\end{prop}
\begin{proof}
Induction over $n$. Following~\autoref{th:riemann-rem-sing}, $\rgf{n}$ has a removable singularity at $z = 0$ if and only if the limit $\lim_{z \to 0} z \rgf{n}$ exists and is equal to $0$. In particular, from~\eqref{eq:rn(z)-fun-eq-simplified}
\begin{equation}\label{eq:rn(z)-rem-sing-form}
\lim_{z \to 0} \frac{z}{\sqrt{1-4 z - 4 z^2}} \bigg( z^{k(\alpha)} {C(z)}^{c(\alpha)} \prod_{i=0}^{n-1} {R_i(z)}^{r_i(\alpha)} \bigg) = 0
\end{equation}
for each $\alpha \in \Phi(\rg{n})$.
Let us start with $n = 0$. In this case the product $\prod_{i=0}^{n-1} {R_i(z)}^{r_i(\alpha)}$ vanishes, simplifying~\eqref{eq:rn(z)-rem-sing-form} to
\begin{equation}\label{eq:rn(z)-rem-sing-form-2}
\lim_{z \to 0} \frac{z^{k(\alpha)+1} {C(z)}^{c(\alpha)}}{\sqrt{1-4 z - 4 z^2}} = 0.
\end{equation}
Due to~\autoref{prop:r0(z)-properties}, ${C(z)}^{c(\alpha)}$ has an analytic extension at $z = 0$. It follows that $\lim_{z \to 0} {C(z)}^{c(\alpha)}$ exists, indeed satisfying equation~\eqref{eq:rn(z)-rem-sing-form-2}.
Now, suppose that $n > 0$. Due to the induction hypothesis all $\rgf{0},\ldots,\rgf{n-1}$ have removable singularities at $z = 0$. Using~\autoref{lem:sing-powers}, we can moreover state that so do their powers ${\rgf{0}}^{r_0(\alpha)},\ldots,{\rgf{n-1}}^{r_{n-1}(\alpha)}$. Together with our previous observation that ${C(z)}^{c(\alpha)}$ has an analytic extension at $z = 0$, we conclude that~\eqref{eq:rn(z)-rem-sing-form} is satisfied, finishing the proof.
\end{proof}
\begin{definition}
Let $\alpha \in \Phi(\rg{n})$ for some $n \geq 1$. We say that $\alpha$ is \emph{major} if and only if $\alpha$ references either $\class{\comb{S} \comb{K}}$ or some $\rg{i}$ for $i \in \set{1,\ldots,n-1}$. Otherwise, we say that $\alpha$ is \emph{minor}.
\end{definition}
In the following proposition we use the notions of major and minor productions, showing that major productions contribute to the asymptotic growth rate of $\rgf{n}$'s counting sequence, whereas minor ones are asymptotically negligible.
\begin{prop}\label{prop:form-of-rn(z)}
Let $n \geq 1$. Then each $\rgf{n}$ is in form of $\sqrt{1-8z}\, P(z) + Q(z)$ where both $P(z)$ and $Q(z)$ are analytic in the disk $|z| < \zeta_0 = \frac{\sqrt{2}-1}{2}$ but at $z = 0$.
\end{prop}
\begin{proof}
Induction over $n$. Consider the base case $n=1$.
Let us divide $\Phi(\rg{1})$ into two groups, i.e.~major and minor productions. Suppose that $\alpha \in \rg{1}$ is a major production. Since $\alpha \in \Phi(\rg{1})$, its corresponding generating function~\eqref{eq:ralpha-fun-eq} $\rgf{\alpha}$ is in form of
\begin{equation}\label{eq:rn(z)-asymptotic-form-1}
\rgf{\alpha} = z^{k(\alpha)} {C(z)}^{c(\alpha)} {R_0(z)}^{r_0(\alpha)},
\end{equation}
where in addition $c(\alpha) \geq 1$. Using Propositions~\ref{prop:c(z)-properties} and~\ref{prop:r0(z)-properties}, we can further rewrite~\eqref{eq:rn(z)-asymptotic-form-1} as
\begin{equation*}
\rgf{\alpha} = \sqrt{1-8z}\, \overline{P}(z) + \overline{Q}(z)
\end{equation*}
for functions $\overline{P}(z)$ and $\overline{Q}(z)$ analytic in the disk $|z| < \zeta_0$ but at $z=0$. Similarly, if $\alpha \in \Phi(\rg{1})$ is minor, we can rewrite its generating function as
\begin{equation*}
\rgf{\alpha} = \sqrt{1-4z-4z^2}\, \widehat{P}(z) + \widehat{Q}(z)
\end{equation*}
where $\widehat{P}(z)$ and $\widehat{Q}(z)$ are analytic in some disk $|z| < \zeta_0 + \varepsilon$ for $\varepsilon > 0$ but at $z=0$. The requested form of $\rgf{1}$ follows now from~\autoref{lem:sqrt-commutative-ring} and the fact that $\zeta_1 < \zeta_0$.
Now, suppose that $n > 1$. Again, let us consider an arbitrary major $\alpha \in \Phi(\rg{n})$. Using the induction hypothesis and~\eqref{eq:ralpha-fun-eq}, we can rewrite $\rgf{\alpha}$ as
\begin{equation}\label{eq:rn(z)-asymptotic-form-2}
\rgf{\alpha} = z^{k(\alpha)} {C(z)}^{c(\alpha)} {\rgf{0}}^{r_0(\alpha)} \prod_{i=1}^{n-1} {\bigg(\sqrt{1-8z}\, \overline{P_i}(z) + \overline{Q_i}(z)\bigg)}^{r_i(\alpha)}.
\end{equation}
Using Propositions~\ref{prop:c(z)-properties} and~\ref{prop:r0(z)-properties}, we can further rewrite~\eqref{eq:rn(z)-asymptotic-form-2} as
\begin{equation}\label{eq:rn(z)-asymptotic-form-3}
\rgf{\alpha} = \bigg(\sqrt{1-8z}\, \overline{P}(z) + \overline{Q}(z)\bigg) \prod_{i=1}^{n-1} {\bigg(\sqrt{1-8z}\, \overline{\overline{P_i}}(z) + \overline{\overline{Q_i}}(z)\bigg)}^{r_i(\alpha)}.
\end{equation}
The result follows now easily from~\autoref{lem:sqrt-commutative-ring}.
\end{proof}
Now we are in a position to prove \autoref{th:main}.
\begin{proof}{(\autoref{th:main})}
Let $k>0$. Due to \autoref{prop:form-of-rn(z)}, every function $R_k(z)$ is in form of $\sqrt{1-8z}P_k(z) + Q_k(z)$ for some algebraic functions $P_k(z)$ and $Q_k(z)$ that are analytic in the disk $|z|<\frac{\sqrt{2}-1}{2}$ but at $z=0$. By \autoref{prop:sing-at-0}, every $R_k(z)$ has a removable singularity at $z=0$. Therefore, every function $R_k(z)$ satisfies the assumptions of \autoref{th:singularity-analysis}. Hence
\[ [z^n]\rgf{k} \sim 8^n \frac{\overline{C}_k n^{-\nicefrac{3}{2}}}{\Gamma(-\nicefrac{1}{2})}\]
for some constant $\overline{C}_k$.
\end{proof}
As $\domsing{m} = \nicefrac{1}{8}$ for every $m \geq 1$, we can easily compute the coefficients $\overline{C}_m$ in the asymptotic approximation of $[z^n]\rgf{m}$ using available computer algebra systems, e.g. Mathematica \textregistered~\cite{mathematicaSoft}. The quotient $\nicefrac{\overline{C}_m}{-4}$ (see~\autoref{eq:TL(z)-asymptotic-approx}) yields the desired asymptotic density of $\S \mathbf{K}$-combinators normalising in $m$ normal-order reduction steps in the set of all $\S \mathbf{K}$-combinators. Hence, we obtain the following corollary.
\begin{cor}
For each $m \in \mathbb{N}$, finding the asymptotic density of combinators normalising in $m$ steps in the set of all $\S \mathbf{K}$-combinators is computable.
\end{cor}
Using an implementation of the normal-order reduction grammar algorithm together with Mathematica \textregistered~we were able to compute the densities of combinators reducing in $m$ normal-order reduction steps $\rg{m}$ in $C_{\S \mathbf{K}}$ for $m = 1,\ldots,7$. The results are summarised in the following figure.
\begin{figure}[th!]
\begin{displaymath}
\begin{array}{r|l}
m & \density{\rg{m}}{C_{\S \mathbf{K}}} \\\hline
1 & 0.08961233291075565\\\hline
2 & 0.06417374404832035\\\hline
3 & 0.0501056553007704\\\hline
4 & 0.04131967414765603\\\hline
5 & 0.03570996929825453\\\hline
6 & 0.03119525702124082\\\hline
7 & 0.027987393260263862\\
\end{array}
\end{displaymath}
\caption{Density of $\rg{m}$ in $\class{\S \mathbf{K}}$ for $m=1,\ldots,7$}
\end{figure}
And so, exploiting the finite additivity of asymptotic density we obtain the following improved lower bound:
\begin{equation*}
0.34010402598726164 \leq \densityminus{\mathcal{WN}_{\comb{S}\comb{K}}}{\class{\comb{S}\comb{K}}}.
\end{equation*}
Clearly, the above lower bound can be further improved if we compute the next asymptotic densities $\density{\rg{8}}{\class{\S \mathbf{K}}},\density{\rg{9}}{\class{\S \mathbf{K}}},\density{\rg{10}}{\class{\S \mathbf{K}}}$, etc. Unfortunately, due to the sheer amount of major productions, this process is immensely time and memory consuming, quickly requiring resources exceeding current desktop computer capabilities. Nevertheless, $\density{\rg{m}}{\class{\S \mathbf{K}}} \leq 1$ for each $m$ and hence
$\sum_{m \geq 0} \density{\rg{m}}{\class{\S \mathbf{K}}}$ is necessarily convergent to some value $0 < \zeta < 1$. Moreover, if $\density{\mathcal{WN}_{\comb{S}\comb{K}}}{\class{\S \mathbf{K}}}$ exists, then we have $\zeta \leq \density{\mathcal{WN}_{\comb{S}\comb{K}}}{\class{\S \mathbf{K}}}$. The intriguing problem of determining whether the inequality can be replaced by the equality still remains open.
\section{Experimental results}\label{sec:experimental-results}
Our method developed in~\autoref{sec:sk-combinators} allows us to improve the lower bound on $\densityminus{\mathcal{WN}_{\comb{S}\comb{K}}}{\class{\comb{S}\comb{K}}}$, provided we have enough computational resources to find and manipulate generating functions $\rgf{m}$ for higher $m$. Unfortunately, the current gap between the lower and upper bound on the density of normalising combinators is still quite significant. In this section we present some experimental results regarding the aforementioned density as well as numerical evaluations of the obtained approximation error.
\subsection{Super-computer results}\label{subsec:super-computer-results}
Consider the following experiment scheme $G(s,n,r)$ with three positive integer parameters $s,n,r$. We draw $s$ uniformly random $\S \mathbf{K}$-combinators of size $n$ using, e.g.~R{\'e}my's algorithm (see~\cite{DBLP:journals/ita/Remy85,Knuth:2006:ACP:1121689}) -- drawing a uniformly random plane binary tree with $n$ inner nodes -- combined with a random $\S \mathbf{K}$-labelling. Then we reduce each of the $s$ samples using up to $r$ normal-order reduction steps. We record then the number of normalised samples, with their corresponding reduction lengths. For samples which did not normalise in $r$ reduction steps, we artificially record their reduction lengths as $-1$. Afterwards, we collect the reduction lengths, plotting the obtained function mapping reduction lengths to the number of samples attaining the given reduction length.
We preformed our experiments on the \emph{Prometheus \textregistered}~super-computer cluster granted by ACC Cyfronet AGH in Kraków, Poland (\nth{48} out of 500 world's most powerful supercomputer in 2016, with theoretical computational power of 2.4 Pflops)~\cite{prometheus}. The following figure summarises the experiment result for $G(s = 1200,n = 50000000,r = 1000)$.
\begin{figure}[th!]\label{fig:experiment}
\centering
\resizebox {200px} {!} {
\begin{tikzpicture}
\begin{axis}[
xlabel={normal-order reductions},
ylabel={Number of samples},
xmin=-1, xmax=868,
ymin=0, ymax=176,
legend pos=north west,
ymajorgrids=true,
grid style=dashed,
]
\addplot[
color=blue,
mark=*,
]
coordinates {
(-1,176)
(1,110)
(2,92)
(3,57)
(4,47)
(5,33)
(6,24)
(7,35)
(8,30)
(9,21)
(10,25)
(11,25)
(12,33)
(13,19)
(14,17)
(15,16)
(16,13)
(17,12)
(18,10)
(19,18)
(20,18)
(21,12)
(22,12)
(23,6)
(24,12)
(25,15)
(26,8)
(27,9)
(28,10)
(29,8)
(30,8)
(31,12)
(32,10)
(33,6)
(34,4)
(35,8)
(36,9)
(37,6)
(38,9)
(39,10)
(40,2)
(41,5)
(42,9)
(43,3)
(44,5)
(45,3)
(46,8)
(47,8)
(48,4)
(49,4)
(50,2)
(51,1)
(52,1)
(53,4)
(54,4)
(55,3)
(56,3)
(57,4)
(58,4)
(59,2)
(60,1)
(61,8)
(62,2)
(63,2)
(64,1)
(65,2)
(66,2)
(67,1)
(68,2)
(69,1)
(70,2)
(72,2)
(75,1)
(76,2)
(77,1)
(78,1)
(79,2)
(81,2)
(83,2)
(84,2)
(85,1)
(89,2)
(90,1)
(91,1)
(93,2)
(95,3)
(96,3)
(97,2)
(98,1)
(100,2)
(102,1)
(103,1)
(108,1)
(109,2)
(110,1)
(111,3)
(112,1)
(114,3)
(116,1)
(119,1)
(120,1)
(126,3)
(129,1)
(135,1)
(136,1)
(137,1)
(141,1)
(167,1)
(170,1)
(183,2)
(192,1)
(193,1)
(195,1)
(208,2)
(216,1)
(218,1)
(219,1)
(223,1)
(234,1)
(236,1)
(237,1)
(245,1)
(279,2)
(282,1)
(289,1)
(290,1)
(301,1)
(314,1)
(316,1)
(329,1)
(401,1)
(469,1)
(478,1)
(490,1)
(601,1)
(684,1)
(693,1)
(698,1)
(868,1)
};
\legend{samples}
\end{axis}
\end{tikzpicture}}
\caption{$G(1200, 50000000, 1000)$}
\end{figure}
Even though the number $r = 1000$ bounding the number of performed reductions was significantly smaller than the size of considered samples, the experiment revealed that normalising combinators have short reduction lengths. In fact, the mean reduction length $E(X)$ over all normalising samples is approximately
$E(X) \approx 31.5810$, whereas $\log_2 n \approx 25.5754$. Out of $1200$ samples, only $176$ did not normalise in $r$ steps, yielding approximately $0.146$ percent of all considered samples. Similar results were obtained with different experiment parameters, suggesting that the ratio of normalising $\S \mathbf{K}$-combinators should be approximately equal to $85\%$, whereas the mean reduction length of normalising terms is $\Theta(\log_2 n)$.
Our Haskell implementation of the program, as well as all the obtained data sets are available at~\cite{mb-haskell-implementation-experiments}.
\subsection{Approximation error}\label{subsec:approximation-error}
In~\autoref{sec:sk-combinators} we showed that $[z^n]\rgf{m} = \Theta([z^n]C(z))$ for $m \geq 1$. Nonetheless, techniques used to obtain this result do not yield the convergence rate of sequences in question. Using Mathematica \textregistered~we compared $[z^n]\rgf{m}$ with its approximation $8^n \widetilde{C}_m n^{-\nicefrac{3}{2}}$. The following figures summarise values for $[z^n]\rgf{1}$ and their relative error $\delta([z^n]\rgf{1})$.
\begin{figure}[th!]
\centering
\begin{displaymath}
\begin{array}{r|r|r|r|r}
n & [z^n]C(z) & [z^n]\rgf{1} & \lfloor 8^n \widetilde{C}_1 n^{-\nicefrac{3}{2}} \rfloor & \delta([z^n]\rgf{1}) \\\hline
2 & 16 & 4 & 2 & 0.5 \\
3 & 80 & 32 & 9 & 0.71875 \\
4 & 448 & 200 & 51 & 0.745 \\
5 & 2688 & 1152 & 296 & 0.7430555555555556 \\
6 & 16896 & 6528 & 1803 & 0.7238051470588235 \\
7 & 109824 & 37184 & 11450 & 0.6920718588640276 \\
8 & 732160 & 215328 & 74973 & 0.6518195497102095 \\
9 & 4978688 & 1275520 & 502653 & 0.6059230745107878 \\
10 & 34398208 & 7753472 & 3433386 & 0.5571808345990029 \\
11 & 240787456 & 48412416 & 23808041 & 0.5082244810917926 \\
12 & 1704034304 & 310294272 & 167159405 & 0.4612874935699748 \\
13 & 12171673600 & 2037696512 & 1185980764 & 0.4179796858777761 \\
14 & 87636049920 & 13675532288 & 8489666053 & 0.3792076334425748 \\
15 & 635361361920 & 93532264448 & 61240081391 & 0.3452518042579103 \\
16 & 4634400522240 & 650108973568 & 444715903783 & 0.3159363708790836 \\
17 & 33985603829760 & 4580578080768 & 3248472837654 & 0.2908159668114757 \\
18 & 250420238745600 & 32644683026432 & 23852497067944 & 0.2693298002424797 \\
19 & 1853109766717440 & 234890688573440 & 175955235773882 & 0.2509058709712602 \\
20 & 13765958267043840 & 1703833526784000 & 1303399617705108 & 0.2350193858637788
\end{array}
\end{displaymath}
\caption{$[z^n]\rgf{1} \sim 8^n \widetilde{C}_1 n^{- \nicefrac{3}{2}}$ with $\widetilde{C}_1 \approx 0.10111668957132425$.}
\end{figure}
\begin{figure}[th!]
\centering
\begin{subfigure}{.6\textwidth}
\centering
\begin{tikzpicture}
\begin{axis}[
title={Relative error $\delta([z^n]\rgf{1})$ for $2 \leq n \leq 20$},
xlabel={$n$},
xmin=2, xmax=20,
ymin=0, ymax=1,
legend pos=north west,
ymajorgrids=true,
grid style=dashed,
]
\addplot[
color=blue,
mark=*,
]
coordinates {
(2,0.5)
(3, 0.71875)
(4, 0.745)
(5,0.7430555555555556)
(6,0.7238051470588235)
(7,0.6920718588640276)
(8,0.6518195497102095)
(9,0.6059230745107878)
(10,0.5571808345990029)
(11,0.5082244810917926)
(12,0.4612874935699748)
(13,0.4179796858777761)
(14,0.3792076334425748)
(15,0.3452518042579103)
(16,0.3159363708790836)
(17,0.29081596681147576)
(18,0.26932980024247977)
(19,0.2509058709712602)
(20,0.2350193858637788)
};
\legend{$\delta([z^n]\rgf{1})$}
\end{axis}
\end{tikzpicture}
\end{subfigure
\begin{subfigure}{.4\textwidth}
\centering
\begin{displaymath}
\begin{array}{r|l}
n & \delta([z^n]\rgf{1}) \\\hline
20 & 0.2350193858637788 \\
40 & 0.10914523244529438 \\
60 & 0.07194386101679777 \\
80 & 0.053697148923235606 \\
100 & 0.042841972392710106 \\
120 & 0.03564026685975644 \\
140 & 0.03051237282492831 \\
160 & 0.026674947319897818 \\
180 & 0.023695169712517998 \\
200 & 0.021314357312846963 \\
220 & 0.019368374581756907 \\
240 & 0.017748046896721537 \\
260 & 0.01637792978607196 \\
280 & 0.015204215010418284 \\
300 & 0.014187492529606827
\end{array}
\end{displaymath}
\end{subfigure
\end{figure}
Since $[z^n]\rgf{1} \sim 8^n \widetilde{C}_1 n^{-\nicefrac{3}{2}}$, the relative error $\delta([z^n]\rgf{1})$ is inevitably tending to $0$ as $n \to \infty$. Remarkably, the error $\delta([z^n]\rgf{1})$ converges much slower than one would expect. With $n = 300$ the error is just of order $10^{-2}$. We observed similar results in the relative errors for higher $n$, where the convergence rate is even slower than in the case of $[z^n]\rgf{1}$.
Our Mathematica \textregistered~scripts and an implementation of the algorithm computing $\rgf{m}$ are available at~\cite{mb-haskell-implementation}.
\section{Conclusion}\label{sec:conclusion}
We presented several basis-independent results regarding asymptotic properties of combinatory logic. In particular, we generalised previously known results for $\S \mathbf{K}$-combinators from~\cite{Bendkowski2015,dgkrtz}, showing that they span to arbitrary Turing-complete combinator bases. Exploiting the results of~\cite{Bendkowski16}, we gave a systematic approach to finding better lower bounds on the density of normalising $\S \mathbf{K}$-combinators, improving the previously best known lower bound from about $3\%$ to approximately $34\%$. Performed super-computer experiments suggest that the searched density, if it exists, should be approximately $85\%$, conjecturing that both lower and upper bounds are still quite far from the actual density. Naturally, the following interesting question emerges -- is our approach converging to the actual density? As the asymptotic density is always bounded, the series $\sum_{m\geq 0} \density{R_m}{\class{\comb{S}\comb{K}}}$ converges to some value $\zeta \in (0,1)$. Is $\zeta$ the desired asymptotic density of normalising $\S \mathbf{K}$-combinators? If not, how far is it from the actual density? Due to the immense computational resources required in the computations and the sheer amount of major normal-order reduction grammar productions, our approach renders brute-force methods of closing the density gap virtually impossible. We expect that more sophisticated techniques are required in order to address this intriguing open problem.
\bibliographystyle{plain}
| 5fa96eaf15096a15ed790bc5a6822030d8cfdf78 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section*{1. Introduction}
The Riemann zeta function $\zeta(s)$ which basically was known already to Euler establishes the most important link between number theory and analysis. The proof of the Riemann hypothesis is a longstanding problem since it was formulated by Riemann \cite{rie} in 1859. The Riemann hypothesis is the conjecture that all nontrivial zeros of the Riemann zeta function $\zeta(s)$ for complex $s=\sigma+\I t$ are positioned on the line $s=\frac{1}{2}+\I t$ that means on the line parallel to the imaginary axis through real value $\sigma = \frac{1}{2}$ in the complex plane and in extension that all zeros are simple zeros \cite{whitt,titch,chand,edw,parschin,patt,cart,ivic,havil,riben,borw,apost0,weisst,
lang,
spekt,stew} (with extensive lists of references in some of the cited sources, e.g., \cite{chand,edw,ivic,borw,weisst}). The book of
Edwards \cite{edw} is one of the best older sources concerning most problems connected with the Riemann zeta function. There are also mathematical tables and chapters in works about Special functions which contain information about the Riemann zeta function and about number analysis,
e.g., Whittaker and Watson \cite{whitt} (chap. 13), Bateman and Erd\'{e}lyi \cite{bate1} (chap. 1) about zeta functions, \cite{bate3} (chap. 17) about number analysis, and Apostol \cite{apost,nist} (chaps. 25 and 27). The book of Borwein, Choi, Rooney and Weirathmueller \cite{borw} gives on the first 90 pages a short account about achievements concerning the Riemann hypothesis and its consequences for number theory and on the following about 400 pages it reprints important original papers and expert witnesses in the field.
Riemann has put aside the search for a proof of his hypothesis 'after some fleeting vain attempts' and emphasizes that 'it is not necessary for the immediate objections of his investigations' \cite{rie} (see \cite{edw}).
The Riemann hypothesis was taken by Hilbert as the 8-th problem in his representation of 23 fundamental unsolved problems in pure mathematics and axiomatic physics in a lecture hold on 8 August in 1900 at the Second Congress of Mathematicians in Paris \cite{hilb,linnik}.
The vast experience with the Riemann zeta function in the past and the progress in numerical calculations of the zeros (see, e.g., \cite{edw,havil,riben,spekt,stew,conrey,bomb}) which all confirmed the Riemann hypothesis suggest that it should be true corresponding to the opinion of most of the specialists in this field but not of all specialists (arguments for doubt are discussed in \cite{ivic1}).
The Riemann hypothesis is very important for prime number theory and a number of consequences is derived under the unproven assumption that it is true. As already said a main role plays a function $\zeta(s)$ which was known already to Euler for real variables $s$ in its product representation (Euler product) and in its series representation and was continued to the whole complex $s$-plane by Riemann and is now called Riemann zeta function. The Riemann hypothesis as said is the conjecture that all nontrivial zeros of the zeta function $\zeta(s)$ lie on the axis parallel to the imaginary axis and intersecting the real axis at $s=\frac{1}{2}$. For the true hypothesis the representation of the Riemann zeta function after exclusion of its only singularity at $s=1$ and of the trivial zeros at $s=-2n,\,(n=1,2,\ldots)$ on the negative real axis is possible by a Weierstrass product with factors which only vanish on the critical line $\sigma=\frac{1}{2}$. The function which is best suited for this purpose is the so-called xi function $\xi(s)$ which is closely related to the zeta function $\zeta(s)$ and which was also introduced by Riemann \cite{rie}. It contains all information about the nontrivial zeros and only the exact positions of the zeros on this line are not yet given then by a closed formula which, likely, is hardly to find explicitly but an approximation for its density was conjectured already by Riemann \cite{rie} and proved by von Mangoldt \cite{mang2}.
The "(pseudo)-random" character of this distribution of zeros on the critical line remembers somehow the "(pseudo)-random" character of the distribution of primes where one of the differences is that the distribution of primes within the natural numbers becomes less dense with increasing integers whereas the distributions of zeros of the zeta function on the critical line becomes more dense with higher absolute values with slow increase and approaches to a logarithmic function in infinity.
There are new ideas for analogies to and application of the Riemann zeta function in other regions of mathematics and physics. One direction is the theory of random matrices \cite{spekt,conrey} which shows analogies in their eigenvalues to the distribution of the nontrivial zeros of the Riemann zeta function. Another interesting idea founded by Voronin \cite{voronin} (see also \cite{spekt,steud}) is the universality of this function in the sense that each holomorphic function without zeros and poles in a certain circle with radius less $\frac{1}{2}$ can be approximated with arbitrary required accurateness in a small domains of the zeta function to the right of the critical line within $\frac{1}{2} \le s \le 1$. An interesting idea is elaborated in articles of Neuberger, Feiler, Maier and Schleich \cite{schl1,schl2}. They consider a simple first-order ordinary differential equation with a real variable $t$ (say the time) for given arbitrary analytic functions $f(z)$ where the time evolution of the function for every point $z$ finally transforms the function in one of the zeros $f(z)=0$ of this function in the complex $z$-plane and illustrate this process graphically by flow curves which they call Newton flow and which show in addition to the zeros the separatrices of the regions of attraction to the zeros. Among many other functions they apply this to the Riemann zeta function $\zeta(z)$ in different domains of the complex plane. Whether, however, this may lead also to a proof of the Riemann hypothesis is more than questionable.
Number analysis defines some functions of a continuous variable, for example, the number of primes $\pi(x)$ less a given real number $x$ which last is connected with the discrete prime number distribution (e.g., \cite{titch,chand,edw,patt,ivic,riben}) and establishes the connection to the Riemann zeta function $\zeta(s)$. Apart from the product representation of the Riemann zeta function the representation by
a type of series which is now called Dirichlet series was already known to Euler. With these Dirichlet series in number theory are connected some discrete functions over the positive integers $n=1,2,\ldots$ which play a role as coefficients in these series and are called arithmetic functions (see, e.g., Chandrasekharan \cite{chand} and Apostol \cite{apost0}). Such functions are the M\"{o}bius function $\mu(n)$ and the Mangoldt function ${\rm \Lambda}(n)$ as the best known ones. A short representation of the connection of the Riemann zeta function to number analysis and of some of the functions defined there became now standard in many monographs about complex analysis (e.g., \cite{lang}).
Our means for the proof of the Riemann hypothesis in present article are more conventional and "old-fashioned" ones, i.e. the Real Analysis and the Theory of Complex Functions which were developed already for a long time.
The most promising way for a proof of the Riemann hypothesis as it seemed to us in past is via the already mentioned entire function $\xi(s)$ which is closely related to the Riemann zeta function $\zeta(s)$. It contains all important elements and information of the last but excludes its trivial zeros and its only singularity and, moreover, possesses remarkable symmetries which facilitate the work with it compared with the Riemann zeta function. This function $\xi(s)$ was already introduced by Riemann \cite{rie} and dealt with, for example, in the classical books of Titchmarsh \cite{titch}, Edwards \cite{edw} and in almost all of the sources cited at the beginning. Present article is mainly concerned with this xi function $\xi(s)$ and its investigation in which, for convenience, we displace the imaginary axis by $\frac{1}{2}$ to the right that means to the critical line and call this Xi function ${\mit\Xi}(z)$ with $z=x+\I y$. We derive some representations for it among them novel ones and discuss its properties, including its derivatives, its specialization to the critical line and some other features. We make an approach to this function via the second mean value theorem of analysis (Gauss-Bonnet theorem, e.g., \cite{courant,widder}) and then we apply an operator identity for analytic functions which is derived in Appendix B and which is equivalent to a somehow integrated form of the Cauchy-Riemann equations. This among other not so successful trials (e.g., via moments of function ${\mit \Omega}(u)$) led us finally to a proof of the Riemann hypothesis embedded into a proof for a more general class of functions.
Our approach to a proof of the Riemann hypothesis in article in rough steps is as follows:\\
First we shortly represent the transition from the Riemann zeta function $\zeta(s)$ of complex variable $s=\sigma +\I t$ to the xi function $\xi(s)$ introduced already by Riemann and derive for it by means of the Poisson summation formula a representation which is convergent in the whole complex plane (Section 2 with main formal part in Appendix A).
Then we displace the imaginary axis of variable $s$ to the critical line at $s=\frac{1}{2}+\I t$ by $s\rightarrow z = s-\frac{1}{2}$ that is purely for convenience of further working with the formulae. However, this has also the desired subsidiary effect that it brings us into the fairway of the complex analysis usually represented with the complex variable $z=x+\I y$. The transformed $\xi(s)$ function is called ${\mit \Xi}(z)$ function.
The function ${\mit \Xi}(z)$ is represented as an integral transform of a real-valued function ${\mit \Omega}(u)$ of the real variable $u$ in the form ${\mit \Xi}(z)= \int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(u z)$ which is related to a Fourier transform (more exactly to Cosine Fourier transform). If the Riemann hypothesis is true then we have to prove that all zeros of the function ${\mit \Xi}(z)$ occur for $x=0$.
To the Xi function in mentioned integral transform we apply the second mean-value theorem of real analysis first on the imaginary axes and discuss then its extension from the imaginary axis to the whole complex plane. For this purpose we derive in Appendix B in operator form relations of the complex mean-value parameter in our application of the second mean-value theorem to this parameter on the imaginary axis which are equivalents in integral form to the Cauchy-Riemann equations in differential form and apply this in specific form to the Xi function (Sections 3 and 4).
Then in Section 5 we accomplish the proof with the discussion and solution of the two most important equations (\ref{zeroscond1}) and (\ref{zeroscond2}) for the last and decisive stage of the proof. These two equations are derived before in preparation of this last stage of the proof. From these equations it is seen that the obtained two real equations admit zeros of the Xi function only on the imaginary axis. This proves the Riemann hypothesis by the equivalence of the Riemann zeta function $\zeta(s)$ to the Xi function ${\mit \Xi}(z)$ and embeds it into a whole class of functions with similar properties and positions of their zeros.
The Sections 6-7 serve for illustrations and graphical representations of the specific parameters (e.g., mean-value parameters) for the Xi function to the Riemann hypothesis and for other functions which in our proof by the second mean-value problem are included for the existence of zeros only on the imaginary axis. This is in particular the whole class of modified Bessel functions ${\rm I}_{\nu}(z), \; \left(-\frac{1}{2} < \nu < +\infty\right)$ with real indices $\nu$ which possess zeros only on the imaginary axis $y$ and where a proof by means of the differential equations exists.
\setcounter{chapter}{2}
\setcounter{equation}{0}
\section*{2. From Riemann zeta function ${\cf{\zeta(s)}}$ to related xi function ${\cf{\xi(s)}}$ and its argument displacement to function ${\cf{{\mit \Xi(z)}}} $}
In this Section we represent the known transition from the Riemann zeta function $\zeta(s)$ to a function $\xi(s)$ and finally to a function ${\mit \Xi}(z)$ with displaced complex variable $s\rightarrow z=s-\frac{1}{2}$ for rational effective work and establish some of the basic representations of these functions, in particular, a kind of modified Cosine Fourier transformations of a function ${\mit \Omega}(u)$ to the function ${\mit \Xi}(z)$.
As already expressed in the Introduction, the most promising way for a proof of the Riemann hypothesis as it seems to us is the way via a certain integral representation of the related xi function $\xi(s)$. We sketch here the transition from the Riemann zeta function $\zeta(s)$ to the related xi function $\xi(s)$ in a short way because, in principle, it is known and we delegate some aspects of the derivations to Appendix A
Usually, the starting point for the introduction of the Riemann zeta function $\zeta(s)$ is the following relation between the Euler product and an infinite series continued to the whole complex $s$-plane
\begin{eqnarray}
\zeta(s) \equiv \prod_{n=1}^\infty
\left(1-\frac{1}{p_n^{\,s}}\right)^{-1} = \sum_{n=1}^\infty
\frac{1}{n^s},\quad \left(\sigma \equiv {\rm Re}(s)
>1\right),\label{eulerprod}
\end{eqnarray}
where $p_n$ denotes the ordered sequence of primes ($p_1=2,p_2=3,p_3=5,\ldots$). The transition from the product formula to the sum representation in (\ref{eulerprod}) via transition to the Logarithm of $\zeta(s)$ and Taylor series expansion of the factors $\log\left(1-\frac{1}{p_n^{\,s}}\right)^{-1}$ in powers of $\frac{1}{p_n^{\,s}}$ using the uniqueness of the prime-number decomposition is well known and due to Euler in 1737. It leads to a special case of a kind of series later introduced and investigated in more general form and called Dirichlet series. The Riemann zeta function $\zeta(s)$ can be analytically continued into the whole complex plane to a meromorphic function that was made and used by Riemann. The sum in (\ref{eulerprod}) converges uniformly for complex variable
$s=\sigma +\I t$ in the open semi-planes with arbitrary $\sigma > 1$ and arbitrary $t$. The only singularity of the function $\zeta(s)$ is a simple pole at $s=1$ with residue $1$ that we discuss below.
The product form (\ref{eulerprod}) of the zeta function $\zeta(s)$ shows that it involves all prime numbers $p_n$ exactly one times and therefore it contains information about them in a coded form. It proves to be possible to regain information about the prime number distribution from this function. For many purposes it is easier to work with meromorphic and, moreover, entire functions than with infinite sequences of numbers but in first case one has to know the properties of these functions which are determined by their zeros and their singularities together with their multiplicity.
From the well-known integral representation of the Gamma function
\begin{eqnarray}
{\rm \Gamma}(z) &=& \int_{0}^{+\infty}dt\,t^{z-1}\E^{-t},\quad
\left({\rm Re}(z) > 0\right),\label{gammadef}
\end{eqnarray}
follows by the substitutions $t=n^{\mu}x,\,\mu z =s$ with an appropriately fixed parameter $\mu > 0$ for arbitrary natural numbers $n$
\begin{eqnarray}
\frac{1}{n^s} &=& \frac{1}{{\rm
\Gamma}\left(\frac{s}{\mu}\right)}\int_{0}^{+\infty}dx\,x^{\frac{s}{\mu}-1}
\E^{-n^{\mu}x},\quad \left({\rm
Re}\left(\frac{s}{\mu}\right)>0\right).\label{gammadefcons}
\end{eqnarray}
Inserting this into the sum representation (\ref{eulerprod}) and changing the order of summation and integration, we obtain for choice $\mu=1$ of the parameter using the sum evaluation of the geometric series
\begin{eqnarray}
\zeta(s) &=& \frac{1}{{\rm
\Gamma}(s)}\int_{0}^{+\infty}dp\,\frac{p^{s-1}}{\E^{p}-1},\quad
\left({\rm Re}(s) > 1\right),\label{zetarep1}
\end{eqnarray}
and for choice $\mu=2$ with substitution $p=\pi q^2$ of the integration variable (see \cite{rie} and, e.g.,
\cite{titch,chand,edw,patt,ivic})
\begin{eqnarray}
\zeta(s) &=& \frac{\pi^{\frac{s}{2}}s}{{\rm
\Gamma}\left(\frac{s}{2}+1\right)}\int_{0}^{+\infty}dq\,q^{s-1}
\sum_{n=1}^\infty\exp\left(-\pi n^2q^2\right),\quad \left({\rm
Re}(s) > 1\right). \label{zetarep2}
\end{eqnarray}
Other choice of $\mu$ seems to be of lesser importance. Both representations (\ref{zetarep1}) and (\ref{zetarep2}) are closely related to a Mellin transform $\hat{f}(s)$ of a function $f(t)$ which together with its inversion is generally defined by (e.g., \cite{lang,erd,zem,brpr,bert})
\begin{eqnarray}
f(t) \rightarrow \hat{f}(s) &\equiv & \int_{0}^{+\infty} dt\,t^{s-1}\,f(t),\quad
\Rightarrow \quad f(\lambda t) \rightarrow \frac{1}{\lambda^s}\hat{f}(s), \quad \left(\,\lambda >0\,\right),\nonumber\\ \hat{f}(s) \rightarrow f(t) &=& \frac{1}{\I 2\pi}\int_{c-\I
\infty}^{c+\I \infty}ds\,t^{-s}\,\hat{f}(s), \quad
\Rightarrow \quad \hat{f}(s-s_0) \rightarrow t^{s_0}f(t),\label{mellin}
\end{eqnarray}
where $c$ is an arbitrary real value within the convergence strip of $\hat{f}(s)$ in complex $s$-plane.
The Mellin transform $\hat{f}(s)$ of a function $f(t)$ is closely related to the Fourier transform $\tilde{\varphi}(y)$ of the function $\varphi(x) \equiv f(\E^x)$ by variable substitution $t=\E^{x}$ and $y=\I s$. Thus the Riemann zeta function $\zeta(s)$ can be represented, substantially (i.e., up to factors depending on $s$), as the Mellin transforms of the functions $f(t)=\sum_{n=1}^\infty \E^{-nt}=\frac{1}{\E^{t}-1}$ or of $f(t)=\sum_{n=1}^\infty\exp\left(-\pi n^2 t^2\right)$, respectively. The kernels of the Mellin transform are the eigenfunctions of the differential operator $t\frac{\partial}{\partial t}$ to eigenvalue $s-1$ or, correspondingly, of the integral operator $\exp\left(\alpha\, t\frac{\partial}{\partial t}\right)$ of the multiplication of the argument of a function by a factor $\E^{\alpha}$ (scaling of argument).
Both representations (\ref{zetarep1}) and (\ref{zetarep2}) can be used for the derivation of further representations of the Riemann zeta function and for the analytic continuation. The analytic continuation of the Riemann zeta function can also be obtained using the Euler-Maclaurin
summation formula for the series in (\ref{eulerprod}) (e.g., \cite{edw,riben,lang}).
Using the Poisson summation formula, one can transform the representation (\ref{zetarep2}) of the Riemann zeta function to the following form
\begin{eqnarray}
\zeta(s) &=&
\frac{\pi^{\frac{s}{2}}}{\left(\frac{s}{2}\right)!(s-1)}\left\{\frac{1}{2}-s(1-s) \int_{1}^{+\infty}dq\,\frac{q^s+q^{1-s}}{q}\sum_{n=1}^\infty \exp\left(-\pi n^2
q^2\right)\right\}.\label{zetarep3}
\end{eqnarray}
This is known \cite{rie,titch,chand,edw,patt,ivic} but for convenience and due to the importance of this representation for our purpose we give a derivation in Appendix A. From (\ref{zetarep3})
which is now already true for arbitrary complex $s$ and, therefore, is an analytic continuation of the representations (\ref{eulerprod}) or (\ref{zetarep2}) we see that the Riemann zeta function satisfies a functional equation for the transformation of the argument $s\rightarrow 1-s$. In simplest form it appears by 'renormalizing' this function via introduction of the xi function $\xi(s)$ defined by Riemann according to \cite{rie} and to \cite{edw,apost}\footnote{Riemann \cite{rie} defines it more specially for argument $s=\frac{1}{2}+\I t$ and writes it $\xi(t)$ with real $t$ corresponding to our $\xi\left(\frac{1}{2}+\I t\right)$. Our definition agrees, e.g., with Eq. (1) in Section 1.8 on p. 16 of Edwards \cite{edw} and with \cite{apost} and many others.}
\begin{eqnarray}
\xi(s) & \equiv & \frac{(s-1)\left(\frac{s}{2}\right)!} {\pi^{\frac{s}{2}}}\zeta(s),\label{xirep1}
\end{eqnarray}
and we obtain for it the following representation converging in the whole complex plane of $s$ (e.g., \cite{rie,chand,edw,ivic,patt})
\begin{eqnarray}
\xi(s) &=& \frac{1}{2}-s(1-s)\int_{1}^{+\infty}dp\,\frac{p^s+p^{1-s}}{p}\sum_{n=1}^\infty
\exp\left(-\pi n^2 p^2\right),\label{xirep2}
\end{eqnarray}
with the 'normalization'
\begin{eqnarray}
\xi(0) \;=\; \xi(1) \;=\; -\zeta(0)\;=\; \frac{1}{2}.\label{normxi}
\end{eqnarray}
For $s=\frac{1}{2}$ the xi function and the zeta function possess the (likely transcendental) values
\begin{eqnarray}
\xi\left(\frac{1}{2}\right) \;=\; - \frac{\left(\frac{1}{4}\right)!}{2\pi^{\frac{1}{4}}}\,\zeta
\left(\frac{1}{2}\right)\;=\; 0.4971207782,\quad \zeta\left(\frac{1}{2}\right) \;=\; -1.4603545088 \label{xizero}
\end{eqnarray}
Contrary to the Riemann zeta function $\zeta(s)$ the function $\xi(s)$ is an entire function. The only singularity of $\zeta(s)$ which is the simple pole at $s=1$, is removed by multiplication of $\zeta(s)$ with $s-1$ in the definition (\ref{xirep1}) and the trivial zeros of $\zeta(s)$ at $s=-2n,\,(n=1,2,\ldots)$ are also removed by its multiplication with $\left(\frac{s}{2}\right)! \equiv \,{\rm \Gamma}\left(\frac{s}{2}+1\right)$ which possesses simple poles there.
The functional equation
\begin{eqnarray}
\xi(s) &=& \xi(1-s),\label{funceqxi}
\end{eqnarray}
from which follows for the $n$-th derivatives
\begin{eqnarray}
\xi^{(n)}(s) &=& (-1)^n \xi^{(n)}(1-s),\quad \Rightarrow \quad \xi^{(2m+1)}\left(\frac{1}{2}\right) \;=\;0,\quad (n,m=0,1,2,\ldots),\label{funceqxider}
\end{eqnarray}
and which expresses that $\xi(s)$ is a symmetric function with respect to $s=\frac{1}{2}$ as it
is immediately seen from (\ref{xirep2}) and as it was first derived by Riemann \cite{rie}. It can be easily converted into the following functional equation for the Riemann zeta function $\zeta(s)$\footnote{According to Havil \cite{havil}, (p. 193), already Euler correctly conjectured this relation for the zeta function $\zeta(s)$ which is equivalent to relation (\ref{funceqxi}) for the function $\xi(s)$ but could not prove it. Only Riemann proved it first.}
\begin{eqnarray}
\zeta(s) &=& \frac{\left(2\pi\right)^{s}}{2\,(s-1)!\cos\left(\pi\frac{s}{2}\right)}\,\zeta(1-s).\label{funceqzeta}
\end{eqnarray}
Together with $\xi\left(s\right) =6 \left(\xi\left(s^*\right)\right)^*$ we find by combination with (\ref{funceqxi})
\begin{eqnarray}
\xi\left(s\right) \;=\; \xi(1-s) \;=\; \left(\xi\left(1-s^*\right)\right)^* \;=\;\left(\xi\left(s^*\right)\right)^*,\label{xisymmrel}
\end{eqnarray}
that combine in simple way, function values for 4 points $(s,1-s,1-s^*,s^*)$ of the complex plane.
Relation (\ref{xisymmrel}) means that in contrast to the function $\zeta(s)$ which is only real-valued on the real axis the function $\xi(s)$ becomes real-valued on the real axis ($s=s^*$) and on the imaginary axis ($s=-s^*$).
As a consequence of absent zeros of the Riemann zeta function $\zeta(\sigma +\I t)$ for $\sigma \equiv {\rm Re}(s) > 1$ together with the functional relation (\ref{funceqzeta}) follows that all nontrivial zeros of this function have to be within the strip $0\le \sigma \le 1$ and the Riemann hypothesis asserts that all zeros of the related xi function $\xi(s)$ are positioned on the so-called critical line $s=\frac{1}{2}+\I t,\,(-\infty < t < +\infty)$. This is, in principle, well known.
We use the functional equation (\ref{funceqxi}) for a simplification of the notations in the following considerations and displace the imaginary axis of the complex variable $s=\sigma +\I t$ from $\sigma=0$ to the value $\sigma=\frac{1}{2}$ by introducing the entire function ${\mit \Xi}(z)$ of the complex variable $z=x+\I y$ as follows
\begin{eqnarray}
{\mit \Xi}\left(z\right) &\equiv & \xi\left(\frac{1}{2}+z\right),\quad z=x+\I y = \sigma-\frac{1}{2}+\I t = s -\frac{1}{2},\label{capxidef}
\end{eqnarray}
with the 'normalization' (see (\ref{normxi}) and (\ref{xizero}))
\begin{eqnarray}
{\mit \Xi}\left(\pm\frac{1}{2}\right) \;=\; \frac{1}{2},\quad {\mit \Xi}\left(0\right) \;=\; \xi\left(\frac{1}{2}\right) \;\approx\; 0.4971207782.\label{normcapxi}
\end{eqnarray}
following from (\ref{normxi}). Thus the full relation of the Xi function $\Xi(z)$ to the Riemann zeta function $\zeta(s)$ using definition (\ref{xirep1}) is
\begin{eqnarray}
{\mit \Xi}(z) &=& \frac{\left(z-\frac{1}{2}\right)\left(\frac{1+2z}{4}\right)!}{\pi^\frac{1+2z}{4}}
\zeta\left(\frac{1}{2}+z\right).\label{xizeta}
\end{eqnarray}
We emphasize again that the argument displacement (\ref{capxidef}) is made in the following only for convenience of notations and not for some more principal reason.
The functional equation (\ref{funceqxi}) together with (\ref{funceqxider}) becomes
\begin{eqnarray}
{\mit \Xi}\left(z\right) &=& {\mit \Xi}\left(-z\right),\quad {\mit \Xi}^{(n)}\left(z\right) \;=\; (-1)^n {\mit \Xi}^{(n)}\left(-z\right), \label{funceqcapxi}
\end{eqnarray}
and taken together with the symmetry for the transition to complex conjugated variable
\begin{eqnarray}
{\mit \Xi}\left(z\right) \;=\; {\mit \Xi}\left(-z\right) \;=\; \left({\mit \Xi}\left(-z^*\right)\right)^* \;=\; \left({\mit \Xi}\left(z^*\right)\right)^*.\label{xisymm}
\end{eqnarray}
This means that the Xi function ${\mit \Xi}\left(z\right)$ becomes real-valued on the imaginary axis $z=\I\,y$ which becomes the critical line in the new variable $z$
\begin{eqnarray}
{\mit \Xi}\left(\I y\right) \;=\; {\mit \Xi}\left(-\I y\right) \;=\; \left({\mit \Xi}\left(\I y\right)\right)^* \;=\; \left({\mit \Xi}\left(-\I y\right)\right)^*.
\end{eqnarray}
Furthermore, the function ${\mit \Xi}(z)$ becomes a symmetrical function and a real-valued one on the real axis $z=x$
\begin{eqnarray}
{\mit \Xi}\left(x\right) = {\mit \Xi}\left(-x\right) = \left({\mit \Xi}\left(-x\right)\right)^* = \left({\mit \Xi}\left(x\right)\right)^*.
\end{eqnarray}
In contrast to this the Riemann zeta function $\zeta(s)$ the function is not a real-valued function on the
critical line $s=\frac{1}{2}+\I t$ and is real-valued but not symmetric on the real axis.
This is represented in Fig. 2.1. (calculated with "Mathematica 6" such as the further figures too).
\begin{figure}[h]
\includegraphics[width=16.0cm]{d1.pdf}\hspace{1cm}
\caption{\foot Real and imaginary part and absolute value of Riemann zeta function on critical line. \newline \scriptsize{The position of the zeros of the whole function $\zeta\left(\frac{1}{2}+\I t\right)$ on the critical line are shown by grid lines. One can see that not all zeros of the real part are also zeros of the imaginary part and vice versa. The figures are easily to generate by program "Mathematica" and are published in similar forms already in literature.}}
\end{figure}
We see that not all of the zeros of the real part ${\rm Re}\left(\zeta\left(\frac{1}{2}+\I t\right)\right)$ are also zeros of the imaginary part ${\rm Im}\left(\zeta\left(\frac{1}{2}+\I t\right)\right)$ and, vice versa, that not all of the zeros of the imaginary part are also zeros of the real part and thus genuine zeros of the function $\zeta\left(\frac{1}{2}+\I t\right)$ which are signified by grid lines. Between two zeros of the real part which are genuine zeros of $\zeta\left(\frac{1}{2}+\I t\right)$ lies in each case (exception first interval) an additional zero of the imaginary part, which almost coincides with a maximum of the real part.
Using (\ref{xirep2}) and definition (\ref{capxidef}) we find the following representation of ${\mit \Xi(z)}$
\begin{eqnarray}
{\mit \Xi}\left(z\right) &=& \frac{1}{2}-\left(\frac{1}{4}-z^2\right) \int_{1}^{+\infty}dq\,\frac{q^z+q^{-z}}{\sqrt{q}} \sum_{n=1}^\infty \exp\left(-\pi n^2 q^2\right).\label{capxirep1}
\end{eqnarray}
With the substitution of the integration variable $q= \E^u $ (see also (\ref{yex}) in Appendix A) representation (\ref{capxirep1}) is transformed to
\begin{eqnarray}
{\mit \Xi}\left(z\right) &\equiv & \frac{1}{2}-2\left(\frac{1}{4}-z^2\right)\int_{0}^{+\infty}du\,{\rm
ch}\left(uz\right)\,\E^{\frac{u}{2}} \sum_{n=1}^\infty\exp\left(-\pi n^2 \E^{2u}\right). \label{capxirep2}
\end{eqnarray}
In Appendix A we show that (\ref{capxirep2}) can be represented as follows (see also Eq. (2) on p. 17 in \cite{edw} which possesses a similar principal form)
\begin{eqnarray}
{\mit \Xi}\left(z\right) &=& \int_{0}^{+\infty}du\,{\mit
\Omega}\left(u\right)\,{\rm ch}\left(uz\right),\label{capxirep3}
\end{eqnarray}
with the following explicit form of the function ${\mit \Omega}\left(u\right)$ of the real variable $u$
\begin{eqnarray}
{\mit \Omega}\left(u\right) &\equiv & 4\E^{\frac{u}{2}}\sum_{n=1}^\infty \pi n^2\E^{2u}\left(2\pi
n^2\E^{2u}-3\right)\exp\left(-\pi n^2 \E^{2u}\right)\;>\;0,\quad (-\infty < u < +\infty). \label{defcapomega}
\end{eqnarray}
The function ${\mit \Omega}\left(u\right)$ is symmetric
\begin{eqnarray}
{\mit \Omega}\left(u\right) = +{\mit \Omega}\left(-u\right) = {\mit
\Omega}\left(\left|u\right|\right),\label{omegasym}
\end{eqnarray}
that means it is an even function although this is not immediately seen from representation (\ref{defcapomega})\footnote{It was for us for the first time and was very surprising to meet a function where its symmetry was not easily seen from its explicit representation. However, if we substitute in (\ref{defcapomega}) $u\rightarrow -u$ and calculate and plot the part of ${\mit \Omega}(u)$ for $u\ge 0$ with the obtained formula then we need much more sum terms for the same accurateness than in case of calculation with (\ref{defcapomega}).)}. We prove this in Appendix B. Due to this symmetry, formula (\ref{capxirep3}) can be also represented by
\begin{eqnarray}
{\mit \Xi}\left(z\right) &=& \frac{1}{2}\int_{-\infty}^{+\infty}du\,{\mit
\Omega}\left(u\right)\,{\rm ch}\left(uz\right) \;=\; \frac{1}{2}\int_{-\infty}^{+\infty}du\,{\mit
\Omega}\left(u\right)\,\E^{uz}.\label{capxirep4}
\end{eqnarray}
In the formulation of the right-hand side the function ${\mit \Xi}\left(z\right)$ appears as analytic continuation of the Fourier transform of the function ${\mit \Omega}\left(u\right)$ written with imaginary argument $z=\I y$ or, more generally, with substitution $z\rightarrow \I z'$ and complex $z'$. From this follows as inversion of the integral transformation (\ref{capxirep4}) using (\ref{omegasym})
\begin{eqnarray}
{\mit \Omega}\left(u\right) &=& \frac{1}{\pi}\int_{-\infty}^{+\infty}dy\,{\mit \Xi}\left(\I y\right)\E^{-\I u y} \;=\; \frac{1}{\pi}\int_{-\infty}^{+\infty}dy\,{\mit \Xi}\left(\I y\right)\cos\left(u y\right),
\end{eqnarray}
or due to symmetry of the integrand in analogy to (\ref{capxirep3})
\begin{eqnarray}
{\mit \Omega}\left(u\right) &=& \frac{2}{\pi}\int_{0}^{+\infty}dy\,{\mit \Xi}\left(\I y\right)\cos\left(u y\right),
\end{eqnarray}
where ${\mit \Xi}\left(\I y\right)$ is a real-valued function of the variable $y$ on the imaginary axis
\begin{eqnarray}
{\mit \Xi}\left(\I y\right) &=& \int_{0}^{+\infty}du\,{\mit
\Omega}\left(u\right)\cos\left(uy\right),\quad \left({\mit \Xi}\left(\I y\right)\right)^*={\mit \Xi}\left(\I y\right),\label{xisym}
\end{eqnarray}
due to (\ref{capxirep3}).
A graphical representation of the function $\Omega(u)$ and of its first derivatives $\Omega^{(1)}(u),(n=1,2,3)$ is given in Fig. 2.2.
\begin{figure}[h]
\includegraphics[width=16.0cm]{g1.pdf}\hspace{1cm}
\caption{\foot Function $\Omega(u)$ and its first derivative $\Omega^{(1)}(u),\,(n=1,2,3)$ (see (\ref{capxirep3}) and (\ref{capxirep5})).\newline \scriptsize{The function ${\mit \Omega}(u)$ is positive for $ 0 \le u < +\infty $ and since its first derivative ${\mit \Omega}^{(1)}(u)$ is negative for $0 < u < +\infty$ the function ${\mit \Omega}(u)$ is monotonically decreasing on the real positive axis. It vanishes in infinity more rapidly than any exponential function with a polynomial in the exponent.}}
\end{figure}
The function ${\mit \Omega}\left(u\right)$ is monotonically decreasing for $0\le u < +\infty $ due to the non-positivity of its first derivative ${\mit \Omega}^{(1)}(u) \equiv \frac{\partial {\mit \Omega(u)}}{\partial u}$ which explicitly is (see also Appendix A)
\begin{eqnarray}
{\mit \Omega}^{(1)}(u) &=& -2 \E^{\frac{u}{2}}\sum_{n=1}^\infty \pi n^2\E^{2u}\left(8\left(\pi n^2\E^{2u}\right)^2-30\pi n^2\E^{2u}+15\right)\exp\left(-\pi n^2 \E^{2u}\right) \nonumber\\ &\le & 0,\qquad \left(0\le u < +\infty\right),\label{dercapomega}
\end{eqnarray}
with one relative minimum at $u_{\rm min}=0.237266$ of depth ${\mit \Omega}^{(1)}\left(u_{\rm min}\right)= -4.92176$. Moreover, it is very important for the following that due to presence of factors $\exp\left(-\pi n^2 \E^{2u}\right)$ in the sum terms in (\ref{defcapomega}) or in (\ref{dercapomega}) the functions ${\mit \Omega}(u)$ and ${\mit \Omega}^{(1)}(u)$ and all their higher derivatives are very rapidly decreasing for $u\rightarrow +\infty$, more rapidly than any exponential function with a polynomial of $u$ in the argument. In this sense the function ${\mit \Omega}(u)$ is more comparable with functions of finite support which vanish from a certain $u \ge u_0$ on than with any exponentially decreasing function.
From (\ref{omegasym}) follows immediately that the function ${\mit \Omega}^{(1)}(u)$ is antisymmetric
\begin{eqnarray}
{\mit \Omega}^{(1)}\left(u\right) = -{\mit \Omega}^{(1)}\left(-u\right) = \frac{u}{|u|}\frac{\partial}{\partial |u|}{\mit \Omega}\left(\left|u\right|\right),\quad \Rightarrow \quad {\mit \Omega}^{(1)}\left(0\right) =0,\label{omega1sym}
\end{eqnarray}
that means it is an odd function.
It is known that smoothness and rapidness of decreasing in infinity of a function change their role in Fourier transformations. As the Fourier transform of the smooth (infinitely continuously differentiable) function ${\mit \Omega}(u)$ the Xi function on the critical line ${\mit \Xi}(\I y)$ is rapidly decreasing in infinity. Therefore it is not easy to represent the real-valued function ${\mit \Xi}(\I y)$ with its rapid oscillations under the envelope of rapid decrease for increasing variable $y$ graphically in a large region of this variable $y$. An appropriate real amplification envelope is seen from (\ref{xizeta}) to be $\alpha(y)=\left|\frac{1}{\left(\frac{1+\I\,2y}{4}\right)!}\right|\frac{2\pi^{\frac{1}{4}}}{\sqrt{1+4y^2}}$ which rises ${\mit \Xi}(\I y)$ to the level of the Riemann zeta function $\zeta\left(\frac{1}{2}+\I t\right)$ on the critical line $z=\I y$. This is shown in Fig. 2.3.
\begin{figure}[h]
\includegraphics[width=16.0cm]{d3.pdf}\hspace{1cm}
\caption{\foot Xi Function ${\mit \Xi}(\I y)$ on the imaginary axis $z=\I y$ (corresponding to $s=\frac{1}{2}+\I y$).
\newline \scriptsize{The envelope over the oscillations of the real-valued function ${\mit \Xi}(\I y)$ decreases extremely rapidly with increase of the variable $y$ in the shown intervals. This behavior makes it difficult to represent this function graphically for large intervals of the variable $y$. By an enhancement factor which rises the amplitude to the level of the zeta function $\zeta(s)$ we may see the oscillations under the envelope (last partial picture). A similar picture one obtains for the modulus of the Riemann zeta function $\left|\zeta\left(\frac{1}{2}+\I y\right)\right|$ only with our negative parts folded to the positive side of the abscissa, i.e. $\left|{\mit \Xi}(\I y)\right|=\left|\zeta\left(\frac{1}{2}+\I y\right)\right|$ (see also Fig. 2.1 (last partial picture)). The given values for the zeros at $\frac{1}{2}\pm \I y_n$ were first calculated by J.-P. Gram in 1903 up to $y_{15}$ \cite{edw}. We emphasize here that the shown very rapid decrease of the Xi function at the beginning of $y$ and for $y\rightarrow \pm \infty$ is due to the 'very high' smoothness of ${\mit \Omega}(u)$ for arbitrary $u$.}}
\end{figure}
The partial picture for $\alpha(y){\mit \Xi}(\I y)$ in Fig. 2.3. with negative part folded up is identical with the absolute value $\left|\zeta\left(\frac{1}{2}+\I t\right)\right|$ of the Riemann zeta function $\zeta(s)$ on the imaginary axis $s=\frac{1}{2}+\I t$ (fourth partial picture in Fig. 2.1).
We now give a representation of the Xi function by the derivative of the Omega function.
Using $\,{\rm ch}\left(uz\right)= \frac{1}{z}\frac{\partial}{\partial u}\,{\rm sh}\left(uz\right)$ one obtains from (\ref{capxirep3}) by partial integration the following alternative representation of the function ${\mit \Xi}\left(z\right)$
\begin{eqnarray}
{\mit \Xi}\left(z\right) &=& -\frac{1}{z}\int_{0}^{+\infty}du\,{\mit \Omega}^{(1)}(u)\,{\rm sh}\left(uz\right),\label{capxirep5}
\end{eqnarray}
that due to antisymmetry of $\,{\mit \Omega}^{(1)}(u)$ and $\,{\rm sh}\left(uz\right)$ with respect to $u\rightarrow -u$ can also be written
\begin{eqnarray}
{\mit \Xi}\left(z\right) &=& -\frac{1}{2z}\int_{-\infty}^{+\infty}du\,{\mit
\Omega}^{(1)}(u)\,{\rm sh}\left(uz\right) \;=\; -\frac{1}{2z}\int_{-\infty}^{+\infty}du\,{\mit
\Omega}^{(1)}\left(u\right)\,\E^{uz}. \label{capxirep6}
\end{eqnarray}
Figure 2.2 gives a graphical representation of the function ${\mit \Omega}\left(u\right)$ and of its first derivative ${\mit \Omega}^{(1)}(u) \equiv \frac{\partial {\mit \Omega}}{\partial u}(u)$ which due to rapid convergence of the sums is easily to generate by computer. One can express ${\mit \Xi}\left(z\right)$ also by higher derivatives ${\mit \Omega}^{(n)}(u) \equiv \frac{\partial^n {\mit \Omega}}{\partial u^n}(u)$ of the Omega function ${\mit \Omega}(u)$
according to
\begin{eqnarray}
{\mit \Xi}\left(z\right) &=& \frac{1}{z^{2m}}\int_{0}^{+\infty}du\,{\mit \Omega}^{(2m)}(u)\,{\rm ch}(uz) \nonumber\\ &=& -\frac{1}{z^{2m+1}}\int_{0}^{+\infty}du\,{\mit \Omega}^{(2m+1)}(u)\,{\rm sh}(uz),\quad \left(m=0,1,2,\ldots \right),\label{capxirep7}
\end{eqnarray}
with the symmetries of the derivatives of the function ${\mit \Omega}(u)$ for $u\leftrightarrow -u$
\begin{eqnarray}
{\mit \Omega}^{(2m)}(u) \;=\; +{\mit \Omega}^{(2m)}(-u),\quad {\mit \Omega}^{(2m+1)}(u) \;=\; -{\mit \Omega}^{(2m+1)}(-u),\;\; \Rightarrow \;\; {\mit \Omega}^{(2m+1)}(0) \;=\; 0,\quad \left(m=0,1,\ldots\right).\quad
\end{eqnarray}
This can be seen by successive partial integrations in (\ref{capxirep3}) together with complete induction. The functions ${\mit \Omega}^{(n)}(u)$ in these integral transformations are for $n \ge 1$ not monotonic functions.
We mention yet another representation of the function ${\mit \Xi}(z)$. Using the transformations
\begin{eqnarray}
t_n\equiv \pi n^2 \E^{2u}, \quad \Rightarrow \quad dt_n \;=\;2\pi n^2 \E^{2u}du \;=\; 2t_ndu,\quad du\;=\;\frac{dt_n}{2u},
\end{eqnarray}
the function ${\mit \Xi}(z)$ according to (\ref{capxirep4}) with the explicit representation of the function ${\mit \Omega}(u)$ in (\ref{defcapomega}) can now be represented in the form
\begin{eqnarray}
{\mit \Xi}(z) &=& \sum_{n=1}^{\infty}\frac{1}{(\pi n^2)^{\frac{1}{4}}}\left\{2\left(\left(\pi n^2\right)^{-\frac{\scr{z}}{2}}{\rm \Gamma}\left(\frac{9}{4}+\frac{z}{2},\pi n^2\right)+\left(\pi n^2\right)^{\frac{\scr{z}}{2}}{\rm \Gamma}\left(\frac{9}{4}-\frac{z}{2},\pi n^2\right)\right)\right. \nonumber\\&& \left. -3\left(\left(\pi n^2\right)^{-\frac{\scr{z}}{2}}{\rm \Gamma}\left(\frac{5}{4}+\frac{z}{2},\pi n^2\right)+\left(\pi n^2\right)^{\frac{\scr{z}}{2}}{\rm \Gamma}\left(\frac{5}{4}-\frac{z}{2},\pi n^2\right)\right)\right\},\label{xiincomplgamma}
\end{eqnarray}
where ${\rm \Gamma}(\alpha,x)$ denotes the incomplete Gamma function defined by (e.g., \cite{bate1,nist,paris}
\begin{eqnarray}
{\rm \Gamma}(\alpha,x) & \equiv & \int_{x}^{+\infty}dt\,\E^{-t}t^{\alpha-1} \;\equiv\; {\rm \Gamma}(\alpha)-{\rm \gamma}(\alpha,x).
\end{eqnarray}
However, we did not see a way to prove the Riemann hypothesis via the representation (\ref{xiincomplgamma}).
The Riemann hypothesis for the zeta function $\zeta(s = \sigma +\I t)$ is now equivalent to the hypothesis that all zeros of the related entire function ${\mit \Xi}\left(z=x+\I y\right)$ lie on the imaginary axis $z=\I y$ that means on the line to real part $x=0$ of $z=x+\I y$ which becomes now the critical line. Since the zeta function $\zeta(s)$ does not possess zeros in the convergence region $\sigma >1$ of the Euler product (\ref{eulerprod}) and due to symmetries (\ref{omegasym}) and (\ref{xisym}) it is only necessary to prove that ${\mit \Xi}(z)$ does not
possess zeros within the strips $-\frac{1}{2}\le x <0$ and $0 < x \le +\frac{1}{2}$ to both sides of the imaginary axis $z=\I y$ where for symmetry the proof for one of these strips would be already sufficient. However, we will go another way where the restriction to these strips does not play a role for the proof.
\setcounter{chapter}{3}
\setcounter{equation}{0}
\section*{3. Application of second mean-value theorem of calculus to Xi function}
After having accepted the basic integral representation (\ref{capxirep3}) of the entire function ${\mit \Xi}(z)$ according to
\begin{eqnarray}
{\mit \Xi}(z) & \equiv & \int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(uz),\label{basiccapxi0}
\end{eqnarray}
with the function ${\mit \Omega}(u)$ explicitly given in (\ref{defcapomega}) we concentrate us on its further treatment. However, we do this not with this specialization for the real-valued function ${\mit \Omega}(u)$ but with more general suppositions for it. Expressed by real part $U(x,y)$ and imaginary part $V(x,y)$ of ${\mit \Xi}(z)$
\begin{eqnarray}
{\mit \Xi}(x+\I y) \; \equiv \; U(x,y) +\I V(x,y), \quad U(x,y) \;=\; \left(U(x,y)\right)^*, \quad V(x,y) \;=\; \left(V(x,y)\right)^*,
\end{eqnarray}
we find from (\ref{basiccapxi0})
\begin{eqnarray}
U(x,y) \;=\; \int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(ux)\cos(uy), \quad V(x,y) \;=\; \int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm sh}(ux)\sin(uy).\label{reimpartxi}
\end{eqnarray}
We suppose now as necessary requirement for ${\mit \Omega}(u)$ and satisfied in the special case (\ref{defcapomega})
\begin{eqnarray}
{\mit \Omega}(u) \; \ge\; 0,\quad (0\le u < +\infty),\quad \Rightarrow \quad {\mit \Omega}(0) \; > \; 0.\label{condsecmvth}
\end{eqnarray}
Furthermore, ${\mit \Xi}(z)$ should be an entire function that requires that the integral (\ref{basiccapxi0}) is finite for arbitrary
complex $z$ and therefore that ${\mit \Omega}(u)$ is rapidly decreasing in infinity, more precisely
\begin{eqnarray}
\lim_{u\rightarrow +\infty}\frac{{\mit \Omega}(u)}{\exp\left(-\lambda u\right)} = 0, \quad 0 < 0 \le \lambda < +\infty ,\label{rapiddecr}
\end{eqnarray}
for arbitrary $\lambda \ge 0$. This means that the function ${\mit \Omega}(u)$ should be a nonsingular function which is rapidly decreasing in infinity, more rapidly than any exponential function $\E^{-\lambda u}$ with arbitrary $\lambda >0$. Clearly, this is satisfied for the special function ${\mit \Omega}(u)$ in (\ref{defcapomega}).
Our conjecture for a longer time was that all zeros of ${\mit \Xi}(z)$ lie on the imaginary axis $z=\I y$ for a large class of functions ${\mit \Omega}(u)$ and that this is not very specific for the special function ${\mit \Omega}(u)$ given in (\ref{defcapomega}) but is true for a much larger class. It seems that to this class belong all non-increasing functions ${\mit \Omega}(u)$, i.e such functions for which holds ${\mit \Omega}^{(1)}(u) \le 0$ for its first derivative and which rapidly decrease in infinity. This means that they vanish more rapidly in infinity than any power functions $|u|^{-n},\,(n=1,2,\ldots)$ (practically they vanish exponentially). However, for the convergence of the integral (\ref{basiccapxi0}) in the whole complex $z$-plane it is necessary that the functions have to decrease in infinity also more rapidly than any exponential function $\exp(-\lambda u)$ with arbitrary $\lambda >0$ expressed in (\ref{rapiddecr}). In particular, to this class belong all rapidly decreasing functions ${\mit \Omega}(u)$ which vanish from a certain $u\ge u_0$ on and which may be called non-increasing finite functions (or functions with compact support). On the other side, continuity of its derivatives ${\mit \Omega}^{(n)}(u),\,(n=1,2,\ldots)$ is not required. The modified Bessel functions ${\rm I}_{\nu}(z)$ 'normalized' to the form of entire functions $\left(\frac{2}{z}\right)^{\nu}{\rm I}_{\nu}(z)$ for $\nu \ge \frac{1}{2}$ possess a representation of the form (\ref{basiccapxi0}) with a function ${\mit \Omega}(u)$ which vanishes from $u \ge 1$ on but a number of derivatives of ${\mit \Omega}(u)$ for the functions is not continuous at $u=1$ depending on the index $\nu$. It is valuable that here an independent proof of the property that all zeros of the modified Bessel functions ${\rm I}_{\nu}(u)$ lie on the imaginary axis can be made using their differential equations via duality relations. We intend to present this in detail in a later work.
Furthermore, to the considered class belong all monotonically decreasing functions with the described rapid decrease in infinity. The fine difference of the decreasing functions to the non-increasing functions ${\mit \Omega}(u)$ is that in first case the function ${\mit \Omega}(u)$ cannot stay on the same level in a certain interval that means we have ${\mit \Omega}^{(1)}(u) <0$ for all points $u > 0$ instead of ${\mit \Omega}^{(1)}(u) \le 0$ only. A function which decreases not faster than $\E^{-\lambda u}$ in infinity does not fall into this category as, for example, the function ${\rm sech}(z) \equiv \frac{1}{{\rm ch}(z)}$ shows. On the other side, also some simply calculable discrete superpositions such as $a_1\,{\rm ch}(u)+a_2\,{\rm ch}(2u)$ or $a_1\,{\rm ch}(z) +a_3\,{\rm ch}(3z)$ as function ${\mit \Xi}(z)$ with positive amplitudes $a_n$ do not provide a counterexample that the zeros lie outside the imaginary axis but show that if the amplitudes $a_n$ do not possess a definite sign then they may possess zeros outside the imaginary axis.
To apply the second mean-value theorem it is necessary to restrict us to a class of functions ${\mit \Omega}(u) \rightarrow f(u)$ which are non-increasing that means for which for all $u_1 < u_2$ in considered interval holds
\begin{eqnarray}
f(a) \ge f(u_1) \ge f(u_2) \ge f(b) \ge 0,\quad (a \le u_1 \le u_2 \le b),
\end{eqnarray}
or equivalently in more compact form
\begin{eqnarray}
f^{(1)}(u) & \le & 0, \quad (a \le u \le b).
\end{eqnarray}
In case of $f^{(1)}(u)=0$ for certain $u$ the next higher non-vanishing derivative should be negative.
The monotonically decreasing functions in the interval $a\le u \le b$, in particular, belong to the class of non-increasing functions with the fine difference that here
\begin{eqnarray}
f(a) > f(u_1) > f(u_2) > f(b) > 0,\quad (a < u_1 < u_2 < b),
\end{eqnarray}
is satisfied. If furthermore $g(u)$ is a continuous function in the interval $a \le u \le b$ the second mean-value theorem (often called theorem of Bonnet (1867) or Gauss-Bonnet theorem) states an equivalence for the following integral on the left-hand side to the expression on the right-hand side according to (see some monographs about calculus or real analysis; we recommend the monographs of Courant \cite{courant} (Appendix to chap IV) and of Widder \cite{widder} who called it Weierstrass form of Bonnet's theorem (chap. 5, \S 4))
\begin{eqnarray}
\int_{a}^{b}du\,f(u)g(u) &=& f(a)\int_{a}^{u_0}du\,g(u) +f(b)\int_{u_0}^{b}du\,g(u),\quad (a\le u_0 \le b),\label{secmeanvth}
\end{eqnarray}
where $u_0 $ is a certain value within the interval boundaries $a < b $ which as a rule we do not exactly know. It holds also for non-decreasing functions which include the monotonically increasing functions as special class in analogous way. The proof of the second mean-value theorem is comparatively simple by applying a substitution in the (first) mean-value theorem of integral calculus \cite{courant,widder}.
Applied to our function $f(u) = {\mit \Omega}(u)$ which in addition should rapidly decrease in infinity according to (\ref{rapiddecr}) this means in connection with monotonic decrease that it has to be positively semi-definite if ${\mit \Omega}(0) > 0$ and therefore
\begin{eqnarray}
{\mit \Omega}(0) \ge {\mit \Omega}(u) \ge 0,\quad {\mit \Omega}^{(1)}(u) \le 0,\quad \left(0\le u \le +\infty\right),\quad
{\mit \Omega}(u\rightarrow +\infty) \rightarrow 0,\label{secmeanvth0}
\end{eqnarray}
and the theorem (\ref{secmeanvth}) takes on the form
\begin{eqnarray}
\int_{0}^{+\infty}du\,{\mit \Omega}(u) g(u) &=& {\mit \Omega}(0)\int_{0}^{u_0}du\,g(u),\quad (0 \le u_0 < +\infty), \label{secmeanvth1}
\end{eqnarray}
where the extension to an upper boundary $b\rightarrow +\infty$ in (\ref{secmeanvth}) for $f(+\infty)=0$ and in case of existence of the integral is unproblematic.
If we insert in (\ref{secmeanvth}) for $g(u)$ the function ${\rm ch}(uz)$ which apart from the real variable $u$ depends in parametrical way on the complex variable $z$ and is an analytic function of $z$ we find that $u_0$ depends on this complex parameter also in an analytic way as follows
\begin{eqnarray}
{\mit \Xi}(z) & \equiv & \int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(uz) \;=\; {\mit \Omega}(0)\int_{0}^{w_0(z)}du\,{\rm ch}(uz) \nonumber\\ &=& {\mit \Omega}(0)\frac{\,{\rm sh}\left(w_0(z)z\right)}{z}, \quad w_0(z) \;=\; u_0(x,y) +\I v_0(x,y),\label{secmeanvth2compl}
\end{eqnarray}
where $w_0(x+\I z)=u_0(x,y)+\I v_0(x,y)$ is an entire function with $u_0(x,y)$ its real and $v_0(x,y)$ its imaginary part. The condition for zeros $z\neq 0$ is that ${\rm sh}\left(w_0(z)z\right){z}$ vanishes that leads to
\begin{eqnarray}
w_0(z)z &=& \left(u_0(x,y)+\I v_0(x,y)\right)\left(x+\I y\right) \;=\; \I n\pi, \quad (n=0,\pm 1,\pm 2,\ldots), \label{zcuvxy0}
\end{eqnarray}
or split in real and imaginary part
\begin{eqnarray}
u_0(x,y)x -v_0(x,y)y \;=\; 0,\label{zcuvxyre}
\end{eqnarray}
for the real part and
\begin{eqnarray}
u_0(x,y)y +v_0(x,y)x \;=\; n\pi, \quad (n=0,\pm 1,\pm 2,\ldots),\label{zcuvxyim}
\end{eqnarray}
for the imaginary part.
The multi-valuedness of the mean-value functions in the conditions (\ref{zcuvxy0}) or (\ref{zcuvxyim}) is an interesting phenomenon which is connected with the periodicity of the function $g(u)=\,{\rm ch}(uz)$ on the imaginary axis $z=\I y$ in our application (\ref{secmeanvth2compl}) of the second mean-value theorem (\ref{secmeanvth1}). To our knowledge this is up to now not well studied. We come back to this in the next Sections 4 and, in particular, Section 7 brings some illustrative clarity when we represent the mean-value functions graphically. At present we will say only that we can choose an arbitrary $n$ in (\ref{zcuvxyim}) which provides us the whole spectrum of zeros $z_1,z_2,\ldots$ on the upper half-plane and the corresponding spectrum of zeros $z_{-1}=-z_1,z_{-2}=-z_2,\ldots $ on the lower half-plane of ${\mathbb C}$ which as will be later seen lie all on the imaginary axis. Since in computer calculations the values of the Arcus Sine function are provided in the region from $-\frac{\pi}{2}$ to $+\frac{\pi}{2}$ it is convenient to choose $n=0$ but all other values of $n$ in (\ref{zcuvxyim}) lead to equivalent results.
One may represent the conditions (\ref{zcuvxyre}) and (\ref{zcuvxyim}) also in the following equivalent form
\begin{eqnarray}
u_0(x,y)\;=\; \frac{y}{x^2+y^2}n\pi,\quad v_0(x,y)\;=\; \frac{x}{x^2+y^2}n\pi,\label{zcuvxy2}
\end{eqnarray}
from which follows
\begin{eqnarray}
\left(u^2_0(x,y)+v^2_0(x,y)\right)\left(x^2+y^2\right) \;=\;(n\pi)^2,\quad \frac{v_0(x,y)}{u_0(x,y)} \;=\; \frac{x}{y}.\label{zcuvxy3}
\end{eqnarray}
All these forms (\ref{zcuvxyre})--(\ref{zcuvxy3}) are implicit equations with two variables $(x,y)$ which cannot be resolved with respect to one variable (e.g., in forms $y=y_k(x)$ for each fixed $n$ and branches $k$) and do not provide immediately the necessary conditions for zeros in explicit form but we can check that (\ref{zcuvxy2}) satisfies the Cauchy-Riemann equations as a minimum requirement
\begin{eqnarray}
\frac{\partial u_0(x,y)}{\partial x} \;=\; \frac{\partial v_0(x,y)}{\partial y},\quad \frac{\partial u_0(x,y)}{\partial y} \;=\; -\frac{\partial v_0(x,y)}{\partial x}.
\end{eqnarray}
We have to establish now closer relations between real and imaginary part $u_0(x,y)$ and $v_0(x,y)$ of the complex mean-value parameter $w_0(z=x+\I y)$. The first step in preparation to this aim is the consideration of the derived conditions on the imaginary axis.
\setcounter{chapter}{4}
\setcounter{equation}{0}
\section*{4. Specialization of second mean-value theorem to Xi function on imaginary axis}
By restriction to the real axis $y=0$ we find from (\ref{reimpartxi}) for the function ${\mit \Xi}(z)$
\begin{eqnarray}
{\mit \Xi}(x) &=& U(x,0), \quad V(x,0) \;=\; 0,
\end{eqnarray}
with the following two possible representations of $U(x,0)$ related by partial integration
\begin{eqnarray}
U(x,0) &=& \int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(ux) \;=\; -\frac{1}{x} \int_{0}^{+\infty}du\,{\mit \Omega}^{(1)}(u)\,{\rm sh}(ux) \; > \; 0.\label{reimpartofimxi}
\end{eqnarray}
The inequality $U(x,0)>0$ follows according to the supposition ${\mit \Omega}(u) \ge 0, {\mit \Omega}(0) >0$ from the non-negativity of the integrand that means from ${\mit \Omega}(u)\,{\rm ch}(ux) \ge 0$. Therefore, the case $y=0$ can be excluded from the beginning in the further considerations for zeros of $U(x,y)$ and $V(x,y)$.
We now restrict us to the imaginary axis $x=0$ and find from (\ref{reimpartxi}) for the function ${\mit \Xi}(z)$
\begin{eqnarray}
{\mit \Xi}(\I y) &=& U(0,y), \quad V(0,y) \;=\; 0.
\end{eqnarray}
with the following two possible representations of $U(0,y)$ related by partial integration
\begin{eqnarray}
U(0,y) &=& \int_{0}^{+\infty}du\,{\mit \Omega}(u)\cos(uy) \;=\; -\frac{1}{y} \int_{0}^{+\infty}du\,{\mit \Omega}^{(1)}(u)\sin(uy).\label{reimpartofimxi1}
\end{eqnarray}
From the obvious inequality
\begin{eqnarray}
-1 \; \le \; \cos(uy) \; \le +1,
\end{eqnarray}
together with the supposed positivity of ${\mit \Omega}(u)$ one derives from the first representation of $U(0,y)$ in (\ref{reimpartofimxi1}) the inequality
\begin{eqnarray}
-{\mit \Omega}_0 \; \le\; U(0,y) \; \le\;+{\mit \Omega}_0,\quad {\mit \Omega}_0 \; =\; U(0,0) \;\equiv \; \int_{0}^{+\infty}du\,{\mit \Omega}(u) \; \ge \; 0.\label{rexiineq1}
\end{eqnarray}
In the same way by the inequality
\begin{eqnarray}
-1 \; \le \; \sin(uy) \; \le +1,
\end{eqnarray}
one derives using the non-positivity of ${\mit \Omega}^{(1)}(u)$ (see (\ref{secmeanvth0})) together with the second representation of $U(0,y)$ in $(\ref{reimpartofimxi1})$ the inequality
\begin{eqnarray}
-{\mit \Omega}(0) \; \le \; U(0,y)y \; \le \; +{\mit \Omega}(0),\quad {\mit \Omega}(0) \;=\; -\int_{0}^{+\infty}du\,{\mit \Omega}^{(1)}(u) \; \ge \; 0.\label{rexiineq2}
\end{eqnarray}
which as it is easily seen does not depend on the sign of $y$. Therefore we have two non-negative parameters, the zeroth moment ${\mit \Omega}_0$ and the value ${\mit \Omega}(0)$, which according to (\ref{rexiineq1}) and (\ref{rexiineq2}) restrict the range of values of $U(0,y)$ to an interior range both to (\ref{rexiineq1}) and to (\ref{rexiineq2}) at once.
For mentioned purpose we now consider the restriction of the mean-value parameter $w_0(z)$ to the imaginary axis $z=\I y$ for which $g(u) = {\rm ch}(u(\I y))=\cos(uy)$ is a real-valued function of $y$. For arbitrary fixed $y$ we find by the second mean-value theorem a parameter $u_0$ in the interval $0 \le y < +\infty$ which naturally depends on the chosen value $y$ that means $u_0=u_0(0,y)$. The extension from the imaginary axis $z=\I y$ to the whole complex plane ${\mathbb C}$ can be made then using methods of complex analysis. We discuss some formal approaches to this in Appendix B. Now we apply (\ref{secmeanvth2compl}) to the imaginary axis $z=\I y$.
The second mean-value theorem (\ref{secmeanvth2compl}) on the imaginary axis $z=\I y$ (or $x=0$) takes on the form
\begin{eqnarray}
\int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(u(\I y)) &=& \int_{0}^{+\infty}du\,{\mit \Omega}(u)\cos(u y) \;=\; {\mit \Omega}(0) \int_{0}^{u_0(0,y)}du\,\cos(uy) \nonumber\\ &=& {\mit \Omega}(0)\frac{\sin\left(u_0(0,y) y\right)}{y},\quad \left(u_0(0,y) \neq 0,\quad v_0(0,y) =0 \right).
\label{realsecmeanvth}
\end{eqnarray}
As already said since the left-hand side is a real-valued function the right-hand side has also to be real-valued and the parameter function $w_0(\I y)$ is real-valued and therefore it can only be the real part $u_0(0,y)$ of the complex function $w_0(z= x +\I y)= u_0(x,y) +\I v_0(x,y)$ for $x=0$.
The second mean-value theorem states that $u_0(0,y)$ lies between the minimal and maximal values of the integration borders that is here between $0$ and $+\infty$ and this means that $u_0(0,y)$ should be positive. Here arises a problem which is connected with the periodicity of the function $g(u)=\cos(uy)$ as function of the variable $u$ for fixed variable $y$ in the application of the mean-value theorem. Let us first consider the special case $y=0$ in (\ref{realsecmeanvth}) which leads to
\begin{eqnarray}
\int_{0}^{+\infty}du\,{\mit \Omega}(u) &=& {\mit \Omega}(0) \int_{0}^{u_0}du \;=\; {\mit \Omega}(0)u_0 \;=\;
{\mit \Omega}(0)u_0\underbrace{\lim_{y\to 0}\frac{\sin(u_0y)}{u_0y}}_{=\,1},\quad u_0\,\equiv \; u_0(0,0)\;>\;0.
\end{eqnarray}
From this relation follows $u_0 \equiv u_0(0,0)>0$ and it seems that all is correct also with the continuation to $u_0(0,y) > 0$ for arbitrary $y$. One may even give the approximate values ${\mit \Omega}(0) \approx 1.78679 $ and $u_0 \approx 0.27822 $ and therefore ${\mit \Omega}_0 \equiv {\mit \Omega}(0)u_0 \approx 0.49712 $ which, however, are not of importance for the later proofs. If we now start from $u_0(0,0) > 0$ and continue it continuously to $u_0(0,y)$ then we see that $u_0(0,y)$ goes monotonically to zero and approaches zero approximately at $y=y_1\approx 14.135$ that is at the first zero of the function ${\mit \Xi}(\I y)$ on the positive imaginary axis and goes then first beyond zero and oscillates then with decreasing amplitude for increasing $y$ around the value zero with intersecting it exactly at the zeros of ${\mit \Xi}(\I y)$. We try to illustrate this graphically in Section 7. All zeros lie then on the branch $u_0(0,y)y=n\pi$ with $n=0$. That $u_0(0,y)$ goes beyond zero seems to contradict the content of the second mean-value theorem according which $u_0(0,y)$ has to be positive in our application. Here comes into play the multi-valuedness of the mean-value function $u_0(0,y)$. For the zeros of $\sin(u_0(0,y)y)$ in (\ref{realsecmeanvth}) the relations $u_0(0,y)y =n\pi$ with different integers $n$ are equivalent and one may find to values $u_0(0,y) <0$ equivalent curves $u_0(n;0,y)$ with $u_0(n;0,y)>0$ and all these curves begin with $u_0(n\neq 0;0,0) \rightarrow \infty$ for $y\rightarrow 0$. However, we cannot continue $u_0(0,0)$ in continuous way to only positive values for $u_0(0,y)$.
For $|y| \rightarrow \infty$ the inequality (\ref{rexiineq2}) is stronger than (\ref{rexiineq1}) and characterizes the restrictions of $U(0,y)$ and via the equivalence $U(0,y)y= {\mit \Omega}(0)\sin(u_0(0,y)y)$ follows from (\ref{rexiineq2})
\begin{eqnarray}
\left(n-\frac{1}{2}\right)\pi \;\le \; u_0(0,y)y \;=\; \arcsin\left(\frac{U(0,y)y}{{\mit \Omega}(0)}\right) \; \le \; \left(n+\frac{1}{2}\right)\pi,\quad (n=0,\pm 1, \pm 2,\ldots),\label{basicinterval}
\end{eqnarray}
where the choice of $n$ determines a basis interval of the involved multi-valued function $\arcsin(z)$ and the inequality says that it is in every case possible to choose it from the same interval of length $\pi$. The zeros $y_k$ of the Xi function ${\mit \Xi}(x+\I y)$ on the imaginary axis $x=0$ (critical line) are determined alone by the (multi-valued) function $u_0(0,y)$ whereas $v_0(0,y)$ vanishes automatically on the imaginary axis in considered special case and does not add a second condition. Therefore, the zeros are the solutions of the conditions
\begin{eqnarray}
u_0(0,y)y \;=\; n\pi,\quad (n=0,\pm 1, \pm2, \ldots),\quad (v_0(0,y) \;=\;0).\label{zerosimax}
\end{eqnarray}
It is, in general, not possible to obtain the zeros $y_k$ on the critical line exactly from the mean-value function $u_0(0,y)$ in (\ref{realsecmeanvth}) since generally we do not possess it explicitly.
In special cases the function $u_0(0,y)$ can be calculated explicitly that is the case, for example, for all (modified) Bessel functions $\left(\frac{2}{z}\right)^{\nu}{\rm I}_{\nu}(z)$. The most simple case among these is the case $\nu=\frac{1}{2}$ when the corresponding function ${\mit \Omega}(u)$ is a step function
\begin{eqnarray}
&& {\mit \Omega}(u) \;=\; {\mit \Omega}(0)\theta(u_0-u),\label{omegastepf}
\end{eqnarray}
where $\theta(x) = \left\{\begin{array}{cc} 0,& x<0 \\ 1,& x>0 \end{array}\right.$ is the Heaviside step function. In this case follows
\begin{eqnarray}
&& {\mit \Xi}(z) \;=\; {\mit \Omega}(0)\int_{0}^{u_0}du\,{\rm ch}(uz) \;=\; {\mit \Omega}(0)u_0\, \frac{{\rm sh}(u_0 z)}{u_0z} \;=\; \underbrace{{\mit \Omega}(0)u_0}_{=\,{\mit \Omega}_0}\, \left(\frac{\pi}{2u_0z}\right)^{\frac{1}{2}}\,{\rm I}_{\frac{1}{2}}(u_0z),\label{stepfunckernel}
\end{eqnarray}
where ${\mit \Omega}(0)u_0 =\int_{0}^{+\infty}du\,{\mit \Omega}(u)$ is the area under the function ${\mit \Omega}(u)= {\mit \Omega}(0)\theta(u_0-u)$ (or the zeroth-order moment of this function. For the squared modulus of the function ${\mit \Xi}(z) $ we find
\begin{eqnarray}
{\mit \Xi}(z)\left({\mit \Xi}(z)\right)^* \;=\; \left({\mit \Omega}(0)\right)^2\frac{{\rm sh}^2(u_0x)+\sin^2(u_0y)}{x^2+y^2}
\;=\; \left({\mit \Omega}(0)\right)^2 \frac{{\rm ch}(2u_0x)-\cos(2u_0y)}{2\left(x^2+y^2\right)},\label{xixistarstep}
\end{eqnarray}
from which, in particular, it is easy to see that this special function ${\mit \Xi}(x+\I y)$ possesses zeros only on the imaginary axis $z=\I y$ or $x=0$ and that they are determined by
\begin{eqnarray}
u_0y_n &=& n\pi, \quad \Rightarrow \quad y_n \;=\; \frac{n\pi}{u_0},\quad (n=\pm 1,\pm 2,\ldots).
\end{eqnarray}
The zeros on the imaginary axis are here equidistant but the solution $y_0=0$ is absent since then also the denominators in (\ref{xixistarstep}) are vanishing. The parameter $w_0(z)$ in the second mean-value theorem is here a real constant $u_0$ in the whole complex plane
\begin{eqnarray}
&& w_0(x+\I y) \;=\; u_0(x,y)+\I v_0(x,y) \;=\;u_0,\quad \Rightarrow \nonumber\\ && u_0(x,y) \;=\; u(0,y) \;=\; u_0,\quad v_0(x,y)\;=\;v_0(0,y) \;=\;0.\label{stepfunckernel1}
\end{eqnarray}
Practically, the second mean-value theorem compares the result for an arbitrary function ${\mit \Omega}(u)$ under the given restrictions with that for a step function ${\mit \Omega}(u)={\mit \Omega}(0)\,\theta(u_0-u)$ by preserving the value ${\mit \Omega}(0)$ and making the parameter $u_0$ depending on $z$ in the whole complex plane. Without discussing now quantitative relations the formulae (\ref{stepfunckernel1}) suggest that $v_0(x,y)$ will stay a 'small' function compared with $u_0(x,y)$ in the neighborhood of the imaginary axis (i.e. for $|x|\ll |y|$) in a certain sense.
We will see in next Section that the function $u_0(0,y)$ taking into account $v_0(0,y)=0$ determines the functions $u_0(x,y)$ and $v_0(x,y)$ and thus $w_0(z)$ in the whole complex plane via the Cauchy-Riemann equations in an operational approach that means in an integrated form which we did not found up to now in literature. The general formal part is again delegated to an Appendix B.
\setcounter{chapter}{5}
\setcounter{equation}{0}
\section*{5. Accomplishment of proof for zeros of Xi functions on imaginary axis alone}
In last Section we discussed the application of the second mean-value theorem to the function ${\mit \Xi}(z)$ on the imaginary axis $z=\I y$. Equations (\ref{zcuvxyre}) and (\ref{zcuvxyim})
or their equivalent forms (\ref{zcuvxy2}) or (\ref{zcuvxy3}) are not yet sufficient to derive conclusions about the position of the zeros on the imaginary axis in dependence on $x\neq 0$. We have yet to derive more information about the mean-value functions $w_0(z)$ which we obtain by relating the real-valued function $u_0(x,y)$ and $v_0(x,y)$ to the function $u_0(0,y)$ on the imaginary axis ($v_0(0,y)=0$).
The general case of complex $z$ can be obtained from the special case $z=\I y$ in (\ref{realsecmeanvth}) by application of the displacement operator $\exp\left(-\I x \frac{\partial}{\partial y}\right)$ to the function ${\mit \Xi}(\I y)$ according to
\begin{eqnarray}
{\mit \Xi}(x+\I y) &=& \exp\left(-\I x\frac{\partial}{\partial y}\right){\mit \Xi}(\I y) \exp\left(\I x\frac{\partial}{\partial y}\right) \;=\; \exp\left(-\I x\frac{\partial}{\partial y}\right)\int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(\I uy) \exp\left(\I x\frac{\partial}{\partial y}\right) \nonumber\\ &=& \exp\left(-\I x\frac{\partial}{\partial y}\right){\mit \Omega}(0)\frac{\,{\rm sh}\Big(\I u_0(0,y)y\Big)}{\I y}\exp\left(\I x\frac{\partial}{\partial y}\right)\;=\;
{\mit \Omega}(0)\frac{\,{\rm sh}\Big(u_0(0,y-\I x)\I (y-\I x)\Big)}{\I (y-\I x)}.\qquad \label{xixy}
\end{eqnarray}
The function $u_0(0,y-\I x) = w_0(x+\I y) = u_0(x,y) +\I v_0(x,y)$ is related to $u_0(0,y)$ as follows
\begin{eqnarray}
&& u_0(x,y) \;=\;
\cos\left(x\frac{\partial}{\partial y}\right)u_0(0,y), \quad v_0(x,y) \;=\;
-\sin\left(x\frac{\partial}{\partial y}\right)u_0(0,y), \label{uvxycossin}
\end{eqnarray}
or in more compact form
\begin{eqnarray}
w_0(x+\I y) &=& \exp\left(-\I x\frac{\partial}{\partial y}\right)u_0(0,y) \;=\; u_0(0,y-\I x).\label{uvxycossin1}
\end{eqnarray}
This is presented in Appendix B in more general form for additionally non-vanishing $v_0(0,y)$ and arbitrary holomorphic functions. It means that we may obtain $u_0(x,y) $ and $v_0(x,y)$ by applying the operators $\cos\left(x\frac{\partial}{\partial y}\right)$ and $-\sin\left(x\frac{\partial}{\partial y}\right)$, respectively, to the function $u_0(0,y)$ on the imaginary axis (remind $v_0(0,y)=0$ vanishes there in our case). Clearly, equations (\ref{uvxycossin}) are in agreement with the Cauchy-Riemann equations $\frac{\partial u_0}{\partial x}= \frac{\partial v_0}{\partial y}$ and $\frac{\partial u_0}{\partial y}= -\frac{\partial v_0}{\partial x}$ as a minimal requirement.
We now write ${\mit \Xi}(z)$ in the form equivalent to (\ref{xixy})
\begin{eqnarray}
{\mit \Xi}(x+ \I y) &=& {\mit \Omega}(0) \frac{{\rm sh}\Big(\big(u_0(x,y)+\I v_0(x,y)\big)(x+\I y)\Big)}{x+ \I y} \nonumber\\ &=& {\mit \Omega}(0) \frac{{\rm sh}\Big(\big(u_0(x,y)x- v_0(x,y)y\big)+\I\big(u_0(x,y)y + v_0(x,y)x\big)\Big)}{x+\I y}.\label{secmeanvth5}
\end{eqnarray}
The denominator $x+\I y$ does not contribute to zeros. Since the Hyperbolic Sine possesses zeros only on the imaginary axis we see from (\ref{secmeanvth5}) that we may expect zeros only for such related variables $(x,y)$ which satisfy the necessary condition of vanishing of its real part of the argument that leads as we already know to (see (\ref{zcuvxyre}))
\begin{eqnarray}
u_0(x,y)x- v_0(x,y)y &=& 0. \label{neccondzeros}
\end{eqnarray}
The zeros with coordinates $(x_k,y_k)$ themselves can be found then as the (in general non-degenerate) solutions of the following equation (see (\ref{zcuvxyim}))
\begin{eqnarray}
u_0(x,y)y + v_0(x,y)x &=& n\pi,\quad (n=0,\pm 1,\pm 2,\ldots), \label{zerosequ}
\end{eqnarray}
if these pairs $(x,y)$ satisfy the necessary condition (\ref{neccondzeros}). Later we will see provides the whole spectrum of solutions for the zeros but we can also obtain each $(x_k,y_k)$ separately from one branch $n$ and would they then denote by $(x_n,y_n)$.
Thus we have first of all to look for such pairs $(x,y)$ which satisfy the condition (\ref{neccondzeros}) off the imaginary axis that is for $x\neq 0$ since we know already that these functions may possess zeros on the imaginary axis $z=\I y$.
Using (\ref{uvxycossin}) we may represent the necessary condition (\ref{neccondzeros}) for the proof by the second mean-value theorem in the form
\begin{eqnarray}
x\cos\left(x\frac{\partial}{\partial y}\right)u_0(0,y) +y\sin\left(x\frac{\partial}{\partial y}\right)u_0(0,y) &=& 0,\label{zeroscondnec}
\end{eqnarray}
and equation (\ref{zerosequ}) which determines then the position of the zeros can be written with equivalent values $n$
\begin{eqnarray}
y\cos\left(x\frac{\partial}{\partial y}\right)u_0(0,y) - x\sin\left(x\frac{\partial}{\partial y}\right)u_0(0,y) &=& n\pi,\quad (n=0,\pm 1,\pm 2,\ldots). \label{zerosequ0}
\end{eqnarray}
We may represent Eqs. (\ref{zeroscondnec}) and (\ref{zerosequ0}) in a simpler form using the following operational identities
\begin{eqnarray}
x\cos\left(x\frac{\partial}{\partial y}\right)+y\sin\left(x\frac{\partial}{\partial y}\right) &=& \sin\left(x\frac{\partial}{\partial y}\right)y,\nonumber\\ y\cos\left(x\frac{\partial}{\partial y}\right)-x\sin\left(x\frac{\partial}{\partial y}\right) &=& \cos\left(x\frac{\partial}{\partial y}\right)y,
\end{eqnarray}
which are a specialization of the operational identities (\ref{fydisplfxyop}) in Appendix B with $w(z) =u(x,y)+\I v(x,y) \rightarrow z=x+\I y $ and therefore $u(x,y)\rightarrow x,\,v(x,y) \rightarrow y$.
If we multiply (\ref{zeroscondnec}) and (\ref{zerosequ0}) both by the function $u_0(0,y)$ then we may write (\ref{zeroscondnec}) in the form (changing order $yu_0(0,y)=u_0(0,y)y$)
\begin{eqnarray}
\sin\left(x\frac{\partial}{\partial y}\right)\left(u_0(0,y)y\right) &=& 0,\label{zeroscond1}
\end{eqnarray}
and (\ref{zerosequ0}) in the form
\begin{eqnarray}
\cos\left(x\frac{\partial}{\partial y}\right)\left(u_0(0,y)y\right) &=& n\pi,\quad (n=0,\pm 1,\pm 2,\ldots). \label{zeroscond2}
\end{eqnarray}
The left-hand side of these conditions possess the general form for the extension of a holomorphic function $W(z)=U(x,y)+\I V(x,y)$ from the functions $U(0,y)$ and $V(0,y)$ on the imaginary axis to the whole complex plane in case of $V(0,y)=0$ and if we apply this to the function $U(0,y)=u_0(0,y)y$. Equations (\ref{zeroscond1}) and (\ref{zeroscond2}) possess now the most simple form, we found, to accomplish the proof for the exclusive position of zeros on the imaginary axis. All information about the zeros of the Xi function ${\mit \Xi}(z) = U(x,y)+\I V(x,y)$ for arbitrary $x$ is now contained in the conditions (\ref{zeroscond1}) and (\ref{zeroscond2}}) which we now discuss.
Since $\cos\left(x\frac{\partial}{\partial y}\right)$ is a nonsingular operator we can multiply both sides of equation (\ref{zeroscond2}) by the inverse operator $\cos^{-1}\left(x\frac{\partial}{\partial y}\right)$ and obtain
\begin{eqnarray}
u_0(0,y)y &=& \cos^{-1}\left(x\frac{\partial}{\partial y}\right) n\pi \;=\; \left\{1+\frac{1}{2}\left(x\frac{\partial}{\partial y}\right)^2 +\ldots\right\}n\pi \;=\;n\pi,\quad (n=0,\pm 1,\ldots).\qquad\label{zeroimaxcond1}
\end{eqnarray}
This equation is yet fully equivalent to (\ref{zeroscond2}) for arbitrary $x$ but it provides only the same solutions for the values $y$ of zeros as for zeros on the imaginary axis. This alone already suggests that it cannot be that zeros with $x\neq 0$ if they exist possess the same values of $y$ as the zeros on the imaginary axis. But in such form the proof of the impossibility of zeros off the imaginary axis seemed to be not satisfactory and we present in the following some slightly different variants which go deeper into the details of the proof.
In analogous way by multiplication of (\ref{zeroscond1}) with the operator $\sin\left(x\frac{\partial}{\partial y}\right)$ and (\ref{zeroscond2}) with the operator $\cos\left(x\frac{\partial}{\partial y}\right)$ and addition of both equations we also obtain condition (\ref{zeroimaxcond1}) that means
\begin{eqnarray}
u_0(0,y)y &=& \left\{\sin^2\left(x\frac{\partial}{\partial y}\right)+\cos^2\left(x\frac{\partial}{\partial y}\right)\right\}u_0(0,y)y \;=\; n\pi,\quad (n=0,\pm 1,\pm 2,\ldots),\label{zeroimaxcond2}
\end{eqnarray}
The equal conditions (\ref{zeroimaxcond1}) and (\ref{zeroimaxcond2}) which are identical with the condition for zeros on the imaginary axis are a necessary condition for all zeros. For each chosen equivalent $n$ (remind $u_0(0,y)$ depends then on $n$ which we do not mention by the notation) one obtains an infinite series of solutions $y_k$ for the zeros of the function ${\mit \Xi}(\I y)$
\begin{eqnarray}
u_0(0,y_k)y_k &=& n\pi, \qquad \{u_0(0,y)y\}_{y\neq y_k} \;\neq\; n\pi, \label{yk0}
\end{eqnarray}
whereas for $y\neq y_k$ equation (\ref{zeroimaxcond1}), by definition of $y_k$, is not satisfied. Supposing that we know $u_0(0,y)$ that is as a rule not the case, we could solve for each $n = 0,\pm 1, \pm 2,\ldots$ the usually transcendental equations (\ref{zeroimaxcond2}) graphically, for example, by drawing the equivalent functions $u_0(0,y)y$ over
variable $y$ as abscissa and looking for the intersections points with the lines $n\pi$ over $y$ (Section 7). These intersection points $y=y_k$ are the solutions for zeros $y_k$ on the imaginary axis. Choosing $x=0$ the second condition (\ref{zeroscond1}) is identically satisfied.
Now we have to look for additional zeros $(x,y_k)$ with $x\neq 0$. Whereas for zeros with $x=0$ the condition (\ref{zeroscond1}) is identically satisfied we have to examine this condition for zeros with $x\neq 0$. In the case of $x\neq 0$ we may divide both sides of the condition (\ref{zeroscond1}) by $x$ and obtain
\begin{eqnarray}
\frac{\sin\left(x\frac{\partial}{\partial y}\right)}{x}\left(u_0(0,y)y\right) &=& \frac{\sin\left(x\frac{\partial}{\partial y}\right)}{x\frac{\partial}{\partial y}} \frac{\partial}{\partial y}\left(u_0(0,y)y\right) \;=\; 0.\label{zeroscond3}
\end{eqnarray}
Since $\frac{\sin\left(x\frac{\partial}{\partial y}\right)}{x\frac{\partial}{\partial y}} $ is a nonsingular operator (in contrast to $\sin\left(x\frac{\partial}{\partial y}\right)$ which possesses $0$ as eigenvalue to eigenfunction $f(x,y)={\rm const}x^n,(n=0,1,\ldots)$) we may multiply equation (\ref{zeroscond3})
by the inverse operator $\frac{x\frac{\partial}{\partial y}}{\sin\left(x\frac{\partial}{\partial y}\right)}$ and obtain
\begin{eqnarray}
\frac{\partial}{\partial y}\left(u_0(0,y)y\right) &=& \frac{\partial u_0}{\partial y}(0,y)y +u_0(0,y) \;=\; 0.
\end{eqnarray}
This condition has also to be satisfied for the solution $y=y_k$ of (\ref{zeroimaxcond1}) which make this equation to the identity (\ref{yk0}) that means that
\begin{eqnarray}
\frac{\partial u_0}{\partial y}(0,y_k)y_k +u_0(0,y_k) &=& \frac{\partial u_0}{\partial y}(0,y_k)y_k +\frac{n\pi}{y_k} \;=\; 0,
\end{eqnarray}
has to be identically satisfied. Moreover, if we apply the operator $\frac{\sin^m\left(x\frac{\partial}{\partial y}\right)}{x^{m+1}}$ to condition (\ref{zeroscond1}) we obtain
\begin{eqnarray}
\frac{\sin^{m+1}\left(x\frac{\partial}{\partial y}\right)}{x^{m+1}}\left(u_0(0,y)y\right) &=& \left(\frac{\sin\left(x\frac{\partial}{\partial y}\right)}{x\frac{\partial}{\partial y}}\right)^{m+1} \frac{\partial^{m+1}}{\partial y^{m+1}}\left(u_0(0,y)y\right) \;=\; 0,\;\; (m=0,1,2,\ldots),\qquad\; \label{zeroscond4}
\end{eqnarray}
and by multiplication with the inverse operator to the nonsingular operator $\left(\frac{x\frac{\partial}{\partial y}}{\sin\left(x\frac{\partial}{\partial y}\right)}
\right)^{m+1} $ we find
\begin{eqnarray}
\frac{\partial^{m+1}}{\partial y^{m+1}}\left(u_0(0,y)y\right) \;=\; 0,\quad (m=0,1,2,\ldots).\label{condzerosnplus1}
\end{eqnarray}
All these conditions have to be satisfied for the solutions $y=y_k$ with $x\neq 0$ in (\ref{zeroimaxcond2}), i.e.
\begin{eqnarray}
\frac{\partial^{m+1}}{\partial y^{m+1}}\left\{\left(u_0(0,y)y\right)\right\}_{y=y_k} \;=\; 0,\quad (m=0,1,2,\ldots). \label{dnyk0}
\end{eqnarray}
The same conditions follow also from (\ref{zeroscond2}) combined with (\ref{zeroscond2}) by Taylor series expansion with respect to $x\frac{\partial}{\partial y}$ for $x\neq 0$ according to
\begin{eqnarray}
n\pi &=&\exp\left(\pm \I x\frac{\partial}{\partial y}\right)\left(u_0(0,y)y\right) \;=\; u_0(0,y)y+\sum_{m=0}^\infty\frac{\left(\pm \I x\right)^{m+1}}{(m+1)!}\frac{\partial^{m+1}}{\partial y^{m+1}}\left(u_0(0,y)y\right),
\end{eqnarray}
by setting all sum terms proportional to $x^{m+1}$ equal to zero.
If we make now a Taylor series expansion of the function $u_0(0,y)y$ in the neighborhood $y=y_k$ of a solution which obeys all conditions (\ref{yk0}) and (\ref{dnyk0}) then we find
\begin{eqnarray}
u_0(0,y)y &=& u_0(0,y_k)y_k +\sum_{m=0}^{\infty}\frac{1}{(m+1)!}\frac{\partial^{m+1}}{\partial y^{m+1}}\left\{\left(u_0(0,y)y\right)\right\}_{y=y_k}\left(y-y_k\right)^{m+1} \;=\; n\pi.
\end{eqnarray}
Thus we can find zeros for $x\neq 0$ that means off the imaginary axis if the mean value function $u_0(0,y)$ on the imaginary axis possess one of the forms
\begin{eqnarray}
u_0(0,y)y = n\pi, \quad \Leftrightarrow \quad u_0(0,y) = \frac{n\pi}{y},\quad (n=0,\pm 1,\pm 2,\ldots),
\end{eqnarray}
for a certain integer $n$. According to (\ref{uvxycossin}) the whole mean-value functions $u_0(x,y)$ and $v_0(x,y)$ are then
\begin{eqnarray}
u_0(x,y) &=& \cos\left(x\frac{\partial}{\partial y}\right)\frac{n\pi}{y} \;=\; \frac{n\pi}{2}\left\{\exp\left(\I x\frac{\partial}{\partial y}\right) +\exp\left(-\I x\frac{\partial}{\partial y}\right)\right\}\frac{1}{y} \nonumber\\ &=& \frac{n\pi}{2}\left(\frac{1}{y+\I x}+\frac{1}{y-\I x} \right) \;=\; n\pi \frac{y}{x^2+y^2}, \nonumber\\ v_0(x,y) &=& -\sin\left(x\frac{\partial}{\partial y}\right)\frac{k\pi}{y} \;=\; -\frac{n\pi}{\I 2}\left\{\exp\left(\I x\frac{\partial}{\partial y}\right) -\exp\left(-\I x\frac{\partial}{\partial y}\right)\right\}\frac{1}{y} \nonumber\\ &=& -\frac{n\pi}{\I 2}\left(\frac{1}{y+\I x}-\frac{1}{y-\I x} \right) \;=\; n\pi \frac{x}{x^2+y^2},
\end{eqnarray}
or in compact form
\begin{eqnarray}
w_0(z) &=& u_0(x,y)+\I v_0(x,y) \;=\; n\pi\frac{y+\I x}{x^2+y^2} \;=\; \I\frac{n\pi}{x+\I y},\quad \Leftrightarrow \quad w_0(z)z \;=\; \I n\pi.
\end{eqnarray}
If we insert $w_0(z)z= \I n\pi$ into equation (\ref{secmeanvth2compl}) then we get ${\mit \Xi}(z) = 0$ for all $n=0,\pm 1,\pm2,\ldots$. This means that all conditions for zeros with $x\neq 0$ together do not lead to a solution for arbitrary ${\mit \Xi}(z) \neq 0$. Thus we have proved that all zeros of Xi functions ${\mit \Xi}(z)$ lie on the imaginary axis $x=0$.
Recognizing equations (\ref{zeroscond1}) and (\ref{zeroscond2}) as correct ones one may find modifications of the given proof of impossibility of solutions for zeros with $x\neq 0$. For example, one can make Taylor series expansions and obtain from equation (\ref{zeroscond2})
\begin{eqnarray}
u_0(0,y)y+\sum_{m=1}^{\infty}\frac{(-1)^m x^{2m}}{(2m)!}\frac{\partial^{2m}}{\partial y^{2m}}\left(u_0(0,y)y\right) &=& n\pi, \quad (n=0,\pm 1,\pm 2,\ldots),\label{taylu00y1}
\end{eqnarray}
and from (\ref{zeroscond1})
\begin{eqnarray}
\sum_{m=0}^{\infty}\frac{(-1)^m x^{2m+1}}{(2m+1)!}\frac{\partial^{2m+1}}{\partial y^{2m+1}}\left(u_0(0,y)y\right) &=& 0. \label{taylu00y2}
\end{eqnarray}
The right-hand sides of these equations are independent of variable $x$ and therefore the left-hand sides must it too. On the imaginary axis for $x=0$ this leads to the condition (\ref{zeroimaxcond2}) as the only condition which determines the zeros on this axis. For $x\neq 0$
one has as additional condition the vanishing of the coefficients in front of powers of $x^{n+1},\,(n=0,1,2,\ldots)$ in (\ref{taylu00y1}) and (\ref{taylu00y2}) that taken together leads to the conditions (\ref{condzerosnplus1}). The further discussion of the impossibility of this case is the same as before.
A third variant to show the impossibility for zeros in case of $x\neq 0$ is to make the transition to the Fourier transform of $u_0(0,y)y$
and to solve the equation arising from \ref{zeroscond3}) by generalized functions and then making the inverse Fourier transformation. One may show then that this is not compatible with the general solution of (\ref{zeroscond2}) which determines the position of the zeros. We do not present this variant here.
We have now finally proved that all Xi functions ${\mit \Xi}(z)$ of the form (\ref{basiccapxi0}) for which the second mean-value theorem is applicable (function ${\mit \Omega}(u)$ positively semi-definite and non-increasing (or also negatively semi-definite and non-decreasing)
may possess zeros only on the imaginary axis.
\setcounter{chapter}{6}
\setcounter{equation}{0}
\section*{6. Consequences for proof of the Riemann hypothesis}
The given proof for zeros only on the imaginary axis $x=0$ for the considered Xi function ${\mit \Xi}(z)={\mit \Xi}(x+\I y)$ includes as special case the function ${\mit \Omega}(u)$ to the Riemann hypothesis which is given in (\ref{defcapomega}). However, it includes also the whole class of modified Bessel functions of imaginary argument ${\rm I}_{\nu}(z)$ which possess zeros only on the imaginary axis and if we make the substitution $z\leftrightarrow \I z$ also the usual Bessel function ${\rm J}_{\nu}(z)$ which possess zeros only on the real axis.
We may ask about possible degeneracies of the zeros of the Xi functions ${\mit \Xi}(z)$ on the imaginary axis $z=\I y$. Our proof does not give a recipe to see whether such degeneracies are possible or not. In case of the Riemann zeta function ${\mit \Xi}(z) \leftrightarrow \xi(s) \leftrightarrow \zeta(s)$ one cannot expect a degeneracy because the countable number of all nontrivial zeros are (likely)
irrational (transcendental, proof?) numbers but we do not know a proof for this.
For ${\mit \Xi}(z)$ as an entire function one may pose the question of its factorization with factors of the form $1-\frac{z}{z_n} $ where $z_n$ goes through all roots where in case of degeneracy the same factors are taken multiple times according to the degeneracy.
It is well known that an entire function using its ordered zeros $z_n,\, (|z_n|\le |z_{n+1}|)$ can be represented in Weierstrass product form multiplied by an exponential function $\E^{h(z)}$ with an entire function function $h(z)$ in the exponent with the result that $\E^{h(z)}$ is an entire function without zeros. This possesses the form (e.g., \cite{lang})
\begin{eqnarray}
{\mit \Xi}(z) &=& \E^{h(z)} z^{m}\prod_{n}\left(1-\frac{z}{z_{n}}\right)\exp\left(P_{k_n}\left(\frac{z}{z_n}\right)\right),\label{wei}
\end{eqnarray}
with a polynomial $P_{k}(w)$ of degree $k$ which depending on the roots $z_n$ must be appropriately chosen to guarantee the convergence of the product. This polynomial is defined by first $k$ sum terms in the Taylor series for $-\log(1-w)$\footnote{Sometimes our $P_{k-1}(w)$ is denoted by $P_{k}(w)$.}
\begin{eqnarray}
P_k(w) & \equiv & w+\frac{w^2}{2}+\frac{w^3}{3}+\ldots+\frac{w^{k}}{k} \;=\; -\log(1-w)-\sum_{l=k+1}^{\infty}\frac{w^l}{l}.
\end{eqnarray}
By means of these polynomials the Weierstrass factors are defined as the functions
\begin{eqnarray}
E_k(w) &\equiv & (1-w)\exp\left(P_k(w)\right),
\end{eqnarray}
from which follows
\begin{eqnarray}
\log\left(E_k(w)\right) \;=\; \log(1-w)-\left(\log(1-w)+\sum_{l=k+1}^{\infty}\frac{w^l}{l}\right) \;=\; -\sum_{l=k+1}^{\infty}\frac{w^l}{l}.
\end{eqnarray}
From this form it is seen that $E_k(w)$ possesses the following initial terms of the Taylor series
\begin{eqnarray}
E_k(w) &=& (1-w)\exp\left(-\sum_{l=k+1}^{\infty}\frac{w^{l}}{l}\right) \;=\; 1-\frac{w^{k+1}}{k+1} +\ldots\;,
\end{eqnarray}
and is a function with a zero at $w =1$ but with a Taylor series expansion which begins with the
terms $1-\frac{w^{k+1}}{k+1}$.
Hadamard made a precision of the Weierstrass product form by connecting the degree $k_n$ of the polynomials in (\ref{wei}) with the order $\rho$ of growth of the entire function and showed that $k_n$ can be chosen independently of the $n$-th root $z_n$ by $k_n \rightarrow k \ge \rho -1$
The order of ${\mit \Xi}(z)$ which is equal to $1$ is not a strict order $\rho = 1$ (for this last notion see \cite{lang}). However, this does not play a role in the Hadamard product representation of ${\mit \Xi}(z)$ and the polynomials
$P_{k_n}(w)$ in (\ref{wei}) can be chosen as $P_0(w)$ that means equal to $0$ according to $k_n= k =\rho -1$. The entire function $h(z)$ in the exponent in (\ref{wei}) can be only a constant since in other case it would introduce a higher growth of ${\mit \Xi}(z)$. Thus the product representation of ${\mit \Xi}(z)$ possesses the form
\begin{eqnarray}
{\mit \Xi}(z) &=& {\mit \Xi}(0){{\prod}'}_{n=-\infty}^{+\infty} \left(1-\frac{z}{z_n}\right) \;=\; \int_{0}^{+\infty}du\,{\mit \Omega}(u) \prod_{n=1}^{+\infty}
\left(1-\frac{z^2}{z_n^2}\right) \nonumber\\ &=& {\mit \Omega}_{0} \prod_{n=1}^{+\infty}\left(1+\frac{z^2}{y_n^2}\right), \quad {\mit \Xi}(0) \;=\; {\mit \Omega}_0 \;\equiv \; \int_{0}^{+\infty}du\,{\mit \Omega}(u) \;=\; \xi\left(\frac{1}{2}\right) \;=\; 0.49712...\;,\label{prodxi}
\end{eqnarray}
where we took into account the symmetry $z_{-n} = -z_n =(z_n)^*$ of the zeros and the proof $z_n=\I y_n$ that all zeros lie on the imaginary axis and a zero $z_0=0$ is absent. With ${\mit \Omega}_0$ we denoted the first moment of the function ${\mit \Omega}(u)$.
Formula (\ref{prodxi}) in connection with his hypothesis was already used by Riemann in \cite{rie}
and later proved by von Mangoldt where the product representation of entire functions by Weierstrass which was later stated more precisely by Hadamard plays a role. There is another formula for an approximation to the number of nontrivial zeros of $\zeta(s)$ or $\xi(s)$ which in application to the number of zeros $N(Y)$ of ${\mit \Xi}(z)$ on the imaginary axis $z=\I y$ (critical line) in the interval between $y=0$ and $y=Y$. It takes on the form ($Y$ for ${\mit \Xi}(z)$ is equivalent to usual $T$ for $\zeta(s)$)
\begin{eqnarray}
N(Y) &=& \int_{0}^{Y}dy\,\nu(y) \;\approx\;\frac{Y}{2\pi}\log\left(\frac{Y}{2\pi}\right)-\frac{Y}{2\pi},\quad (Y \gg 0),\label{xinumbzeros}
\end{eqnarray}
with the logarithmically growing density
\begin{eqnarray}
\nu(y) &\approx & \frac{1}{2\pi}\log\left(\frac{y}{2\pi}\right),\quad (y \gg 1). \label{xidenszeros}
\end{eqnarray}
As long as the Riemann hypothesis was not proved it was formulated for the critical strip $ 0 \le \sigma \le 1$ of the complex coordinate $s=\sigma +\I t$ in $\xi(s)$ parallel to the imaginary axis and with $t$ between $t=0$ and $t=T$ (with $T$ equal to our $Y$ in (\ref{xinumbzeros})). It was already suggested by Riemann \cite{rie} but not proved in detail there and was later proved by von Mangoldt in 1905. A detailed proof by means of the argument principle can be found in \cite{borw}. The result of Hardy (1914) (cited in \cite{edw}) that there exist an infinite number of zeros on the critical line is a step to the full proof of the Riemann hypothesis. Section 4 of present article may be considered as involving such proof of this last statement.
We have now proved that functions ${\mit \Xi}(z)$ defined by integrals of the form (\ref{basiccapxi0}) with non-increasing functions ${\mit \Omega}(u)$ which decrease in infinity sufficiently rapidly in a way that ${\mit \Xi}(z)$ becomes an entire function of $z$ possess zeros only on the imaginary axis $z=\I y$. This did not provide a recipe to see in which cases all zeros on the imaginary axis are simple zeros but it is unlikely that within a countable sequence of irregularly chosen (probably) transcendental numbers (the zeros) two of them are coincident (it seems to be difficult to formulate last statement in a more rigorous way). It also did not provide a direct formula for the number of zeros in an interval $\left[(0,0),(0,Y)\right]$ from zero to $Y$ on the imaginary axis or of its density there but, as already said, Riemann \cite{rie} suggested for this an approximate formula and von Mangoldt proved it
The proof of the Riemann hypothesis is included as the special case (\ref{defcapomega}) of the function ${\mit \Omega}(u)$ into a wider class of functions with an integral representation of the form (\ref{basiccapxi0}) which under the discussed necessary conditions allowing the application of the second mean-value theorem of calculus possess zeros only on the imaginary axis.
The equivalent forms (\ref{capxirep6}) and (\ref{capxirep7}) of the integral (\ref{basiccapxi0}) where the functions, for example ${\mit \Omega}^{(1)}(u)$, are no more generally non-increasing suggest that conditions for zeros only on the imaginary axis are existent for more general cases than such prescribed by the second mean-value theorem. A certain difference may happen, for example, for $z=0$ because powers of it are in the denominators in the representations in (\ref{capxirep7}).
\setcounter{chapter}{7}
\setcounter{equation}{0}
\section*{7. Graphical illustration of mean-value parameters to Xi function for the Riemann hypothesis}
\begin{figure}[h]
\includegraphics[width=16.0cm]{d15.pdf}\hspace{1cm}
\caption{\foot Mean value parameters $w_0(\I y) = u_0(0,y)$ and $w_0(x) = u_0(x,0)$ for the Xi function in the proof of the Riemann hypothesis. \newline \scriptsize{It is not to see in the chosen scale that the curve $u_0(0,y)$ goes beyond the $y$-axis and oscillates around it due to extremely rapid vanishing of the envelope of $u_0(0,y)$ with increasing $y$ but we do not resolves this here by additional graphics because this behavior is better to see in the later considered case of modified Bessel functions. Using (\ref{w0zriem}) we calculate numerically $w_0(0)=u_0(0,0) \approx 0.27822$ that is the value which we call the optimal value for the moment series expansion (see Section 11). The part in the second partial figure which at the first glance looks like a straight line as asymptote is not such.}}
\end{figure}
\begin{figure}[h]
\includegraphics[width=16.0cm]{d12.pdf}\hspace{1cm}
\caption{\foot Mean value parameters in the proof of the Riemann hypothesis. \newline \scriptsize{On the left-hand side there are shown the mean value parameters $\frac{w_0(\I y)}{w(0)} = \frac{u_0(0,y)}{u_0(0,0)}$ for the Xi function to the Riemann hypothesis if we do not take the values of the function ${\rm \arcsin(t)}$ in the basic range $-\frac{\pi}{2} \le t \le \frac{\pi}{2}$ but in equivalent ranges according to (\ref{arcsinequ}). On the right-hand side are
shown the corresponding functions $u_0(0,y)y$ which according to $\cos(x+n\pi)=(-1)^n\sin(x)$ and the condition for zeros $\cos\left(u_0(0,y)y\right)=0$ lead to equivalent ranges
$k\pi =u_0(0,y)y \cong u_0(0,y)y +n\pi= (k+n)\pi, (k=\pm 1, \pm 2,\ldots;\,n=0,\pm 1,\pm2,\ldots)$ (see (\ref{zerosimax})) determine the zeros of the Xi function on the imaginary axis. We see that the multi-valuedness of the ${\rm arcsin(x)}$ function does not spoil a unique result for the zeros because every branch find the corresponding $n$ of $n\pi$ where then all zeros lie. Due to extremely rapid decrease of the function $u_0(0,y)$ with increasing $y$ this is difficult to see (position of first three zero at $y_1\approx 14.1,\,y_2 \approx 21.0,\,y_3 \approx 25.0$ is shown) but if we separate small intervals of $y$ and enlarge the range of values for $y_0(0,y)$ this becomes visible (similar as in Fig. 2.3). We do not make this here because this effect is better visible for the modified Bessel functions which we intend to consider at another place.}}
\end{figure}
To get an imagination how the mean-value function $w_0(z) =w_0(x+\I y)$ looks like we calculate it for the imaginary axis and for the real axis for the case of the function ${\mit \Omega}(u)$ in (\ref{defcapomega}) that is possible numerically. From the two equations for general $z$ and for $z=0$
\begin{eqnarray}
{\mit \Omega}(0) \frac{\,{\rm sh}\left(w_0(z)z\right)}{z} &=& \int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(uz), \nonumber\\ {\mit \Omega}(0)w_0(0) &=& \int_{0}^{+\infty}du\,{\mit \Omega}(u),
\end{eqnarray}
follows
\begin{eqnarray}
w_0(z) &=& \frac{1}{z}\,{\rm Arsh}\left(\frac{z}{{\mit \Omega}(0)}\int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(uz)\right), \nonumber\\ w_{0}(0) &=& \frac{1}{{\mit \Omega}(0)}\int_{0}^{+\infty}du\,{\mit \Omega}(u), \label{w0zriem}
\end{eqnarray}
with the two initial terms of the Taylor series
\begin{eqnarray}
w_0(z) &=& \frac{1}{z}\left\{\frac{z}{{\mit \Omega}(0)}\int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(uz) -\frac{1}{6}\left(\frac{z}{{\mit \Omega}(0)}\int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(uz)\right)^3 +\ldots\right\},\label{w0zriemtaylor}
\end{eqnarray}
and with the two initial terms of the asymptotic series
\begin{eqnarray}
w_0(z) \;=\; \frac{1}{z}\left\{\log\left(2\frac{z}{{\mit \Omega}(0)}\int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(uz)\right) +\frac{1}{4}\left(\frac{z}{{\mit \Omega}(0)}\int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(uz)\right)^{-2}+\ldots \right\}. \label{w0zriemasympt}
\end{eqnarray}
From (\ref{w0zriem}) follows
\begin{eqnarray}
\frac{w_0(z)}{w_0(0)} &=& \frac{{\mit \Omega}(0)}{z\int_{0}^{+\infty}du\,{\mit \Omega}(u)}\,{\rm Arsh}\left(\frac{z}{{\mit \Omega}(0)}\int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(uz)\right).\label{w0zriem1}
\end{eqnarray}
This can be numerically calculated from the explicit form (\ref{defcapomega}) of ${\mit \Omega}(u)$. For $y=0$ and for $x=0$ (and only for these cases) the function $\frac{w_0(z)}{w_0(0)}$ is real-valued, in particular, for $y=0$
\begin{eqnarray}
\frac{w_0(x)}{w_0(0)} &=& \frac{{\mit \Omega}(0)}{x\,\int_{0}^{+\infty}du\,{\mit \Omega}(u)}\,{\rm Arsh}\left(\frac{x}{{\mit \Omega}(0)}\int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(ux)\right) \nonumber\\ &=& \frac{\int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(ux)}{\int_{0}^{+\infty}du\,{\mit \Omega}(u)}\left\{1-\frac{1}{6}\left(\frac{x}{{\mit \Omega}(u)}\int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}(ux)\right)^2+\ldots \right\} \nonumber\\ &=& \frac{u_0(x,0)}{u_0(0,0)},\label{meanvalwxriem}
\end{eqnarray}
and for $x=0$
\begin{eqnarray}
\frac{w_0(\I y)}{w_0(0)} &=& \frac{{\mit \Omega}(0)}{y\,\int_{0}^{+\infty}du\,{\mit \Omega}(u)}\,{\rm arcsin}\left(\frac{y}{{\mit \Omega}(0)}\int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm cos}(uy)\right) \nonumber\\ &=&
\frac{\int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm cos}(uy)}{\int_{0}^{+\infty}du\,{\mit \Omega}(u)}\left\{1+\frac{1}{6}\left(\frac{y}{{\mit \Omega}(u)}\int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm cos}(uy)\right)^2+\ldots \right\} \nonumber\\ &=& \frac{u_0(0,y)}{u_0(0,0)} ,\qquad\label{meanvalwyriem}
\end{eqnarray}
where we applied the first two terms of the Taylor series expansion of ${\rm arcsin}(x)$ in powers of $x$. A small problem is here that we get the value for this multi-valued function in the range $-\frac{\pi}{2} \le {\rm arcsin}(x) \le +\frac{\pi}{2}$. Since $\frac{1}{x}\, {\rm arcsin}(x)$ is
an even function with only positive coefficients in its Taylor series the term in braces is in every case positive that becomes important below.
The two curves which we get for $\frac{w_0(x)}{w_0(0)}$ and for $\frac{w_0(\I y)}{w_0(0)}$ are shown in Fig. 7.4. The function for $w_0(x,0)$ on the real axis $y=0$ (second partial picture) is not very exciting. The necessary condition $xu_0(x,0)=0$ (see (\ref{neccondzeros})) can be satisfied only for $x=0$ but it is easily to see from ${\mit \Xi}(0) = \int_{0}^{+\infty}du\;{\mit \Omega}(u) \approx 1.7868 \neq 0$ that there is no zero. For the function $w_0(0,y)$ on the imaginary axis $x=0$ the necessary condition $yv_0(0,y)=0$ (see (\ref{neccondzeros})) is trivially satisfied since $v_0(0,y)=0$ and does not restrict the solutions for zeros. In this case only the sufficient condition $yu_0(0,y) = n\pi,(n=0,\pm 1,\pm2,\ldots)$ determines the position of the zeros on the imaginary axis. The first two pairs of zeros are at $y_1\approx \pm 14.135,\;y_2 \approx \pm 21.022$ and the reason that we do not see them in Fig. 7.4 is the rapid decrease of the function $u_0(0,y)$ with increasing $y$. If we enlarge this range we see that the curve goes beyond the $y$-axis after the first root at $14.135$ of the Xi function. As a surprise for the second mean-value method we see that the parameter $u_0(0,y)$ becomes oscillating around this axis.
This means that the roots which are generally determined by the equation $yu_0(0,y)=n\pi$ (see (\ref{zerosequ})) are determined here by the value $n=0$ alone. The reason for this is the multi-valuedness of the ArcSine function according to
\begin{eqnarray}
{\rm arcsin}(x) & \cong & n\pi+(-1)^n {\rm arcsin}(x),\quad \left(-1 \le x \le +1,\quad -\frac{\pi}{2} \le {\rm arcsin}(x) \le +\frac{\pi}{2},\quad n \in {\mathbb Z} \right).\qquad
\label{arcsinequ}
\end{eqnarray}
If we choose the values for the ${\rm arcsin(x)}$-function not in the basic interval $-\frac{\pi}{2} \le x \le +\frac{\pi}{2}$ for which the Taylor series provides the values but from other equivalent intervals according to (\ref{arcsinequ}) we get other curves for $u_0(0,y)$ and $yu_0(0,y)$ from which we also may determine the zeros (see Fig. 7.5), however, with other values $n$ in the relation $yu_0(0,y)=n\pi,\,(n=0,\pm 1,\pm 2,\ldots)$ and the results are invariant with respect to the multi-valuedness. This is better to see in case of the modified Bessel functions for which the curves vanish less rapidly with increasing $y$ as we intend to show at another place. All these considerations do not touch the proof of the non-existence of roots off the imaginary axis but should serve only for better understanding of the involved functions. It seems that the specific phenomenons of the second mean-value theorem (\ref{secmeanvth}) if the functions $g(u)$ there are oscillating functions (remind, only continuity is required) are not yet well illustrated in detail.
We now derive a few general properties of the function $u_0(0,y)$ which can be seen in the Figures. From (\ref{realsecmeanvth}) written in the form and by Taylor series expansion according to
\begin{eqnarray}
\int_{0}^{+\infty}du\,{\mit \Omega}(u)\cos(u y) &=& {\mit \Omega}(0)u_0(0,y) \frac{\sin\left(u_0(0,y)y\right)}{u_0(0,y)y} \nonumber\\ &=& {\mit \Omega}(0)u_0(0,y) \left(1-\frac{y^2}{6}\left(u_0(0,y)\right)^2 +\ldots \right),\label{secmeanvth2real2}
\end{eqnarray}
follows from the even symmetry of the left-hand side that $u_0(0,y)$ also has to be a function of the variable $y$ with even symmetry (notation $\frac{\dis{\partial^n u_0(0,y)}}{\dis{\partial y^n}} \equiv u_0^{(n)}(0,y)$)
\begin{eqnarray}
u_0(0,y) &=& +u_0(0,-y),\label{secmeanvth2real3}
\end{eqnarray}
with the consequence
\begin{eqnarray}
u_0(0,y) \;=\; u_0(0,0)+\sum_{m=1}^{\infty}\frac{u_0^{(2m)}(0,0)}{(2m)!}y^{2m},\quad u_0^{(2m+1)}(0,0)\;=\; 0.\qquad \label{secmeanvth2real30}
\end{eqnarray}
Concretely, we obtain by $n$-fold differentiation of both sides of (\ref{secmeanvth2real2}) at $y=0$ for the first coefficients of the Taylor series
\begin{eqnarray}
\int_{0}^{+\infty}du\,{\mit \Omega}(u) &=& {\mit \Omega}(0)u_0(0,0), \quad 0 \;=\; {\mit \Omega}(0)u_0^{(1)}(0,0), \nonumber\\ -\int_{0}^{+\infty}du\,{\mit \Omega}(u)u^2 &=& {\mit \Omega}(0)\left(u_0^{(2)}(0,0)-\frac{1}{3}\left(u_0(0,0)\right)^3\right),
\end{eqnarray}
from which follows
\begin{eqnarray}
u_0(0,0) &=& \frac{1}{{\mit \Omega}(0)}\int_{0}^{+\infty}du\,{\mit \Omega}(u), \quad u_0^{(1)}(0,0) \;=\; 0, \quad \nonumber\\ u_0^{(2)}(0,0) &=& \underbrace{-\frac{1}{{\mit \Omega}(0)}\int_{0}^{+\infty}du\,{\mit \Omega}(u)u^2}_{<\,0} +\underbrace{\frac{1}{3}\left(\frac{1}{{\mit \Omega}(0)}\int_{0}^{+\infty}du\,{\mit \Omega}(u)\right)^3}_{>\,0}.
\end{eqnarray}
Since the first sum term on the right-hand side is negative and the second is positive it depends from their values whether or not $u_0^{(2)}(0,0)$ possesses a positive or negative value. For the special function ${\mit \Omega}(u)$ in (\ref{defcapomega}) which plays a role in the Riemann hypothesis we find approximately
\begin{eqnarray}
&& {\mit \Omega}_0 \;\equiv\; \int_{0}^{+\infty}du\,{\mit \Omega}(u) \;\approx \; 0.497121,\quad 2!{\mit \Omega}_2 \;\equiv \; \int_{0}^{+\infty}du\,{\mit \Omega}(u)u^2 \;\approx \; 0.0229719,\nonumber\\ && {\mit \Omega}(0) \;\approx\; 1.78679,\quad
u_0(0,0) \;\approx \; 0.278220,\quad u_0^{(2)}(0,0) \;\approx\; -0.00567784,\label{momentnumer}
\end{eqnarray}
meaning that the second coefficient in the expansion of $u_0(0,y)$ in a Taylor series in powers of $y$ is negative that can be seen in the first part of Fig. 7.4. However, as we have seen the proof of the Riemann hypothesis is by no means critically connected with some numerical values.
In principle, the proof of the Riemann hypothesis is accomplished now and illustrated and we will stop here. However, for a deeper understanding of the proof it is favorable to consider some aspects of the proof such as, for example, analogues to other functions with a representation of the form (\ref{basiccapxi0}) and with zeros only on the imaginary axis and some other approaches although they did not lead to the full proof that,
however, we cannot make here.
\setcounter{chapter}{8}
\setcounter{equation}{0}
\section*{8. Equivalent formulations of the main theorems in a summary}
In present article we proved the following main result\\[1mm]
{\large{\bf Theorem 1:}}\\
Let ${\mit \Omega}(u)$ be a real-valued function of variable $u$ in the interval $0 \le u < +\infty$ which is positive semi-definite in this interval and non-increasing and is rapidly vanishing in infinity, more rapidly than any exponential function $\exp\left(-\lambda u\right)$, that means
\begin{eqnarray}
{\mit \Omega}(u) \; \ge 0, \quad {\mit \Omega}^{(1)}(u) &\le & 0,\quad (0\le u <+\infty),\quad \lim_{u\rightarrow +\infty}\frac{{\mit \Omega}(u)}{\exp\left(-\lambda u\right)} \;= \;0,\quad (\lambda \in {\mathbb R}^{+}).
\end{eqnarray}
Then the following integral with arbitrary complex parameter $z=x+\I y$
\begin{eqnarray}
{\mit \Xi}(z) &=& \int_{0}^{+\infty}du\,{\mit \Omega}(u) \,{\rm ch}(uz), \quad (z \in {\mathbb C}), \label{basiccapxi2}
\end{eqnarray}
is an entire function of $z$ with possible zeros $z=z_0=x_0+\I y_0$ only on the imaginary axis $x=0$ that means
\begin{eqnarray}
{\mit \Xi}(z_0) &=& 0, \quad \Rightarrow \quad z_0\;=\; \I y_0,\quad ({\rm or}\;\; x_0=0).
\end{eqnarray}
{\large{\bf Proof:}}\\[1mm]
The proof of this theorem for non-increasing functions ${\mit \Omega}(u)$ takes on Sections 3--5 of this article. The function ${\mit \Omega}(u)$ in (\ref{defcapomega}) satisfies these conditions and thus provides a proof of the Riemann hypothesis.
\\[1mm]
{\large{\bf Remark:}}\\
A analogous theorem is obviously true by substituting in (\ref{basiccapxi2}) ${\rm ch}(u) \leftrightarrow \cos(u)$ and by interchanging the role of the imaginary and of the real axis $y\leftrightarrow x$.
Furthermore, a similar theorem with a few peculiarities is true for substituting ${\rm ch}(uz)$ in (\ref{basiccapxi2}) by ${\rm sh}(uz)$ .\\
\indent
{\bf{Theorem 1}} can be formulated in some equivalent ways which lead to interesting consequences\footnote{Some of these equivalences now formulated as consequences originate from trials to prove the Riemann hypothesis in other way.}. The Mellin transformation ${\hat{f}}(s)$ of an arbitrary function $f(t)$ together with its inversion is defined by \cite{erd,zem,brpr,bert}
\begin{eqnarray}
{\hat{f}}(s) \;=\; \int_{0}^{+\infty}dt\,f(t)t^{s-1},\quad f(t)\;=\; \frac{1}{\I 2\pi}\int_{c-\I \infty}^{c+\I \infty}ds\,{\hat{f}}(s)t^{-s},\label{mellintransform}
\end{eqnarray}
where the real value $c$ has only to lie in the convergence strip for the definition of ${\hat{f}}(s)$ by the integral.
Formula (\ref{basiccapxi2}) is an integral transform of the function ${\rm ch}(z)$ and can be considered as the application of an integral operator to the function ${\rm ch}(z)$ which using the Mellin transform ${\hat{\mit \Omega}}(s)$ of the function ${\mit \Omega}(u)$ can be written in the following convenient form
\begin{eqnarray}
{\mit \Xi}(z) &=& {\hat{\mit \Omega}}\left(\frac{\partial}{\partial z}z\right) \,{\rm ch}(z).\label{theoremmellin}
\end{eqnarray}
This is due to
\begin{eqnarray}
\int_{0}^{+\infty}du\,{\mit \Omega}(u) \,{\rm ch}(uz) &=& \int_{0}^{+\infty}du\,{\mit \Omega}(u) u^{z\frac{\partial}{\partial z}}\,{\rm ch}(z) \;=\; \int_{0}^{+\infty}du\,{\mit \Omega}(u) u^{\frac{\partial}{\partial z}z-1}\,{\rm ch}(z),
\end{eqnarray}
where $u^{z\frac{\partial}{\partial z}}$ is the operator of multiplication of the argument of an arbitrary function $g(z)$ by the number $u$, i.e. it transforms as follows
\begin{eqnarray}
g(z) \;\rightarrow \; g(uz) &=& u^{z\frac{\partial}{\partial z}} g(z),
\end{eqnarray}
according to the following chain of conclusions starting from the property that all functions $z^n,\,(n=0,1,2,\ldots)$ are eigenfunctions of $z\frac{\partial}{\partial z}$ to eigenvalue $n$
\begin{eqnarray}
&& z\frac{\partial}{\partial z}z^{n} \;=\; nz^{n},\quad \Rightarrow \quad f\left(z\frac{\partial}{\partial z}\right) z^n \;=\; f(n)z^n, \quad \Rightarrow \quad \exp\left(\lambda z\frac{\partial}{\partial z}\right)z^n \;=\; \left(\E^{\lambda} z\right)^n, \quad \Rightarrow \nonumber\\ && \exp\left(\lambda z\frac{\partial}{\partial z}\right)g(z) \;=\; g\left(\E^\lambda z\right),\quad \Rightarrow \quad u^{z\frac{\partial}{\partial z}} g(z) \;=\; g(uz),\quad (\E^{\lambda}\equiv u).
\end{eqnarray}
This chain is almost obvious and does not need more explanations. The operators $u^{z\frac{\partial}{\partial z}}$ are linear operators in linear spaces depending on the considered set of numbers $u$.
Expressed by real variables $(x,y)$ and by $\left(\frac{\partial}{\partial z},\frac{\partial}{\partial z^*}\right)=\left(\frac{1}{2}\left(\frac{\partial}{\partial x}-\I \frac{\partial}{\partial y}\right),\frac{1}{2}\left(\frac{\partial}{\partial x}+\I \frac{\partial}{\partial y}\right)\right)$ we find from (\ref{theoremmellin})
\begin{eqnarray}
{\mit \Xi}(x+\I y) &=& {\hat{\mit \Omega}}\left\{1+\frac{1}{2}\left(x\frac{\partial}{\partial x}+y\frac{\partial}{\partial y}\right)+\frac{\I}{2}\left(y\frac{\partial}{\partial x}-x\frac{\partial}{\partial y}\right)\right\} \,{\rm ch}(x+\I y).\label{theoremmellinreal}
\end{eqnarray}
From this formula follows that ${\mit \Xi}(\I y)$ may be obtained by transformation of ${\rm ch}(\I y)=\cos(y)$ alone via
\begin{eqnarray}
{\mit \Xi}(\I y) &=& \hat{\mit \Omega}\left\{1+y\frac{\I}{2}\left(\frac{\partial}{\partial x}-\I \frac{\partial}{\partial y}\right) \right\}\,{\rm ch}\left(x+\I y\right) \;=\; \hat{\mit \Omega}\left(1+\I y\frac{\partial}{\partial z}\right)\,{\rm ch}(z) \nonumber\\ &=& \hat{\mit \Omega}\left(\frac{\partial}{\partial y}y\right)\cos(y).
\end{eqnarray}
On the right-hand side we have a certain redundance since in analytic functions the information which is contained in the values of the function on the imaginary axis is fully contained also in other parts of the function (here of ${\rm ch}(z)$).
The most simple transformation of ${\rm ch}(z)$ is by a delta function $\delta(u-u_0)$ as function ${\mit \Omega}(u)$ which stretches only the argument of the Hyperbolic Cosine function ${\rm ch}(z) \rightarrow {\rm ch}(u_0z)$. The next simple transformation is with a function
function ${\mit \Omega}(u)$ in form of a step function $\theta(u_0-u)$ which leads to the transformation ${\rm ch}(z) \rightarrow \frac{1}{z}\,{\rm sh}(u_0z)$. Our application of the second mean-value theorem reduced other cases under the suppositions of the theorem to this case, however, with parameter $u_0 =u_0(z)$ depending on complex variable $z$.
The great analogy between displacement operators (infinitesimal $-\I\frac{\partial}{\partial x}$ ) of the argument of a function and multiplication operator (infinitesimal $ x\frac{\partial}{\partial x}$) of the argument of a function with respect to the role of Fourier transformation and of Mellin transformation can be best seen from the following two relations
\begin{eqnarray}
\int_{-\infty}^{+\infty}dy\,f(y)g(x-y) &=& \tilde{f}\left(-\I \frac{\partial}{\partial x}\right)g(x),\quad \tilde{f}(t) \;\equiv \; \int_{-\infty}^{+\infty}dy\,f(y)\E^{-\I t y}, \nonumber\\ \int_{0}^{+\infty}du\,f(u)g(ux) &=& \hat{f}\left(\frac{\partial}{\partial x}x\right)g(x),\quad \hat{f}(s) \; \equiv \; \int_{0}^{+\infty}du\,f(u)\,u^{s-1}.\label{fourmellequ}
\end{eqnarray}
We remind you that Mellin and Fourier transform are related by substituting the integration variables $u=\E^y$ and the independent variables $s=-\I t$ and by the substitutions $f(\E^y) \leftrightarrow f(y)$ and $\hat{f}(-\I t) \leftrightarrow \tilde{f}(t)$ in (\ref{fourmellequ}).
Using the discussed Mellin transformation {\bf{Theorem 1}} can be reformulated as follows\\[1mm]
{\large{\bf Theorem $1^{\star}$:}}\\[1mm]
The mapping of the function ${\rm ch}(z)$ of the complex variable $z$ into the function ${\mit \Xi}(z)$ by an operator ${\hat{\mit \Omega}}\left(\frac{\partial}{\partial z}z\right)$
according to
\begin{eqnarray}
{\mit \Xi}(z) &=& {\hat{\mit \Omega}}\left(\frac{\partial}{\partial z}z\right) \,{\rm ch}(z),\quad
{\hat{\mit \Omega}}(s) \;\equiv\; \int_{0}^{+\infty}du\,{\mit \Omega}(u) u^{s-1},
\end{eqnarray}
where ${\hat{\mit \Omega}}(s)$ is the Mellin transformation of the function ${\mit \Omega}(u)$ which last possesses the properties given in {\bf{Theorem 1}} maps the function ${\rm ch}(z)$ with zeros only on the imaginary axis again into a function ${\mit \Xi}(z)$ with zeros only on the imaginary axis.\\
{\large{\bf Proof:}}\\[1mm]
It is proved as a reformulation of the {\bf{Theorem 1}} which is supposed here to be correctly proved.\\[2mm]
\indent
It was almost evident that the theorem may be formulated for more general functions ${\mit \Omega}(u)$ as supposed for the application of the second mean-value theorem that was made in Section 6. Under the suppositions of the theorem the integral on the left-hand side of (\ref{theoremmellin}) can be transformed by partial integration to (notation:
${\hat{\mit \Omega}^{(1)}}(s) \equiv \int_{0}^{+\infty}du\,{\mit \Omega}^{(1)}(u)u^{s-1}$)
\begin{eqnarray}
{\mit \Xi}(z) &=& -\frac{1}{z}\int_{0}^{+\infty}du\,{\mit \Omega}^{(1)}(u)\,{\rm sh}(uz) \;=\; -\frac{1}{z}\,{\hat{\mit \Omega}^{(1)}}\left(\frac{\partial}{\partial z}z\right)\,{\rm sh}(z).\label{theoremmellinsh}
\end{eqnarray}
The derivative ${\mit \Omega}^{(1)}(u)$ of the function ${\mit \Omega}(u)$ to the Riemann hypothesis although semi-definite (here negatively) and rapidly vanishing in infinity is not monotonic and possesses a minimum (see (\ref{defcapomega}) and Fig. 2.2).
In case of the (modified) Bessel functions we find by partial integration (e.g., \cite{erd})
\begin{eqnarray}
\nu!\left(\frac{2}{z}\right)^{\nu}\,{\rm I}_{\nu}\left(z\right) &=& \frac{\nu!} {\left(\nu-\frac{1}{2}\right)!\left(\frac{1}{2}\right)!}\int_{0}^{+1}du\,
\left(1-u^2\right)^{\nu-\frac{1}{2}}\,{\rm ch}\left(uz\right)\nonumber\\ &=& \frac{2\,\nu!}{\left(\nu-\frac{3}{2}\right)!\left(\frac{1}{2}\right)!}\,\frac{1}{z} \int_{0}^{+1}du\,\left(1-u^2\right)^{\nu-\frac{3}{2}}\,u\,{\rm sh}\left(uz\right), \label{intreprbessel}
\end{eqnarray}
where the functions in the second transform $\left(1-u^2\right)^{\nu-\frac{3}{2}}\,u$ for $\nu > \frac{3}{2}$ are non-negative but not monotonic and possess a maximum for a certain value $u_{\rm max}$ within the interval $0 < u_{\rm max} <1$. The forms (\ref{theoremmellinsh}) for ${\mit \Xi}(z)$ and (\ref{intreprbessel}) suggest that there should be true a similar theorem to the integral in (\ref{basiccapxi2}) with substitution ${\rm ch}(uz) \rightarrow {\rm sh}(uz)$ and that monotonicity of the corresponding functions should not be the ultimate requirement for the zeros in such transforms on the imaginary axis.
Another consequence of the {\bf{Theorem 1}} follows from the non-negativity of the squared modulus of the function ${\mit \Xi}(z)$ resulting in the obvious inequality
\begin{eqnarray}
0 & \le & {\mit \Xi}(z)\left({\mit \Xi}(z)\right)^* \nonumber\\ &=& \frac{1}{2}\int_{0}^{+\infty}du_1\int_{0}^{+\infty}du_2\,{\mit \Omega}(u_1){\mit \Omega}(u_2) \nonumber\\ && \cdot \Big(\,{\rm ch}((u_1+u_2)x)\cos((u_1-u_2)y)+\,{\rm ch}((u_1-u_2)x)\cos((u_1+u_2)y)\Big),\label{ineq1}
\end{eqnarray}
which can be satisfied with the equality sign only on the imaginary axis $z=\I y$ for discrete values $y=y_k$ (the zeros of ${\mit \Xi}(z=x+\I y)$).
By transition from Cartesian coordinates $(u_1,u_2)$ to inertial-point coordinates $(u,\Delta u)$ according to
\begin{eqnarray}
&& u=\frac{u_1+u_2}{2},\quad \Delta u =u_2-u_1, \quad u_1=u-\frac{\Delta u}{2},\quad u_2= u+\frac{\Delta u}{2},\quad du_1 \wedge du_2 =du\wedge d(\Delta u),\qquad
\end{eqnarray}
equation (\ref{ineq1}) can be also written
\begin{eqnarray}
0 & \le & \frac{1}{2}\int_{0}^{+\infty}du\int_{-2u}^{+2u}d(\Delta u)\,{\mit
\Omega}\left(u-\frac{\Delta u}{2}\right){\mit \Omega}\left(u+\frac{\Delta u}{2}\right)
\nonumber\\ && \cdot
\Big(\,{\rm ch}\left(2u\,x\right)\cos\left(\Delta u\,y\right)+\,{\rm ch}\left(\Delta u\,x\right)\cos\left(2u\,y\right)\Big). \label{ineq2}
\end{eqnarray}
As already said the case of the equality sign in (\ref{ineq1}) or (\ref{ineq2}) can only be obtained for $x=0$ and then only for discrete values of $y=y_k$ by solution of this inequality with the specialization for $x=0$
\begin{eqnarray}
0 & = & {\mit \Xi}(\I y)\left({\mit \Xi}(\I y)\right)^* \nonumber\\ &=& \frac{1}{2}\int_{0}^{+\infty}du_1\int_{0}^{+\infty}du_2\,{\mit \Omega}(u_1){\mit \Omega}(u_2) \Big(\cos((u_1-u_2)y)+\cos((u_1+u_2)y)\Big) \nonumber\\ & = & \frac{1}{2}\int_{0}^{+\infty}du\int_{-2u}^{+2u}d(\Delta u)\,{\mit
\Omega}\left(u-\frac{\Delta u}{2}\right){\mit \Omega}\left(u+\frac{\Delta u}{2}\right)
\Big(\cos\left(\Delta u\,y\right)+\cos\left(2u\,y\right)\Big).\qquad \label{ineq3}
\end{eqnarray}
A short equivalent formulation of the inequality (\ref{ineq1}) and (\ref{ineq2}) together with (\ref{ineq3}) is the following\\
{\large{\bf Theorem 2:}}\\
If the function ${\mit \Omega}(u)$ satisfies the suppositions in {\bf{Theorem 1}} then with $y_{-k} = -y_{k}$
\begin{eqnarray}
{\mit \Xi}(x+\I y)\left({\mit \Xi}(x+ \I y)\right)^* \;\; \left\{\begin{array}{llll} > 0,& \Rightarrow & \left\{\begin{array}{cl} \forall x\neq 0,& \\ \;\, x=0,& \forall y\neq y_k, \quad (k=\pm 1,\pm 2,\ldots), \end{array} \right. &
\\ =0, & \Rightarrow & \quad \;\;\, x=0, \quad \exists y=y_k, \quad (k=\pm 1,\pm 2,\ldots).\end{array} \right. \label{ineq4}
\end{eqnarray}
{\large{\bf Proof:}}\\[1mm]
As a consequence of proved {\bf{Theorem 1}} it is also proved. \\
The sufficient condition that this inequality is satisfied with the equality sign is that we first set $x=0$ in the expressions on the right-hand side of (\ref{ineq1}) and that we then determine the zeros $y=y_k$ of the obtained equation for ${\mit \Xi}(\I y)\left({\mit \Xi}(\I y)\right)^* =0$. In case of indefinite ${\mit \Omega}(u)$ there are possible in addition zeros on the $x$-axis.\\[1mm]
{\large{\bf Remark:}}\\[1mm]
Practically, (\ref{ineq1}) is an inequality for which it is difficult to prove in another way that it can be satisfied with the equality sign only for $x=0$. Proved in another way with specialization (\ref{defcapomega}) for ${\mit \Omega}(u)$ it would be an independent proof of the Riemann hypothesis.
\setcounter{chapter}{9}
\setcounter{equation}{0}
\section*{Conclusion}
We proved in this article the Riemann hypothesis embedded into a more general theorem for a class of functions ${\mit \Xi}(z)$ with a representation of the form (\ref{basiccapxi0}) for real-valued functions ${\mit \Omega}(u)$ which are positive semi-definite and non-increasing in the interval $0 \le u < +\infty$ and which are vanishing in infinity more rapidly than any exponential function $\exp\left(-\lambda u\right)$ with $\lambda > 0$. The special Xi function ${\mit \Xi}(z)$ to the function ${\mit \Omega}(u)$ given in (\ref{defcapomega}) which is essentially the xi function $\xi(s)$ equivalent to the Riemann zeta function $\zeta(s)$ concerning the hypothesis belongs to the described class of functions.
Modified Bessel functions of imaginary argument 'normalized' to entire functions $(\frac{2}{\I z})^{\nu}\,{\rm J}_{\nu}(\I z) =(\frac{2}{z})^{\nu}\,{\rm J}_{\nu}(z)$ for $\nu \ge \frac{1}{2}$ belong also to this class of functions with a representation of the form (\ref{basiccapxi0}) with ${\mit \Omega}(u)$ which satisfy the mentioned conditions and in this last case it is well known and proved in independent way that their zeros lie only on the imaginary axis corresponding to the critical line in the Riemann hypothesis. Knowing this property of the modified Bessel functions we looked from beginning for whole classes of functions including the Riemann zeta function which satisfy analogous conditions as expressed in the Riemann hypothesis. The details of the approach to Bessel functions and also to certain classes of almost-periodic functions we prepare for another work.
The numerical search for zeros of the Riemann zeta function $\zeta(s),\,s=\sigma + \I t$ in the critical strip, in particular, off the critical line may come now to an end by the proof of the Riemann hypothesis since its main purpose was, in our opinion, to find a counter-example to the Riemann hypothesis and thus to disprove it. We did not pay attention in this article to methods of numerical calculation of the zeros with (ultra-)high precision and for very high values of the imaginary part. However, the proof if correct may deliver some calculators now from their pain to calculate more and more zeros of the Riemann zeta function.
We think that some approaches in this article may possess importance also for other problems. First of all this is the operational approach of the transition from real and imaginary part of a function on the real or imaginary axis to an analytic function in the whole complex plane. In principle, this is possible using the Cauchy-Riemann equations but the operational approach integrates this to two integer instead of differential equations. We think that this is possible also in curved coordinates and is in particular effective starting from curves of constant real or imaginary part of one of these functions on a curve.
One of the fascinations of prime number theory is the relation of the apparently chaotic distribution function of prime numbers $\pi(x)$ on the real axis $x\ge 0$ to a fully well-ordered analytic function, the Riemann zeta function $\zeta(s)$, at least, in its representation in sum form as a special Dirichlet series and thus providing the relations between multiplicative and additive representations of arithmetic functions.
{\setcounter{section}{20}}
\section*{Acknowledgement}
I am very grateful to my son Arne W\"{u}nsche for different help. He helped me very much with the arrangement of my computer and if there were problems with it and also with the installation and use of programmes and by often simple questions he gave me new suggestions.
{\setcounter{chapter}{21}
\setcounter{equation}{0}
\aeqn
\section*{Appendix A.\\ Transformation of the Xi function\\}
In this Appendix we transform the function $\xi(s)$ defined in (\ref{xirep1}) by means of the zeta function $\zeta(s)$ from the form taken from (\ref{zetarep2}) to the form (\ref{xirep2}) using the Poisson summation formula. The Poisson summation formula is the transformation of a sum over a lattice into a sum over the reciprocal lattice. More generally, in one-dimensional case the decomposition of a special periodic function $F(q)=F(q+a)$ with period $a$ defined by the following series over functions $f(q+na)$
\begin{eqnarray}
F(q) \;\equiv\; \sum_{n=-\infty}^{+\infty}f(q+na) \;=\; F(q+a),
\end{eqnarray}
can be transformed into the reciprocal lattice providing a Fourier series as follows. For this purpose we expand $F(q)$ in a Fourier series with Fourier coefficients $F_m$ and make then obvious transformations ($q'+na \rightarrow q'',\,\dis{\sum_{n=-\infty}^{+\infty}\int_{na}^{(n+1)a}dq'' \rightarrow \int_{-\infty}^{+\infty}dq''} $ and changing the order of summation and integration) according to
\begin{eqnarray}
F(q) &=& \sum_{m=-\infty}^{+\infty}\underbrace{\left(\frac{1}{a}\int_{0}^{a}dq'\sum_{n=-\infty}^{+\infty}f(q'+na)\exp\left(-\I m\frac{2\pi q'}{a}\right)\right)}_{=\,F_m}\exp\left(\I m \frac{2\pi q}{a}\right) \nonumber\\ &=&
\frac{1}{a}\sum_{m=-\infty}^{+\infty}\left(\sum_{n=-\infty}^{+\infty}\int_{na}^{(n+1)a}dq''\,f(q'')\exp\left(-\I m\frac{2\pi q''}{a}\right)\right) \exp\left(\I m\frac{2\pi q}{a}\right) \nonumber\\ &=& \frac{1}{a}\sum_{m=-\infty}^{+\infty} \tilde{f}\left(m\frac{2\pi}{a} \right)\exp\left(\I m\frac{2\pi q}{a}\right),\label{poisson1}
\end{eqnarray}
where the coefficients $F_{m}$ of the decomposition of $F(q)$ are given by the Fourier transform $\tilde{f}(k)$ of the function $f(q)$ defined in the following way
\begin{eqnarray}
\tilde{f}(k) \;=\; \int_{-\infty}^{+\infty}dx f(q)\exp\left(-\I kq\right),\quad \Rightarrow \quad
\tilde{f}\left(m\frac{2\pi}{a}\right) \;=\; \int_{-\infty}^{+\infty}dq f(q)\exp\left(-\I m \frac{2\pi q}{a}\right).\label{fourcoeff}
\end{eqnarray}
Using the period $b= \frac{2\pi}{a}$ of the reciprocal lattice relation on the right-hand side of (\ref{poisson1}) it may be written in the forms
\begin{eqnarray}
\sum_{n=-\infty}^{+\infty}f(q+na) &=& \sum_{n=-\infty}^{+\infty}f\left(q+n\frac{2\pi}{b}\right)\;=\; \frac{b}{2\pi}\sum_{m=-\infty}^{+\infty} \tilde{f}\left(mb\right)\exp\left(\I mb q\right) \nonumber\\ &=& \frac{1}{a}\sum_{m=-\infty}^{+\infty} \tilde{f}\left(m\frac{2\pi}{a} \right)\exp\left(\I m\frac{2\pi}{a}q\right), \qquad ab = 2\pi.\qquad \label{poisson3}
\end{eqnarray}
In the special case $q=0$ one obtains from (\ref{poisson1}) the well-known basic form of the Poisson summation formula
\begin{eqnarray}
F(0) \;\equiv \; \sum_{n=-\infty}^{+\infty}f(na) \;=\;
\frac{b}{2\pi}\sum_{m=-\infty}^{+\infty}\tilde{f}\left(mb\right), \qquad ab=2\pi.\label{poisson2}
\end{eqnarray}
Formula (\ref{poisson2}) applied to the sum $\dis{\sum_{n=-\infty}^{+\infty}\exp\left(-\pi n^2q^2\right)}$ corresponding to $f(q)=\exp\left(-\pi q^2\right)$ with Fourier transform $\tilde{f}(k)=\exp\left(-\frac{k^2}{4\pi}\right)$ provides a relation which can be written in the following symmetric form (we need it in the following only for $q\ge 0$)
\begin{eqnarray}
{\mit \Psi}(q) &\equiv & \sqrt{\left|q\right|}\left\{\frac{1}{2}+\sum_{n=1}^\infty
\exp\left(-\pi n^2 q^2\right)\right\} \;=\;\frac{\sqrt{\left|q\right|}}{2} \sum_{n=-\infty}^{+\infty} \exp\left(-\pi n^2 q^2\right) \nonumber\\ &=& \frac{1}{2\sqrt{\left|q\right|}} \sum_{m=-\infty}^{+\infty}
\exp\left(-\pi \frac{m^2}{q^2}\right) \;=\; \frac{1}{\sqrt{\left|q\right|}} \left\{\frac{1}{2}+\sum_{m=1}^\infty \exp\left(-\pi \frac{m^2}{q^2}\right)\right\} \;\equiv\; {\mit \Psi}\left(\frac{1}{q}\right).
\label{defcappsi}
\end{eqnarray}
This is essentially a transformation of the Theta function $\vartheta_3\left(u,q\right)$ in special case $\frac{2{\mit \Psi}(q)}{\scr{\sqrt{\left|q\right|}}} \equiv \vartheta_3\left(0,\E^{-\pi q^2}\right)$. We now apply this to a transformation of the function $\xi(s)$.
From (\ref{xirep2}) and (\ref{zetarep2}) follows
\begin{eqnarray}
\xi(s) &=& s(s-1)\int_{0}^{+\infty}dq\,q^{s-1} \sum_{n=1}^\infty\exp\left(-\pi n^2q^2\right) \nonumber\\ &=& s(s-1)\left\{\int_{0}^{1}dq\,q^{s-1} \sum_{n=1}^\infty\exp\left(-\pi
n^2q^2\right)+\int_{1}^{+\infty}dq\,q^{s-1} \sum_{n=1}^\infty\exp\left(-\pi n^2q^2\right) \right\}.\label{xi1}
\end{eqnarray}
The second term in braces is convergent for arbitrary $q$ due to the rapid vanishing of the summands of the sum for $q\rightarrow \infty$. To the first term in braces we apply the Poisson summation formula (\ref{poisson2}) and obtain from the special result
(\ref{defcappsi})
\begin{eqnarray}
\int_{0}^{1}dq\,q^{s-1} \sum_{n=1}^\infty\exp\left(-\pi n^2q^2\right) &=& \int_{0}^{1}dq\,q^{s-1} \left\{\frac{1}{2}\left(\frac{1}{q}-1\right)+ \frac{1}{q}\sum_{m=1}^{\infty}
\exp\left(-\pi\frac{m^2}{q^2}\right)\right\}\nonumber\\ &=& \frac{1}{2s(s-1)}+\int_{1}^{+\infty}dq'\, q'^{-s}\sum_{m=1}^\infty \exp\left(-\pi m^2 q'^2\right),
\end{eqnarray}
with the substitution $q=\frac{1}{q'}$ of the integration variable made in last line. Thus from (\ref{xi1}) we find
\begin{eqnarray}
\xi(s) &=&
\frac{1}{2}-s(1-s)\int_{1}^{+\infty}dq\,\frac{q^{s}+q^{1-s}}{y} \sum_{n=1}^\infty\exp\left(-\pi n^2q^2\right).
\end{eqnarray}
With the substitution of the integration variable
\begin{eqnarray}
q=\E^u,\quad \left(q\ge 0,\; \Leftrightarrow \; -\infty < u <+\infty\right),\label{yex}
\end{eqnarray}
and with displacement of the complex variable $s$ to $z \equiv s+\frac{1}{2}$ and introduction of ${\mit \Xi}\left(z\right)$ instead of $\xi(s)$ this leads to the representation
\begin{eqnarray}
{\mit \Xi}\left(z\right)&\equiv & \xi\left(\frac{1}{2}+z\right)\;=\; \frac{1}{2}
-2\left(\frac{1}{4}-z^2\right)\int_{0}^{+\infty}du\,{\rm ch}\left(u z\right)\,\E^{\frac{u}{2}} \sum_{n=1}^\infty \exp\left(-\pi n^2 \E^{2u}\right),\label{capxirepa3}
\end{eqnarray}
given in (\ref{capxirep2}). In Appendix B we transform this representation by means of partial integration to a form which due to symmetries is particularly appropriate for the further
considerations about the Riemann zeta function.
Using the substitution (\ref{yex}) we define a function ${\mit \Phi}(u)$ by means of the function ${\mit \Psi}(y)$ in (\ref{defcappsi}) as follows
\begin{eqnarray}
{\mit \Phi}\left(u\right) & \equiv & {\mit \Psi}\left(\E^{u}\right),\quad \Rightarrow \quad {\mit
\Phi}\left(0\right) \;=\; {\mit \Psi}\left(1\right) \;=\; 0.543217\label{defcapphi},
\end{eqnarray}
and explicitly due to Poisson summation formula
\begin{eqnarray}
{\mit \Phi}\left(u\right) & \equiv & \E^{\frac{u}{2}}\left\{\frac{1}{2}+\sum_{n=1}^\infty \exp\left(-\pi n^2 \E^{2u}\right)\right\} \;=\; \frac{1}{2}\E^{\frac{u}{2}}\sum_{n=-\infty}^{+\infty} \exp\left(-\pi
n^2 \E^{2u}\right) \nonumber\\ &=& \frac{1}{2}\E^{-\frac{u}{2}}\sum_{n=-\infty}^{+\infty}
\exp\left(-\pi n^2 \E^{-2u}\right) \;=\; \E^{-\frac{u}{2}}\left\{\frac{1}{2}+\sum_{n=1}^\infty \exp\left(-\pi n^2 \E^{-2u}\right)\right\}.
\end{eqnarray}
From ${\mit \Psi}(y) = {\mit \Psi}\!\left(\frac{1}{y}\right)$ according to (\ref{defcappsi}) follows that ${\mit \Phi}\left(u\right)$ is a symmetric function
\begin{eqnarray}
{\mit \Phi}\left(u\right) = {\mit \Phi}\left(-u\right) = {\mit \Phi}\left(\left|u\right|\right).
\end{eqnarray}
Therefore, all even derivatives of ${\mit \Phi}(u)$ are also symmetric functions, whereas all odd derivatives of ${\mit \Phi}(u)$ are antisymmetric functions (we denote these derivatives by ${\mit
\Phi}^{\left(n\right)}(u)\equiv \frac{\partial^n {\mit \Phi}}{\partial u}(u)$)
\begin{eqnarray}
{\mit \Phi}^{\left(2m\right)}\left(u\right) &=& +{\mit \Phi}^{\left(2m\right)}\left(-u\right)\;=\; {\mit \Phi}^{\left(2m\right)}\left(\left|u\right|\right),\nonumber\\ {\mit \Phi}^{\left(2m+1\right)}\left(u\right) &=& -{\mit \Phi}^{\left(2m+1\right)}\left(-u\right),\quad \Rightarrow \quad {\mit \Phi}^{\left(2m+1\right)}\left(0\right)\;=\;0, \quad \left(m=0,1,2,\ldots\right).\label{symmcapphi}
\end{eqnarray}
Explicitly, one obtains for the first two derivatives
\begin{eqnarray}
{\mit \Phi}^{\left(1\right)}\left(u\right) &=& \frac{1}{2}\E^{\frac{u}{2}}\left\{\frac{1}{2}+ \sum_{n=1}^\infty \left(1-4\pi n^2\E^{2u}\right) \exp\left(-\pi n^2 \E^{2u}\right)\right\},\nonumber\\ {\mit \Phi}^{\left(2\right)}\left(u\right) &=&
\frac{1}{4}\E^{\frac{u}{2}}\left\{\frac{1}{2}+ \sum_{n=1}^\infty \left(1-6\left(4\pi n^2\E^{2u}\right)+\left(4\pi n^2\E^{2u}\right)^2\right) \exp\left(-\pi n^2 \E^{2u}\right)\right\}.
\label{derivcapphi}
\end{eqnarray}
As a subsidiary result we obtain from vanishing of the odd derivatives of ${\mit \Phi}(u)$ at $u=0$ that means from ${\mit \Phi}^{\left(2m+1\right)}\left(0\right)=0,\,(m=0,1,2,\ldots)$ an infinite sequence of special sum evaluations from which the first three are
\begin{eqnarray}
&& \hspace{-5mm}\sum_{n=1}^\infty \left(4\,\pi n^2-1\right)\exp\left(-\pi n^2\right) \;=\; \frac{1}{2},\nonumber\\ && \hspace{-5mm} \sum_{n=1}^\infty \left(\left(4\,\pi n^2\right)^3-15\left(4\,\pi n^2\right)^2 +31\left(4\,\pi n^2\right)-1\right)\exp\left(-\pi n^2\right) \;=\; \frac{1}{2}.
\label{derivcapphizero}
\end{eqnarray}
We checked relations (\ref{derivcapphizero}) numerically by computer up to a sufficiently high precision. We also could not find (\ref{gentheta}) among the known transformations of theta functions. The interesting feature of these sum evaluations is that herein power functions as well as exponential functions containing the transcendental number $\pi$ in the exponent are involved in a way which finally leads to a rational number that should also be attractive for recreation mathematics. In contrast, in the well-known series for the trigonometric functions one obtains for certain rational multiples of $\pi$ as argument also rational numbers but one has involved there only power functions with rational coefficients that means rational functions although an infinite number of them.
Using the function ${\mit \Phi}(u)$ the function ${\mit \Xi}(z)$ in (\ref{capxirepa3}) can be represented as
\begin{eqnarray}
{\mit \Xi}\left(z\right) &=& \frac{1}{2} +2\left(z^2-\frac{1}{4}\right)\int_{0}^{+\infty}du\,\left({\mit \Phi}\left(u\right)-\frac{1}{2}\,\E^{\frac{u}{2}}\right)\,{\rm ch}\left(uz\right) \nonumber\\ &=& \frac{1}{2} +2\int_{0}^{+\infty}du\,\left({\mit \Phi}\left(u\right)-\frac{1}{2}\,\E^{\frac{u}{2}}\right) \left\{\left(\frac{\partial^2}{\partial
u^2}-\frac{1}{4}\right)\,{\rm ch}\left(uz\right)\right\}.\label{capxirepa4}
\end{eqnarray}
From this we obtain by partial integration
\begin{eqnarray}
{\mit \Xi}\left(z\right) &=& 2\int_{0}^{+\infty}du\,\left\{\left(\frac{\partial^2}{\partial
u^2}-\frac{1}{4}\right)\left({\mit \Phi}\left(u\right)-\frac{1}{2}\,\E^{\frac{u}{2}}\right)\right\}\,{\rm ch}\left(uz\right),\label{capxirepa5}
\end{eqnarray}
where the contribution from the lower integration limit at $u=0$ has exactly canceled the constant term $\frac{1}{2}$ on the right-hand side of (\ref{capxirepa4}) and the contributions from the upper limit $x\rightarrow +\infty$ is vanishing. Using (\ref{derivcapphi}) we find with abbreviation ${\mit \Omega}(u)$ according to
\begin{eqnarray}
{\mit \Omega}(u) &\equiv & 2\left(\frac{\partial^2}{\partial u^2}-\frac{1}{4}\right)\left({\mit
\Phi}\left(u\right)-\frac{1}{2}\,\E^{\frac{u}{2}}\right),
\end{eqnarray}
the following basic structural form of the Xi function
\begin{eqnarray}
{\mit \Xi}\left(z\right) &=& \int_{0}^{+\infty}du\,{\mit \Omega}(u)\,{\rm ch}\left(uz\right),\label{capxirepa6}
\end{eqnarray}
with the following explicit representation of ${\mit \Omega}(u)$
\begin{eqnarray}
{\mit \Omega}(u) &=& 2{\mit \Phi}^{\left(2\right)}\left(u\right)-\frac{1}{2}\,{\mit
\Phi}\left(u\right) \nonumber\\ &=& 4 \E^{\frac{u}{2}}\sum_{n=1}^\infty \pi n^2\E^{2u}\left(2\pi n^2 \E^{2u}-3\right)\exp\left(-\pi n^2\E^{2u}\right) \;>\;0.\label{defcapomega1}
\end{eqnarray}
Since according to (\ref{symmcapphi}) the even derivatives of ${\mit \Phi}(u)$ are symmetric functions it follows from relation (\ref{defcapomega1}) that ${\mit \Omega}(u)$ is also a symmetric function and (\ref{omegasym}) holds. This is not immediately seen from the explicit
representation (\ref{defcapomega1}). Furthermore, ${\mit \Omega}(u)$ is positively definite for $-\infty < u < +\infty $ since the factor $\left(2\pi n^2 \E^{2u}-3\right)$ in
(\ref{defcapomega1}) is positive for $n\ge 1$ and $u\ge 0$ and all other factors too. It goes rapidly to zero for $u\rightarrow \pm\infty$, more rapidly than any exponential function $\exp\left(-\gamma|x|^\rho\right)$ with arbitrary $\gamma>0$ and arbitrary $\rho >0$ due to factors $\exp\left(-\pi n^2 \E^{2u}\right)$ in the sum terms in (\ref{defcapomega1}). For
the first derivative of ${\mit \Omega}(u)$ we find
\begin{eqnarray}
{\mit \Omega}^{\left(1\right)}(u) &=& 2{\mit \Phi}^{\left(3\right)}\left(u\right)-\frac{1}{2}\,{\mit \Phi}^{\left(1\right)}\left(u\right) \nonumber\\ &=& -2 \E^{\frac{u}{2}}\sum_{n=1}^\infty \pi n^2\E^{2u}\left(8\left(\pi
n^2\E^{2u}\right)^2-30\pi n^2\E^{2u}+15\right)\exp\left(-\pi n^2 \E^{2u}\right) \nonumber\\ &=& -{\mit \Omega}^{\left(1\right)}(-u), \quad \Rightarrow \quad {\mit \Omega}^{\left(1\right)}(0)=0,\quad {\mit \Omega}^{\left(1\right)}(u) \;<\; 0,\quad \left(u > 0\right).
\end{eqnarray}
It is vanishing for $u=0$ due to its antisymmetry and negatively definite for $u > 0$ as the negative sign of ${\mit \Omega}^{\left(2\right)}(0)$ together with considerations of the sum
for $u>0$ show (i.e., the polynomial $f(N)\equiv 8\pi^2 N^2-30\pi N+15 \ge 0 $ for $N\ge \frac{15+\sqrt{105}}{8\pi} \approx 1.00454 $ and negativity is already obtained taking the first two sum terms to $n=1$ and $n=2$ alone). Thus ${\mit \Omega}(u)$ is monotonically decreasing for $u\ge 0$. A few approximate numerical values of parameters for the function ${\mit \Omega}(u)$ are
\begin{eqnarray}
&& {\mit \Omega}_0 \;\equiv \; \int_{0}^{+\infty}du\,{\mit \Omega}(u) \;=\; -\int_{0}^{+\infty}du\,{\mit \Omega}^{(1)}(u)u \;=\; 0.497121,\nonumber\\ && {\mit \Omega}(0) \;=\; - \int_{0}^{+\infty}dx\,{\mit \Omega}^{(1)}(u) \;=\; 1.78679,\quad {\mit \Omega}^{\left(1\right)}(0) \;=\; - \int_{0}^{+\infty}du\,{\mit \Omega}^{(2)}(u) \;=\; 0.
\label{numcapomega}
\end{eqnarray}
In next Appendix we consider the transition from analytic functions given on the real or imaginary axis to the whole complex plane.
}
{\setcounter{chapter}{15}
\beqn
\setcounter{equation}{0}
\section*{Appendix B.\\ Transition from analytic functions on real or imaginary axis to whole complex plane}
The operator $\frac{\partial}{\partial x}$ is the infinitesimal displace operator and $\exp\left(-x_0\frac{\partial}{\partial x}\right)$ the finite displacement operator for the displacement of the argument of a function $f(x) \rightarrow f(x-x_0)$. In complex analysis the
real variable $x$ can be displaced with sense for an analytic function to the complex variable $z=x+\I y$ in the whole complex plane by
\begin{eqnarray}
\exp\left(\I y \frac{\partial}{\partial x}\right)x \exp\left(-\I y \frac{\partial}{\partial x}\right) &=& x+\frac{\I y}{1!}\left[\frac{\partial}{\partial x},x \right] + \frac{(\I y)^2}{2!}\left[\frac{\partial}{\partial x}\left[\frac{\partial}{\partial x},x\right]\right]+\ldots \;=\; x+\I y, \qquad\label{xdisply0}
\end{eqnarray}
where $\left[A,B\right] \equiv AB-BA $ denotes the commutator of two operators $A$ and $B$, in particular $\left[\frac{\partial}{\partial x},x\right] =1$ and (\ref{xdisply0}) may be written in the form
\begin{eqnarray}
\exp\left(\I y \frac{\partial}{\partial x}\right)x &=& (x+\I y)\exp\left(\I y \frac{\partial}{\partial x}\right).\label{xdisply}
\end{eqnarray}
Analogously, the transition from the variable $y$ on the imaginary axis $\I y$ to the variable $z=\I(y-\I x)=x+\I y $ in the whole complex plane may be written as
\begin{eqnarray}
\exp\left(-\I x \frac{\partial}{\partial y}\right)\I y &=& (x+\I y)\exp\left(-\I x \frac{\partial}{\partial y}\right),\qquad \; \label{ydisplx}
\end{eqnarray}
In the following we consider only the case (\ref{xdisply}) since the case (\ref{ydisplx}) is
completely analogous with simple substitutions.
We wrote the equations (\ref{xdisply0}), (\ref{xdisply}) and (\ref{ydisplx}) in a form which we call operational form and meaning that they may be applied to further functions on the left-hand and correspondingly right-hand side\footnote{Non-operational form would be if we write, for example, $\exp\left(\I y \frac{\partial}{\partial x}\right)x = x+\I y$ instead of (\ref{xdisply}) which is correct but cannot be applied to further functions $f(x) \neq const\cdot 1 $, for example to $f(x)=x$.)}. It is now easy to see that an analytic function $w(z)=w(x+\I y); \frac{\partial}{\partial z^*}w(z) =0$ can be generated from $w(x)$ on the $x$-axis in operational form by
\begin{eqnarray}
\exp\left(\I y \frac{\partial}{\partial x}\right)w(x) &=& w(x+\I y)\exp\left(\I y \frac{\partial}{\partial x}\right),\label{fxdisplfxy}
\end{eqnarray}
and analogously from $w(\I y)$ on the imaginary axis by
\begin{eqnarray}
\exp\left(-\I x \frac{\partial}{\partial x}\right)w(\I y) &=& w(x+\I y)\exp\left(-\I x \frac{\partial}{\partial y}\right).\label{fydisplfxy}
\end{eqnarray}
Writing the function $w(z)$ with real part $u(x,y)$ and imaginary part $v(x,y)$ in the form
\begin{eqnarray}
&& w(x+\I y)= u(x,y)+\I v(x,y),\quad \left(w(x+\I y)\right)^* = u(x,y)-\I v(x,y), \quad \Rightarrow \nonumber\\&& u(x,y) \;=\; \frac{1}{2}\Big(w(x+\I y)+ \left(w(x+\I y)\right)^*\Big),\quad v(x,y) \;=\; -\frac{\I}{2}\Big(w(x+\I y)-\left(w(x+\I y)\right)^*\Big),\qquad
\end{eqnarray}
we find from (\ref{fxdisplfxy})
\begin{eqnarray}
\exp\left(\I y \frac{\partial}{\partial x}\right)w(x) &=& \exp\left(\I y \frac{\partial}{\partial x}\right)\Big(u(x,0)+\I v(x,0)\Big) \nonumber\\ &=& \Big(u(x+\I y,0)+ \I v(x+\I y,0)\Big)\exp\left(\I y \frac{\partial}{\partial x}\right) \nonumber\\ &=& \Big(u(x,y)+ \I v(x,y)\Big)\exp\left(\I y \frac{\partial}{\partial x}\right),\label{fxdisplfxyplus}
\end{eqnarray}
and correspondingly
\begin{eqnarray}
\left(\exp\left(\I y \frac{\partial}{\partial x}\right)w(x)\right)^* &=& \exp\left(-\I y \frac{\partial}{\partial x}\right)\Big(u(x,0)-\I v(x,0)\Big) \nonumber\\ &=& \Big(u(x-\I y,0)- \I v(x-\I y,0)\Big)\exp\left(-\I y \frac{\partial}{\partial x}\right) \nonumber\\ &=& \Big(u(x,y)- \I v(x,y)\Big)\exp\left(-\I y \frac{\partial}{\partial x}\right).\label{fxdisplfxyminus}
\end{eqnarray}
From (\ref{fxdisplfxyplus}) and (\ref{fxdisplfxyminus}) follows forming the sum and the difference
\begin{eqnarray}
\cos\left(y \frac{\partial}{\partial x}\right)u(x,0) -\sin\left(y \frac{\partial}{\partial x}\right)v(x,0) &=& u(x,y)\cos\left(y \frac{\partial}{\partial x}\right) -v(x,y)\sin\left(y \frac{\partial}{\partial x}\right), \nonumber\\ \sin\left(y \frac{\partial}{\partial x}\right)u(x,0) +\cos\left(y \frac{\partial}{\partial x}\right)v(x,0) &=& u(x,y)\sin\left(y \frac{\partial}{\partial x}\right) +v(x,y)\cos\left(y \frac{\partial}{\partial x}\right).\qquad
\label{fxdisplfxyop}
\end{eqnarray}
These are yet operational identities which can be applied to arbitrary functions $f(x)$. Applied to the function $f(x)=1$ follows
\begin{eqnarray}
\cos\left(y \frac{\partial}{\partial x}\right)u(x,0) -\sin\left(y \frac{\partial}{\partial x}\right)v(x,0) &=& u(x,y), \nonumber\\ \sin\left(y \frac{\partial}{\partial x}\right)u(x,0) +\cos\left(y \frac{\partial}{\partial x}\right)v(x,0) &=& v(x,y).
\label{fxdisplfxynonop}
\end{eqnarray}
In full analogy we may derive the continuation of an analytic function from the imaginary axes $z=\I y$ to the whole complex plane $z=x+\I y$ in operational form
\begin{eqnarray}
\cos\left(x\frac{\partial}{\partial y}\right)u(0,y) +\sin\left(x\frac{\partial}{\partial y}\right)v(0,y) &=& u(x,y)\cos\left(x\frac{\partial}{\partial y}\right) +v(x,y)\sin\left(x\frac{\partial}{\partial y}\right), \nonumber\\ -\sin\left(x\frac{\partial}{\partial y}\right)u(0,y) +\cos\left(x\frac{\partial}{\partial y}\right)v(0,y) &=& -u(x,y)\sin\left(x\frac{\partial}{\partial y}\right) +v(x,y)\cos\left(x\frac{\partial}{\partial y}\right),\qquad \label{fydisplfxyop}
\end{eqnarray}
and this applied to the function $f(y)=1$
\begin{eqnarray}
\cos\left(x\frac{\partial}{\partial y}\right)u(0,y) +\sin\left(x\frac{\partial}{\partial y}\right)v(0,y) &=& u(x,y), \nonumber\\ -\sin\left(x\frac{\partial}{\partial y}\right)u(0,y) +\cos\left(x\frac{\partial}{\partial y}\right)v(0,y) &=& v(x,y). \label{fydisplfxynonop}
\end{eqnarray}
It is easy to check that both (\ref{fxdisplfxynonop}) and (\ref{fydisplfxynonop}) satisfy the Cauchy-Riemann equations
\begin{eqnarray}
\frac{\partial u}{\partial x}(x,y) \;=\; \frac{\partial v}{\partial y}(x,y),\quad \frac{\partial u}{\partial y}(x,y) \;=\; -\frac{\partial v}{\partial x}(x,y),\label{creq}
\end{eqnarray}
and it is even possible to derive these relations from these equations by Taylor series expansions of $u(x,y)$ and $v(x,y)$ in powers of $y$ or $x$ in dependence from which axis we make the continuation to the whole complex plane. For example, in expansion in powers of $y$ we obtain
using (\ref{creq}) (and the resulting equations $\left(\frac{\partial^2}{\partial x^2}+\frac{\partial^2}{\partial y^2}\right)u(x,y) =\left(\frac{\partial^2}{\partial x^2}+\frac{\partial^2}{\partial y^2}\right)v(x,y) = 0$ from them)
\begin{eqnarray}
w(x+\I y) &=& \sum_{m=0}^{\infty}\frac{y^{2m}}{(2m)!}\left(\frac{\partial^{2m}}{\partial y^{2m}}u(x,y)\right)_{y=0} + \sum_{m=0}^{\infty}\frac{y^{2m+1}}{(2m+1)!}\left(\frac{\partial^{2m+1}}{\partial y^{2m+1}}u(x,y)\right)_{y=0} \nonumber\\ && +\I\left\{\sum_{m=0}^{\infty}\frac{y^{2m}}{(2m)!}\left(\frac{\partial^{2m}}{\partial y^{2m}}v(x,y)\right)_{y=0} + \sum_{m=0}^{\infty}\frac{y^{2m+1}}{(2m+1)!}\left(\frac{\partial^{2m+1}}{\partial y^{2m+1}}v(x,y)\right)_{y=0} \right\},\qquad\quad
\end{eqnarray}
that can be written in compact form
\begin{eqnarray}
w(x+\I y) &=& \cos\left(y\frac{\partial}{\partial x}\right)u(x,0)-\sin\left(y\frac{\partial}{\partial x}\right)v(x,0) \nonumber\\ && +\I\left\{\cos\left(y\frac{\partial}{\partial x}\right)v(x,0)+\sin\left(y\frac{\partial}{\partial x}\right)u(x,0)\right\}\;=\; u(x,y) +\I v(x,y),\label{fxdisplfxynonop2}
\end{eqnarray}
and is equivalent to (\ref{fxdisplfxynonop}). Analogously by expansion in powers of $x$ as intermediate step we obtain
\begin{eqnarray}
w(x+\I y) &=& \cos\left(x\frac{\partial}{\partial y}\right)u(0,y)+\sin\left(x\frac{\partial}{\partial y}\right)v(0,y) \nonumber\\ && +\I\left\{\cos\left(x\frac{\partial}{\partial y}\right)v(0,y)-\sin\left(x\frac{\partial}{\partial y}\right)u(0,y)\right\} \;=\; u(x,y)+\I v(x,y),\label{fydisplfxyop2}
\end{eqnarray}
that is equivalent to (\ref{fydisplfxyop}).
Therefore, relations (\ref{fxdisplfxynonop2}) and (\ref{fydisplfxyop2}) represent some integral forms of the Cauchy-Riemann equations.
In cases if one of the functions $u(x,0)$ or $v(x,0)$ in (\ref{fxdisplfxynonop}) or $u(0,y)$ or $v(0,y)$ in (\ref{fydisplfxynonop}) is vanishing these formulae simplify and the case $v(0,y) = 0$ is applied in Sections 4--6.
We did not find up to now such representations in textbooks to complex analysis but it seems possible that they are somewhere.
}
\small{
| e5d1299409238970185a8114a49fbc73b409222a | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
In this paper we study a parameter estimation problem for the drift/viscosity coefficient of a linear parabolic stochastic partial differential equation (SPDE) driven by an additive space-time noise (possibly colored in space). We assume that the observable variable is one path of the solution, observed continuously on a finite time-interval as an element of an infinite dimensional Hilbert space. More precisely, we assume that the first $N$ Fourier modes of the solution are observed continuously on the time-interval $[0,T]$, and we investigate the asymptotics of the proposed estimators, as $N\to\infty$. Most of the existing literature on statistical inference for SPDEs has a similar spectral approach, starting with the seminal paper by Huebner and Rozovskii~\cite{HuebnerRozovskii}, and essentially explore the Maximum Likelihood Estimator (MLE) approach. The main idea can be summarized as follows: the measures generated by the solution $u$ of a parabolic SPDE that corresponds to different drift parameters are singular to each other (under some appropriate subordination assumption), which indicates that the true parameter can be found exactly. Indeed, by taking an appropriate finite dimensional projection of the solution, one obtains a finite dimensional system of stochastic ordinary differential equations (SODEs) for which an $\textrm{MLE}$ for the drift coefficient exists. One can show that as the dimension $N$ of the projection increases, the $\textrm{MLE}$ converges to the true parameter and it is asymptotically normal. The MLE type estimators for SPDEs are well understood, to some extent, and for more details on this method we refer to the survey~\cite{Lototsky2009Survey} and to the monograph \cite{Bishwal2008} on linear SPDEs, to~\cite{IgorNathanAditiveNS2010} for an adaptation of this method to a nonlinear SPDE, to~\cite{CXu2014,CXu2015} for the hypothesis testing for stochastic fractional heat equation, and to~\cite{MishraBishwal1995,PiterbargRozovskii1997, Markussen2003, PrakasaRao2003, MishraPrakasaRao2002} for discrete time sampling. For linear, diagonalizable\footnote{A diagonalizable SPDE is an equation for which the first $N$ Fourier coefficients of the solution form an $N$-dimensional \textit{decoupled} system of ordinary stochastic differential equation. For a formal definition, in terms of the differential operators and the structure of the noise term, see for instance~\cite{Lototsky2009Survey}.} SPDEs, the $\textrm{MLE}$ can be computed explicitly, and of course, from a statistical point of view there is no need to study other type of estimators. In general, this statement does not hold true for non-diagonalizable equations such as nonlinear SPDEs or SPDEs driven by a multiplicative noise. While the parameter estimation problem for SODEs went way beyond the MLEs (cf. the monograph~\cite{KutoyantsBook2004}), there exist a limited number of works dedicated to the non-MLE statistical inference for SPDEs. For example, in~\cite{CialencoLototsky2009} the authors explore the singularity of corresponding probability measures and derive a closed-form estimators for the drift coefficient for some linear parabolic SPDEs driven by a multiplicative noise (of special structure). This study is the first attempt to investigate different type of estimators, and it is related to what is known in the literature the trajectory fitting estimators (TFEs). Using as observations the first $N$ Fourier modes, we construct the TFE by analogy to the TFE for continuously observed finite dimensional ergodic diffusion processes first introduced by Y.~A.~Kutoyants \cite{Kutoyants1991}; see also \cite[Section 1.3 \& Section 2.3]{KutoyantsBook2004} and references therein. It would be fair to call this estimator Kutoyants Estimator, but it seems that TFE is already a well established name and thus we will follow this terminology. The TFE can be also viewed as an analog of the least squares estimators widely used in time-series analysis.
As already mentioned, we study the asymptotic properties of the TFE as $N\to\infty$, in contrast to the diffusion processes where the asymptotics are done for the large-time regime. Surely one can investigate the large-time asymptotics for SPDEs too, but we find this case to be too similar to the estimators for diffusion processes and we omitted it here. In this study, we consider a fairly simple, although general, class of SPDEs: linear, parabolic, diagonalizable equations, driven by an additive space-time noise. The diagonalizable nature of these equations, allows to derive explicit expressions for the considered estimators and for the asymptotics of their first two moments, and hence to investigate the rate of convergence of these estimators. While simple, these equations can be viewed as a good approximation of some more complicated and practically important models. On the other hand, the obtained results will serve as benchmarks for future studies of more complicated and realistic models which will be addressed in the sequel. Under some general structural assumptions we prove consistency and asymptotic normality of the proposed estimators.
The paper is organized as follows. In Section~\ref{sec:Setup} we setup the problem and give some auxiliary results. The TFE is derived in Section~\ref{sec:TFE-construction}. Section~\ref{sec:mainResults} is devoted to the main results -- consistency of TFE (Theorem~\ref{thm:Consistency}) and asymptotic normality (Theorem~\ref{thm:AsymNormal}). In Section~\ref{sec:Examples} we discuss the TFE for two particular illustrative examples of SPDEs. Due to the nature of the proofs, some proofs required nontrivial and tedious computations for which, as appropriate, we used symbolic computations in a mathematical software. All technical proofs and some auxiliary results are deferred to the Appendix.
\section{Setup of the problem and some auxiliary results}\label{sec:Setup}
Let $(\Omega,\mathscr{F},\set{\mathscr{F}_t}_{t\geq 0},\mathbb{P})$ be a stochastic basis with usual assumptions, and let $H$ be a separable Hilbert space, with the corresponding inner product $(\,\cdot\,,\,\cdot\,)_{H}$. We consider the following evolution equation
\begin{align}\label{eq:mainAbstract}
du(t)+\left(\theta\mathcal{A}_{1}+\mathcal{A}_{0}\right)u(t)\,dt&=\sigma\,dW(t),
\end{align}
with initial condition $u(0)=u_{0}\in H$, and where $\mathcal{A}_{0}$ and $\mathcal{A}_{1}$ are linear operators on $H$, $W:=\{W(t)\}_{t\geq 0}$ is a cylindrical Brownian motion in $H$, and $\sigma,\theta\in\mathbb{R}_{+}:=(0,\infty)$. We will take the continuous-time observation framework by assuming that the solution $u$, as an object in $H$, or a finite dimensional projection in $H$ of it, is observed continuously in time for all $t\in[0,T]$, and for some fixed horizon $T$. In this framework, the parameter $\sigma$ can be found exactly, using a quadratic variation argument, and thus we will assume that $\sigma$ is known. We are interested in estimating the unknown parameter $\theta\in\Theta\subset\mathbb{R}_{+}$.
We start with some structural assumptions on \eqref{eq:mainAbstract} related to the well-posedness of the solution. Throughout the paper, we assume that:
\begin{enumerate}[(i)]
\item The operators $\mathcal{A}_{0}$ and $\mathcal{A}_{1}$ have only point spectra, and a common system of eigenfunctions $\{h_{k}\}_{k\in\mathbb{N}}$ that form a complete, orthonormal system in $H$. We denote the corresponding eigenvalues of $\mathcal{A}_{0}$ and $\mathcal{A}_{1}$ by $\rho_{k}$ and $\nu_{k}$, $k\in\mathbb{N}$, respectively.
\item The sequence $\{\mu_{k}(\theta)\}_{k\in\mathbb{N}}$, where $\mu_{k}(\theta):=\theta\nu_{k}+\rho_{k}$, is such that
\begin{align}
\lim_{k\rightarrow\infty}\mu_{k}(\theta)=+\infty,
\end{align}
where the convergence is uniform in $\theta\in\Theta$.
\item There exist universal constants $J\in\mathbb{N}$ and $c_{0}>0$ such that, for any $k\geq J$ and any $\theta\in\Theta$,
\begin{align}
\frac{\nu_{k}}{\mu_{k}(\theta)}\leq c_{0}.
\end{align}
\item The sequence $\{\nu_{k}\}_{k\in\mathbb{N}}$ is such that $\lim_{k\rightarrow\infty}\nu_{k}=+\infty.$\footnote{Without loss of generality, we will assume that $\nu_{k}\geq 0$, for all $k\in\mathbb{N}$.}
\item The noise term $W$ is a cylindrical Brownian motion in $H$, and has the following form
\begin{align}\label{eq:W}
W(t)=\sum_{k=1}^{\infty}\lambda_{k}^{-\gamma}h_{k}w_{k}(t),\quad t\geq 0,
\end{align}
for some $\gamma\geq 0$, where $\lambda_{k}:=\nu_{k}^{1/(2m)}$, $k\in\mathbb{N}$, for some $m\geq 0$,\footnote{Of course, one can consider at once just $\lambda_{k}=\nu_{k}$. Our choice to consider $m$ is to put the results on par with the notations from the existing literature. As mentioned later, if $\mathcal{A}_{0}$ and $\mathcal{A}_{1}$ are some pseudo-differential operators, then it is convenient to denote by $2m$ the order of the leading order operator.} and where $w_{k}:=\{w_{k}(t)\}_{t\geq 0}$, $k\in\mathbb{N}$, is a collection of independent standard Brownian motions.
\end{enumerate}
Conditions (i)--(v) imply that the equation \eqref{eq:mainAbstract} is linear, diagonalizable, parabolic, and that the solution exists and is unique; this can be established by standard methods from theory of SPDEs and we refer, for instance, to~\cite{Lototsky2009Survey, HuebnerLototskyRozovskii97, HuebnerRozovskii} for similar setup, or to~\cite{RozovskiiBook,ChowBook} for a general theory. Of course, one class of operators $\mathcal{A}_{0}$ and $\mathcal{A}_{1}$ that satisfy the above conditions are pseudo-differential operators on bounded domains with appropriate boundary conditions, with $\mathcal{A}_{0}$ being subordinated to $\mathcal{A}_{1}$ -- see Section \ref{sec:Examples} for some particular examples.
It is out of the scope of this work to go through the constructions of the scale of Sobolev space, or to investigate the exact order of regularity of the solution. We will only mention that generally speaking the series \eqref{eq:W} may diverge, but one can enlarge the underlying space $H$, such that $W$ is a square-integrable martingale. For example, let $\mathcal{X}$ be the closure of $H$ with respect to the norm
$\left\|f\right\|_{\mathcal{X}}:=\big(\sum_{k=1}^{\infty}k^{-2}\left(f,h_{k}\right)_{H}^{2}\big)^{1/2}$. Then, the unique solution to \eqref{eq:mainAbstract}, with initial condition $u(0)=u_{0}$, is a continuous (in mean-square sense) $\mathcal{X}$-valued stochastic process given by
\begin{align}
u(t)=\sum_{k=1}^{\infty}u_{k}(t)h_{k},\quad t\geq 0,
\end{align}
where, for each $k\in\mathbb{N}$, $u_{k}:=\{u_{k}(t)\}_{t\geq 0}$ satisfies the following ordinary stochastic differential equation (SDE)
\begin{align}\label{eq:FModes}
du_{k}(t)+ \mu_{k}(\theta)\,u_{k}(t)\,dt=\sigma\lambda_{k}^{-\gamma}\,dw_{k}(t),
\end{align}
with initial condition $u_{k}(0)=(u_{0},h_{k})_{H}$. The stochastic processes $u_{k}$, $k\in\mathbb{N}$, are the Fourier modes of the solution $u$ with respect to the basis $\{h_{k}\}_{k\in\mathbb{N}}$ of $H$, i.e., $u_{k}(t)=(u(t),h_{k})_{H}$, $t\geq 0$, $k\in\mathbb{N}$. Note that the SDEs of the form \eqref{eq:FModes}, for $k\in\mathbb{N}$, provide an infinite system of decoupled/independent Ornstein--Uhlenbeck processes. As already mentioned, the decoupling nature of Fourier modes, or the diagonalizable property of the original equation, will play a critical role in our study, and it is essentially guaranteed by the assumptions (i) and (v). By It\^{o}'s formula, clearly we have
\begin{align}\label{eq:FModeSols}
u_{k}(t)=e^{-\mu_{k}(\theta)t}u_{k}(0)+\sigma\lambda_{k}^{-\gamma}e^{-\mu_{k}(\theta)t}\int_{0}^{t}e^{\mu_{k}(\theta)s}\,dw_{k}(s),\quad t\geq 0,\quad k\in\mathbb{N}.
\end{align}
\subsection{Trajectory Fitting Estimators}\label{sec:TFE-construction}
The trajectory fitting estimators for continuous-time diffusion processes can be viewed as an analog of the least squares estimators widely used in time-series analysis. Following~\cite[Section 1.3 \& Section 2.3]{KutoyantsBook2004}, we will briefly describe the TFEs for finite-dimensional diffusions. Assume that the observed process $S(\theta):=\{S(t;\theta)\}_{t\geq 0}$ follows the dynamics
\begin{align}\label{eq:SDES}
dS(t;\theta)=b(\theta,S(t;\theta))dt+\sigma(S(t;\theta))\,dB(t),
\end{align}
where $B:=\{B(t)\}_{t\geq 0}$ is a one-dimensional standard Brownian motion, and $\theta$ is the parameter of interest. We assume that the drift $b$ and the volatility $\sigma$ are known, and that the solution to \eqref{eq:SDES} (with certain initial condition $S(0,\theta)=S_{0}$) exists and is unique, for any $\theta\in\Theta$. Let $F:\mathbb{R}\rightarrow\mathbb{R}$ be a twice continuously differentiable function. By It\^{o}'s formula,
\begin{align}
F(S(t;\theta))&=F(S_{0})+\int_{0}^{t}\left(F'(S(s;\theta))b\left(\theta,S(s;\theta)\right)+\frac{1}{2}F''(S(s;\theta))\sigma^{2}(S(s;\theta))\right)ds\\
&\quad\,\,+\int_{0}^{t}F'(S(s;\theta))\sigma(S(s;\theta))\,dB(s).
\end{align}
For any $\theta\in\Theta$ and $t\in[0,T]$, let
\begin{align}
\widetilde{F}(t;\theta):=F(S_{0})+\int_{0}^{t}\left(F'(S(s;\theta))b\left(\theta,S(s;\theta)\right)+\frac{1}{2}F''(S(s;\theta))\sigma^{2}(S(s;\theta))\right)ds.
\end{align}
The \emph{trajectory fitting estimator}\footnote{The terminology comes from the fact that the estimator is obtained by fitting the observed trajectory with the artificial one.} $\widetilde{\theta}_{T}$ of $\theta$ is then defined as the solution to the minimization problem
\begin{align}\label{eq:TFESDE}
\widetilde{\theta}_{T}:=\arginf_{\theta\in\Theta}\int_{0}^{T}\left(F(S(t;\theta))-\widetilde{F}(t;\theta)\right)^{2}dt.
\end{align}
The choice of function $F$ depends on the underlying models, and has to be taken such that the estimator satisfies the desired asymptotic properties (consistency, asymptotic normality, etc).
In this study, it is enough to consider the quadratic function $F(x)=x^{2}$, the choice we make throughout. For each Fourier mode $u_{k}$, $k\in\mathbb{N}$, by It\^{o}'s formula, we have
\begin{equation}\label{eq:uk2}
u_{k}^{2}(t)=u_{k}^{2}(0)+\int_{0}^{t}\left(\sigma^{2}\lambda_{k}^{-2\gamma}-2\mu_{k}(\theta)u_{k}^{2}(s)\right)ds+2\sigma\lambda_{k}^{-\gamma}\int_{0}^{t}u_{k}(s)\,dw_{k}(s),\quad t\geq 0.
\end{equation}
By \eqref{eq:TFESDE}, one can easily construct a TFE for $\theta$ based on the trajectory on $[0,T]$, for some fixed horizon $T>0$, of each Fourier mode $u_{k}$. The \emph{long-time} asymptotic behavior of such estimators as $T\rightarrow\infty$ has been well investigated (cf.~\cite{KutoyantsBook2004}), and thus we will omit it here.
By analogy to the construction of maximum likelihood estimators for SPDEs (cf.~\cite{Lototsky2009Survey}), we will construct a TFE for the unknown parameter $\theta$ based on the trajectories of the first $N$ Fourier modes of the solution. Moreover, for a fixed horizon $T>0$, we will study the \emph{large-space} asymptotic behavior of the TFE as the number of the Fourier modes increases, which is a distinguished feature for an infinite dimensional evolution system. Specifically, fix any $T>0$, and for any $\theta\in\Theta$, let
\begin{equation}\label{eq:uhat}
V_{k}(t;\theta):=u_{k}^{2}(0)+\int_{0}^{t}\left(\sigma^{2}\lambda_{k}^{-2\gamma}-2\mu_{k}(\theta)u_{k}^{2}(s)\right)ds,
\quad k\in\mathbb{N},\quad t\in [0,T].
\end{equation}
The TFE for the unknown parameter $\theta$ is defined as
\begin{equation}
\widetilde{\theta}_{N} =\widetilde{\theta}_{N}(\gamma,T,\sigma,m):=\arginf_{\theta\in\Theta}\sum_{k=1}^{N}\int_{0}^{T}\left(V_{k}(t;\theta)-u_{k}^{2}(t)\right)^{2}dt.
\end{equation}
We are interested in the asymptotic properties of $\widetilde{\theta}_{N}$, as $N\to\infty$, with $T$ being fixed.
In what follows, we will make use of the following notations. For any $t\in[0,T]$, let
\begin{align}\label{eq:notation}
\xi_{k}(t)&:=\!\int_{0}^{t}u_{k}^{2}(s)\,ds,\,\,\,\,X_{k}(t):=\!\int_{0}^{t}s\xi_{k}(s)\,ds
,\,\,\,\,Y_{k}(t):=\!\int_{0}^{t}\xi_{k}(s)\,ds,\,\,\,\,Z_{k}(t):=\!\int_{0}^{t}\xi_{k}^{2}(s)\,ds.
\end{align}
One advantage of the TFE is that it can be given by an explicit formula that does not contain a stochastic integral. Indeed, by \eqref{eq:uk2} and \eqref{eq:uhat},
\begin{equation}
\sum_{k=1}^{N}\int_{0}^{T}\left(V_{k}(t;\theta)-u_{k}^{2}(t)\right)^{2}dt=\sum_{k=1}^{N}\int_{0}^{T}\left(u_{k}^{2}(0)+\sigma^{2}\lambda_{k}^{-2\gamma}t-2\rho_{k}\xi_{k}(t)-u_{k}^{2}(t)-2\theta\nu_{k}\xi_{k}(t)\right)^{2}dt.
\end{equation}
The maximizer of the last expression, with respect to $\theta$, can be computed simply by finding the root of the first-order derivative. This yields the following explicit expression for the TFE
\begin{equation}\label{eq:thetaMain}
\widetilde{\theta}_{N}=-\frac{\sum_{k=1}^{N}\nu_{k}\left(\frac{1}{2}\xi_{k}^{2}(T)-u_{k}^{2}(0)Y_{k}(T)-\sigma^{2}\lambda_{k}^{-2\gamma}X_{k}(T)+2\rho_{k}Z_{k}(T)\right)}{2\sum_{k=1}^{N}\nu_{k}^{2}Z_{k}(T)}.
\end{equation}
\section{Main results}\label{sec:mainResults}
In what follows, we will denote by $\theta$ the true parameter. For notational simplicity, the $T$ variable in $\xi_{k}(T)$, $X_{k}(T)$, $Y_{k}(T)$, $Z_{k}(T)$ and $A_{k}(T)$ will be omitted from now on. A straightforward algebraic computation leads to
\begin{align}\label{eq:thetaDif}
\widetilde{\theta}_{N}-\theta=-\frac{\sum_{k=1}^{N}\nu_{k}\left(\frac{1}{2}\xi_{k}^{2}-u_{k}^{2}(0)Y_{k}-\sigma^{2}\lambda_{k}^{-2\gamma}X_{k}+2\mu_{k}(\theta)Z_{k}\right)}{2\sum_{k=1}^{N}\nu_{k}^{2}Z_{k}}=:-\frac{\sum_{k=1}^{N}\nu_{k}A_{k}}{2\sum_{k=1}^{N}\nu_{k}^{2}Z_{k}}.
\end{align}
As usual, for two sequences of positive numbers $\{a_{n}\}_{n\in\mathbb{N}}$ and $\{b_{n}\}_{n\in\mathbb{N}}$, we will write $a_{n}\sim b_{n}$ if $\lim_{n\rightarrow\infty}a_{n}/b_{n}=1$, and will write $a_{n}\asymp b_{n}$ if there exist universal constants $K_{2}>K_{1}>0$, such that $K_{1}b_{n}\leq a_{n}\leq K_{2}b_{n}$ for $n\in\mathbb{N}$ large enough.
We start with a technical result regarding the leading order terms of the means and variances of $A_{k}$ and $Z_{k}$, as $k\to\infty$. The proof will be deferred to the Appendix.
\begin{proposition}\label{prop:asymp}
Let the assumptions \emph{(i) - (v)} be satisfied. Then, as $k\rightarrow\infty$,
\begin{align}\label{eq:ezk}
\mathbb{E}(Z_{k})&\asymp\frac{1}{\mu_{k}^{2}(\theta)}\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right)^{2},\\
\label{eq:varzk} \emph{Var}(Z_{k})&\asymp\frac{\lambda_{k}^{-2\gamma}}{\mu_{k}^{5}(\theta)}\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right)^{3},\\
\label{eq:eak} \mathbb{E}(A_{k})&\asymp\frac{\lambda_{k}^{-2\gamma}}{\mu_{k}^{2}(\theta)}\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right),\\
\label{eq:varak}
\emph{Var}(A_{k})&\asymp\frac{\lambda_{k}^{-2\gamma}}{\mu_{k}^{3}(\theta)}\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right)^{3}.
\end{align}
\end{proposition}
\begin{remark}\label{rem:asymp}
In the above proposition, the exact asymptotic behavior of the means and variances of $Z_{k}$ and $A_{k}$, as $k\rightarrow\infty$, can be obtained as follows:
\begin{align}\label{eq:ezkExact} \mathbb{E}(Z_{k})&\sim\frac{u_{k}^{4}(0)T}{4\mu_{k}^{2}(\theta)}+\frac{\sigma^{2}\lambda_{k}^{-2\gamma}u_{k}^{2}(0)T^{2}}{4\mu_{k}^{2}(\theta)}+\frac{\sigma^{4}\lambda_{k}^{-4\gamma}T^{3}}{12\mu_{k}^{2}(\theta)},\\
\label{eq:varzkExact} \text{Var}(Z_{k})&\sim\frac{\sigma^{2}\lambda_{k}^{-2\gamma}u_{k}^{6}(0)T^{2}}{2\mu_{k}^{5}(\theta)}+\frac{2\sigma^{4}\lambda_{k}^{-4\gamma}u_{k}^{4}(0)T^{3}}{3\mu_{k}^{5}(\theta)}+\frac{\sigma^{6}\lambda_{k}^{-6\gamma}u_{k}^{2}(0)T^{4}}{3\mu_{k}^{5}(\theta)}+\frac{\sigma^{8}\lambda_{k}^{-8\gamma}T^{5}}{15\mu_{k}^{5}(\theta)},\\
\label{eq:eakExact} \mathbb{E}(A_{k})&\sim\frac{\sigma^{2}\lambda_{k}^{-2\gamma}u_{k}^{2}(0)T}{\mu_{k}^{2}(\theta)}+\frac{\sigma^{4}\lambda_{k}^{-4\gamma}T^{2}}{2\mu_{k}^{2}(\theta)},\\
\label{eq:varakExact} \text{Var}(A_{k})&\sim\frac{\sigma^{2}\lambda_{k}^{-2\gamma}u_{k}^{6}(0)T^{2}}{2\mu_{k}^{3}(\theta)}+\frac{2\sigma^{4}\lambda_{k}^{-4\gamma}u_{k}^{4}(0)T^{3}}{3\mu_{k}^{3}(\theta)}+\frac{\sigma^{6}\lambda_{k}^{-6\gamma}u_{k}^{2}(0)T^{4}}{3\mu_{k}^{3}(\theta)}+\frac{\sigma^{8}\lambda_{k}^{-8\gamma}T^{5}}{15\mu_{k}^{3}(\theta)}.
\end{align}
These formulas will be used to obtain the exact asymptotic bias and the exact rate of convergence in the proof of asymptotic normality.
\end{remark}
\subsection{The Consistency of TFE}
In this subsection we prove the large-space consistency of the TFE $\widetilde{\theta}_{N}$, as $N\rightarrow\infty$. The proof relies on the classical version of the Strong Law of Large Numbers (cf. \cite[Theorem IV.3.2]{ShiryaevBookProbability}), which we state in the Appendix for sake of completeness. With this at hand, we now present the first main result of this paper.
\begin{theorem}[Consistency of TFE]\label{thm:Consistency}
Let the assumptions \emph{(i) - (v)} be satisfied. Moreover, assume that
\begin{align}\label{eq:zk}
\sum_{k=1}^{\infty}\nu_{k}^{2}\, \mathbb{E}(Z_{k})=\infty.
\end{align}
Then,
\begin{align}
\lim_{N\rightarrow\infty}\widetilde{\theta}_{N}=\theta,\quad\mathbb{P}-\text{a.}\,\text{s.}.
\end{align}
\end{theorem}
\begin{proof}
By \eqref{eq:thetaDif},
\begin{align}\label{eq:thetaDif2}
\widetilde{\theta}_{N}-\theta=-\frac{\sum_{k=1}^{N}\nu_{k}A_{k}}{2\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}\cdot\frac{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}{\sum_{k=1}^{N}\nu_{k}^{2}Z_{k}}.
\end{align}
We first study the second factor in \eqref{eq:thetaDif2}. Clearly,
\begin{align}
\sum_{N=1}^{\infty}\frac{\nu_{N}^{4}\,\text{Var}(Z_{N})}{\left(\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})\right)^{2}}&\leq\frac{\text{Var}(Z_{1})}{\left(\mathbb{E}(Z_{1})\right)^{2}}+\sum_{N=2}^{\infty}\frac{\nu_{N}^{4}\,\text{Var}(Z_{N})}{\sum_{k=1}^{N-1}\nu_{k}^{2}\,\mathbb{E}(Z_{k})\cdot\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}Z_{k}}\\
\label{eq:DecompVarZNSqSumExpZk} &=\frac{\text{Var}(Z_{1})}{\left(\mathbb{E}(Z_{1})\right)^{2}}\!+\!\!\sum_{N=2}^{\infty}\!\frac{\nu_{N}^{2}\text{Var}(Z_{N})}{\mathbb{E}(Z_{N})}\!\left(\!\frac{1}{\sum_{k=1}^{N-1}\!\nu_{k}^{2}\mathbb{E}(Z_{k})}\!-\!\frac{1}{\sum_{k=1}^{N}\!\nu_{k}^{2}\mathbb{E}(Z_{k})}\!\right).
\end{align}
By \eqref{eq:ezk}, \eqref{eq:varzk} and the assumption (iii), as $N\rightarrow\infty$,
\begin{align}
\frac{\nu_{N}^{2}\,\text{Var}(Z_{N})}{\mathbb{E}(Z_{N})}=O\left(\nu_{N}^{2}\cdot\frac{\frac{\lambda_{N}^{-2\gamma}}{\mu_{N}^{5}(\theta)}\left(u_{N}^{2}(0)+\sigma^{2}T\lambda_{N}^{-2\gamma}\right)^{3}}{\frac{1}{\mu_{N}^{2}(\theta)}\left(u_{N}^{2}(0)+\sigma^{2}T\lambda_{N}^{-2\gamma}\right)^{2}}\right)=O\left(\frac{\lambda_{N}^{-2\gamma}}{\mu_{N}(\theta)}\left(u_{N}^{2}(0)+\sigma^{2}T\lambda_{N}^{-2\gamma}\right)\right).
\end{align}
Since $u_{0}\in H$, we have
\begin{align}\label{eq:LimitSquN}
\lim_{N\rightarrow\infty}u_{N}^{2}(0)=0.
\end{align}
Together with the assumptions (ii), (iv) and (v),
\begin{align}
\lim_{N\rightarrow\infty}\frac{\lambda_{N}^{-2\gamma}}{\mu_{N}(\theta)}\left(u_{N}^{2}(0)+\sigma^{2}T\lambda_{N}^{-2\gamma}\right)=\lim_{N\rightarrow\infty}\frac{1}{\mu_{N}(\theta)\nu_{N}^{\gamma/m}}\left(u_{N}^{2}(0)+\frac{\sigma^{2}T}{\nu_{N}^{\gamma/m}}\right)=0.
\end{align}
Hence, there exists a universal constant $C_{1}>0$ such that
\begin{align}
\frac{\nu_{N}^{2}\,\text{Var}(Z_{N})}{\mathbb{E}(Z_{N})}\leq C_{1}\quad\text{for all }\,N\in\mathbb{N},
\end{align}
which, together with \eqref{eq:DecompVarZNSqSumExpZk}, implies that
\begin{align}
\sum_{N=1}^{\infty}\frac{\nu_{N}^{4}\,\text{Var}(Z_{N})}{\left(\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})\right)^{2}}&\leq\frac{\text{Var}(Z_{1})}{\left(\mathbb{E}(Z_{1})\right)^{2}}+C_{1}\sum_{N=2}^{\infty}\left(\frac{1}{\sum_{k=1}^{N-1}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}-\frac{1}{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}\right)\\
\label{eq:EstVarZNSqSumExpZk} &=\frac{\text{Var}(Z_{1})}{\left(\mathbb{E}(Z_{1})\right)^{2}}+\frac{C_{1}}{\nu_{1}^{2}\,\mathbb{E}(Z_{1})}<\infty.
\end{align}
Combining \eqref{eq:zk} with \eqref{eq:EstVarZNSqSumExpZk}, we conclude by Remark \ref{rem:SLLN} that
\begin{align}\label{eq:Consist1}
\lim_{N\rightarrow\infty}\frac{\sum_{k=1}^{N}\nu_{k}^{2}Z_{k}}{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}=1,\quad\mathbb{P}-\text{a.}\,\text{s.}.
\end{align}
Next, we will analyze the asymptotic behavior of the first factor in \eqref{eq:thetaDif2}. By \eqref{eq:ezk}, \eqref{eq:varak}, \eqref{eq:LimitSquN} and the assumptions (ii), (iv) and (v), as $N\rightarrow\infty$, we get that
\begin{align}
\frac{\text{Var}(A_{N})}{\mathbb{E}(Z_{N})}=O\left(\frac{\frac{\lambda_{N}^{-2\gamma}}{\mu_{N}^{3}(\theta)}\left(u_{N}^{2}(0)+\sigma^{2}T\lambda_{N}^{-2\gamma}\right)^{3}}{\frac{1}{\mu_{N}^{2}(\theta)}\left(u_{N}^{2}(0)+\sigma^{2}T\lambda_{N}^{-2\gamma}\right)^{2}}\right)=O\left(\frac{1}{\mu_{N}(\theta)\nu_{N}^{\gamma/m}}\left(u_{N}^{2}(0)+\frac{\sigma^{2}T}{\nu_{N}^{\gamma/m}}\right)\right)\rightarrow 0.
\end{align}
An argument similar to that of \eqref{eq:DecompVarZNSqSumExpZk} and \eqref{eq:EstVarZNSqSumExpZk} shows that, there exists a universal constant $C_{2}>0$, such that
\begin{align}
\sum_{N=1}^{\infty}\frac{\nu_{N}^{2}\text{Var}(A_{N})}{\left(\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})\right)^{2}}&\leq\frac{\text{Var}(A_{1})}{\nu_{1}^{2}\left(\mathbb{E}(Z_{1})\right)^{2}}+\sum_{N=2}^{\infty}\frac{\text{Var}(A_{N})}{\mathbb{E}(Z_{N})}\left(\frac{1}{\sum_{k=1}^{N-1}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}-\frac{1}{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}\right)\\
\label{eq:EstVarANSqSumExpZk} &\leq\frac{\text{Var}(A_{1})}{\nu_{1}^{2}\left(\mathbb{E}(Z_{1})\right)^{2}}+\frac{C_{2}}{\nu_{1}^{2}\,\mathbb{E}(Z_{1})}<\infty.
\end{align}
In view of Theorem \ref{thm:SLLN}, \eqref{eq:zk} and \eqref{eq:EstVarANSqSumExpZk} imply that
\begin{align}
\lim_{N\rightarrow\infty}\frac{\sum_{k=1}^{N}\nu_{k}\left(A_{k}-\mathbb{E}(A_{k})\right)}{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}=0,\quad\mathbb{P}-\text{a.}\,\text{s.}.
\end{align}
If the series $\sum_{k=1}^{N}\nu_{k}\,\mathbb{E}(A_{k})$ converges, then by \eqref{eq:zk}, we clearly have that
\begin{align}\label{eq:LimitSumExpAkSumExpZk}
\lim_{N\rightarrow\infty}\frac{\sum_{k=1}^{N}\nu_{k}\,\mathbb{E}(A_{k})}{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}=0,
\end{align}
On the other hand, if the series in the numerator of \eqref{eq:LimitSumExpAkSumExpZk} diverges, then by Stolz--Ces\`{a}ro Theorem
\begin{align}
\lim_{N\rightarrow\infty}\frac{\sum_{k=1}^{N}\nu_{k}\,\mathbb{E}(A_{k})}{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}=\lim_{N\rightarrow\infty}\frac{\nu_{N}\,\mathbb{E}(A_{N})}{\nu_{N}^{2}\,\mathbb{E}(Z_{N})},
\end{align}
and by \eqref{eq:ezk}, \eqref{eq:eak} and the assumption (iv), as $N\rightarrow\infty$, we deduce that
\begin{align}
\frac{\mathbb{E}(A_{N})}{\nu_{N}\,\mathbb{E}(Z_{N})}=O\left(\frac{\frac{\lambda_{N}^{-2\gamma}}{\mu_{N}^{2}(\theta)}\left(u_{N}^{2}(0)+\sigma^{2}T\lambda_{N}^{-2\gamma}\right)}{\frac{\nu_{N}}{\mu_{N}^{2}(\theta)}\left(u_{N}^{2}(0)+\sigma^{2}T\lambda_{N}^{-2\gamma}\right)^{2}}\right)=O\left(\frac{1}{\nu_{N}\left(u_{N}^{2}(0)\lambda_{N}^{2\gamma}+\sigma^{2}T\right)}\right)=O\left(\nu_{N}^{-1}\right)\rightarrow 0.
\end{align}
Combining the above, we conclude that
\begin{align}\label{eq:Consist2}
\lim_{N\rightarrow\infty}\frac{\sum_{k=1}^{N}\nu_{k}A_{k}}{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}=0,\quad\mathbb{P}-\text{a.}\,\text{s.}.
\end{align}
Finally, by \eqref{eq:thetaDif2}, \eqref{eq:Consist1} and \eqref{eq:Consist2} we conclude the proof.
\end{proof}
\begin{remark} A note on condition \eqref{eq:zk} is in order. The divergence of the series \eqref{eq:zk} is needed for the Law of Large Numbers to hold true. In view of \eqref{eq:ezk}, the condition \eqref{eq:zk} can be equivalently stated in terms of the known primary objects -- the initial data $u(0)$, the asymptotics of the eigenvalues of $\mathcal{A}_{0}$ and $\mathcal{A}_{1}$, and the covariance structure of the noise (see Proposition \ref{prop:EquiDivConds} below). In particular, the consistency of the TFE does not depend on the regularity of the initial data.
\end{remark}
\subsection{The Asymptotic Normality of TFE}
In this subsection, we will investigate the asymptotic normality of the TFE $\widetilde{\theta}_{N}$. The proof is based on classical Central Limit Theorem (CLT) for independent random variables with the Lyapunov condition (which is a sufficient condition for the Lindeberg condition to hold). For convenience, we list this result in the Appendix, and the complete proof can be found, for instance, in~\cite[Section III.4]{ShiryaevBookProbability}.
In what follows we will make use of the following technical lemma; the proof is deferred to the Appendix.
\begin{lemma}\label{lemma:MomentEstxi}
Let $\xi_{k}(t)$, $k\in\mathbb{N}, \ t\in[0,T]$, be defined as in \eqref{eq:notation}. Then, for any $n\in\mathbb{N}$, there exist a constant $D_{n}=D_{n}(t)>0$, depending only on $n$ and $t$, such that, for every $k\in\mathbb{N}$,
\begin{align}
\mathbb{E}\left(\xi_{k}^{n}(t)\right)\leq D_{n}\left(\frac{u_{k}^{2}(0)+\sigma^{2}t\,\lambda_{k}^{-2\gamma}}{\mu_{k}(\theta)}\right)^{n},\quad t\in[0,T].
\end{align}
\end{lemma}
Now we present a version of the large-space asymptotic normality of the TFE $\widetilde{\theta}_{N}$.
\begin{theorem}[Asymptotic Normality of TFE]\label{thm:AsymNormal}
In addition to the conditions of Theorem \ref{thm:Consistency}, assume further that
\begin{align}\label{eq:SumVarAk}
\sum_{k=1}^{\infty}\nu_{k}^{2}\,\emph{Var}\left(A_{k}\right)=\infty.
\end{align}
Then, as $N\rightarrow\infty$,
\begin{align}
\frac{\widetilde{\theta}_{N}-\theta+a_{N}}{b_{N}}\;{\stackrel{d}{\longrightarrow}}\;\mathcal{N}(0,1),
\end{align}
where
\begin{align}
a_{N}:=\frac{\sum_{k=1}^{N}\nu_{k}\,\mathbb{E}(A_{k})}{2\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})},\quad b_{N}:=\frac{\sqrt{\sum_{k=1}^{N}\nu_{k}^{2}\,\emph{Var}(A_{k})}}{2\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})},
\end{align}
and where $\;{\stackrel{d}{\longrightarrow}}\;$ denotes the convergence in distribution.
\end{theorem}
\begin{proof} The proof is split in two steps.
\medskip
\noindent
\textit{Step 1.} We will first show that the sequence $\set{\nu_{k}A_{k}}_{k\in\mathbb{N}}$ satisfies the Lyapunov condition \eqref{eq:Lyapunov} with $\delta=2$. Clearly,
\begin{align}
\mathbb{E}\left(\left(A_{k}-\mathbb{E}(A_{k})\right)^{4}\right)&=\mathbb{E}\left(A_{k}^{4}\right)-4\,\mathbb{E}\left(A_{k}^{3}\right)\mathbb{E}(A_{k})+6\,\mathbb{E}\left(A_{k}^{2}\right)(\mathbb{E}(A_{k}))^{2}-3\left(\mathbb{E}(A_{k})\right)^{4}\\
&=\mathbb{E}\left(A_{k}^{4}\right)-4\,\mathbb{E}\left(A_{k}^{3}\right)\mathbb{E}(A_{k})+6\,\text{Var}(A_{k})\left(\mathbb{E}(A_{k})\right)^{2}+3\left(\mathbb{E}(A_{k})\right)^{4}.
\label{eq:fourthPower}
\end{align}
We will estimate each term in the above expression separately. To begin with, for every $k\in\mathbb{N}$, let $\zeta_{k}:=(\zeta_{k}(t))_{t\in[0,T]}$, where
\begin{align}
\zeta_{k}(t):=\int_{0}^{t}u_{k}(s)\,dw_{k}(s),\quad t\in[0,T].
\end{align}
By \eqref{eq:uk2} and \eqref{eq:notation}, for any $k\in\mathbb{N}$ and $t\in[0,T]$,
\begin{align}
u_{k}^{2}(t)=u_{k}^{2}(0)+\sigma^{2}\lambda_{k}^{-2\gamma}t-2\mu_{k}(\theta)\xi_{k}(t)+2\sigma\lambda_{k}^{-\gamma}\zeta_{k}(t),
\end{align}
which, when multiplied by $\xi_{k}(t)$, and then integrated on $[0,T]$, leads to
\begin{align}
A_{k}=2\sigma\lambda_{k}^{-\gamma}\int_{0}^{T}\zeta_{k}(t)\xi_{k}(t)\,dt,\quad k\in\mathbb{N}.
\end{align}
Hence, by the Cauchy--Schwartz inequality and the definition of $\xi_{k}$, for any $k\in\mathbb{N}$,
\begin{align}
\mathbb{E}\left(A_{k}^{4}\right)&\leq 16\sigma^{4}\lambda_{k}^{-4\gamma}\,\mathbb{E}\left(\left(\int_{0}^{T}\zeta_{k}^{2}(t)\,dt\cdot\int_{0}^{T}\xi_{k}^{2}(t)\,dt\right)^{2}\right)\\
&\leq 16\sigma^{4}\lambda_{k}^{-4\gamma}\left(\mathbb{E}\left(\left(\int_{0}^{T}\zeta_{k}^{2}(t)dt\right)^{4}\right)\mathbb{E}\left(\left(\int_{0}^{T}\xi_{k}^{2}(t)dt\right)^{4}\right)\right)^{1/2}\\
&\leq 16\sigma^{4}\lambda_{k}^{-4\gamma}\left(T^{2}\,\mathbb{E}\left(\left(\int_{0}^{T}\zeta_{k}^{4}(t)\,dt\right)^{2}\right)\cdot T^{4}\,\mathbb{E}\left(\xi_{k}^{8}\right)\right)^{1/2}\\
&\leq 16\sigma^{4}T^{7/2}\lambda_{k}^{-4\gamma}\left(\mathbb{E}\left(\int_{0}^{T}\zeta_{k}^{8}(t)\,dt\right)\mathbb{E}\left(\xi_{k}^{8}\right)\right)^{1/2}.
\end{align}
Moreover, by the Burkholder--Davis--Gundy inequality, there exists a constant $C_{1}=C_{1}(T)>0$, depending only on $T$, such that
\begin{align}
\mathbb{E}\left(\sup_{t\in[0,T]}\zeta_{k}^{8}(t)\right)\leq C_{1}\,\mathbb{E}\left(\left[\zeta_{k},\zeta_{k}\right]^{4}(T)\right)=C_{1}\,\mathbb{E}\left(\xi_{k}^{4}\right).
\end{align}
Together with Lemma \ref{lemma:MomentEstxi}, we obtain that, for any $k\in\mathbb{N}$,
\begin{align}\label{eq:EstAk4Moment}
\mathbb{E}\left(A_{k}^{4}\right)\leq 16\,C_{1}\sigma^{4}\, T^{4}\lambda_{k}^{-4\gamma}\sqrt{\mathbb{E}\left(\xi_{k}^{4}\right)\mathbb{E}\left(\xi_{k}^{8}\right)}\leq C_{2}\,\frac{\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right)^{8}}{\mu_{k}^{6}(\theta)},
\end{align}
where $C_{2}:=16\,C_{1}\sqrt{D_{4}D_{8}}T^{2}>0$ is a constant depending only on $T$.
Next, we will study the last three terms of \eqref{eq:fourthPower}. In view of \eqref{eq:eak} and \eqref{eq:varak}, there exists a universal constant $C_{3}>0$, such that for any $k\in\mathbb{N}$,
\begin{align}\label{eq:EstAkVarAk}
\mathbb{E}(A_{k})\leq C_{3}\,\frac{\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right)^{2}}{\mu_{k}^{2}(\theta)},\quad\text{Var}(A_{k})\leq C_{2}\,\frac{\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right)^{4}}{\mu_{k}^{3}(\theta)}.
\end{align}
Hence, it suffices to estimate $\mathbb{E}(A_{k}^{3})$. By the definition of $A_{k}$ in \eqref{eq:thetaDif},
\begin{equation}
-u_{k}^{2}(0)Y_{k}-\sigma^{2}\lambda_{k}^{-2\gamma}X_{k}\leq A_{k}\leq\frac{1}{2}\xi_{k}^{2}+2\mu_{k}(\theta)Z_{k}.
\end{equation}
Moreover, since $\xi_{k}(t)$ is increasing in $t$, we deduce that
\begin{align}
Y_{k}=\int_{0}^{T}\xi_{k}(t)\,dt\leq T\xi_{k},\quad X_{k}=\int_{0}^{T}t\xi_{k}(t)\,dt\leq T^{2}\xi_{k},\quad Z_{k}=\int_{0}^{T}\xi_{k}^{2}(t)\,dt\leq T\xi_{k}^{2},
\end{align}
and thus,
\begin{align}
-T\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right)\xi_{k}\leq A_{k}\leq\left(\frac{1}{2}+2\mu_{k}(\theta)T\right)\xi_{k}^{2}.
\end{align}
Together with Lemma \ref{lemma:MomentEstxi}, we obtain that
\begin{align}\label{eq:EstAk3Moment}
- D_{3}T\frac{\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right)^{6}}{\mu_{k}^{3}(\theta)}\leq\mathbb{E}(A_{k}^{3})\leq D_{6}(2T+1)\frac{\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right)^{6}}{\mu_{k}^{3}(\theta)}.
\end{align}
Combining \eqref{eq:fourthPower}--\eqref{eq:EstAk3Moment}, we conclude that there exists a constant $C_{4}=C_{4}(T)>0$, depending only on $T$, such that for any $k\in\mathbb{N}$,
\begin{align}
\mathbb{E}\left(\left(\nu_{k}A_{k}-\mathbb{E}(\nu_{k}A_{k})\right)^{4}\right)\leq C_{4}\,\frac{\nu_{k}^{4}}{\mu_{k}^{5}(\theta)}\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right)^{8}.
\end{align}
On the other hand, by \eqref{eq:varak} again, we can find a universal constant $C_{5}>0$, such that
\begin{align}
\text{Var}\left(\nu_{k}A_{k}\right)\geq C_{5}\frac{\nu_{k}^{2}}{\mu_{k}^{3}(\theta)}\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right)^{4},
\quad \text{for all }\,k\in\mathbb{N}.
\end{align}
In view of the assumptions (ii)-(v), and since $\lim_{k\rightarrow\infty}u_{k}^{2}(0)=0$, we deduce that there exists a constant $C_{6}=C_{6}(c_{0},\sigma,T)>0$, depending on $c_{0}$, $\sigma$ and $T$, such that
\begin{align}
\mathbb{E}\left(\left(\nu_{k}A_{k}-\mathbb{E}(\nu_{k}A_{k})\right)^{4}\right)\leq C_{6}\text{Var}\left(\nu_{k}A_{k}\right),\quad\text{for all }\,k\in\mathbb{N}.
\end{align}
Finally, by \eqref{eq:SumVarAk}, we obtain that
\begin{align}
\lim_{N\rightarrow\infty}\frac{\sum_{k=1}^{N}\mathbb{E}\left(\left(\nu_{k}A_{k}-\mathbb{E}(\nu_{k}A_{k})\right)^{4}\right)}{\left(\sum_{k=1}^{N}\text{Var}\left(\nu_{k}A_{k}\right)\right)^{2}}\leq\lim_{N\rightarrow\infty}\frac{C_{6}}{\sum_{k=1}^{N}\text{Var}\left(\nu_{k}A_{k}\right)}=0.
\end{align}
\smallskip
\noindent
\textit{Step 2:} Note that
\begin{align}\label{eq:expandofAll}
\frac{\widetilde{\theta}_{N}-\theta+a_{N}}{b_{N}}=-\frac{\sum_{k=1}^{N}\nu_{k}\left(A_{k}-\mathbb{E}(A_{k})\right)}{2\,b_{N}\sum_{k=1}^{N}\nu_{k}^{2}Z_{k}}-\frac{\sum_{k=1}^{N}\nu_{k}\,\mathbb{E}(A_{k})}{2\,b_{N}\sum_{k=1}^{N}\nu_{k}^{2}Z_{k}}+\frac{a_{N}}{b_{N}}.
\end{align}
For the first term in \eqref{eq:expandofAll}, by \eqref{eq:Consist1}, Step 1 and Theorem \ref{thm:LyapunovCLT}, as $N\rightarrow\infty$, we get
\begin{align}\label{eq:CLTLyapunov}
-\frac{\sum_{k=1}^{N}\nu_{k}\left(A_{k}-\mathbb{E}(A_{k})\right)}{2\,b_{N}\sum_{k=1}^{N}\nu_{k}^{2}Z_{k}}=-\frac{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}{\sum_{k=1}^{N}\nu_{k}^{2}Z_{k}}\cdot\frac{\sum_{k=1}^{N}\nu_{k}\left(A_{k}-\mathbb{E}(A_{k})\right)}{\sqrt{\sum_{k=1}^{N}\nu_{k}^{2}\,\text{Var}(A_{k})}}\;{\stackrel{d}{\longrightarrow}}\;\mathcal{N}(0,1).
\end{align}
Moreover, for the last two terms in \eqref{eq:expandofAll}, we derive that
\begin{align}
\frac{a_{N}}{b_{N}}-\frac{\sum_{k=1}^{N}\nu_{k}\,\mathbb{E}(A_{k})}{2\,b_{N}\sum_{k=1}^{N}\nu_{k}^{2}Z_{k}}&=\frac{2\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}{\sqrt{\sum_{k=1}^{N}\nu_{k}^{2}\,\text{Var}(A_{k})}}\left(\frac{\sum_{k=1}^{N}\nu_{k}\,\mathbb{E}(A_{k})}{2\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}-\frac{\sum_{k=1}^{N}\nu_{k}\,\mathbb{E}(A_{k})}{2\sum_{k=1}^{N}\nu_{k}^{2}Z_{k}}\right)\\
\label{eq:expandOfTail} &=\frac{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}{\sum_{k=1}^{N}\nu_{k}^{2}Z_{k}}\cdot\frac{\sum_{k=1}^{N}\nu_{k}\,\mathbb{E}(A_{k})}{\sqrt{\sum_{k=1}^{N}\nu_{k}^{2}\,\text{Var}(A_{k})}}\cdot\frac{\sum_{k=1}^{N}\nu_{k}^{2}\left(Z_{k}-\mathbb{E}(Z_{k})\right)}{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}.
\end{align}
In light of \eqref{eq:Consist1}, we only need to show that the product of the last two factors above converges to zero in probability, as $N\rightarrow\infty$. Note that, by the independence of $Z_{k}$, $k\in\mathbb{N}$,
\begin{align}
\mathbb{E}\!\left(\!\left(\frac{\sum_{k=1}^{N}\nu_{k}\,\mathbb{E}(A_{k})}{\sqrt{\sum_{k=1}^{N}\nu_{k}^{2}\,\text{Var}(A_{k})}}\cdot\frac{\sum_{k=1}^{N}\nu_{k}^{2}\left(Z_{k}-\mathbb{E}(Z_{k})\right)}{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}\right)^{2}\right)=\left(\frac{\sum_{k=1}^{N}\nu_{k}\,\mathbb{E}(A_{k})}{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}\right)^{2}\frac{\sum_{k=1}^{N}\nu_{k}^{4}\,\text{Var}(Z_{k})}{\sum_{k=1}^{N}\nu_{k}^{2}\,\text{Var}(A_{k})}.
\end{align}
By \eqref{eq:varzk}, \eqref{eq:varak} and the assumption (iii), there exists a universal constant $C_{7}>0$, such that
\begin{align}
\frac{\sum_{k=1}^{N}\nu_{k}^{4}\,\text{Var}(Z_{k})}{\sum_{k=1}^{N}\nu_{k}^{2}\,\text{Var}(A_{k})}&\leq C_{7}\frac{\sum_{k=1}^{N}\frac{\nu_{k}^{4}\lambda_{k}^{-2\gamma}}{\mu_{k}^{5}(\theta)}(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma})^3}{\sum_{k=1}^{N}\frac{\nu_{k}^{2}\lambda_{k}^{-2\gamma}}{\mu_{k}^{3}(\theta)}(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma})^3}\leq C_{7}c_{0},
\end{align}
Similarly, by \eqref{eq:ezk}, \eqref{eq:eak} and the assumption (iii),
\begin{align}\label{eq:RatioSumAkSumZk}
\frac{\sum_{k=1}^{N}\nu_{k}\,\mathbb{E}(A_{k})}{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}&\leq C_{8}\frac{\sum_{k=1}^{N}\frac{\nu_{k}\lambda_{k}^{-2\gamma}}{\mu_{k}^{2}(\theta)}(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma})}{\sum_{k=1}^{N}\frac{\nu_{k}^{2}}{\mu_{k}^{2}(\theta)}(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma})^{2}},
\end{align}
where $C_{8}>0$ is some universal constant. Using \eqref{eq:ezk} and \eqref{eq:zk}, we conclude that the series in the denominator on the right-hand side of \eqref{eq:RatioSumAkSumZk} diverges. Hence, the right-hand side of \eqref{eq:RatioSumAkSumZk} converges to 0, as $N\rightarrow\infty$, if the series in the numerator on the right-hand side of \eqref{eq:RatioSumAkSumZk} converges. Now assume that the numerator on the right-hand side of \eqref{eq:RatioSumAkSumZk} diverges. By Stolz--Ces\`{a}ro Theorem,
\begin{align}
\lim_{N\rightarrow\infty}\frac{\sum_{k=1}^{N}\frac{\nu_{k}\lambda_{k}^{-2\gamma}}{\mu_{k}^{2}(\theta)}\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right)}{\sum_{k=1}^{N}\frac{\nu_{k}^{2}}{\mu_{k}^{2}(\theta)}\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right)^{2}}=\lim_{N\rightarrow\infty}\frac{\frac{\nu_{N}\lambda_{N}^{-2\gamma}}{\mu_{N}^{2}(\theta)}\left(u_{N}^{2}(0)+\sigma^{2}T\lambda_{N}^{-2\gamma}\right)}{\frac{\nu_{N}^{2}}{\mu_{N}^{2}(\theta)}\left(u_{N}^{2}(0)+\sigma^{2}T\lambda_{N}^{-2\gamma}\right)^{2}}\leq\lim_{N\rightarrow\infty}\frac1{\sigma^{2}T\nu_{N}}=0.
\end{align}
Therefore, we have shown that
\begin{align}\label{eq:L2CovCLT}
\frac{\sum_{k=1}^{N}\nu_{k}\,\mathbb{E}(A_{k})}{\sqrt{\sum_{k=1}^{N}\nu_{k}^{2}\,\text{Var}(A_{k})}}\cdot\frac{\sum_{k=1}^{N}\nu_{k}^{2}\left(Z_{k}-\mathbb{E}(Z_{k})\right)}{\sum_{k=1}^{N}\nu_{k}^{2}\,\mathbb{E}(Z_{k})}\rightarrow 0\quad\text{in }\,L^{2}(\Omega),\quad N\rightarrow\infty.
\end{align}
Combining \eqref{eq:Consist1}, \eqref{eq:expandofAll}, \eqref{eq:expandOfTail} and \eqref{eq:L2CovCLT} completes the proof.
\end{proof}
The next result provides some equivalent conditions, given explicitly in terms of the model coefficients, for \eqref{eq:zk} and \eqref{eq:SumVarAk} to hold. In particular, we note that the consistency and the asymptotic normality of the TFE do not depend on the regularity of the initial data.
\begin{proposition}\label{prop:EquiDivConds}
Under the assumptions \emph{(i) - (v)},
\begin{align}\label{eq:EquiCondExpZk} \sum_{k=1}^{\infty}\nu_{k}^{2}\,\mathbb{E}(Z_{k})=\infty\quad&\Leftrightarrow\quad\sum_{k=}^{\infty}\frac{\nu_{k}^{2}\lambda_{k}^{-4\gamma}}{\mu_{k}^{2}(\theta)}=\infty,\\
\label{eq:EquiCondVarAk}
\sum_{k=1}^{\infty}\nu_{k}^{2}\,\emph{Var}(A_{k})=\infty\quad&\Leftrightarrow\quad\sum_{k=1}^{\infty}\frac{\nu_{k}^{2}\lambda_{k}^{-8\gamma}}{\mu_{k}^{3}(\theta)}=\infty.
\end{align}
\end{proposition}
\begin{proof}
We will only present the proof for \eqref{eq:EquiCondVarAk}, as \eqref{eq:EquiCondExpZk} can be obtained similarly. Clearly, \eqref{eq:varak} implies the ``$\Leftarrow$" direction in \eqref{eq:EquiCondVarAk}. Now assume that
\begin{align}
\sum_{k=1}^{\infty}\nu_{k}^{2}\,\text{Var}(A_{k})=\infty,
\end{align}
which, by \eqref{eq:varak}, is equivalent to
\begin{align}
\infty=\sum_{k=1}^{\infty}\frac{\nu_{k}^{2}\lambda_{k}^{-2\gamma}}{\mu_{k}^{3}(\theta)}\left(u_{k}^{2}(0)+\sigma^{2}T\lambda_{k}^{-2\gamma}\right)^{3}&=\sum_{k=1}^{\infty}\frac{\nu_{k}^{2}\lambda_{k}^{-2\gamma}u_{k}^{6}(0)}{\mu_{k}^{3}(\theta)}+3\sigma^{2}T\sum_{k=1}^{\infty}\frac{\nu_{k}^{2}\lambda_{k}^{-4\gamma}u_{k}^{4}(0)}{\mu_{k}^{3}(\theta)}\\
&\quad\,\,+3\sigma^{4}T^{2}\sum_{k=1}^{\infty}\frac{\nu_{k}^{2}\lambda_{k}^{-6\gamma}u_{k}^{2}(0)}{\mu_{k}^{3}(\theta)}+\sigma^{6}T^{3}\sum_{k=1}^{\infty}\frac{\nu_{k}^{2}\lambda_{k}^{-8\gamma}}{\mu_{k}^{3}(\theta)}.
\end{align}
Hence, it suffices to show that the first three series on the right-hand side above are all convergent. We will only check the first series, and the other two can be verified using a similar argument. Indeed, by the assumptions (ii) - (v) and since $u(0)\in H$ (so that $\lim_{k\rightarrow\infty}u_{k}(0)=0$), there exists a universal constant $C>0$ such that
\begin{align}
\sum_{k=1}^{\infty}\frac{\nu_{k}^{2}\lambda_{k}^{-2\gamma}u_{k}^{6}(0)}{\mu_{k}^{3}(\theta)}\leq c_{0}^{2}\,C\sum_{k=1}^{\infty}u_{k}^{2}(0)<\infty.
\end{align}
This concludes the proof.
\end{proof}
We conclude this section by providing the asymptotically equivalent formulas for the sequences $\{a_{N}\}_{N\in\mathbb{N}}$ and $\{b_{N}\}_{N\in\mathbb{N}}$ in Theorem~\ref{thm:AsymNormal}, given in terms of the model coefficients. In light of Proposition~\ref{prop:EquiDivConds}, the relations \eqref{eq:zk} and \eqref{eq:SumVarAk} imply that each of the last terms in \eqref{eq:ezkExact}--\eqref{eq:varakExact} give the exact leading order term for $\mathbb{E}(Z_{k})$, $\text{Var}(Z_{k})$, $\mathbb{E}(A_{k})$ and $\text{Var}(A_{k})$, respectively. The following result follows immediately from Stolz--Ces\`{a}ro Theorem.
\begin{corollary}
Under the conditions of Theorem \ref{thm:AsymNormal}, as $N\rightarrow\infty$, we have
\begin{align}
a_{N}\sim \frac3{T}\frac{\sum_{k=1}^{N}\frac{\nu_{k}\lambda_{k}^{-4\gamma}}{\mu_{k}^{2}(\theta)}}{\sum_{k=1}^{N}\frac{\nu_{k}^{2}\lambda_{k}^{-4\gamma}}{\mu_{k}^{2}(\theta)}},\quad b_{N}\sim \sqrt{\frac{12}{5T}}\frac{\sqrt{\sum_{k=1}^{N}\frac{\nu_{k}^{2}\lambda_{k}^{-8\gamma}}{\mu_{k}^{3}(\theta)}}}{\sum_{k=1}^{N}\frac{\nu_{k}^{2}\lambda_{k}^{-4\gamma}}{\mu_{k}^{2}(\theta)}}.
\end{align}
\end{corollary}
\section{Examples}\label{sec:Examples}
In this section, we will present two examples that illustrate the theoretical results obtained in Section~\ref{sec:mainResults}. Throughout this section, let $G\subseteq\mathbb{R}^{d}$ be a smooth and bounded domain, $H:=L^{2}(G)$, and let $\Delta$ be the Laplace operator on $G$ with zero boundary condition. It is known (cf.~\cite{Shubin}) that $\Delta$ has a complete orthonormal system of eigenfunctions $\{h_{k}\}_{k\in\mathbb{N}}$ in $H$. Moreover, the corresponding eigenvalues $\{\tau_{k}\}_{k\in\mathbb{N}}$ can be arranged such that $0\leq -\tau_{1}\leq -\tau_{2}\leq\cdots$, and there exists a universal constant $c_{1}>0$ so that
\begin{align}
\lim_{k\rightarrow\infty}\left|\tau_{k}\right|\cdot k^{-2/d}=c_{1}.
\end{align}
\begin{example}\label{eg:FracStoHeatEq}
Consider the following fractional stochastic heat equation driven by an additive noise, possibly colored in space,
\begin{align}
du(t,x)+\theta(-\Delta)^{\beta}u(t,x)\,dt=\sigma\sum_{k=1}^{\infty}\lambda_{k}^{-\gamma}h_{k}(x)\,dw_{k}(t),\quad t\in[0,T],\quad x\in G,
\end{align}
with initial condition $u(0,x)=u_{0}(x)\in H$, where $\theta>0$, $\beta>0$, $\gamma\geq 0$ and $\sigma\in\mathbb{R}\setminus\{0\}$ are constants, and where $\lambda_{k}:=\sqrt{-\tau_{k}}$, $k\in\mathbb{N}$. In this case, $\rho_{k}=0$ for all $k\in\mathbb{N}$, and
\begin{align}
\nu_{k}\sim c_{1}k^{2\beta/d},\quad\mu_{k}(\theta)\sim c_{1}\theta\,k^{2\beta/d},\quad\lambda_{k}\sim\sqrt{c_{1}}\,k^{1/d},\quad k\rightarrow\infty.
\end{align}
Together with Proposition \ref{prop:EquiDivConds}, the conditions \eqref{eq:zk} and \eqref{eq:SumVarAk} are equivalent to
\begin{align}
\frac{1}{c_{1}^{2\gamma}\theta^{2}}\sum_{k=1}^{\infty}\frac{1}{k^{4\gamma/d}}
=\infty,
\quad\text{and}\quad\frac{1}{c_{1}^{1+4\gamma}\theta^{3}}\sum_{k=1}^{\infty}\frac{1}{k^{(2\beta+8\gamma)/d}}=\infty,
\end{align}
respectively. Therefore, the consistency and the asymptotic normality hold for the TFE $\tilde{\theta}_N$ given by \eqref{eq:thetaMain}, whenever
\begin{align}
2\beta+8\gamma\leq d.
\end{align}
\end{example}
\begin{example}\label{eg:DiagSPDE}
Let us consider the following SPDE, with the parameter of interest $\theta$ in front of lower order differential operator,
\begin{align}
du(t,x)+\left(\Delta u(t,x)+\theta u(t,x)\right)dt=\sum_{k=1}^{\infty}h_{k}(x)\,dw_{k}(t),\quad t\in[0,T],\quad x\in G,
\end{align}
with initial condition $u(0,x)=u_{0}(x)\in H$. In this case, $\gamma=0$, $\nu_{k}\equiv 1$ for all $k\in\mathbb{N}$, and
\begin{align}
\rho_{k}\sim c_{1}\,k^{2/d},\quad\mu_{k}(\theta)\sim\theta+c_{1}\,k^{2/d},\quad k\rightarrow\infty.
\end{align}
Together with Proposition \ref{prop:EquiDivConds}, the conditions \eqref{eq:zk} and \eqref{eq:SumVarAk} are equivalent to
\begin{align}
\sum_{k=1}^{\infty}\frac{1}{\left(\theta+c_{1}k^{2/d}\right)^{2}}=\infty\quad\text{and}\quad\sum_{k=1}^{\infty}\frac{1}{\left(\theta+c_{1}k^{2/d}\right)^{3}}=\infty,
\end{align}
respectively. Therefore, in order for the consistency and the asymptotic normality of $\tilde{\theta}_N$ to hold true, we need to have $d\geq 6$.
\end{example}
{\small
\section{Appendix}
\noindent
\textit{Proof of Theorem \ref{prop:asymp}.} Due to the nature of desired asymptotic results, the underlying computations are somehow extensive and tedious. For simplicity of brevity, we will only provide the proof for the special case when $u_{0}=0$, $\gamma=0$ and $\sigma=1$, but the genera case can be verified using similar arguments and the details can be obtained from the authors upon request. Most of the evaluations were performed using symbolic computations in Mathematica. For each $k\in\mathbb{N}$, when $u_{0}=0$, $\gamma=0$ and $\sigma=1$,
\begin{align}
u_{k}(t)=e^{-\mu_{k}(\theta)t}\int_{0}^{t}e^{\mu_{k}(\theta)s}\,dw_{k}(s),\quad t\in[0,T],
\end{align}
and with the notations introduced in \eqref{eq:notation} and \eqref{eq:thetaDif}, we get
\begin{align}
A_{k}=\frac{1}{2}\xi_{k}^{2}-X_{k}+2\mu_{k}(\theta)Z_{k}.
\end{align}
Note that for any $t\in[0,T]$, $u_{k}(t)$ is a centered normal random variable with variance $(1-e^{-2\mu_{k}(\theta)t})/(2\mu_{k}(\theta))$, and thus,
\begin{align}\label{eq:CenNormalMoments}
\mathbb{E}\left(u_{k}^{2n}(t)\right)=(2n-1)!!\cdot\left(\frac{1-e^{-2\mu_{k}(\theta)t}}{2\mu_{k}(\theta)}\right)^{n},
\quad n\in\mathbb{N}.
\end{align}
We first verify \eqref{eq:ezk}, which now reduces to
\begin{align}\label{eq:newezk}
\mathbb{E}(Z_{k})\asymp\frac{T^{2}}{\mu_{k}^{2}(\theta)},\quad k\rightarrow\infty,
\end{align}
by computing
\begin{align}\label{eq:DecompEzk}
\mathbb{E}(Z_{k})=\mathbb{E}\left(\int_{0}^{T}\xi_{k}^{2}(t)dt\right)=\int_{0}^{T}\mathbb{E}\left(\xi_{k}^{2}(t)\right)dt
=2\int_{0}^{T}\int_{0}^{t}\mathbb{E}\left(\xi_{k}(s)u_{k}^{2}(s)\right)ds\,dt,
\end{align}
where the last equality follows from the definition of $\xi_{k}$ in \eqref{eq:notation}. By It\^{o}'s formula,
\begin{align}
d\xi_{k}(t)u_{k}^{2}(t)= \left(u_{k}^{4}(t)+\xi_{k}(t)\right)dt-2\mu_{k}(\theta)\xi_{k}(t)u_{k}^{2}(t)\,dt+2u_{k}(t)\xi_{k}(t)dw_{k}(t).
\end{align}
Taking the expectations on both sides above, using the definition of $\xi_{k}(t)$ in \eqref{eq:notation} and \eqref{eq:CenNormalMoments}, we obtain that, for $t\in[0,T]$,
\begin{align}\label{eq:Exikuk2}
\mathbb{E}\left(\xi_{k}(t)u_{k}^{2}(t)\right)=\frac{1-e^{-2\mu_{k}(\theta)t}}{8\mu_{k}^{3}(\theta)}-\frac{5te^{-2\mu_{k}(\theta)t}}{4\mu_{k}^{2}(\theta)}+\frac{3\left(1-e^{-2\mu_{k}(\theta)t}\right)e^{-2\mu_{k}(\theta)t}}{8\mu_{k}^{3}(\theta)}+\frac{t}{4\mu_{k}^{2}(\theta)}.
\end{align}
Therefore, by \eqref{eq:DecompEzk},
\begin{align}\label{eq:ExactEzk}
\mathbb{E}(Z_{k})=\frac{35-3e^{-4\mu_{k}(\theta)T}-32e^{-2\mu_{k}(\theta)T}}{64\mu_{k}^{5}(\theta)}-\frac{9T+10T\,e^{-2\mu_{k}(\theta)T}}{16\mu_{k}^{4}(\theta)}+\frac{T^{2}}{8\mu_{k}^{3}(\theta)}+\frac{T^{3}}{12\mu_{k}^{2}(\theta)},
\end{align}
which leads to \eqref{eq:newezk}, since by the assumption (ii), the first three terms above all have higher orders than the last term, as $k\rightarrow\infty$, and since $T>0$ is a fixed constant.
Next, we study the asymptotic order of $\text{Var}(Z_{k})$, $k\rightarrow\infty$, given by \eqref{eq:varzk}, which now reduces to
\begin{align}\label{eq:NewVarzk}
\text{Var}(Z_{k})\asymp\frac{T^{3}}{\mu_{k}^{5}(\theta)},\quad k\rightarrow\infty.
\end{align}
In light of \eqref{eq:ExactEzk}, we are left to compute $\mathbb{E}(Z_{k}^{2})$. By It\^{o}'s formula, and since the It\^{o} integral terms have zero expectation, we have
\begin{align}\label{eq:DecompEzk2}
\mathbb{E}(Z_{k}^{2})=2\!\int_{0}^{T}\!\mathbb{E}\!\left(Z_{k}(t)\xi_{k}^{2}(t)\right)dt=2\!\int_{0}^{T}\!\!\int_{0}^{t}\mathbb{E}\!\left(\xi_{k}^{4}(s)\right)\!ds\,dt+4\!\int_{0}^{T}\!\!\int_{0}^{t}\mathbb{E}\!\left(Z_{k}(s)\xi_{k}(s)u_{k}^{2}(s)\right)\!ds\,dt.
\end{align}
To compute the first expectation in \eqref{eq:DecompEzk2}, by It\^{o}'s formula again, we continue
\begin{align}
d\xi_{k}^{4}(t)&=4\xi_{k}^{3}(t)u_{k}^{2}(t)\,dt,\\
d\!\left(\xi_{k}^{3}(t)u_{k}^{2}(t)\right)&=\left(3\xi_{k}^{2}(t)u_{k}^{4}(t)+\xi_{k}^{3}(t)-2\mu_{k}(\theta)\xi_{k}^{3}(t)u_{k}^{2}(t)\right)dt+2\xi_{k}^{3}(t)u_{k}(t)\,dw_{k}(t),\\
d\xi_{k}^{3}(t)&=3\xi_{k}^{2}(t)u_{k}^{2}(t)\,dt,\\
d\!\left(\xi_{k}^{2}(t)u_{k}^{4}(t)\right)&=2\left(\xi_{k}(t)u_{k}^{6}(t)+3\xi_{k}^{2}(t)u_{k}^{2}(t)-2\mu_{k}(\theta)\xi_{k}^{2}(t)u_{k}^{4}(t)\right)dt+4\xi_{k}^{2}(t)u_{k}^{3}(t)\,dw_{k}(t).
\end{align}
Hence, we only need to compute $\mathbb{E}(\xi_{k}(s)u_{k}^{6}(s))$ and $\mathbb{E}(\xi_{k}^{2}(s)u_{k}^{2}(s))$. Again by It\^{o}'s formula, we get
\begin{align}
d\xi_{k}(t)u_{k}^{6}(t)&=\left(u_{k}^{8}(t)+15\xi_{k}(t)u_{k}^{4}(t)-6\mu_{k}(\theta)\xi_{k}(t)u_{k}^{6}(t)\right)dt+6\xi_{k}(t)u_{k}^{5}(t)\,dw_{k}(t),\\
d\xi_{k}^{2}(t)u_{k}^{2}(t)&=\left(2\xi_{k}(t)u_{k}^{4}(t)+\xi_{k}^{2}(t)-2\mu_{k}(\theta)\xi_{k}^{2}(t)u_{k}^{2}(t)\right)dt+2\xi_{k}^{2}(t)u_{k}(t)\,dw_{k}(t),\\
d\xi_{k}(t)u_{k}^{4}(t)&=\left(u_{k}^{6}(t)+6\xi_{k}(t)u_{k}^{2}(t)-4\mu_{k}(\theta)\xi_{k}(t)u_{k}^{4}(t)\right)dt+4\xi_{k}(t)u_{k}^{3}(t)\,dw_{k}(t).
\end{align}
Therefore, by \eqref{eq:CenNormalMoments} and \eqref{eq:Exikuk2}, we can obtain first $\mathbb{E}(\xi_{k}(s)u_{k}^{6}(s))$ and $\mathbb{E}(\xi_{k}^{2}(s)u_{k}^{2}(s))$, and then $\mathbb{E}(\xi_{k}^{3}(s))$, $\mathbb{E}(\xi_{k}^{2}(s)u_{k}^{4}(s))$ and $\mathbb{E}(\xi_{k}^{3}(s)u_{k}^{2}(s))$, and finally $\mathbb{E}(\xi_{k}^{4}(s))$. A similar argument leads to the computation of the second expectation in \eqref{eq:DecompEzk2}.
To sum up, with the help of Mathematica, we obtain that
\begin{align}
\text{Var}(Z_{k})&=-\frac{16917}{512\,\mu_{k}^{10}(\theta)}+\frac{3\,e^{-8\mu_{k}(\theta)T}}{128\,\mu_{k}^{10}(\theta)}+\frac{79\,e^{-6\mu_{k}(\theta)T}}{128\,\mu_{k}^{10}(\theta)}+\frac{2953\,e^{-4\mu_{k}(\theta)T}}{512\,\mu_{k}^{10}(\theta)}+\frac{3409\,e^{-2\mu_{k}(\theta)T}}{128\,\mu_{k}^{10}(\theta)}\\
&\quad\,\,+\frac{1093T}{32\,\mu_{k}^{9}(\theta)}+\frac{45T\,e^{-6\mu_{k}(\theta)T}}{64\,\mu_{k}^{9}(\theta)}+\frac{1165T\,e^{-4\mu_{k}(\theta)T}}{128\,\mu_{k}^{9}(\theta)}+\frac{2321T\,e^{-2\mu_{k}(\theta)T}}{64\,\mu_{k}^{9}(\theta)}\\
&\quad\,\,-\frac{659T^{2}}{64\,\mu_{k}^{8}(\theta)}+\frac{53T^{2}e^{-4\mu_{k}(\theta)T}}{16\,\mu_{k}^{8}(\theta)}+\frac{71T^{2}e^{-2\mu_{k}(\theta)T}}{8\,\mu_{k}^{8}(\theta)}-\frac{5T^{3}}{12\,\mu_{k}^{7}(\theta)}-\frac{5T^{3}e^{-4\mu_{k}(\theta)T}}{8\,\mu_{k}^{7}(\theta)}\\
&\quad\,\,-\frac{113T^{3}e^{-2\mu_{k}(\theta)T}}{24\,\mu_{k}^{7}(\theta)}+\frac{23T^{4}}{48\,\mu_{k}^{6}(\theta)}-\frac{5T^{4}e^{-2\mu_{k}(\theta)T}}{2\mu_{k}^{6}(\theta)}+\frac{T^{5}}{15\mu_{k}^{5}(\theta)},
\end{align}
which clearly implies \eqref{eq:NewVarzk}, and thus completes the proof.\hfill $\Box$
\medskip
\noindent
\textit{Proof of Lemma \ref{lemma:MomentEstxi}:} By \eqref{eq:FModeSols} and Cauchy--Schwartz inequality, for any $0\leq s\leq t\leq T$,
\begin{align}
u_{k}^{2}(s)&=e^{-2\mu_{k}(\theta)s}\left(u_{k}(0)+\sigma\lambda_{k}^{-\gamma}\int_{0}^{s}e^{\mu_{k}(\theta)r}\,dw_{k}(r)\right)^{2}\\
&\leq e^{-2\mu_{k}(\theta)s}\left(u_{k}^{2}(0)+\sigma^{2}\lambda_{k}^{-2\gamma}t\right)\left(1+\frac{1}{t}\left(\int_{0}^{s}e^{\mu_{k}(\theta)r}\,dw_{k}(r)\right)^{2}\right).
\end{align}
Hence, for any $t\in[0,T]$, and $n\in\mathbb{N}$,
\begin{align}
\xi_{k}^{n}(t)&\leq \left(u_{k}^{2}(0)+\sigma^{2}\lambda_{k}^{-2\gamma}t\right)^{n}\left\{\int_{0}^{t}e^{-2\mu_{k}(\theta)s}\left[1+\frac{1}{t}\left(\int_{0}^{s}e^{\mu_{k}(\theta)r}\,dw_{k}(r)\right)^{2}\right]ds\right\}^{n}\\
&=\left(u_{k}^{2}(0)+\sigma^{2}\lambda_{k}^{-2\gamma}t\right)^{n}\left[\frac{1-e^{-2\mu_{k}(\theta)t}}{2\mu_{k}(\theta)}+\frac{1}{t}\int_{0}^{t}e^{-2\mu_{k}(\theta)s}\left(\int_{0}^{s}e^{\mu_{k}(\theta)r}\,dw_{k}(r)\right)^{2}ds\right]^{n}\\
&\leq\left(u_{k}^{2}(0)+\sigma^{2}\lambda_{k}^{-2\gamma}t\right)^{n}\left\{\left(\frac{1-e^{-2\mu_{k}(\theta)t}}{\mu_{k}(\theta)}\right)^{n}\!\!+\frac{2^{n}}{t^{n}}\left[\int_{0}^{t}e^{-2\mu_{k}(\theta)s}\left(\int_{0}^{s}e^{\mu_{k}(\theta)r}\,dw_{k}(r)\right)^{2}\!ds\right]^{n}\right\}.
\end{align}
By~\cite[Theorem 2.1]{Lototsky2009Survey}, there exists a constant $\widetilde{D}_{n}=\widetilde{D}_{n}(t)>0$, such that
\begin{align}
\mathbb{E}\left(\left[\int_{0}^{t}e^{-2\mu_{k}(\theta)s}\left(\int_{0}^{s}e^{\mu_{k}(\theta)r}\,dw_{k}(r)\right)^{2}ds\right]^{n}\right)\leq\frac{\widetilde{D}_{n}}{\mu_{k}^{n}(\theta)}.
\end{align}
Therefore, for any $t\in[0,T]$ and $n\in\mathbb{N}$,
\begin{align}
\mathbb{E}\left(\xi_{k}^{n}(t)\right)\leq\left(u_{k}^{2}(0)+\sigma^{2}\lambda_{k}^{-2\gamma}t\right)^{n}\left(\frac{1}{\mu_{k}^{n}(\theta)}+\frac{2^{n}\widetilde{D}_{n}}{t^{n}}\frac1{\mu_{k}^{n}(\theta)}\right)=D_{n}\left(\frac{u_{k}^{2}(0)+\sigma\lambda_{k}^{-2\gamma}t}{\mu_{k}(\theta)}\right)^{n},
\end{align}
where $D_{n}=D_{n}(t):=1+2^{n}t^{-n}\widetilde{D}_{n}$.\hfill $\Box$
\medskip
We conclude the appendix by listing, for sake of completeness, a version of law of large numbers and a version of central limit theorem used in the proofs of the main results in this paper.
For the detailed proofs of these results we refer the reader to \cite{ShiryaevBookProbability}.
\begin{theorem}\label{thm:SLLN} \textbf{\emph{(Strong Law of Large Number)}}$\,$
Let $\{\eta_{n}\}_{n\in\mathbb{N}}$ be a sequence of independent random variables, and let $\{b_{n}\}_{n\in\mathbb{N}}$ be a sequence of non-decreasing positive numbers such that $\lim_{n\rightarrow\infty}b_{n}=\infty$. If
\begin{align}
\sum_{n=1}^{\infty}\frac{\emph{Var}\left(\eta_{n}\right)}{b_{n}^{2}}<\infty,
\end{align}
then
\begin{align}
\lim_{n\rightarrow\infty}\frac{1}{b_{n}}\sum_{k=1}^{n}\left(\eta_{k}-\mathbb{E}(\eta_{k})\right)=0,\quad\mathbb{P}-\text{a.}\,\text{s.}.
\end{align}
\end{theorem}
\begin{remark}\label{rem:SLLN}
As an immediate corollary, if $\{\eta_{n}\}_{n\in\mathbb{N}}$ is a sequence of independent non-negative random variables with
\begin{align}
\sum_{n=1}^{\infty}\mathbb{E}(\eta_{n})=\infty\quad\text{and}\quad\sum_{n=1}^{\infty}\frac{\text{Var}\left(\eta_{n}\right)}{\left(\sum_{k=1}^{n}\mathbb{E}(\eta_{k})\right)^{2}}<\infty,
\end{align}
then
\begin{align}
\lim_{n\rightarrow\infty}\frac{\sum_{k=1}^{n}\eta_{k}}{\sum_{k=1}^{n}\mathbb{E}(\eta_{k})}=1,\quad\mathbb{P}-\text{a.}\,\text{s.}.
\end{align}
\end{remark}
\begin{theorem}\label{thm:LyapunovCLT} \textbf{\emph{(Lyapunov Central Limit Theorem)}}
Let $\{\eta_{n}\}_{n\in\mathbb{N}}$ be a sequence of independent random variables with finite second moments. If there exists some $\delta>0$, such that
\begin{align}\label{eq:Lyapunov}
\lim_{n\rightarrow\infty}\frac{1}{\left(\sum_{k=1}^{n}\emph{Var}(\eta_{k})\right)^{2+\delta}}\sum_{k=1}^{n}\mathbb{E}\left(\left|\eta_{k}-\mathbb{E}(\eta_{k})\right|^{2+\delta}\right)=0,
\end{align}
then
\begin{align}
\frac{\sum_{k=1}^{n}\left(\eta_{k}-\mathbb{E}(\eta_{k})\right)}{\sqrt{\sum_{k=1}^{n}\emph{Var}(\eta_{k})}}\;{\stackrel{d}{\longrightarrow}}\;\mathcal{N}(0,1),\quad n\rightarrow\infty.
\end{align}
\end{theorem}
\section*{Acknowledgments}
Part of the research was performed while Igor Cialenco was visiting the Institute for Pure and Applied Mathematics (IPAM), which is supported by the National Science Foundation. The authors would like to thank the anonymous referees, the associate editor and the editor for their helpful comments and suggestions which improved greatly the final manuscript.
\bibliographystyle{alpha}
\def$'${$'$}
| 5cfed8e5835c99870eccaaa8a94f587cb4f2ef4c | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Consider the standard stochastic approximation scheme given by,
\begin{equation}\label{recstd}
X_{n+1}=X_n+a(n)\left(h(X_n)+M_{n+1}\right)
\end{equation}
where $\left\{X_n\right\}_{n\geq0}$ is a sequence of $\mathbb{R}^d$-valued random variables, $h:\mathbb{R}^d\rightarrow\mathbb{R}^d$ is a Lipschitz
continuous drift function and $\left\{M_{n}\right\}_{n\geq1}$ is a sequence of $\mathbb{R}^d$-valued random variables which denote the additive noise
terms. In \cite{benaim2}, it was shown that under certain assumptions the asymptotic behavior of recursion \eqref{recstd} can be determined by
the asymptotic behavior of the \it{o.d.e.}\rm
\begin{equation*}
\frac{dx}{dt}=h(x).
\end{equation*}
This method is known as the ODE method and the central idea of this method is to show that the linearly interpolated trajectory of recursion
\eqref{recstd} \it{``tracks'' }\rm the flow of the o.d.e. This idea was later generalized in \cite{benaim3} to analyze the asymptotic behavior
of stochastic processes with continuous sample paths for which a tracking argument as in the ODE method could be established. It was here that
the notion of \it{asymptotic pseudotrajectory} \rm was introduced. For a precise definition we refer the reader to chapter 3 of \cite{benaim3}.
In many applications arising in machine learning and optimization, the drift function $h$ is set-valued or is single-valued and does not
satisfy the assumption of Lipschitz continuity (or even continuity). In such cases the recursion studied takes the form,
\begin{equation}\label{recsvm}
X_{n+1}-X_{n}-a(n)M_{n+1}\in a(n)H(X_n),
\end{equation}
where $H:\mathbb{R}^d\rightarrow\left\{\text{subsets of }\mathbb{R}^d\right\}$ is a set-valued map satisfying certain assumptions. Such
recursions were first studied in \cite{benaim1} by showing that the linearly interpolated trajectory of recursion \eqref{recsvm} \it{``tracks'' }\rm
the flow of the differential inclusion given by,
\begin{equation*}
\frac{dx}{dt}\in H(x).
\end{equation*}
The notion of asymptotic pseudotrajectory was also extended to the set-valued case and a limit set theorem established.
In many applications such as methods relying on Monte Carlo simulation where sampling exactly from a distribution (or family of distributions)
is not possible and instead Markov chain Monte Carlo methods are used, the recursion consists of an additional non-additive iterate-dependent
Markov noise component. The recursion takes the general form,
\begin{equation}\label{recmark1}
X_{n+1}-X_n=a(n)\left(h(X_n,S_n)+M_{n+1}\right),
\end{equation}
where $\{S_n\}_{n\geq0}$ denotes the Markov noise component. The other quantities in the above recursion have the same interpretation as those
in \eqref{recstd}. The analysis of such a recursion was first performed in \cite{benven}. In \cite{benven}, the analysis is
carried out under the assumptions that the drift function $h$ is continuous, the Markov chain defined by the transition kernel associated with
the Markov noise terms in the recursion admits a unique stationary distribution as well as a solution to the Poisson equation among others. The
case with discontinuities in the drift function was considered in \cite{tadic} but the analysis required the function $h(\cdot,s)$ to be Holder continuous.
Recently in \cite{matti}, such a recursion has been analyzed when the drift function is just measurable. In all the above references, however, the Markov
noise terms are assumed to have a unique stationary distribution and the assumptions imply an existence of a solution to the Poisson equation.
Another set of assumptions on the Markov noise terms under which the recursion \eqref{recmark1} is studied can be found in \cite{borkarmark}.
In \cite{borkarmark}, the transition kernel defining the Markov noise terms is required to be continuous in both the iterate and the state variables.
The advantage in this case is that the set of stationary distributions need not be unique and one does not need the Markov chain associated with
the transition kernel for every value of the iterate, to be aperiodic and irreducible. Further the analysis in \cite{borkarmark} enables one to study
the recursion \eqref{recmark1} when the noise terms are not Markov by themselves, but their lack of Markov property comes through the dependence on
a control sequence.
In this paper we extend the above to the case where the drift function $h$ in recursion \eqref{recmark1}, is a set-valued map. The recursion now
takes the form,
\begin{equation*}
X_{n+1}-X_n-a(n)M_{n+1}\in a(n)H(X_n,S_n),
\end{equation*}
where $H:\mathbb{R}^d\times\mathcal{S}\rightarrow\left\{\text{subsets of }\mathbb{R}^d\right\}$ is a set-valued map and $\mathcal{S}$ denotes the
state space of the Markov noise terms which we assume to be a compact metric space. The assumption on the Markov noise terms is similar to
\cite{borkarmark}. We shall show that the linearly interpolated trajectory of the above recursion is an asymptotic pseudotrajectory for the flow
of a limiting differential inclusion obtained by averaging the set-valued map w.r.t. the stationary distributions of the Markov noise terms.
The main idea is to approximate the set-valued drift function with continuous set-valued maps which admit a single-valued parametrization, there
by enabling us to write the recursion as a standard stochastic approximation scheme with single-valued maps for which an asymptotic
pseudotrajectory argument is easy to establish. Later we invoke the limit set theorem in \cite{benaim1} to characterize the limit set of
the above recursion.\newline
\it{Organization of the paper:}\rm
In section \ref{rec_ass}, we formally define the recursion and state the assumptions imposed, which is then followed by a brief discussion on each of
the assumptions.
In section \ref{backgrd}, we review certain results from set-valued analysis which will be used later to analyze the recursion and define the
limiting differential inclusion.
In section \ref{ldi}, the limiting differential inclusion is defined and properties of the same are established.
In section \ref{sopmvf}, we define the space of probability measure valued functions and define an appropriate topology on this space. A metrization
lemma for the above mentioned space is stated and a continuous function on such a space needed later in analyzing the recursion is defined.
In section \ref{recanal}, we first state some preliminaries and later state our main result followed by the limit set theorem.
In section \ref{siacmnc}, we state two variants of the Markov noise assumption under which the analysis of the stochastic approximation scheme
can be carried out along similar lines.
In section \ref{app}, we state four applications where the recursion analyzed in this paper naturally appears. We finally conclude in section
\ref{cadffw} by providing a few interesting directions for future work.
\section{Proof of the continuous embedding lemma (Lemma \ref{ctem})}
\label{prf_ctem}
The proof of this lemma is similar to the proof of Theorem 1, chapter 1, section 13 of \cite{aubindi}. We shall provide a brief outline here
for the sake of completeness.
For any $\epsilon>0$, for every $(x_0,s_0)\in\mathbb{R}^d\times\mathcal{S}$, let
$B(\epsilon,x_0,s_0):=\left\{(x,s):\parallel x-x_0\parallel<\epsilon,\ d_{\mathcal{S}}(s,s_0)<\epsilon\right\}$. Let
$\left\{\epsilon_l:=\frac{1}{3^l}\right\}_{l\geq1}$. Then for every $l\geq1$, $\mathscr{C}_l:=\left\{B(\epsilon_l,x_0,s_0):(x_0,s_0)\in
\mathbb{R}^d\times\mathcal{S}\right\}$ is an open covering of $\mathbb{R}^d\times\mathcal{S}$. Since $\mathbb{R}^d\times\mathcal{S}$ is a
metric space, it is paracompact (see Defn. 4 and Theorem 1 in chapter 0, section 1 of \cite{aubindi}). Therefore for every $l\geq1$, there
exists a locally finite open refinement of the covering $\mathscr{C}_l$ and let it be denoted by $\tilde{\mathscr{C}}_l:=
\left\{C_i^l\right\}_{i\in I^{l}}$ where $I^l$ is an arbitrary index set. By Theorem 2, chapter 0, section 1 of \cite{aubindi}, there exists a
continuous partition of unity, $\left\{\psi_i^{l}\right\}_{i\in I^l}$, subordinated to the covering $\tilde{\mathscr{C}}_l$. Therefore,
for every $l\geq1$, for every $i\in I^l$, there exists $(x_i^l,s_i^l)$, such that $support(\psi_i^l)\subseteq C_i^l\subseteq
B(\epsilon_l,x_i^l,s_i^l)$. For every $l\geq1$, for every $(x,s)$, let $I^l(x,s):=\left\{i\in I^l:\psi_i^l(x,s)>0\right\}$ and by definition
of $\psi_i^l$, we have that $|I^l(x,s)|<\infty$ and $\sum_{i\in I^l(x,s)}\psi_i^l(x,s)=1$.
For every $l\geq1$, define the set valued map $H^{(l)}:\mathbb{R}^d\times\mathcal{S}\rightarrow\left\{\text{subsets of }\mathbb{R}^d\right\}$,
such that for every $(x,s)$, $H^{(l)}(x,s):=\sum_{i\in I^l(x,s)}\psi_i^{l}(x,s)A_i^{l}$, where $A_i^{l}:=\bar{co}\left(H\left(B\left(2\epsilon_l,
x_i^l,s_i^l\right)\right)\right)$.
Fix $l\geq1$. For every $i\in I^l$, by assumption $(A1)(ii)$, $\sup_{z\in H\left(B\left(2\epsilon_l,x_i^l,s_i^l\right)\right)}\parallel z
\parallel\leq K(1+\parallel x_i^l\parallel +2\epsilon_l)$ from which we can deduce that $\sup_{z\in A_i^l}\parallel z\parallel \leq
K(1+\parallel x_i^l\parallel+2\epsilon_l)$. Hence $A_i^l$ is convex and compact. Therefore $H^{(l)}(x,s)$ is a convex combination of compact
and convex subsets of $\mathbb{R}^d$ and hence is \bf{convex and compact}\rm.
Fix $l\geq1$ and $(x,s)$. Then for every $i\in I^l(x,s)$, $(x,s)\in support(\psi_i^l)\subseteq C_i^l\subseteq B(\epsilon_l,x_i^l,s_i^l)\subseteq
B(2\epsilon_l,x_i^l,s_i^l)$. Therefore for every $i\in I^l(x,s)$, $H(x,s)\subseteq H\left(B\left(2\epsilon_l,x_i^l,s_i^l\right)\right)\subseteq
A_i^l$ and since $H(x,s)$ is convex we have that $H(x,s)=\sum_{i\in I^l(x,s)}\psi_i^l(x,s)H(x,s)\subseteq\sum_{i\in I^l(x,s)}\psi_i^l(x,s)A_i^l=
H^{(l)}(x,s)$. Thus $\bf{H(x,s)\subseteq H^{(l)}(x,s)}\rm$.
Fix $l\geq1$ and $(x,s)$. Then for every $i\in I^{l+1}(x,s)$, $\parallel x-x_i^{l+1}\parallel<\epsilon_{l+1}$ and $d_{\mathcal{S}}(s,s_i^{l+1})
<\epsilon_{l+1}$. Similarly for every $j\in I^{l}(x,s)$, $\parallel x-x_j^{l}\parallel<\epsilon_{l}$ and $d_{\mathcal{S}}(s,s_j^{l})
<\epsilon_{l}$. Therefore for every $i\in I^{l+1}(x,s)$, for every $j\in I^l(x,s)$, $\parallel x_i^l-x_i^{l+1}\parallel<\epsilon_l+\epsilon_{l+1}=
\epsilon_l+\frac{\epsilon_l}{3}=\frac{4\epsilon_l}{3}$ and $d_{\mathcal{S}}(s_i^l,s_i^{l+1})<\frac{4\epsilon_l}{3}$. For every $i\in
I^{l+1}(x,s)$, for every $j\in I^{l}(x,s)$, for every $(x',s')\in B(2\epsilon_{l+1},x_i^{l+1},s_i^{l+1})$, $\parallel x'-x_j^{l}\parallel\leq
\parallel x'-x_i^{l+1}\parallel+\parallel x_i^{l+1}-x_j^l\parallel<2\epsilon_{l+1}+\frac{4\epsilon_l}{3}=\frac{2\epsilon_l}{3}+
\frac{4\epsilon_l}{3}=2\epsilon_l$ and $d_{\mathcal{S}}(s',s_j^{l})<2\epsilon_l$. Thus for every $i\in I^{l+1}(x,s)$, for every
$j\in I^{l}(x,s)$, for every $(x',s')\in B(2\epsilon_{l+1},x_i^{l+1},s_i^{l+1})$, $H(x',s')\subseteq H\left(B\left(2\epsilon_l,
x_j^l,s_j^l\right)\right)\subseteq A_j^l$. Therefore for every $i\in I^{l+1}(x,s)$, for every $j\in I^{l}(x,s)$, $A_i^{l+1}\subseteq A_j^l$
and by using convexity of $A_i^{l+1}$ we get that for every $i\in I^{l+1}(x,s)$, $A_i^{l+1}=\sum_{j\in I^{l}(x,s)}\psi_j^{l}(x,s)A_i^{l+1}\subseteq
\sum_{j\in I^{l}(x,s)}\psi_j^{l}(x,s)A_j^{l}=H^{(l)}(x,s)$. By using convexity of $H^{(l)}(x,s)$ we get that
$H^{(l+1)}(x,s)=\sum_{i\in I^{l+1}(x,s)}\psi_i^{l+1}(x,s)A_i^{l+1}\subseteq\sum_{i\in I^{l+1}(x,s)}\psi_i^{l+1}(x,s)H^{(l)}(x,s)=H^{(l)}(x,s)$.
Therefore, $\bf{H^{(l+1)}(x,s)\subseteq H^{(l)}(x,s)}\rm$.
Fix $(x,s)$. Clearly $H(x,s)\subseteq \cap_{l\geq1}H^{(l)}(x,s)$. By u.s.c. of $H$, we have that for every $\epsilon>0$, there exists
$\delta(\epsilon,x,s)>0$ such that for every $(x',s')$ satisfying $\parallel x'-x\parallel<\delta(\epsilon,x,s)$ and $d_{\mathcal{S}}(s',s)<
\delta(\epsilon,x,s)$ we have $H(x',s')\subseteq H(x,s)+\epsilon U$. Clearly we can find $L$ such that for every $l\geq L$, $3\epsilon_l<
\delta(\epsilon,x,s)$. Therefore for every $\epsilon>0$, for $L$ as above, we have that for every $l\geq L$, for every $i\in I^l(x,s)$,
$H\left(B\left(2\epsilon_l,x_i^l,s_i^l\right)\right)\subseteq H(x,s)+\epsilon U$. Since $H(x,s)+\epsilon U$ is convex we get, for every
$\epsilon>0$, there exists $L$ such that for every $l\geq L$, for every $i\in I^l(x,s)$, $A_i^l\subseteq H(x,s)+\epsilon U$. Again by convexity
of $H(x,s)+\epsilon U$, we have that for every $\epsilon>0$, there exists $L$ such that $l\geq L$, $H^{(l)}(x,s)=\sum_{i\in I^l(x,s)}
\psi^l_i(x,s)A_i^l\subseteq H(x,s)+\epsilon U$. Therefore for every $\epsilon>0$, $\cap_{l\geq1}H^{(l)}(x,s)\subseteq H(x,s)+\epsilon U$. Now
it follows that $\bf{H(x,s)=\cap_{l\geq1}H^{(l)}(x,s)}\rm$.
Fix $l\geq1$ and $(x,s)$. We have already shown that for every $i\in I^l(x,s)$, $\sup_{z\in A_i^l}\parallel z\parallel\leq K(1+\parallel x_i^l
\parallel+2\epsilon_l)$. Using the fact that for every $i\in I^l(x,s)$, $\parallel x_i^l-x\parallel<\epsilon_l$ we get that for every $i\in
I^l(x,s)$, $\sup_{z\in A_i^l}\parallel z\parallel\leq K(1+\parallel x\parallel+3\epsilon_l)$. Since $\left\{\psi_i^l(x,s)\right\}_{i\in
I^l(x,s)}$ are convex combining coefficients, we get that $\bf{\sup_{z\in H^{(l)}(x,s)}\parallel z\parallel\leq K^{(l)}(1+\parallel x
\parallel)}\rm$, where $K^{(l)}:=K+3\epsilon_l$.
Fix $l\geq1$ and $(x,s)$. Since $\tilde{\mathscr{C}}_l$ is a locally finite open covering of $\mathbb{R}^d\times\mathcal{S}$, there exists
$\delta>0$ such that $|I^l(\delta,x,s)|<\infty$ where $I^l(\delta,x,s):=\left\{i\in I^l:C_i^l\cap B(\delta,x,s)\neq\emptyset\right\}$. Since
for every $i\in I^l$, $support(\psi_i^l)\subseteq C_i^l$, for every $(x',s')\in B(\delta,x,s)$ we have that $H^{(l)}(x',s')=\sum_{i\in
I^l(\delta,x,s)}\psi_i^l(x',s')A_i^l$. Let $z\in H^{(l)}(x,s)$ and $\left\{\left(x_n,s_n\right)\right\}_{n\geq1}$ be some sequence converging
to $(x,s)$. Then for every $i\in I^l(\delta,x,s)$ there exists $a_i^l\in A_i^l$ such that $z=\sum_{i\in I^l(\delta,x,s)}\psi_i^l(x,s)a_i^l$
and there exists $N$, such that for every $n\geq N$, $(x_n,s_n)\in B(\delta,x,s)$. For every $n\leq N$, define $z_n$ to be any value in
$H^{(l)}(x_n,s_n)$ and for every $n\geq N$, define $z_n=\sum_{i\in I^l(\delta,x,s)}\psi_i^l(x_n,s_n)a_i^l\in H^{(l)}(x_n,s_n)$ where $a_i^l$
are as defined above. Then clearly for every $n\geq1$, $z_n\in H^{(l)}(x_n,s_n)$ and $\left\{z_n\right\}_{n\geq1}$ converges to $z$. Thus
$H^{(l)}$ is l.s.c. at $(x,s)$. Upper simecontinuity of the map $H^{(l)}$ at $(x,s)$ follows from Theorem 1, chapter 1, section 13 of
\cite{aubindi}. Since $(x,s)$ is arbitrary the map $H^{(l)}$ is both u.s.c. and l.s.c at every $(x,s)$ and hence is \bf{continuous}\rm.
\section{Measurability of set-valued map $G$ in Lemma \ref{chint}}
\label{prf_msb}
Recall that $G:\mathcal{S}\rightarrow\left\{\text{subsets of }U\right\}$ is such that for every $s\in\mathcal{S}$, $G(s):=\left\{u\in U:
h^{(l)}(x,s,u)=f(s)\right\}$. Let $C\subseteq U$ be a closed subset. Then $G^{-1}(C)=\left\{s\in\mathcal{S}:G(s)\cap C\neq\emptyset\right\}$.
Thus $s\in G^{-1}(C)$ if and only if there exists $u\in C$, such that $h^{(l)}(x,s,u)=f(s)$ or equivalently $s\in G^{-1}(C)$ if and only if
$f(s)\in h^{(l)}(x,s,C)$, where $h^{(l)}(x,s,C):=\left\{h^{(l)}(x,s,u):u\in C\right\}$. Since $h_x^{(l)}:\mathcal{S}\times C\rightarrow
\mathbb{R}^d$ is continuous, by Proposition 1, chapter 1, section 2 of \cite{aubindi}, the set valued map $s\rightarrow h_x^{(l)}(s,C)
$ is continuous and its measurability now follows from arguments similar to Lemma \ref{msble}(ii). Therefore, $G^{-1}(C)=\left\{s\in\mathcal{S}
:f(s)\in h^{(l)}(x,s,C)\right\}=\left\{s\in\mathcal{S}:d(f(s),h^{(l)}(x,s,C))=0\right\}$. By Lemma \ref{mtool},
$\left\{s\in\mathcal{S}:d(f(s),h^{l}(x,s,C))=0\right\}\in\mathscr{B}(\mathcal{S})$ and hence $G^{-1}(C)\in\mathscr{B}(\mathcal{S})$.
\section{Proof of uniform convergence in Lemma \ref{march1}(iii)}
\label{prf_conv}
Since $\left\{x_{n_k}\right\}_{k\geq1}$ converges to $x$, there exists $r>0$, such that $\sup_{k\geq1}\parallel x_{n_k}\parallel\leq r$.
Fix $\epsilon>0$. We know that $h^{(l)}(\cdot)|_{2rU\times \mathcal{S}\times U}$ is uniformly continuous and hence there exists $\delta>0$
(depending on $\epsilon$), such that for every $x'\in\mathbb{R}^d$ satisfying $\parallel x'-x\parallel<\delta$,
$\sup_{(s,u)\in\mathcal{S}\times U}\parallel h_x'^{(l)}(s,u)-h_x^{(l)}(s,u)\parallel<\epsilon$. Hence there exists $K$, such that for every
$k\geq K$, $\parallel x_{n_k}-x\parallel<\delta$ and therefore for every $k\geq K$,
$\sup_{(s,u)\in\mathcal{S}\times U}\parallel h_{x_{n_k}}^{(l)}(s,u)-h_x^{(l)}(s,u)\parallel<\epsilon$.
\section{Applications}
\label{app}
In this section we consider four applications of the foregoing. These applications presented are natural extensions of the ones presented
in chapter 5.3 of \cite{borkartxt} to the case with Markov noise. All the applications presented in this section are stated with Markov noise
terms $\left\{S_n\right\}_{n\geq0}$ satisfying assumption $(A2)$. One may also consider the applications presented here with the Markov noise
terms satisfying assumptions $(A2)'$ or $(A2)''$ as well.
\subsection{Controlled stochastic approximation}
Consider the iteration
\begin{equation}\label{reccsa}
X_{n+1}-X_n-a(n)M_{n+1}=a(n)h(X_n,Z_n,S_n),
\end{equation}
where, $\{Z_n\}_{n\geq0}$ is a random sequence taking values in a compact metric space $\mathcal{Z}$ and $h:\mathbb{R}^d\times\mathcal{Z}
\times\mathcal{S}\rightarrow\mathbb{R}^d$ is continuous and Lipschitz in the first argument uniformly w.r.t. the second and third, i.e.,
there exists $L>0$ such that, for every $(z,s)\in\mathcal{Z}\times\mathcal{S}$ and for every $x_1,x_2\in\mathbb{R}^d$,
\begin{equation*}
\parallel h(x_1,z,s)-h(x_2,z,s)\parallel\leq L\parallel x_1-x_2\parallel.
\end{equation*}
$\{Z_n\}_{n\geq0}$ is viewed here as a control sequence, that is, $Z_n$ is chosen by the agent running the algorithm at time $n\geq0$ based on
the observed history and possibly extraneous independent randomization as is usual in stochastic control problems. It could also be an
unknown random process that affects the measurements in addition to the additive noise $\{M_n\}_{n\geq0}$. We shall assume that
assumptions $(A2)-(A5)$ hold.
Define the set valued map $H:\mathbb{R}^d\times\mathcal{S}\rightarrow\{\text{subsets of }\mathbb{R}^d\}$ such that, for every $(x,s)\in
\mathbb{R}^d\times\mathcal{S}$,
\begin{equation*}
H(x,s):=\bar{co}(\left\{h(x,z,s):z\in\mathcal{Z}\right\}).
\end{equation*}
Then it can be shown that the set valued map $H$ satisfies assumption $(A1)$ (see Lemma 5 in chapter 5.3 of \cite{borkartxt}). Then
recursion \eqref{reccsa} can be written as,
\begin{equation*}
X_{n+1}-X_n-a(n)M_{n+1}\in a(n)H(X_n,S_n)
\end{equation*}
and the asymptotic behavior of the same can be analyzed as in section \ref{recanal} of this paper.
\subsection{Stochastic subgradient descent of a parametrized convex function}
Consider the function, $J:\mathbb{R}^d\times\mathcal{S}\rightarrow\mathbb{R}$, continuous and for every $s\in\mathcal{S}$, $J(\cdot,s)$
is a convex function. Suppose $J(\cdot,s)$ is not continuously differentiable everywhere, so that $\nabla J(x,s)$ is not defined at all
$x\in\mathbb{R}^d$, where $\nabla$ denotes gradient w.r.t. $x$. A natural generalization of gradient to this non-smooth case is the notion
of subdifferential w.r.t. $x$ denoted by $\partial J(x,s)$, defined as the set of all $y\in\mathbb{R}^d$ such that for every
$z\in\mathbb{R}^d$,
\begin{equation*}
J(z,s)\geq J(x,s)+\langle y,z-x\rangle.
\end{equation*}
Then for every $(x,s)\in\mathbb{R}^d\times\mathcal{S}$, $\partial J(x,s)$ is nonempty, convex and compact (see Proposition 5.4.1 in
\cite{bertsekas}). From the definition of subdifferential and the continuity of function $J(\cdot)$, it can be easily shown that the set
valued map $(x,s)\rightarrow\partial J(x,s)$ has a closed graph.
Define the set valued map $H:\mathbb{R}^d\times\mathcal{S}\rightarrow\{\text{subsets of }\mathbb{R}^d\}$, such that for every
$(x,s)\in\mathbb{R}^d\times\mathcal{S}$,
\begin{equation}\label{sbgrd}
H(x,s)=-\partial J(x,s).
\end{equation}
We further assume that the set valued map $H$ satisfies the linear growth property (that is assumption $(A1)(ii)$). Then by arguments in
the preceding paragraph the set valued map $H$ satisfies assumption $(A1)$.
Consider the recursion \eqref{rec} with the set valued map defined above. The recursion,
\begin{equation*}
X_{n+1}-X_{n}-a(n)M_{n+1}\in a(n)H(X_n,S_n),
\end{equation*}
where $\{M_{n}\}_{n\geq1}$ can be interpreted as subgradient estimation error. One case of interest where assumption $(A4)$ associated
with the additive noise terms is satisfied is when $M_n$ are i.i.d. zero mean with finite variance (more generally martingale difference
terms satisfying assumption $(A3)$ in chapter 2.1 of \cite{borkartxt}).
Suppose the assumptions $(A2)-(A5)$ are satisfied then the asymptotic behavior of the above recursion can be analyzed as
in section \ref{recanal} of this paper. If in addition to assumption $(A2)$, we know that for every $x\in\mathbb{R}^d$,
$|D(x)|=1$ (that is the stationary distribution is unique for every $x$), then the set valued map associated with the limiting DI,
$\hat{H}$ as defined in \eqref{lim1} is such that for every $x\in\mathbb{R}^d$,
\begin{equation*}
\hat{H}(x)\subseteq-\partial\mathbb{E}_{s}[J(\cdot);\mu_x](x),
\end{equation*}
where $\{\mu_x\}=D(x)$, $\mathbb{E}_{s}[J(\cdot);\mu_x]$ denotes a real valued map on $\mathbb{R}^d$ such that, for every
$y\in\mathbb{R}^d$, $\mathbb{E}_{s}[J(\cdot);\mu_x](y)=\int_{\mathcal{S}}J(y,s)\mu_x(ds)$ (which is clearly a convex function) and
$\partial\mathbb{E}_{s}[J(\cdot);\mu_x](x)$ denotes the subdifferential of the above defined map at $x$. If in addition we have that the
state space of the Markov noise is finite (that is $|\mathcal{S}|<\infty$) then by Proposition 5.4.6 in \cite{bertsekas} we have that for
every $x\in\mathbb{R}^d$,
\begin{equation*}
\hat{H}(x)=-\partial\mathbb{E}_{s}[J(\cdot);\mu_x](x).
\end{equation*}
As a consequence of the above we have that the global attractor of the DI,
\begin{equation*}
\frac{dx}{dt}\in-\partial\mathbb{E}_{s}[J(\cdot);\mu_x](x),
\end{equation*}
is also a global attractor of the limiting DI (that is DI \eqref{di2})and hence by the limit set theorem (that is Theorem \ref{ls}), we
can conclude that the iterates $\{X_n\}_{n\geq0}$ will almost surely converge to such an attractor.
Another important application is the minimization of the convex function obtained by averaging the parameter w.r.t. a particular
probability distribution say, $\mu\in\mathcal{P}(\mathcal{S})$. Formally the minimization of function $J_{\mu}:\mathbb{R}^d\rightarrow
\mathbb{R}$, where for every $x\in\mathbb{R}^d$, $J_{\mu}(x):=\int_{\mathcal{S}}J(x,s)\mu(ds)$. Consider recursion \eqref{rec}, where the
set-valued map $H$ is as in equation \eqref{sbgrd} and Markov noise terms (that is $\{S_n\}_{n\geq0}$) satisfy assumption $(A2)$
with the additional condition that for every $x\in\mathbb{R}^d$, $D(x)=\{\mu\}/D_{\phi}(x)=\{\mu\}$. Then under additional assumptions of
$(A1),(A3)-(A5)$ we have that the linearly interpolated trajectory of recursion \eqref{rec} is an APT for the flow of DI \eqref{di2}. The set
valued map $\hat{H}$ associated with DI \eqref{di2}, can be shown to satisfy,
\begin{equation*}
\hat{H}(x)\subseteq-\partial J_{\mu}(x),
\end{equation*}
for every $x\in\mathbb{R}^d$. Further if the state space of the Markov noise is finite (that is $|\mathcal{S}|<\infty$) then by
Proposition 5.4.6 in \cite{bertsekas} we have that for every $x\in\mathbb{R}^d$,
\begin{equation*}
\hat{H}(x)=-\partial J_{\mu}(x).
\end{equation*}
\subsection{Approximate drift problem}
Consider the continuous function $h:\mathbb{R}^d\times\mathcal{S}\rightarrow\mathbb{R}^d$, satisfying the linear growth property, that is,
there exists $K>0$, such that for every $(x,s)\in\mathbb{R}^d\times\mathcal{S}$, $\parallel h(x,s)\parallel\leq
K(1+\parallel x\parallel)$. In many applications one wishes to implement the recursion,
\begin{equation*}
X_{n+1}-X_{n}-a(n)M_{n+1}=a(n)h(X_n,S_n),
\end{equation*}
where assumptions $(A2)/(A2)'/(A2)'',(A3)-(A5)$ are assumed to hold. But in practice one has access only to approximate value of the
function $h(\cdot)$. Further the magnitude of the error in approximation of $h(\cdot)$ is only known to be bounded by a constant, say
$\epsilon>0$. Then the recursion implemented becomes,
\begin{equation}\label{recapd}
X_{n+1}-X_{n}-a(n)M_{n+1}=a(n)(h(X_n,S_n)+\eta_n),
\end{equation}
where $\eta_n$ represents the error in estimation of $h(\cdot)$ and we have that for every $n\geq0$, $\parallel\eta_n\parallel\leq
\epsilon$.
Define the set valued map $H:\mathbb{R}^d\times\mathcal{S}\rightarrow\{\text{subsets of }\mathbb{R}^d\}$ such that for every $(x,s)\in
\mathbb{R}^d\times\mathcal{S}$,
\begin{equation}\label{hapd}
H(x,s):=h(x,s)+\epsilon U,
\end{equation}
where $U$ denotes the closed unit ball in $\mathbb{R}^d$. Then recursion \eqref{recapd} can be written as,
\begin{equation*}
X_{n+1}-X_{n}-a(n)M_{n+1}\in a(n)H(X_n,S_n),
\end{equation*}
where the set valued map $H$ is as defined in equation \eqref{hapd}. It can be easily shown that the set valued map $H$ satisfies
assumption $(A1)$. Under assumptions $(A2)-(A5)$ one may analyze the asymptotic behavior of recursion \eqref{recapd} as
in section \ref{recanal} of this paper.
\subsection{Discontinuous dynamics}
Consider the recursion,
\begin{equation}\label{recdd}
X_{n+1}-X_{n}-a(n)M_{n+1}=a(n)h(X_n,S_n),
\end{equation}
where $h(\cdot)$ is merely measurable, satisfying the linear growth property that is there exists $K>0$, such that for every
$(x,s)\in\mathbb{R}^d\times\mathcal{S}$, $\parallel h(x,s)\parallel\leq K(1+\parallel x\parallel)$.
Define the set valued map $H:\mathbb{R}^d\times\mathcal{S}\rightarrow\{\text{subsets of }\mathbb{R}^d\}$ such that for every $(x,s)\in
\mathbb{R}^d\times\mathcal{S}$,
\begin{equation}\label{hdd}
H(x,s):=\cap_{\epsilon>0}\bar{co}\left(\left\{h(y,s'):\parallel x-y\parallel<\epsilon,\ d_{\mathcal{S}}(s,s')<\epsilon\right\}\right).
\end{equation}
Then the set valued map defined above can be shown to satisfy assumption $(A1)$. Now the recursion \eqref{recdd} can be written in the form
of recursion \eqref{rec} with the set valued map $H$ as defined in equation \eqref{hdd}. Under assumptions $(A2)-(A5)$,
one may analyze the asymptotic behavior of recursion \eqref{recdd} as in section \ref{recanal} of this paper.
\section{Recursion and assumptions}
\label{rec_ass}
Let $(\Omega,\mathscr{F},\mathbb{P})$ be a probability space and $\left\{X_n\right\}_{n\geq0}$ be a sequence of $\mathbb{R}^d$-valued
random variables on $\Omega$ satisfying
\begin{equation}\label{rec}
X_{n+1}-X_{n}-a(n)M_{n+1}\in a(n)H(X_n,S_n)
\end{equation}
where,
\begin{itemize}
\item [(A1)]$H:\mathbb{R}^d\times\mathcal{S}\rightarrow\left\{\text{subsets of }\mathbb{R}^d\right\}$ is a set valued map on
$\mathbb{R}^d\times\mathcal{S}$ where $(\mathcal{S},d_{\mathcal{S}})$ is a compact metric space and the map $H(\cdot)$ satisfies
\begin{itemize}
\item [(i)] for every $(x,s)\in\mathbb{R}^d\times\mathcal{S}$, $H(x,s)$ is a non-empty, compact, convex subset of $\mathbb{R}^d$,
\item [(ii)] there exists $K>0$ such that, for every
$(x,s)\in\mathbb{R}^d\times\mathcal{S},\\ \sup_{z\in H(x,s)}\parallel z\parallel\leq K(1+\parallel x\parallel)$,
\item [(iii)] for every $\mathbb{R}^d\times\mathcal{S}$ valued sequence, $\left\{\left(x_n,s_n\right)\right\}_{n\geq1}$ converging to $(x,s)$, for every
sequence $\left\{z_n\in H(x_n,s_n)\right\}_{n\geq1}$ converging to $z$, we have, $z\in H(x,s)$.
\end{itemize}
\end{itemize}
Throughout this paper $\mathcal{P}(\cdots)$, denotes the set of probability measures on a compact metric space \lq$\cdots$' with the
Prohorov topology (see chapter 2 in \cite{borkarap}).
\begin{itemize}
\item [(A2)]$\left\{S_n\right\}_{n\geq0}$ is a sequence of $\mathcal{S}$ valued random variables on $\Omega$ such that,
for every $n\geq0$, for every $A\in \mathscr{B}(\mathcal{S})$, $\mathbb{P}(S_{n+1}\in A|S_m,X_m,m\leq n)=
\mathbb{P}(S_{n+1}\in A|S_n,X_n)=\Pi(X_n,S_n)(A)$ $a.s.$ where $\Pi:\mathbb{R}^d\times\mathcal{S}\rightarrow\mathcal{P}(\mathcal{S})$,
is continuous.
\item [(A3)] $\left\{a(n)\right\}_{n\geq0}$ is a sequence of real numbers satisfying
\begin{itemize}
\item[(i)]$a(0)\leq1$ and for every $n\geq0$, $a(n)\geq a(n+1)$,
\item[(ii)] $\sum_{n\geq0}a(n)=\infty$ and $\sum_{n\geq0}a(n)^2<\infty$.
\end{itemize}
\item [(A4)] $\left\{M_n\right\}_{n\geq1}$ is a sequence of $\mathbb{R}^d$ valued random variables such that, $a.e. (\omega)$, for every $T>0$
$\\\lim_{n\to\infty}\sup_{n\leq k\leq\tau(n,T)}\parallel\sum_{m=n}^{k}a(m)M_{m+1}\parallel=0$ where,
$\tau(n,T):=\min\left\{m>n:\sum_{k=n}^{m-1}a(k)\geq T\right\}$.
\item[(A5)] $\mathbb{P}(\sup_{n\geq0}\parallel X_n \parallel<\infty)=1$.
\end{itemize}
Assumption $(A1)$ is an extension of the assumption imposed on the set valued map for the case without Markov noise in \cite{benaim1} to the
case with Markov noise. The only difference is in $(A1)(ii)$, since in our case we expect the constant $K$ to be independent of Markov noise
terms. This strengthening of the assumption allows us to obtain integrable measurable selections and also to prove certain properties of the
limiting differential inclusion in future sections.
Assumption $(A2)$ ensures that for every $x\in \mathbb{R}^d$ the Markov chain defined by the transition kernel $\Pi(x,\cdot)(\cdot)$ possesses
weak Feller property (see chapter 6.1.1 of \cite{meyntweed}). In addition to the above, since the state space of the Markov chain is a compact
metric space, by Theorem 12.1.2(ii) in \cite{meyntweed}, for every $x\in\mathbb{R}^d$, the Markov chain defined by the transition kernel
$\Pi(x,\cdot)(\cdot)$ admits at least one stationary distribution ($\mu\in\mathcal{P}(\mathcal{S})$ is stationary for the Markov chain defined
by the transition kernel $\Pi(x,\cdot)(\cdot)$ if, for every $A\in\mathscr{B}(\mathcal{S})$, $\mu(A)=\int_{\mathcal{S}}\Pi(x,s)(A)\mu(ds)$).
For every $x\in\mathbb{R}^d$, let $D(x)\subseteq\mathcal{P}(\mathcal{S})$ denote the set of stationary distributions of the Markov chain
defined by the transition kernel $\Pi(x,\cdot)(\cdot)$. It can be easily shown that,
\begin{itemize}
\item [(i)] for every $x\in\mathbb{R}^d$, $D(x)$ is a convex and compact subset of $\mathcal{P}(\mathcal{S})$.
\item [(ii)] the map $x\rightarrow D(x)$ has a closed graph, that is the set
\begin{equation*}
\left\{(x,\mu)\in\mathbb{R}^d\times\mathcal{P}(\mathcal{S}): x\in\mathbb{R}^d,\ \mu\in D(x)\right\},
\end{equation*}
is a closed subset of $\mathbb{R}^d\times\mathcal{P}(\mathcal{S})$(for a proof of the above two properties we refer the reader to page
69 in \cite{borkartxt}).
\end{itemize}
It is worth noting here that for every $x\in\mathbb{R}^d$, we do \it{not }\rm require that the associated Markov chain be aperiodic,
irreducible or possess a unique stationary distribution.
Assumption $(A3)$ is the standard step size assumption. The assumptions of a non-increasing sequence and square summability are required in
the analysis of the Markov noise component.
Assumption $(A4)$ is the same additive noise assumption as in Proposition 1.3, section 1.5 of \cite{benaim1}. The assumption ensures that the
contribution of the additive noise is eventually negligible. For various sufficient conditions under which $(A4)$ is satisfied we refer the
reader to \cite{benaim1}.
Assumption $(A5)$ ensures that the sample path of iterates of recursion \eqref{rec} remain bounded. This assumption is known as the stability
assumption and sufficient conditions for ensuring the same for the single-valued case (see \cite{borkarmeyn}, \cite{andvih}) and for
the set-valued case without Markov noise(see \cite{arunstab}) exist.
\section{Background}
\label{backgrd}
In this section we shall review some results from the theory of set valued maps and their integration. First we shall present a result on
approximation of upper semi-continuous set valued maps with continuous set valued maps. We will then use this approximation to obtain
single valued maps that represent these approximate continuous set valued maps thus enabling us to rewrite the stochastic recursive
inclusion, \eqref{rec}, as a standard stochastic approximation scheme with single valued maps.
Secondly, in order to define the set valued map associated with the limiting differential inclusion (also known as mean field) and prove
properties of the same we review definitions of measurability of set valued maps, its integral and some properties of the integral.
\input{Upper_semicontinuous_set_valued_maps_and_their_approximation}
\input{measurable_set_valued_maps_and_integration}
\section{Iterate-independent and controlled Markov noise cases}
\label{siacmnc}
In this section we shall introduce two variants of assumption $(A2)$ under which a similar analysis as presented in the previous section
can be carried out (assumptions $(A1),\ (A3)-(A5)$ are assumed to hold). The first case is of iterate-independent Markov noise i.e.,
where the transition probabilities do not depend on the iterate $x$ while second case is of controlled Markov noise i.e., the
so-called noise process is not Markov by itself, but its lack of Markov property comes through its dependence on some possibly imperfectly known
time varying process which is viewed as a control for analysis purposes. Throughout this section we assume that
$(A1),\ (A3)-(A5)$ hold.
\subsection{Iterate-independent Markov noise}
First present the iterate-independent Markov noise assumption.
\begin{itemize}
\item [$(A2)'$] $\left\{S_n\right\}_{n\geq0}$ is a sequence of $\mathcal{S}$ valued random variables on $\Omega$ such that,
for every $n\geq0$, for every $A\in \mathscr{B}(\mathcal{S})$, $\mathbb{P}(S_{n+1}\in A|S_m,X_m,m\leq n)=
\mathbb{P}(S_{n+1}\in A|S_n)=\Pi(S_n)(A)$ $a.s.$ where $\Pi:\mathcal{S}\rightarrow\mathcal{P}(\mathcal{S})$ is continuous.
\end{itemize}
Assumption $(A2)'$ and the fact that the state space of Markov noise is a compact metric space gives us that
the Markov chain defined by the transition kernel, $\Pi(\cdot)(\cdot)$, admits a stationary distribution ($\mu\in\mathcal{P}(\mathcal{S})$ is
stationary for the Markov chain defined by the transition kernel $\Pi(\cdot)(\cdot)$ if, for every $A\in\mathscr{B}(\mathcal{S})$,
$\mu(A)=\int_{\mathcal{S}}\Pi(s)(A)\mu(ds)$).
Let $D\subseteq\mathcal{P}(\mathcal{S})$ denote the set of stationary distributions of the Markov chain defined by the transition kernel
$\Pi(\cdot)(\cdot)$. It is easy to show that $D$ is a convex and compact subset of $\mathcal{P}(\mathcal{S})$. Define the set valued map,
$\hat{H}_{i}:\mathbb{R}^d\rightarrow\left\{subsets\ of\ \mathbb{R}^d\right\}$, such that, for every $x\in\mathbb{R}^d$,
\begin{equation*}
\hat{H}_{i}(x):=\cup_{\mu\in D}\int_{\mathcal{S}}H_x(s)\mu(ds),
\end{equation*}
where for every $x\in\mathbb{R}^d$, the set valued map $H_x$ is as defined in section \ref{ldi}. Similar to Lemma \ref{march2}, it can be
shown that the set valued map $\hat{H}$ defined above is a Marchaud map. By the same set of arguments as in section \ref{recanal}, we obtain
the following result.
\begin{theorem}\emph{[APT for iterate-independent Markov noise]}
Under assumptions $(A1),\ (A2)',\ (A3)-(A5)$, for almost every $\omega$, the linearly interpolated trajectory of recursion \eqref{rec}, is
an asymptotic pseudotrajectory for the flow of the DI,
\begin{equation}\label{di3}
\frac{dx}{dt}\in\hat{H}_{i}(x).
\end{equation}
\end{theorem}
As a consequence of the above theorem, a limit set theorem similar to Theorem \ref{ls} can be established which under assumptions $(A1),(A2)',
(A3)-(A5)$ characterizes the limit set of recursion \eqref{rec} in terms of the dynamics of DI\eqref{di3}.
\subsection{Controlled Markov noise}
Let $\left\{Z_n\right\}_{n\geq0}$ be a sequence of random variables(control sequence) on $\Omega$, taking values in a compact metric space,
$\mathcal{Z}$. The controlled Markov noise assumption is as follows.
\begin{itemize}
\item [$(A2)''$]$\left\{S_n\right\}_{n\geq0}$ is a sequence of $\mathcal{S}$ valued random variables on $\Omega$ such that,
for every $n\geq0$, for every $A\in \mathscr{B}(\mathcal{S})$, $\mathbb{P}(S_{n+1}\in A|S_m,Z_m,X_m,m\leq n)=
\mathbb{P}(S_{n+1}\in A|S_n,Z_n,X_n)=\Pi(X_n,S_n,Z_n)(A)$ a.s. with $\Pi:\mathcal{S}\rightarrow\mathcal{P}(\mathcal{S})$,
satisfying,
\begin{itemize}
\item [(i)] $\Pi$ is continuous.
\item [(ii)] $\left\{Z_n\right\}_{n\geq0}$ is a stationary randomized control, i.e., for every $n\geq0$, for every $A\subseteq\mathscr{B}(
\mathcal{Z}$), $\mathbb{P}(Z_n\in A|X_m,S_m,Z_{m-1},m\leq n)=\mathbb{P}(Z_n\in A|S_n)=\phi(S_n)(A)$ $a.s.$ with $\phi:\mathcal{S}
\rightarrow\mathcal{P}(Z)$, measurable.
\end{itemize}
Let $\Pi_{\phi}:\mathbb{R}^d\times\mathcal{S}\rightarrow\mathcal{P}(\mathcal{S})$ be such that for every $x\in\mathbb{R}^d$, for every
$s\in\mathcal{S}$, $\Pi_{\phi}(x,s)(ds')=\int_{\mathcal{Z}}\Pi(x,z,s)(ds')\phi(s)(dz)$.
\begin{itemize}
\item [(iii)] For every $x\in\mathbb{R}^d$ the Markov chain with state space $\mathcal{S}$ defined by the transition kernel,
$\Pi_{\phi}(x,\cdot)(\cdot)$ admits
a stationary distribution ($\mu\in\mathcal{P}(\mathcal{S})$ is stationary for the Markov chain defined by the transition kernel
$\Pi_{\phi}(x,\cdot)(\cdot)$ if, for every $A\in\mathscr{B}(\mathcal{S})$, $\mu(A)=\int_{\mathcal{S}}\Pi_{\phi}(x,s)(A)\mu(ds)$).
\end{itemize}
\end{itemize}
For every $x\in\mathbb{R}^d$, let $D_{\phi}(x)\subseteq\mathcal{P}(\mathcal{S})$ denote the set of stationary distributions of the Markov
chain defined by the transition kernel, $\Pi_{\phi}(x,\cdot)(\cdot)$. The convexity, compactness of $D_{\phi}(x)$ and the closed graph
property of the set valued map $x\rightarrow D_{\phi}(x)$ follow from arguments similar to those in page 69 of \cite{borkartxt}. Define the
set valued map, $\hat{H}_{\phi}:\mathbb{R}^d\rightarrow\left\{\text{subsets of }\mathbb{R}^d\right\}$ such that for every $x\in\mathbb{R}^d$,
\begin{equation*}
\hat{H}_{\phi}(x):=\cup_{\mu\in D_{\phi}(x)}\int_{\mathcal{S}}H_x(s)\mu(ds),
\end{equation*}
where for every $x\in\mathbb{R}^d$, the set valued map $H_x$ is as defined in section \ref{ldi}.
By a proof similar to Lemma \ref{march2}, one can show that the map $\hat{H}_{\phi}$ is a Marchaud map. Then by a similar set of arguments
as in section \ref{recanal}, one obtains the following result.
\begin{theorem}\emph{[APT for controlled Markov noise]}
Under assumptions $(A1),\ (A2)'',\ (A3)-(A5)$, for almost every $\omega$, the linearly interpolated trajectory of recursion \eqref{rec} is
an asymptotic pseudotrajectory for the flow of the DI,
\begin{equation}\label{di4}
\frac{dx}{dt}\in\hat{H}_{\phi}(x).
\end{equation}
\end{theorem}
As a consequence of the above theorem, a limit set theorem similar to Theorem \ref{ls} can be established which under assumptions $(A1),(A2)'',
(A3)-(A5)$ characterizes the limit set of recursion \eqref{rec} in terms of the dynamics of DI\eqref{di4}.
\section{Limiting differential inclusion and its properties}
\label{ldi}
In this section we shall define the limiting differential inclusion (or the mean field) whose flow the iterates of recursion \eqref{rec} are
expected to track. Analogous to the single valued case as in \cite{borkarmark}, the set valued map associated with the limiting differential
inclusion is nothing but the set valued map obtained by averaging the set valued map $H$ in equation \eqref{rec} w.r.t. the stationary
distributions of the Markov noise. The above shall be made precise in this section. Throughout this section we assume that $H$ denotes the
set valued map satisfying $(A1)$ and $\left\{H^{(l)}\right\}_{l\geq1}$, $\left\{h^{(l)}\right\}_{l\geq1}$ are as in Lemma \ref{ctem} and Lemma \ref{param}
respectively.
First we define slices of the set valued maps $H$ and $H^{(l)}$ for every $x\in\mathbb{R}^d$.
\begin{definition}
For every $x\in \mathbb{R}^d$, define,
\begin{itemize}
\item [(i)] $H_x:\mathcal{S}\rightarrow\left\{\text{subsets of }\mathbb{R}^d\right\}$ such that, for every $s\in\mathcal{S}$, $H_x(s):=H(x,s)$.
\item [(ii)] for every $l\geq1$, $H^{(l)}_x:\mathcal{S}\rightarrow\left\{\text{subsets of }\mathbb{R}^d\right\}$ such that, for every
$s\in\mathcal{S}$, $\\H^{(l)}_x(s):=H^{(l)}(x,s)$.
\item [(iii)] for every $l\geq1$, $h^{(l)}_x:\mathcal{S}\times U\rightarrow\mathbb{R}^d$ such that, for every
$(s,u)\in\mathcal{S}\times U$, $\\h^{(l)}_x(s,u):=h^{(l)}(x,s,u)$.
\end{itemize}
\end{definition}
The lemma below summarizes properties that the set valued maps $H_x$ and $H_x^{(l)}$ inherit from the set valued maps $H$ and $H^{(l)}$
respectively.
\begin{lemma}\label{msble}
For every $x\in \mathbb{R}^d$,
\begin{itemize}
\item [(i)] $H_x$ is a measurable set valued map and for every $s\in\mathcal{S}$, $H_x(s)$ is a convex and compact subset of
$\mathbb{R}^d$. Further there exists $C_x:=K(1+\parallel x\parallel)>0$, such that
$\\\sup_{s\in\mathcal{S}}\sup_{z\in H_x(s)}\parallel z\parallel\leq C_x$ (which means $H_x$ is bounded).
\item [(ii)] for every $l\geq1$, $H^{(l)}_x$ is a measurable set valued map and for every $s\in\mathcal{S}$, $H^{(l)}_x(s)$
is a convex and compact subset of $\mathbb{R}^d$. Further for every $l\geq1$, there exists
$C^{(l)}_x:=K^{(l)}(1+\parallel x\parallel)>0$, such that
$\sup_{s\in\mathcal{S}}\sup_{z\in H^{(l)}_x(s)}\parallel z\parallel\leq C^{(l)}_x$ (which means $H^{(l)}_x$ is bounded).
\item [(iii)] For any probability measure $\mu$ on $(\mathcal{S},\mathscr{B}(\mathcal{S}))$, all measurable selections of $H_x$
are $\mu$-integrable and hence $H_x$ is $\mu$-integrable.
\item [(iv)] For any probability measure $\mu$ on $(\mathcal{S},\mathscr{B}(\mathcal{S}))$, all measurable selections of $H^{(l)}_x$
are $\mu$-integrable and hence $H^{(l)}_x$ is $\mu$-integrable.
\item [(v)] for every $l\geq1$, $h^{(l)}_x$ is continuous, bounded by the constant $C^{(l)}_x$ where $C^{(l)}_x$ is as in part
$(ii)$ of this lemma and for every $s\in\mathcal{S}$, $h^{(l)}_x(s,U)=H^{(l)}_x(s)$.
\end{itemize}
\end{lemma}
\begin{proof} Fix some $x\in\mathbb{R}^d$.
\begin{itemize}
\item [(i)] For every $s\in\mathcal{S}$, the convexity and compactness of $H_x(s)$ follows from the definition of $H_x$ and
assumption $(A1)(i)$. Boundedness of $H_x$ follows from the definition of $H_x$ and assumption $(A1)(ii)$. From definition
of $H_x$ and assumption $(A1)(iii)$, we deduce that for every sequence $\left\{s_n\right\}_{n\geq1}$ converging to $s$ and for every sequence
$\left\{z_n\in H_x(s_n)\right\}_{n\geq1}$ converging to $z$, $z\in H_x(s)$ (which means that the graph of $H_x$ is closed).
Let $C$ be some closed subset of $\mathbb{R}^d$. Consider a sequence $\left\{s_n\right\}_{n\geq1}$ converging to $s$ such that for every
$n\geq1$, $s_n\in \left\{s\in \mathcal{S}:H_x(s)\cap C\neq \emptyset\right\}$. For every $n\geq1$, let $z_n\in H_x(s_n)\cap C$. Since $H_x$
is bounded, $\left\{z_n\right\}_{n\geq1}$ is a bounded sequence in $\mathbb{R}^d$ and hence has a convergent subsequence, say
$\left\{z_{n_k}\right\}_{k\geq1}$ converging to $z$. Since $C$ is closed and for every $k\geq1$, $z_{n_k}\in C$, we have $z\in C$.
Since the graph of $H_x$ is closed, we have, $z\in H_x(s)$. Hence $H_x(s)\cap C\neq\emptyset$. Therefore
$\left\{s\in \mathcal{S}:H_x(s)\cap C\neq \emptyset\right\}$ is closed and clearly belongs to $\mathscr{B}(\mathcal{S})$, which establishes the
measurability of $H_x$.
\item [(ii)] Fix some $l\geq1$. For every $s\in\mathcal{S}$, the convexity and compactness of $H^{(l)}_x(s)$ follows from the
definition of $H^{(l)}_x$ and Lemma \ref{ctem}$(i)$. Boundedness of $H^{(l)}_x$ follows from the definition of $H^{(l)}_x$ and
Lemma \ref{ctem}$(iii)$. Let $\left\{s_n\right\}_{n\geq1}$ be a sequence in $\mathcal{S}$ converging to $s$ and
$\left\{z_n\in H^{(l)}_x(s_n)\right\}_{n\geq1}$ converging to $z$. Using the continuity of $H^{(l)}$ and the definition of $H^{(l)}_x$,
we have, for every $\epsilon>0$, there exists $\delta>0$ such that
$\cup_{\left\{n:d(s_n,s)<\delta\right\}}H^{(l)}_x(s_n)\subseteq H^{(l)}_x(s)+\epsilon U$. Since $H^{(l)}_x(s)+\epsilon U$ is closed, we have,
$\overline{\cup_{\left\{n:d(s_n,s)<\delta\right\}}H^{(l)}_x(s_n)}\subseteq H^{(l)}_x(s)+\epsilon U$. Hence for every $\epsilon>0$,
$z\in H^{(l)}_x(s)+\epsilon U$. Since $H^{(l)}_x(s)$ is compact we have, $z\in H^{(l)}_x(s)$. Hence the graph of $H^{(l)}_x$ is
closed and by argument similar to the one in part $(i)$ of this proof we can prove the measurability of $H^{(l)}_x$.
\item [(iii)] Since $H_x$ is bounded, for every $f\in\mathscr{S}(H_x)$, we have
$\sup_{s\in\mathcal{S}}\parallel f(s)\parallel\leq C_x$. Hence $\int_{\mathcal{S}}\parallel f(s)\parallel\mu(ds)\leq C_x$.
\item [(iv)] Argument is same as in part $(iii)$ of this proof.
\item [(v)] Follows from definition of $h^{(l)}_x$ and Lemma \ref{param}.\qed
\end{itemize}
\end{proof}
Recall that $\mathcal{P}(\mathcal{S}\times U)$ denotes the set of probability measures on
$(\mathcal{S}\times U,\mathscr{B}(\mathcal{S})\otimes\mathscr{B}(U))$ with the Prohorov's topology.
For any probability measure $\nu\in\mathcal{P}(\mathcal{S}\times U)$, let $\nu_{\mathcal{S}}\in\mathcal{P}(\mathcal{S})$ denote the
image of $\nu$ under the projection, $\mathcal{S}\times U\rightarrow\mathcal{S}$ (i.e. for every $A\in\mathscr{B}(\mathcal{S})$,
$\nu_{\mathcal{S}}(A):=\int_{A\times U}\nu(ds,du)$). The next lemma provides a useful characterization of the integral of set valued maps
$H^{(l)}_x$ in terms of their parametrization, $h^{(l)}_x$.
\begin{lemma}\label{chint}
For every $x\in\mathbb{R}^d$, for every $l\geq1$, for every probability measure $\mu$ on $(\mathcal{S},\mathscr{B}(\mathcal{S}))$,
\begin{equation*}
\int_{\mathcal{S}}H^{(l)}_x(s)\mu(ds)=\left\{\int_{\mathcal{S}\times U}h^{(l)}_x(s,u)\nu(ds,du):
\nu\in \mathcal{P}(\mathcal{S}\times\ U),\ \nu_{\mathcal{S}}=\mu\right\}
\end{equation*}
\end{lemma}
\begin{proof}
Fix some $x\in\mathbb{R}^d$, $l\geq1$ and $\mu\in\mathcal{P}(\mathcal{S})$.
Let $z\in\int_{\mathcal{S}}H^{(l)}_x(s)\mu(ds)$. There exists $f\in\mathscr{S}(H^{(l)}_x)$ such that, $z=\int_{\mathcal{S}}f(s)\mu(ds)$.
Define a set valued map, $G:\mathcal{S}\rightarrow\left\{\text{subsets of }U\right\}$ such that, for every $s\in\mathcal{S}$,
$G(s):=\left\{u\in U: h^{(l)}(x,s,u)=f(s)\right\}$. Clearly for every $s\in\mathcal{S}$, $G(s)$ is non-empty and from the continuity of
$h^{(l)}$ we have that $G(s)$ is a closed subset of $U$. Moreover $G$ is a measurable set valued map (the proof of measurability has
been provided in appendix \ref{prf_msb}). Since $G$ is measurable, by Lemma \ref{msel}$(i)$ we have, $\mathscr{S}(G)\neq\emptyset$. For any
$g\in\mathscr{S}(G)$, we have, for every $s\in\mathcal{S}$, $h^{(l)}(x,s,g(s))=f(s)$. Define
$\hat{g}:\mathcal{S}\rightarrow\mathcal{S}\times U$ such that, for every $s\in\mathcal{S}$, $\hat{g}(s)=(s,g(s))$. Let
$\nu:=\mu\hat{g}^{-1}$ (push-forward measure). Clearly $\nu_{\mathcal{S}}=\mu$ and
$\int_{\mathcal{S}\times U}h^{(l)}(x,s,u)\nu(ds,du))=\int_{\mathcal{S}\times U}h^{(l)}(x,s,u)\mu\hat{g}^{-1}(ds,du))=
\int_{\mathcal{S}}h^{(l)}(x,\hat{g}(s))\mu(ds)=\int_{\mathcal{S}}h^{(l)}(x,s,g(s))\mu(ds)=
\int_{\mathcal{S}}f(s)\mu(ds)=z$.
Let $\nu\in\mathcal{P}(\mathcal{S}\times U)$ such that, $\nu_{\mathcal{S}}=\mu$. By Corollary 3.1.2 in \cite{borkarap}, there exists a
$\mu$-a.s. unique measurable map $q:\mathcal{S}\rightarrow\mathcal{P}(U)$ such that, $\nu(ds,du)=\mu(ds)q(s,du)$. By definition of
$h^{(l)}_x$ and Lemma \ref{param}, we have, for every $s\in\mathcal{S}$, $H^{(l)}_x(s)=h_x^{(l)}(s,U)=\left\{h_x^{(l)}(s,u):u\in U\right\}$ is
convex and compact subset of $\mathbb{R}^d$. Hence for every $s\in\mathcal{S}$, $\int_{U}h_x^{(l)}(s,u)q(s,du)\in H^{(l)}_x(s)$.
Let $f:\mathcal{S}\rightarrow\mathbb{R}^d$ be such that for every $s\in\mathcal{S}$, $f(s)=\int_{U}h_x^{(l)}(s,u)q(s,du)$. Then,
clearly $f$ is measurable and $f\in\mathscr{S}(H^{(l)}_x)$. Therefore, $\int_{\mathcal{S}\times U}h^{(l)}_x(s,u)\nu(ds,du)=
\int_{\mathcal{S}}\big{[}\int_{U}h^{(l)}_x(s,u)q(s,du)\big{]}\mu(ds)=\int_{\mathcal{S}}f(s)\mu(ds)\in
\int_{\mathcal{S}}H^{(l)}_x(s)\mu(ds)$.\qed
\end{proof}
Recall that for every $x\in\mathbb{R}^d$, $D(x)$ denotes the set of stationary distributions associated with the Markov chain whose
transition probability kernel is given by $\Pi(x,\cdot)(\cdot)$ as in assumption $(A2)$. We shall now define the set valued maps associated
with the limiting differential inclusion which is obtained by averaging the set valued map $H$ w.r.t. the stationary distributions of the
Markov noise. We shall perform a similar operation on set valued maps $H^{(l)}$ which approximate $H$.
\begin{definition}:
\begin{itemize}
\item [(i)] Define $\hat{H}:\mathbb{R}^d\rightarrow\left\{\text{subsets of }\mathbb{R}^d\right\}$ such that, for every $x\in\mathbb{R}^d$,
\begin{equation}\label{lim1}
\hat{H}(x):=\cup_{\mu\in D(x)}\int_{\mathcal{S}}H_x(s)\mu(ds).
\end{equation}
\item [(ii)] For every $l\geq1$, define $\hat{H}^{(l)}:\mathbb{R}^d\rightarrow\left\{\text{subsets of }\mathbb{R}^d\right\}$ such that, for every
$x\in\mathbb{R}^d$,
\begin{equation}\label{lim2}
\hat{H}^{(l)}(x):=\cup_{\mu\in D(x)}\int_{\mathcal{S}}H^{(l)}_x(s)\mu(ds).
\end{equation}
\end{itemize}
\end{definition}
For every $l\geq1$, the differential inclusion (DI) associated with the set valued map $\hat{H}^{(l)}$ is given by,
\begin{equation}\label{di1}
\frac{dx}{dt}\in \hat{H}^{(l)}(x).
\end{equation}
Similarly the differential inclusion (DI) associated with the set valued map $\hat{H}$ is given by,
\begin{equation}\label{di2}
\frac{dx}{dt}\in \hat{H}(x).
\end{equation}
A function $\bf{x}\rm:\mathbb{R}\rightarrow\mathbb{R}^d$ is said to be a \bf{solution }\rm of DI \eqref{di1} (or DI \eqref{di2}) with initial condition
$x_0\in\mathbb{R}^d$ if, $\bf{x}\rm$ is absolutely continuous, $\bf{x}\rm(0)=x_0$ and for a.e. $t\in\mathbb{R}$,
$\frac{d\bf{x}\rm(t)}{dt}\in\hat{H}^{(l)}(\bf{x}\rm(t))$ (or $\frac{d\bf{x}\rm(t)}{dt}\in\hat{H}(\bf{x}\rm(t))$).
For every $l\geq1$, for every initial condition $x_0\in\mathbb{R}^d$, let $\Sigma^{(l)}(x_0)$ denote the set of solutions to DI \eqref{di1}
with initial condition $x_0$, that is,
\begin{equation*}
\Sigma^{(l)}(x_0)=\left\{\bf{x}\rm:\mathbb{R}\rightarrow\mathbb{R}^d: \bf{x}\rm\ is\ a\ solution\ of\ DI\ \eqref{di1},\ \bf{x}\rm(0)=x_0\right\}.
\end{equation*}
Similarly, let $\Sigma(x_0)$ denote the set of solutions of DI \eqref{di2} with initial condition $x_0\in\mathbb{R}^d$. For any
$A\subseteq\mathbb{R}^d$, for every $l\geq1$, $\Sigma^{(l)}(A):=\cup_{x_0\in A}\Sigma^{(l)}(x_0)$ (similarly,
$\Sigma(A):=\cup_{x_0\in A}\Sigma(x_0)$).
The next lemma establishes some properties of the set valued map $\hat{H}^{(l)}$ which ensure the existence of solutions to DI \eqref{di1}.
\begin{lemma}\label{march1}
For every $l\geq1$, $\hat{H}^{(l)}$ satisfies the following,
\begin{itemize}
\item [(i)] for $K^{(l)}$ as in Lemma \ref{ctem}$(iii)$, we have, for every $x\in\mathbb{R}^d$,
$\\\sup_{z\in\hat{H}^{(l)}(x)}\parallel z\parallel\leq K^{(l)}(1+\parallel x\parallel)$.
\item [(ii)] for every $x\in\mathbb{R}^d$, $\hat{H}^{(l)}(x)$ is a non-empty, convex and compact subset of $\mathbb{R}^d$.
\item [(iii)] for every sequence $\left\{x_n\right\}_{n\geq1}$ in $\mathbb{R}^d$ converging to $x$, for every sequence
$\left\{z_n\in\hat{H}^{(l)}(x_n)\right\}_{n\geq1}$ converging to $z$, we have, $z\in\hat{H}^{(l)}(x)$.
\end{itemize}
\end{lemma}
\begin{proof}
Fix $l\geq1$.
\begin{itemize}
\item [(i)] For every $x\in\mathbb{R}^d$, from Lemma \ref{msble}$(ii)$, we have, for every $f\in\mathscr{S}(H^{(l)}_x)$, for every
$\mu\in D(x)$, $\parallel\int_{\mathcal{S}}f(s)\mu(ds)\parallel\leq\int_{\mathcal{S}}\parallel f(s)\parallel\mu(ds)\leq C^{(l)}_x=
K^{(l)}(1+\parallel x\parallel)$. Using the above and from the definition of $\hat{H}^{(l)}$ we have,
$\sup_{z\in \hat{H}^{(l)}(x)}\parallel z\parallel=\\\sup_{\mu\in D(x)}\sup_{f\in\mathscr{S}(H^{(l)}_x)}\parallel
\int_{\mathcal{S}}f(s)\mu(ds)\parallel\leq K^{(l)}(1+\parallel x\parallel)$.
\item [(ii)] For every $x\in\mathbb{R}^d$, $\hat{H}^{(l)}(x)$ is clearly non-empty. Let $z_1,z_2\in\hat{H}^{(l)}(x)$ and
$\theta\in(0,1)$. By definition of $\hat{H}^{(l)}$ and by Lemma \ref{chint}, we have that there exist $\nu^{1},\nu^{2}\in
\mathcal{P}(\mathcal{S}\times U)$ such that, for $i\in\left\{1,2\right\}$, $z_i=\int_{\mathcal{S}\times U}h^{(l)}_x(s,u)\nu^i(ds,du)$ and
$\nu^{i}_{\mathcal{S}}\in D(x)$. Hence,
\begin{align*}
\theta z_1+(1-\theta)z_2&=\theta\int_{\mathcal{S}\times U}h^{(l)}_x(s,u)\nu^1(ds,du)+
(1-\theta)\int_{\mathcal{S}\times U}h^{(l)}_x(s,u)\nu^2(ds,du)\\
&=\int_{\mathcal{S}\times U}h^{(l)}_x(s,u)(\theta\nu^1+(1-\theta)\nu^2)(ds,du),
\end{align*}
where, $\theta\nu^1+(1-\theta)\nu^2\in\mathcal{P}(\mathcal{S}\times U)$ such that, for every
$A\in\mathscr{B}(\mathcal{S})\otimes \mathscr{B}(U)$, $\\(\theta\nu^1+(1-\theta)\nu^2)(A)=\theta\nu^1(A)+(1-\theta)\nu^2(A)$. Further
$(\theta\nu^1+(1-\theta)\nu^2)_{\mathcal{S}}=\theta\nu^1_{\mathcal{S}}+(1-\theta)\nu^2_{\mathcal{S}}$ and it belongs to
$D(x)$, since for every $x\in\mathbb{R}^d$, $D(x)$ is convex.
By part $(i)$ of this proof, for every $x\in\mathbb{R}^d$, $\hat{H}^{(l)}(x)$ is bounded. Hence in order to show compactness it is enough
to prove that for every $x\in\mathbb{R}^d$, $\hat{H}^{(l)}(x)$ is closed. Let $\left\{z_n\right\}_{n\geq1}$ be a sequence in $\hat{H}^{(l)}(x)$
for some $x\in\mathbb{R}^d$, converging to $z$. By Lemma \ref{chint}, we know that for every $n\geq1$, there exists
$\nu^n\in\mathcal{P}(\mathcal{S}\times U)$ such that, $z_n=\int_{\mathcal{S}\times U}h^{(l)}_x(s,u)\nu^n(ds,du)$ and
$\nu^n_{\mathcal{S}}\in D(x)$. Since $\mathcal{S}\times U$ is a compact metric space, by Prohorov's theorem (see theorem 2.3.1 in
\cite{borkarap}), there exists a subsequence $\left\{n_k\right\}_{k\geq1}$ such that, $\left\{\nu^{n_k}\right\}_{k\geq1}$ converges to some
$\nu\in\mathcal{P}(\mathcal{S}\times U)$. Clearly $\left\{z_{n_k}\right\}_{k\geq1}$ converges to $z$ and $\left\{\nu^{n_k}_{\mathcal{S}}\right\}_{k\geq1}$
converges to $\nu_{\mathcal{S}}$ in $\mathcal{P}(\mathcal{S})$. Since for every $x\in\mathbb{R}^d$, $D(x)$ is a closed subset of
$\mathcal{P}(\mathcal{S})$, we have $\nu_{\mathcal{S}}\in D(x)$. By theorem 2.1.1$(ii)$ in \cite{borkarap} and continuity of
$h^{(l)}_x$ we have $\left\{\int_{\mathcal{S}\times U}h^{(l)}_x(s,u)\nu^{n_k}(ds,du)\right\}_{k\geq1}$ converges to
$\int_{\mathcal{S}\times U}h^{(l)}_x(s,u)\nu(ds,du)$ and hence $z=\int_{\mathcal{S}\times U}h^{(l)}_x(s,u)\nu(ds,du)$ with
$\nu_{\mathcal{S}}\in D(x)$. Therefore $z\in \hat{H}^{(l)}(x)$ and hence $\hat{H}^{(l)}(x)$ is closed.
\item [(iii)] Let $\left\{x_n\right\}_{n\geq1}$ be a sequence in $\mathbb{R}^d$ converging to $x$ and $\left\{z_n\in \hat{H}^{(l)}(x_n)\right\}_{n\geq1}$
converging to $z$. By Lemma \ref{chint}, we have, for every $n\geq1$, there exists $\nu^n\in\mathcal{P}(\mathcal{S}\times U)$,
such that $z_n=\int_{\mathcal{S}\times U}h_{x_n}^{(l)}(s,u)\nu^n(ds,du)$ and $\nu^n_{\mathcal{S}}\in D(x_n)$. Since
$\mathcal{S} \times U$ is a compact metric space, by Prohorov's theorem we have that, there exists a subsequence, $\left\{n_k\right\}_{k\geq1}$
such that, $\left\{\nu^{n_k}\right\}_{k\geq1}$ converges to some $\nu\in\mathcal{P}(\mathcal{S}\times U)$. Since $\left\{x_{n_k}\right\}_{k\geq1}$ converges
to $x$, we have $\left\{h^{(l)}_{x_{n_k}}\right\}_{k\geq1}$ converges uniformly to $h^{(l)}_x$ (for a proof see appendix \ref{prf_conv}). By uniform convergence
of $h^{(l)}_{x_{n_k}}$ to $h^{(l)}_x$ and Theorem 2.1.1$(ii)$ in \cite{borkarap} we have that the sequence
$\left\{\int_{\mathcal{S}\times U}h^{(l)}_{x_{n_k}}(s,u)\nu^{n_k}(ds,du)\right\}_{k\geq1}$ converges to
$\int_{\mathcal{S}\times U}h^{(l)}_x(s,u)\nu(ds,du)$. Hence $z=\int_{\mathcal{S}\times U}h^{(l)}_x(s,u)\nu(ds,du)$. Clearly
$\left\{\nu^{n_k}_{\mathcal{S}}\right\}_{k\geq1}$ converges to $\nu_{\mathcal{S}}$ and by closed graph property of the map $x\rightarrow D(x)$, we
have $\nu_{\mathcal{S}}\in D(x)$. Hence $z=\int_{\mathcal{S}\times U}h^{(l)}_x(s,u)\nu(ds,du)$ and $\nu_{\mathcal{S}}\in D(x)$.
Therefore by Lemma \ref{chint} and definition of $\hat{H}^{(l)}$, we have, $z\in \hat{H}^{(l)}(x)$.\qed
\end{itemize}
\end{proof}
The set valued map satisfying properties stated in Lemma \ref{march1}, is called a Marchaud map (see page 62 of \cite{aubinvt}).
By Lemma \ref{march1}, we know that for every $l\geq1$, the set valued maps $\hat{H}^{(l)}$ are Marchaud maps. For such maps their
associated DIs are known to admit at least one solution through every initial condition (see \cite{aubindi}, chapter 2.1 or
\cite{benaim1}, section 1.2). Thus for every $l\geq1$, for every initial condition $x_0\in\mathbb{R}^d$, $\Sigma^{(l)}(x_0)\neq\emptyset$.
The next lemma establishes relation between $\hat{H}$ and $\hat{H}^{(l)}$.
\begin{lemma}\label{approx1}
For every $x\in\mathbb{R}^d$,
\begin{itemize}
\item [(i)] for every $l\geq1$, $\hat{H}(x)\subseteq\hat{H}^{(l+1)}(x)\subseteq\hat{H}^{(l)}(x)$.
\item [(ii)]$\cup_{\mu\in D(x)}\cap_{l\geq1}\int_{\mathcal{S}}H_x^{(l)}(s)\mu(ds)=\cap_{l\geq1}\hat{H}^{(l)}(x)$.
\item [(iii)] $\hat{H}(x)=\cap_{l\geq1}\hat{H}^{(l)}(x)$.
\end{itemize}
\end{lemma}
\begin{proof}
Fix $x\in\mathbb{R}^d$.
\begin{itemize}
\item [(i)] Fix $l\geq1$.
By Lemma \ref{ctem}$(ii)$, for every $(x,s)\in\mathbb{R}^d\times\mathcal{S}$, $H(x,s)\subseteq H^{(l+1)}(x,s)\subseteq
H^{(l)}(x,s)$. Hence, for every $x\in\mathbb{R}^d$, by definition of $H_x$ and $H^{(l)}_x$ we have, for every $s\in\mathcal{S}$,
$H_x(s)\subseteq H^{(l+1)}_x(s)\subseteq H^{(l)}_x(s)$. Therefore, $\mathscr{S}(H_x)\subseteq\mathscr{S}(H_x^{(l+1)})\subseteq
\mathscr{S}(H^{(l)}_x)$. From the above we obtain that, for every $\mu\in D(x)$, $\int_{\mathcal{S}}H_x(s)\mu(ds)\subseteq
\int_{\mathcal{S}}H_x^{(l+1)}(s)\mu(ds)\subseteq\int_{\mathcal{S}}H^{(l)}_x(s)\mu(ds)$ which leads to the desired conclusion.
\item [(ii)] Clearly by definition of the set valued map $\hat{H}^{(l)}$ we have that
$\cup_{\mu\in D(x)}\cap_{l\geq1}\int_{\mathcal{S}}H_x^{(l)}(s)\mu(ds)\subseteq\cap_{l\geq1}\hat{H}^{(l)}(x)$. Let
$z\in\cap_{l\geq1}\hat{H}^{(l)}(x)$. Then for every $l\geq1$, there exists $\mu^{(l)}\in D(x)$ such that $z\in\int_{\mathcal{S}}
H^{(l)}_x(s)\mu^{(l)}(ds)$. By Prohorov's theorem the sequence of probability measures $\left\{\mu^{(l)}\right\}_{l\geq1}$ is relatively
compact in $\mathcal{P}(\mathcal{S})$ and hence has a limit point, say $\mu$. Let $\left\{\mu^{(l_k)}\right\}_{k\geq1}$ denote a subsequence
with $\mu$ as its limit and since $D(x)$ is a closed subset of $\mathcal{P}(\mathcal{S})$, we have that $\mu\in D(x)$. For every
$l\geq1$, for every $k$ such that $l_k\geq l$, we have that $\mathscr{S}(H_x^{(l_k)})\subseteq\mathscr{S}(H_x^{(l)})$. Thus for
every $l\geq1$, for every $k$ such that $l_k\geq l$, we have that $z\in\int_{\mathcal{S}}H^{(l)}_x(s)\mu^{(l_k)}(ds)$. By Lemma
\ref{chint}, we have that for every $l\geq1$, for every $k$ such that $l_k\geq l$, there exists $\nu^{(l,l_k)}\in
\mathcal{P}(\mathcal{S}\times U)$ with $\nu^{(l,l_k)}_{\mathcal{S}}=\mu^{(l_k)}$ and $z=\int_{\mathcal{S}\times U}h^{(l)}(x,s,u)
\nu^{(l,l_k)}(ds,du)$.
Fix $l\geq1$. Since $\mathcal{S}\times U$ is a compact metric space, by Prohorov's theorem we have that the
sequence $\left\{\nu^{(l,l_k)}\right\}_{k\geq1}$ is relatively compact in $\mathcal{P}(\mathcal{S}\times U)$. Let $\nu^{(l)}$ denote a limit
point of the above sequence and its clear that $\nu^{(l)}_{\mathcal{S}}=\mu$. Then clearly $z=\int_{\mathcal{S}\times U}
h^{(l)}(x,s,u)\nu^{(l)}(ds,du)$ and hence by Lemma \ref{chint}, $z\in\int_{\mathcal{S}}H^{(l)}_x(s)\mu(ds)$.
Since $l\geq1$ is arbitrary, we get that for $\mu\in D(x)$, for every $l\geq1$, $z\in\int_{\mathcal{S}}H^{(l)}_x(s)\mu(ds)$.
Thus $z\in\cup_{\mu\in D(x)}\cap_{l\geq1}\int_{\mathcal{S}}H_x^{(l)}(s)\mu(ds)$ and hence $\cap_{l\geq1}\hat{H}^{(l)}(x)\subseteq
\cup_{\mu\in D(x)}\cap_{l\geq1}\int_{\mathcal{S}}H_x^{(l)}(s)\mu(ds)$.
\item [(iii)] By part$(i)$ of this lemma we have, for every $x\in\mathbb{R}^d$, $\hat{H}(x)\subseteq\cap_{l\geq1}\hat{H}^{(l)}(x)$.
Therefore its enough to show that for every $x\in\mathbb{R}^d$, $\cap_{l\geq1}\hat{H}^{(l)}(x)\subseteq\hat{H}(x)$. Fix
$x\in\mathbb{R}^d$ and $\mu\in D(x)$. Let $z\in\cap_{l\geq1}\int_{\mathcal{S}}H^{(l)}_x(s)\mu(ds)$. For every $l\geq1$, there
exists $f^{(l)}\in\mathscr{S}(H^{(l)}_x)$ such that $z=\int_{\mathcal{S}}f^{(l)}(s)\mu(ds)$. Let
$d(z,\int_{\mathcal{S}}H_x(s)\mu(ds)):=\inf\left\{\parallel z-y\parallel: y\in\int_{\mathcal{S}}H_x(s)\mu(ds)\right\}$. Then, for every
$l\geq1$,
\begin{align*}
d(z,\int_{\mathcal{S}}H_x(s)\mu(ds))&=\inf_{f\in\mathscr{S}(H_x)}\parallel\int_{\mathcal{S}}(f^{(l)}(s)-f(s))\mu(ds)\parallel\\
&\leq\inf_{f\in\mathscr{S}(H_x)}\int_{\mathcal{S}}\parallel f^{(l)}(s)-f(s)\parallel\mu(ds)\\
&=\int_{\mathcal{S}}\inf_{f\in\mathscr{S}(H_x)}\parallel f^{(l)}(s)-f(s)\parallel\mu(ds),
\end{align*}
where the last equality follows from Lemma 1.3.12 in \cite{shoumei}. By Lemma \ref{msble}$(i)$, $H_x$ is a measurable set valued
map and hence by Lemma \ref{msel}$(ii)$, $H_x$ admits a Castaing representation, $\left\{f_n\right\}_{n\geq1}$. Hence for every
$s\in\mathcal{S}$, $d(f^{(l)}(s),H_x(s))\leq\inf_{f\in\mathscr{S}(H_x)}\parallel f^{(l)}(s)-f(s)\parallel\leq
\inf_{n\geq1}\parallel f^{(l)}(s)-f_n(s)\parallel=d(f^{(l)}(s),H_x(s))$ where $s\rightarrow d(f^{(l)}(s),H_x(s))$ is as in Lemma
\ref{mtool}. Therefore, for every $l\geq1$,
\begin{equation*}
d(z,\int_{\mathcal{S}}H_x(s)\mu(ds))\leq\int_{\mathcal{S}}d(f^{(l)}(s),H_x(s))\mu(ds).
\end{equation*}
By observation (b) stated after Lemma \ref{ctem}, we have that, for every $s\in\mathcal{S}$,
$\\\lim_{l\to\infty}d(f^{(l)}(s),H_x(s))=0$. Further by observation (a) stated after Lemma \ref{ctem}, Lemma \ref{msble}$(i)-(ii)$,
we have that, for every $s\in\mathcal{S}$, $d(f^{(l)}(s),H_x(s))\leq(\tilde{K}+K)(1+\parallel x\parallel)$. Thus, by
bounded convergence theorem we get, $\lim_{l\to\infty}\int_{\mathcal{S}}d(f^{(l)}(s),H_x(s))\mu(ds)=0$. Hence,
$d(z,\int_{\mathcal{S}}H_x(s)\mu(ds))=0$. By Lemma \ref{cldint}, we have that $\int_{\mathcal{S}}H_x(s)\mu(ds)$ is closed.
Therefore $z\in\int_{\mathcal{S}}H_x(s)\mu(ds)$.
From the arguments in the previous paragraph, we have that for every $x\in\mathbb{R}^d$, for every $\mu\in D(x)$,
$\cap_{l\geq1}\int_{\mathcal{S}}H^{(l)}_x(s)\mu(ds)\subseteq\int_{\mathcal{S}}H_x(s)\mu(ds)$. Thus, for every $x\in\mathbb{R}^d$,
$\\\cup_{\mu\in D(x)}\cap_{l\geq1}\int_{\mathcal{S}}H^{(l)}_x(s)\mu(ds)\subseteq\cup_{\mu\in D(x)}\int_{\mathcal{S}}H_x(s)\mu(ds)$.
Therefore, by part $(ii)$ of this lemma we have that for every $x\in\mathbb{R}^d$,
$\cap_{l\geq1}\hat{H}^{(l)}(x)\subseteq\hat{H}(x)$.\qed
\end{itemize}
\end{proof}
In the next lemma we show that the set valued map $\hat{H}$ is a Marchaud map.
\begin{lemma}\label{march2}
The set valued map $\hat{H}$ satisfies the following:
\begin{itemize}
\item [(i)] For $K>0$ as in Lemma \ref{msble}$(i)$, for every $x\in\mathbb{R}^d$,
$\sup_{z\in\hat{H}(x)}\parallel z\parallel\leq K(1+\parallel x\parallel)$.
\item [(ii)] For every $x\in\mathbb{R}^d$, $\hat{H}(x)$ is a non-empty, convex and compact subset of $\mathbb{R}^d$.
\item [(iii)] For every sequence $\left\{x_n\right\}_{n\geq1}$ in $\mathbb{R}^d$ converging to $x$, for every sequence
$\left\{z_n\in\hat{H}(x_n)\right\}_{n\geq1}$ converging to $z$, we have, $z\in\hat{H}(x)$.
\end{itemize}
\end{lemma}
\begin{proof}:
\begin{itemize}
\item [(i)] For every $x\in\mathbb{R}^d$, from Lemma \ref{msble}$(i)$, we have, for every $f\in\mathscr{S}(H_x)$, for every
$\mu\in D(x)$, $\parallel\int_{\mathcal{S}}f(s)\mu(ds)\parallel\leq\int_{\mathcal{S}}\parallel f(s)\parallel\mu(ds)\leq C_x=
K(1+\parallel x\parallel)$. Using the above and from the definition of $\hat{H}$ we have,
$\sup_{z\in \hat{H}(x)}\parallel z\parallel=\sup_{\mu\in D(x)}\sup_{f\in\mathscr{S}(H_x)}\parallel
\int_{\mathcal{S}}f(s)\mu(ds)\parallel\\\leq K(1+\parallel x\parallel)$.
\item [(ii)] For every $x\in\mathbb{R}^d$, by Lemma \ref{msble}$(iii)$, for every $\mu\in D(x)$, $H_x$ is $\mu$-integrable. Hence
$\hat{H}_x$ is non-empty. By Lemma \ref{march1}$(ii)$ and Lemma \ref{approx1}$(iii)$, we have that, $\hat{H}(x)$ is a convex and
compact subset of $\mathbb{R}^d$.
\item [(iii)] Let $\left\{x_n\right\}_{n\geq1}$ be a sequence in $\mathbb{R}^d$, converging to $x$ and $\left\{z_n\in \hat{H}(x_n)\right\}_{n\geq1}$
converging to $z$. By Lemma \ref{approx1}$(i)$, for every $l\geq1$, for every $n\geq1$, $z_n\in\hat{H}^{(l)}(x_n)$. Hence by Lemma
\ref{march1}$(iii)$, we have that, for every $l\geq1$, $z\in\hat{H}^{(l)}(x)$. Therefore by Lemma \ref{approx1}$(iii)$, we have,
$z\in\hat{H}(x)$.\qed
\end{itemize}
\end{proof}
By Lemma \ref{march2}, we know that the set valued map $\hat{H}$ is a Marchaud map. Hence for every $x_0\in\mathbb{R}^d$,
$\Sigma(x_0)\neq\emptyset$.
By definition, $\Sigma^{(l)}(x_0)$ and $\Sigma(x_0)$ are subsets of $\mathcal{C}(\mathbb{R},\mathbb{R}^d)$, the space of all $\mathbb{R}^d$
valued continuous functions on $\mathbb{R}$. $\mathcal{C}(\mathbb{R},\mathbb{R}^d)$ is a complete metric space for distance $\bf{D}\rm$,
defined by,
\begin{equation*}
\bf{D}\rm(\bf{x}\rm,\bf{z}\rm):=\sum_{k=1}^{\infty}\frac{1}{2^k}\min(\parallel \bf{x}\rm-\bf{z}\rm\parallel_{[-k,k]},1),
\end{equation*}
where $\parallel\cdot\parallel_{[-k,k]}$ denotes the sup norm on $\mathcal{C}([-k,k],\mathbb{R}^d)$.
The next lemma summarizes some properties of the solutions of DI \eqref{di1},\eqref{di2} and the relation between them.
\begin{lemma}\hfill
\begin{itemize}
\item [(i)] For every $A\subseteq\mathbb{R}^d$ compact, $\Sigma(A)$ and for every $l\geq1$, $\Sigma^{(l)}(A)$ are compact subsets of
$\mathcal{C}(\mathbb{R},\mathbb{R}^d)$.
\item [(ii)] For every $x_0\in\mathbb{R}^d$, for every $l\geq1$, $\Sigma(x_0)\subseteq\Sigma^{(l+1)}(x_0)\subseteq\Sigma^{(l)}(x_0)$.
\item [(iii)] For every $x_0\in\mathbb{R}^d$, $\Sigma(x_0)=\cap_{l\geq1}\Sigma^{(l)}(x_0)$.
\end{itemize}
\end{lemma}
\begin{proof}:
\begin{itemize}
\item [(i)] Follows from Lemma 3.1 in \cite{benaim1}.
\item [(ii)] Fix some $x_0\in\mathbb{R}^d$ and $l\geq1$. Let $\bf{x}\rm\in\Sigma(x_0)$. Then for $a.e.\ t\in\mathbb{R}$,
$\frac{d\bf{x}\rm(t)}{dt}\in\hat{H}(\bf{x}\rm(t))$. By Lemma \ref{approx1}$(i)$, we have, for every $t\geq0$,
$\hat{H}(\bf{x}\rm(t))\subseteq\hat{H}^{(l)}(\bf{x}\rm(t))$. Therefore, for $a.e.\ t\in\mathbb{R}$,
$\frac{d\bf{x}\rm(t)}{dt}\in\hat{H}^{(l)}(\bf{x}\rm(t))$. Hence, $\bf{x}\rm\in\Sigma^{(l)}(x_0)$. A similar argument gives us that,
$\Sigma^{(l+1)}(x_0)\subseteq\Sigma^{(l)}(x_0)$.
\item [(iii)] By part $(ii)$ of this lemma we have that, for every $x_0\in\mathbb{R}^d$,
$\Sigma(x_0)\subseteq\cap_{l\geq1}\Sigma^{(l)}(x_0)$. Therefore, it is enough to show that, for every $x_0\in\mathbb{R}^d$,
$\cap_{l\geq1}\Sigma^{(l)}(x_0)\subseteq\Sigma(x_0)$. Let $\bf{x}\rm\in\cap_{l\geq1}\Sigma^{(l)}(x_0)$ for some $x_0\in\mathbb{R}^d$.
Then for every $l\geq1$, for $a.e.\ t\in\mathbb{R}$, $\frac{d\bf{x}\rm(t)}{dt}\in\hat{H}^{(l)}(\bf{x}\rm(t))$. Thus for
$a.e.\ t\in\mathbb{R}$, $\frac{d\bf{x}\rm(t)}{dt}\in\cap_{l\geq1}\hat{H}^{(l)}(\bf{x}\rm(t))$. By Lemma \ref{approx1}$(iii)$, we have
that, for $a.e.\ t\in\mathbb{R}$, $\frac{d\bf{x}\rm(t)}{dt}\in\hat{H}(\bf{x}\rm(t))$. Hence $\bf{x}\rm\in\Sigma(x_0)$.\qed
\end{itemize}
\end{proof}
\subsection{Measurable set valued maps and integration}
Let $(\mathcal{Y},\mathscr{F}_{\mathcal{Y}})$ denote a measurable space and $F:\mathcal{Y}\rightarrow\left\{\text{subsets of }\mathbb{R}^d\right\}$
be a set valued map such that, for every $y\in\mathcal{Y}$, $F(y)$ is a non-empty closed subset of $\mathbb{R}^d$. Throughout this
section $F$ refers to the set valued map as defined above.
\begin{definition}
\emph{[measurable set-valued map]} A set valued map $F$ is measurable if for every $C\subseteq \mathbb{R}^d$, closed,
\begin{equation*}
F^{-1}(C):=\left\{y\in\mathcal{Y}:F(y)\cap C\neq\emptyset\right\}\in \mathscr{F}_{\mathcal{Y}}.
\end{equation*}
\end{definition}
We refer the reader to Theorem 1.2.3 in \cite{shoumei} for other notions of measurability and their relation to the definition above.
\begin{definition}
\emph{[measurable selection]} A function $f:\mathcal{Y}\rightarrow\mathbb{R}^d$ is a measurable selection of a set valued map $F$ if
$f$ is measurable and for every $y\in\mathcal{Y}$, $f(y)\in F(y)$.
\end{definition}
For a set valued map $F$ let, $\mathscr{S}(F)$ denote the set of all measurable selections. The next lemma summarizes some standard
results about measurable set valued maps and their measurable selections.
\begin{lemma}\label{msel}
For any measurable set valued map $F$,
\begin{itemize}
\item [(i)] $\mathscr{S}(F)\neq \emptyset$.
\item [(ii)] \it{(Castaing representation) }\rm there exists $\left\{f_n\right\}_{n\geq1}\subseteq\mathscr{S}(F)$ such that, for every
$y\in\mathcal{Y}$, $F(y)=cl(\left\{f_n(y)\right\}_{n\geq1})$, where $cl(\cdot)$ denotes the closure of a set.
\end{itemize}
\end{lemma}
We refer the reader to Theorem 1.2.6 and Theorem 1.2.7 in \cite{shoumei} for the proofs of Lemma \ref{msel}$(i)$ and $(ii)$
respectively.
\begin{definition}
\emph{[$\mu$-integrable set-valued map]}
Let $\mu$ be a probability measure on $(\mathcal{Y},\mathscr{F}_{\mathcal{Y}})$. A measurable set valued map $F$ is said to be
$\mu$-integrable if, there exists $f\in\mathscr{S}(F)$ which is $\mu$-integrable.
\end{definition}
\begin{definition}
\emph{[Aumann's integral]} Let $\mu$ be a probability measure on $(\mathcal{Y},\mathscr{F}_{\mathcal{Y}})$. The integral of an
integrable set valued map $F$ is defined as,
\begin{equation*}
\int_{\mathcal{Y}} F(y)\mu(dy):=\left\{\int_{\mathcal{Y}} f(y)\mu(dy):\ f\in \mathscr{S}(F),\ f\ is\ \mu-integrable\right\}.
\end{equation*}
\end{definition}
The next lemma states a useful result on the properties of the integral of a set valued map which is convex and compact set valued.
\begin{lemma}\label{cldint}
Let $\mu$ be a probability measure on $(\mathcal{Y},\mathscr{F}_{\mathcal{Y}})$ and $F$ a $\mu$-integrable set valued map such that,
for every $y\in\mathcal{Y}$, $F(y)$ is convex and compact. Then, $\int_{\mathcal{Y}} F(y)\mu(dy)$ is a convex and closed subset of $\mathbb{R}^d$.
\end{lemma}
For a proof of the above lemma we refer the reader to Theorem 2.2.2 in \cite{shoumei}. The next lemma is a useful tool to prove
measurability of set valued maps that arise later in this paper.
\begin{lemma}\label{mtool}
Let $g:\mathcal{Y}\rightarrow\mathbb{R}^d$ be a measurable function and $F$ a measurable set valued map. Then the map
$y\rightarrow d(g(y),F(y))$ where $d(g(y),F(y)):=\inf_{z\in F(y)}\parallel g(y)-z \parallel$, is measurable.
\end{lemma}
\begin{proof}
Since $F$ is measurable, there exists $\left\{f_n\right\}_{n\geq1}$ a Castaing representation of $F$. Clearly for every $n\geq1$, the map
$y\rightarrow\parallel g(y)-f_n(y)\parallel$ is measurable. Hence, the map $y\rightarrow\inf_{n\geq1}\parallel g(y)-f_n(y)\parallel$
is measurable. Since for every $y\in\mathcal{Y}$, $cl(\left\{f_n(y)\right\}_{n\geq1})=F(y)$ we have,
$\inf_{n\geq1}\parallel g(y)-f_n(y)\parallel=d(g(y),F(y))$. Therefore the map $y\rightarrow d(g(y),F(y))$ is measurable.\qed
\end{proof}
\section{Recursion analysis}
\label{recanal}
Before we present the analysis of recursion \eqref{rec} we begin with some preliminaries in the next subsection. Later we present the main
result of the paper followed by the limit set theorem which characterizes the limit set of the recursion in terms of the dynamics of the
limiting DI. Throughout this section we shall assume that assumptions $(A1)-(A5)$ are satisfied.
\subsection{Preliminaries}\label{prelim}
Define $t(0):=0$ and for every $n\geq1$, $t(n):=\sum_{k=0}^{n-1}a(k)$. Define the stochastic process
$\bar{X}:\Omega\times [0,\infty)\rightarrow\mathbb{R}^d$ such that, for every $(\omega,t)\in\Omega\times [0,\infty)$,
\begin{equation*}
\bar{X}(\omega,t):=\left(\frac{t-t(n)}{t(n+1)-t(n)}\right)X_{n+1}(\omega)+\left(\frac{t(n+1)-t}{t(n+1)-t(n)}\right)X_{n}(\omega),
\end{equation*}
where $n$ is such that $t\in[t(n),t(n+1))$.
Recall that recursion \eqref{rec} is given by,
\begin{equation*}
X_{n+1}-X_{n}-a(n)M_{n+1}\in H(X_n,S_n),
\end{equation*}
for every $n\geq0$. By Lemma \ref{ctem}, we have that, for every $l\geq1$, for every $n\geq0$, $H(X_n,S_n)\subseteq H^{(l)}(X_n,S_n)$.
Therefore, for every $l\geq1$, recursion \eqref{rec} can be written as,
\begin{equation}\label{rec1}
X_{n+1}-X_{n}-a(n)M_{n+1}\in H^{(l)}(X_n,S_n),
\end{equation}
for every $n\geq0$. By Lemma \ref{param}, we know that for every $l\geq1$, the set valued map $H^{(l)}$, admits a parametrization. The
next lemma allows us write the recursion in terms of the parametrization of $H^{(l)}$.
\begin{lemma}
For every $l\geq1$, for every $n\geq0$, there exists a $U$-valued random variable on $\Omega$, say $U^{(l)}_n$, such that,
\begin{equation*}
X_{n+1}-X_{n}-a(n)M_{n+1}= h^{(l)}(X_n,S_n,U^{(l)}_n),
\end{equation*}
where $h^{(l)}$ is as in Lemma \ref{param}.
\end{lemma}
\begin{proof}Fix $l\geq1$ and $n\geq0$. Define a set valued map $G:\Omega\rightarrow\left\{\text{subsets of }U\right\}$, such that, for every
$\omega\in\Omega$, $G(\omega):=\left\{u\in U: h^{(l)}(X_n(\omega),S_n(\omega),u)=X_{n+1}(\omega)-X_n(\omega)-a(n)M_{n+1}(\omega)\right\}$. By
Lemma \ref{param} and by $\eqref{rec1}$, we have that for every $\omega\in\Omega$, $G(\omega)\neq\emptyset$. By continuity of
$h^{(l)}(\cdot)$, we have that for every $\omega\in\Omega$, $G(\omega)$ is closed. For any $C\subseteq U$, closed,
$G^{-1}(C)=\left\{\omega\in\Omega:d(X_{n+1}(\omega)-X_n(\omega)-a(n)M_{n+1}(\omega),h^{(l)}(X_n(\omega),S_n(\omega),C))=0\right\}$. Clearly the set
valued map $\omega\rightarrow h^{(l)}(X_n(\omega),S_n(\omega),C)$ is measurable and hence by Lemma \ref{mtool}, the map
$\omega\rightarrow d(X_{n+1}(\omega)-X_n(\omega)-a(n)M_{n+1}(\omega),h^{(l)}(X_n(\omega),S_n(\omega),C))$ is measurable. Thus
$G^{-1}(C)\in\mathscr{F}$. Therefore by Lemma \ref{msel}, $\mathscr{S}(G)\neq\emptyset$. Set $U^{(l)}_n$ to be any measurable selection
of $G$.\qed
\end{proof}
For every $l\geq1$, define $\Gamma^{(l)}:\Omega\times[0,\infty)\rightarrow\mathcal{P}(\mathcal{S}\times U)$, such that, for every
$(\omega,t)\in\Omega\times[0,\infty)$,
\begin{equation}\label{diracm}
\Gamma^{(l)}(\omega,t)=\delta_{S_n(\omega)}\otimes\delta_{U^{(l)}_n(\omega)},
\end{equation}
where $n$ is such that $t\in[t(n),t(n+1))$, $\delta_{S_n(\omega)}\in\mathcal{P}(\mathcal{S})$ denotes the Dirac measure in
$\mathcal{P}(\mathcal{S})$(i.e. for any $A\in\mathscr{B}(\mathcal{S})$, $\delta_{S_n(\omega)}(A)=1$ if $S_n(\omega)\in A$ or $0$ otherwise)
and $\delta_{U^{(l)}_n(\omega)}\in\mathcal{P}(U)$ denotes the Dirac measure in $\mathcal{P}(U)$.
For every $l\geq1$, for every $\nu\in\mathcal{P}(\mathcal{S}\times U)$, define $\tilde{h}^{(l)}_{\nu}:\mathbb{R}^d\rightarrow\mathbb{R}^d$,
such that for every $x\in\mathbb{R}^d$, $\tilde{h}^{(l)}_{\nu}(x):=\int_{\mathcal{S}\times U}h^{(l)}(x,s,u)\nu(ds,du)$ where
$h^{(l)}(\cdot)$ are as in Lemma \ref{param}. The next lemma provides an equicontinuity result used later.
\begin{lemma}\label{paramuse}
For every $l\geq1$, for every $r>0$, $\left\{\tilde{h}^{(l)}_{\nu}(\cdot)|_{rU}:\nu\in\mathcal{P}(\mathcal{S}\times U)\right\}$ is an
equicontinuous family where $\tilde{h}^{(l)}_{\nu}(\cdot)|_{rU}$ denotes the restriction of $\tilde{h}^{(l)}_\nu(\cdot)$ to the closed
ball of radius $r$.
\end{lemma}
\begin{proof}
Fix $l\geq1$ and $r>0$. Since $rU\times \mathcal{S}\times U$ is compact, by Heine-Cantor theorem,
$h^{(l)}(\cdot)|_{rU\times \mathcal{S}\times U}$ is uniformly continuous. Fix $\epsilon>0$. Then, there exits $\delta>0$
(depending on $l$, $r$ and $\epsilon$), such that, for every $x,x'\in rU$, for every $s,s'\in\mathcal{S}$, for every $u,u'\in U$
satisfying, $\parallel x-x'\parallel<\delta$, $d_{\mathcal{S}}(s,s')<\delta$, $\parallel u-u'\parallel<\delta$, we have,
$\parallel h^{(l)}(x,s,u)-h^{(l)}(x',s',u')\parallel<\epsilon$. Therefore for $\delta$ as above, for $x,x'\in rU$ satisfying
$\parallel x-x'\parallel<\delta$, we have that for every $\nu\in\mathcal{P}(\mathcal{S}\times U)$, $\parallel \tilde{h}^{(l)}_{\nu}(x)-
\tilde{h}^{(l)}_{\nu}(x')\parallel=\parallel \int_{\mathcal{S}\times U}(h^{(l)}(x,s,u)-h^{(l)}(x',s,u))\nu(ds,du)\parallel\leq
\\\int_{\mathcal{S}\times U}\parallel h^{(l)}(x,s,u)-h^{(l)}(x',s,u)\parallel\nu(ds,du)<\int_{\mathcal{S}\times U}\epsilon\nu(ds,du)=
\epsilon$.\qed
\end{proof}
For every $l\geq1$, define the stochastic process, $Y^{(l)}:\Omega\times[0,\infty)\rightarrow\mathbb{R}^d$, such that, for every
$(\omega,t)\in\Omega\times[0,\infty)$, $Y^{(l)}(\omega,t):=h^{(l)}(X_n(\omega),S_n(\omega),U^{(l)}_n(\omega))$, where $n$ is such that
$t\in[t(n),t(n+1))$.
In what follows, in the next two subsections, most of the arguments are sample path wise. We omit $\omega$ from our notation and use lower
case letters to denote the above defined quantities along a particular sample path, for example, $\bar{x}(t),\ x_n,\ s_n,\ m_n,\
u^{(l)}_n,\ y^{(l)}(t)$ and $\gamma^{(l)}(t)$ are to be understood as $\bar{X}(\omega,t),\ X_n(\omega),\ S_n(\omega),\ M_n(\omega),\
U^{(l)}_n(\omega),\ Y^{(l)}(\omega,t)$ and $\Gamma^{(l)}(\omega,t)$ respectively for some $\omega$ fixed.
\subsection{Main result - Asymptotic pseudotrajectory}
For every $\omega\in\Omega$, for every $l\geq1$, for every $\tilde{t}\geq0$, let $\tilde{x}^{(l)}(\cdot;\tilde{t})$ denote the solution
of the o.d.e.,
\begin{equation}
\label{ode}
\dot{\tilde{x}}^{(l)}(t;\tilde{t})=y^{(l)}(t+\tilde{t}),
\end{equation}
for every $t\geq0$, with initial condition $\tilde{x}^{(l)}(0;\tilde{t})=\bar{x}(\tilde{t})$.
First we shall get rid of the additive noise terms, $M_n$. Let
\begin{equation*}
\Omega_{a,s}:=\left\{\omega\in\Omega: (A4)\ and\ (A5)\ hold\right\},
\end{equation*}
i.e., the set of sample paths where the iterates are stable and the additive noise terms are eventually negligible. By assumption $(A4)$
and $(A5)$, we have $\mathbb{P}(\Omega_{a,s})=1$.
\begin{lemma}\label{addnois}
For every $l\geq1$, almost surely for every $\omega$, for every $T>0$,
\begin{equation*}
\lim_{t\to\infty}\sup_{0\leq q\leq T}\parallel \bar{x}(t+q)-\tilde{x}^{(l)}(q;t)\parallel=0.
\end{equation*}
\end{lemma}
\begin{proof}
Fix $l\geq1$, $\omega\in\Omega_{a,s}$ and $T>0$. We shall prove the claim along the sequence $\left\{t(n)\right\}_{n\geq0}$ as defined in section
\ref{prelim}. The general claim easily follows from this special case.
Fix $n\geq0$. Let $\tau(n,T):=\min\left\{k>n:t(k)\geq t(n)+T\right\}$. Let $q\in[0,T]$. Then, there exists $k$ such that $t(n)+q\in[t(k),t(k+1))$
and $n\leq k\leq \tau(n,T)-1$. By definition of $\bar{x}(\cdot)$ and $\tilde{x}^{(l)}(\cdot;t(n))$, we have that,
$\bar{x}(t(n)+q)=\alpha x_k+(1-\alpha)x_{k+1}$ and $\tilde{x}^{(l)}(q;t(n))=\alpha\tilde{x}^{(l)}(t(k)-t(n);t(n))+\\(1-\alpha)
\tilde{x}^{(l)}(t(k+1)-t(n);t(n))$ where $\alpha=\frac{t(k+1)-t(n)-q}{t(k+1)-t(k)}$. Since $\tilde{x}^{(l)}(\cdot;t(n))$ is a solution of the
o.d.e. $\eqref{ode}$, we have that, for every $k\geq n$, $\tilde{x}^{(l)}(t(k)-t(n);t(n))=x_n+\sum_{j=n}^{k-1}a(j)h^{(l)}(x_j,s_j,u^{(l)}_j)
$ and by Lemma \ref{paramuse}, we have that, for every $k\geq n$, $x_k=\bar{x}(t(k))=x_n+\sum_{j=n}^{k-1}a(j)h^{(l)}(x_j,s_j,u^{(l)}_j)+
\sum_{j=n}^{k-1}a(j)m_{j+1}$. Thus,
\begin{align*}
\parallel\bar{x}(t(n)+q)-\tilde{x}^{(l)}(q;t(n))\parallel&\leq \parallel\alpha\sum_{j=n}^{k-1}a(j)m_{j+1}+(1-\alpha)
\sum_{j=n}^{k}a(j)m_{j+1}\parallel\\
&\leq\alpha\parallel\sum_{j=n}^{k-1}a(j)m_{j+1}\parallel+(1-\alpha)\parallel\sum_{j=n}^{k}a(j)m_{j+1}
\parallel\\
&\leq \sup_{n\leq k\leq \tau(n,T)}\parallel\sum_{j=n}^{k}a(j)m_{j+1}\parallel.
\end{align*}
Since the r.h.s. of the above inequality is independent of $q\in[0,T]$, we have, $\\\sup_{0\leq q\leq T}\parallel\bar{x}(t(n)+q)-
\tilde{x}^{(l)}(q;t(n))\parallel\leq\sup_{n\leq k\leq \tau(n,T)}\parallel\sum_{j=n}^{k}a(j)m_{j+1}\parallel$. Therefore,
$\\\lim_{n\to\infty}\sup_{0\leq q\leq T}\parallel\bar{x}(t(n)+q)-\tilde{x}^{(l)}(q;t(n))\parallel\leq\lim_{n\to\infty}
\sup_{n\leq k\leq \tau(n,T)}\parallel\sum_{j=n}^{k}a(j)m_{j+1}\parallel$. Now the claim follows follows from assumption $(A4)$.\qed
\end{proof}
\begin{lemma}\label{eqcont}
For every $l\geq1$, for almost every $\omega$, $\left\{\tilde{x}^{(l)}(\cdot;t)\right\}_{t\geq0}$ is relatively compact in
$\mathcal{C}([0,\infty),\mathbb{R}^d)$.
\end{lemma}
\begin{proof}
Fix $l\geq1$, $\omega\in\Omega_{a,s}$. By assumption $(A5)$, we know that there exists $r>0$ such that
$\sup_{n\geq0}\parallel x_n\parallel\leq r$ and hence $\sup_{t\geq0}\tilde{x}^{(l)}(0;t)=\sup_{t\geq0}\bar{x}(t)\leq r$.
For any $t\geq0$, let $[t]:=\max\left\{n\geq0:t(n)\leq t\right\}$. For every $t\geq0$ and $q_1,q_2\in[0,\infty)$ (w.l.o.g. assume $q_1<q_2$)
we have,
\begin{align*}
\parallel \tilde{x}^{(l)}(q_1;t)-\tilde{x}^{(l)}(q_2;t)\parallel&=\parallel\int_{q_1}^{q_2}h^{(l)}(x_{[t+q]},s_{[t+q]},u^{(l)}_{[t+q]})
dq\parallel\\
&\leq\int_{q_1}^{q_2}\parallel h^{(l)}(x_{[t+q]},s_{[t+q]},u^{(l)}_{[t+q]})\parallel dq\\
&\leq\int_{q_1}^{q_2}K^{(l)}(1+\parallel x_{[t+q]}\parallel)dq\\
&\leq C^{(l)}(q_2-q_1),
\end{align*}
where $C^{(l)}:=K^{(l)}(1+r)$ and $r>0$ is such that, $\sup_{n\geq0}\parallel x_n\parallel\leq r$. Thus
$\left\{\tilde{x}^{(l)}(\cdot;t)\right\}_{t\geq0}$ is an equicontinuous family. Now the claim follows from Arzella-Ascoli theorem.
\qed
\end{proof}
From Lemma \ref{addnois} and Lemma \ref{eqcont} we conclude the following for almost every $\omega$.
\begin{itemize}
\item [(1)] The family of functions, \bf{$\left\{\bar{x}(\cdot+t)\right\}_{t\geq0}$ is relatively compact }\rm in
$\mathcal{C}([0,\infty),\mathbb{R}^d)$, because if not, there exist $t_n\to\infty$ such that $\left\{\bar{x}(\cdot+t_n)\right\}_{n\geq0}$ does not
have a limit point in $\mathcal{C}([0,\infty),\mathbb{R}^d)$. Then by Lemma \ref{addnois}, $\left\{\tilde{x}^{(l)}(\cdot;t_n)\right\}_{n\geq0}$ does
not have a limit point in $\mathcal{C}([0,\infty),\mathbb{R}^d)$ which contradicts Lemma \ref{eqcont}.
\item [(2)] Since $\left\{\bar{x}(\cdot+t)\right\}_{t\geq0}$ is relatively compact in $\mathcal{C}([0,\infty),\mathbb{R}^d)$, by Arzella-Ascoli
theorem, we have that for every $T>0$, $\left\{\bar{x}(\cdot+t)|_{[0,T]}\right\}_{t\geq0}$ is equicontinuous. Set $T=1$ and fix $\epsilon>0$. Then
for any $t_0\geq\frac{1}{2}$, there exists $t\geq0$ such that $t+\frac{1}{2}=t_0$. By equicontinuity of $\left\{\bar{x}(\cdot+t)|_{[0,T]}\right\}_{t\geq0}$,
we can obtain a $\delta>0$ (independent of $t$ and hence $t_0$) such that, for every $t'$ satisfying
$|t'-t_0|<\min\left\{\delta,\frac{1}{2}\right\}$, we have $\parallel \bar{x}(t')-\bar{x}(t_0)\parallel<\epsilon$. Since $\epsilon$ was arbitrary
and $\bar{x}(\cdot)|_{[0,\frac{1}{2}]}$ is uniformly continuous we obtain that the function \bf{$\bar{x}(\cdot)$ is uniformly continuous
}\rm on $[0,\infty)$.
\end{itemize}
\begin{proposition}\label{main1}
For $a.e.\ \omega$, every limit point $x^*(\cdot)$ of $\left\{\bar{x}(\cdot+t)\right\}_{t\geq0}$ satisfies the following.
\begin{itemize}
\item [(i)] For every $l\geq1$, there exists $\tilde{\gamma}^{(l)}\in\mathcal{M}(\mathcal{S}\times U)$ such that, for every $t\geq0$,
\begin{equation*}
x^*(t)=x^*(0)+\int_{0}^{t}\left[\int_{\mathcal{S}\times U}h^{(l)}(x^*(q),s,u)\tilde{\gamma}^{(l)}(q)(ds,du)\right]dq.
\end{equation*}
\item [(ii)] For every $l\geq1$, $\tilde{\gamma}^{(l)}$ as in part $(i)$ of this lemma is such that, for almost every $t\geq0$,
\begin{equation*}
\Lambda(\tilde{\gamma}^{(l)})(t)\in D(x^*(t)).
\end{equation*}
\item [(iii)] $x^*(\cdot)$ is absolutely continuous and for almost every $t\geq0$,
\begin{equation*}
\frac{dx^*(t)}{dt}\in\hat{H}(x^*(t)).
\end{equation*}
\end{itemize}
\end{proposition}
\begin{proof}
Fix $\omega\in\Omega_{a,s}$, and let $t_n\to\infty$, such that $\bar{x}(\cdot+t_n)\to x^*(\cdot)$ in
$\mathcal{C}([0,\infty),\mathbb{R}^d)$.
\begin{itemize}
\item [(i)] Fix $l\geq1$. Consider the $\mathcal{M}(\mathcal{S}\times U)$ valued sequence $\left\{\gamma^{(l)}(\cdot+t_n)\right\}_{n\geq1}$. Since
$\mathcal{M}(\mathcal{S}\times U)$ is a compact metric space, there exists a subsequence of the above that converges. Set
$\tilde{\gamma}^{(l)}(\cdot)$ to be some limit point and w.l.o.g. assume $\left\{\gamma^{(l)}(\cdot+t_n)\right\}_{n\geq1}$ converges to
$\tilde{\gamma}^{(l)}(\cdot)$. Since the sequence $\left\{\bar{x}(\cdot+t_n)\right\}_{n\geq1}$ converges to $x^*(\cdot)$, by Lemma \ref{addnois},
we have that $\left\{\tilde{x}^{(l)}(\cdot;t_n)\right\}_{n\geq1}$ also converges to $x^*(\cdot)$ in $\mathcal{C}([0,\infty),\mathbb{R}^d)$. For every
$n\geq1$, by definition of $\tilde{x}^{(l)}(\cdot;t_n)$ we have that for every $t\geq0$,
\begin{align*}
\tilde{x}^{(l)}(t;t_n)&=\bar{x}(t_n)+\int_{0}^{t}y^{(l)}(q)dq\\
&=\bar{x}(t_n)+\int_{0}^{t}h^{(l)}(x_{[t_n+q]},s_{[t_n+q]},u^{(l)}_{[t_n+q]})dq.
\end{align*}
By definition of $\gamma^{(l)}(\cdot)$(see \eqref{diracm} and recall that $\gamma^{(l)}(\cdot)=\Gamma^{(l)}(\omega,\cdot)$) for every
$n\geq1$ and for every $t\geq0$ we can write the above as,
\begin{equation*}
\tilde{x}^{(l)}(t;t_n)=\bar{x}(t_n)+\int_{0}^{t}\left[\int_{\mathcal{S}\times U}h^{(l)}(x_{[t_n+q]},s,u)\gamma^{(l)}(q+t_n)(ds,du)\right]dq.
\end{equation*}
Therefore, for every $t\geq0$,
\begin{align}\label{tmp0}
\lim_{n\to\infty}[\tilde{x}^{(l)}(t;t_n)-\bar{x}(t_n)]&=\lim_{n\to\infty}\int_{0}^{t}\left[\int_{\mathcal{S}\times U}
h^{(l)}(x_{[t_n+q]},s,u)\gamma^{(l)}(q+t_n)(ds,du)\right]dq.\nonumber\\
x^*(t)-x^*(0)&=\lim_{n\to\infty}\int_{0}^{t}\left[\int_{\mathcal{S}\times U}
h^{(l)}(x_{[t_n+q]},s,u)\gamma^{(l)}(q+t_n)(ds,du)\right]dq.
\end{align}
Since $\gamma^{(l)}(\cdot+t_n)\to\tilde{\gamma}^{(l)}(\cdot)$ and by our choice of the topology for $\mathcal{M}(\mathcal{S}\times U)$,
we have,
\begin{equation*}
\int_{0}^{t}\left[\int_{\mathcal{S}\times U}\tilde{f}(q,s,u)\gamma^{(l)}(q+t_n)(ds,du)\right]dq-
\int_{0}^{t}\left[\int_{\mathcal{S}\times U}\tilde{f}(q,s,u)\tilde{\gamma}^{(l)}(q)(ds,du)\right]dq\to0,
\end{equation*}
for all bounded continuous $\tilde{f}:[0,t]\times\mathcal{S}\times U\rightarrow\mathbb{R}$ of the form,
\begin{equation*}
\tilde{f}(q,s,u)=\sum_{m=1}^{N}a_mg_m(q)f_m(s,u),
\end{equation*}
for some $N\geq1$, scalars $a_m$ and bounded continuous functions $g_m,\ f_m$ on $[0,t],\ \mathcal{S}\times U$ respectively,
for $1\leq m\leq N$. By the Stone-Weierstrass theorem, such functions can uniformly approximate any function in
$\mathcal{C}([0,t]\times\mathcal{S}\times U,\mathbb{R})$. Thus the above convergence holds true for all real valued continuous functions
on $[0,t]\times\mathcal{S}\times U$, implying that $\\t^{-1}\gamma^{(l)}(q+t_n)(ds,du)dq\to t^{-1}\tilde{\gamma}^{(l)}(q)(ds,du)dq$ in
$\mathcal{P}([0,t]\times\mathcal{S}\times U)$. Thus,
\begin{small}
\begin{equation}\label{tmp1}
\parallel\int_{0}^{t}\left[\int_{\mathcal{S}\times U}h^{(l)}(x^*(q),s,u)\gamma^{(l)}(q+t_n)(ds,du)\right]dq-
\int_{0}^{t}\left[\int_{\mathcal{S}\times U}h^{(l)}(x^*(q),s,u)\tilde{\gamma}^{(l)}(q)(ds,du)\right]dq\parallel\to0
\end{equation}
\end{small}
as $n\to\infty$. Since $\left\{\bar{x}(\cdot+t_n)|_{[0,t]}\right\}_{n\geq1}$ converges uniformly to $x^*(\cdot)|_{[0,t]}$ we have that, the
function $q\rightarrow x_{[t_n+q]}$ converges uniformly to $x^*(\cdot)|_{[0,t]}$ on $[0,t]$. Using the above and by
Lemma \ref{paramuse}, we have that for every $\epsilon>0$, there exists $N$ (depending on $\epsilon$) such that, for every $n\geq N$, for
every $q\in[0,t]$, we have,
\begin{equation}\label{tmp2}
\parallel\int_{\mathcal{S}\times U}h^{(l)}(x_{[t_n+q]},s,u)\gamma^{(l)}(q+t_n)(ds,du)-
\int_{\mathcal{S}\times U}h^{(l)}(x^*(q),s,u)\gamma^{(l)}(q+t_n)(ds,du)\parallel<\epsilon.
\end{equation}
Now,
\begin{small}
\begin{align*}
\parallel\int_{0}^{t}\left[\int_{\mathcal{S}\times U}h^{(l)}(x_{[t_n+q]},s,u)\gamma^{(l)}(q+t_n)(ds,du)\right]dq&-
\int_{0}^{t}\left[\int_{\mathcal{S}\times U}h^{(l)}(x^*(q),s,u)\tilde{\gamma}^{(l)}(q)(ds,du)\right]dq\parallel\\
&\leq\\
\parallel\int_{0}^{t}\int_{\mathcal{S}\times U}h^{(l)}(x_{[t_n+q]},s,u)\gamma^{(l)}(q+t_n)(ds,du)dq&-
\int_{0}^{t}\int_{\mathcal{S}\times U}h^{(l)}(x^*(q),s,u)\gamma^{(l)}(q+t_n)(ds,du)dq\parallel\\
&+\\
\parallel\int_{0}^{t}\left[\int_{\mathcal{S}\times U}h^{(l)}(x^*(q),s,u)\gamma^{(l)}(q+t_n)(ds,du)\right]dq&-
\int_{0}^{t}\left[\int_{\mathcal{S}\times U}h^{(l)}(x^*(q),s,u)\tilde{\gamma}^{(l)}(q)(ds,du)\right]dq\parallel.
\end{align*}
\end{small}
Taking limit on both sides as $n\to\infty$ in the above equation and using equations \eqref{tmp1} and \eqref{tmp2} we obtain,
\begin{small}
\begin{equation*}
\lim_{n\to\infty}\parallel\int_{0}^{t}\left[\int_{\mathcal{S}\times U}h^{(l)}(x_{[t_n+q]},s,u)\gamma^{(l)}(q+t_n)(ds,du)\right]dq-
\int_{0}^{t}\left[\int_{\mathcal{S}\times U}h^{(l)}(x^*(q),s,u)\tilde{\gamma}^{(l)}(q)(ds,du)\right]dq\parallel\leq\epsilon t,
\end{equation*}
\end{small}
for every $\epsilon>0$. Therefore, for every $t\geq0$,
\begin{small}
\begin{equation*}
\lim_{n\to\infty}\int_{0}^{t}\left[\int_{\mathcal{S}\times U}h^{(l)}(x_{[t_n+q]},s,u)\gamma^{(l)}(q+t_n)(ds,du)\right]dq=
\int_{0}^{t}\left[\int_{\mathcal{S}\times U}h^{(l)}(x^*(q),s,u)\tilde{\gamma}^{(l)}(q)(ds,du)\right]dq.
\end{equation*}
\end{small}
Substituting the above limit in equation \eqref{tmp0}, we get, for every $t\geq0$,
\begin{equation*}
x^*(t)-x^*(0)=\int_{0}^{t}\left[\int_{\mathcal{S}\times U}h^{(l)}(x^*(q),s,u)\tilde{\gamma}^{(l)}(q)(ds,du)\right]dq.
\end{equation*}
\item [(ii)] The proof of this part is similar to the proof of Lemma 6, chapter 6.3 of \cite{borkartxt}. We shall present a proof here for
the sake of completeness.
Let $\left\{f_i\right\}$ be a countable set of real valued continuous functions on $\mathcal{S}$ that is a convergence determining class for
$\mathcal{P}(\mathcal{S})$. By replacing each $f_i$ by $a_if_i+b_i$ for suitable scalars $a_i,\ b_i>0$, we may suppose that
$0\leq f_i(\cdot)\leq 1$ for all $i$. For each $i$,
\begin{equation*}
\zeta_n^{i}:=\sum_{k=0}^{n-1}a(k)(f_i(S_{k+1})-\int_{\mathcal{S}}f_i(s')\Pi(X_k,S_k)(ds')),
\end{equation*}
is a square integrable zero mean martingale w.r.t. the filtration $\left\{\mathscr{F}_n:=\sigma(S_k,X_k:0\leq k\leq n-1)\right\}_{n\geq1}$ and for
almost every $\omega$, $\sum_{n=1}^{\infty}\mathbb{E}[(\zeta^i_{n+1}-\zeta^i_{n})^2|\mathscr{F}_n]\leq 2\sum_{n=0}^{\infty}a(n)^2<\infty$
. Hence by martingale convergence theorem (see Appendix C, Theorem 11 in \cite{borkartxt}), for almost every $\omega$,
$\left\{\zeta_n^i\right\}_{n\geq1}$ converges. Let $\Omega_{m}:=\left\{\omega\in\Omega: \forall i, \left\{\zeta_n^i\right\}_{n\geq1}\ converges\right\}$. Then
$\mathbb{P}(\Omega_{m})=1$. Define,
\begin{equation}
\Omega^*:=\Omega_m\cap\Omega_{a,s}
\end{equation}
and clearly $\mathbb{P}(\Omega^*)=1$. Recall, that for every $T\geq0$ and $n\geq0$, $\tau(n,T):=\min\left\{k\geq n:t(k)\geq t(n)+T\right\}$. Then
for every $\omega\in\Omega^*$, for every $i$, for every $T>0$, as $n\to\infty$,
\begin{equation*}
\sum_{k=n}^{\tau(n,T)}a(k)(f_i(s_{k+1})-\int_{\mathcal{S}}f_i(s')\Pi(x_k,s_k)(ds'))\to0.
\end{equation*}
By the choice of $\left\{f_i\right\}_{i\geq1}$ and the fact that $\left\{a(n)\right\}_{n\geq0}$ are non-increasing (see assumption $(A3)(i)$) we get, that for
every $\omega\in\Omega^*$, for every $i$, for every $T>0$, as $n\to\infty$,
\begin{equation*}
|\sum_{k=n}^{\tau(n,T)}(a(k)-a(k+1))f_i(s_{k+1})|\leq a(n)-a(\tau(n,T)+1)\to0.
\end{equation*}
Thus, for every $\omega\in\Omega^*$, for every $i$, for every $T>0$, as $n\to\infty$,
\begin{equation}\label{tmp3}
\sum_{k=n}^{\tau(n,T)}a(k)(f_i(s_k)-\int_{\mathcal{S}}f_i(s')\Pi(x_k,s_k)(ds'))\to0.
\end{equation}
Fix $l\geq1$. Define for every $\omega$, $\mu^{(l)}(\cdot):=\Lambda(\gamma^{(l)})\in\mathcal{M}(\mathcal{S})$. Then by definition of
$\gamma^{(l)}(\cdot)$ (see equation \eqref{diracm} and recall $\gamma^{(l)}(\cdot):=\Gamma^{(l)}(\omega,\cdot)$) and $\Lambda(\cdot)$ (see
Lemma \ref{lambd}) we have that for every $\omega$, for every $t>0$, $\mu^{(l)}(t)=\delta_{s_n}$ where $n$ is such that,
$t\in[t(n),t(n+1))$. Using the definition of $\mu^{(l)}(\cdot)$ in equation \eqref{tmp3}, we get that for every $\omega\in\Omega^*$, for
every $i$, for every $T>0$, as $n\to\infty$,
\begin{equation*}
\int_{0}^{t(\tau(n,T))-t(n)}\int_{\mathcal{S}}\left[f_i(s)-\int_{\mathcal{S}}f_i(s')\Pi(x_{[t(n)+q]},s)(ds')\right]\mu^{(l)}(q+t(n))
(ds)dq\to0,
\end{equation*}
where, for every $t\geq0$, $[t]:=\max\left\{n\geq0:t(n)\leq t\right\}$. From the above it can easily be shown that for every $\omega\in\Omega^*$,
for every $i$, for every $T>0$,
\begin{equation*}
\lim_{t\to\infty}\int_{0}^{T}\int_{\mathcal{S}}\left[f_i(s)-\int_{\mathcal{S}}f_i(s')\Pi(x_{[t+q]},s)(ds')\right]\mu^{(l)}(q+t)(ds)dq=0.
\end{equation*}
By assumption $(A2)$, $\Pi(\cdot)|_{2rU\times\mathcal{S}}$ is uniformly continuous, where $r=\sup_{n\geq0}\parallel x_n\parallel$.
Thus the function $(x,s)\rightarrow f_i(s)-\int_{\mathcal{S}}f_i(s')\Pi(x,s)(ds')$ is uniformly continuous on $2rU\times\mathcal{S}$ for
every $i$. Using the above, the fact that $\lim_{n\to\infty}(t(n+1)-t(n))=0$ and uniform continuity of $\bar{x}(\cdot)$ we get that, for
every $\omega\in\Omega^*$, for every $i$, for every $T>0$,
\begin{equation*}
\lim_{t\to\infty}\int_{0}^{T}\int_{\mathcal{S}}\left[f_i(s)-\int_{\mathcal{S}}f_i(s')\Pi(\bar{x}(t+q),s)(ds')\right]\mu^{(l)}(q+t)(ds)
dq=0.
\end{equation*}
Fix $\omega\in\Omega^*$. From part $(i)$ of this lemma we have a sequence $t_n\to\infty$ such that $\left\{\bar{x}(\cdot+t_n)\right\}_{n\geq1}$
converges to $x^*(\cdot)$ in $\mathcal{C}([0,\infty),\mathbb{R}^d)$ and $\left\{\gamma^{(l)}(\cdot+t_n)\right\}_{n\geq1}$ converges to
$\tilde{\gamma}^{(l)}(\cdot)$ in $\mathcal{M}(\mathcal{S}\times U)$. By continuity of the map $\Lambda(\cdot)$ (see Lemma \ref{lambd}),
we have that $\mu^{(l)}(\cdot+t_n)\to\tilde{\mu}^{(l)}(\cdot)=\Lambda(\tilde{\gamma}^{(l)})$ in $\mathcal{M}(\mathcal{S})$. Using the
convergence above and the fact that the family of functions, $\left\{x\in2rU\rightarrow\int_{\mathcal{S}}\left[f_i(s)-\int_{\mathcal{S}}
f_i(s')\Pi(x,s)(ds')\right]\nu(ds):\nu\in\mathcal{P}(\mathcal{S})\right\}$ is equicontinuous (which can be shown by arguments similar to Lemma
\ref{paramuse}) we get that for every $i$, for every $T>0$,
\begin{equation*}
\int_{0}^{T}\int_{\mathcal{S}}\left[f_i(s)-\int_{\mathcal{S}}f_i(s')\Pi(x^*(q),s)(ds')\right]\tilde{\mu}^{(l)}(q)(ds)dq=0.
\end{equation*}
An application of Lesbesgue's theorem (see chapter 11.1.3 in \cite{borkartxt}), we get that for almost every $t\geq0$, for every $i$,
\begin{equation*}
\int_{\mathcal{S}}\left[f_i(s)-\int_{\mathcal{S}}f_i(s')\Pi(x^*(t),s)(ds')\right]\tilde{\mu}^{(l)}(t)(ds)=0.
\end{equation*}
By our choice of $\left\{f_i\right\}_{i\geq1}$, we get that for almost every $t\geq0$,
\begin{equation*}
\tilde{\mu}^{(l)}(t)(ds)=\int_{\mathcal{S}}\Pi(x^*(t),s')(ds)\tilde{\mu}^{(l)}(t)(ds').
\end{equation*}
Therefore for almost every $t\geq0$, $\tilde{\mu}^{(l)}(t)=\Lambda(\tilde{\gamma}^{(l)})(t)\in D(x^*(t))$.
\item [(iii)] Fix $l\geq1$. From part $(i)$ of this lemma we have that for every $t\geq0$,
\begin{equation*}
x^*(t)=x^*(0)+\int_{0}^{t}\left[\int_{\mathcal{S}\times U}h^{(l)}(x^*(q),s,u)\tilde{\gamma}^{(l)}(q)(ds,du)\right]dq.
\end{equation*}
Clearly $x^*(\cdot)$ is absolutely continuous and for almost every $t\geq0$,
\begin{equation}\label{tmp4}
\frac{dx^*(t)}{dt}=\int_{\mathcal{S}\times U}h^{(l)}(x^*(t),s,u)\tilde{\gamma}^{(l)}(t)(ds,du).
\end{equation}
By part $(ii)$ of this lemma we have that for almost every $t\geq0$, $\Lambda(\tilde{\gamma}^{(l)})(t)\in D(x^*(t))$. By definition of
the map $\Lambda(\cdot)$ (see Lemma \ref{lambd}), we have that for almost every $t\geq0$, $\tilde{\gamma}^{(l)}_{\mathcal{S}}(t)\in
D(x^*(t))$. By Lemma \ref{chint} and by definition of $\hat{H}^{(l)}$ (see equation \eqref{lim2}) we have that for almost every $t\geq0$,
\begin{equation*}
\int_{\mathcal{S}\times U}h^{(l)}(x^*(t),s,u)\tilde{\gamma}^{(l)}(t)(ds,du)\in
\cup_{\mu\in D(x^*(t))}\int_{\mathcal{S}}H^{(l)}_{x^*(t)}(s)\mu(ds)=\hat{H}^{(l)}(x^*(t)).
\end{equation*}
Using the above in equation \eqref{tmp4} we obtain that for almost every $t\geq0$,
\begin{equation*}
\frac{dx^*(t)}{dt}\in\hat{H}^{(l)}(x^*(t)).
\end{equation*}
Since $l\geq1$ that was fixed was arbitrary, the above holds for every $l\geq1$. Therefore, for almost every $t\geq0$,
\begin{equation*}
\frac{dx^*(t)}{dt}\in\cap_{l\geq1}\hat{H}^{(l)}(x^*(t))=\hat{H}(x^*(t)),
\end{equation*}
where the equality follows from Lemma \ref{approx1}$(iii)$.\qed
\end{itemize}
\end{proof}
Before we proceed further we shall briefly recall the definition of asymptotic pseudotrajectories (APT) for set-valued dynamics
introduced in \cite{benaim1}. The translation flow $\Theta:\mathcal{C}(\mathbb{R},\mathbb{R}^d)\times \mathbb{R}\rightarrow
\mathcal{C}(\mathbb{R},\mathbb{R}^d)$ is the flow defined by,
\begin{equation*}
\Theta^t(\bf{x}\rm)(q):=\bf{x}\rm(q+t).
\end{equation*}
For every $\omega\in\Omega$, extend $\bar{x}(\cdot)$ to $\mathbb{R}$ by letting $\bar{x}(t)=\bar{x}(0)$ for $t<0$. Then $\bar{x}(\cdot)$
is an APT for the flow of DI \eqref{di2} if,
\begin{equation*}
\lim_{t\to\infty}\bf{D}\rm(\Theta^t(\bar{x}),\Sigma)=0,
\end{equation*}
where, $\Sigma:=\cup_{x\in\mathbb{R}^d}\Sigma(x)$ denotes the set of all solutions of DI \eqref{di2}.
In what follows we fix $\omega\in\Omega^*$ and let $\bar{x}(\cdot)$ denote the extension to $\mathbb{R}$ as defined above.
By uniform continuity of $\bar{x}(\cdot)$, we have that the family $\left\{\Theta^t(\bar{x})\right\}_{t\geq0}$ is equicontinuous and by
assumption $(A5)$ is pointwise bounded. Hence $\left\{\Theta^{t}(\bar{x})\right\}_{t\geq0}$ is relatively compact in
$\mathcal{C}(\mathbb{R},\mathbb{R}^d)$. Let $x^*(\cdot)$ be a limit point of $\left\{\Theta^{t}(\bar{x})\right\}_{t\geq0}$. Then by Proposition
\ref{main1}$(iii)$, we have that $x^*(\cdot)|_{[0,\infty)}$ is a solution on $[0,\infty)$ of DI \eqref{di2}. Usually the negative time
argument is omitted since it follows from the positive time argument as follows:
Fix $T>0$. Since $x^*(\cdot)$ is a limit point of $\left\{\Theta^{t}(\bar{x})\right\}_{t\geq0}$, there exists $t_n\to\infty$ such that,
$\left\{\Theta^{t_n}(\bar{x})\right\}_{n\geq1}$ converges to $x^*(\cdot)$ in $\mathcal{C}(\mathbb{R},\mathbb{R}^d)$. Then
$\left\{\Theta^{t_n-T}(\bar{x})\right\}_{n\geq1}$ converges to $x^*(\cdot-T)$. By Proposition \ref{main1}$(iii)$, $x^*(\cdot-T)|_{[0,\infty)}$ is a
solution of DI \eqref{di2}. Therefore $x^*(\cdot)|_{[-T,0]}$ is absolutely continuous and for almost every $t\in[-T,0]$,
\begin{equation*}
\frac{dx^*(t)}{dt}\in\hat{H}(x^*(t)).
\end{equation*}
Since $T>0$, is arbitrary, we have that $x^*(\cdot)|_{(-\infty,0]}$ is a solution on $(-\infty,0]$ of DI \eqref{di2}.
Therefore every limit point of $\left\{\Theta^{t}(\bar{x})\right\}_{t\geq0}$ is in $\Sigma$, the set of solutions of DI \eqref{di2}. Then by Theorem
4.1 in \cite{benaim1} we get the following result.
\begin{theorem}\emph{[APT]}\label{apt}
Under assumptions $(A1)-(A5)$, for almost every $\omega$, the linearly interpolated trajectory of recursion \eqref{rec}, $\bar{x}(\cdot)$,
is an asymptotic pseudotrajectory of DI \eqref{di2}.
\end{theorem}
\subsection{Characterization of limit sets}
For every $\omega$, the limit set of recursion \eqref{rec}, denoted by $L(\bar{x})$ is defined as,
\begin{equation}
L(\bar{x}):=\cap_{t\geq0}\overline{\left\{\bar{x}(q+t):q\geq0\right\}}.
\end{equation}
As a consequence of Theorem \ref{apt} we will be able to characterize the limit set of recursion \eqref{rec} in terms of the dynamics
induced by $\hat{H}$. The notions of invariance, internal chain transitivity, attracting sets, basin of attraction and attractors are
taken from \cite{benaim1}. We shall state here definitions of a few of the notions mentioned above for the sake of completeness.
The \bf{flow }\rm of DI$\eqref{di2}$ is given by the set valued map $\Phi:\mathbb{R}^d\times \mathbb{R}\rightarrow
\left\{\text{subsets of }\mathbb{R}^d\right\}$, where for every $(x,t)\in\mathbb{R}^d\times\mathbb{R}$,
$\Phi(x,t):=\left\{\bf{x}\rm(t):\bf{x}\rm\in\Sigma(\it{x}\rm)\right\}$. For any $C\subseteq\mathbb{R}^d$, let $\omega_{\Phi}(C):=\cap_{t\geq0}
\overline{\Phi_{[t,\infty)}(C)}$, where $\Phi_{[t,\infty)}(C):=\cup_{(x,\tau)\in\ C\times[t,\infty)}\Phi(x,\tau)$. For any $A\subseteq\mathbb{R}^d$,
its \bf{basin of attraction }\rm denoted by $B(A)$ is defined as, $B(A):=\left\{x\in\mathbb{R}^d:
\omega_{\Phi}\left(\left\{x\right\}\right)\subseteq A\right\}$.
A set $A\subseteq\mathbb{R}^d$ is said to be \bf{invariant }\rm for DI$\eqref{di2}$ if for all $x\in A$, there exists $\bf{x}\rm\in
\Sigma(\it{x}\rm)$ such that for every $t\in\mathbb{R}$, $\bf{x}\rm(t)\in A$.
A compact set $A\subseteq\mathbb{R}^d$ is an \bf{attracting set }\rm for the flow of DI$\eqref{di2}$ if there exists a neighborhood $O$ of $A$,
with the property that for every $\epsilon>0$ there exists $t_{\epsilon}>0$ such that for every $t\geq t_{\epsilon}$, $\Phi_{t}(O)\subseteq
N^{\epsilon}(A)$, where $N^{\epsilon}(A)$ stands for the $\epsilon$-neighborhood of $A$.
A compact set $A\subseteq\mathbb{R}^d$ is an \bf{attractor }\rm for the flow of DI$\eqref{di2}$ if it is an attracting set and is invariant.
Further if the basin of attraction of attractor $A$ is the whole of $\mathbb{R}^d$, that is $B(A)=\mathbb{R}^d$, then $A$ is a
\bf{global attractor}\rm.
Given a set $A\subseteq\mathbb{R}^d$ and $x,y\in A$, for any $\epsilon>0$ and $T>0$ there exists an $(\epsilon,T)$ chain from $x$ to $y$ for
DI\eqref{di2} if there exists an integer $n\in\mathbb{N}$, solutions $\bf{x}\rm_1,\dots,\bf{x}\rm_n$ to DI\eqref{di2} and real numbers $t_1,\dots,t_n$
greater than $T$ such that
\begin{itemize}
\item for all $i\in\left\{1,\dots,n\right\}$ and for all $q\in[0,t_i]$, $\bf{x}\rm_i(\it{q}\rm)\in A$,
\item for all $i\in\left\{1,\dots,n\right\}$, $\parallel\bf{x}\rm_i(\it{t_i}\rm)-\bf{x}\rm_{i+1}(0)\parallel\leq\epsilon$,
\item $\parallel \bf{x}\rm_1(0)-x\parallel\leq\epsilon$ and $\parallel\bf{x}\rm_n(\it{t_n}\rm)-y\parallel\leq\epsilon$.
\end{itemize}
A compact set $A\subseteq\mathbb{R}^d$ is said to be \bf{internally chain transitive }\rm if for every $x,y\in A$, for every $\epsilon>0$ and
for every $T>0$, there exists $(\epsilon,T)$ chain from $x$ to $y$ for the DI\eqref{di2}.
\begin{theorem}\emph{[Limit set]}\label{ls}
Under assumptions $(A1)-(A5)$, for almost every $\omega$, the following hold.
\begin{itemize}
\item [(i)] $L(\bar{x})$ is a non-empty, compact subset of $\mathbb{R}^d$ and is internally chain transitive.
\item [(ii)] If $A\subseteq\mathbb{R}^d$ is an attracting set for the flow of DI \eqref{di2} with a basin of attraction $B(A)$, such
that $L(\bar{x})\cap B(A)\neq\emptyset$, then $L(\bar{x})\subseteq A$.
\item [(iii)] If $A\subseteq\mathbb{R}^d$ is a global attractor for the flow of DI \eqref{di2}, then $L(\bar{x})\subseteq A$.
\item [(iv)] If $A=\left\{x^*\right\}$ is a global attractor for the flow of DI \eqref{di2}, then $\left\{x_n\right\}_{n\geq0}$ converges to $x^*$.
\end{itemize}
\end{theorem}
\begin{proof}:
\begin{itemize}
\item [(i)] Follows from Theorem \ref{apt} above and Theorem 4.3 of \cite{benaim1}.
\item [(ii)] From part $(i)$ of this theorem we have that $L(\bar{x})$ is internally chain transitive. Now the claim follows from Theorem
3.23 of \cite{benaim1}.
\item [(iii)] Follows from part $(i)$ of this theorem and Corollary 3.24 of \cite{benaim1}.
\item [(iv)] Follows from part $(iii)$ of this theorem with $A={x^*}$.\qed
\end{itemize}
\end{proof}
\section{Space of probability measure valued functions}
\label{sopmvf}
In this section we define the space of probability measure valued measurable functions on $[0,\infty)$ and introduce an appropriate
topology on this space. Such spaces are used in the theory of existence of optimal control for diffusions and can also be found in
\cite{borkarmark} and \cite{borkaropt}. We shall use the following in the analysis of recursion \eqref{rec} presented in the next section.
Let $\mathcal{M}(\mathcal{S}\times U)$ denote the set of all functions $\gamma:[0,\infty)\rightarrow\mathcal{P}(\mathcal{S}\times U)$,
measurable. Formally,
\begin{equation*}
\mathcal{M}(\mathcal{S}\times U):=\left\{\gamma:[0,\infty)\rightarrow\mathcal{P}(\mathcal{S}\times U):\ \gamma\ is\ measurable\right\}.
\end{equation*}
Similarly, define $\mathcal{M}(\mathcal{S})$, the set of all functions $\gamma:[0,\infty)\rightarrow\mathcal{P}(\mathcal{S})$, measurable.
Formally,
\begin{equation*}
\mathcal{M}(\mathcal{S}):=\left\{\gamma:[0,\infty)\rightarrow\mathcal{P}(\mathcal{S}):\ \gamma\ is\ measurable\right\}.
\end{equation*}
Let $\tau_{\mathcal{S}\times U}$ denote the topology on $\mathcal{M}(\mathcal{S}\times U)$ which is the coarsest topology that
renders continuous the maps $\gamma\in\mathcal{M}(\mathcal{S}\times U)\rightarrow\int_{0}^{T}g(t)[\int_{\mathcal{S}\times U}f(s,u)
\gamma(t)(ds,du)]dt\in\mathbb{R}$ for all $f\in\mathcal{C}(\mathcal{S}\times U,\mathbb{R})$, for all $T>0$ and for all
$g\in\mathbb{L}_2[0,T]$.
Similarly let $\tau_{\mathcal{S}}$ denote the topology on $\mathcal{M}(\mathcal{S})$ which is the coarsest topology that renders
continuous the maps $\gamma\in\mathcal{M}(\mathcal{S})\rightarrow\int_{0}^{T}g(t)[\int_{\mathcal{S}}f(s)\gamma(t)(ds)]dt\in\mathbb{R}$
for all $f\in\mathcal{C}(\mathcal{S},\mathbb{R})$, for all $T>0$ and for all $g\in\mathbb{L}_2[0,T]$.
The next result is a well known metrization lemma for the above defined topological spaces.
\begin{lemma}
\emph{[Metrization]}
\begin{itemize}
\item [(i)] The topological space $(\mathcal{M}(\mathcal{S}\times U),\tau_{\mathcal{S}\times U})$ is compact metrizable.
\item [(ii)] The topological space $(\mathcal{M}(\mathcal{S}),\tau_{\mathcal{S}})$ is compact metrizable.
\end{itemize}
\end{lemma}
For a proof of the above lemma we refer the reader to Lemma 2.1 in \cite{borkarmark}. The next lemma provides a continuous map from
$\mathcal{M}(\mathcal{S}\times U)$ to $\mathcal{M}(\mathcal{S})$ which is used later in this paper. Recall that for any probability
measure, $\nu\in\mathcal{P}(\mathcal{S}\times U)$, $\nu_{\mathcal{S}}$ denotes the image of $\nu$ under projection,
$\mathcal{S}\times U\rightarrow\mathcal{S}$ (i.e. $\nu_{\mathcal{S}}\in\mathcal{P}(\mathcal{S})$ such that for every
$A\in\mathscr{B}(\mathcal{S})$, $\nu_{\mathcal{S}}(A)=\int_{A\times U}\nu(ds,du)$).
\begin{lemma}:\label{lambd}
\begin{itemize}
\item [(i)]The map $\lambda:\mathcal{P}(\mathcal{S}\times U)\rightarrow\mathcal{P}(\mathcal{S})$ such that, for every
$\nu\in\mathcal{P}(\mathcal{S}\times U)$, $\lambda(\nu):=\nu_{\mathcal{S}}$, is continuous.
\item [(ii)] For every $\gamma\in\mathcal{M}(\mathcal{S}\times U)$, $\lambda\circ\gamma\in\mathcal{M}(\mathcal{S})$.
\item [(iii)] The map $\Lambda:\mathcal{M}(\mathcal{S}\times U)\rightarrow\mathcal{M}(\mathcal{S})$ such that, for every
$\gamma\in\mathcal{M}(\mathcal{S}\times U)$, $\Lambda(\gamma):=\lambda\circ\gamma$, is continuous.
\end{itemize}
\end{lemma}
\begin{proof}:
\begin{itemize}
\item [(i)] Let $\left\{\nu^n\right\}_{n\geq1}$ be a sequence in $\mathcal{P}(\mathcal{S}\times U)$ converging to
$\nu\in\mathcal{P}(\mathcal{S}\times U)$. Let $\pi:\mathcal{S}\times U\rightarrow\mathcal{S}$ denote the projection map.
Then for every $f\in\mathcal{C}(\mathcal{S},\mathbb{R})$, $f\circ\pi\in\mathcal{C}(\mathcal{S}\times U,\mathbb{R})$. By Theorem 2.1.1
$(ii)$ in \cite{borkarap} and the compactness of $\mathcal{S}\times U$, for every $f\in\mathcal{C}(\mathcal{S},\mathbb{R})$,
$\int_{\mathcal{S}\times U}\left(f\circ\pi\right)(s,u)\nu^n(ds,du)\to\int_{\mathcal{S}\times U}\left(f\circ\pi\right)(s,u)\nu(ds,du)$ as $n\to\infty$. Hence for
every $f\in\mathcal{C}(\mathcal{S},\mathbb{R})$, $\int_{\mathcal{S}}f(s)\nu^n\pi^{-1}(ds)\to\int_{\mathcal{S}}f(s)\nu\pi^{-1}(ds)$.
Observing that for every $n\geq1$, $\nu^n\pi^{-1}=\nu^n_{\mathcal{S}}$ and $\nu\pi^{-1}=\nu_{\mathcal{S}}$ gives us that, for every
$f\in\mathcal{C}(\mathcal{S},\mathbb{R})$, $\int_{\mathcal{S}}f(s)\nu^n_{\mathcal{S}}(ds)\to\int_{\mathcal{S}}f(s)\nu_{\mathcal{S}}(ds)$
. Hence $\lambda(\nu^n)\to\lambda(\nu)$ in $\mathcal{P}(\mathcal{S})$, which gives us continuity of $\lambda(\cdot)$.
\item [(ii)] Composition of two measurable functions is measurable.
\item [(iii)] Let $\left\{\gamma_n\right\}_{n\geq1}$ be a sequence in $\mathcal{M}(\mathcal{S}\times U)$ converging to
$\gamma\in\mathcal{M}(\mathcal{S}\times U)$. Then, for every $f\in\mathcal{C}(\mathcal{S}\times U,\mathbb{R})$, for every $T>0$ and for
every $g\in\mathbb{L}_2[0,T]$, $\int_{0}^{T}g(t)[\int_{\mathcal{S}\times U}f(s,u)\gamma_n(t)(ds,du)]dt\to\int_{0}^{T}g(t)
[\int_{\mathcal{S}\times U}f(s,u)\gamma(t)(ds,du)]dt$ as $n\to\infty$. Hence, for every $f\in\mathcal{C}(\mathcal{S},\mathbb{R})$, for
every $T>0$ and for every $g\in\mathbb{L}_2[0,T]$, $\int_{0}^{T}g(t)[\int_{\mathcal{S}\times U}\left(f\circ\pi\right)(s,u)\gamma_n(t)(ds,du)]dt\to\\
\int_{0}^{T}g(t)[\int_{\mathcal{S}\times U}\left(f\circ\pi\right)(s,u)\gamma(t)(ds,du)]dt$ as $n\to\infty$. By argument similar to part $(i)$ of this
lemma, we have, for every $f\in\mathcal{C}(\mathcal{S},\mathbb{R})$, for every $T>0$ and for every $g\in\mathbb{L}_2[0,T]$,
$\int_{0}^{T}g(t)[\int_{\mathcal{S}}f(s)\left(\lambda\circ\gamma_n\right)(t)(ds)]dt\to\int_{0}^{T}g(t)
[\int_{\mathcal{S}}f(s)\left(\lambda\circ\gamma\right)(t)(ds)]dt$ as $n\to\infty$. Therefore $\left\{\left(\lambda\circ\gamma_n\right)\right\}_{n\geq1}$ converges to
$\left(\lambda\circ\gamma\right)$ in $\mathcal{M}(\mathcal{S})$, which gives us continuity of $\Lambda(\cdot)$.\qed
\end{itemize}
\end{proof}
\section{Conclusions and directions for future work}
\label{cadffw}
We have shown that almost surely the linearly interpolated trajectory of recursion \eqref{rec} is an asymptotic pseudotrajectory for the flow
of DI \eqref{di2}. The asymptotic pseudotrajectory result proved in this paper enables one to characterize limit sets of recursion \eqref{rec}
as internally chain transitive sets of the flow of DI \eqref{di2} which in-turn enables us to guarantee convergence of the iterates to
attractors of DI \eqref{di2} as stated in Theorem \ref{ls}. We have also stated two variants of the Markov noise assumption where the analysis
is very similar to the one presented. Finally applications are presented where the recursion studied in this paper
naturally appears and in the case of subgradient descent we have also been able to interpret the set-valued map associated with DI \eqref{di2} as
subdifferential of a certain averaged function.
Certain extensions and applications which we wish to consider in future are listed below.
\begin{itemize}
\item [(1)] Stochastic approximation schemes on multiple time scales are extensively studied in literature and find use in several
reinforcement learning and optimization applications. For the case without Markov noise, the two time scale stochastic approximation with
single valued maps and the case of set-valued maps have already been studied (see chapter 6 in \cite{borkartxt} for the single valued case
and \cite{leslep} for the set-valued case). One can extend the analysis presented in this paper to obtain a similar result for two time scale
stochastic recursive inclusions with Markov noise.
\item [(2)] Throughout this paper we have assumed that the state space of the Markov noise is a compact metric space. In many practical
scenarios, the Markov noise terms take values in a general Polish space. For such a case we believe that the analysis presented in the paper
can be extended under certain additional assumptions. The idea would be to embed the polish space as a dense subset of a compact metric space
and under some additional assumptions which guarantee an upper semicontinuous extension of set-valued map $H$ and tightness of some
probability measures one might be able to provide an APT argument as in this paper (see \cite{borkarmark}).
\item [(3)] Assumption $(A5)$, which ensures that the iterates are stable usually is hard to verify. The Borkar-Meyn theorem (see
\cite{borkarmeyn}) which provides a sufficient condition for stability of iterates for stochastic approximation schemes with single valued
maps has been extended to the set-valued case without Markov noise (see \cite{arunstab}) and we believe that a similar extension is possible
to the set-valued case with Markov noise. Stability results similar to the one in \cite{andvih} also need to be investigated for possible
extension to the set-valued case.
\item [(4)] In many applications one encounters the recursion,
\begin{equation*}
X_{n+1}\in \text{\bf{P}\rm}\left[X_{n}+a(n)(H(X_n,S_n)+M_{n+1})\right],
\end{equation*}
where \bf{P }\rm denotes the projection of the set $X_{n}+a(n)(H(X_n,S_n)+M_{n+1})$ onto a compact (possibly convex) subset of $\mathbb{R}^d$. Such
recursions are of interest also because the assumption of stability (that is $(A5)$) is naturally satisfied. For the single valued case such
recursions are studied in \cite{dupuis} and the extension of the same to the recursion above needs to be established.
\item [(5)] Another variant of Markov noise assumptions studied for the single valued case is given in \cite{matti}. These set of assumptions
enable one to study behavior of stochastic approximation schemes in the single valued case when the state space is a general Euclidean space.
Since in this paper we study the set-valued case by converting it to the single valued setting it would be interesting to see if those
assumptions can be used for the set-valued case as in this paper.
\end{itemize}
\subsection{Upper semicontinuous set valued maps and their approximation}
First we shall recall the notion of upper semicontinuity, lower semicontinuity and continuity of set valued maps. These notions are
taken from chapter 1, section 1 of \cite{aubindi}.
\begin{definition}
A set valued map $H:\mathbb{R}^d\times\mathcal{S}\rightarrow\left\{\text{subsets of }\mathbb{R}^d\right\}$ is,
\begin{itemize}
\item \it{Upper semicontinuous }\rm(u.s.c.) if, for every $(x_0,s_0)\in\mathbb{R}^d\times\mathcal{S}$, for every $\epsilon>0$, there
exists $\delta>0$ (depending on $(x_0,s_0)$ and $\epsilon$) such that,
\begin{equation*}
\parallel x-x_0\parallel<\delta,\ d_{\mathcal{S}}(s,s_0)<\delta\implies H(x,s)\subseteq H(x_0,s_0)+\epsilon U,
\end{equation*}
where $U$ denotes the closed unit ball in $\mathbb{R}^d$.
\item \it{Lower semicontinuous }\rm(l.s.c) if, for every $(x_0,s_0)\in\mathbb{R}^d\times\mathcal{S}$, for every $z_0\in H(x_0,s_0)$,
for every sequence $\left\{\left(x_n,s_n\right)\right\}_{n\geq1}$ converging to $(x_0,s_0)$, there exists a sequence
$\left\{z_n\in H(x_n,s_n)\right\}$ converging to $z_0$.
\item \it{Continuous }\rm if, it is both u.s.c. and l.s.c.
\end{itemize}
\end{definition}
For set valued maps taking compact set values we have the above mentioned notion of u.s.c. to be equivalent to the standard notion of u.s.c.
(see pg. 45, \cite{aubindi}). In this paper we shall encounter set valued maps which are compact set valued and hence we have chosen to state
the above as the definition of upper semicontinuity.
Since the set valued map $H$ satisfying assumption $(A1)$ has closed graph (i.e. assumption $(A1)(iii)$ holds), the following lemma follows
from Corollary 1 in chapter 1, section 1 of \cite{aubindi}.
\begin{lemma}
\emph{[u.s.c.]} A set valued map $H$ satisfying $(A1)$ is u.s.c.
\end{lemma}
The next lemma gives a sequence of decreasing continuous set valued maps which approximate the set valued map $H$ satisfying assumption
$(A1)$. The statement of the lemma below can be found in page 39 of \cite{aubindi}.
\begin{lemma}
\emph{[continuous embedding]}\label{ctem} Let $H$ be a set valued map satisfying $(A1)$. Then there exists a sequence of set valued
maps, $\left\{H^{(l)}\right\}_{l\geq1}$ where for every $\ l\geq1$,
\begin{itemize}
\item [(i)] $H^{(l)}:\mathbb{R}^d\times\mathcal{S}\rightarrow\left\{\text{subsets of }\mathbb{R}^d\right\}$ is continuous and for
every $(x,s)\in\mathbb{R}^d\times\mathcal{S}$, $H^{(l)}(x,s)$ is a convex and compact subset of $\mathbb{R}^d$,
\item [(ii)] for every $(x,s)\in\mathbb{R}^d\times\mathcal{S}$, $H(x,s)\subseteq H^{(l+1)}(x,s)\subseteq H^{(l)}(x,s)$,
\item [(iii)] there exists $K^{(l)}>0$, such that for every
$(x,s)\in\mathbb{R}^d\times\mathcal{S},\\ \sup_{z\in H^{(l)}(x,s)}\parallel z\parallel\leq K^{(l)}(1+\parallel x\parallel)$.
\end{itemize}
Furthermore, for every $(x,s)\in\mathbb{R}^d\times\mathcal{S}$, $H(x,s)=\cap_{\substack{l\geq1}}H^{(l)}(x,s)$.
\end{lemma}
A brief outline of the proof of the above lemma is provided in appendix \ref{prf_ctem}.
The following are some useful observations from the proof of Lemma \ref{ctem}.
\begin{itemize}
\item [(a)] $\tilde{K}:=\sup_{l\geq1}K^{(l)}$ is finite.
\item [(b)] For every $(x,s)\in\mathbb{R}^d\times\mathcal{S}$, for every $\epsilon>0$, there exists $L$ (depending on $\epsilon$ and
$(x,s)$), such that, for every $l\geq L$, $H^{(l)}(x,s)\subseteq H(x,s)+\epsilon U$.
\end{itemize}
The next lemma gives us a continuous parametrization of the continuous set valued maps $H^{(l)}$ as in Lemma \ref{ctem}. This
lemma follows from Theorem 2 in chapter 1, section 7 of \cite{aubindi}.
\begin{lemma}
\emph{[parametrization]}\label{param} For every $l\geq1$, $H^{(l)}$ be a set valued map as in Lemma \ref{ctem}. Then for every
$l\geq1$, there exists a continuous function $h^{(l)}:\mathbb{R}^d\times\mathcal{S}\times U\rightarrow\mathbb{R}^d$ such that,
\begin{itemize}
\item [(i)] for every $(x,s)\in \mathbb{R}^d\times\mathcal{S}$, $H^{(l)}(x,s)=h^{(l)}(x,s,U)$ where $U$ denotes the closed unit ball
in $\mathbb{R}^d$ and $h^{(l)}(x,s,U):=\left\{h^{(l)}(x,s,u):u\in U\right\}$,
\item[(ii)] for every $(x,s,u)\in\mathbb{R}^d\times\mathcal{S}\times U$,
$\parallel h^{(l)}(x,s,u)\parallel\leq K^{(l)}(1+\parallel x\parallel)$ where $K^{(l)}$ is as in Lemma \ref{ctem}\it{(iii)}\rm.
\end{itemize}
\end{lemma}
Combining Lemma \ref{ctem} and Lemma \ref{param} we obtain the approximation theorem stated below.
\begin{theorem}
\emph{[approximation]} Let $H$ be a set valued map satisfying $(A1)$. Then there exists a sequence of continuous functions,
$\left\{h^{(l)}\right\}_{l\geq1}$, such that, for every $l\geq1$,
\begin{itemize}
\item [(i)] $h^{(l)}:\mathbb{R}^d\times\mathcal{S}\times U\rightarrow \mathbb{R}^d$ is continuous and for every
$(x,s)\in\mathbb{R}^d\times\mathcal{S}$, $h^{(l)}(x,s,U)$ is a convex and compact subset of $\mathbb{R}^d$,
\item [(ii)] for every $(x,s)\in\mathbb{R}^d\times\mathcal{S}$, $H(x,s)\subseteq h^{(l+1)}(x,s,U)\subseteq h^{(l)}(x,s,U)$,
\item [(iii)] there exists $K^{(l)}>0$, such that for every $(x,s,u)\in \mathbb{R}^d\times\mathcal{S}\times U$,
$\\\parallel h^{(l)}(x,s,u)\parallel\leq K^{(l)}(1+\parallel x\parallel)$.
\end{itemize}
Furthermore, for every $(x,s)\in\mathbb{R}^d\times\mathcal{S}$, $H(x,s)=\cap_{l\geq1}h^{(l)}(x,s,U)$.
\end{theorem}
| f5136b7735b2523e973dd0105203a19e90343fe0 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Let $d\geq 3$ and let $G(n,d)$ denote the random $d$-regular graph on $n$ vertices (see e.g.\ the monograph \cite{bollobas}). Equivalently, we can think of $G(n,d)$ as a random model of symmetric $0-1$ matrices in which the row sums are conditioned to be $d$. It is expected that the spectral properties of $G(n,d)$ are closely related to random matrix theory; however, many questions in the area are still open. It is well known that the spectral measure of $G(n,d)$ converges to the so called Kesten--McKay measure in the weak topology as $n$ goes to infinity. This gives an approximate semicircle law if $d$ is large. A famous result by J. Friedman solves Alon's second eigenvalue conjecture showing that $G(n,d)$ is almost Ramanujan \cite{friedman}. Much less is known about the scaled eigenvalue spacing and about the structure of the eigenvectors. Recent results in the area include \cite{charles, doron} on the second eigenvalue; \cite{bhky, bauer, elliot, pal} on eigenvalue spacing, local semicircle law and functional limit theorems; \cite{nalini, brooks, geisinger} on the delocalization of the eigenvectors.
In the present paper we study approximate eigenvectors or shortly: \emph{almost eigenvectors} of $G(n,d)$ i.e.\ vectors that satisfy the eigenvector equation $(A-\lambda I)v=0$ with some small error. Almost eigenvectors are not necessarily close to proper eigenvectors. They are much more general objects. For example any linear combination of eigenvectors with eigenvalues in the interval $[\lambda-\varepsilon,\lambda+\varepsilon]$ is an almost eigenvector with error depending on $\varepsilon$. In general a vector is an almost eigenvector if and only if its spectral measure is close to a Dirac measure in the weak topology.
We show that despite of this generality, almost eigenvectors of $G(n,d)$ have a quite rigid structure if $n$ is big. Our main result implies that every almost eigenvector (with entry sum 0 and normalized to have length $\sqrt{n}$) has an entry distribution close to some Gaussian distribution $N(0,\sigma)$ in the weak topology where $0\leq\sigma\leq 1$. Note that if $\sigma=0$, then the $\ell^2$-weight of the vector is concentrated on a small fraction of the vertices. Such vectors are called localized. Our main result holds even in a stronger sense where joint distributions are considered using the local structure of the graph. In some sense our result is best possible since there are examples for both localized and delocalized almost eigenvectors (see chapter \ref{prelim}). Note that proper eigenvectors are conjectured to be delocalized.
The issue of eigenvector Gaussianity goes back to random matrix theory. It is not hard to show that in the GUE (Gaussian Unitary Ensemble) random matrix model every eigenvector has a near Gaussian entry distribution. It is much harder to analyze the random model when the elements of the matrix are chosen from a non Gaussian distribution. Nevertheless Gaussianity of the eigenvectors is proved under various conditions for generalized Wigner matrices \cite{bouryau, taovu} and also for various other models (see e.g.\ \cite{bloemen} and \cite{antti, surveyvu} for recent surveys). Sparser models are harder to analyze \cite{huang}. This paper deals with the sparsest case, where the matrix is the adjacency matrix of a random $d$-regular graph with some fixed $d$. In this case there is a stronger meaning of eigenvector Gaussianity. For example it is natural to ask about the Gaussianity of the joint distribution of the entries at the two endpoints of a randomly chosen edge in the graph. More generally one can look at the joint distribution of the entries in random balls of radius $r$. Our Gaussianity results for almost eigenvectors are established in this strong sense. Furthermore, it makes sense to study the joint distribution of the entries of several almost eigenvectors. More precisely, a $k$-tuple of almost eigenvectors can be interpreted as a function from the vertices to $\mathbb R^k$. This function evaluated at a randomly chosen vertex gives a probability distribution on $\mathbb R^k$ that we define as the joint distribution of the entries. It is an interesting question whether such joint distributions are Gaussian. We prove this if $k$ is fixed, $n$ is large and the eigenvalues corresponding to the almost eigenvectors are close to each other. Moreover, we also prove this result when the joint distribution of many eigenvectors is considered in random neighborhoods.
The proof of our main theorem is based on the so-called local-global graph limits \cite{BR, HLSz}. However, to keep the paper self-contained, we use a slightly simplified framework (see section \ref{typpro}) optimized for this particular problem. We relate the properties of random regular graphs to random processes on the infinite $d$-regular tree $T_d$.
Most of the work is done in this convenient limiting framework.
An invariant process on $T_d$ is a joint distribution of random variables $\{X_v\}_{v\in T_d}$ labeled by the vertices of $T_d$ such that it is invariant under the automorphism group of $T_d$. A special class of invariant processes, called {\it typical processes}, was introduced in \cite{BSz}. Roughly speaking, an invariant process is typical if it can be obtained as the Benjamini--Schramm limit of colored random regular graphs. There is a correspondence between the properties of typical processes on $T_d$ and the properties of large random $d$-regular graphs.
Our main theorem is equivalent to the statement that if a typical process satisfies the eigenvector equation at every vertex of $T_d$ and has finite variance (at every vertex), then the process is jointly Gaussian. Note that such Gaussian eigenvector processes on $T_d$ are completely characterized and there is a unique one for each possible eigenvalue. A key ingredient in our proof is a general entropy inequality for typical processes. It implies that typical eigenvector processes obey another inequality involving differential entropy. From here we finish the proof using heat propagation on the space of typical eigenvector processes combined with DeBrujin's identity for Fisher information. Gaussianity will follow from the fact that heat propagation converges to a Gaussian distribution.
A well studied subclass of typical processes is the class of factor of i.i.d.\ processes. These processes appeared first in ergodic theory but they are also relevant in probability theory, combinatorics, statistical physics and in computer science. Not every typical process is factor of i.i.d.; this follows from the results of Gamarnik and Sudan \cite{gamarnik}; see also Rahman and Vir\'ag \cite{mustazee}.
Despite of recent progress in the area \cite{cordec, balint, lewis, csoka, robin, damien, harangi, HLSz, russ, nazarov, fiidperc}, a satisfying understanding of factor of i.i.d.\ processes is only available in the case $d=2$ \cite{ornstein}, which is basically equivalent to the framework of classical ergodic theory of $\mathbb{Z}$ actions. Our results imply that if an invariant process (with finite variance) is in the weak closure of factor of i.i.d. processes and satisfies the eigenvector equation then the process is Gaussian. This answers a question of B. Vir\'ag.
{\bf Outline of the paper.} In Chapter \ref{chap:main}, we formulate the main results for finite random $d$-regular graphs. Chapter \ref{prelim} and \ref{chap:eigen} contain general statements about invariant processes, eigenvector processes, entropy and almost eigenvectors. Chapter \ref{typpro} provides the translation of our main result to the infinite setting using typical eigenvector processes on the infinite $d$-regular tree. In Chapter \ref{chap:entr}, we prove a necessary condition for a process to be typical in the form of an entropy inequality. In Chapter \ref{smooth} we reduce the limiting form of the main theorem to a special family of eigenvector processes called smooth eigenvector processes. Chapter \ref{chap:entrtyp} gives a differential entropy inequality for smooth eigenvector processes. In Chapter \ref{chap:cov}, we calculate the eigenvalues of special submatrices of the covariance matrices of eigenvector processes corresponding to balls around vertices and edges on the tree. In Chapter \ref{chap:impr}, we use the results from Chapter \ref{chap:cov} to prove that among smooth typical eigenvector processes the Gaussian minimizes the differential entropy formula from Chapter \ref{chap:entrtyp}. On the other hand, in Chapter \ref{chap:heat}, we show that the Gaussian eigenvector process maximizes the same formula. Moreover, we finish the proof of the main result in Chapter \ref{chap:heat}.
\section{The main theorem}
\label{chap:main}
In this chapter we state the main theorem first in a simpler but weaker form and later in the strong form.
An $\varepsilon$-almost eigenvector of a matrix $A\in\mathbb{R}^{n\times n}$ with eigenvalue $\lambda$ is a vector $v\in\mathbb{R}^n$ such that $\|v\|_2=1$ and $\|Av-\lambda v\|_2\leq\varepsilon$.
To every vector $v$ in $\mathbb{R}^n$ we associate a probability distribution ${\rm distr}(v)$ on $\mathbb{R}$ obtained by choosing a uniform random entry from $v$. If $\|v\|_2=1$, then the second moment of ${\rm distr}(v)$ is $1/n$. Thus, to avoid degeneracy in this case, it is more natural to consider ${\rm distr}(\sqrt{n}v)$ whose second moment is $1$.
We will compare probability distributions on $\mathbb{R}$ using an arbitrary but fixed metrization of the weak convergence of probability measures. Our theorem in a weak form says the following
\begin{theorem}[Weak form of main theorem] For every $\varepsilon>0$ there are constants $N, \delta$ such that if $G$ is a random $d$-regular graph on $n\geq N$ vertices, then with probability at least $1-\varepsilon$ the following holds. We have that every $\delta$-almost eigenvector $v$ of $G$ (with entry sum $0$) has the property that ${\rm distr}(\sqrt{n}v)$ is at most $\varepsilon$-far from some Gaussian distribution $N(0,\sigma)$ in the weak topology where $0\leq\sigma\leq 1$.
\end{theorem}
Note that if ${\rm distr}(\sqrt{n}v)$ is close to the degenerate distribution $N(0,0)$ then most of the $\ell^2$ weight of $v$ is concentrated on $o(n)$ points. Such vectors are called localized. In general, if $\sigma$ is smaller than, $1$ then some of the $\ell^2$ weight is concentrated on $o(n)$ vertices and the rest is Gaussian.
To formulate our main theorem in the strong form we need some more notation. Recall that $T_d$ denotes the infinite $d$-regular tree and $o$ is a distinguished vertex called ${\it root}$ in $T_d$. We will denote the vertex set $V(T_d)$ of $T_d$ by $V_d$. For two vertices in a graph we write $v\sim w$ if they are connected to each other. Let $[n]$ denote the set $\{1,2,\dots,n\}$ and let $G$ be a $d$-regular graph on the vertex set $[n]$. We denote by ${\rm Hom}^*(T_d,G)$ the set of all covering maps from $T_d$ to $G$. In other words ${\rm Hom}^*(T_d,G)$ is the set of maps $\phi:V_d\rightarrow V(G)$ such that for every vertex $v\in V_d$ the neighbors of $v$ are mapped bijectively to the neighbors of $\phi(v)$.
The set ${\rm Hom}^*(T_d,G)$ has a natural probability measure. We first choose the image of $o$ uniformly at random in $V(G)$. Then we recursively extend the map $\phi$ to larger and larger neighborhoods of $o$ in a random (conditionally independent) way preserving the local bijectivity. It is easy to see that this probability distribution is independent from the choice of $o$.
Let $X$ be a topological space and $f:[n]\rightarrow X$ be a function. We define the probability distribution ${\rm distr}^*(f,G)$ on $X^{V_d}$ as the distribution $f\circ \phi$ where $\phi$ is a random covering in ${\rm Hom}^*(T_d,G)$.
In other words ${\rm distr}^*(f,G)$ is a random lift of $f$ to $T_d$ using a random covering of $G$ with $T_d$. By regarding vectors $v\in\mathbb{R}^n$ as functions form $[n]$ to $\mathbb{R}$ it makes sense to use ${\rm distr}^*(v,G)$.
To formulate our main theorem we need the concepts of eigenvector processes and Gaussian waves on $T_d$ (see \cite{wave}). An {\em eigenvector process} with eigenvalue $\lambda$ is a joint distribution $\{X_v\}_{v\in V_d}$ of real valued random variables with variance 1 such that it is ${\rm Aut}(T_d)$ invariant and satisfies the eigenvector equation
\begin{equation}\label{eq:sv}\sum_{v\sim o} X_v=\lambda X_o\end{equation} with probability $1$.
Note that the group invariance implies that the eigenvector equation is satisfied at every vertex on $T_d$.
We call an eigenvector process {\em trivial} if $\mathbb E(X_o)\neq 0$. Notice that the triviality of an eigenvector process implies that $\lambda=d$. This follows by taking expectation in \eqref{eq:sv} and using the invariance of the process. Furthermore, trivial eigenvector processes are constants in the sense that $X_v=X_w$ holds with probability one for every pair of vertices $v,w$ in $V_d$.
A {\em Gaussian wave} is an eigenvector process whose joint distribution is Gaussian. It is proved in \cite{wave} that for every $-d\leq\lambda\leq d$ there is a unique Gaussian wave $\Psi_\lambda$ with eigenvalue $\lambda$.
Let us choose a fix metrization of the weak topology on $\mathbb{R}^{V_d}$. Our main theorem on random $d$-regular graphs is the following.
\begin{theorem}[Main theorem]\label{main} For every $\varepsilon>0$ there exist constants $N, \delta$ such that if $G$ is a random $d$-regular graph on $n\geq N$ vertices then with probability at least $1-\varepsilon$ the following holds. For every $\delta$-almost eigenvector $v$ of $G$ (with entry sum $0$) has the property that ${\rm distr}^*(\sqrt{n}v,G)$ is at most $\varepsilon$-far (in the weak topology) from some Gaussian wave $\Psi_\lambda$ with $|\lambda|\leq 2\sqrt{d-1}$.
\end{theorem}
\begin{corollary} For every $\varepsilon>0$ and $k\in \mathbb N$ there exist constants $N, \delta$ such that if $G$ is a random $d$-regular graph on $n\geq N$ vertices then with probability at least $1-\varepsilon$ the following holds. For every $k$-tuple of $\delta$-almost eigenvectors $Q=(v^{(1)}, v^{(2)}, \ldots, v^{(k)})$ of $G$ (with entry sum $0$) corresponding to eigenvalues $\lambda_1, \ldots, \lambda_k$ with the property $|\lambda_i-\lambda_j|<\delta$ we have that ${\rm distr}^*(\sqrt{n}Q, G)$ is at most $\varepsilon$-far (in the weak topology) from some Gaussian wave $\Psi_\lambda$ with $|\lambda|\leq 2\sqrt{d-1}$.
\end{corollary}
\section{Preliminaries}
\label{prelim}
\subsubsection*{Invariant processes}
For a separable metric space $Y$ we denote by $I_d(Y)$ the set of Borel probability measures on $Y^{V_d}$ that are invariant under the automorphisms of the tree. More precisely, for every $\tau\in \mathrm{Aut}(T_d)$ (not necessarily fixing the root), the probability measure on $Y^{V_d}$ is required to be invariant under the natural $Y^{V_d}\rightarrow Y^{V_d}$ map induced by $\tau$. Note that eigenvector processes are in $I_d(\mathbb{R})$. If $\mu\in I_d(Y)$ and $F\subseteq V_d$, then we denote by $\mu_F$ the marginal distribution of $\mu$ at $F$. We can equivalently think of $\mu\in I_d(Y)$ as a joint distribution $\{X_v\}_{v\in V_d}$ of $Y$-valued random variables that is invariant under the automorphism group of $T_d$. In this language $\mu_F$ is the same as the joint distribution $\{X_v\}_{v\in F}$. If both $F$ and $Y$ are finite then $\mu_F$ is a probability distribution on the finite set $Y^F$. In this case we denote the entropy of $\mu_F$ by $\mathbb{H}(F)$. By invariance of $\mu$ the quantity $\mathbb{H}(F)$ depends only on the isomorphism class of $F$.
We will consider convergence in $I_d(Y)$ with respect to the weak topology. Since the weak topology is metrizable we can always choose a fixed metrization of it in advance.
\subsubsection*{Almost eigenvectors}
Let $A\in \mathbb R^{n\times n}$ be a matrix. An $\varepsilon$-almost eigenvector of $A$ (with eigenvalue $\lambda$) is a vector $v\in \mathbb R^n$ such that $\|v\|_2=1$ and $\|Av-\lambda v\|_2\leq \varepsilon$.
\begin{lemma} \label{lem:almost}Let $A$ be the adjacency matrix of a $d$-regular graph on $n$ vertices. Let $\lambda_2(A)$ denote the second largest (in absolute value) eigenvalue of $A$. Let $v$ be an $\varepsilon$-almost eigenvector with eigenvalue $\lambda$ such that the entry sum of $v$ is $0$. Then $|\lambda|\leq \lambda_2(A)+\varepsilon$.
\end{lemma}
\begin{proof}
Since $v$ has $0$ entry sum, we can write $v=\sum a_i v_i$, where each $v_i$ is a nonconstant eigenvector of $A$ with eigenvalue $\lambda_i$. It follows that $(A-\lambda I)v=\sum a_i (\lambda_i-\lambda)v_i$, where $|\lambda_i|\leq \lambda_2(A)$. Thus, $\varepsilon^2\geq \|(A-\lambda I)v\|^2_2=\sum a_i^2 (\lambda_i-\lambda)^2$. Suppose that $|\lambda|\geq \lambda_2(A)$ (otherwise the statement is trivial). Then $|\lambda_i-\lambda|\geq |\lambda|-\lambda_2(A)$. Therefore $\varepsilon^2\geq (\sum a_i^2)(|\lambda|-\lambda_2(A))^2$, which completes the proof by using that $\sum a_i^2=\|v\|_2^2=1$. \hfill $\square$
\end{proof}
As we mentioned in the introduction, we will give examples for both localized and delocalized almost eigenvectors on essentially large girth $d$-regular graphs. (Note that a graph is called essentially large girth if most vertices are not contained in short cycles.) The purpose of these examples is to show that our results on the almost eigenvectors of random regular graphs are best possible in the sense that all $0\leq \sigma \leq 1$ can indeed occur in the statement.
We will need some preparation. For $k\geq 1$ and $x\in [-1,1]$ let
\begin{equation}\label{eq:fkx}f(k,x)=\frac{1}{\sqrt{d(d-1)^{k-1}}}q_k(x),\end{equation} where
\begin{equation}\label{eq:qkx}q_k(x)=\sqrt{\frac{d-1}{d}}U_k(x)-\frac{1}{\sqrt{d(d-1)}}U_{k-2}(x); \qquad U_k(\cos \vartheta)=\frac{\sin((k+1)\vartheta)}{\sin \vartheta}.\end{equation}
($U_k(x)$ is the Chebyshev polynomial of the second kind.) Let $g_{\lambda}: V_d\rightarrow \mathbb R$ be the function defined by \[g_{\lambda}(v)=f(|v|, \lambda/(2\sqrt{d-1})),\] where $|v|$ denotes the distance of $v$ and the root $o$. It is easy to see (and well known) that $g_{\lambda}$ satisfies the eigenvector equation with eigenvalue $\lambda$ at every vertex $v$ (for $\lambda\in [-2\sqrt{d-1}, 2\sqrt{d-1}]$).
Now let $G$ be a $d$-regular graph such that there is a vertex $w\in V(G)$ with the property that the shortest cycle containing $w$ has length at least $2n$. We define the function $g'_{\lambda}: V(G)\rightarrow \mathbb R$ by $g'_{\lambda}(v)
=f(d(w,v), \lambda/{2\sqrt{d-1}})$ if $d(v,w)<n$ and $0$ otherwise. It is easy to see that $u=g'_{\lambda}/\|g'_{\lambda}\|_2$ is an almost eigenvector with eigenvalue $\lambda$ with error tending to $0$ as $n\rightarrow \infty$. Furthermore, if $|V(G)|$ is much larger than $n$, then $u$ is close to the constant $0$ distribution in the weak topology. Thus we obtain examples for completely localized almost eigenvectors for $d$-regular graphs (for all $\lambda\in [-2\sqrt{d-1}, 2\sqrt{d-1}]$), which corresponds to the case $\sigma=0$.
We switch to the delocalized example. In \cite{harangi}, the authors construct eigenvector processes on $T_d$ for every $\lambda\in [-2\sqrt{d-1}, 2\sqrt{d-1}]$ that are weak limits of factor of i.i.d.\ processes. These processes have the property that they can be arbitrarily well approximated on any essential large girth $d$-regular graph. It is easy to see that these approximations are almost eigenvectors that are completely delocalized. This corresponds to the case $\sigma=1$. Finally, every $\sigma$ occurs by mixing completely localized and completely delocalized almost eigenvectors corresponding to the same $\lambda$.
\section{Eigenvectors and eigenvector processes on the tree}
\label{chap:eigen}
For a vertex set $F\subseteq V_d$ we denote by $B_k(F)$ the neighborhood of radius $k$ around $F$.
Let $F\subseteq V_d$ be a subset of the vertices of the tree, and let $f\in \mathbb R^{F}$. We say that $v$ satisfies the eigenvector equation with eigenvalue $\lambda$ if for every $v\in F$ with $B_1(v)\subseteq F$ we have that $\lambda f(v)=\sum_{w\sim v} f(w)$. It is clear that for a fixed $\lambda$ these vectors form a linear subspace of $\mathbb R^F$ that we denote by $W_{\lambda}(F)$. We will need a formula for ${\rm dim}\, W_{\lambda}(F)$ for a family of special finite sets $F$.
Given $F$, we say that $F_0\subseteq F$ is a basis if for all $f\in \mathbb R^{F_0}$ and $\lambda \in \mathbb R$ the subspace $W_{\lambda}(F)$ contains exactly one extension of $f$ to $F$. It is clear that if $F_0$ is a basis in $F$, then ${\rm dim}\, W_{\lambda}(F)=|F_0|$ for all $\lambda\in \mathbb R$.
\begin{lemma}Let $F_0$ be a basis of a path-connected set $F$. Suppose that $v\in F$ and $|F\cap B_1(v)|=2$. Furthermore, let $D\subseteq B_1(v)$ such that $|D|=d-2$ and $D\cap F=\emptyset$. Then $F_0\cup D$ is a basis of $F\cup B_1(v)$. \label{lem:comb}
\end{lemma}
\begin{proof}
Let $f$ be a function from $F_0\cup D$ to $\mathbb R$. By assumption, we have that $f|_{F_0}$ extends to a unique function $\tilde f$ on $F$. Now using the eigenvector equation at $v$, we obtain a unique value for $B_1(v)\setminus (D \cup F)$. Note that the connectivity of $F$ implies that the function constructed this way is in $W_{\lambda}(F\cup B_1(v))$.
\end{proof}\hfill $\square$
We will use $C$ to denote the star $B_1(o)$ and we will use $e$ to denote a distinguished edge in $T_d$.
It is clear that if $v$ is a neighbor of $o$, then $C\setminus \{v\}$ is a basis of $C$. Similarly $e$ is a basis of itself. Using Lemma \ref{lem:comb} and induction, we obtain that
\begin{equation}\label{eq:dim} {\rm dim}\, W_{\lambda}(B_k(C))=|\partial B_k(C)|=d(d-1)^{k}; \qquad {\rm dim}\, W_{\lambda}(B_k(e))=|\partial B_k(e)|=2(d-1)^k \end{equation}
holds for every $\lambda \in \mathbb R$ and $k \in \mathbb N$, where $\partial F$ denotes the boundary of a set $F$.
Recall that an invariant process $\{X_v\}_{v\in V_d}$ is an eigenvector process with eigenvalue $\lambda$ if it satisfies the eigenvector equation \eqref{eq:sv} with probability $1$ and $\mathrm{Var}(X_v)=1$. Notice that if $F$ is any vertex set in $T_d$, then $\{X_v\}_{v\in F}$ is supported on $W_{\lambda}(F)$. In the rest of this chapter we investigate the joint distribution $\{X_v\}_{v\in S}$ in a non-trivial eigenvector process where $S$ is one of $B_k(C)$ or $B_k(e)$ for some $k\in\mathbb{N}$. Our goal is to find uncorrelated linear combinations of the variables $\{X_v\}_{v\in S}$ with the property that they linearly generate every random variable in $\{X_v\}_{v\in S}$. Observe that since the covariance matrix of $\{X_v\}_{v\in S}$ has rank at most ${\rm dim}(W_\lambda(S))=|\partial S|$, it is enough to find that many uncorrelated linear combinations with nontrivial variance to guarantee that they generate everything.
Let $Y=(Y_1, Y_2,\dots,Y_n)$ be an $n$-dimensional distribution with $0$ mean such that the covariance matrix is of full rank. We can always find a linear transformation $T:\mathbb{R}^n\rightarrow\mathbb{R}^n$ with the property that the covariance matrix of $TY$ is the identity matrix. The probability distribution $TY$ is unique up to an orthogonal transformation on $\mathbb{R}^n$. We call it the standardized version of $Y$.
Let $\mu\in I_d(\mathbb{R})$ represented by an invariant joint distribution $\{X_v\}_{v\in V_d}$ of random variables. Assume that $\mathbb{E}(X_o)=0$. With every directed edge $(v,w)$ of $T_d$ we associate a probability distribution on $\mathbb{R}^{d-2}$ defined up to an orthogonal transformation. Let $\{v_i\}_{i=1}^{d-1}$ be the set of neighbors of $w$ different from $v$. We denote by $A_{v,w}$ the standardized version of $\{X_{v_i}-X_{v_{d-1}}\}_{i=1}^{d-2}$. It is easy to see that (the orthogonal equivalence class of) $A_{v,w}$ does not depend on the labeling of the vectors $v_i$.
We introduce the symmetric relation $r$ on directed edges of $T_d$ such that $((x,y),(v,w))\in r$ if and only if the unique shortest path connecting $y$ and $w$ contains at least one of $x$ and $v$. The next lemma implies that if $((x,y),(v,w))\in r$ then $A_{x,y}$ and $A_{v,w}$ are uncorrelated and so if $\mu$ is Gaussian then $A_{x,y}$ and $A_{v,w}$ are independent.
\begin{lemma}\label{arcuncorr} Let $(v,w)$ be a directed edge in $T_d$. Let $\{v_i\}_{i=1}^{d-1}$ be the set of neighbors of $w$ different from $v$. Let $X=\sum_{j=1}^k a_jX_{u_j}$ be a linear combination such that the shortest path connecting $w$ and $u_j$ contains $v$ for every $1\leq j\leq k$. Then $\mathbb{E}(X(X_{v_i}-X_{v_{d-1}}))=0$ holds for $1\leq i\leq d-2$.
\end{lemma}
\begin{proof} The condition on the vertices $u_j$ guarantees that for every fix $j$ the distance of $u_j$ from $v_i$ does not depend on $i$. Using this and the automorphism invariance of $\mu$ it follows that $\mathbb{E}(X_{u_j}(X_{v_i}-X_{v_{d-1}}))=0$ holds for every $1\leq j\leq k$ and $1\leq i\leq d-2$. By linearity of expected value the proof is complete. \hfill $\square$
\end{proof}
Let $S\subset V_d$ be either $B_k(C)$ or $B_k(e)$ for some $k\in\mathbb{N}$. Assume that $\mu$ is an eigenvector process. Let $p\in S$ be such that it has distance at least one from the boundary $\partial S$.
Let $D$ denote the set of directed edges $(v,w)$ inside $S$ with the following three properties: $(a)$ the unique shortest path connecting $p$ and $w$ contains $v$, $(b)$ $p\notin\{v,w\}$, $(c)$ $B_1(w)\subset S$. Let $B_p$ denote the joint distribution $(X_v-X_p)_{v\sim p}$. We denote by $\mathcal{Q}(S,p,\mu)$ the joint distribution of the random variables $\{A_{v,w}\}_{(v,w)\in D}$ and $B_p$. Lemma \ref{arcuncorr} implies that the components of $\mathcal{Q}(S,p,\mu)$ are uncorrelated multidimensional random variables. It is easy to see that if $|\lambda|<d$, then the correlation matrix of each such multidimensional random variable is of full rank. This implies that if $|\lambda|<d$, then $\mathcal{Q}(S, p, \mu)$ provides a linear basis for $\{X_v\}_{v\in S}$. By counting dimensions, this yields the following corollary.
\begin{corollary}\label{cor:qsp}
Let either $S=B_k(C)$ or $S=B_k(e)$ and $\mu$ an eigenvector process with eigenvalue $|\lambda|<d$. Then we have that
\[\langle {\rm supp}\ (\mu_S)\rangle_{\mathbb R}=W_{\lambda}(S).\]
\end{corollary}
\section{Typical processes and the limiting form of the main theorem}
\label{typpro}
Now we describe a limiting form of our main theorem using typical processes on $T_d$. Typical processes on $T_d$ were first introduced in \cite{BSz} to study the properties of random regular graphs via ergodic theory.
Here we use a slightly different definition which extends the original notion to processes that take values in a separable, metrizable space $Y$.
\begin{definition} Let $Y$ be separable, metrizable topological space and let $\mu$ be a Borel probability distribution on $Y^{V_d}$. We say that $\mu$ is a typical process if there is a growing sequence of natural numbers $\{n_i\}$ with the following property. Assume that $\{G_i\}_{i=1}^n$ is a random graph sequence whose elements $G_i$ are independently chosen random $d$-regular graphs on $n_i$ vertices. Then with probability $1$ there are maps $f_i:V(G_i)\rightarrow Y$ such that the distributions ${\rm distr}^*(f_i,G_i)$ are converging to $\mu$ in the weak topology.
\end{definition}
Our main theorem in the limit setting is the following.
\begin{theorem}[Limiting form of the main theorem]\label{mainlim} If $\mu$ is a nontrivial typical eigenvector process with eigenvalue $\lambda$, then $|\lambda|\leq 2\sqrt{d-1}$ and $\mu$ is the Gaussian wave $\Psi_\lambda$.
\end{theorem}
Using the fact that weak limits of factor of i.i.d processes are typical (see \cite{BSz}) we obtain the next corollary which answers a question of B. Vir\'ag.
\begin{corollary}\label{mainlimcorr} If $\mu$ is a nontrivial eigenvector process that is a weak limit of factor of i.i.d.\ processes, then $\mu$ is a Gaussian wave with eigenvalue $|\lambda|\leq 2\sqrt{d-1}$.
\end{corollary}
Note that Corollary \ref{mainlimcorr} implies that if many eigenvector processes corresponding to the same eigenvalue are coupled in a way that the coupling is a weak limit of factor of i.i.d.\ processes, then its distribution is jointly Gaussian.
We emphasize that the first part of the statement in Theorem \ref{mainlim}, namely that $|\lambda|\leq 2\sqrt{d-1}$ is a consequence of Friedman's theorem \cite{friedman}. We prove this implication in Lemma \ref{lem:equiv}. In addition, the main goal of this chapter is to show that Theorem \ref{mainlim} implies Theorem \ref{main}.
\begin{lemma}\label{lem:friedman}
If $\mu$ is a nontrivial typical eigenvector process with eigenvalue $\lambda$, then $|\lambda|\leq 2\sqrt{d-1}$.
\end{lemma}
\begin{proof}
Our first goal is to prove that there exists a sequence of $d$-regular graphs $\{G_i\}_{i=1}^{\infty}$ and vectors $\{f_i: V(G_i)\rightarrow \mathbb R\}_{i=1}^{\infty}$ such that $(a)$ $|\lambda_2(G_i)|\rightarrow 2\sqrt{d-1}$, where $\lambda_2(G)$ denotes the second largest (in absolute value) eigenvalue of a finite graph $G$; $(b)$ $\mathrm{distr}^*(f_i, G_i)\rightarrow \mu$ in the weak topology. Using that $\mu$ is typical, there exists a sequence $\{n_i\}_{i=1}^{\infty}$ of growing natural numbers such that with probability 1, if $\{G'_i\}_{i=1}^{\infty}$ is a sequence of independent random $d$-regular graphs with $|V(G'_i)|=n_i$, then there exists a sequence of functions $\{f'_i\}_{i=1}^{\infty}$ satisfying $(b)$. Friedman's theorem \cite{friedman} implies that with probability 1 we can choose a further subsequence satisfying $(a)$.
We fix an arbitrary $\varepsilon>0$. Let $X_o$ be a random variable with distribution $\mu_o$. Since $\mathbb E(X^2_o)=1$, we can find $k>0$ such that $k$ is a continuity point of the cumulative distribution function of $X_o$ and $k^2\mathbb P(|X_o|>k)<\varepsilon$. Furthermore, if $k$ is sufficiently large, we can also assume that $\mathbb E([X_o]^2_k)\geq 1/2$ and $|\mathbb E([X_o]_k)|\leq \varepsilon$, where $[x]_k=\max(\min(x, k), -k)$. Using that $\mathrm{distr}^*(f_i, G_i)$ converges to the eigenvector process $\mu$, we obtain that if $i$ is large enough, then $(i)$ $k^2\mathbb P(\mathrm{distr}(|f_i|)>k)<2\varepsilon$; $(ii)$ $\mathbb P(\mathrm{distr}(|(G_i-\lambda I)f_i|)\geq \varepsilon)\leq \varepsilon/k^2$; $(iii)$ $\mathbb E(\mathrm{distr}([f_i]_k^2))\geq 1/3$; $(iv)$ $c_i=|\mathbb E(\mathrm{distr}([f_i]_k))|\leq 2\varepsilon$. Note that $(i)$ implies that $f_i\neq [f_i]_k$ holds on a vertex set of density at most $2\varepsilon/k^2$. It follows that $(G_i-\lambda I)[f_i]_k\neq (G_i-\lambda I)f_i$ holds on a vertex of density at most $(d+1)2\varepsilon/k^2$. Furthermore, we have that $\|(G_i-\lambda I)[f_i]_k\|_{\infty}\leq (d+|\lambda|)k$. Putting all this together, we obtain that for $i$ large enough
\[\|(G_i-\lambda I)([f_i]_k-c_i)\|_2\leq \sqrt{\varepsilon^2n_i+(2d+3)(\varepsilon/k^2)\cdot n_i(d+|\lambda|)^2k^2}+|d-\lambda|c_i\sqrt{n_i}\]\[\leq\sqrt{n_i}\big(\sqrt{\varepsilon^2+(2d+3)\varepsilon(d+|\lambda|)^2}+2|d-\lambda|\varepsilon\big).\]
Let $v_i=([f_i]_{k}-c_i)/\|[f_i]_{k}-c_i\|_2$. Using that $(iii)$ implies that $\|[f_i]_k\|_2^2\geq 1/3n_i$, we obtain that for an appropriate choice of small enough $\varepsilon$ and large enough $i$ the quantity $\|(G_i-\lambda I)v_i\|_2$ is arbitrarily small.
Thus, by using $(a)$ and Lemma \ref{lem:almost}, we get that $|\lambda|\leq 2\sqrt{d-1}$.
\hfill $\square$
\end{proof}
\begin{proposition} \label{lem:equiv} Theorem \ref{mainlim} implies Theorem \ref{main}.
\end{proposition}
We need a few notions and lemmas. Let $\mathcal{P}$ denote the set of Borel probability distributions $\mu$ on $\mathbb{R}^{V_d}$ which have a second moment bounded from above by $1$ at each coordinate.
By tightness of $\mathcal P$, we have that $\mathcal{P}$ is compact with respect to the weak topology of measures. Let $m$ be a fixed metrization of the weak topology on $\mathcal{P}$. Let us denote by $\mathcal{T}$ the set of closed subsets in $\mathcal{P}$, and by $d_H$ the the Hausdorff metric on $\mathcal{T}$. We have that $d_H$ induces a compact topology on $\mathcal{T}$.
Let us define the distance $m^*$ for $d$-regular graphs in the following way. If $G_1$ and $G_2$ are $d$-regular graphs then $m^*(G_1,G_2)$ is the infimum of the numbers $\delta$ with the property the if $f_1:V(G_1)\rightarrow \mathbb{R},f_2:V(G_2)\rightarrow \mathbb{R}$ are arbitrary functions with $\mathbb{E}({\rm distr}(f_1^2)),\mathbb{E}({\rm distr}(f_2^2))\leq 1$ then there are functions $f_1':V(G_2)\rightarrow\mathbb{R}, f_2':V(G_1)\rightarrow \mathbb{R}$ with $\mathbb{E}({\rm distr}(f_1'^2)),\mathbb{E}({\rm distr}(f_2'^2))\leq 1$ such that $$m({\rm distr}^*(f_1,G_1),{\rm distr}^*(f_1',G_2))\leq\delta ~~~,~~~m({\rm distr}^*(f_2,G_2),{\rm distr}^*(f_2',G_1))\leq\delta.$$
We can describe the metric $m^*$ in terms of the metric $d_H$ as follows. For a graph $G$, let
\[S(G)=\{\mathrm{distr}^*(f, G)|f\in \mathbb{R}^{V(G)},\mathbb{E}(\mathbb{\rm distr}(f^2))\leq 1\}.\]
We have that $m^*(G_1, G_2)=d_H(S(G_1), S(G_2))$.
\begin{definition} We say that a finite $d$-regular graph $G$ is $\varepsilon$-typical for some $\varepsilon>0$ if with probability at least $1-\varepsilon$ a random $d$-regular graph $G'$ on $|V(G)|$ vertices has the property that $m^*(G, G')<\varepsilon$.
\end{definition}
\begin{lemma} \label{lem:lip}For every $\varepsilon>0$ and $T\in \mathcal T$ there exists $n(\varepsilon)$ such that for all $N>n(\varepsilon)$ there exists a value $c$ satisfying
\begin{equation}\label{eq:dh}\mathbb P(|d_H(S(G), T)-c|>\varepsilon)<\varepsilon.\end{equation}
\end{lemma}
\begin{proof}The proof relies on a certain continuity property of the metric $m^*$ with respect to small changes in a graph. More precisely, we show that for every $\varepsilon_2$ if $N$ is large enough, then for every pair $G, G'$ of $d$-regular graphs on the vertex set $[N]$ satisfying $|E(G)\Delta E(G')|\leq 4$ we have that $m^*(G, G')<\varepsilon_2$. The significance of the number $4$ comes from the fact that $d$-regular graphs can be transformed into each other through a sequence of operations in which two independent edges $(u_1, v_1), (u_2, v_2)$ are replaced by $(u_1, v_2), (u_2, v_1)$. To prove the continuity property, we show that if $N$ is large enough, then the inequality
\[m(\mathrm{distr}^*(f, G), \mathrm{distr}^*(f, G'))<\varepsilon_2\]
holds for every function $f: [N]\rightarrow \mathbb{R}$ with $\mathbb{E}(f^2)\leq 1$. Let $k$ be an arbitrary integer. Observe that the marginal distribution $\mathrm{distr}^*(f, G)|_{B_k(o)}$ can be obtained from the distribution of $\mathbb{R}$-colored neighborhoods of radius $k$ of a random vertex in $G$ (and the analogous statement holds for $G'$). Since $|E(G)\Delta E(G')|$ intersects such a neighborhood with probability tending to 0 as $|V(G)|\rightarrow \infty$, we have that the distance between $\mathrm{distr}^*(f, G)|_{B_k(o)}$ and $\mathrm{distr}^*(f, G')|_{B_k(o)}$ converges to 0 in any metrization of the weak topology of $\mathbb{R}^{B_k(o)}$. This implies the desired continuity property.
We obtain by the above statement and the triangle inequality that if $|E(G)\Delta E(G')|\leq 4$ and $N$ is large enough, then $|d_H(S(G), T)-d_H(S(G'), T)|<\varepsilon_2$.
It is well-known that graph parameters on random $d$-regular graphs that satisfy this Lipschitz property are concentrated around their mean; see \cite[Theorem 2.19]{wormald}.
\end{proof} \hfill $\square$
\begin{lemma}\label{lem:typgraph}
For every $\varepsilon>0$ there exists $n(\varepsilon)$ such that for every $N>n(\varepsilon)$ with probability at least $1-\varepsilon$ a random $d$-regular graph on $N$ vertices is $\varepsilon$-typical.
\end{lemma}
\begin{proof}Let $M$ be a finite $\varepsilon/2$-net in $\mathcal T$. If $G$ is a random $d$-regular graph on $N$ vertices, then there exists $T\in M$ with the property that
\begin{equation}\label{eq:ph}\mathbb P(d_H(S(G), T)<\varepsilon/2)\geq 1/|M|. \end{equation}
We apply Lemma \ref{lem:lip} with $\varepsilon'\leq \varepsilon/4$. Combining inequalities \eqref{eq:dh} and \eqref{eq:ph}, we obtain that for $\varepsilon'<1/|M|$ and $N$ large enough $|c|\leq 3/4\varepsilon$. Then applying \eqref{eq:dh} again, the proof is complete.
\end{proof} \hfill $\square$
Now we enter the proof of Lemma \ref{lem:equiv}.
\begin{proof} We go by contradiction. If Theorem \ref{main} fails then there is a growing sequence of natural numbers $\{n_i\}_{i=1}^\infty$, $\varepsilon>0$ and a sequence $\{\delta_i\}_{i=1}^\infty$ with $\lim_{i\rightarrow\infty}\delta_i=0$ such that the following holds. If $G$ is a random $d$-regular graph on $n_i$ vertices then we have with probability at least $\varepsilon$ that there is an $\delta_i$-almost eigenvector $v$ of $G$ (with entry sum $0$) such that ${\rm distr}^*(v,G)$ is at least $\varepsilon$-separated from any Gaussian wave in the weak topology.
From Lemma \ref{lem:typgraph} we obtain that there is a sequence $\{\varepsilon'_i\}_{i=1}^\infty$ with $\lim_{i\rightarrow \infty} \varepsilon_i'=0$ such that a random $d$-regular graph on $n_i$ vertices is $\varepsilon_i'$-typical with probability at least $1-\varepsilon_i'$ for every $i$. There exists an index $j$ such that for all $i\geq j$ we have $\varepsilon_i'<\varepsilon$. This implies that for all $i\geq j$ we can choose a graph $G_i$ on $n_i$ vertices such that $G_i$ is $\varepsilon_i'$-typical and there exists a $\delta_i$-almost eigenvector $f_i$ of $G_i$ (with entry sum 0) satisfying that ${\rm distr}^*(\sqrt{n_i}f_i, G_i)$ is at least $\varepsilon$-separated from any Gaussian wave in the weak topology.
By choosing a subsequence we can assume (by abusing the notation) that ${\rm distr}^*(\sqrt{n_i}f_i,G_i)$ weakly converges to some measure $\mu\in\mathcal{P}$. It is clear that $\mu$ is a nontrivial eigenvector process which is at least $\varepsilon$-separated from any Gaussian wave in the weak topology. To get a contradiction it remains to show that $\mu$ is typical.
Again by choosing a subsequence we can assume that $\sum_{i=1}^\infty\varepsilon'_i<\infty$. Let $\{G'_i\}_{i=1}^\infty$ be such that $G_i'$ is a random $d$-regular graph on $n_i$ vertices and the terms of the sequence are independent. It follows from the Borel--Cantelli lemma that almost surely all but finitely many indices $i$ satisfy that $m^*(G_i',G_i)\leq\varepsilon'_i$. For such indices we can find $f_i'$ with $m({\rm distr}^*(f_i',G'_i),{\rm distr}^*(\sqrt{n_i}f_i,G_i))\leq\varepsilon'_i$. We obtain that ${\rm distr}^*(f_i',G_i')$ converges to $\mu$ showing that $\mu$ is typical.
\end{proof} \hfill $\square$
\section{Entropy inequality for typical processes}
\label{chap:entr}
Let $X$ be a separable metric space; let $F$ be a finite set and let $\mathcal{P}(F)$ denote the set of probability distributions on $F$ equipped with the topology generated by total variation distance. A continuous discretization of $X$ is a continuous function $\phi:X\rightarrow\mathcal{P}(F)$.
If $\alpha\in X^V$ is an $X$-coloring of a finite or countable set $V$, then we denote by $\phi*\alpha$ the probability distribution on $F^V$ obtained by independently choosing an element from $F$ for each $v\in V$ with distribution $\phi(\alpha(v))$.
If $\mu$ is a probability distribution on $X^V$, then we denote by $\phi*\mu$ the probability distribution obtained by first taking a $\mu$ random element $\alpha:V\rightarrow X$ and then in a second round of randomization we take $\phi*\alpha$.
The main result of this chapter is the next entropy inequality for typical processes.
\begin{theorem}\label{discmain} If $\mu\in I_d(X)$ is a typical process and $\phi:X\rightarrow\mathcal{P}(F)$ is a continuous discretization, then the process $\phi*\mu$ satisfies the next entropy inequality.
$$\mathbb{H}(B_k(C))-(d/2)\mathbb{H}(B_k(e))\geq\mathbb{E}_{\mu_o}(\mathbb{H}(\phi(x))).$$
\end{theorem}
Before proving Theorem \ref{discmain} we need some preparation. Let us fix a metrization of the weak topology on $I_d(X)$. For a finite $d$-regular graph $G$ let $\omega(G)$ denote the infimum of the numbers $\varepsilon>0$ for which it is true that at least $1-\varepsilon$ fraction of the vertices of $G$ are not contained in a cycle of length at most $\lfloor 1/\varepsilon\rfloor$. The quantity $\omega(G)$ measures how similar the graph $G$ is to the tree $T_d$ in the Benjamini--Schramm metric. Throughout this chapter, $G$ is always assumed to be finite.
The operator ${\rm distr}^*$ maps $X$-colored $d$-regular graphs $(\alpha\in X^{V(G)},G)$ to invariant processes in $I_d(X)$. Using this correspondence and the metric on $I_d(X)$ we define the distance of an $X$-colored graph $(\alpha,G)$ and a process $\mu\in I_d(X)$ as $\omega(G)$ plus the distance of ${\rm distr}^*(\alpha,G)$ and $\mu$ in $I_d$. Note that if $(\alpha\in X^{V(G)},G)$ is an $X$-colored $d$-regular graph, then $(\phi*\alpha,G)$ is a probability distribution on $F$-valued colorings of the vertices of $G$. In this case ${\rm distr}^*(\phi*\alpha,G)$ is a probability distribution on $I_d(X)$, while $\phi*{\rm distr}^*(\alpha,G)$ is a single element in $I_d(X)$.
\begin{proposition}\label{discapprox} Let $\mu\in I_d(X)$ be an invariant process and $\phi$ is a continuous discretization of $X$. Then for every $\varepsilon>0$ there is $\delta>0$ such that if a colored $d$-regular graph $(\alpha\in X^{V(G)},G)$ is at most of distance $\delta$ from $\mu$, then with probability at least $1-\varepsilon$ we have that $(\phi*\alpha,G)$ is at most of distance $\varepsilon$ from $\phi*\mu$.
\end{proposition}
The proof of Proposition \ref{discapprox} relies on the next technical lemma.
\begin{lemma}\label{discapproxl} Let $\mu\in I_d(X)$ be an invariant process. Let $r\in\mathbb{N}$, $B=B_r(o)\subset V_d$ and let $\beta:F^B\rightarrow\mathbb{R}$ be any automorphism invariant function. Then for every $\varepsilon>0$ there is $\delta>0$ such that if a colored $d$-regular graph $(\alpha\in X^{V(G)},G)$ is at most of distance $\delta$ from $\mu$ then with probability at least $1-\varepsilon$ we have that
$$|W|^{-1}\sum_{v\in W} \beta(\gamma|_{B_r(v)})$$ is at most $\varepsilon$-far from $\mathbb{E}(\beta(\phi*\mu_B))$ where $\gamma=\phi*\alpha$ and $W=\{v: v\in V(G), B_r(v)\simeq B\}$.
\end{lemma}
\begin{proof} Assume first that $(\alpha\in X^{V(G)},G)$ is an arbitrary $d$-regular $X$-colored graph whose distance is $\delta'$ from $\mu$ and let $\gamma,W$ as in the statement of the lemma. For a vertex $v\in W$ let $Y_v$ be the random variable with value $\beta(\gamma|_{B_r(v)})$. Let $g: X^B\rightarrow\mathbb{R}$ be the function defined by $g(h)=\mathbb{E}(\beta(\phi*h))$. We have for $v\in W$ that $\mathbb{E}(Y_v)$ is equal to $g(\alpha|_{B_r(v)})$. Let $Y=|W|^{-1}\sum_{v\in W}Y_v$. It follows that
\begin{equation}\label{tpei1} \mathbb{E}(Y)=|W|^{-1}\sum_{v\in W} g(\alpha|_{B_r(v)})=\int_{X^B}g~ d\nu_G,
\end{equation}
where $\nu_G$ describes the probability distribution of the isomorphism classes of $\alpha|_{B_r(v)}$ where $v\in W$ is a uniform random point. Using the fact that $g$ is a continuous function we obtain that if $\delta'$ is small enough then the right hand side of (\ref{tpei1}) is at most $\varepsilon/2$ far from $\int_{X^B}~g~d\mu_B=\mathbb{E}(\beta(\phi*\mu_B))$. Observe that $Y_v$ and $Y_w$ are independent if $v$ and $w$ have distance at least $2r+1$ in $G$. It follows that there are at most $|B_{2r+1}(o)||W|$ correlated pairs in $\{Y_v\}_{v\in W}$. We obtain that the variance of $Y$ is at most $|B_{2r+1}(o)|^{1/2}|W|^{-1/2}\max |\beta|$. We use that $\omega(G)$ goes to $0$ as $\delta'$ goes to $0$ and thus $|W|$ tends to infinity. This implies that if $\delta'$ is sufficiently small then the variance of $Y$ is at most $\varepsilon^2/3$. Now by Chebyshev's inequality we have that $\mathbb{P}(|Y-\mathbb{E}(Y)|\geq \varepsilon/2)\leq \varepsilon^2.$ It follows that $\mathbb{P}(|Y-\mathbb{E}(\beta(\phi*\mu_B))|\geq\varepsilon)\leq \varepsilon^2$ which completes the proof. \hfill $\square$
\end{proof}
We continue with the proof of Proposition \ref{discapprox}.
\begin{proof} Let $\delta'$ be an arbitrary positive number. Let $(\alpha\in X^{V(G)},G)$ be an $X$-colored $d$-regular graph of distance $\delta'$ from $\mu$. Let $(\gamma\in F^{V(G)},G)$ be chosen according to the probability distribution $(\phi*\alpha,G)$. Let $\varepsilon'>0$, $r=\lfloor 1/\varepsilon'\rfloor$ and $W=\{v:v\in V(G), B_r(v)\simeq B\}$. It follows from Lemma \ref{discapproxl} that there is $c=c(\varepsilon')>0$ such that if $\delta'<c$ then the condition of Lemma \ref{discapproxl} holds for $(\gamma,G)$ simultaneously for every $0-1$ valued $\beta$ with probability at least $1-\varepsilon'$. (Here we use the fact that there are finitely many such functions $\beta$.)
If $c$ is small enough, then it also guarantees that $|W|/|V(G)|\geq 1-\varepsilon'$. Now it is clear that if $\varepsilon'$ is small enough, then these properties imply that $(\gamma,G)$ is at most $\varepsilon$ far from $\phi*\mu$.
\end{proof} \hfill $\square$
\medskip
For the next lemma let $R(d,n)$ denote the number of $d$-regular graphs on the vertex set $[n]$. In case $d$ is odd we will always assume that the number of vertices is even.
\begin{lemma}\label{countcolor} Let $F$ be a finite set and $\mu\in I_d(F)$. Let $N(n,\varepsilon)$ denote the number of $d$-regular $F$-colored graphs on the vertex set $[n]$ whose distance from $\mu$ is at most $\varepsilon$. Assume that $\{n_i\}_{i=1}^\infty$ is a growing sequence of natural numbers. Then for every $\varepsilon>0$ and $k\in\mathbb{N}$ we have that
\begin{equation}\label{cceq}
\mathbb{H}(B_k(C))-(d/2)\mathbb{H}(B_k(e))\geq \lim_{\varepsilon\rightarrow 0}\Bigl(\limsup_{i\rightarrow\infty} n_i^{-1}\log(N(n_i,\varepsilon)/R(d,n_i))\Bigr).
\end{equation}
\end{lemma}
\begin{proof} First we prove the statement for $k=0$. In this case we use a formula from \cite{BSz} that approximates the number $N'(n,\varepsilon)$ of $F$-colored graphs on $[n]$ in which the statistics of colored $1$-neighborhoods is $\varepsilon$-close to $\mu_C$. We have that
$$N'(n,\varepsilon)=R(d,n)\mathbb{H}(C)^{(n(1+o(1))}\mathbb{H}(e)^{-(dn/2)(1+o(1))}$$ where $o(1)$ is a quantity which goes to $0$ when first $n\rightarrow\infty$ and then $\varepsilon\rightarrow 0$. This implies that the right hand side of (\ref{cceq}) is equal to the left hand side when $N(n_i,\varepsilon)$ is replaced by $N'(n_i,\varepsilon)$. Now we use that for every $\varepsilon>0$ there exists $\varepsilon'>0$ such that $N'(n_i,\varepsilon)\geq N(n_i,\varepsilon')$ holds for all $i$. This finishes the proof of the first part.
The idea of the proof in case of $k>0$ is to generate a new process from $\mu$ in which the color of every vertex $v\in V_d$ is replaced by the isomorphism type of the colored neighborhood of $v$ of radius $k$. The main difficulty in this approach is that the isomorphism type describes the neighborhood only up to automorphisms which leads to extra constants in the entropy formulas. To control these constants (and to eventually get rid of them) we add some extra randomness to the process.
Let us introduce new processes $\mu_{r,m}$ on $T_d$ for every $k,m\in\mathbb{N}$. If $r=0$, then $\mu_{0,m}$ denotes the $F\times [m]$ valued process in which we generate a $\mu$-random coloring on $T_d$ and then we add a second coordinate from $[m]$ to every vertex independently and uniformly. In general $\mu_{r,m}$ denotes the process obtained from $\mu_{0,m}$ by coloring $v\in V_d$ with the isomorphism class of the coloring of $B_r(v)$ in $\mu_{0,m}$. Let $a_{r,m,k}$ denote the left hand side (resp. $b_{r,m}$ denote the right hand side) of (\ref{cceq}) evaluated for $\mu_{r,m}$ and $k$. It is easy to see that $b_{r,m}$ does not depend on $r$. On the other hand the independence of the two coordinates implies that $b_{0,m}=b_{0,1}+\log m$. All together this means that $b_{r,m}=b_{0,1}+\log m$. It is clear that $a_{0,m,k}=a_{0,1,k}+\log m$. From the case $k=0$ we have that $a_{r,m,0}\geq b_{r,m}$. We can write this as $$ a_{0,1,r}+(a_{0,m,r}-a_{0,1,r})+(a_{r,m,0}-a_{0,m,r})\geq b_{0,1}+\log m$$ and thus $a_{0,1,r}+c_{r,m}\geq b_{0,1}$ holds for every $m$ where $c_{r,m}=a_{r,m,0}-a_{0,m,r}$. Since the inequality $a_{0,1,r}\geq b_{0,1}$ is equivalent to the statement of the lemma for $k=r$ it remains to show that $\lim_{m\rightarrow\infty} c_{r,m}=0$ holds for every $r$.
Let $t_r$ denote the size of the automorphism group of the rooted $d-1$-regular tree of depth $r$.
We claim that $\mathbb{H}_{\mu_{0,m}} (B_r(C))-\mathbb{H}_{\mu_{r,m}}(C)=d\log t_r + o(1)$ and that $\mathbb{H}_{\mu_{0,m}} (B_r(e))-\mathbb{H}_{\mu_{r,m}}(e)=2\log t_r + o(1)$ as $m\rightarrow\infty$. It is clear that this claim implies $c_{r,m}=o(1)$. We show the proof of the first claim. (The proof of the second one is almost identical.)
If $m$ is large enough then in the process $\mu_{0,m}$ restricted to $B_{r+1}(o)$ all labels are different with probability converging to $1$. In such a case knowing the isomorphism classes of the colored neighborhoods of radius $r-1$ of vertices in $C$ is equivalent with knowing the colored version of $B_{r+1}(o)$ up to an isomorphism that fixes $C$. The stabilizer of $C$ in the automorphism group of $B_{r+1}(o)$ is the $d$-th power of the automorphism group of the rooted $d-1$ regular tree of depth $r$. Thus the entropy loss of $\mathbb{H}_{\mu_{r,m}}(C)$ compared to $\mathbb{H}_{\mu_{0,m}} (B_r(C))$ is converging to $d\log t_r$. \hfill $\square$
\end{proof}
Now we arrived to the proof of Theorem \ref{discmain}.
\begin{proof} According to Lemma \ref{countcolor} it is enough to show that the process $\phi*\mu$ satisfies
$$\lim_{\varepsilon\rightarrow 0}\Bigl(\limsup_{i\rightarrow\infty} n_i^{-1}\log(N(n_i,\varepsilon)/R(d,n))\Bigr)\geq\mathbb{E}_{\mu_o}(\mathbb{H}(\phi(x)))$$ for some growing sequence $\{n_i\}_{i=1}^\infty$ of natural numbers.
For $n\in\mathbb{N},\varepsilon>0$ let $a(n,\varepsilon)$ denote the number of $d$-regular graphs $G$ on $[n]$ with the property that there exists an $X$-coloring $\alpha$ of $[n]$ such that $(G,\alpha)$ is of distance at most $\varepsilon$ from $\mu$. The fact that $\mu$ is typical is equivalent to the fact that there is a sequence $\{n_i\}_{i=1}^\infty$ such that $\lim_{i\rightarrow\infty}a(n_i,\varepsilon)/R(d,n_i)=1$ holds for every $\varepsilon>0$.
From Proposition \ref{discapprox} we obtain that for every $\varepsilon_2>0$ there is $\varepsilon>0$ such that if a graph $G$ has an $X$ coloring $\alpha$ of distance at most $\varepsilon$ from $\mu$, then with probability at least $1-\varepsilon_2$ we have that $(\phi*\alpha,G)$ is of distance $\varepsilon_2$ from $\phi*\mu$. It follows that $G$ has at least $\exp(\mathbb{H}(\phi*\alpha,G)+o(1))$
$F$-colorings of distance at most $\varepsilon_2$ from $\phi*\mu$. On the other hand, we have $|V(G)|^{-1}\mathbb{H}(\phi*\alpha,G)=|V(G)|^{-1}\sum_{v\in V(G)}\mathbb{H}(\phi(\alpha(v)))$, which converges to $\mathbb{E}_{\mu_o}(\mathbb{H}(\phi(x)))$ as $\alpha$ converges to $\mu$.
New let $N(n,\varepsilon)$ defined as in Lemma \ref{countcolor} for the process $\phi*\mu$. From the above observations we obtain that $n_i^{-1}\log (N(n_i,\varepsilon)/R(d,n_i))$ can be estimated from below by $\mathbb{E}_{\mu_o}(\mathbb{H}(\phi(x)))-o(1)$ as $n_i$ goes to infinity. Then Lemma \ref{countcolor} finishes the proof. \hfill $\square$
\end{proof}
\section{Smooth eigenvector processes}
\label{smooth}
Let $\mu$ be an eigenvector process. If $F\subseteq V_d$ is a finite set then the distribution of $\mu$, when restricted to $F$, is concentrated on the subspace $W_{\lambda}(F)$ (recall Chapter \ref{chap:eigen}). We denote by $\mathbb D_{\rm sp}(F,\mu)$ the differential entropy of $\mu_F$ measured inside this subspace using the Euclidean structure inherited from $\mathbb{R}^F$. We say that $\mu$ is {\it smooth} if $\mathbb{D}_{\rm sp}(B_k(C))$ and $\mathbb{D}_{\rm sp}(B_k(e))$ are finite for every $k$. In this chapter we reduce Theorem \ref{mainlim} to smooth eigenvector processes. The reduction will rely on the following statement.
\begin{proposition}\label{prop:masodik}
Let $\{X_v\}_{v\in V_d}$ be a typical eigenvector process with eigenvalue $\lambda$ and $\{Y_v\}_{v\in V_d}$ the unique Gaussian wave $\Psi_{\lambda}$ with eigenvalue $\lambda$. Then the independent sum $\{X_v+aY_v\}_{v\in V_d}$ is smooth for $a>0$.
\end{proposition}
\begin{proof}
Let $S$ be one of $B_k(C)$ and $B_k(e)$.
Both $\{X_v\}_{v\in S}$ and $\{Y_v\}_{v\in S}$ are supported on $W_{\lambda}(S)$.
Moreover, by Corollary \ref{cor:qsp}, we obtain that the support of $\Psi_{\lambda}|_S$ is equal to $W_{\lambda}(S)$. Using Lemma \ref{lem:veges} inside the space $W_{\lambda}(S)$, we get that the differential entropy of the joint distribution $\{X_v+aY_v\}_{v\in S}$ is finite. \hfill $\square$
\end{proof}
Assume that Theorem \ref{mainlim} holds for smooth typical eigenvector processes. Using Proposition \ref{prop:masodik} we obtain that if $\{X_v\}_{v\in V_d}$ is an arbitrary typical eigenvector process then
$\{X_v+aY_v\}_{v\in V_d}$ is smooth for all $a>0$. In addition, by Proposition \ref{prop:elso} and Proposition \ref{prop:hv}, $\{X_v+aY_v\}_{v\in V_d}$ is typical. By our assumption, we obtain the Gaussianity of $\{X_v+aY_v\}_{v\in V_d}$ for all $a>0$. This implies the Gaussianity of $\{X_v\}_{v\in V_d}$ by going to $0$ with $a$.
\medskip
\section{Entropy Inequality for typical eigenvector processes}
\label{chap:entrtyp}
In this chapter we prove the next theorem.
\begin{theorem}\label{mainineq} Let $\mu\in I_d(\mathbb{R})$ be a smooth typical eigenvector process. Then
$$\mathbb D_{\rm sp}(B_k(C))-\frac{d}{2}\mathbb D_{\rm sp}(B_k(e))\geq 0$$ holds for every $k\geq 0$ integer.
\end{theorem}
To prove the above theorem we will need some preparation.
For $a\in\mathbb{N}$ let us define the continuous discretization $t_{0,a}$ of $\mathbb{R}$ in the following way.
If $x>a$ (resp. $x<-a$), then $t_{0,a}(x)=a$ (resp. $t_{0,a}(x)=-a$) with probability $1$. Otherwise let $t_{0,a}(x)$ denote the probability distribution that takes $\lfloor x*a\rfloor/a$ with probability $1+\lfloor x*a\rfloor-x*a$ and takes $1/a+\lfloor x*a\rfloor/a$ with probability $x*a-\lfloor x*a\rfloor$. For $\sigma>0$ we define the discretization $t_{\sigma,a}$ by $t_{\sigma,a}(x)=t_{0,a}(x+\sigma N)$ where $N$ is a random variable with standard normal distribution. We denote by $t_{\sigma,a}^n$ the continuous discretization of $\mathbb{R}^n$ obtained by the coordinatewise independent application of $t_{\sigma,a}$.
\begin{lemma}\label{finombecsles} Let $X$ be a random variable with values in $\mathbb{R}^n$ with finite variance, i.e.\ $\mathbb{E}(\|X\|_2^2)<\infty$. Then we have for every fixed $\sigma>0$ that
$$\mathbb{H}(t_{\sigma,a}^n(X))=n\log a+\mathbb D(X+\sigma M)+o(1)$$
as $a\rightarrow\infty$, where $M$ is independent of $X$ and has standard normal distribution on $\mathbb{R}^n$.
\end{lemma}
The main difficulty of the proof of Lemma \ref{finombecsles} comes from the fact that the support of $X$ is not necessarily compact. We have to treat a situation where we refine and increase the interval of discretization simultaneously.
\begin{proof} By Lemma \ref{lem:veges}, the finite variance of $X$ guarantees that $\mathbb D(X+\sigma M)$ exists and is a finite quantity. Let $S_a=\{r/a | r\in\mathbb{Z}^n, \|r\|_\infty\leq a^2\}$ and let $S'_a=\{x|x\in S_a, \|x\|_\infty<a\}$. For $x\in S_a$ let $p_a(x)$ denote the probability of $x$ in the distribution $t_{\sigma,a}^n(X)$. We have that $$\mathbb{H}(t_{\sigma,a}^n(X))=\sum_{x\in S_a}-p_a(x)\log p_a(x).$$
Let $q_a$ denote the quantity $\mathbb{P}(t^n_{\sigma, a}(X)\in S_a\setminus S'_a)=\sum_{x\in S_a\setminus S'_a} p_a(x)$. The construction of $t_{\sigma,a}^n$ shows that $$q_a\leq \mathbb{P}(\|X+\sigma M\|_\infty\geq a-a^{-1}).$$
By Chebyshev's inequality we have that $$\mathbb{P}(\|X+\sigma M\|_\infty\geq a-a^{-1})=O(a^{-2})$$ and so $q_a=O(a^{-2})$. It follows that \[\sum_{x\in S_a\setminus S'_a}-p_a(x)\log p_a(x)\leq -q_a\log q_a+q_a\log |S_a\setminus S'_a|\leq -q_a\log q_a + q_a\log |S_a|\]\[=-q_a\log q_a+q_a n \log (2a^2+1)=o(1).\]
For $x=(x_1 ,x_2, \dots, x_n)\in\mathbb{R}^n$ let $g_a(x)=\prod_{i=1}^n\max\{1-a|x_i|,0\}$. For every $x\in S'_a$ we have that $p_a(x)=\int_{z\in\mathbb{R}^n}g_a(z)f(x-z)$ where $f$ is the density function of $X+\sigma M$ on $\mathbb{R}^n$. Using that $a^n\int_{\mathbb{R}^n} g_a=1$ we have that $a^np_a(x)$ is a weighted average of the values of $f$ in an $L^\infty$-ball of radius $1/a$. It follows that for every $x\in S'_a$ there is $\alpha(x)\in\mathbb{R}^n$ with the property that $\|x-\alpha(x)\|_\infty\leq 1/a$ and $a^np_a(x)=f(\alpha(x))$ (using that $f$ is continuous). Now we have that
\begin{equation}\label{eqa1}
\sum_{x\in S'_a} -p_a(x)\log p_a(x)=(n\log a)\sum_{x\in S'_a}p_a(x)-a^{-n}\sum_{x\in S'_a} f(\alpha(x))\log f(\alpha(x)).
\end{equation}
It follows from $q_a=O(a^{-2})$ that
$$(n\log a)\sum_{x\in S'}p_a(x)=n(\log a)(1-q_a)=n\log a + o(1).$$
It remains to bound the second part of (\ref{eqa1}).
From the equation
\begin{equation}\label{eqa2}
\int_{z\in[-a+a^{-1},a]^n}-f(z)\log f(z)=\int_{z\in[0,a^{-1}]^n}\sum_{x\in S'_a}-f(x+z)\log f(x+z)
\end{equation}
we obtain that there is a fixed $\gamma\in [0,a^{-1}]^n$ such that
\begin{equation}\label{eqa2b}
a^{-n}\sum_{x\in S'_a}-f(x+\gamma)\log f(x+\gamma)
\end{equation}
is equal to the left hand side of (\ref{eqa2}). On the other hand, the left hand side of (\ref{eqa2}) is equal to $\mathbb D(X+\sigma M)+o(1)$. It remains to show that
\begin{equation}\label{eqa3}
a^{-n}\sum_{x\in S'_a}\bigl(f(\alpha(x))\log f(\alpha(x))-f(\beta(x))\log f(\beta(x))\bigr)=o(1)
\end{equation}
where $\beta(x)=x+\gamma$. (Note that $\alpha, \beta, \gamma$ all depend on $a$.) We will use that $\|\alpha(x)-\beta(x)\|_\infty\leq 2/a$ holds for every $x\in S'_a$ and thus $r(x):=\|\alpha(x)-\beta(x)\|_2\leq 2\sqrt n/a$. Let $t_x=f(\alpha(x))/f(\beta(x))$. We have by Lemma \ref{arany} that for every $\varepsilon>0$ if $a$ is big enough, then $t_x\geq 1-\varepsilon$ (resp. $t_x^{-1}\geq 1-\varepsilon$) provided that $f(\beta(x))> c$ (resp. $f(\alpha(x))>c$) where $c= \sigma^{-n/2}\exp(-a/(32n\sigma^2)).$ This implies that for every $a>0$ we can choose $\varepsilon=\varepsilon(a)$ such that $\lim_{a\rightarrow \infty}\varepsilon(a)=0$ and the previous property holds with $\varepsilon$. We will assume that $a$ is so large that $\varepsilon(a)<1/3$.
Let $T_1$ denote the sum of the terms in (\ref{eqa3}) where $f(\beta(x))\leq 2c$ and let $T_2$ denote the sum of the remaining terms. According to Lemma \ref{arany} either $f(\alpha(x))<c$ or $t_x^{-1}\geq 1-\varepsilon$. If $f(\beta(x))\leq 2c$, then we have $f(\alpha(x))\leq 2c/(1-\varepsilon)\leq 3c$ in both cases. It follows that $T_1\leq 3c\log(3c)(2a^2+1)^n/a^n=o(1)$.
Now we estimate $T_2$. From now on we assume that $f(\beta(x))>2c$. By Lemma \ref{arany} we obtain $t_x\geq 1-\varepsilon$ holds and thus $f(\alpha(x))\geq (1-\varepsilon)2c>c$. This implies again by Lemma \ref{arany} that $t_x^{-1}\geq 1-\varepsilon$ and so $|1-t_x|\leq 2\varepsilon$. We have that $$f(\alpha(x))\log f(\alpha(x))-f(\beta(x))\log f(\beta(x))=f(\alpha(x)) \log t_x-(1-t_x)f(\beta(x))\log f(\beta(x)).$$
Using that $f(\alpha(x))=a^np_a(x)$ and $|\log t_x|\leq -\log(1-2\varepsilon)$ we get that
$$\biggl| a^{-n}\sum_{x\in S'_a, f(\beta(x))>2c} f(\alpha(x))\log t_x\biggr| \leq\sum_{x\in S_a} p_a(x)(-\log(1-2\varepsilon)) =-\log(1-2\varepsilon)= o(1).$$
It is now clear that the following claim finishes the proof of the lemma.
\noindent{\it Claim:}~$a^{-n}\sum_{x\in S'_a} |f(\beta(x))\log f(\beta(x))| = O(1).$
By Lemma \ref{korlat} there is $b\in\mathbb{R}^+$ such that $f(x)<1$ whenever $\|x\|_\infty>b-1$. Let $B=[-b,b]^n$.
By the finiteness of $B$ we have that
\begin{equation}\label{equa4}
a^{-n}\sum_{x\in S'_a\cap B} -\beta(x)\log(\beta(x))=\int_B -f\log f + o(1)
\end{equation}
and that
\begin{equation}\label{equa5}
a^{-n}\sum_{x\in S'_a\cap B} |\beta(x)\log(\beta(x))|=\int_B |f\log f| + o(1)
\end{equation}
It follows from (\ref{equa4}) and the property that the left hand side of (\ref{eqa2}) is equal to (\ref{eqa2b}) that
\begin{equation}\label{equa6}
a^{-n}\sum_{x\in S'_a\setminus B} -\beta(x)\log(\beta(x))=\int_{\overline{B}} -f\log f+o(1).
\end{equation}
By (\ref{equa5}) and (\ref{equa6}) we get that
$$a^{-n}\sum_{x\in S'_a} |\beta(x)\log(\beta(x))|=\int_{\mathbb{R}^n} |f\log f|+o(1)=O(1).\qquad \qquad \square$$
\end{proof}
\bigskip
In the following lemma, first we calculate the entropy of the random variable $t_{\sigma, a}(x)$ for every fixed $x$, and then we take the expectation of this quantity with respect to the distribution of a random variable $X$.
\begin{lemma}\label{varhato} Let $X$ be a real valued random variable with finite variance and with distribution $\nu$. Then $\mathbb{E}_{\nu}(\mathbb{H}(t_{\sigma,a}(x)))=\log a+\mathbb D(N(0,\sigma))+o(1)$ for every fixed $\sigma>0$ as $a\rightarrow \infty$.
\end{lemma}
\begin{proof} Let $\nu_a$ denote the conditional distribution of $X$ when $|X|<a/2$ and let $\nu_a'$ denote the conditional distribution when $|X|\geq a/2$. We have that
$$\mathbb{E}_{\nu}(\mathbb{H}(t_{\sigma,a}(x)))=\mathbb{P}(|X|<a/2)\mathbb{E}_{\nu_a}(\mathbb{H}(t_{\sigma,a}(x)))+\mathbb{P}(|X|\geq a/2)\mathbb{E}_{\nu_a'}(\mathbb{H}(t_{\sigma,a}(x))).$$
By Chebyshev's inequality we obtain that $\mathbb{P}(|X|\geq a/2)=O(a^{-2})$. It follows from the trivial uniform bound $\mathbb{H}(t_{\sigma,a}(x))\leq\log(a^2+1)$ that $\mathbb{P}(|X|\geq a/2)\mathbb{E}_{\nu_a'}(\mathbb{H}(t_{\sigma,a}(x)))=o(1)$. Similarly by $\mathbb{P}(|X|< a/2)=1-O(a^{-2})$ we obtain that $$\mathbb{P}(|X|<a/2)\mathbb{E_{\nu}}(\mathbb{H}(t_{\sigma,a}(x)))=\mathbb{E}_{\nu_{a}}(\mathbb{H}(t_{\sigma,a}(x)))+o(1).$$
It is now enough to prove that if $|x|\leq a/2$ then $\mathbb{H}(t_{\sigma,a}(x))=\log a +\mathbb D(N(0,\sigma))+o(1)$, where the $o(1)$ error term does not depend on $x$ but tends to $0$ as $a\rightarrow \infty$.
Lemma \ref{finombecsles} implies that $\mathbb H(t_{\sigma, a}(0))=\log a+\mathbb D(N(0, \sigma))+o(1)$; this is the $X=0$ case.
Next, suppose that $x\in S_a=\{r/a|r\in \mathbb Z, \| r\|_{\infty}\leq a^2\}$ and $0<x\leq a/2$. Notice that if $y\in S_a$ and $-a< y< a-x$, then $\mathbb P(t_{\sigma, a}(0)=y)=\mathbb P(t_{\sigma, a}(x)=y+x)$, because the distance of $0$ and $x$ is a multiple of the distance of the points in the grid $S_a$ that we used for discretization. Hence the difference $\mathbb H(t_{\sigma, a}(x))-\mathbb H(t_{\sigma, a}(0))$ contains only terms corresponding to $|y-x|>a/2, y\in S_a$ in the first entropy expression (for $x$) and $|y|>a/2, y\in S_a$ in the second one (for $0$). The facts that $t_{\sigma, a}$ is supported on a set of at most $a^2+1$ elements and that the probability that a Gaussian distribution is farther from its expectation than $a/2$ is $O(\exp(-a^2)/2)$ imply that $\mathbb H(t_{\sigma, a}(x))-\mathbb H(t_{\sigma, a}(0))=o(1)$ uniformly in $0<x<a/2$ as $a\rightarrow \infty$, when $x\in S_a$. A similar argument works for $-a/2<x<0$ if $x$ is an element of $S_a$.
Finally, let $x\in [-a/2, a/2]$ arbitrary, and $\overline x$ be the closest element of $S_a$ to $x$. As it is well known (e.g.\ as a consequence of Pinsker's inequality), the total variation distance of $\textrm N(x, \sigma)$ and $\textrm N(\overline x, \sigma)$ is of order $O(1/a)$ provided $|x-\overline x|\leq 1/a$. By choosing an appropriate coupling of these two distributions and applying the same discretization, it follows that $d_{\rm TV}(t_{\sigma,a}(x), t_{\sigma, a}(\overline x))=O(1/a)$. Applying Theorem 17.3.3. of \cite{cover} we obtain that
\[|\mathbb H(t_{\sigma, a}(x))-\mathbb H(t_{\sigma, a}(\overline x))|\leq -d_{\rm TV}(t_{\sigma,a}(x), t_{\sigma, a}(\overline x))\log \frac{d_{\rm TV}(t_{\sigma,a}(x), t_{\sigma, a}(\overline x))}{a^2+1}=o(1), \]
and the error term does not depend on $x$. This concludes the proof. \hfill $\square$
\end{proof}
We finish this chapter with the proof of Theorem \ref{mainineq}.
Let $\mu_{\sigma,a}$ denote the process in which we pointwise discretize $\mu$ using $t_{\sigma,a}$ (using the notation of Chapter \ref{typpro}, we have $\mu_{\sigma, a}=t_{\sigma, a}*\mu$) and let $\mu_\sigma$ denote the process obtained from $\mu$ by adding $\sigma$ times the i.i.d normal distribution.
By Lemma \ref{finombecsles} and $|B_k(C)|-(d/2)|B_k(e)|=1$ we obtain that
$$\mathbb{H}(B_k(C),\mu_{\sigma,a})-\frac{d}{2}\mathbb{H}(B_k(e),\mu_{\sigma,a})=\log a + \mathbb D(B_k(C),\mu_\sigma)-\frac{d}{2}\mathbb D(B_k(e),\mu_\sigma)+o(1).$$
By Theorem \ref{discmain} and the typicality of $\mu$ we get that
$$\mathbb{H}(B_k(C),\mu_{\sigma,a})-\frac{d}{2}\mathbb{H}(B_k(e),\mu_{\sigma,a})\geq \mathbb{E}_{\mu_o}(\mathbb{H}(t_{\sigma,a}(x))).$$
Using the previous formulas, Lemma \ref{varhato} and the limit $a\rightarrow\infty$ we obtain that
\begin{equation}\label{eq:eht}\mathbb D(B_k(C),\mu_\sigma)-\frac{d}{2} \mathbb D(B_k(e),\mu_\sigma)\geq D(N(0,\sigma)\end{equation} for every $\sigma>0$.
Let $S\subset V_d$ be either $B_k(C)$ or $B_k(e)$. We denote by $\mu_{S, \sigma}$ the probability measure obtained by convolving the measure $\mu_S$ with the standard normal distribution on $W_{\lambda}(S)$ (for the definition of $W_\lambda(S)$ see chapter \ref{chap:eigen}), where $\lambda$ is the eigenvalue corresponding to $\mu$. Observe that the standard normal distribution on $\mathbb R^S$ is the independent sum of the standard normal distribution on $W_{\lambda}(S)$ and on $W_{\lambda}(S)^{\perp}$. Then by using that $\mathrm{dim}\, W_{\lambda}(S)^{\perp}=|S|-|\partial S|$, we have
\[\mathbb D(S, \mu_{\sigma})=\mathbb D_{\rm sp}(\mu_{S, \sigma})+(|S|-|\partial S|)\mathbb D(N(0,\sigma)).\]
Using this formula for $S=B_k(C)$ and $S=B_k(e)$ in \eqref{eq:eht} together with $|\partial B_k(C)|=(d/2)|\partial B_k(e)|$, we obtain that
\[\mathbb D_{\rm sp}(\mu_{B_k(C), \sigma})-\frac{d}{2}\mathbb D_{\rm sp}(\mu_{B_k(e), \sigma})\geq 0.\]
If $\sigma\rightarrow 0$, then we obtain the statement of Theorem \ref{mainineq}.
\hfill $\square$
\section{Eigenvalues of the covariance operator}
\label{chap:cov}
The main goal of this section is to prove the following statement on the differential entropies of Gaussian waves. Recall the definition of $\mathbb D_{\rm sp}$ from Chapter \ref{smooth}. In addition, by $\mathrm{det}_{\rm sp}$ we mean the product of the non-zero eigenvalues of a matrix.
\begin{theorem} \label{thm:nulla} Let $\Psi_{\lambda}$ be a Gaussian wave function on the $d$-regular tree with $\lambda\in [-2\sqrt{d-1}, 2\sqrt{d-1}]$. Then we have
\[\mathbb D_{\rm sp}(B_k(C), \Psi_{\lambda})-\frac d2 \mathbb D_{\rm sp}(B_k(e), \Psi_{\lambda})\rightarrow 0 \qquad (k\rightarrow \infty).\]
\end{theorem}
Let $\Sigma_k$ be the covariance matrix of the joint distribution of $\Psi_{\lambda}$ restricted to the ball $B_k(C)$, and $\Sigma'_k$ be the similar covariance matrix on $B_k(e)$. The differential entropy of a multivariate normal random variable with covariance matrix $\Sigma$ of rank $m$ is given by $\frac 12 \log \big((2\pi e)^m \det_{\rm sp} \Sigma)\big)$ if we measure differential entropy inside the support of the variable. Equation \eqref{eq:dim} and Corollary \ref{cor:qsp} imply that the rank of $\Sigma_k$ is $d/2$ times the rank of $\Sigma_k'$. Hence we need to prove that
\begin{equation}\label{eq:logdet}\log \det\!_{\rm sp}\, \Sigma_k-\frac{d}{2}\log \det\!_{\rm sp} \, \Sigma_k'\rightarrow 0 \qquad (k\rightarrow\infty).\end{equation}
Notice that if $s$ is an eigenvalue of both $\Sigma_k$ and $\Sigma_k'$, and its multiplicity in the first case is $d/2$ times its multiplicity in the second case, then it is canceled out in the difference. In order to find the eigenvalues that do not cancel out, we decompose both $\mathbb R^{|B_k(C)|}$ and $\mathbb R^{|B_k(e)|}$ as a union of orthogonal subspaces that are invariant under the corresponding covariance operators.
First we need some notation. We will use the genealogical labeling of the vertices in $B_k(C)$ and in $B_k(e)$ (in this section, we will not distinguish vertices and labels).
In $B_k(C)$, the root gets label $\emptyset$, and we put the labels on the vertices such that the labels of neighbors differ only in the last coordinate (i.e.\ $1, 2, \ldots, d$ are the neighbors of the root; $11, 12, \ldots, 1(d-1)$ are the further neighbors of $1$, and so on). For a vertex $v$ the length of its label is denoted by $|v|$, which is its distance from $\emptyset$. For a vertex $v$ and a sequence $y$, by $vy$ we mean the vertex with the label obtained by concatenating $v$ and $y$. We say that $v$ is an ancestor of $w$ (denoted by $v\rightarrow w$), if $w=vy$ for some $y\neq \emptyset$. As for $B_k(e)$, we use a similar notation, but keeping track of symmetry with respect to the central edge $e$. The endpoints of $e$ have labels $\emptyset$ and $\emptyset'$. The descendants of $\emptyset$ have labels $1, \ldots, d-1$, their descendants have labels $11, 12, \ldots, 1(d-1), 21, \ldots$ and so on. Similarly, the descendants of $\emptyset'$ have labels $1', \ldots, (d-1)'$, their descendants have labels $11', 12', \ldots, 1(d-1)', 21', \ldots$ and so on. Note that $|v|$ still denotes the length of the label.
We assign a linear subspace to each vertex in $B_k(C)\setminus \partial B_k(C)$ and $B_k(e)\setminus \partial B_k(e)$. Fix $v\in B_k(C)\setminus \partial B_k(C)$. Let $\mathcal E_v$ be the elements $\alpha\in\mathbb R^{B_k(C)}$ for which the following hold. $(i)$ $\alpha_w=0$ if $v$ is not an ancestor of $w$. $(ii)$ Suppose that $1\leq j\leq (d-1)$ and $y, z$ are labels with $|y|=|z|$. Then $\alpha_{vjy}=\alpha_{vjz}$. (To put it in another way, for descendants of $v$, the value of $\alpha$ depends only on the first coordinate after $v$ and the distance from $v$.) $(iii)$ We have $\sum_{y: |y|=r} \alpha_{vy}=0$ for $r\geq 1$.
In addition, we introduce the following subspace:
\[\mathcal G=\{\alpha\in\mathbb R^{B_k(C)}: \alpha_v=\alpha_w \text{ if } |v|=|w|\}.\]
We will also refer to $\mathcal E_v'$ (when $v\in B_k(e)\setminus \partial B_k(e)$), which are linear subspaces of $\mathbb R^{B_k(e)}$ defined similarly.
The definition of the complement subspace is somewhat different:
\[\mathcal G_1'=\{\alpha\in\mathbb R^{B_k(e)}: \alpha_v=\alpha_w \text{ if } |v|=|w|\}.\]
\[\mathcal G_2'=\{\alpha\in\mathbb R^{B_k(e)}: \alpha_v=\alpha_w \text{ if } \emptyset \rightarrow v, \emptyset \rightarrow w, |v|=|w|;\ \alpha_{v'}=-\alpha_v \text{ if } v=\emptyset \text{ or } \emptyset \rightarrow v\}.\]
\begin{lemma}\label{lem:ev}
The following hold for the linear subspaces defined above.
\begin{enumerate}[$(a)$]
\item $\mathcal E_v, v\in B_k(C)\setminus \partial B_k(C)$ and $\mathcal G$ are invariant under $\Sigma_k$. Similarly, $\mathcal E_v', v\in B_k(e)\setminus \partial B_k(e)$ and $\mathcal G_1', \mathcal G_2'$ are invariant under $\Sigma_k'$.
\item $\mathcal E_v, v\in B_k(C)\setminus \partial B_k(C)$ and $\mathcal G$ are pairwise orthogonal. Similarly, $\mathcal E_v', v\in B_k(e)\setminus \partial B_k(e)$ and $\mathcal G_1', \mathcal G_2'$ are pairwise orthogonal.
\item $\mathbb R^{B_k(C)}=\mathcal G + \sum_{v\in B_k(C)} \mathcal E_v$ and $\mathbb R^{B_k(e)}=\mathcal G_1'+ \mathcal G_2'+ \sum_{v\in B_k(e)} \mathcal E'_v$.
\end{enumerate}
\end{lemma}
\begin{proof} Before going into the proof, we note that we will only use the property that every entry $(i,j)$ of $\Sigma_k$ depends only on the distance of $i$ and $j$. $(a)$ First observe $\mathcal G$ consists of all vectors that are invariant under the full automorphism group of $B_k(C)$. This property is preserved by $\Sigma_k$ and thus $\mathcal G$ is invariant under $\Sigma_k$.
Take any $\alpha\in \mathcal E_v$. For $\Sigma_k \alpha$, property $(ii)$ is preserved because the entries of $\Sigma_k$ depend only on the distance of the two corresponding vertices (as it is the covariance matrix of an invariant random process). Putting this together with the third property we get that $(i)$ also holds for $\Sigma_k \alpha$. Property $(iii)$ means orthogonality to $\mathcal G$; using the invariance of $\mathcal G$, this will also be satisfied by $\Sigma_k\alpha$, which is thus in $\mathcal E_v$. Similar arguments work for the other two linear subspaces.
\end{proof}
$(b)$ Fix $v_1, v_2\in B_k(C)$. If none of them is ancestor of the other one, then the support of any vector of $\mathcal E_{v_1}$ is disjoint from the support of any vector in $\mathcal E_{v_2}$, which implies orthogonality. If $v_1\rightarrow v_2$, then the value of a vector in $\mathcal E_{v_1}$ is the same at all vertices of type $v_2y$ with $|y|$ fixed. Multiplying this by the values of a vector in $\mathcal E_{v_2}$ and summing this up for different $y$s (of fixed length) we get $0$, because of property $(iii)$. This implies the orthogonality. The other cases are similar; we omit the details.
$(c)$ The dimensions of these subspaces are as follows.
\[\mathrm{dim}\ \mathcal E_{v}=(k-|v|+1)(d-2) \qquad \text{for } \emptyset \neq v\in B_k(C)\setminus \partial B_k(C);\]
\[\mathrm{dim}\ \mathcal E_{\emptyset}=(k+1)(d-1); \qquad {\rm dim}\ \mathcal G=k+2;\]
\[\mathrm{dim}\ \mathcal E'_{v}=(k-|v|+1)(d-2) \qquad \text{for } \emptyset \neq v\in B_k(e)\setminus \partial B_k(e);\]
\[\mathrm{dim}\ \mathcal G'_{1}=k+1; \qquad {\rm dim}\ \mathcal G_2'=k+1.\]
The following equalities are easy to check by induction on $k$:
\[1+(k+1)d+\sum_{j=1}^{k} d(d-1)^{j-1} (k+1-j)(d-2)=|B_k(C)|;\]
\[2(k+1)+2\sum_{j=1}^{k-1} (d-1)^{j-1} (k-j)(d-2)=|B_k(e)|. \]
Hence the sum of the dimension of the linear subspaces $\mathcal E_v$ and $\mathcal G$ is equal to the dimension of the space $\mathbb R^{B_k(C)}$. Since the subspaces are pairwise orthogonal by part $(b)$ of the lemma, this implies that the sum must be equal to $\mathbb R^{B_k(C)}$. A similar argument works for $B_k(e)$.
\hfill $\square$
For the following lemma, recall the definition of $f(k, x)$ from equation \eqref{eq:fkx}. Furthermore, the calculation about this recurrence relation in \cite{alon} imply that if we take $\lambda=2\sqrt{d-1}x$, then the covariance of the values at distance $k$ in the Gaussian wave $\Psi_{\lambda}$ is equal to $f(k,x)$.
\begin{lemma}\label{lem:s1}
Let $f(k, x)$ be defined by equation \eqref{eq:fkx}. We define
\[l(k, x)=1+\sum_{j=1}^{k-1}(d-2)(d-1)^{j-1} f(2j, x).\]
Then the eigenvalues of $\Sigma_k$ corresponding to $\mathcal E_{\emptyset}$ and $\mathcal G$ are as follows.
\[s_1(k, x)=1+\sum_{j=1}^k (d-1)^j f(2j, x)+ \sum_{j=1}^k l(j,x)\quad \text{ with multiplicity }\mathrm{1};\]
\[s_2(k, x)=\bigg(1-s_1(k, x)+\sum_{j=1}^k dl(j,x)\bigg)/(d-1) \quad \text{ with multiplicity }\mathrm{d-1}.\]
The eigenvalues of $\Sigma_k'$ corresponding to $\mathcal G_{1}'$ and $\mathcal G_2'$ are as follows.
\[s_3(k, x)=\sum_{j=1}^k l(j, x)+(d-1)^{j-1}f(2j-1, x) \quad \text{ with multiplicity }\mathrm{1};\]
\[s_4(k, x)=\sum_{j=1}^k l(j, x)-(d-1)^{j-1}f(2j-1, x)\quad \text{ with multiplicity }\mathrm{1}.\]
\end{lemma}
\begin{proof} Using the notation from Chapter \ref{chap:eigen}, let $\mathcal S_k=W_{\lambda}(B_k(C))$ and $\mathcal S'_k=W_{\lambda}(B_k(e))$.
First we consider $\mathcal G$. Since $f(k,x)$ is the covariance of the values at distance $k$ in a (nontrivial) Gaussian wave, we have by linearity that every row of $\Sigma_k$ is in $S_k.$ It follows that $\mathrm{Im}(\Sigma_k)\subseteq \mathcal S_k$. On the other hand, by the previous lemma, $\mathcal G$ is invariant under $\Sigma_k$. Notice that $\mathcal G\cap \mathcal S_k$ is one dimensional: given the value at the root, the common value of its neighbors is determined (even for $\lambda=0$), and this can be continued. It follows that $\Sigma_k|_\mathcal G$ has rank one, and the eigenvalue corresponding to $\mathcal G$ can be obtained by calculating the trace of the matrix of $\Sigma_k|_\mathcal G$ in an arbitrary basis. By choosing the basis of the indicator functions of the spheres of radius $0, 1, \ldots, k$ around the root, elementary calculation shows that this eigenvalue is equal to $s_1(k, x)$.
The three eigenvalues corresponding to the invariant subspaces $\mathcal E_{\emptyset}$, $\mathcal G_1'$ and $\mathcal G_2'$ can be obtained by similar arguments, by identifying the image of $\Sigma_k$ (or $\Sigma_k'$) restricted to the given subspace and calculating the trace of its matrix. In the second case, $\mathcal E_{\emptyset}'\cap \mathcal S_k$ has dimension $d-1$: 0 is assigned to the root; the values of the $d$ neighbors of $\emptyset$ have to sum up to 0, but there are no other conditions; given these values, all the others are uniquely determined. The only nonzero eigenvalue has multiplicity $d-1$, which makes it possible to calculate it based on the trace of the matrix. As for the last two cases, $\mathcal G_1'\cap \mathcal S_k'$ and $\mathcal G_2'\cap \mathcal S_k'$ both have dimension 1 again: given the value at $\emptyset$, the value at $\emptyset'$ has to be the same or the opposite. Then the eigenvalue equation and the equality conditions in $\mathcal G_1'$ and $\mathcal G_2'$ uniquely determine all the other values. By choosing appropriate bases in these subspaces, it is straightforward to obtain the eigenvalues in the lemma. \end{proof}\hfill $\square$
\begin{lemma}\label{lem:kul}
Using the notation of the previous lemma, for every $x\in [-1, 1]$, we have
\[\log s_1(k, x)+(d-1)\log s_2(k, x)-\frac d2 \log s_3(k, x)-\frac d2 \log s_4(k, x)\rightarrow 0 \qquad (k\rightarrow \infty).\]
\end{lemma}
\begin{proof}
First we calculate the middle term of $s_1(k, x)$. Using equation \eqref{eq:fkx}, we obtain that
\[T(k, x):=\sum_{j=1}^k (d-1)^j f(2j, x)=\sum_{j=1}^k \frac{(d-1)^j}{\sqrt{d(d-1)^{2j-1}}}q_{2j}(x),\]
where the polynomials $q$ are defined by equation \eqref{eq:qkx}.
Straightforward calculation shows that with $x=\cos \vartheta$ we have
\[T(k, x)=\frac{d-1}{d}U_{2k}(x)+\frac{d-2}{d\sin \vartheta} \mathrm{Im}\frac{e^{3i\vartheta}(e^{(2k-2)i\vartheta}-1)}{e^{2i\vartheta}-1}-\frac 1d,\]
if $\sin \vartheta\neq 0$ and $e^{2\vartheta}\neq 1$. (We deal with the exceptional cases at the end of the proof.)
Using this formula, we obtain that
\begin{equation*}l(k, x)=1-\frac{d-2}{(d-1)d}+\frac{d-2}{d}\,U_{2k-2}(x)+\frac{(d-2)^2}{d(d-1)\sin\vartheta}\,\mathrm{Im}\ \frac{e^{3i\vartheta}(e^{(2k-4)i\vartheta}-1)}{e^{2i\vartheta}-1} \qquad (k\geq 2).\end{equation*}
Notice that the last term is bounded in $k$ for every fixed $x$. From now on, $O(1)$ will denote a quantity which depends both on $x$ and $k$ such that {\em for every fixed $x$ it is bounded in $k$}. We emphasize that in Theorem \ref{thm:nulla} the limit is taken for fixed $\lambda$ (which is equal to $x\cdot \sqrt{d-1}$). Thus in the proofs of this chapter we always think of $x$ as a fixed quantity while tending to infinity with $k$.
Continuing our calculations, we obtain that
\[s_1(k, x)=1+T(k, x)+\sum_{j=1}^k l(j, x)=\bigg(1-\frac{d-2}{d(d-1)}-\frac{(d-2)^2}{d(d-1)\sin \vartheta}\mathrm{Im}\, \frac{e^{3i\vartheta}}{e^{2i\vartheta}-1}\bigg)k+O(1).\]
On the other hand, we have
\[s_1(k, x)-s_2(k, x)=\frac{d}{d-1}s_1(k,x)-\frac{d}{d-1}\sum_{j=1}^k l(j, x)-\frac{1}{d-1}=O(1);\]
\[s_1(k, x)-s_3(k, x)=1+\sum_{j=1}^k (d-1)^j f(2j, x)-(d-1)^j f(2j-1, x)=O(1);\]
\[s_3(k, x)-s_4(k, x)=2\sum_{j=1}^k (d-1)^j f(2j-1, x)=O(1).\]
To put it in another way, with
\[A=1-\frac{d-2}{d(d-1)}-\frac{(d-2)^2}{d(d-1)\sin \vartheta}\,\mathrm{Im}\, \frac{e^{3i\vartheta}}{e^{2i\vartheta}-1}\]
the expressions $s_1(k, x)$, $s_2(k, x)$, $s_3(k, x)$ and $s_4(k, x)$ are all in the form $Ak+O(1)$. If $A>0$, then we get that \[\log s_j(k, x)=\log A + \log k+o(1) \qquad (j=1, 2, 3, 4),\] which implies the statement of the lemma. Hence, in the rest of the proof, we check that $A>0$ holds.
First notice that $1-\frac{d-2}{d(d-1)}>\frac{(d-2)^2}{d(d-1)}$ is satisfied for all $d$. In addition, by using elementary trigonometric identities we obtain
\[\mathrm{Im}\,\frac{e^{3i\vartheta}}{\sin \vartheta(e^{2i\vartheta}-1)}=\frac{\sin \vartheta-\sin 3\vartheta}{2\sin\vartheta(1-\cos 2\vartheta)}=\frac{1-\cos 2\vartheta-2\cos^2\vartheta}{4 \sin^2\vartheta}=\frac{\sin^2\vartheta-\cos^2\vartheta}{2\sin^2 \vartheta}\leq 1.\]
Putting this together, we get the positivity of $A$, which concludes the proof.
We have to deal with the remaining special cases. First, if $|x|=1$, then $\sin \vartheta=0$, but we still have $U_k(\cos \vartheta)=k+1$. This implies that $l(k, x)$ is a quadratic polynomial, and $s_1(k, x)$ is a cubic polynomial (with leading coefficient $1-(d-2)^2/d/(d-1)>0$). Moreover, the differences $s_1-s_2, s_1-s_3, s_1-s_4$ are all of order $O(k^2)$, which implies the statement of the lemma.
The last case is when $e^{2i\vartheta}=1$. This implies $U_k(\cos \vartheta)=1$ for all $k$. That is, $q_k(x)=(d-2)/\sqrt{d(d-1)}$, and $l(k, x)$ is of the form $Bk+O(1)$ for some nonzero $B$. Now $s_1, s_2, s_3, s_4$ are all quadratic polynomials as a function of $k$, while their differences are linear. It follows again that the expression in the lemma goes to 0 as $k\rightarrow\infty$. \hfill $\square$
\end{proof}
\noindent{\it Proof of Theorem \ref{thm:nulla}.} As we have discussed, it is sufficient to show that \eqref{eq:logdet} holds. First fix $k$, and recall Lemma \ref{lem:ev}. Notice that for every $1\leq r\leq k-1$, if we take two vertices $v_1, v_2$ in $B_k(C)$ such that $|v_1|=|v_2|=r$, then the linear transformation $\Sigma_k$ restricted to $\mathcal E_{v_1}$ is isomorphic to the linear transformation $\Sigma_k$ restricted to $\mathcal E_{v_2}$ (we use again that the entries of the covariance matrix depend only on the distance of the vertices). Hence the set of eigenvalues of $\Sigma_k$ corresponding to the invariant subspaces $\mathcal E_{v_1}$ and $\mathcal E_{v_2}$ are the same. Furthermore, this linear transformation is
isomorphic to the linear transformation $\Sigma_k'$ restricted to $\mathcal E_{v}'$, if $|v|=r$ holds. For every $1\leq r\leq k-1$ we have
\[\big|\{v: |v|=r,\, v\in B_k(C)\}\big|=\frac d2 \big|\{v: |v|=r,\, v\in B_k(e)\}\big|.\]
This means that all eigenvalues belonging to the spaces $\mathcal E_v$ and $\mathcal E'_v$ cancel out in the expression in \eqref{eq:logdet} for $|v|>0$.
Therefore only the eigenvalues corresponding to $\mathcal G, \mathcal E_{\emptyset}', \mathcal G_1'$ and $\mathcal G_2'$ are left. These are calculated with multiplicities in Lemma \ref{lem:s1}, and hence Lemma \ref{lem:kul} finishes the proof by showing that the difference goes to zero as $k\rightarrow \infty$. \hfill $\square$
\section{Improved differential entropy inequality}
\label{chap:impr}
In this chapter we use a combination of Theorem \ref{mainineq} and Theorem \ref{thm:nulla} to prove an improved version of Theorem \ref{mainineq} for the case $k=0$. We will use the notation from Chapter \ref{chap:eigen}. If $S$ is either $B_k(C)$ or $B_k(e)$ for some $k\geq 0$ and $p$ is in $S\setminus\partial S$, then we have for the Gaussian wave $\Psi_\lambda$ with distribution $\mu$ that
\begin{equation}\label{dfor}
\mathbb{D}(\mathcal{Q}(S,p,\mu))=\mathbb{D}(B_p)+\sum_{(v,w)\in D}\mathbb{D}(A_{v,w}).
\end{equation}
We obtain from (\ref{dfor}) that the differential entropy $\mathbb{D}(\mathcal{Q}(S,p,\mu))$ does not depend on $p$ for a Gaussian wave with distribution $\mu$. Observe that changing the vertex $p$ to $p'$ results in a linear transformation $T_{p,p'}$ in the system $\mathcal{Q}(S,p,\mu)$. The invariance of the differential entropy shows by Lemma \ref{difftrdet} that $T_{p,p'}$ has determinant $1$.
Now applying Lemma \ref{difftrdet} for an arbitrary smooth eigenvector process $\nu$ with eigenvalue $\lambda$ we obtain that the value of $\mathcal{D}(\mathcal{Q}(S,p,\nu))$ is independent of $p$ since the transformations $T_{p,p'}$ depend only on the quadruple $p,p',\lambda,S$ and can be calculated from the eigenvector equation. For a general smooth eigenvector process $\nu$ we define $\mathbb{D}(S,\nu)$ as this unique differential entropy of $\mathbb{D}(\mathcal{Q}(S,p,\nu))$. We will need the next definition.
\begin{definition} A process $\{Y_v\}_{v\in V_d}$ in $I_d(X)$ is called $2$-Markov if for an arbitrary edge $e$ the distributions $\{Y_v\}_{v\in W_1}$ and $\{Y_v\}_{v\in W_2}$ are conditionally independent with respect to $\{Y_v\}_{v\in e}$ where $W_1$ and $W_2$ are the set of vertices on the two sides of $e$. (With this notation $V_d=W_1\cup e\cup W_2$.)
\end{definition}
Note that the $2$-Markov property implies that the marginal distribution $Y_C=\{Y_v\}_{v\in C}$ determines the whole process because we can build up the distribution $\{Y_v\}_{v\in V_d}$ using iterated conditionally independent couplings of $Y_C$ along edges. More precisely, if for some connected subgraph $K$ of $T_d$ the distribution $\{Y_v\}_{v\in V(K)}$ is already constructed and $w\in V_d$ is a vertex such that the star $B_1(w)$ intersects $K$ in a single edge $e$, then the joint distribution $\{Y_v\}_{v\in V(K)\cup B_1(w)}$ is the conditionally independent coupling of $\{Y_v\}_{v\in V(K)}$ and $\{Y_v\}_{v\in B_1(w)}$ with respect to $\{Y_v\}_{v\in e}$. The invariance of the process implies that $\{Y_v\}_{v\in B_1(w)}$ has the same distribution as $Y_C$. By iterating this procedure we can build up the marginal distribution on any finite connected subgraph of $T_d$ and thus the whole process in uniquely determined.
\begin{lemma}\label{imprlem2} We have for every $k\geq 1$ and smooth eigenvector process $\nu$ the following three inequalities. $$\mathbb{D}(B_k(C),\nu)\leq d\mathbb{D}(B_k(e),\nu)-(d-1)\mathbb{D}(B_{k-1}(C),\nu),$$
$$\mathbb{D}(B_k(e),\nu)\leq 2\mathbb{D}(B_{k-1}(C),\nu)-\mathbb{D}(B_{k-1}(e),\nu),$$
$$\mathbb{D}(B_k(C),\nu)-(d/2)\mathbb{D}(B_k(e),\nu)\leq\mathbb{D}(B_{k-1}(C),\nu)-(d/2)\mathbb{D}(B_{k-1}(e),\nu).$$ If $\nu$ is Gaussian, then we have equality everywhere. Furthermore if we have equality everywhere (for every $k$), then $\nu$ is $2$-Markov.
\end{lemma}
\begin{proof} The proof is based on the general fact (see Lemma \ref{diffincform}) that for a joint distribution $(X,Y,Z)$ we have that $\mathbb{D}(X,Z)+\mathbb{D}(Y,Z)-\mathbb{D}(Z)\geq\mathbb{D}(X,Y,Z)$ holds with equality if and only if $X$ and $Y$ are conditionally independent with respect to $Z$. To see the first inequality let us place $p$ to the root of $B_k(C)$. We can cover $B_k(C)$ in a rotational symmetric way by $d$ copies of $B_k(e)$ in a way that all of them contain $B_{k-1}(C)$ and they are disjoint outside of $B_{k-1}(C)$. Each copy of $B_k(e)$ covers a subset of the variables $A_{v,w}$ and $B_p$ such that the joint differential entropy of this subset of variables is equal to $\mathbb{D}(B_k(e),\nu)$. Now Lemma \ref{diffincform} finishes the proof of the first inequality. The other two inequalities can be seen in a similar way. If we have equality everywhere for every $k$, then by lemma \ref{diffincform} we get that the joint distribution of $\nu$ on $B_k(C)$ is the conditionally independent coupling of $d$ copies of $B_k(e)$ over $B_{k-1}(C)$ and the joint distribution on $B_k(e)$ is the conditionally independent coupling of $2$ copies of $B_ {k-1}(C)$ over $B_{k-1}(e)$. By induction the $2$-markov property follows inside $B_k(C)$ for every $k$ and thus for the whole process.
\end{proof}
Let $\alpha(S,\nu)$ denote the difference $\mathbb{D}(\mathcal{Q}(S,p,\nu))-\mathbb{D}(\mathcal{Q}(S,p,\mu))$ where $\mu$ is the Gaussian eigenvector process with the same eigenvalue as $\nu$. If we apply the same change of basis to both $\mathbb{D}(\mathcal{Q}(S,p,\nu))$ and $\mathbb{D}(\mathcal{Q}(S,p,\mu))$, they change with the same additive constant by lemma \ref{difftrdet} and thus $\alpha(S,\nu)$ remains unchanged. This shows the basis independence of $\alpha(S,\nu)$. In particular we have that if $\nu$ is a smooth eigenvector process then $\alpha(S,\nu)=\mathbb{D}_{\rm sp}(S,\nu)-\mathbb{D}_{\rm sp}(S,\mu)$ and so
$$\alpha(B_k(C),\nu)-(d/2)\alpha(B_k(e),\nu)=$$
\begin{equation}\label{imprlem3eq}
\Bigl(\mathbb{D}_{\rm sp}(B_k(C),\nu)-(d/2)\mathbb{D}_{\rm sp}(B_k(e),\nu)\Bigr)-\Bigl(\mathbb{D}_{\rm sp}(B_k(C),\mu)-(d/2)\mathbb{D}_{\rm sp}(B_k(e),\mu)\Bigr).
\end{equation}
Using Theorem \ref{mainineq} and Theorem \ref{thm:nulla} we get the following consequence.
\begin{proposition}\label{improp} If $\nu$ is a smooth typical eigenvector process then $$\limsup_{k\rightarrow\infty} \alpha(B_k(C),\nu)-(d/2)\alpha(B_k(e),\nu)=\limsup_{k\rightarrow\infty}\mathbb{D}_{\rm sp}(B_k(C),\nu)-(d/2)\mathbb{D}_{\rm sp}(B_k(e),\nu)\geq 0.$$
\end{proposition}
\begin{proof} By Theorem \ref{mainineq} the first term in (\ref{imprlem3eq}) is non negative and by Theorem \ref{thm:nulla} the second term converges to $0$. This completes the proof.
\end{proof}
The main theorem of this chapter is the following.
\begin{theorem}\label{thm:imprmain} If $\nu$ is a smooth typical eigenvector process, then $\alpha(C,\nu)-(d/2)\alpha(e,\nu)\geq 0$ and equality implies that $\nu$ is $2$-Markov.
\end{theorem}
\begin{proof} By iterating the third inequality in Lemma \ref{imprlem2} we obtain that
$$\mathbb{D}(B_k(C),\nu)-(d/2)\mathbb{D}(B_k(e),\nu)\leq\mathbb{D}(B_l(C),\nu)-(d/2)\mathbb{D}(B_l(e),\nu)$$
holds for every $k\geq l\geq 0$. Furthermore if we replace $\nu$ in the above formula by the Gaussian wave $\mu$ then we get equality. This implies that
\begin{equation}\label{imprtheq}
\alpha(B_k(C),\nu)-(d/2)\alpha(B_k(e),\nu)\leq\alpha(B_l(C),\nu)-(d/2)\alpha(B_l(e),\nu)
\end{equation}
holds for every $k\geq l\geq 0$. By Proposition \ref{improp} we get the inequality of the theorem by applying \eqref{imprtheq} with $l=0$ and $k\rightarrow\infty$.
For the second statement assume that $\alpha(C,\nu)-(d/2)\alpha(e,\nu)=0$. By (\ref{imprtheq}) this is only possible if $\alpha(B_k(C),\nu)-(d/2)\alpha(B_k(e),\nu)=0$ holds for every $k$ and thus the inequalities of Lemma \ref{imprlem2} are all equalities. This implies that $\nu$ is $2$-Markov. \hfill $\square$
\end{proof}
\section{Heat equation and the proof of the main theorem}
\label{chap:heat}
\begin{definition}\label{stardist} For $\lambda\in [-2\sqrt{d-1},2\sqrt{d-1}]$ and $d\geq 3$ let $\mathcal{F}_{d,\lambda}$ denote the set of joint distributions $F=(X_1,X_2,\dots,X_d,Z)$ of real valued random variables such that
\begin{enumerate}
\item $\mathbb{E}(X_iX_j)=(\lambda^2-d)d^{-1}(d-1)^{-1}$, $\mathbb{E}(X_iZ)=\lambda/d$, $\mathbb E(X_i)=\mathbb E(Z)=0$ and $\mathbb{E}(X_i^2)=\mathbb{E}(Z^2)=1$ holds for $1\leq i,j\leq d$ and $i\neq j$;
\item the joint distribution $(X_1,X_2,\dots,X_d,Z)$ is symmetric under every permutation that fixes $Z$;
\item for every $1\leq i\leq d$ the joint distribution $(X_i,Z)$ is the same as the joint distribution $(Z,X_i)$;
\item the quantities $\mathbb D_{\rm sp}(X_1,X_2,\dots,X_d,Z)$ and $\mathbb D(X_1,Z)$ are both finite.
\end{enumerate}
We define the function $\mathcal{D}:\mathcal{F}_{d,\lambda}\rightarrow\mathbb{R}$ by $\mathcal{D}(F)=\mathbb D_{\rm sp}(X_1,X_2,\dots,X_d,Z)-\frac{d}{2}\mathbb D(X_1,Z)$.
\end{definition}
Notice that the covariance conditions of definition \ref{stardist} guarantee that $\mathbb{E}((X_1+X_2+\dots+X_d-\lambda Z)^2)=0$ and thus $X_1+X_2+\dots+X_d=\lambda Z$ holds with probability $1$. This implies that the joint distribution $F$ is concentrated on the $1$ co-dimensional ($d$-dimensional) subspace $W_\lambda(C)$ in $\mathbb{R}^{d+1}$. The subspace differential entropy in definition \ref{stardist} is measured in this subspace.
In this chapter we think of $\lambda\in[-2\sqrt{d-1},2\sqrt{d-1}]$ and $d$ as fixed values and most of the times our notation will not indicate the dependence on these values even if there is such a dependence. Our goal is to solve the extremal problem of maximizing $\mathcal{D}$ inside $\mathcal{F}_{d,\lambda}$ (see Theorem \ref{gausscond}). This will provide the last step in the proof of our main theorems (see Theorem \ref{main} and Theorem \ref{mainlim}) as we will explain at the end of this chapter.
It will be important that there is a unique element $F^*=(X_1^*,X_2^*,\dots,X_d^*,Z^*)$ in $\mathcal{F}_{d,\lambda}$ such that $F^*$ is Gaussian. (The covariances define the Gaussian system uniquely, which clearly satisfies the symmetry conditions.) Note that $F^*$ depends on both $d$ and $\lambda$.
Most of this chapter deals with the proof of the the next theorem which says that the entropy formula $\mathcal{D}$ in $\mathcal{F}_{d,\lambda}$ is maximized by the Gaussian distribution $F^*$.
\begin{theorem}\label{gausscond} For every $F\in\mathcal{F}_{d,\lambda}$ we have that $\mathcal{D}(F)\leq\mathcal{D}(F^*)$ and equality holds if and only if $F=F^*$.
\end{theorem}
To get rid of the subspace differential entropy, we apply a change of basis to the systems in $\mathcal{F}_{d,\lambda}$. We can choose a fix linear transformation $T:\mathbb{R}^{d+1}\rightarrow \mathbb{R}^d$ (depending on $d$ and $\lambda$) such that for every $F\in\mathcal{F}_{d,\lambda}$ the system $T(F)=(B_1,B_2,\dots,B_d)$ satisfies $\mathbb{E}(B_iB_j)=\delta_{i,j}$ for $1\leq i,j\leq d$. Using that $M=T(F^*)$ is Gaussian we obtain that $M$ is the standard normal distribution on $\mathbb{R}^d$.
Using that linear transformations change differential entropy with a fix constant (depending on the transformation; see Lemma \ref{difftrdet}), the statement of the theorem is equivalent to the fact that
$$\mathbb D(T(F))-(d/2)\mathbb D(X_1,Z)\leq \mathbb D(M)-(d/2) \mathbb D(X_1^*,Z^*)$$ and equality holds if and only if $T(F)=M$. The proof of Theorem \ref{gausscond} relies on the following proposition.
\medskip
\begin{proposition}\label{gaussprop} Let $F=(X_1,X_2,\dots,X_d,Z)\in\mathcal{F}_{d,\lambda}$ and $F_t=F+\sqrt{2t}F^*$ (using independent sum) for every $t>0$. Then the function $$\Lambda_F(t)=\mathbb D(T(F_t))-(d/2)\mathbb D(X_1+\sqrt{2t}X_1^*,Z+\sqrt{2t}Z^*)$$ satisfies $\Lambda_F'(0)\geq 0$. If $F$ is not Gaussian then $\Lambda_F'(t)> 0$ for some $t\geq 0$.
\end{proposition}
\medskip
Note that the choice of $F_t$ comes from the heat equation in $\mathbb{R}^d$ (see chapter \ref{heatedrandom}). We first show that Proposition \ref{gaussprop} implies Theorem \ref{gausscond}. The joint distribution $F_t$ does not satisfy the covariance conditions of definition \ref{stardist} but it is clear that the scaled version $F_t(1+2t)^{-1/2}$ is in $\mathcal{F}_{d,\lambda}$. Notice that scaling does not change the differential entropy formula because the extra additive constants coming from scaling exactly cancel each other. By using the claim for $G=F_t(1+2t)^{-1/2}$ we obtain from $\Lambda_G'(0)\geq 0$ that $\Lambda_F'(t)\geq 0$ holds for $F$ with every $t\geq 0$. Since $F_t(1+2t)^{-1/2}=F(1+2t)^{-1/2}+F^*(2t/(1+2t))^{1/2}$ converges to $F^*$ as $t\rightarrow \infty$, we obtain that $F^*$ maximizes $\mathcal{D}$. To see that only the Gaussian system $F^*$ attains the maximum assume that $F$ attains the maximum. In this case $\Lambda_F'(t)\geq 0$ is only possible if $\Lambda_F'(t)=0$ holds for every $t\geq 0$. This implies that $F$ is Gaussian by the second part of Proposition \ref{gaussprop}.
It remains to prove Proposition \ref{gaussprop}. We start with some notation and lemmas. Assume that the measure $\mu$ is the distribution and $f$ is the density function of $B=T(F)$.
We can choose a matrix $Q\in \mathbb R^{(d+1)\times d}$ such that $X_i=\sum_k Q_{i,k}B_k$ holds for $i=1, \ldots, d$, and $Z=\sum_k Q_{d+1, k} B_k$ is also satisfied. We define $v_i=(Q_{i, 1}, \ldots, Q_{i, d})$ for $i=1, \ldots, d$ and $w=(Q_{d+1, 1}, \ldots, Q_{d+1, d})$. Notice that the covariance matrix of $F$ is $QQ^T$.
Hence the vectors $\{v_i\}_{i=1}^d$ and $w$ are unit vectors such that $(v_i,v_j)=(\lambda^2-d)d^{-1}(d-1)^{-1}$ and $(v_i,w)=\lambda/d$ for every $1\leq i,j\leq d$. The joint distribution $((v_1,B),(v_2,B),\dots,(v_d,B),(w,B))$ is the same as $F$. We have that $\mu$ is invariant with respect to every orthogonal transformation permuting the system $\{v_i\}_{i=1}^d$ and fixing $w$. Furthermore we have that $\mu$ projected to the space spanned by $v_i$ and $w$ is invariant with respect to the reflection interchanging $v_i$ and $w$. We can choose two real numbers $\alpha$ and $\beta$ such that the vectors $a_i=\alpha w+\beta v_i$ and $b_i=\alpha v_i+\beta w$ satisfy $(a_i,b_i)=0$ and $\|a_i\|_2=\|b_i\|_2=1$ for every $1\leq i\leq d$ (in particular, we need that $(\alpha^2+\beta^2)\lambda/d+2\alpha \beta=0$, and $(\alpha^2+\beta^2)+2\alpha\beta\lambda/d=1$). The choice of $\alpha$ and $\beta$ is unique up to multiplying both by $-1$ or switching them. Their values can be determined using elementary geometry.
Note that construction of the vector system $\{v_i\}_{i=1}^d, w, \{a_i\}_{i=1}^d, \{b_i\}_{i=1}^d$ in $\mathbb{R}^d$ is purely linear algebraic. Such system, with the scalar products given above, can be constructed for an arbitrary $|\lambda|\leq d$ however in the case of $|\lambda|\leq2\sqrt{d-1}$ they satisfy a useful geometric property expressed in the following lemma.
\begin{lemma}\label{wr} If $|\lambda|\leq2\sqrt{d-1}$, then there are numbers $t_1,t_2$ with $t_1,t_2\geq 0$ and $t_1+t_2=1$ such that for every $u\in\mathbb{R}^d$ we have
$$\|u\|_2^2=\sum_{i=1}^d \Bigl(t_1(u,a_i)^2+t_2(u,b_i)^2\Bigr).$$
\end{lemma}
\begin{proof} The proof follows from two observations. The first one is the following. Let $\{v'_i\}_{i=1}^d$ be a system of unit vectors such that all pairwise scalar products are equal and for all $1\leq i\leq d$ we have $(v'_i,w)=c$ for some $c\leq 1$. (The system $\{a_i\}_{i=1}^d$ satisfies the conditions with $c=\alpha+\beta \lambda/d$, and $\{b_i\}_{i=1}^d$ with $c=\alpha\lambda/d+\beta$.) Then
\begin{equation}\label{umbrella}
\sum_{i=1}^d (u,v_i')^2=(1-c^2)d(d-1)^{-1}\|u-(u,w)w\|_2^2+c^2d(u,w)^2
\end{equation}
holds for every $u\in\mathbb{R}^d$.
To see this, first notice that $\sum_i v_i'=cd w$, which implies that $(v_i', v_j')=(c^2d-1)/(d-1)$ for $1\leq i<j\leq d$. It follows that we can choose $\gamma \in \mathbb R$ such that the equality $(v_i'-\gamma w, v_j'-\gamma w)=0$ holds for $1\leq i<j\leq d$. On the other hand, we have $\sum_i (u-(u,w)w, v'_i)=0$. Therefore
\[\sum_{i=1}^d (u, v_i')^2=\sum_{i=1}^d \big((u-(u,w)w, v_i')^2+(u, w)^2(w, v_i')^2\big).\]
The second term is equal to the second term of $\eqref{umbrella}$. In the first term, we can replace $v_i'$ with $v_i'-\gamma w$. If the latter is equal to zero, then we are done. Otherwise, since $\{v_i'-\gamma w\}_{i=1}^d$ is an orthogonal basis in $w^{\bot}$, we obtain
\[\sum_{i=1}^d (u, v_i')^2=\|v_i'-\gamma w\|_2^2\|u-(u,w)w\|_2^2+c^2d(u,w)^2.\]
On the other hand, for symmetry reasons, $\|v_i'-\gamma w\|_2^2$ does not depend on $i$. By substituting $u=v_1'$ and using $(v_1', v_j')=(c^2d-1)/(d-1)$ again, we get that the value of this constant is the same as in equation \eqref{umbrella}.
The second observation says that if $|\lambda|\leq 2\sqrt{d-1}$ then there exist constants $t_1,t_2\geq 0$ with $t_1+t_2=1$ such that $t_1(a_i,w)^2+t_2(b_i,w)^2=1/d$. First of all note the symmetries of the vector system imply that $(a_i,w)^2,(b_i,w)^2$ are independent from $i$. Elementary calculation shows that the two values $(a_i,w)^2$ and $(b_i,w)^2$ are equal to $(1\pm\sqrt{1-(\lambda/d)^2})/2$. This shows the existence of the constants $t_1,t_2$. We get the statement of the lemma by taking the convex combination of (\ref{umbrella}) applied for $\{a_i\}_{i=1}^d$ and $\{b_i\}_{i=1}^d$ with coefficients $t_1$ and $t_2$.
\end{proof}\hfill $\square$
\medskip
Now we return to the proof of Proposition \ref{gaussprop}. For $1\leq i\leq d$ let $f_i$ denote the orthogonal projection of $f$ to the two dimensional space $V_i=\langle w,v_i\rangle_{\mathbb{R}}$. This means that $$f_i(x)=\int_{z\in V_i^{\perp}}f(x+z)$$ for $x\in V_i$.
Let $T_2:\mathbb{R}^2\rightarrow\mathbb{R}^2$ denote the linear transformation $T_2(x,y)=(\alpha x+\beta y, \alpha y+\beta x)$ with $\alpha$ and $\beta$ defined above. We have that $f_i$ (when written in the orthonormal basis $a_i,b_i$) is the density function of $T_2(X_i,Z)$. We can write $\Lambda_F(t)$ as
\begin{equation}
k+\mathbb D(T(F_t))-(d/2)\mathbb D(T_2(X_1+\sqrt{2t}X_1^*,Z+\sqrt{2t}Z^*))
\end{equation}
where the constant $k$ comes from the change of basis $T_2$.
Then by the de Bruijn identity (see equation \eqref{diffheat} and Lemma \ref{korlat}) we get $$\Lambda_F'(0)=\int_{\mathbb{R}^d} \|\triangledown f\|_2^2/f-(d/2)\int_{V_1} \|\triangledown f_1\|_2^2/f_1.$$
From Lemma \ref{wr} we have that
\begin{equation}
\|\triangledown f\|_2^2=\sum_{i=1}^d (t_1(\partial_{a_i}f)^2+t_2(\partial_{b_i}f)^2)
\end{equation}
holds for some $t_1, t_2\geq 0$ such that $t_1+t_2=1$.
Using
$$\|\triangledown f_1\|_2^2=(\partial_{a_1}f_1)^2+(\partial_{b_1}f_1)^2$$
and the above equations it follows that
$$\Lambda_F'(0)=\sum_{i=1}^d \int_{\mathbb{R}^d} (t_1(\partial_{a_i}f)^2+t_2(\partial_{b_i}f)^2)/f-(d/2)\int_{V_1} ((\partial_{a_1}f_1)^2+(\partial_{b_1}f_1)^2)/f_1.$$
By the symmetries of $f$ we have that the terms in the above sum are all the same and thus
\begin{equation}\label{eq:aa11}
\Lambda_F'(0)=d\int_{\mathbb{R}^d} (t_1(\partial_{a_1}f)^2+t_2(\partial_{b_1}f)^2)/f-(d/2)\int_{V_1} ((\partial_{a_1}f_1)^2+(\partial_{b_1}f_1)^2)/f_1.
\end{equation}
\begin{lemma}\label{csch} If $u\in V_1$, then $$\int_{\mathbb{R}^d} (\partial_uf)^2/f \geq \int_{V_1}(\partial_uf_1)^2/f_1.$$ Equality holds if and only if the function $g=\partial_u f(x)/f(x)=\partial_u \log f$ satisfies $g(x)=g(x+z)$ for every pair $x\in\mathbb{R}^d$ and $z\in V_1^\perp$.
\end{lemma}
\begin{proof} $$\int_{\mathbb{R}^d} (\partial_uf)^2/f=\int_{x\in V_1}f_1(x)\int_{z\in V_1^\perp} (f(x+z)/f_1(x))(\partial_u f(x+z)/f(x+z))^2$$
$$\geq \int_{x\in V_1}f_1(x)\Bigl(\int_{z\in V_1^\perp} (f(x+z)/f_1(x))(\partial_u f(x+z)/f(x+z))\Bigr)^2$$
$$=\int_{x\in V_1}f_1(x)\Bigl(\int_{z\in V_1^\perp} (\partial_u f(x+z)/f_1(x))\Bigr)^2=\int_{V_1} (\partial_uf_1)^2/f_1.$$ To see the inequality in the above calculation notice that $f(x+z)/f_1(x)$ is the density function of a probability measure on $x+V_1$ . We can apply the Cauchy--Schwarz inequality using this density function to get the inequality.
It also shows that equality holds in the statement of the lemma if and only if $\partial_u f(x+z)/f(x+z)$ is constant almost everywhere on $x+V_1$ for almost every $x$. Since we work with continuous functions the almost can be omitted. \hfill $\square$
\end{proof}
\medskip
We apply Lemma \ref{csch} for $a_1$ and $b_1$ and (\ref{eq:aa11}) to obtain that
\begin{equation}\label{eq:ab1}\Lambda_F'(0)\geq (dt_1-d/2)\int_{V_1} (\partial_{a_1}f_1)^2/f_1+(dt_2-d/2)\int_{V_1} (\partial_{b_1}f_1)^2/f_1.\end{equation}
Using the symmetry of $(X_1,Z)$ we obtain that
\begin{equation}\label{eq:ab2}\int_{V_1} (\partial_{a_1}f_1)^2/f_1=\int_{V_1} (\partial_{b_1}f_1)^2/f_1.\end{equation}
It follows that $\Lambda_F'(0)\geq 0$. The proof of the first part of Proposition \ref{gaussprop} is now complete.
We arrived to the second part of Proposition \ref{gaussprop}. Assume that $F$ satisfies $\Lambda_F'(t)=0$ for every $t\geq 0$. Using the notation from Lemma \ref{wr} we have by $t_1+t_2=1$ that at least one of $t_1>0$ and $t_2>0$ holds. Without loss of generality we assume that $t_1>0$. Let $g_t:\mathbb{R}^d\rightarrow\mathbb{R}$ denote the logarithm of the density function of $B_t=T(F_t)$.
This implies by Lemma \ref{csch} that $\partial_{a_1}g_t$ satisfies the property that $\partial_{a_1}g_t(x)=\partial_{a_1}g_t(x+z)$ holds whenever $x\in \mathbb R^d$ and $z\in V_1^{\perp}$. For convenience, we will write every element $x\in \mathbb R^d$ as a triple $(\alpha(x), \beta(x), \gamma(x))$, where $\alpha(x)=(x, a_1)$, $\beta(x)=(x, b_1)$ and $\gamma(x)$ is the projection of $x$ to $V_1^{\perp}$. Using this notation, we have that $\partial_{a_1}g_t(x)=h_t(\alpha(x), \beta(x))$. This means that there exists a function $\widehat{h_t}: \mathbb R^2\rightarrow \mathbb R$ such that $\partial_{a_1} \widehat{ h_t}(\alpha(x), \beta(x))=\partial_{a_1}g_t(x)$. We have by $\partial_{a_1}(g_t-\widehat{h_t})=0$ that $g_t(x)-\widehat{h_t}(\alpha(x), \beta(x))=s_t(\beta(x), \gamma(x))$ for some function $s_t$. We obtain that the density function of $B_t$ can be written in the following form.
\[\exp(\widehat{h_t}(\alpha(x), \beta(x)))\cdot \exp(s_t(\beta(x), \gamma(x)).\]
In other words, this means that the random variables of $\alpha(B_t)$ and $\gamma(B_t)$ are conditionally independent with respect to $\beta(B_t)$. This implies by Lemma \ref{twoposs} that one of the following two possibilities holds: either $\alpha(B)$ is independent of $(\beta(B), \gamma(B))$ or $\gamma(B)$ is independent of $(\alpha(B), \beta(B))$. In the first case we obtain (using the terminology of Lemma \ref{indepgauss}) that $a_1$ is an independent direction for $B$. By symmetries of $B$ we obtain that $\{a_i\}_{i=1}^d$ are all independent directions for $B$. If $t_1<1$ then $(a_i,a_j)\neq 0$ for every pair $1\leq i<j\leq d$ and Lemma \ref{indepgauss} finishes the proof. If $t_1=1$, then $\{a_i\}_{i=1}^d$ is an orthonormal basis in $\mathbb{R}^d$ and $b_1=\sum_{i=2}^{d-1}a_i(d-1)^{-1/2}$. We have that $(B,a_i)$ are identically distributed independent random variables and that $(B,b_1)=\sum_{i=2}^{d-1}(B,a_i)(d-1)^{-1/2}$ has the same distribution. This is only possible if this $(B,a_i)$ is normal for every $i$.
We obtain that $B$ is Gaussian.
In the case when $\gamma(B)$ is independent of $(\alpha(B), \beta(B))$ we have that $\partial_u g(x)=\partial_ug(x+z)$ holds for every triple $u\in V_1, x\in\mathbb{R}^d,z\in V_1^\perp$. The symmetries of $f$ imply that $\partial_u g(x)=\partial_ug(x+z)$ holds for every triple $u\in V_i, x\in\mathbb{R}^d,z\in V_i^\perp$.
Let $r_i$ denote the orthogonal projection of $a_i$ to $w^\perp$ for $1\leq i\leq d$. Note that the vector system $\{r_i\}_{i=1}^d$ is completely symmetric in the sense that the origin is the center of a regular simplex whose vertices are given by these vectors.
We have for every $1\leq i\leq d$ that $\partial_{r_i} g(x)=h_i((x,r_i),(x,w))$ for some two variable function $h_i:\mathbb{R}^2\rightarrow\mathbb{R}$. The symmetries of $f$ imply that $h_i$ does not depend on $i$ and thus $h_i=h$ for some $h$ for every $i$.
The next step is to prove that $h(x,y)=xh^*(y)$ for some one variable function $h^*$.
We have by $\sum_{i=1}^d r_i=0$ that $\sum_{i=1}^d\partial_{r_i}g(x)=0$ and thus $\sum_{i=1}^d h_i((x,r_i),(x,w))=0$ holds for every $x\in\mathbb{R}^d$. For arbitrary numbers $x_1,x_2,\dots,x_d,y\in\mathbb{R}$ with $\sum_{i=1}^d x_i=0$ we can choose $x\in\mathbb{R}^d$ such that $(x,r_i)=x_i$ and $(x,w)=y$. It follows that $\sum_{i=1}^d h(x_i,y)=0$ holds for arbitrary numbers with $\sum_{i=1}^d x_i=0$. Assume first that all $x_i$ is $0$ then we have that $dh(0,w)=0$ and thus $h(0,w)=0$ for every $w$. Then assume that $x_1=a,x_2=-a$ and $x_i=0$ if $i\geq 3$. We obtain that $h(a,w)+h(-a,w)=0$ and thus $h(-a,w)=-h(a,w)$ holds for every $a$ and $w$. Finally let $x_1=a,x_2=b,x_3=-a-b$ and $x_i=0$ if $i\geq 4$. We obtain that $h(a,w)+h(b,w)=-h(-a-b,w)=h(a+b,w)$.
Since $h$ is additive and continuous in the first coordinate a well known fact implies that $h$ is a linear function in the first coordinate and thus we obtain $h(x,y)=xh^*(y)$.
Now we have $\partial_{r_i} g(x)=(x,r_i)h^*((x,w))$ for every $1\leq i\leq d$. It is easy to see that this implies that $g(x)=\|x-(x,w)w\|_2^2h^*((x,w))+c^*((x,w))$ where $c^*:\mathbb{R}\rightarrow\mathbb{R}$ is some function.
We have that $\partial_w g(x)=\|x-w(x,w)\|_2^2(h^*)'((x,w))+(c^*)'((x,w))$.
On the other hand we have that $\partial_w g(x)=\partial_w g(x+z)$ holds whenever $z\in V_i^\perp$ for every $1\leq i\leq d$. It follows that $(h^*)'=0$ everywhere and so $g(x)=c\|x-(x,w)w\|_2^2+c^*((x,w))$ with some constant $c$. Now from $f(x)=\exp(c\|x-(x,w)w\|_2^2+c^*((x,w)))$ we have that $B-(B,w)w$ and $(B,w)$ are independent random variables. Furthermore $B-(B,w)w$ is a Gaussian distribution concentrated on the orthogonal space of $w$. This means that the pair $(B,r_1),(B,w)$ of random variables is independent and $(B,r_1)$ is Gaussian. We know that $(B,v_1)$ is a linear combination of $(B,r_1)$ and $(B,w)$ (with a non-zero coefficient for $(B,r_1)$) and its distribution is the same as the distribution of $(B,w)$ (here we use the symmetries of $B$). It follows that $(B,w)$ is also Gaussian and thus $c^*((x,w))=c_2(x,w)^2+c_3$ for some constants $c_2,c_3$. Thus we have that $B$ is a Gaussian joint distribution implying that $F$ is also joint Gaussian, as it is a linear function of $B$. \hfill $\square$
\medskip
\noindent{\bf Proof of Theorem \ref{main} and Theorem \ref{mainlim}}~~From Proposition \ref{lem:equiv} we have that theorem \ref{mainlim} implies theorem \ref{main}. Let $\mu$ be a smooth typical eigenvector process corresponding to eigenvalue $\lambda$ represented by a system of random variables $\{X_v\}_{v\in V_d}$. According to the results in chapter \ref{smooth} it is enough to show that $\mu$ is a Gaussian wave. We have by Lemma \ref{lem:friedman} that $\lambda\in [-2\sqrt{d-1},2\sqrt{d-1}]$. Let $F=\{X_v\}_{v\in C}$. It is clear that $F\in\mathcal{F}_{d,\lambda}$. We have by Theorem \ref{thm:imprmain} that $\mathcal{D}(F)\geq\mathcal{D}(F^*)$ and thus by Theorem \ref{gausscond} we obtain that $F=F^*$. Again by Theorem \ref{thm:imprmain} we have that $\mu$ is $2$-Markov and so $\{X_v\}_{v\in V_d}$ can be obtained by iterating conditionally independent couplings of $C$ along edges (see chapter \ref{chap:impr}). This shows the Gaussianity of the whole system $\{X_v\}_{v\in V}$.
\section{Appendix A: On heated random variables}\label{heatedrandom}
Let $X$ be a random variable with values in $\mathbb{R}^n$ and let $M$ be the standard normal distribution on $\mathbb{R}^n$. Let $f_t$ denote the density function of $X+\sqrt{2t}M$ and let $\mu_t$ denote the corresponding measure on $\mathbb{R}^n$. The standard heat equation says that $\partial_tf_t=\triangle f_t$ holds for every $t>0$. It is useful to compute the variation of the differential entropy $\mathbb D(f_t)$. The de Bruijn identity (see e.g.\ \cite{cover}) says that
\begin{equation}\label{diffheat}
\partial_t(\mathbb D(f_t))=\partial_t \int_{\mathbb{R}^n} -f_t\log f_t=\int_{\mathbb{R}^n} -\triangle f_t (1+\log f_t)=\int_{\mathbb{R}^n} \|\triangledown f_t\|_2^2/f_t.
\end{equation}
However the validity of (\ref{diffheat}) relies on the fact that both $\partial_i f_t$ and $\partial_i f_t\log f_t$ vanish at infinity for every $1\leq i\leq n$. This fact is proved in Lemma \ref{korlat}.
Notice that if $X$ has finite variance then also $X+\sqrt{2t}M$ has finite variance and if $t>0$ then $\mathbb D(f_t)$ is a finite quantity (Lemma \ref{lem:veges}).
In general if $\sigma>0$ then the density function $f$ of $X+\sigma M$ is smooth, non-vanishing and analytic restricted to every line in $\mathbb{R}^n$. More precisely, if $p,q\in\mathbb{R}^n$ then the real function $\lambda\mapsto f(p+\lambda(q-p))$ extends to an entire analytic function on $\mathbb{C}$. Furthermore every partial derivative of $f$ has this property. In the rest of this appendix we prove several other facts about heated random variables.
\begin{lemma}\label{appineq} Let $X$ be a random variable with values in $\mathbb{R}^n$ and let $M$ be the standard normal distribution on $\mathbb{R}^n$. Let $f$ be the density function of the independent sum $X+\sigma M$ for some $\sigma>0$. Then for every $1\leq i\leq n$ and $x\in\mathbb{R}^n$ we have that $|\partial_i f(x)|\leq f(x)a(1+|\log(bf(x))|^{1/2})$ for some constants $a,b$ depending on $n$ and $\sigma$.
\end{lemma}
\begin{proof} Let $\Phi$ be the density function of $\sigma M$ and let $\mu$ be the distribution of $X$. Let $r\in\mathbb{R}^+$ the smallest positive real number such that $\partial_i\Phi(z)\leq |f(x)|$ for every $z$ satisfying $|z|\geq r$.
It can be shown that $r\leq c_2(1+|\log (c_1f(x))|^{1/2})$ for some constants (depending on $n$ and $\sigma$).
Let $D=\{z : |x-z|\leq r\}$. We have that $\int_{z\in \overline D} \partial_i\Phi(x-z) d\mu\leq |f(x)|$. On the other hand, by $f(x)=\int_{z\in \mathbb R} \Phi(x-z) d\mu$ we obtain that $$\int_{z\in D}\partial_i\Phi(x-z) d\mu\leq |f(x)|\max_{|y|\leq r}\partial_i\Phi(y)/\Phi(y).$$ Using that $\partial_if(x)=\int_z \partial_i\Phi(x-z)d\mu$ and that $\partial_i\Phi(z)/\Phi(z)=O(z)$ the proof is complete.
\hfill $\square$
\end{proof}
\begin{lemma}\label{korlat} Let $X$ be a random variable with values in $\mathbb{R}^n$ and let $M$ be a random variable with standard normal distribution on $\mathbb{R}^n$. Let $f$ be the density function of the independent sum $X+\sigma M$ for some $\sigma>0$. Then for $1\leq i\leq n$ the functions $f$, $\partial_i f$ and $\partial_i f\log f$ vanish at infinity.
\end{lemma}
\begin{proof} We start with $f$. For contradiction, assume that $D=\{x : f(z)\geq c\}$ is unbounded for some $c>0$. Let $\Phi$ be the density function of $\sigma M$ and let us choose $r\in\mathbb{R}^+$ such that $\Phi(x)\leq c/2$ whenever $\|x\|_2\geq r$. Let $\mu$ be the probability distribution of $X$. We have that if $f(x)\geq c$ and $Q_x=\{z: \|z-x\|_2< r\}$, then $\int_{z\in Q_x}\Phi(z-x) d\mu\geq c/2$ and thus $\mu(Q_x)\geq \|\Phi\|_\infty^{-1}c/2$. From the unboundedness of $D$ we conclude that there is an infinite set of points $\{p_i\}_{i=1}^\infty$ in $D$ such that $\|p_i-p_j\|_2>2r$ holds for every pair $i\neq j$ in $\mathbb{N}$. This contradicts the fact that $\mu$ is finite. The statement for $\partial_i f$ and $\partial_i f\log f$ follows from Lemma \ref{appineq}.\hfill $\square$
\end{proof}
\begin{lemma}\label{lem:veges} Let $X$ be a random variable with values in $\mathbb R^n$ with finite covariance matrix. Let $M$ be independent of $X$, with standard normal distribution on $\mathbb R^n$. Then, for every $\sigma>0$, $X+\sigma M$ has finite differential entropy.
\end{lemma}
\begin{proof} The random variable $X+\sigma M$ has finite covariance matrix. As it is well-known, among the distributions with a given covariance matrix, Gaussian distribution maximizes differential entropy. Hence $\mathbb D(X+\sigma M)<\infty$. On the other hand, as in the previous lemma, let $f$ be the density function of $X+\sigma M$. Lemma \ref{korlat} implies that $\{t: f(t)>1\}$ is a compact set. The continuity of $f$ implies that $\int_{\mathbb R^n} f(t) \log f (t) dt<\infty$. Thus we also have $\mathbb D(X+\sigma M)>-\infty$.
\end{proof}\hfill $\square$
\begin{lemma}\label{arany} Let $X$ be a random variable with values in $\mathbb{R}^n$ and let $M$ be the standard normal distribution on $\mathbb{R}^n$. Let $f$ be the density function of the independent sum $X+\sigma M$ for some $\sigma>0$. Then for every $\varepsilon>0$ there is $\varepsilon'>0$ such that for every pair $a,b\in\mathbb{R}^n$ with $\|a-b\|_2=r\leq\varepsilon'$ and $f(b)>c$ we have that $f(a)/f(b)\geq 1-\varepsilon$, where $c= \sigma^{-(n-1)/2}\exp(-r^{-1}/(16\sigma^2)).$
\end{lemma}
\begin{proof} We start by general estimates for a pair $a,b\in\mathbb{R}^n$ with $r=\|a-b\|_2\leq 1/4$. We have that $f(x)=\int_{y\in\mathbb{R}^n}\Phi(x-y) d\mu$ where $\mu$ is the probability distribution of $X$ and $\Phi$ is the density function of $\sigma M$. Let $D=\{z : \|z-a\|_2\leq r^{-1/2}\}$. Let $f_1(x)=\int_{y\in\mathbb{R}^n}1_D\Phi(x-y) d\mu$ and $f_2(x)=f(x)-f_1(x)$. We have that $f_2(x)\leq\sup_{z\in\overline{D}}\Phi(x-z)$. It follows that $|f_2(a)|,|f_2(b)|\leq (2\pi \sigma^2)^{-(n-1)/2}\Phi_0(r^{-1/2}-r)$ where $\Phi_0$ is the density function of the one dimensional normal distribution $N(0,\sigma)$. Thus using $r^{-1/2}-r\geq r^{-1/2}/2$ and $1/\sqrt{2\pi}<1$ we have
\begin{equation}\label{eq:f2}|f_2(a)|,|f_2(b)|\leq(2\pi \sigma^2)^{-(n-1)/2}\Phi_0(r^{-1/2}/2)<c^2.\end{equation}
To estimate $f_1(a)/f_1(b)$ we use
$$\min_{z\in D}\Phi(z-a)/\Phi(z-b)\leq f_1(a)/f_1(b).$$
From $$\Phi(z-a)/\Phi(z-b)=\exp((\|z-b\|_2^2/2-\|z-a\|_2^2/2)/\sigma^2)=\exp(((a-z,b-a)+r^2/2)/\sigma^2)$$
it follows that
\begin{equation}\label{eq:fab}f_1(a)/f_1(b) \geq\exp((-r^{1/2}+r^2/2)/\sigma^2)\geq 1-r^{1/2}/\sigma^2 .\end{equation}
Inequality \eqref{eq:f2} implies that $f(b)=f_1(b)+f_2(b)\leq f_1(b)+c^2$. Using this and $f_1(a)\leq f(a)$ we obtain $$f(a)/f(b)\geq f_1(a)/(f_1(b)+c^2)=(f_1(a)/f_1(b))(1+c^2/f_1(b))^{-1}.$$ If $f(b)>c$, then $f_1(b)=f(b)-f_2(b)>c-c^2$ and thus by \eqref{eq:fab} we get $$f(a)/f(b)>(f_1(a)/f_1(b))(1+c^2/(c-c^2))^{-1}\geq (1-r^{1/2})(1-c).$$
The quantity $c$ goes to $0$ with $r$ and so if $r$ is small enough we have that $f(a)/f(b)\geq 1-\varepsilon$.
\end{proof} \hfill $\square$
\medskip
Let $X$ be a random variable with values in $\mathbb{R}^d$. We say that $v\in\mathbb{R}^d$ is an independent direction for $X$ if the $\mathbb{R}$-valued random variable $(X,v)$ is independent from the projection of $X$ to the $d-1$ dimensional space $v^{\perp}$. Note that every direction is independent for the standard normal distribution.
\begin{lemma}\label{indepgauss} Let $X$ be a random variable with values in $\mathbb{R}^d$ and with $\mathbb{E}(X)=0$. Assume that $\{v_i\}_{i=1}^d$ is a basis in $\mathbb{R}^d$ such that each $v_i$ is an independent direction for $X$ and furthermore for every $1\leq i\leq d$ there is $1\leq j_i\leq d$ such that $(v_i,v_{j_i})\neq 0$. Then $X$ is Gaussian.
\end{lemma}
\begin{proof} Let $N$ be the standard normal distribution. It is clear that the independent sum $X+\varepsilon N$ has the same independence property as $X$ for every $\varepsilon\geq 0$. Furthermore it is enough to prove that the heated version $X+\varepsilon N$ of $X$ is Gaussian for every $\varepsilon>0$. Let us fix $\varepsilon>0$. The advantage of working with $X+\varepsilon N$ is that it has a strictly positive smooth density function $f$ on $\mathbb{R}^d$ and so we can work with logarithms and with partial derivatives. The independence property now says that $\partial_{v_i} \log f (x)$ is equal to $h_i((x,v_i))$ for some smooth function $h_i:\mathbb{R}\rightarrow\mathbb{R}$.
We obtain that
$$\partial_{v_{j_i}}\partial_{v_i}f=(v_i,v_{j_i})h_i''((x,v_i))~~,~~\partial_{v_i}\partial_{v_{j_i}}f=(v_i,v_{j_i})h_{j_i}''((x,v_{j_i})).
$$
and so $h_i''((x,v_i))=h_{j_i}''((x,v_{j_i}))$. Since $v_i$ and $v_{j_i}$ are independent for every pair $a,b\in\mathbb{R}$ there is $x\in\mathbb{R}^d$ such that $(x,v_i)=a$ and $(x,v_{j_i})=b$. This implies that $h_i''(a)=h_{j_i}''(b)$ holds for every $a,b$ and so each $h_i''$ is a constant function for every $i$.
Consequently $h_i$ is linear for every $i$ and thus (using that $\{v_i\}_{i=1}^d$ is a basis) $\partial_u \log f$ is a linear function for every $u\in\mathbb{R}^d$. It follows that $\log f$ satisfies $\partial_{u_1}\partial_{u_2}\partial_{u_3} \log f =0$ for every $u_1,u_2,u_3\in\mathbb{R}^d$. This means that $\log f$ is given by a quadratic $d$-variate polynomial $Q$ on $\mathbb{R}^d$. We obtain that $f=c\exp(Q(x))$ holds and thus $f$ is the density function of some Gaussian distribution.
\end{proof}
\begin{lemma}\label{twoposs} Let $(X,Y,Z)$ be a joint distribution with $X,Y\in\mathbb{R}$ and $Z\in\mathbb{R}^{d-2}$. Let $(X_t,Y_t,Z_t)$ be the triple obtained by running the heat equation for time $t$ with $X_0=X,Y_0=Y$ and $Z_0=Z$. Assume that for every $t\geq 0$ we have that $X_t$ and $Z_t$ are conditionally independent with respect to $Y_t$. Then either $(X,Y)$ is independent from $Z$ or $(Y,Z)$ is independent from $X$.
\end{lemma}
\begin{proof}
We parametrize $\mathbb{R}^d$ with triples $(x,y,\underline{z})$ where the coordinates $x$ and $y$ are real numbers and $\underline z$ is a $d-2$-dimensional vector. By $\Delta_z$ we mean the sum of the second partial derivates with respect to the coordinates belonging to $\underline{z}$. We denote by $f_t,h_t,g_t$ and $m_t$ the density functions of $(X_t,Y_t,Z_t),(X_t,Y_t),(Z_t,Y_t)$ and $Y_t$, respectively. We also introduce $s_t(\underline z, y)= g_t(\underline z, y)/m_t(y)$. Using the conditional independence and the heat equation we obtain the following equations.
\[f_t(x, y, \underline z)=h_t(x, y)g_t(\underline z,y)/m_t(y)=h_t(x,y)s_t(\underline z, y).\]
\[ \partial_tf_t=\Delta f_t; \qquad \partial_th_t=\Delta h_t; \qquad \partial_tg_t=\Delta g_t=\partial_{yy}g_t+\Delta_z g_t; \qquad \partial_tm_t=\Delta m_t. \]
By abusing the notation we will omit $t$ from $f_t,h_t,g_t,m_t$ and $s_t$ in the following calculations.
We start from the first equality, and use the other three one after the other.
\[\partial_{xx} f+\partial_{yy}f+\Delta_{z}f=\partial_th \cdot s+h\cdot\partial_ts.\]
\[\partial_{xx}h\cdot s+\partial_{yy}f+\Delta_{z}f=\partial_{xx}h\cdot s+\partial_{yy}h\cdot s+h\cdot \partial_ts.\]
\[\partial_{yy}f+h\cdot \Delta_{z}g/m=\partial_{yy}h\cdot s+h\cdot\big((\partial_tg)/m-(g\cdot \partial_tm)/m^2\big). \]
\begin{equation}\label{eq:partial1}\partial_{yy}h\cdot s+2\partial_{y}h\cdot \partial_{y} s+h\cdot\partial_{yy} s= \partial_{yy}h\cdot s+(h\cdot \partial_{yy}g)/m-(gh\cdot\partial_{yy}m)/m^2.\end{equation}
Before continuing this, we calculate the partial derivatives of $s$ with respect to $y$.
\[\partial_{y}s=(\partial_yg)/m-(g\cdot\partial_ym)/m^2.\]
\[\partial_{yy}s=(\partial_{yy}g)/m-2(\partial_y g\cdot \partial_y m)/m^2-(g\cdot\partial_{yy}m)/m^2+2g\cdot(\partial_y m)^2/m^3.\]
Now we substitute this into equation \eqref{eq:partial1}.
\begin{multline*}2(\partial_yh\cdot\partial_yg)/m-2g(\partial_yh\cdot\partial_ ym)/m^2+(h\cdot \partial_{yy}g)/m\\-2h(\partial_y g\cdot \partial_y m)/m^2-(gh\cdot \partial_{yy}m)/m^2+2hg(\partial_ym)^2/m^3=(h\cdot \partial_{yy}g)/m-(gh\cdot\partial_{yy}m)/m^2.\end{multline*}
\[(\partial_yh\cdot\partial_yg)/m-(g\cdot\partial_y h+h\cdot\partial_y g)\cdot(\partial_ym)/m^2+gh(\partial_ ym)^2/m^3=0.\]
\[(\partial_yh\cdot\partial_yg)/m^2-(g\cdot\partial_y h+h\cdot\partial_y g)\cdot(\partial_ym)/m^3+gh(\partial_ ym)^2/m^4=0.\]
\[\Big((\partial_y h)/m-(h\cdot\partial_y m)/m^2\Big)\Big((\partial_yg)/m-(g\cdot \partial_y m)/m^2\Big)=0.\]
\[\big(\partial_y(h/m)\big)\cdot \big(\partial_y(g/m)\big)=0.\]
We obtain that at least one of $\partial_y(h/m)=0$ and $\partial_y(g/m)=0$ holds on an open set. Assume that (without loss of generality) $\partial_y(h/m)=0$ holds on an open set $U$ of the domain $D$ of $h/m$ which is $\mathbb{R}^2$. This is equivalent to $m\partial_y h - h\partial_ym=0$ on $U$. Let $p\in U$ and $q\in D$ be arbitrary. Let us define the function $r:\mathbb{R}\rightarrow\mathbb{R}$ by $r(\lambda)=(m\partial_yh-h\partial_y m)(p+\lambda(q-p))$. Then $r$ has an analytic extension to $\mathbb{C}$. In addition, $r=0$ in a small neighborhood of $0$ in $\mathbb{R}$. It follows that $r$ is constant $0$ and thus $\partial_y (h/m)$ is $0$ at every $q\in D$. This implies that $X_t$ is independent of $Y_t$. Similarly if $\partial_y(g/m)=0$ holds on an open set we obtain that $Z_t$ is independent of $Y_t$. The conditional independence of $X_t$ and $Z_t$ with respect to $Y_t$ finishes the proof.
\hfill $\square$
\end{proof}
\section{Appendix B: Differential entropy}
Differential entropy is defined as follows for absolutely continuous random vectors. Some properties of the discrete entropy are preserved (e.g.\ it is additive if we put together independent random variables), others do not hold any more; an essential difference is that differential entropy does not have to be nonnegative.
\begin{definition}Let $(X_1, X_2, \ldots, X_n)$ be a family of random variables. Suppose that their joint distribution is absolutely continuous, and their joint density function is $f$. Then their differential entropy is defined as follows (provided that the integral exists):
\[\mathbb D(X_1, X_2, \ldots, X_n)=-\int_{\mathbb R^n} f(t_1, \ldots, t_n) \log f(t_1, \ldots, t_n)dt_1\ldots dt_n.\]
\end{definition}
To see the connection between entropy in the discrete case and differential entropy, recall Theorem 9.3.1\ from \cite{cover}. This says that if we divide the range of $X$ into bins of length $\delta$, and $X^{\delta}$ denotes the quantized version of $X$ with respect to this grid, then
\[\mathbb H(X^{\delta})+\log \delta\rightarrow \mathbb H(X) \]
as $\delta\rightarrow 0$, assuming that the density of $X$ is Riemann integrable.
The following well-known lemma shows how the differential entropy is modified when we apply a linear transformation to the random vector (see e.g. corollary to Theorem 9.6.4 in \cite{cover}.
\begin{lemma}\label{difftrdet} Let $\underline X=(X_1, \ldots, X_n)$ be a family of random variables, and let $A\in \mathbb R^{n\times n}$ be an invertible matrix. Then
\[\mathbb D(A\underline X)=\mathbb D(\underline X)+\log |\!\det (A)|.\]
\end{lemma}
The following lemma is equivalent to the fact that the nonnegativity of conditional mutual information holds for differential entropy as well. We include a proof for completeness.
\begin{lemma}\label{diffincform} Let $X, Y, Z$ be random variables such that their differential entropy exist. Then we have
\[\mathbb D(X, Y, Z)\leq \mathbb D(X, Z)+\mathbb D(Y, Z)-\mathbb D(Z).\]
\end{lemma}
\begin{proof}
Let $g(x,y)=f(x, y, z)/f(z)$ on the support of $Z$, and 0 otherwise. Then $g$ is a density function on $\mathbb R^2$. As the nonnegativity of mutual information is satisfied for differential entropy (see e.g.\ Corollary to Theorem 9.6.1.\ in \cite{cover}), we have
\[-\int g(x,y) \log g(x,y) dx dy \leq -\int g_1(x)\log g_1(x) dx-\int g_2(y) \log g_2(y) dy, \]
where $g_1$ and $g_2$ are the marginal densities of $g$. Multiplying both sides by $f(z)$ and integrating with respect to $z$ we get the statement of the lemma.
\end{proof} \hfill $\square$
\section{Appendix C: Factor of i.i.d.\ processes}
Let $f:[0,1]^{V_d}\rightarrow Y$ be a measurable function such that it is invariant under root preserving automorphisms. We can use $f$ to construct an invariant process in the following way. First we put independent uniformly random elements from $[0,1]$ on the vertices of $T_d$. Then, at each vertex $v$, we evaluate $f$ for this random labeling such that the root is placed to $v$. If $f$ depends only on finitely many coordinates, then the corresponding process is called a block factor of i.i.d.\ process.
\begin{proposition}\label{prop:elso}
If $\{X_v\}_{v\in V_d}$ is a real-valued typical process and $\{Y_v\}_{v\in V_d}$ is a weak limit of factor of i.i.d.\ processes, then their independent sum $\{X_v+Y_v\}_{v\in V_d}$ is a typical process.
\end{proposition}
\begin{proof}
Notice that the family of typical processes is closed with respect to the weak topology. On the other hand, every process that is a weak limit of factor i.i.d.\ processes is also a weak limit of block factor of i.i.d.\ processes \cite{russ}. Hence it is enough to prove the statement in the case when $\{Y_v\}_{v\in V_d}$ is a block factor of i.i.d.\ process. It is well known that block factor of i.i.d.\ processes can be approximated with the corresponding local algorithm computed on graphs with sufficiently large (essential) girth. Since large random $d$-regular graphs have large essential girth, the independent sum of the approximation of $\{X_v\}_{v\in V_d}$ and the local algorithm approximating $\{Y_v\}_{v\in V_d}$ is an approximation of the process $\{X_v+Y_v\}_{v\in V_d}$.
\hfill $\square$
\end{proof}
The next proposition was proved by Harangi and Vir\'ag in \cite{harangi}.
\begin{proposition}\label{prop:hv}
For $|\lambda|\leq 2\sqrt{d-1}$ the unique Gaussian wave $\Psi_\lambda$ is a weak limit of factor of i.i.d processes (but not a factor of i.i.d.\ process itself).
\end{proposition}
\subsection*{Acknowledgement.} The research leading to these results has received funding from the European Research Council under the European Union's Seventh Framework Programme (FP7/2007-2013) / ERC grant agreement n$^{\circ}$617747. The research was partially supported by the MTA R\'enyi Institute Lend\"ulet Limits of Structures Research Group.
| 5a0665e9d0c2712bbf177440e0dd0cfe20a66b6f | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
\subsection{Introduction}
The Fisher--Rao metric on the space $\on{Prob}(M)$ of probability densities is invariant under the action of the diffeomorphism group $\on{Diff}(M)$.
Restricted to finite-dimensional submanifolds of $\on{Prob}(M)$,
so-called statistical manifolds, it is called Fisher's information metric \cite{Ama1985}.
A uniqueness result was established \cite[p. 156]{Cen1982} for Fisher's information
metric on finite sample spaces and \cite{AJLS2014} extended it to infinite sample spaces.
The Fisher--Rao metric on the infinite-dimensional manifold of all positive probability densities
was studied in \cite{Fri1991}, including the computation of its curvature.
In \cite{BBM2016} it was proved that any $\on{Diff}(M)$-invariant Riemannian metric on the space $\on{Dens}_+(M)$ of smooth positive densities on a compact manifold $M$ without boundary is of the form
\begin{equation}\label{Fisher-Rao-metric}
G_\mu(\alpha,\beta)=C_1(\mu(M)) \int_M \frac{\alpha}{\mu}\frac{\beta}{\mu}\,\mu + C_2(\mu(M)) \int_M\alpha \cdot \int_M\beta
\end{equation}
for some smooth functions $C_1,C_2$ of the total volume $\mu(M)$.
This implies that the Fisher--Rao metric on $\on{Prob}(M)$ is, up to a multiplicative constant, the unique $\on{Diff}(M)$-invariant metric. By Cauchy--Schwarz the metric \eqref{Fisher-Rao-metric} is positive definite if and only if
$C_2(m)>-\frac1m C_1(m)$ for all $m>0$.
\subsection{The setting}
Let $M^m$ be a smooth compact manifold. It may have boundary or it may even be a manifold with corners; i.e., modelled on open subsets of quadrants in $\mathbb R^m$.
For a detailed description of the line bundle of smooth densities we refer to \cite{BBM2016} or \cite[10.2]{Mic2008}.
We let $\on{Dens}_+(M)$ denote the space of smooth positive densities on $M$, i.e.,
$\on{Dens}_+(M) =
\{ \mu \in \Gamma(\on{Vol}(M)) \,:\, \mu(x) > 0\; \forall x \in M\}$. Let $\on{Prob}(M)$ be the
subspace of positive densities with integral 1 on $M$. Both spaces are smooth Fr\'echet manifolds;
in particular they are open subsets of the affine spaces of all densities and densities of integral
1 respectively. For $\mu \in \on{Dens}_+(M)$ we have
$T_\mu \on{Dens}_+(M) = \Gamma(\on{Vol}(M))$ and for $\mu\in \on{Prob}(M)$ we have
$$
T_\mu\on{Prob}(M)=\{\alpha\in \Gamma(\on{Vol}(M)): \int_M\alpha =0\}.
$$
The Fisher--Rao metric, given by
$G^{\operatorname{FR}}_\mu(\alpha,\beta) = \int_M \frac{\alpha}{\mu}\frac{\beta}{\mu}\mu$ is a Riemannian metric on $\on{Prob}(M)$; it is invariant under the natural action of the group $\on{Diff}(M)$ of all diffeomorphisms of $M$.
If $M$ is compact without boundary of dimension $\ge 2$, the Fisher-Rao metric is the unique $\on{Diff}(M)$-invariant metric up to a multiplicative constant.
This follows, since any $\on{Diff}(M)$-invariant Riemannian metric on $\on{Dens}_+(M)$ is of the form \eqref{Fisher-Rao-metric} as proved in \cite{BBM2016}.
\subsection{Overview}
\label{overview}
We will study four different representations of the metric $G$ in \eqref{Fisher-Rao-metric}.
The first representation is $G$ itself on the space $\on{Dens}_{+}(M)$.
Next we fix a density $\mu_0 \in \on{Prob}(M)$ and consider the mapping
\[
R:\on{Dens}_+(M)\to C^\infty(M,\mathbb R_{>0})\,, \qquad R(\mu) = f = \sqrt{\frac{\mu}{\mu_{0}}}\,.
\]
This map is a diffeomorphism with inverse $R^{-1}(f)= f^2\mu_0$, and we will denote the induced metric by $\tilde G = \left(R^{-1}\right)^\ast G$; it is given by the formula
\[
\tilde G_{f}(h,k) = 4 C_1(\| f\|^2) \langle h, k \rangle
+ 4 C_2(\| f \|^2) \langle f, h \rangle \langle f, k \rangle\,,
\]
with $\| f\|^2 = \int_M f^2 \mu_0$ denoting the $L^2(\mu_0)$-norm, and this formula makes sense for $f \in C^{\infty}(M,\mathbb R)$. See Sect.\ \ref{geodesics} for calculations.
Next we take the pre-Hilbert space $(C^{\infty}(M,\mathbb R), \langle\;,\;\rangle_{L^2(\mu_0)})$ and pass to polar coordinates.
Let
$S = \{ \varphi \in L^2(M,{\mathbb R}) \,:\, \int_M \varphi^2 \mu_0 = 1\}$ denote the $L^2$-sphere. Then
\[
\Phi : C^\infty(M, {\mathbb R}_{>0}) \to {\mathbb R}_{>0} \times (S \cap C^\infty_{>0})\,,\qquad
\Phi(f) = (r,\varphi) = \left(\| f \|, \frac f{\|f\|}\right)\,,
\]
is a diffeomorphism, where $C^\infty_{>0} = C^\infty(M,{\mathbb R}_{>0})$; its inverse is $\Phi^{-1}(r,\varphi)= r.\varphi$. We set $\bar G = \left(\Phi^{-1}\right)^\ast \tilde G$; the metric $\bar G$ has the expression
\[
\bar G_{r,\varphi} = g_1(r) \langle d\varphi, d\varphi \rangle + g_2(r) dr^2\,,
\]
with $g_1(r) = 4C_1(r^2) r^2$ and $g_2(r) = 4\left(C_1(r^2) + C_2(r^2) r^2\right)$.
Finally we change the coordinate $r$ diffeomorphically to
\[
s= W(r) = \int_1^r \sqrt{g_2(\rho)} \,d\rho\,.
\]
Then, defining $a(s) = 4 C_1(r(s)^2) r(s)^2$, we have
\[
\bar G_{s,\varphi} = a(s) \langle d\varphi, d\varphi \rangle + ds^2\,.
\]
We will use $\bar G$ to denote the metric in both $(r,\varphi)$ and $(s,\varphi)$ coordinates.
Let $W_- = \lim_{r\to 0+} W(r)$ and $W_+ = \lim_{r\to \infty} W(r)$. Then $W: {\mathbb R}_{>0} \to (W_-, W_+)$ is a diffeomorphism. This completes the first row in Fig.~\ref{fig:diagram}. The geodesic equation of $G$ in the various representations will be derived in Sect.~\ref{geodesics}. The formulas for the geodesic equation and later for curvature are infinite-dimensional analoga of the corresponding formulas for warped products; see \cite[p.~204ff]{ONeill1983} or \cite[Chap.~7]{Chen2011}.
The four representations are summarized in the following diagram.
\[
\xymatrix @C-0.5pc {
\on{Dens}_+(M) \ar[r]^-R & C^\infty(M,{\mathbb R}_{>0}) \ar[r]^-\Phi &
{\mathbb R}_{>0} \times (S \cap C^\infty_{>0}) \ar[r]^-{W \times \on{Id}} &
(W_-, W_+) \times (S \cap C^\infty_{>0})\,.
}
\]
\begin{figure}
\[
\xymatrix @C-1pc {
\xyoption{rotate}
\on{Dens}_+(M) \ar[r]^-R \ar[d] & C^\infty(M,\mathbb R_{>0}) \ar[r]^-\Phi \ar[d] &
\mathbb R_{>0} \!\times\! S \cap C^\infty_{>0} \ar[r]^-{W \times \on{Id}} \ar[d] &
(W_-, W_+) \!\times\! S \cap C^\infty_{>0} \ar[d] \\
\Gamma_{C^1}(\on{Vol}(M)) \!\setminus\! \{0\} \ar[d] &
C^\infty(M,\mathbb R) \!\setminus\! \{0\} \ar[l]_{\;\;R^{-1}} \ar[r]^-\Phi \ar[d] &
\mathbb R_{>0} \!\times\! S \cap C^\infty\ar[r]^-{W \times \on{Id}} \ar[d] & \mathbb R \!\times\! S \cap C^\infty \ar[d] \\
\Gamma_{L^1}(\on{Vol}(M)) \!\setminus\! \{0\} \ar[r]^-R &
L^2(M,\mathbb R) \!\setminus\! \{0\} \ar[r]^-\Phi &
\mathbb R_{>0} \!\times\! S \ar[r]^-{W \times \on{Id}} & \mathbb R \!\times\! S
}
\]
\caption{Representations of $\on{Dens}_+(M)$ and its completions. In the second and third rows we assume that $(W_-, W_+) = (-\infty,+\infty)$ and we note that $R$ is a diffeomorphism only in the first row.}
\label{fig:diagram}
\end{figure}
Since $\bar G$ induces the canonical metric on $(W_-, W_+)$, a necessary condition for $\bar G$
to be geodesically complete is $(W_-, W_+) = (-\infty, +\infty)$. Rewritten in terms of the functions $C_1$ and $C_2$ this becomes
\[
W_+ = \infty
\Leftrightarrow\left(
\int_1^\infty r^{-1/2} \sqrt{C_1(r)} \,dr = \infty
\text{ or }
\int_1^\infty \sqrt{C_2(r)} \,dr = \infty
\right)\,,
\]
and similarly for $W_-=-\infty$, with the limits of integration being 0 and 1.
If $\bar G$ is incomplete, i.e., $W_->-\infty$ or $W_+<\infty$, there are sometimes geodesic completions. See Sect.~\ref{completions} for details.
We now assume that $(W_-, W_+) = (-\infty, +\infty)$. The metrics $\bar G$ and $\tilde G$ can be extended to the spaces ${\mathbb R}\times S \cap C^\infty$ and $C^\infty(M,{\mathbb R}) \setminus \{0\}$ and the last two maps in the diagram
\[
\xymatrix{
\Gamma_{C^1}(\on{Vol}(M)) \!\setminus\! \{0\} &
C^\infty(M,{\mathbb R}) \!\setminus\! \{0\} \ar[l]_{\qquad R^{-1}} \ar[r]^-\Phi &
{\mathbb R}_{>0} \times S \cap C^\infty \ar[r]^-{W \times \on{Id}}& {\mathbb R} \times S \cap C^\infty
}
\]
are bijections. The extension of $R^{-1}$ is given by $R^{-1}(f) = f |f| \mu_0$; it does not map into smooth densities any more, but only into $C^1$-sections of the volume bundle; however, $R^{-1}$ is not surjective into $C^1$-sections, because the loss of regularity for $R^{-1}(f)$ occurs only at point where $f$ is $0$.
The last two maps, $\Phi$ and $W \times \on{Id}$, are diffeomorphisms.
The following will be shown in Sect.~\ref{completeness}: $(W_-, W_+) = (-\infty, +\infty)$ implies that $({\mathbb R} \times S \cap C^\infty, \bar G)$ is geodesically complete and hence so are $({\mathbb R}_{>0} \times S \cap C^\infty, \bar G)$ and $(C^\infty(M,{\mathbb R}) \setminus \{0\}, \tilde G)$.
Finally we consider the metric completions, still assuming that $(W_-, W_+) = (-\infty, +\infty)$.
For $\bar G$ this is ${\mathbb R} \times S$ or ${\mathbb R}_{>0} \times S$ in $(s,\varphi)$ or $(r,\varphi)$-coordinates, respectively, as shown in Sect.~\ref{completeness}. The metrics and maps can be extended to
\[
\xymatrix{
\Gamma_{L^1}(\on{Vol}(M)) \setminus \{0\} \ar[r]^-R &
L^2(M,{\mathbb R}) \setminus \{0\} \ar[r]^-\Phi &
{\mathbb R}_{>0} \times S \ar[r]^-{W \times \on{Id}} & {\mathbb R} \times S
}\,.
\]
Here $\Gamma_{L^1}$ denotes the space of $L^1$-sections. The extension of $R$ is given by $R(\mu) = \on{sgn}(\mu) \sqrt{|\mu|/\mu_0}$ and its inverse is $R^{-1}(f) = f |f| \mu_0$ as before. The last two maps are diffeomorphisms and hence $(L^2(M,{\mathbb R}) \setminus \{0\}, \tilde G)$ is metrically complete. The extension of $R$ is bijective, but not a diffeomorphism. It is continuous, but not $C^1$, and its inverse is $C^1$, but not $C^2$; furthermore $DR^{-1}(f)$ is not surjective if $f=0$ on a set of positive measure. However we can use $R$ to pull back the geodesic distance function from $L^2(M,{\mathbb R}) \setminus \{0\}$ to $\Gamma_{L^1}(\on{Vol}(M)) \setminus \{0\}$ to obtain a complete metric on the latter space, that is compatible with the standard topology.
\subsection{The inverse $R^{-1}$ and geodesic completeness}
There is more than one choice for the extension of $R^{-1}(f) = f^2 \mu_0$ from $C^\infty(M,{\mathbb R}_{>0})$ to $C^\infty(M,{\mathbb R})$. The choice $R^{-1}(f) = f |f| \mu_0$ remains injective and can be further extended to a bijection on the metric completion $L^2(M,{\mathbb R}) \setminus \{0\}$. We can consider the equally natural extension $Q$ and its factorizarion given by
\begin{gather*}
\xymatrix{
C^\infty(M, \mathbb R) \ar[rr]^{Q} \ar[dr]_{Q_1} && \Gamma_{\ge 0}(\on{Vol}(M))\\
& \{|f|: f\in C^\infty(M,\mathbb R)\} \ar[ur]_{Q_2}&
}
\\
Q(f) = f^2\mu_0\,,\quad Q_1(f) = |f|\,, \quad Q_2(|f|) = |f|^2\mu_0\,.
\end{gather*}
into the space of smooth, nonnegative sections. The map $Q$ is not surjective; see \cite{KLM04} for a discussion of smooth non-negative functions admitting smooth square roots.
The image $ \{|f|: f\in C^\infty(M,\mathbb R)\}$ of $Q_1$ looks somewhat like the orbit space of a discrete reflection group:
An example of a codimension 1 wall of the image could be $\{|f|: f\in C^\infty(M,\mathbb R), f(x)=0\}$ for one fixed point $x\in M$. Since this is dense in the $L^2$-completion of $T_f C^\infty(M,\mathbb R)$ with respect to $\tilde G_f$, we do not have a reflection at this wall.
Fixing $\varphi_0\in S\cap C^\infty$ and considering $\{(r,\varphi) \in \mathbb R_{>0}\times S\cap C^\infty: \langle \varphi_0, \varphi\rangle = 0 \}$ we can write the orthogonal reflection $(r, t_1 \varphi_0 + t_2 \varphi)\mapsto (r, -t_1\varphi_0 + t_2\varphi)$.
Geodesics in $(C^\infty(M,\mathbb R),\tilde G)$ are mapped by $Q_1$ to curves that are geodesics in the interior $C^\infty(M,\mathbb R_{>0})$, and that are reflected following Snell's law at any hyperplanes in the boundary for which the angle makes sense. The mapping $Q_2$ then smoothes out the reflection to a `quadratic glancing of the boundary' if one can describe the smooth structure of the boundary. It is tempting to paraphrase this as: \emph{The image of $Q$ is geodesically complete.}
But note that: \thetag{1} The metric $G$ becomes ill-defined on the boundary. \thetag{2} The boundary is very complicated; each closed subset of $M$ is the zeroset of a smooth non-negative function and thus corresponds to a `boundary component'. Some of them `look like reflection walls'. One could try to set up a theory of infinite dimensional stratified Riemannian manifolds and geodesics on them to capture this notion of geodesic completeness, similarly to \cite{Michor03orbit}. But the situation is quite clear geometrically, and we prefer to consider the geodesic completion described by the inverse $R^{-1}$ used in this paper, which is perhaps more natural.
\subsection{Geodesics of the Fisher-Rao metric on $\on{Dens}_+(M)$}\label{geodesics}
In \cite{Fri1991} it was shown that $\on{Prob}(M)$ has constant sectional curvature for the Fisher-Rao metric. For fixed
$\mu_{0}\in \on{Prob}(M)$ we consider the mapping
$$R:\on{Dens}_+(M)\to C^\infty(M,\mathbb R_{>0}), \qquad R(\mu) = \sqrt{\frac{\mu}{\mu_{0}}}\,.$$
The inverse $R^{-1}: C^{\infty}(M,\mathbb R_{>0})\to \on{Dens}_{+}(M)$ is given by $R^{-1}(f)=f^{2}\mu_{0}$; its tangent mapping is $T_{f}R^{-1}.h = 2fh\mu_{0}$.
\begin{remark*}
In \cite{KLMP2013} it was shown that for $C_1 \equiv 1$ and $C_2\equiv 0$ the rescaled map
$R(\mu) = 2 \sqrt{\frac{\mu}{\mu_{0}}}$ is an isometric diffeomorphism from $\on{Prob}(M)$ onto the open subset $C^\infty(M, {\mathbb R}_{>0}) \cap \{ f \,:\, \int f^2 \mu_0 = 4\}$ of the $L^2$-sphere of radius $2$ in the pre-Hilbert space
$(C^{\infty}(M,\mathbb R), \langle\;,\;\rangle_{L^{2}(\mu_{0})})$. For a general function $C_1$ the same holds for
$R(\mu) = \lambda \sqrt{\frac{\mu}{\mu_{0}}}$ and the $L^2$-sphere of radius $\lambda$, where $\lambda>0$ is a solution of the equation $\lambda^2 = 4C_1(\lambda^{-2})$.
\end{remark*}
The Fisher--Rao metric
induces the following metric on the open convex cone
$C^{\infty}(M,\mathbb R_{>0})\subset C^{\infty}(M,\mathbb R)$:
\begin{multline*}\tag{a}
\left((R^{-1})^{*}G\right)_{f}(h,k) = G_{R^{-1}(f)}(T_{f}R^{-1}.h, T_{f}R^{-1}.k) = G_{f^2\mu_0}(2fh\mu_0,2fk\mu_0)
\\
= C_1(\|f\|^2_{L^2(\mu_0)}) \int \frac{2fh\mu_0}{f^2\mu_0}\frac{2fk\mu_0}{f^2\mu_0} f^2\mu_0
+ C_2(\|f\|^2_{L^2(\mu_0)})\int 2fh\mu_0\cdot \int 2fk\mu_0
\\
= 4C_1(\|f\|^2) \int hk\mu_0 + 4C_2(\|f\|^2)\int fh\mu_0\cdot \int fk\mu_0
\\= 4C_1(\|f\|^2) \langle h,k\rangle + 4C_2(\|f\|^2)\langle f,h\rangle \langle f,k\rangle
\\
\shoveleft{= 4C_1(\|f\|^2) \Big\langle h-\frac{\langle f,h\rangle}{\|f\|^2} f,k-\frac{\langle f,k\rangle}{\|f\|^2} f\Big\rangle\; +}
\\
+ 4\big(C_2(\|f\|^2).\|f\|^2 + C_1(\|f\|^2)\big)\Big\langle \frac{f}{\|f\|},h\Big\rangle \Big\langle \frac{f}{\|f\|},k\Big\rangle\,,
\end{multline*}
where in the last expression we split $h$ and $k$ into the parts perpendicular to $f$ and multiples of $f$.
We now switch to
\emph{polar coordinates on the pre-Hilbert space}: Let $S = \{\varphi\in L^2(M,{\mathbb R}): \int \varphi^2\mu_0 =1 \}$ denote the sphere, and let $S\cap C^{\infty}_{>0}$ be the intersection with the positive cone. Then
$C^{\infty}(M,\mathbb R) \setminus \{0\}\cong \mathbb R_{>0}\times S\cap C^{\infty}$ via
\[
\Phi : C^\infty(M, {\mathbb R}) \setminus \{0\} \to {\mathbb R}_{>0} \times S\,,\qquad
\Phi(f) = (r,\varphi) = \left(\| f \|, \frac f{\|f\|}\right)\,.
\]
Note that $\Phi(C^{\infty}(M,\mathbb R_{>0}))=\mathbb R_{>0}\times S\cap C^{\infty}_{>0}$.
We have $f = \Phi^{-1}(r,\varphi) = r.\varphi$
thus $df = r\,d\varphi +\varphi\,dr $,
where $r\,d\varphi(h)= h- \langle\varphi,h\rangle\varphi$ is the orthogonal projection onto the tangent space of $S$ at $\varphi$ and $dr(h)=\langle\varphi,h\rangle$. The Euclidean (pre-Hilbert) metric in polar coordinates is given by
\begin{align*}
\langle df,df \rangle &= \langle \varphi.dr + r.d\varphi,\varphi.dr + r.d\varphi\rangle
= \langle\varphi,\varphi\rangle dr^2 + 2r.\langle\varphi,d\varphi\rangle.dr + r^2\langle d\varphi,d\varphi\rangle
\\&
= dr^2 + r^2\langle d\varphi,d\varphi\rangle\,.
\end{align*}
The pullback metric is then
\begin{align*}\tag{b}
\bar G=\left((\Phi^{-1})^\ast \tilde G\right)
&= 4C_1(r^2) r^2 \langle d\varphi,d\varphi\rangle + 4\left( C_2(r^2) r^2 + C_1(r^2)\right) dr^2
\\
&= g_1(r) \langle d\varphi,d\varphi\rangle + g_2(r) dr^2
\\
&= a(s) \langle d\varphi,d\varphi\rangle + ds^2 \,,
\end{align*}
where we introduced the functions
$$ g_1(r) = 4C_1(r^2) r^2
\quad \text{ and }\quad
g_2(r) = 4\left( C_2(r^2) r^2 + C_1(r^2)\right)\,,$$
and where in the last expression we changed the coordinate $r$
diffeomorphically to
$$
s(r) = 2 \int_1^r \sqrt{ C_2(\rho^2) \rho^2 + C_1(\rho^2)}\,d\rho
\quad\text{ and let }a(s) = 4C_1(r(s)^2) r(s)^2.
$$
The resulting metric is
a radius dependent scaling of the metric on the sphere times a different radius dependent scaling of the metric on ${\mathbb R}_{>0}$. Note that the metric \thetag{b} (as well as the metric in the last expression of \thetag{a}) is actually well-defined on $C^{\infty}(M,\mathbb R)\setminus \{0\} \cong \mathbb R_{>0}\times S \cap C^\infty$; this leads to a (partial) geodesic completion of $(\on{Dens}_+(M), G)$.
Geodesics for the metric \thetag{b} follow great circles on the sphere with some time dependent stretching, since reflection at any hyperplane containing this great circle is an isometry.
We derive the geodesic equation. Let
$[0,1]\times(-\varepsilon,\varepsilon)\ni (t,s)\mapsto (r(t,s),\varphi(t,s))$ be a smooth variation with fixed ends of a curve $(r(t,0),\varphi(t,0))$.
The energy of the curve and its derivative with respect to the variation parameter $s$ are as follows, where $\nabla^S$ is the covariant derivative on the sphere $S$.
\begin{align*}
E(r,\varphi) &=
\int_0^1 \left( \frac 12 g_1(r) \langle \varphi_t, \varphi_t \rangle
+ \frac 12 g_2(r). r_t^2 \right) dt \\
\partial_s E(r,\varphi) &=
\int_0^1 \Big( \frac 12 g_1'(r).r_s \langle \varphi_t, \varphi_t \rangle +
g_1(r) \langle \nabla^S_{\partial_s} \varphi_t, \varphi_t \rangle + \\
&\qquad\qquad
+\frac 12 g_2'(r).r_s.r_t^2 + g_2(r).r_t.r_{ts} \Big) dt \\
&= \int_0^1 \Big( \frac 12 g_1'(r).r_s \langle \varphi_t, \varphi_t \rangle
- g_1'(r).r_t \langle \varphi_s, \varphi_t \rangle
- g_1(r) \langle \varphi_s, \nabla^S_{\partial_t} \varphi_t \rangle + \\
&\qquad\qquad
+\frac 12 g_2'(r).r_s.r_t^2 - g_2'(r).r_t^2.r_s - g_2(r).r_{tt}.r_s \Big) dt \\
&= \int_0^1 \Big( \frac 12 g_1'(r) \langle \varphi_t, \varphi_t \rangle
- \frac 12 g_2'(r).r_t^2 - g_2(r).r_{tt} \Big) r_s \\
&\qquad\qquad
- \Big( g_1'(r).r_t \langle \varphi_s, \varphi_t \rangle
+ g_1(r) \langle \varphi_s, \nabla^S_{\partial_t} \varphi_t \rangle \Big) dt\,.
\end{align*}
Thus the geodesic equation is
\begin{equation*}\tag{c}\boxed{\;
\begin{aligned}
\nabla^S_{\partial_t}\varphi_{t} &=
- \partial_t\left( \log g_1(r) \right) \varphi_t
\\
r_{tt} &=
\frac 12 \frac{ g_1'(r)}{g_2(r)} \langle \varphi_t, \varphi_t \rangle
-\frac 12 \partial_t \left( \log g_2(r) \right) r_t
\end{aligned}
\;}\end{equation*}
Using the first equation we get:
\begin{align*}
\partial_t\langle \varphi_t,\varphi_t\rangle
&= 2\langle\nabla_{\partial_t}\varphi_t,\varphi_t\rangle
= -2\, \partial_t\left( \log g_1(r) \right) \langle \varphi_t, \varphi_t \rangle
\\
\partial_t \left( \log \langle \varphi_t, \varphi_t \rangle \right)
&= -2\, \partial_t\left( \log g_1(r) \right)
\\
\log(\|\varphi_t\|^2) &= -2 \log g_1(r) + 2\log A_0
\quad\text{ with }\quad
A_0 = g_1(r)\, \| \varphi_t \| \,,
\end{align*}
which describes the speed of $\varphi(t)$ along the great circle in terms of $r(t)$; note that the quantity $g_1(r) \| \varphi_t\|$ is constant in $t$. The geodesic equation \thetag{c} simplifies to
\begin{equation*}\tag{d}\boxed{\;
\begin{aligned}
\nabla^S_{\partial_t}\varphi_{t} &=
-\partial_t\left( \log g_1(r) \right) \varphi_t
\\
r_{tt} &= \frac {A_0^2}2 \frac{g_1'(r)}{g_1(r)^2 g_2(r)}
-\frac 12 \partial_t \left( \log g_2(r) \right) r_t
\end{aligned}
\;}\end{equation*}
with $g_1(r) = 4C_1(r^2) r^2$ and
$g_2(r) = 4\left( C_2(r^2) r^2 + C_1(r^2)\right)$.
We can solve equation \thetag{d} for $\varphi$ explicitely. Given initial conditions $\varphi_0, \psi_0$, the geodesic $\tilde \varphi(t)$ on the sphere with radius $1$ satisfying $\tilde \varphi(0) = \varphi_0$, $\tilde \varphi_t(0) = \psi_0$ is
\[
\tilde\varphi(t) = \cos(\| \psi_0\| t) \varphi_0 + \sin(\|\psi_0\| t ) \frac{\psi_0}{\|\psi_0\|}\,.
\]
We are looking for a reparametrization $\varphi(t) = \tilde \varphi(\alpha(t))$. Inserting this into the geodesic equation we obtain
\begin{align*}
\partial_t^2 \left(\tilde \varphi(\alpha) \right)
- \left\langle \partial_t^2 \left(\tilde \varphi(\alpha)\right),
\frac{\tilde \varphi(\alpha)}{\| \tilde \varphi(\alpha) \|} \right\rangle \tilde \varphi(\alpha)
&= -\partial_t \left(\log g_1(r) \right) \partial_t \left(\tilde \varphi(\alpha)\right) \\
\left(\nabla^S_{\partial_t}\tilde \varphi_t\right)(\alpha) \alpha_t^2 + \tilde\varphi_t(\alpha) \alpha_{tt}
- \left\langle \tilde\varphi_t(\alpha) \alpha_{tt},
\frac{\tilde \varphi(\alpha)}{\| \tilde \varphi(\alpha) \|} \right\rangle \tilde \varphi(\alpha)
&= -\partial_t \left(\log g_1(r) \right) \tilde\varphi_t(\alpha) \alpha_t \\
\alpha_{tt} = \partial_t \left(\log g_1(r) \right) \alpha_t\,.
\end{align*}
With intial conditions $\alpha(0) = 0$ and $\alpha_t(0) = 1$ this equation has the solution
\[
\alpha(t) ={g_1(r_0)} \int_0^{t} \frac{1}{g_1(r(\tau))} \,d\tau\,,
\]
where $r_0 = r(0)$ is the initial condition for the $r$-component of the geodesic.
If the metric is written in the form $\bar G = ds^2 + a(s)\langle d\varphi,d\varphi\rangle$, equation \thetag{d} becomes
\begin{equation*}
s_{tt} = \frac {A_0^2}2 \frac{a'(s)}{a(s)^2}\,,\quad
\text{ for } A_0 = a(s) \|\varphi_t\|\,,
\end{equation*}
where $\varphi(t)$ is given explicitly as above. This can be integrated into the form
\begin{equation*}\tag{e}\boxed{\;
s_{t}^2 = -\frac {A_0^2}{a(s)} + A_1\,,\quad A_1 \text{ a constant.}
\;}\end{equation*}
\subsection{Relation to hypersurfaces of revolution} \label{revolution}
We consider the metric $\bar G$
on $(W_-,W_+)\times S\cap C^\infty$ where
$\bar G_{s,\varphi}=a(s) \langle d\varphi, d\varphi \rangle + ds^2$ and
$a(s) = 4C_1(r(s)^2) r(s)^2$.
Then the map $\Psi$ is an isometric embedding (remember $\langle \varphi,d\varphi\rangle=0$ on
$S\cap C^\infty$),
\begin{align*}
&\Psi: ((W_-,W_+)\times S\cap C^\infty, \bar G) \to \big(\mathbb R\times C^{\infty}(M,\mathbb R),
du^2 + \langle df,df\rangle\big)\,,
\\&
\Psi(s,\varphi) = \Big(\int_0^s \sqrt{1 -\frac{ a'(\sigma)^2}{4 a(\sigma)}}\,d\sigma\;,\; \sqrt{ a(s)}\varphi\Big)\,,
\end{align*}
In fact it is defined and smooth only on the open subset
$$\left\{(s,\varphi)\in (W_-,W_+)\times S\cap C^{\infty}: a'(s)^2 < 4 a(s)\right\}\,.$$
We will see in Sect.~\ref{curvature} that the condition $a'(s)^2 < 4 a(s)$ is equivalent to a sign condition on the sectional curvature; to be precise
\[
a'(s)^2 < 4 a(s) \Leftrightarrow \on{Sec}_{(s,\varphi)}(\on{span}(X,Y)) > 0\,,
\]
where $X,Y \in T_\varphi S$ is any $\bar G$-orthonormal pair of tangent vectors.
Fix some $\varphi_0\in S\cap C^\infty$ and consider the generating curve
\[
\gamma(s) = \Big(\int_0^s \sqrt{1 -\frac{ a'(\sigma)^2}{4 a(\sigma)}}\,d\sigma\;,\; \sqrt{ a(s)} \varphi_0\Big)\in {\mathbb R} \times C^\infty(M,{\mathbb R})\,;
\]
then $\gamma(s)$ is already arc-length parametrized!
Any arc-length parameterized curve $I\ni s\mapsto (c_1(s),c_2(s))$ in $\mathbb R^2$ generates a hypersurface of revolution
$$\{(c_1(s), c_2(s)\varphi): s\in I, \varphi\in S\cap C^\infty\}\subset \mathbb R\times C^\infty(M,\mathbb R)\,,$$
and the induced metric in the $(s,\varphi)$-parameterization is $c_2(s)^2\langle d\varphi,d\varphi\rangle + ds^2$.
This suggests that the moduli space of hypersurfaces of revolution is naturally embedded in the moduli space of all metrics of the form $a(s) \langle d\varphi, d\varphi \rangle + ds^2$.
Let us make this more precise in an example: In the case of $S=S^1$ and the tractrix $(c_1,c_2)$, the surface of revolution is the pseudosphere (curvature $-1$) whose universal cover is only part of the hyperbolic plane. But in polar coordinates we get a space whose universal cover is the whole hyperbolic plane. In detail: the arc-length parametrization of the tractrix and the induced metric are
\begin{align*}
&c_1(s) = \int_0^s\sqrt{1-e^{-2\sigma}}\,d\sigma = \on{Arcosh}\big({e^{s}}\big) - \sqrt{1-e^{-2s}}
,\quad c_2(s) = e^{-s},\quad s> 0
\\
&a(s)\,d\varphi^2 + ds^2 = e^{-2s}d\varphi^2 + ds^2, \qquad s\in \mathbb R\,.
\end{align*}
\subsection{Completeness}
\label{completeness}
In this section we assume that $(W_-, W_+) = (-\infty, +\infty)$, which is a necessary and sufficient condition for completeness. First we have the following estimate for the geodesic distance $\on{dist}$ of the metric $\bar G$, which is valid on bounded metric balls. Let $\on{dist}_{S}$ denote the geodesic distance on $S$ with respect to the standard metric.
\begin{lemma*}
\label{lem:dist_est}
Let $(W_-, W_+) = (-\infty, +\infty)$, $(s_0, \varphi_0) \in {\mathbb R} \times S$ and $R>0$. Then there exists $C>0$, such that
\begin{multline*}
C^{-1}\left( \on{dist}_S(\varphi_1, \varphi_2) + |s_1 - s_2| \right)
\leq \on{dist}\left((s_1, \varphi_1), (s_2, \varphi_2)\right) \leq \\
\leq C \left( \on{dist}_S(\varphi_1, \varphi_2) + |s_1 - s_2| \right)\,,
\end{multline*}
holds for all $(s_i, \varphi_i)$ with $\on{dist}\left((s_0, \varphi_0), (s_i, \varphi_i)\right) < R$, $i=1,2$.
\end{lemma*}
\begin{proof}
First we observe that
\[
|s_1 - s_2| \leq \int_0^1 |s_t(t)| \,dt
\leq \int_0^1 \sqrt{ a(s) \| \varphi_t \|^2 + s_t^2 } \,dt = \on{Len}(s, \varphi)\,,
\]
and hence by taking the infimum over all paths,
\[
|s_1 - s_2| \leq \on{dist}\left((s_1, \varphi_1), (s_2, \varphi_2)\right) < 2R\,.
\]
Thus $s$ is bounded on bounded geodesic balls.
Now let $(s_i, \varphi_i)$ be chosen according to the assumptions and let $(s(t), \varphi(t))$ be a path connecting $(s_1, \varphi_1)$ and $(s_2, \varphi_2)$ with $\on{Len}(s,\varphi) < 2 \on{dist}\left((s_1, \varphi_1), (s_2, \varphi_2)\right)$. Then for $t \in [0,1]$,
\[
\on{dist}\left((s_0, \varphi_0), (s(t), \varphi(t))\right)
\leq \on{dist}\left((s_0, \varphi_0), (s_1, \varphi_1)\right)
+ 2\on{dist}\left((s_1, \varphi_1), (s_2, \varphi_2)\right)
\leq 5 R\,.
\]
In particular the path remains in a bounded geodesic ball.
Thus there exists a constant $C > 1$, such that $C^{-1} \leq a(s) \leq C$ holds along $(s(t),\varphi(t))$. From there we obtain
\[
C^{-1} \int_0^1 \| \varphi_t \|^2 + s_t^2 \,dt
\leq \int_0^1 a(s) \| \varphi_t \|^2 + s_t^2 \,dt
\leq C \int_0^1 \| \varphi_t \|^2 + s_t^2 \,dt\,,
\]
and by taking the infimum over paths connecting $(s_1, \varphi_1)$ and $(s_2, \varphi_2)$ the desired result follows.
\end{proof}
\begin{proposition*} If $(W_-, W_+) = (-\infty, +\infty)$,
the space $({\mathbb R} \times S, \bar G)$ is metrically and geodesically complete. The subspace $({\mathbb R}\times S \cap C^\infty, \bar G)$ is geodesically complete.
\end{proposition*}
\begin{proof}
Given a Cauchy sequence $(s_n, \varphi_n)_{n \in \mathbb N}$ in ${\mathbb R} \times S$ with respect to the geodesic distance, the lemma shows that $(s_n)_{n \in \mathbb N}$ and $(\varphi_n)_{n \in \mathbb N}$ are Cauchy sequences in ${\mathbb R}$ and $S$ respectively. Hence they have limits $s$ and $\varphi$ and by the lemma the sequence $(s_n, \varphi_n)_{n \in \mathbb N}$ converges to $(s,\varphi)$ in the geodesic distance as well. It is shown in \cite[Prop.~6.5]{Lang1999} that a metrically complete, strong Riemannian manifold is geodesically complete.
Since the $\varphi$-part of a geodesic in ${\mathbb R} \times S$ is a reparametrization of a great circle, if the initial conditions lie in ${\mathbb R} \times S \cap C^\infty$, so will the whole geodesic. Hence ${\mathbb R} \times S \cap C^\infty$ is geodesically complete.
\end{proof}
The map $W \times \on{Id} \circ \Phi : L^2(M,{\mathbb R}) \setminus \{ 0 \} \to {\mathbb R} \times S$ is a diffeomorphism and an isometry with respect to the metrics $\tilde G$ and $\bar G$.
\begin{corollary*} If $(W_-, W_+) = (-\infty, +\infty)$,
the space $(L^2(M,{\mathbb R}) \setminus \{ 0\}, \tilde G)$ is metrically and geodesically complete. The subset $(C^\infty(M,{\mathbb R}) \setminus \{0\}, \tilde G)$ is geodesically complete.
\end{corollary*}
It remains to consider the existence of minimal geodesics.
\begin{theorem*} If $(W_-, W_+) = (-\infty, +\infty)$, then
any two points $(s_0,\varphi_0)$ and $(s_1,\varphi_1)$ in $\mathbb R\times S$ can be joined by a minimal geodesic.
If $\varphi_0$ and $\varphi_1$ lie in $S\cap C^\infty$, then the minimal geodesic also lies in $\mathbb R\times S\cap C^\infty$.
\end{theorem*}
\begin{proof}
If $\varphi_0$ and $\varphi_1$ are linearly independent, we consider the 2-space $V=V(\varphi_0,\varphi_1)$ spanned by $\varphi_0$ and $\varphi_1$ in $L^2$. Then $\mathbb R\times V\cap S$ is totally geodesic since it is the fixed point set of the isometry $(s,\varphi)\mapsto (s, \mathfrak s_V(\varphi))$ where $\mathfrak s_V$ is the orthogonal reflection at $V$. Thus there is exists a minimizing geodesic between $(s_0,\varphi_0)$ and $(s_1,\varphi_1)$ in the complete 3-dimensional Riemannian submanifold $\mathbb R\times V\cap S$.
This geodesic is also length-minimizing in the strong Hilbert manifold $\mathbb R\times S$ by the following argument:
Given any smooth curve $c = (s,\varphi):[0,1]\to \mathbb R\times S$ between these two points,
there is a subdivision $0=t_0<t_1<\dots<t_N=1$
such that the piecewise geodesic $c_1$ which first runs along a geodesic from $c(t_0)$ to $c(t_1)$, then to $c(t_2)$, \dots, and finally to $c(t_N)$, has length
$\on{Len}(c_1)\le \on{Len}(c)$.
This piecewise geodesic now lies in the totally geodesic $(N+2)$-dimensional submanifold
$\mathbb R\times V(\varphi(t_0),\dots,\varphi(t_N))\cap S$. Thus there exists a geodesic $c_2$ between the two points $(s_0,\varphi_0)$ and $(s_1,\varphi_1)$ which is length-minimizing in this $(N+2)$-dimensional submanifold.
Therefore $\on{Len}(c_2)\le \on{Len}(c_1)\le \on{Len}(c)$. Moreover, $c_2=(s\circ c_2,\varphi\circ c_2)$ lies in $\mathbb R\times V(\varphi_0, (\varphi\circ c_2)'(0))\cap S$ which also contains $\varphi_1$, thus $c_2$ lies in
$\mathbb R\times V(\varphi_0,\varphi_1)\cap S$.
If $\varphi_0=\varphi_1$, then $\mathbb R\times \{\varphi_0\}$ is a minimal geodesic. If $\varphi_0=-\varphi_0$ we choose a great circle between them which lies in a 2-space $V$ and proceed as above. When $\varphi_0, \varphi_1 \in C^\infty$, then the 3-dimensional submanifold ${\mathbb R}\times V \cap S$ lies in ${\mathbb R} \times S\cap C^\infty$ and hence so does the minimal geodesic.
\end{proof}
\subsection{Some geodesic completions}\label{completions}
The relation to hypersurfaces of revolution in Sect.~\ref{revolution} suggests that there are functions $C_1$ and $C_2$ such that geodesic incompleteness of the metric $G$ is
due to a `coordinate singularity' at $W_-$ or at $W_+$. Let us write $I=(W_-,W_+)$.
We work in polar coordinates on the infinite-dimensional manifold $I \times (S\cap C^\infty)$
with the metric $\bar G = ds^2 + a(s)\langle d\varphi,d\varphi\rangle$.
\emph{Example.} For $I=(0,\infty)$ the metric $ds^2 + s^2\langle d\varphi,d\varphi\rangle$ describes the flat space
$C^\infty(M,\mathbb R)\setminus \{0\}$ with the $L^2$-metric in polar coordinates. Putting $0$ back in geodesically completes the space.
Moreover, for $\beta\in (0,\pi/4]$ the metric $ds^2 + \sin^2(\beta) s^2 \langle d\varphi,d\varphi\rangle$ describes the cone with radial opening angle $\beta$. Putting in 0 generates a tip; sectional curvature is a delta distribution at the tip of size $2(1-\sin(\beta))\pi$. This is an orbifold with symmetry group $\mathbb Z/k\mathbb Z$ at the tip if
$\sin(\beta)=1/k$.
More generally, $ds^2 + K^2 s^2 \langle d\varphi,d\varphi\rangle$ describes the generalized cone whose `angle defect' at the tip is $2\pi (1-K)$; there is negative curvature at the tip if $K>1$ in which case we cannot describe it as a surface of revolution.
\emph{Example.} For $I= (0,\pi)$, the metric $ds^2 + \sin^2(s)\langle d\varphi,d\varphi\rangle$ describes the infinite-dimensional round sphere `of 1 dimension higher' with equator $S\cap C^\infty$ and with north- and south-pole omitted.
This can be seen from the formula for sectional curvature from Sect.~\ref{curvature} below, or by transforming it to the hypersurface of revolution according to Sect.~\ref{revolution}. Putting back the two poles gives the geodesic completion.
To realize this on the space of densities, we may choose a smooth and positive function $g_2(r)$ freely, and then put
\begin{align*}
g_1(r) &= \sin^2\Big(\int_1^r g_2(\rho)^{1/2}\,d\rho\Big)\,,
\quad
C_1(m) = \frac{g_1(\sqrt{m})}{4m}\,,
\\
C_2(m) &= \frac1{4m} g_2(\sqrt{m}) - \frac1{4m^2} g_1(\sqrt{m})\,.
\end{align*}
Choosing $g_2(r)= 4r^2$ we get $g_1(r)=\sin^2(r^2-1)$ so that $C_1(m)=\frac1{4m}\sin^2(m-1)$
and $C_2(m)= 1- \frac1{4m^2}\sin^2(m-1)$ .
The general situation can be summarized in the following result:
\begin{theorem*}
If $W_->-\infty$ and if $C_1$ and $C_2$ have smooth extensions to $[0,\infty)$ and $C_1(0) > 0$, then the metric $\bar G$ has a smooth 1-point geodesic completion at $r=0$ (or $s=W_-$).
If $W_+<\infty$ and if $C_1$ and $C_2$ have smooth extensions to $(0,\infty]$ in the coordinate $1/m$, then the metric $\bar G$ has a smooth 1-point geodesic completion at $r=\infty$ (in the coordinate $1/r$), or at $s=W_+$.
\end{theorem*}
\begin{proof}
From the formulas in Sect.~\ref{overview} we get
\begin{multline*}
\Phi^*(g_1(r)\langle d\varphi,d\varphi\rangle + g_2(r)dr^2)
= \frac{g_1(\|f\|)}{\|f\|^2}\langle df,df\rangle +\Big(\frac{g_2(\|f\|)}{\|f\|^2} - \frac{g_1(\|f\|)}{\|f\|^4}\Big)\langle f,df\rangle^2
\\
= 4C_1(\|f\|^2)\langle df,df\rangle + 4C_2(\|f\|^2)\langle f,df\rangle^2\,.
\end{multline*}
By a classical theorem of Whitney the even smooth functions $h(r)$ are exactly the smooth functions of $r^2$. So the metric extends smoothly at 0 to $C^{\infty}(M,\mathbb R)$.
The proof for the case $W_+ <\infty$ is similar.
\end{proof}
\subsection{Covariant derivative and curvature}\label{curvature}
In this section we will write $I = (W_-,W_+)$. In order to calculate the covariant derivative we consider the infinite-dimensional manifold $I \times S$
with the metric
$\bar G = ds^2 + a(s)\langle d\varphi,d\varphi\rangle$ and smooth vector fields $f(s,\varphi)\partial_s + X(s,\varphi)$ where
$X(s,\;)\in \mathfrak X(S)$ is a smooth vector field on the Hilbert sphere $S$.
We denote by $\nabla^S$ the covariant derivative on $S$ and get
\begin{align*}
\partial_s &\bar G\big(g\partial_s+ Y,h\partial_s + Z\big) = \partial_s\big(gh + a\langle Y,Z\rangle\big) =
\\&
= g_s h + g h_s + a_s\langle Y,Z\rangle + a\langle Y_s,Z\rangle + a\langle Y,Z_s\rangle
\\&
= \bar G\big( g_s\partial_s + \frac{a_s}{2a}Y + Y_s, h\partial_s + Z\big)
+ \bar G\big(g\partial_s + Y, h_s\partial_s + \frac{a_s}{2a}Z + Z_s\big)
\\
X &\bar G\big(g\partial_s+ Y,h\partial_s + Z\big) = X\big(gh + a\langle Y,Z\rangle\big)
\\&
= dg(X).h + g.dh(X) + a\langle \nabla^S_XY,Z\rangle + a \langle Y,\nabla^S_XZ,\rangle
\\&
= \bar G\big(dg(X)\partial_s + \nabla^S_XY, h\partial_s+Z\big) + \bar G\big(g\partial_s+Y,dh(X)\partial_s + \nabla^S_XZ\big)\,.
\end{align*}
Thus the following covariant derivative on $I \times S$, which is not the Levi-Civita covariant derivative,
\begin{align*}
\bar\nabla_{f\partial_s+X}(g\partial_s+Y) = f.g_s\partial_s + f\frac{a_s}{2a}Y + fY_s + dg(X)\partial_s + \nabla^S_XY\,,
\end{align*}
respects the metric $ds^2 + a\langle d\varphi,d\varphi\rangle$. But it has torsion which is given by
\begin{align*}
\on{Tor}&(f\partial_s+X,g\partial_s+Y)=
\\&
=\bar\nabla_{f\partial_s+X}(g\partial_s+Y) - \bar\nabla_{g\partial_s+Y}(f\partial_s+X) - [f\partial_s + X, g\partial_s + Y] =
\\&
= \frac{a_s}{2a}(fY - gX)\,.
\end{align*}
To remove the torsion we consider the endomorphisms
\begin{gather*}
\on{Tor}_{f\partial_s +X}, \on{Tor}_{f\partial_s +X}^{\top} : T(I \times S)\to T(I \times S)\,,
\\
\on{Tor}_{f\partial_s +X}(g\partial_s+Y) = \on{Tor}(f\partial_s+X,g\partial_s + Y)\,,
\\
\bar G\big(\on{Tor}_{f\partial_s +X}^\top(g\partial_s + Y), h\partial_s + Z) = \bar G\big(g\partial_s + Y,\on{Tor}_{f\partial_s +X}(h\partial_s+ Z)\big)
\end{gather*}
The endomorphism
\begin{align*}
&A_{f\partial_s+ X}(g\partial_s+Y) :=
\\&
= \tfrac12\big(\on{Tor}(f\partial_s+ X,g\partial_s+Y) - \on{Tor}_{f\partial_s+ X}^\top(g\partial_s+Y)
- \on{Tor}_{g\partial_s+ Y}^\top(f\partial_s+X)\big)
\end{align*}
is then $\bar G$-skew, so that
\begin{equation*}
\nabla_{f\partial_s + X}(g\partial_s + Y) = \bar\nabla_{f\partial_s + X}(g\partial_s + Y) - A_{f\partial_s+ X}(g\partial_s+Y)
\end{equation*}
still respects $\bar G$ and is now torsion free. In detail we get
\begin{align*}
\on{Tor}_{f\partial_s+ X}^\top(g\partial_s+Y) &= - \frac{a_s}2\langle X,Y\rangle \partial_s + \frac{a_s}{2a}fY
\\
A_{f\partial_s+ X}(g\partial_s +Y) &= \frac{a_s}{2}\langle X,Y\rangle\partial_s - \frac{a_s}{2a}gX\,,
\end{align*}
so that $\nabla$ is the Levi-Civita connection of $\bar G$:
$$\boxed{\;
\begin{aligned}
\nabla_{f\partial_s + X}(g\partial_s + Y) &= \big(f.g_s+ dg(X)- \frac{a_s}{2}\langle X,Y\rangle\big)\partial_s
\\&\quad
+ \frac{a_s}{2a}(fY + gX)+ fY_s + \nabla^S_XY\,.
\end{aligned}
\;}$$
For the curvature computation we assume from now on that all vector fields of the form $f\partial_s+X$ have $f$ constant and $X=X(\varphi)$ so that in this case
\begin{align*}
\nabla_{f\partial_s + X}(g\partial_s + Y) &= -\frac{a_s}{2}\langle X,Y\rangle \partial_s
+ \frac{a_s}{2a}(fY + gX) + \nabla^S_XY\,,
\\
[f\partial_s + X,g\partial_s + Y] &= [X,Y]^S\,,
\end{align*}
in order to obtain
\begin{align*}
&\nabla_{f\partial_s + X}\nabla_{g\partial_s + Y}(h\partial_s+Z) =
\nabla_{f\partial_s + X}
\big(-\frac{a_s}{2}\langle Y,Z\rangle \partial_s + \frac{a_s}{2a}(gZ + hY) + \nabla^S_YZ\big)
\\&
=\Big(-f\frac{a_{ss}}{2}\langle Y,Z\rangle-\frac{a_s}{2}\langle \nabla^S_XY,Z\rangle
-\frac{a_s}{2}\langle Y,\nabla^S_XZ\rangle
\\&\qquad\qquad
-\frac{a_s^2}{4a}g\langle X, Z\rangle -\frac{a_s^2}{4a}h\langle X,Y\rangle
-\frac{a_s}2\langle X, \nabla^S_YZ\rangle
\Big)\partial_s
\\&\quad
+ \frac{a_s^2}{4a^2}fgZ + \frac{a_s^2}{4a^2}fhY + \frac{a_s}{2a}f\nabla^S_YZ -\frac{a_s^2}{4a}\langle Y,Z\rangle X
\\&\quad
+\big(\frac{a_s}{2a}\big)_sfgZ + \big(\frac{a_s}{2a}\big)_sfhY
+ \frac{a_s}{2a}g\nabla^S_XZ + \frac{a_s}{2a}h\nabla^S_XY + \nabla^S_X\nabla^S_YZ
\\
-&\nabla_{[f\partial_s + X,g\partial_s + Y]}(h\partial_s+Z) = -\nabla_{[X,Y]^S}(h\partial_s+Z)
\\&
= +\frac{a_s}{2}\langle [X,Y]^S,Z\rangle\partial_s - \frac{a_s}{2a}h[X,Y]^S - \nabla^S_{[X,Y]^S}Z
\end{align*}
Summing up we obtain the curvature (for general vector fields, since curvature is of tensorial character)
\begin{align*}
&\mathcal R(f\partial_s + X,g\partial_s + Y)(h\partial_s + Z) =
\\&
= \big(\frac{a_{ss}}{2} - \frac{a_s^2}{4a}\big)\langle gX-fY,Z\rangle\partial_s
+ \mathcal R^S(X,Y)Z
\\&\quad
- \big(\big(\frac{a_s}{2a}\big)_s + \frac{a_s^2}{4a^2}\big)h(gX-fY)
+\frac{a_s^2}{4a}\big(\langle X,Z\rangle Y - \langle Y,Z\rangle X\big)\,.
\end{align*}
and the numerator for sectional curvature
\begin{align*}
&\bar G\big(\mathcal R(f\partial_s + X,g\partial_s + Y)(g\partial_s + Y),f\partial_s +X\big)=
a\langle\mathcal R^S(X,Y)Y,X\rangle
\\&\quad
-\big(\frac{a_{ss}}{2} - \frac{a_s^2}{4a}\big)\big\langle gX-fY ,gX-fY\big\rangle
+\frac{a_s^2}{4}\big(\langle X,Y\rangle^2 - \langle Y,Y\rangle \langle X,X\rangle\big)\,.
\end{align*}
Let us take $X,Y\in T_\varphi S$ with $\langle X,Y\rangle=0$ and
$\langle X,X\rangle=\langle Y,Y\rangle = 1/a(s)$, then
\begin{align*}
\on{Sec}_{(s,\varphi)}(\on{span}(X,Y)) &= \frac1a - \frac{a_s^2}{4a^2}\,, &
\on{Sec}_{(s,\varphi)}(\on{span}(\partial_s,Y)) &= -\frac{a_{ss}}{2a} + \frac{a_s^2}{4a^2}\,,
\end{align*}
are all the possible sectional curvatures.
Compare this with the formulae for the principal curvatures of a hypersurface of revolution in \cite{CollHarrison13} and with the formulas for rotationally symmetric Riemannian metrics in \cite[Sect.~3.2.3]{Petersen2006}.
\subsection{Example}
The simplest case is the choice $C_1(\lambda) = \frac 1\lambda$ and $C_2(\lambda) = 0$. The Riemannian metric is
\[
G_\mu(\alpha, \beta) = \frac{1}{\mu(M)} \int_M \frac \alpha\mu \frac \beta\mu \mu\,.
\]
Then $g_1(r) = 4$ and $g_2(r) = \frac 4{r^2}$.
This metric is geodesically complete on $C^\infty(M,\mathbb R)\setminus \{0\}$.
The geodesic equation \thetag{d} simplifies to
\[
r_{tt} = \frac{r_t^2}{r}\,.
\]
This ODE can be solved explicitely and the solution is given by
\[
r(t) = r(0) \exp \left( \frac{r_t(0)}{r(0)} t \right)\,.
\]
The reparamterization map is $\alpha(t) = t$ and thus the geodesic
\[
\varphi(t) = \cos\left(\| \varphi_t(0)\| t\right) + \sin\left(\| \varphi_t(0)\| t\right)
\frac{\varphi_t(0)}{\| \varphi_t(0)\|}\,,
\]
describes a great circle on the sphere with the standard parametrization. Note that geodesics with $r_t(0) = 0$ are closed with period $2\pi / \| \varphi_t(0)\|$. The spiraling behaviour of the geodesics can be seen in Fig.~\ref{fig:sim}.
\begin{figure}
\centering
\includegraphics[width=.4\textwidth]{figures/simGeodesics30pm.pdf}
\includegraphics[width=.4\textwidth]{figures/simGeodesicsUpwardsPm.pdf}
\caption{Fixing $\varphi(0)$, $\varphi_t(0)$ with $\|\varphi_t(0)\|=1$, the figure shows geodesics $r(t).\varphi(t)$ starting at $r(0)=1$ for various choices of $r_t(0)$; the geodesics are shown in the orthonormal basis $\{\varphi(0), \varphi_t(0)\}$. A periodic geodesic can be seen on the right. The coefficients in the metric are $C_1(\lambda) = \lambda^{-1}$ and $C_2(\lambda) = 0$.}
\label{fig:sim}
\end{figure}
\subsection{Example.} By setting $C_1(\lambda) = 1$ and $C_2(\lambda) = 0$ we obtain the Fisher--Rao metric on the space of all densities. The Riemannian metric is
\[
G_\mu(\alpha, \beta) = \int_M \frac \alpha\mu \frac \beta\mu \mu\,.
\]
In this case $g_1(r) = 4 r^2$ and $g_2(r) = 4$.
The metric is incomplete towards 0 on $C^\infty(M,\mathbb R)\setminus \{0\}$. The pullback metric \thetag{b} is
\[
\tilde G = 4 r^2 \langle d\varphi, d\varphi \rangle + 4 dr^2\,,
\]
and hence geodesics are straight lines in $C^\infty(M,{\mathbb R}) \setminus \{0\}$. In terms of the variables $(r,\varphi)$, the geodesic equation \thetag{d} for $r$ is
\[
r_{tt} = \frac{A_0^2}{16} \frac 1{r^3}\,,
\]
with $A_0 = 4 r(0)^2 \| \varphi_t(0)\|$.
\begin{figure}
\centering
\includegraphics[width=.4\textwidth]{figures/extBalls20.pdf}
\includegraphics[width=.4\textwidth]{figures/infBalls15from01.pdf}
\caption{Fixing $\varphi(0)$, $\varphi_t(0)$ with $\|\varphi_t(0)\|=1$, the figure shows geodesics $r(t).\varphi(t)$ for various choices of $r_t(0)$; on the left the extended Fisher--Rao metric with $C_1=C_2=1$ with geodesics starting from $r(0)=1$; on the right the metric with $C_1=\frac 1{r^2}$ with geodesics starting from $r(0)=0.1$.}
\label{fig:ext_inf}
\end{figure}
\subsection{Example} Setting $C_1(\lambda) = 1$ and $C_2(\lambda) = 1$ we obtain the extended metric
\[
G_\mu(\alpha, \beta) = \int_M \frac \alpha\mu \frac \beta\mu \mu + \int_M \alpha \int_M \beta\,.
\]
In this case $g_1(r) = 4r^2$ and $g_2(r) = 4r^2 + 4$. The geodesic equation \thetag{d} is
\[
r_{tt} = \frac{A_0^2 - 16 r^4 r_t^2}{16 r^3 \left( 1+ r^2 \right)}\,.
\]
The metric on $C^\infty(M,\mathbb R)\setminus \{0\}$ is incomplete towards 0. Geodesics for the metric can be seen in Fig.~\ref{fig:ext_inf}. Note that only the geodesic going straight into the origin seems to be incomplete.
\subsection{Example} Setting $C_1(\lambda) = \frac{1}{\lambda^2}$ and $C_2(\lambda) = 0$ we obtain the metric
\[
G_\mu(\alpha, \beta) = \frac{1}{\mu(M)^2} \int_M \frac \alpha\mu \frac \beta\mu \mu\,,
\]
which is complete towards 0, but incomplete towards infinity on $C^\infty(M,\mathbb R)\setminus \{0\}$.
We have $g_1(r) = 4/r^2$ and $g_2(r) = 4 / r^2$. The geodesic equation \thetag{d} is
\[
r_{tt} = \frac{2 r_t^2 - A_0^2 r^6}{16 r}\,.
\]
Examples of geodesics can be seen in Fig.~\ref{fig:ext_inf}. Note that the geodesic ball extends more towards infinity than towards the origin.
| 7acd8c9dd197da9b3a56c7ac85b40693c29191e9 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Galaxy clusters are the largest virialized structures in the Universe and form
via aggregation of less massive systems \citep[e.g.][]{press74}. During merger
events, the intra-cluster medium (ICM) is heated by shocks and is believed to
become turbulent. Part of the energy involved in these processes is converted
into non-thermal phenomena that exhibit themselves in the radio band as
halo and relic emissions \citep[e.g.][for a review]{brunetti14rev}. Both radio
sources are diffuse cluster-scale sources with steep
spectra\footnote{$S_\nu \propto \nu^{-\alpha}$, with $\alpha$ spectral index.}
($\alpha\gtrsim1$). Radio halos are generally morphologically connected with
the
X-ray emission of the hosting cluster, whereas radio relics are elongated,
polarized and found in cluster peripheries \citep[e.g.][for an observational
overview]{feretti12rev}. In particular, radio relics are believed to form at
the gigantic shocks that are generated in major mergers, where cosmic ray
electrons (CRe) are (re)accelerated \citep[see][for reviews]{bruggen12rev,
brunetti14rev}. This scenario is supported by the arc-shaped morphologies of
relics, their high level of polarization and by the fact that an increasing
number of shocks have been detected at the location of radio relics
\citep[e.g.][]{akamatsu13systematic, bourdin13, shimwell15, eckert16,
botteon16}. The main difficulty in the understanding of the origin of radio
relics resides in the low Mach number ($\mathcal{M}\lesssim 3-4$) associated
with merger shocks. The acceleration efficiency at these weak shocks is indeed
expected to be small and in several cases it is in tension with the
observational requirements \citep[e.g.][]{markevitch05, macario11, kang12,
pinzke13, kang14, vanweeren16toothbrush, botteon16}. \\
\indent
ACT-CL J0102--4915 is the most massive cluster detected in the far Universe, at
a redshift of $z=0.87$ \citep{menanteau12}. For its extraordinary mass of
$M_{500}\sim8.8\times10^{14}$~$\rm{M_\odot}$ \citep{planck14xxix}, it is also
known with the
nickname of `El Gordo'. The cluster was firstly discovered by
its strong Sunyaev-Zel'dovich (SZ) signal \citep{marriage11} and later
confirmed through optical and X-ray observations. The system is in a complex
merger state, as revealed by the double peaked galaxy distribution and the
elongated morphology of its hot ($kT\sim15$ $\rm{keV}$) ICM
\citep{menanteau10, menanteau12}. In the radio band, a tenuous halo and a
double relic system at the cluster NW and SE X-ray boundaries were discovered
\citep{lindner14}. \\
\indent
In this paper we report the discovery of a strong shock associated with a radio
relic in `El Gordo' cluster. In particular, our joint \chandra\ and \gmrt\
(GMRT) analysis provides interesting insights about the origin of the relic.
Throughout the paper, we assume a concordance $\Lambda$CDM cosmology with $H_0
= 70$~$\rm{km\,s^{-1}\,Mpc^{-1}}$, $\Omega_{m}=0.3$ and $\Omega_{\Lambda}=0.7$,
in which $1'' = 7.713$~$\rm{kpc}$ at the cluster redshift ($z=0.87$). Reported
uncertainties are 68\%, unless stated otherwise.
\section{Observations and data reduction}
\begin{figure*}
\centering
\includegraphics[width=.45\textwidth]{fig1_a.pdf}
\includegraphics[width=.45\textwidth]{fig1_b.pdf}\\
\vspace{1cm}
\includegraphics[width=.72\textwidth]{fig1_c_whi.pdf}
\caption{`El Gordo' galaxy cluster. \textit{a}) \chandra\ 0.5-2~$\rm{keV}$
band exposure-corrected image smoothed on a $3''$ scale. \textit{b}) GMRT
610~$\rm{MHz}$ radio emission at a resolution of $11'' \times 4.8''$. The
$1\sigma$ noise level is 50~$\rm{\mu Jy\,b^{-1}}$, contours are drawn at
levels of $3\sigma \times (-1,1,2,4,8,16)$. Circles denote the compact radio
sources identified in \citet{lindner14} as cluster members (red) or not
(blue). \textit{c}) In the \chandra/GMRT comparison the location of the radio
relics at the cluster X-ray boundaries and the spatial connection between the
halo and the northern X-ray tail are evident.}
\label{fig:gordo_cluster}
\end{figure*}
\subsection{X-ray data reduction}
`El Gordo' was observed three times (ObsID: 12258, 14022, 14023) with \chandra\
in ACIS-I configuration and \texttt{VFAINT} mode for a total exposure time of
360~$\rm{ks}$. We carried out the standard data reduction by using
\texttt{CIAO 4.7} and \chandra\ \texttt{CALDB 4.6.9}. In particular,
soft proton flares were inspected analyzing the light curves extracted from the
S3 chip in the 0.5-2~$\rm{keV}$ band for each ObsID and removed using the
\texttt{lc\_clean} routine. We then used the \texttt{merge\_obs} task to make
the final 0.5-2~$\rm{keV}$ cleaned image (340~$\rm{ks}$) shown in
Fig.~\ref{fig:gordo_cluster}a. \\
\indent
We created a single exposure-corrected point spread function (PSF) map
with
minimum size for the merged image by combining the PSF and exposure maps of
each
ObsID. Once the PSF of the instrument is known, the \texttt{wavdetect} task
allows to identify discrete sources in the surface brightness (SB) image of the
cluster. These were detected using wavelet radii of 1, 2, 4, and 8 pixels,
confirmed by eye and excluded in the SB profile analysis. In order to create a
single background image, the \texttt{reproject\_event} task was used to match
the background templates to the corresponding event files for every ObsID. This
single background image was normalized by counts in the band 9.5-12~$\rm{keV}$
and subtracted during the SB analysis. \\
\indent
Dealing with spectral analysis of low SB sources as in the case
of cluster outskirts requires a detailed treatment of the astrophysical and
instrumental background emission. In this respect, we modeled the sky
component due to the
Galactic emission with two thermal plasmas with $kT_1 = 0.14$~$\rm{keV}$ and
$kT_2 = 0.25$~$\rm{keV}$ , the cosmic X-ray background with an absorbed
power-law with photon index
$\Gamma=1.4$ and the ACIS-I particle background by using the analytical
approach prosed by \citet{bartalucci14}. Spectra were extracted in the same
region for every ObsID and simultaneously fitted in the 0.5-11~$\rm{keV}$
energy
band with the package \texttt{XSPEC v12.9.0o}. Since the low X-ray count rate,
we
kept the metal abundance of the \texttt{APEC} model, which accounts for the ICM
thermal emission, fixed at the value of 0.3~$\rm{Z_\odot}$ (solar abundance
table by \citealt{anders89}) and used Cash statistics during the fits.
\subsection{Radio data reduction}
Archival GMRT 610~$\rm{MHz}$ observations of `El Gordo' (project code
$22\_001$, PI: R.~R.~Lindner) taken on 26 August 2012 were analyzed using the
Astronomical Image Processing System (\texttt{AIPS}). The GMRT Software Backend
was used to record the parallel polarization products RR and LL with a
bandwidth of 33.3~$\rm{MHz}$ divided into 256 channels. The source 3C48 was
used for flux and bandpass calibration and the calibrator $0024-420$ was used
for phase calibration towards the target. The total on-target observing time
was 170 minutes. Standard steps of flagging (excision of bad data) and
calibration were carried out. The resulting calibrated visibilities towards the
target were split and used for imaging. A few rounds of phase-only
self-calibration and a round of amplitude and phase self-calibration were
carried out to improve the sensitivity of the image. The final image with
visibilities weighted according to \texttt{ROBUST 0} in the task \texttt{IMAGR}
and resolution $11'' \times 4.8''$ (position angle $4.8$$^\circ$) is presented in
Fig.~\ref{fig:gordo_cluster}b. The image was corrected for the GMRT primary
beam using the task \texttt{PBCOR}. The off-source noise level is
50~$\rm{\mu Jy\,b^{-1}}$ and a 10\% error on the absolute flux calibration
scale was assumed. \\
As a preliminary result of new GMRT data, we also used observations
taken at 327~$\rm{MHz}$ (project code $25\_023$, PI: R.~Kale) to perform
spectral analysis. The complete analysis of the new radio dataset will be
presented in the forthcoming paper (Kale et al., in preparation).
\section{Results}
\subsection{X-ray/radio analysis}\label{ch:x-radio}
`El Gordo' X-ray emission remarkably recalls the famous `Bullet' cluster
\citep{markevitch02bullet}: a dense cool core ($kT\sim 6$~$\rm{keV}$) is
moving in the SE-NW direction producing a prominent cold front
\citep{menanteau12} which is expected to follow a shock wave
\citep[e.g.][]{vikhlinin01a, markevitch02bullet}. The cluster is elongated along
the merger direction and presents a couple of X-ray tails that give to the
system a comet-like morphology (Fig.~\ref{fig:gordo_cluster}a). \\
\indent
Our 610~$\rm{MHz}$ radio image of `El Gordo' recovers extended emission better
than previously done by \citet[Fig.~2 and 15]{lindner14} as we considered
baselines down to 0.2~${\rm k\lambda}$ (instead of 0.48~${\rm k\lambda}$). This
allows to study the morphology of the diffuse sources in more detail. In our
image shown in Fig.~\ref{fig:gordo_cluster}b, the prominent and elongated radio
halo connects a pair of radio relics, located in opposite directions at the NW
and SE edges of the cluster X-ray emission (Fig.~\ref{fig:gordo_cluster}c). The
strongest part of the halo coincides with the disrupted cluster core, whereas a
radio tail appears to remarkably follow the northern tail visible in the
X-rays (Fig.~\ref{fig:gordo_cluster}c). \\
\indent
Our work is focused on the NW radio relic, whose flux densities at 610
and 327~$\rm{MHz}$ are $F_{610} = 27.5\pm2.8$~$\rm{mJy}$ and $F_{327} =
64.6\pm6.6$~$\rm{mJy}$, respectively. These result in a radio spectral index
$\alpha=1.37\pm0.20$. The flux density of the relic at 2.1~$\rm{GHz}$ measured
by \citet{lindner14} with the \atca\ implies $\alpha\sim1.5$ from 2100 to
327~$\rm{MHz}$, which is consistent with what we estimated in the narrower
frequency range. Nevertheless, we will use the spectral index from 610 to
327~$\rm{MHz}$ since it is taken from two high sensitivity images obtained from
GMRT observations with matched inner-\textit{uv} coverage (\textit{uv}$_{min}$
= 0.2~${\rm k\lambda}$).
\begin{figure}
\centering
\includegraphics[width=\hsize]{unsharp2.pdf}
\caption{Unsharp-masked \chandra\ images for `El Gordo' cluster created by
subtracting images convolved with Gaussians with $\sigma_1$ and
$\sigma_2$ and dividing by the sum of the two. From top left panel in clockwise
order $(\sigma_1,\sigma_2) = (3'',5''), (3'',20''), (7'',30''), (5'',20'')$.}
\label{fig:un-mask}
\end{figure}
\begin{figure}
\centering
\includegraphics[width=\hsize]{overlay-region_yellow.pdf}
\caption{Radio/X-ray overlay of `El Gordo'. Red sectors delineate the surface
brightness extracting regions. Spectral analysis was performed in the yellow
sectors. Colors and contours are the same reported in
Fig.~\ref{fig:gordo_cluster}.}
\label{fig:sectors}
\end{figure}
\subsection{Relics and shocks}
Double relics have been observed in different systems
\citep[e.g.][]{degasperin14} and are believed to form in mergers between
two clusters with similar mass where diametrically opposite shocks move
outwards along the merger axis (re)accelerating particles
\citep{vanweeren11double}. \citet{menanteau12} pointed out the possible
presence of a couple of shocks by analyzing a $60\:\rm{ks}$ \chandra\
unsharp-masked image of `El Gordo'. For these reasons we created the
unsharp-masked images shown in Fig.~\ref{fig:un-mask} and searched for sharp
edges in the X-ray SB image, identifying at least one
discontinuity in the cluster. We used \texttt{PROFFIT v1.3.1} \citep{eckert11}
to extract the SB profiles in the red sectors shown in Fig.~\ref{fig:sectors},
where the NW relic stands out. An underlying broken power-law density profile
is
usually adopted to describe the SB in presence of a discontinuity. In the case
of spherical symmetry, the downstream density $\rho_d$ is higher by a factor of
$\mathcal{C} \equiv \rho_d/\rho_u$, with $\rho_u$ upstream density, at the
shock putative distance $r_{sh}$. In formula
\begin{equation}\label{eq:break-pl}
\begin{array}{ll}
\rho_d (r) = \mathcal{C} \rho_0 \left( \frac{r}{r_{sh}} \right)^{a_1}, &
\mbox{if} \quad r \leq r_{sh} \\
\\
\rho_u (r) = \rho_0 \left( \frac{r}{r_{sh}} \right)^{a_2}, & \mbox{if} \quad r
> r_{sh}
\end{array}
\end{equation}
\noindent
where $\rho_0$ is the density normalization, $a_1$ and $a_2$ are the power-law
indices and $r$ is the radius from the center of the sector (in
Fig.~\ref{fig:sectors} RA: $+15^\circ.7275$, DEC: $-49^\circ.2724$, J2000). We
used this density shape to fit the X-ray SB keeping all parameters of the model
free to vary. \\
We firstly report results concerning sector 1+2 (opening angle (OA):
$30^\circ-98^\circ$) because it covers the whole extension of the feature
shown in Fig.~\ref{fig:un-mask} and it gives the maximum SB drop with the best
statistics (a discussion on the sector choice is presented in
Section~\ref{ch:syste}). In Fig.~\ref{fig:sb_profile} we report
the best broken power-law model fit, which is in excellent agreement with data.
We detect a large SB drop, corresponding to a density compression factor
$\mathcal{C}=3.4^{+0.4}_{-0.3}$, co-spatially located with the relic. For a
shock, the Rankine-Hugoniot (RH) jump conditions for a monatomic gas
\begin{equation}\label{eq:mach-from-dens}
\mathcal{C} \equiv \frac{\rho_d}{\rho_u} =
\frac{4\mathcal{M}_{\rm{SB}}^2}{\mathcal{M}_{\rm{SB}}^2 + 3}
\end{equation}
\noindent
would lead to a Mach number $\mathcal{M}_{\rm{SB}}=4.1^{+3.4}_{-0.9}$. A
$\mathcal{M}>3$ shock is quite a rarity in galaxy clusters and so far only two
of them have been detected ($\mathcal{M}=3.0\pm0.4$ in the `Bullet' cluster,
\citealt{markevitch06}; $\mathcal{M}=3.0\pm0.6$ in A665,
\citealt{dasadia16a665}).
\begin{figure}
\centering
\includegraphics[width=\hsize]{shock_NW_radio.pdf}
\caption{X-ray SB profile in the 0.5-2~$\rm{keV}$ band extracted in sector
1+2 (Fig.~\ref{fig:sectors}). The data were rebinned to reach
a minimum signal-to-noise ratio of 7. The green dashed line shows the NW
radio relic brightness profile (in arbitrary units).}
\label{fig:sb_profile}
\end{figure}
Shocks heat the downstream gas and this allows to distinguish them from cold
fronts, other kinds of SB discontinuities found in galaxy clusters with
inverted temperature jumps \citep[e.g.][]{markevitch07rev}. For
this reason we performed spectral analysis in the yellow sectors shown in
Fig.~\ref{fig:gordo_cluster}c. Spectra for sector 1+2 are reported in
Fig.~\ref{fig:spectra}. We found evidence of a very high downstream
temperature, $kT_d=17.9^{+3.3}_{-2.8}$~$\rm{keV}$, while only a lower limit to
the upstream one was obtained, $kT_u>6.1$~$\rm{keV}$. In principle, this is not
enough to confirm the shock nature of the discontinuity but, similarly to the E
shock in the `Bullet' cluster \citep{shimwell15}, the presence of a cold front
is very unlikely because it would imply a too high temperature ($kT_u
> 20$~$\rm{keV}$) at such a large cluster distance. \\
\indent
Although current data do not allow to measure a temperature jump at the
position of the shock, we can use the lower limit to such a jump to provide
independent constraints on the shock Mach number. According to RH conditions,
the upstream and downstream temperature are related by
\begin{equation}\label{eq:mach-from-temp}
\frac{T_d}{T_u} = \frac{5\mathcal{M}_{\rm kT}^4 + 14\mathcal{M}_{\rm kT}^2
-3}{16\mathcal{M}_{\rm kT}^2}
\end{equation}
\noindent
which implies $\mathcal{M}_{\rm{kT}}<2.9$ if we insert the upper $1\sigma$
limit of $T_d$ and the lower limit on $T_u$. We anticipate that this value is
consistent with the Mach number inferred from SB jump once systematic errors
are taken into account (see the following Section) \\
\indent
A visual inspection of Fig.~\ref{fig:gordo_cluster}a suggests the presence of a
drop in SB also at the position of the SE relic. A shock in this region is
expected due to the presence of the radio relic and in analogy with the
`Bullet' cluster \citep{markevitch02bullet}. However, current data do not allow
us to characterize statistically the SB drop because of the low X-ray counts in
this region. Nevertheless we found evidence for high temperature in the
putative downstream gas, $kT_d=30.1^{+10.5}_{-6.2}$~$\rm{keV}$ in the yellow
sector in the SE (Fig.~\ref{fig:sectors}), somewhat supporting this
possibility. Since typical temperatures in cluster outskirts are of a
few keV, such an high $kT_d$ would likely imply a $\mathcal{M}\gtrsim3.5-4$
shock.
\begin{figure}
\centering
\includegraphics[width=.73\hsize,angle=-90]{spectrum_down.pdf} \\
\includegraphics[width=.73\hsize,angle=-90]{spectrum_up.pdf}
\caption{Downstream (top) and upstream (bottom) spectra of sector 1+2. Data
points are shown in black together with the best fitting model. Different
colors highlight the model components: the cluster emission (in solid red), the
particle background (in dashed blue) and the sky background (in dotted green).
The c-stat/d.o.f. of the fits are 203/168 and 128/115 for the downstream and
upstream spectrum, respectively. Although spectra were simultaneously fitted,
only one ObsID was reported in order to avoid confusion in the plot.}
\label{fig:spectra}
\end{figure}
\subsubsection{Systematic errors on X-ray analysis}\label{ch:syste}
\begin{table*}
\centering
\caption{Results of the SB and spectral fits of the regions shown in
Fig.~\ref{fig:sectors}. Fits in the $2.0'-4.9'$ radial range were made
keeping $r_{sh}$ frozen at the best fit value achieved in the wider range.}
\label{tab:sectors-sb}
\begin{tabular}{lcccccccc}
\hline
Sector & Radial range & $r_{sh}\:(')$ & $\mathcal{C}$ &
$\mathcal{M}_{\rm{SB}}$ & $\chi^2 /\rm{d.o.f.}$ & $kT_d$~$\rm{(keV)}$ &
$kT_u$~$\rm{(keV)}$ & $\mathcal{M}_{\rm{kT}}$ \\
\hline
1 & \begin{tabular}{@{}c@{}}$1.2'-4.9'$ \\
$2.0'-4.9'$\end{tabular} &
$2.359^{+0.006}_{-0.004}$ &
\begin{tabular}{@{}c@{}}$3.5^{+0.7}_{-0.5}$ \\
$3.1^{+0.7}_{-0.5}$ \end{tabular} &
\begin{tabular}{@{}c@{}}$>3.0$ \\
$3.2^{+4.3}_{-0.8}$ \end{tabular} &
\begin{tabular}{@{}c@{}}20.4/32 \\
0.9/6 \end{tabular} &
$18.3^{+4.1}_{-3.2}$ & $6.8^{+10.8}_{-2.8}$ & $1.9-3.4$ \\
2 & \begin{tabular}{@{}c@{}}$1.2'-4.9'$ \\
$2.0'-4.9'$\end{tabular} &
$2.321^{+0.065}_{-0.041}$ &
\begin{tabular}{@{}c@{}}$3.7^{+1.2}_{-0.7}$ \\
$4.2^{+1.6}_{-0.9}$ \end{tabular} &
\begin{tabular}{@{}c@{}}$>3.0$ \\
$>3.8$ \end{tabular} &
\begin{tabular}{@{}c@{}}35.4/24 \\
7.3/2 \end{tabular} &
$15.8^{+7.9}_{-3.9}$ & $>6.1$ & $<3.1$ \\
1+2 & \begin{tabular}{@{}c@{}}$1.2'-4.9'$ \\
$2.0'-4.9'$\end{tabular} &
$2.338^{+0.007}_{-0.005}$ &
\begin{tabular}{@{}c@{}}$3.4^{+0.4}_{-0.3}$ \\
$3.4^{+0.5}_{-0.4}$ \end{tabular} &
\begin{tabular}{@{}c@{}}$4.1^{+3.4}_{-0.9}$ \\
$4.1^{+6.7}_{-1.1}$ \end{tabular} &
\begin{tabular}{@{}c@{}}18.9/34 \\
5.0/8 \end{tabular} &
$17.9^{+3.3}_{-2.8}$ & $>6.1$ & $<2.9$ \\
\hline
\end{tabular}
\end{table*}
Results in the previous Section are based on measurements obtained for a
particular sector (1+2). This entirely covers the feature found in the
unshap-mask images (Fig.~\ref{fig:un-mask}) and allows the best
characterization of the SB jump due to the statistics of the fit.\\
\indent
We checked the impact due to the choice of the SB extracting region in the
determination of the NW X-ray discontinuity and the resulting Mach number.
Firstly, we re-performed SB and spectral analysis by splitting the red and
yellow sectors of Fig.~\ref{fig:sectors} in two sub-regions; the dashed
line distinguishes between sector 1 (OA: $60^\circ.5-98^\circ$), which
is oriented in the N direction, to sector 2 (OA:
$30^\circ-60^\circ.5$), which is in the NW direction and better overlaps the
relic. In both
regions, the SB profile is well described by a compression factor $\mathcal{C}
\gtrsim 3$, implying $\mathcal{M}_{\rm{SB}} \gtrsim 3$. \\
We then repeated the SB analysis by excluding data at $r<2'$ and keeping the
discontinuity distance frozen at the values found in the $1.2'-4.9'$ radial
range. Although with a large error, spectral analysis allowed to constrain the
upstream temperature in sector 1, implying a 68\% confidence interval estimate
for the Mach number $\mathcal{M}_{\rm{kT}}=1.9-3.4$ (taking into account the
asymmetric errors on the two temperatures), whereas only lower limits to $kT_u$
can be obtained in sectors 1+2 and 2. The results of the fits obtained for the
three regions are summarized in Tab.~\ref{tab:sectors-sb}. \\
Finally, we checked the variation on $\mathcal{M}_{\rm{SB}}$ in sector
1+2 due to different shock curvature radii from the best fit value found in
Fig.~\ref{fig:sb_profile} ($r_{curv} \sim 1$~$\rm{Mpc}$). Results are reported
in Tab.~\ref{tab:curvature} and the impact of $r_{curv}$ on the shock
compression factor is presented in Fig.~\ref{fig:r_curv}.\\
\indent
Spectral analysis requires a careful determination of the background sources
and its systematic uncertainties. In this respect, we varied background
normalization levels within $\pm1\sigma$ and re-performed spectral fits. We
achieved results consistent with the reported cluster parameters within
$1\sigma$. Nonetheless, we highlight that the measurement of high temperatures
is critical with \chandra\ given its low effective area at energies higher than
$5$~$\rm{keV}$, in particular the estimated confidence range may not reflect
entirely the true statistical and systematic error range. \\
\indent
As a final test, a more complex model of a two-temperature
thermal plasma was adopted to fit the downstream spectra. In this case, the
high-$T$ component is not constrained while the low-$T$ component gives
unreasonably low temperatures (e.g.\ $kT_{high}>21.2$~$\rm{keV}$ and
$kT_{low}=2.2^{+4.0}_{-1.2}$~$\rm{keV}$, for sector 1+2). As pointed out in the
case of the Coma cluster \citep{gastaldello15}, the low-$T$ component mitigates
the fit residuals at low energy rather than describing a physical condition.
The high-$T$ spectral component instead supports the presence of a high
temperature plasma in the downstream region.
\begin{table}
\centering
\caption{Impact on the SB profile fits in sector 1+2 due to different
shock curvature radii. Note that $r_{curv} \sim 1$~$\rm{Mpc}$ in
Fig.~\ref{fig:sb_profile}.}
\label{tab:curvature}
\begin{tabular}{lccc}
\hline
$r_{curv}$~$\rm{(Mpc)}$ & $\mathcal{C}$ &
$\mathcal{M}_{\rm{SB}}$ & $\chi^2 /\rm{d.o.f.}$ \\
\hline
0.6 & $2.5\pm0.2$ & $2.2^{+0.3}_{-0.2}$ & 96.1/48 \\
0.8 & $3.1\pm0.3$ & $3.2^{+0.9}_{-0.6}$ & 91.5/52 \\
1.2 & $3.4^{+0.4}_{-0.3}$ & $4.1^{+3.4}_{-0.9}$ & 41.0/45 \\
1.4 & $3.2^{+0.4}_{-0.3}$ & $3.5^{+1.7}_{-0.6}$ & 24.7/28 \\
1.6 & $3.1\pm0.3$ & $3.2^{+0.9}_{-0.6}$ & 37.5/27 \\
1.8 & $2.9^{+0.3}_{-0.2}$ & $2.8^{+0.6}_{-0.3}$ & 43.1/25 \\
\hline
\end{tabular}
\end{table}
\begin{figure}
\centering
\includegraphics[width=\hsize]{curv-rad.pdf} \\
\includegraphics[width=\hsize,trim={0.1cm 11cm 0.5cm
0},clip]{curvature.pdf}
\caption{In the top panel we show the difference between the best fit
curvature radius of $\sim 1$~$\rm{Mpc}$ (dashed line) and the two extreme cases
with $r_{curv}=600$ and 1800~$\rm{kpc}$ (lower and upper solid lines,
respectively). In the bottom panel we compare compression factors achieved for
different values of $r_{curv}$ (see Tab.~\ref{tab:curvature}).}
\label{fig:r_curv}
\end{figure}
\subsection{Constraints on the downstream magnetic field}\label{ch:ic}
Relativistic electrons scattering with the cosmic microwave
background (CMB) photons is expected to produce inverse Compton (IC)
emission. From the ratio between radio and X-ray emission it is possible to
constraints the magnetic field in the source region
\citep[e.g.][]{blumenthal70rev}. For this reason we preformed spectral analysis
in a region enclosing the NW relic and introduced, in addition to the canonical
thermal model for the ICM, a power-law in the spectral fit. \\
\indent
We assume that the IC spectrum is a power-law with photon index related to the
synchrotron spectral index via $\Gamma=\alpha+1$. Initially we set
$\Gamma=2.37$ (see Section~\ref{ch:x-radio}) and kept it frozen in the fit
while thermal parameters were free to vary. In this case we obtain
0.5-2~$\rm{keV}$ upper limit to the non-thermal component
$F_{[0.5-2\:\rm{keV}]}
< 6.76 \times 10^{-15}$~$\rm{erg\,s^{-1}\,cm^{-2}}$. \\
\indent
The IC measurement is a very complicated issue and can be influenced by several
factors. We investigated the impact on IC flux estimation by: using the IC
power-law slope in the range $2.17-2.57$ (consistently with the values
reported in Section~\ref{ch:x-radio}), keeping the temperature frozen at 13.5
and 17~$\rm{keV}$ (which covers a range of $kT$ obtained for
different sector choices in the relic region), varying background
normalization levels within $\pm1\sigma$
and re-performing the fits in the 0.7-11 and 0.9-11~$\rm{keV}$ energy bands.
In summary, we found upper limits in the range $(2.95-8.51) \times
10^{-15}$~$\rm{erg\,s^{-1}\,cm^{-2}}$ for the IC flux, representing
$\lesssim50$\%
of the thermal model flux in the same energy band (0.5-2~$\rm{keV}$). However,
in the case of temperature frozen at 17~$\rm{keV}$, the fits formally result in
a IC detection both for $\Gamma=2.17$ and 2.57; we do not consider these
detections solid enough due to the systematics related to the
presence of multi-temperature components and background characterization. We
further explore the
possibility of IC detection in Section~\ref{ch:ic-bump}. \\
\indent
By using our upper limits to IC flux, we conservatively
obtained\footnote{Calculations were obtained using Eq.~32 in
\citet{brunetti14rev}.} the following lower limits to the downstream magnetic
field strength: $B\geq3.1$~$\rm{\mu G}$ for $\Gamma= 2.17$, $B\geq4.9$~$\rm{\mu
G}$ for $\Gamma= 2.37$ and $B\geq7.6$~$\rm{\mu G}$ for $\Gamma= 2.57$. These
values are in line with other estimates for radio relics
\citep[e.g.][]{bonafede09double, finoguenov10, vanweeren10,
vanweeren11zwcl0008}.
\subsection{Acceleration efficiency}
\begin{figure}
\centering
\includegraphics[width=\hsize,trim={3cm 0cm 3cm
0},clip]{etaetaGORDO.jpg}
\caption{Electron acceleration efficiency versus magnetic field downstream in
the NW shock in `El Gordo'. Black lines represent efficiencies evaluated for a
Mach number with $\mathcal{M}=2.5$ (top) and 3.5 (bottom). Calculations were
obtained with $p_{min}=0.1 m_e c$ in Eq.~\ref{eq:n_inj}. Vertical lines denote
the lower limits on the downstream magnetic field strength achieved from the
lack of IC emission from the relic.}
\label{fig:eff}
\end{figure}
The relic--shock connection is nowadays supported by many observational
studies. Nevertheless, theoretical models of relic formation are challenged by
the low Mach numbers associated with cluster shocks. In fact, the commonly
adopted diffusive shock acceleration (DSA) model is severely challenged for
weak shocks if CRe are accelerated by the thermal pool \citep[e.g.][for
review]{brunetti14rev}. A connected problem is the ratio of cosmic ray protons
and electrons that would be generated at these shocks and that current
$\gamma$-ray \fermi\ limits constrain at values that are significantly smaller
than that in supernova remnants \citep[e.g.][]{vazza15efficiency, vazza16}. \\
If the downstream synchrotron luminosity emitted at frequency $\geq \nu_0$
originates from electrons in steady state conditions, the bolometric ($\geq
\nu_0$) synchrotron luminosity that is generated via shock acceleration from a
shock with speed $V_{sh}$ and surface $S$ can be estimated as
\begin{equation}\label{eq:luminosity}
\int_{\nu_0} L(\nu)\,d\nu \simeq {1 \over 2 } \eta_e \Psi \rho_u
V_{sh}^3 \left( 1 - {1 \over{\mathcal{C}^2}} \right) {{ B^2 }\over{B_{cmb}^2 +
B^2}} S
\end{equation}
\noindent
where $\eta_e$ is the efficiency of electron acceleration,
\begin{equation}\label{eq:psi}
\Psi = {{\int_{p_{min}} N_{inj}(p) E\,dp}\over{\int_{p_0} N_{inj}(p) E\,dp}}
\end{equation}
\noindent
accounts for the ratio of the energy flux injected in ``all'' electrons and
those visible in the radio band ($\nu \geq \nu_0$), $p_0$ is the momentum of
the relativistic electrons emitting the synchrotron frequency $\nu_0$ in a
magnetic field $B$ and $B_{cmb} = 3.25(1 + z)^2$~$\rm{\mu G}$ accounts for IC
scattering of CMB photons. The injection spectrum of accelerated CRe is given
by
\begin{equation}\label{eq:n_inj}
N_{inj}(p) = (\delta_{inj} +2 ) p^{-\delta_{inj}} \int_{p_{min}}^{p}
x^{\delta_{inj}-1} N_u(x)\,dx
\end{equation}
\noindent
where $N_u$ is the spectrum of seed particles upstream. In the case of CRe
acceleration from the thermal pool this is $N_u \propto p^{-\delta_{inj}}$,
where $\delta_{inj} = 2 (\mathcal{M}^2+1)/(\mathcal{M}^2-1)$, implying a
synchrotron spectral index $\alpha=\delta_{inj}/2$ in the case of steady state
conditions \citep[e.g.][]{blandford87rev}. \\
\indent
In Fig.~\ref{fig:eff} we report the acceleration efficiency that is necessary
to explain the radio luminosity observed in the NW relic assuming DSA of
thermal electrons upstream. \\
We assumed shock Mach numbers $\mathcal{M}=3.5$ and 2.5, in line with the
values derived from the X-ray analysis. These Mach numbers would imply a
synchrotron spectrum of the relic $\alpha = 1.18$ and $1.38$, respectively,
that are in line with radio measurements (Section \ref{ch:x-radio}).
Calculations were obtained assuming upstream number density and temperature
$2.4 \times 10^{-4}$~$\rm{cm^{-3}}$ and 6.1~$\rm{keV}$, respectively, and a
surface of the relic $S=\pi \times 350^2$~$\rm{kpc^2}$. \\
In Fig.~\ref{fig:eff} we also show the lower limits to the magnetic field in
the relic that are derived from the upper limits to the IC flux assuming the
two values of the spectral index (Section~\ref{ch:ic}). \\
\indent
Despite we are dealing with a high-velocity shock, $V_{sh} \sim
4000$~$\rm{km\,s^{-1}}$, we note that the efficiency of CRe acceleration that
is requested to explain the radio relic is large. This is due to the fact that
the NW relic in the `El Gordo' is one of the most luminous radio relics known
so
far and because, for few $\rm{\mu G}$ magnetic fields, most of the CRe energy
is radiated via IC emission (due to the high redshift of the cluster). Still,
contrary to the case of weaker shocks (see e.g.\ A754, \citealt{macario11}; 1RXS
J0603.3+4214, \citealt{vanweeren16toothbrush}; A115, \citealt{botteon16}), we
conclude that in this case DSA of thermal electrons is still a viable option.
Indeed for Mach number $\geq 3.5$ the electron acceleration efficiency appears
energetically viable $\eta_e \leq 0.01$, whereas for $\mathcal{M}\sim 3-3.5$
additional mechanisms of pre-acceleration of thermal electrons downstream
(see \citealt{guo14a,guo14b}) may be required. \\
\indent
The other possibility is that the NW relic is due to shock re-acceleration of
seeds (relativistic or supra-thermal) electrons. In this case the efficiency
necessary to explain the radio emission is much smaller simply because the bulk
of the energy is channelled directly into highly relativistic particles
(Eq.~\ref{eq:n_inj}, e.g.\ \citealt{markevitch05, kang12}). Seeds can be broadly
distributed in the cluster outskirts where the life time of 100~$\rm{MeV}$
electrons is very long (e.g.\ \citealt{pinzke13, donnert16arx})
or they can be in radio ghost/clouds generated by past AGN activity
\citep[e.g.][]{kang16reacc}. The two possibilities have different predictions
on the upstream synchrotron emission that in principle can be tested with very
deep radio observations, but that are well beyond the aim of our paper.
\section{Discussion}
\subsection{Overall considerations}
`El Gordo' is a high redshift ($z=0.87$, \citealt{menanteau12}) and high mass
($M_{500}\sim8.8\times10^{14}$~$\rm{M_\odot}$, \citealt{planck14xxix}) galaxy
cluster. It is the most
distant massive cluster with the brightest X-ray and SZ emission and the
farthest hosting diffuse radio sources (halo and relics). Our study makes it
is also the most distant cluster where a shock (with one of the highest
Mach number) has been detected. \\
\indent
Optical and X-ray observations revealed that `El Gordo' is in a merging state
\citep{menanteau10,menanteau12}. Recent numerical simulations were able to
reproduce the overall system properties assuming a nearly head on major merger
\citep{donnert14, molnar15, zhang15, ng15}. \\
\indent
Double relic systems are expected to trace shocks moving outwards in cluster
outskirts. So far, studies on `El Gordo' were mainly focused on the SE relic
located in front of the dense cool core, which is expected to follow a shock in
analogy with the well know `Bullet' cluster case \citep{markevitch02bullet}.
However, current X-ray data do not allow to characterize the jump in this
region
because of the low statistics. We instead discovered a $\mathcal{M}\gtrsim3$
shock spatially coincident with the NW relic. Our detection is based on the
\chandra\ SB jump. Although with large uncertainties, spectral analysis
is also consistent with the presence of a strong shock in the ICM. Further
indications of the shock are given by the unsharp-masked images of
Fig.~\ref{fig:un-mask}. We also mention the striking similarity between `El
Gordo' shock/reverse-shock and X-ray morphology with A2146 \citep{russell10,
russell12} even though the latter is a less massive system
($M_{500}\sim3.8\times10^{14}$~$\rm{M_\odot}$, \citealt{planck14xxix}) and does
not host any radio relics at the sensitivity level of current observations
\citep{russell11}.
\subsection{Hints of inverse Compton emission?}\label{ch:ic-bump}
\begin{figure}
\centering
\includegraphics[width=\hsize]{NW_ic_cfr_bump.pdf}
\caption{X-ray SB profile in the 0.5-2~$\rm{keV}$ band extracted in a
region enclosing the NW relic. The broken power-law fit is reported in dashed
red. The SB excess just before the discontinuity, i.e.\ at the relic location,
for which we speculate a IC origin was modeled by adding a Gaussian component
to the broken power-law model (solid blue line). Residuals of the
latter model are displayed in the bottom panel. The data were rebinned to reach
a minimum signal-to-noise ratio of 3.}
\label{fig:ic-bump}
\end{figure}
The search for IC emission from galaxy clusters has been undertaken for many
years with many instruments. However, no confirmed detection has been obtained
so far. The most famous dispute in this field regards the case of the nearby
Coma cluster \citep{rephaeli99, rephaeli02, fusco99, fusco04, fusco07,
rossetti04, wik09, wik11, gastaldello15}. The excellent spatial resolution and
good spectral capabilities of the \chandra\ satellite allow to minimize the
contamination from the thermal X-ray emission and open to the possibility to
search for non-thermal emission also in the soft X-ray band
\citep[e.g.][]{million09}. \\
\indent
`El Gordo' is a perfect target to search for IC emission. It hosts a bright
radio relic in an external region with low thermal SB and it is at high
redshift ($z=0.87$), where the equivalent CMB magnetic field strength is large,
$B_{cmb}=3.25(1+z)^2=11.4$~$\rm{\mu G}$. \\
\indent
We used the deep \chandra\ observations to look for IC signatures in the NW
relic (Section~\ref{ch:ic}). In the case the temperature is frozen at
17~$\rm{keV}$, the spectral analysis provides a formal detection and a
significant fraction (up to $40-50$\%) of the X-ray brightness across a relic
would be contributed by IC emission from the relic itself. This should be
visible in the X-ray images and profiles across the relic. For this reason we
extracted a SB profile in the 0.5-2.0~$\rm{keV}$ across a narrow sector (OA:
$37^\circ.2-67^\circ.5$) containing the NW radio relic. The fit of a
broken power-law model (Eq.~\ref{eq:break-pl}) in this restricted region
provides a good description of the SB jump, as shown in Fig.~\ref{fig:ic-bump}
(dashed red line), leading to $\chi^2 /\rm{d.o.f.} = 48.5/41$. However, a SB
excess is present in the region of the relic. For this reason we also
attempt to fit the SB by adding a Gaussian component to the downstream
power-law. The addition of this Gaussian improves the fit
(Fig.~\ref{fig:ic-bump}, solid blue line) with $\chi^2 /\rm{d.o.f.} = 40.3/38$.
Slight different sector centers and apertures do not influence this excess. The
Gaussian component coincident with the relic could represent an excess due to
IC emission from electrons in the relic. The excess flux associated with the
Gaussian component is in line with that expected from the spectral analysis.
The
combination of this excess with the formal detection of IC emission obtained
from the spectral analysis (with $kT$ frozen) is tantalizing, however deeper
observations (i.e.\ $>500$~$\rm{ks}$) are required to firmly conclude about this
possible detection.
\section{Conclusions}
We presented an X-ray/radio study of the famous `El Gordo' cluster located at
$z=0.87$ focusing on the non-thermal activity in the cluster. \\
\indent
Our GMRT radio observations at 610 and 327~$\rm{MHz}$ confirmed the presence of
a halo and a system of double relics. These represent the most distant diffuse
radio sources detected in a galaxy cluster so far. The halo is quite elongated
in the NW-SE, i.e.\ in the merger direction, and remarkably follows the ICM
emission of the northern X-ray tail. The two relics are found at the boundaries
of the X-ray emission. We focused on the NW relic which has a synchrotron
spectral index $\alpha=1.37\pm0.20$ between 610 and 327~$\rm{MHz}$. \\
\indent
The deep \chandra\ observations (340~$\rm{ks}$) allowed us to discover a shock
at the position of the NW relic. The SB profile
taken is this region abruptly drops at the relic location. The density
compression factor $\mathcal{C} \gtrsim 3$ and the high downstream temperature
provide the indication of a strong shock ($\mathcal{M} \gtrsim 3$) in the ICM.
This is one of the three strongest shocks detected in galaxy clusters and the
most distant ($z=0.87$) observed so far. \\
\indent
The detection of a shock co-spatially located with a relic strongly supports
the relic--shock connection. The NW shock in `El Gordo' cluster allows to study
particle acceleration in a rare regime of strong shock. We found that DSA of
thermal electrons is consistent with measured synchrotron spectrum.
Nonetheless, only shocks with $\mathcal{M} > 3.5$ appear energetically viable
while for weaker shocks re-acceleration models would be preferred. \\
\indent
The presence of relativistic particles emitting a bright synchrotron relic at
$z=0.87$ makes `El Gordo' a suitable cluster candidate to search for IC
emission from the relic. From the X-ray spectral analysis we obtained possible
hints for IC emission from the relic, however we could not firmly
conclude the presence of IC excess and conservatively we derived only lower
limits to the downstream magnetic field that have been used to improve
constraints on particle acceleration. However, we also found hints of
an excess in the 0.5-2~$\rm{keV}$ SB profile across the relic region.
The combination of a possible IC excess in the spectral analysis with the hints
of excess in the SB is tantalizing and certainly deserves deeper \chandra\
observations.
\section*{Acknowledgments}
We thank the anonymous referee for the useful comments on the manuscript. We
thank Dominique Eckert for his help with the \texttt{PROFFIT} SB
analysis. The scientific results reported in this article are based on
observations made by the Chandra X-ray Observatory. We thank the staff of the
GMRT who have made these observations possible. GMRT is run by the National
Centre for Radio Astrophysics of the Tata Institute of Fundamental Research. AB
and GB acknowledge partial support from PRIN-INAF 2014. RK acknowledges
support through the DST-INSPIRE Faculty Award.
\bibliographystyle{mnras}
| b289585d48e6ef0cde2f00d3ddfccdd2031d93c0 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Adaptation is often viewed as a cumulative succession of fixations of beneficial mutations, and the waiting time between two successive beneficial mutations is viewed as one of the main factor limiting adaptation rate \citep{fisher1930, crowkimura1965}. This is however not true in general since several beneficial mutations can compete in a single population, which generates different interference phenomena, such as the Muller's ratchet \citep{muller32, muller64} or more generally Hill-Robertson interfences \citep{hillrobertson66}. Competition between several beneficial mutations is especially expected in large asexual populations with high mutation rates and adaptation is then expected to be slowed down, a phenomenon called ``clonal interference'' \citep{gerrishlenski98}. Clonal interference and its consequences on adaptation rates seems general as they have been observed in bacteria, viruses, yeasts or cancer tumors \citep[e.g.][]{mirallesetal99, devisserrozen06, greavesmaley12, langetal13}. \\
Despite the ubiquity and importance of clonal interference, experiments showed that adaptation is not blocked and keep on even after thousands generations \citep[e.g.][]{wiseretal13}. These observations led more recently to the development of an alternative view of clonal interference, considering that beneficial mutations can occur in a single lineage \citep{desaifisher07}, thus increasing fitness gradually. In large populations with high mutation rates, it is expected that adaptation follows a wave \citep{desaifisher07, goodetal12} bounded by the most and the least fit lineages. Adaptation rate is expected to be constant in this model and depends only on the width of the wave, in other words on the genetic variability present in the population. \\
Models of clonal interference have been useful in understanding the dynamics of adaptation in asexual populations, and have known many empirical successes. However, many experiments have shown that the dynamics of beneficial mutations can be much more complex. Different experiments showed that dynamics can be non-linear: lineages can show multiple frequency peaks during the course of adaptation \citep{leemarx13} and different lineages can coexist for a long time in a single population \citep{langetal11, maddamsettietal15}. This suggests that frequency-dependent fitness can occur. Moreover, cooperation between lineages and niche construction have been observed multiple times in tumoral cancers and bacteria \citep{yangetal14, kinnersleyetal14}, which suggests that taking into account different ecological interactions might be important for a global understanding of the evolution of asexual populations with large population size and high mutation rates. \\
Clonal interference has been mostly theoretically studied using population genetics models with strong assumptions: beneficial mutations are assumed to have simple epistatic interactions, the effect of mutations on fitness is assumed to be transitive, selection is assumed not to depend on the mutants frequency, and the environment is assumed constant. Non-transitive fitness and non-linear dynamics are consequently considered as special cases, of less important interest. There is however a large literature dealing with Lotka-Volterra models with more than two species, especially with non-transitive competitive interactions. These studies try to define the conditions for the coexistence of multiple competitive species in a community \citep[e.g.][]{huismanetal01}. Such studies are generally deterministic and investigate the conditions of stability of dynamical systems. When dealing with stochastic dynamics, simulations are generally performed in spatialized context \citep{laird14} again in order to determine the conditions for the coexistence of multiple competing species. Those studies do not however investigate the probability and time of fixation of mutations and are thus of limited interest regarding the effect of competitive interactions between several clones in the course of adaptation, especially on the adaptation rate. \\
Here we propose a stochastic model with three different lineages under competition, where the competitive interactions are not necessarily transitive, thus relaxing one important assumption of the models studying clonal interference so far. Our model embraces a large variety of phenomena observed in the course of adaptation of asexual species. We recover classical results from clonal interference models and we also show that unexpected behaviours are expected. For instance, we show that in some cases competitive interactions between three clones can lead to a higher rate of adaptation. Our results generally show that non-linear dynamics are likely in large clonal populations, which challenges the interpretation of experimental results. \\
\section{Model and methods}
\subsection{Definitions and assumptions}
We denote $i \in [0,1,2]$ a type of individuals (types can be phenotypes, alleles, strains, clonal species, mutants, etc.). For the sake of simplicity, we will use in the rest of the paper the term \emph{mutant} $i$ when referring to type $i$ individuals. $N(t)=(N_i(t))$ is a vector whose elements are the number of mutants $i$ in the population at time $t$, with $N_i(t)$ a random variable. We assume that the environment has a fixed quantity of available resources: the carrying capacity $K$ modulates the intensity of competition between individuals. We investigate the population dynamics of three clonal types as a birth-death process with competition in continuous time. Each mutant $i$ is characterized by its individual ecological parameters: $\beta_i$ and $\delta_i$ are respectively the individual birth and natural death rates, and $C_{ij}$ is the effect of competition of a single mutant $j$ on a single mutant $i$, assuming $C_{ij} \geq 0$ and $C_{ii}>0$. For simplicity, we assume that competition between individuals affects mortality. The individual death rate of a mutant $i$ thus depends on both an intrinsic component ($\delta_i$) and a component due to competition: $d_i(N(t))=\delta_i+C_{i0} N_0(t)/K +C_{i1}N_1(t)/K+C_{i2}N_2(t)/K$.\\
We suppose that the resident population is only composed of mutants $0$, at its ecological steady-state equilibrium, say $N_0(0)$. A single mutant $1$ is introduced in the population ($N_1(0)=1$). The population of mutants $1$ follows a stochastic dynamics that depends on the ecological parameters and the competitive interactions between mutants $0$ and mutants $1$. The time taken for mutants $1$ to invade the resident population, and eventually get fixed, is of order $\log K$ \citep{champagnat06}. Since the dynamics is stochastic, mutants $1$ can either spread or be lost. Since we are interested in the dynamics of three competing clones, we will focus on cases where neither mutants $0$ nor mutants $1$ are lost when a single mutant $2$ enters the population by mutation (or migration). We assume that the time at which this event occurs is $\alpha \log K$, $\log K$ being the time scale of the whole stochastic dynamics. There are two general cases. Either $\alpha$ is low enough that mutants $1$ are still in too few numbers to affect the invasion of mutants $2$, or $\alpha$ is large enough that mutants $1$ have invaded the population and thus affect the invasion of mutants $2$. We will investigate both situations and show that the time $\alpha \log K$ at which mutant $2$ enters the population is crucial and largely affect the final state of the population.\\
\subsection{The stochastic dynamics is a succession of several phases}
When the population is large, the dynamics followed by the population can be divided into a succession of two kinds of phases. First, when a mutation enters a population, say a mutant $j$ enters in an single copy in a $i$ resident population, the dynamics of mutants $j$ is well approximated by a branching birth-death process without interactions until its population size $N(j)$ is large enough, i.e. when it is of order $K$ \citep{fourniermeleard04, champagnat06}. In the case of the joint dynamics of three interacting clones, the dynamics of two mutants $j$ and $k$ in a resident population $i$ is also well approximated by a branching birth-death process without interactions (see the proofs in the companion paper \cite{billiardsmadi}). Second, when two or three mutants have a population size of order $K$, then the stochastic dynamics of these populations is well approximated by a competitive Lotka-Volterra deterministic system \citep{fourniermeleard04, champagnat06, billiardsmadi}. During this phase, if the population size of one of the mutants is of order lower than $K$, then its population size essentially does not change until the deterministic equilibrium is reached. Even though the whole dynamics is stochastic, we will respectively call these two kinds of phases ``stochastic'' and ``deterministic'', for the sake of simplicity (see Fig. \ref{fig:phases}). The dynamics of the population can finally be described as a succession of ``stochastic'' and ``deterministic'' phases. Note that we only give in this paper the relevant biological results and a summarized version of the model (mathematical proofs and detailed computations are given in a companion paper \cite{billiardsmadi}).\\
{\bf Deterministic phases.} We denote by $n_i$ the size of the population of mutants $i$ when dealing with the deterministic dynamics, while we will keep the notation $N_i$ when dealing with stochastic dynamics ($n_i$ is the population size rescaled by the carrying capacity $N_i / K$ when $N_i$ is of order $K$). When the population size of the three mutants are of order $K$, the dynamics of the rescaled process can be well approximated by the following system of ordinary differential equations,
\begin{equation} \label{EDO}
\left\{\begin{array}{ll}
\dot{n}_0=(\beta_0-\delta_0-C_{0,0}n_0-C_{0,1}n_1-C_{0,2}n_2)n_0,\\
\dot{n}_1=(\beta_1-\delta_1-C_{1,0}n_0-C_{1,1}n_1-C_{1,2}n_2)n_1,\\
\dot{n}_2=(\beta_2-\delta_2-C_{2,0}n_0-C_{2,1}n_1-C_{2,2}n_2)n_2.
\end{array}
\right.
\end{equation}
Under the assumptions that competitive parameters are $C_{ij} \geq 0$ and $C_{ii}>0$ for all $\{i,j\}$, this system of deterministic equations is a three-dimensions Lotka-Volterra competitive model. Such a three species population shows different possible dynamics: different equilibrium states (either monomorphic or polymorphic, with two or three coexistent mutants), or stable limit cycles \citep{zeeman93,zeemanvandendriessche98,zeemanzeeman03}.\\
The fate of a mutant $i$ entering in a single copy a resident $j$ population is associated to the so-called ``invasion fitness'', denoted $S_{ij}=\beta_i-\delta_i-C_{ij} \bar{n}^j$, where $\bar{n}^j=\frac{\beta_j-\delta_j}{C_{jj}}$ is the population size of mutant $j$ at equilibrium when there are only mutants $j$ in the population. The invasion fitness corresponds to the initial growth rate of the mutant when it is rare. If the resident population is composed of both mutants $i$ and $j$ at equilibrium, then the fate of a mutant $k$ entering in a single copy is associated with the invasion fitness denoted $S_{kij}=\beta_k-\delta_k-C_{ki} \bar{n}^{i}_{ij}-C_{kj} \bar{n}^{j}_{ij}$, where
\begin{equation} \label{eqpop}
\begin{array}{ll}\\
\bar{n}^i_{ij}=\frac{C_{jj}(\beta_i-\delta_i)-C_{ij}(\beta_j-\delta_j)} {C_{ii}C_{jj}- C_{ij}C_{ji}}, \bar{n}^j_{ij}=\frac{C_{ii}(\beta_j-\delta_j)-C_{ji}(\beta_i-\delta_i)}{C_{ii}C_{jj}- C_{ij}C_{ji}},
\end{array}
\end{equation}
is the equilibrium of Eq. \ref{EDO} when there are only mutants $i$ and $j$ in the population. If $S_{kij}>0$, mutation $k$ is favorable when rare in the polymorphic resident population $(i,j)$ and can invade. \\
{\bf The stochastic phase.} When the population size of at least one mutant, say $i$, is low (\emph{i.e.} its population size is of order lower than $K$), while the other mutants are at their deterministic steady state (their population size is of order $K$), the dynamics of mutants $i$ is close to a pure birth-death process with birth and death rates respectively $\beta_i$ and $\delta_i+\sum_{j \neq i}C_{ij}n_j$, thus neglicting the effect of the competition between individuals $i$. When a mutant $i$ enters a resident $j$ population in a single copy, the probability of invasion of the mutant $i$, defined as the probability that the population of mutants $i$ reaches a size of order $K$, is $S_{ij}/\beta_i$ when $S_{ij}>0$, and a probability 0 if $S_{ij}\leq 0$. The time taken by a mutant $i$ which enters a $j$ resident population to reach the threshold population size is of order $\log K / S_{ij}$. We can similarly define the probability of invasion of a mutant $k$ in a resident population with both mutants $i$ and $j$ as $S_{kij}/\beta_k$ when $S_{kij}>0$, and a probability 0 if $S_{kij}\leq 0$. The time taken by a mutant $k$ which enters a $i$ and $j$ resident population to reach the threshold population size is of order $\log K / S_{kij}$. In both cases, the stochastic phase ends when the mutant is either lost or reaches a threshold population size of order $K$.
\subsection{Stochastic dynamics and final states with three competitive clones}
When a single favorable mutant enters a resident population, the stochastic dynamics can be decomposed into three successive phases (see Fig. \ref{fig:phases}a and \ref{fig:phases}b): First, a stochastic phase corresponding to the beginning of the mutant invasion and where the mutant has a population size of order lower than $K$; Second, a deterministic phase when the population size of the new mutant is large enough (of order $K$); Third, a new stochastic phase until the resident mutant is lost (if both mutants do not stably coexist at deterministic equilibrium). In the case of competition between three clones with two mutations entering a resident population, the succession of stochastic and deterministic phases must be decomposed into a higher and not limited number of phases.\\
{\bf What determines the different dynamics and final states when there is competition between three clones?} The different dynamics and final states depend on the ecological parameters and also on the following conditions (see details of computation in Appendix A1 and illustrations in Fig. \ref{fig:phases} and \ref{fig:clonalreinforcement}, simulation algorithm given in Appendix A3):\\
i) Does mutant $2$ enter the population during the first (Fig. \ref{fig:phases}a) or second (Fig. \ref{fig:phases}b) stochastic phase (depending on the time $\alpha \log K$)? If mutant 2 enters the population during the first stochastic phase then mutant 1 has a population size of order lower than $K$. Consequently, mutant 2 suffers the competitive effect of mutants 0 only. If mutant 2 enters the population during the second stochastic phase then, assuming mutant 1 is favorable in the resident population, there are two possibilities: either mutants 0 and 1 stably coexist, in which case mutant 2 suffers the competitive effects of both mutants, either mutant 0 has a population size of order lower than $K$, in which case mutant 2 suffers the competitive effects of mutant 1 only; \\
ii) When mutant 2 enters the population during the first stochastic phase (Fig. \ref{fig:phases}a), does mutant 1 or 2 reaches first a population size of order $K$? Since mutants 1 and 2 have a population size lower than order $K$, the speed at which they invade the resident population only depends on their competitive interactions with the resident mutants 0. Hence, which mutants reaches first a population size of order $K$ depends on their invasion fitness $S_{10}$ and $S_{20}$ and on the time when mutant 2 enters the population $\alpha \log K$. The first mutant which reaches a population size of order $K$ determines the initial state of the succeeding deterministic phase; \\
iii) What is the equilibrium of the first deterministic phase: stable coexistence of two mutants, i.e. two mutants have a population size of order $K$, or a single mutant has a population size of order $K$? This only depends on the sign of the invasion fitnesses (Eq. \ref{EDO}). For instance, mutants 0 and 1 stably coexist if $S_{01}>0$ and $S_{10}>0$. \\
iv) What is the population size of all mutants when the second stochastic phase begins? It depends on whether two mutants stably coexist or not at the end of the deterministic phase (step iii), and on the population size of the mutant which did not invade and still has a size lower than order $K$; \\
v) Does a mutant go extinct before the start of the next deterministic phase? When a mutant has a population size of order lower than $K$ and is deleterious in a given context, it is expected to go extinct. However, its time to extinction can take longer than the time for another rare mutation to reach a population size of order $K$. In this case, a new deterministic phase begins. The ecological context of the deleterious mutation can change before it goes extinct, which changes its fate. \\
vi) Steps ii-v are again applied for the further succeeding phases (when applicable) as often as necessary until a final steady state is reached.\\
{\bf The different possible final states and their hitting times.}
Our goal is to investigate how clonal interference might affect the dynamics of mutant populations and thus adaptation. We will thus especially focus on cases where mutation $1$
and $2$
have a positive invasion fitness when 2 enters the population during the first stochastic phase ( $S_{10}>0$and $S_{20}>0$), and on cases where mutation $1$ has a positive invasion fitness( $S_{10}>0$) if mutation 2 enters the population during the second stochastic phase (in the latter case, the invasion fitness of mutant 2 depends on the currect state of the population).
All the possible final states and their hitting times are compiled in Tables \ref{tableass1} and \ref{tableass2}. We do not give all detailed calculations for all cases here, we only give one detailed example in Appendix A2 as an illustration (for complete and detailed computations see the companion paper \citet{billiardsmadi}). Roughly, two classes of final states are possible: either one mutant goes to fixation (it can be either 0, 1 or 2), or two or three mutants coexist (in all possible combinations).
\subsection{ Likelihood of the final states}
We want to estimate the likelihood of the different possible final states assuming ecological parameters are drawn in given prior distributions. The complexity of the model can be reduced to: $\rho_i=\beta_i-\delta_i$, the net individuals reproductive rate of mutants $i$, and $\widetilde{C}_{ij}=\frac{C_{ij}}{C_{jj}}$ the ratio of the between and within competitive interactions. We drew $10^6$ different sets of parameters in prior distributions. For a given set of parameters and a given $\alpha$, the final state is given by Tab. \ref{tableass1} and \ref{tableass2}, which allows to estimate the posterior distribution of the final states among the $10^6$ random sets of parameters. \\
The time $\alpha \log K$ at which the second mutation enters the population has a large impact on the final states. We can determine the final states for a given parameter set for any $\alpha$ using the results of our model (Tab. \ref{tableass1} and \ref{tableass2}). Assuming mutation 2 enters the population during the first stochastic phase, we know that mutant 2 necessarily appears before mutant 1 spreads out, i.e. $\alpha<1/S_{10}$. If mutation 2 enters the population during the second stochastic phase, mutant 2 necessarily appears after mutant 1 spreads out and before mutant 0 goes extinct, i.e. $1/S_{10}<\alpha<1/S_{10}+1/|S_{01}|$. We also know that there are two threshold values $\alpha < 1/S_{10}+1/S_{20}(S_{21}/|S_{01}|-1)$ and $\alpha < S_{02} S_{21}/(S_{10} |S_{12}| |S_{01}|)-1/S_{01}$ (Tab. \ref{tableass1}) which determine which dynamics is followed by the population when mutation 2 enters the population during the first stochastic phase. Similarly, there are two threshold values $1/S_{10}<\alpha < 1/S_{10}+1/|S_{01}|-1/S_{21}$ and $1/S_{10}<\alpha <1/S_{10}+S_{02}/(|S_{12}||S_{01}|)-1/S_{21}$ (Tab. \ref{tableass2}) which determine which dynamics is followed by the population when mutation 2 enters the population during the second stochastic phase (see above). We can thus finally compute the probability of any final states given an ecological parameter set and assuming $\alpha$ is uniformly drawn in the interval $\left[ 0, 1/S_{10}\right]$ or $\left[ 1/S_{10}, 1/S_{10}+1/|S_{01}|\right]$ respectively when mutation 2 enters the population during the first or second stochastic phase. \\
{\bf Effect of mutations on the reproductive rates.} In bacteria, yeasts or some eukaryotes, fitness is generally estimated as the initial growh rate (at low density) of mutants (see Table 2 in \cite{martinlenormand06} and the Appendix in \cite{mannaetal12}). We will thus assume that the effect of mutations on the growth rate of mutant $i$ follows a Fisher's geometric model. Given the net reproduction rate of mutants 0 is $\rho_0$, we assumed that the reproductive rate of mutant $i$ is $\rho_i=\rho_0 + x_i$ with $x_i$, the effect of mutation $i$, being drawn in a shifted negative Gamma distribution which is an approximation of a Fisher's geometric model for adaptation \citep{martinlenormand06}. Note that when mutation 2 enters the population during the second stochastic phase, mutation 2 is assumed to occur in the most frequent mutation at equilibrium: $\rho_2=\rho_1 + x_2$ when mutant 1 is more frequent than mutant 0, $\rho_2=\rho_0 + x_2$ otherwise.\\
{\bf Effect of mutations on competition.} There is, to our knowledge, no theoretical or empirical consensus on the distribution of mutation effects on the competitive abilities $\widetilde{C}_{ij}$. Without any knowledge about the distribution of competitive abilities, we simply assumed that the ratio of competitive interaction $\widetilde{C}_{ij}$ follows arbitrary chosen distributions. First, we assumed that it follows a uniform distribution in the interval $\left[1-\mu, 1+\mu \right]$, with $0 \leq \mu \leq 1$. Second we assumed it follows an exponential distribution with parameter $\mu>0$. When $\mu=0$, all $\widetilde{C}_{ij}=1$, fitnesses are necessarily transitive, while if $\mu>0$, frequency-dependent fitnesses can occur. As $\mu$ increases, the variance of the competitive ratio $\widetilde{C}_{ij}$ also increases, i.e. the more different can the competitive interactions be between mutants. \\
\section{Results }
Given that two favorable mutations successively enter in a single copy a resident population 0, Tables \ref{tableass1} and \ref{tableass2} show all possible dynamics and final states, depending on the sign of the invasion fitnesses $S_{ij} \text{ and } S_{ijk}, \{i,j,k\} \in \{0,1,2 \}$, and the time of appearance of the second mutation $\alpha \log K$. When the second mutation enters the population during the first stochastic phase, Tab. \ref{tableass1} shows that six final states are possible: fixation of either mutation 1 or 2, stable polymorphic equilibrium with two (mutants 0 and 1, 1 and 2 or 0 and 2) or three mutants (mutants 0, 1 and 2). Tab. \ref{tableass1} also shows that a given possible final state can be reached under different conditions, and consequently after different durations. For instance, mutant 2 can fix under three different cases (a subcase of B, and cases E and I), with potentially different fixation times. When mutation 2 enters the population during the second stochastic phase, Tab. \ref{tableass2} shows that a seventh final state is possible: the fixation of mutant 0, even if mutants 1 and 2 are advantageous when they enter the population. Interestingly, Rock-Paper-Scissors cyclical dynamics can only occur if the second mutation enters the population during the second stochastic phase. Assuming three co-occuring competing clones, our model can thus capture a large diversity of dynamics and allows to determine their duration, final states and likeliness. In the following, we first show that despite the complexity and variety of the possible stochastic dynamics with three competing clones, six possible general dynamics can be defined. Second, we focus on several special cases of particular interest. We especially argue that our model, despite its simple assumptions, captures a large range of dynamics diversity observed in experiments.
\subsection{Beyond clonal interference: Six possible dynamics}
Two categories of dynamics have been proposed in the empirical literature to explain observations in experiments with several interacting clones: clonal interference, when adaptation is slowed down because of the interaction between advantageous mutations \citep{gerrishlenski98}, or clonal reinforcement \citep{kinnersleyetal14}, also called niche construction or frequency-dependent selection elsewhere \citep{yangetal14}, when several clones stably coexist. Using our model, focusing on the second mutation (mutant 2), we can embrace the two previous proposed categories, and propose an alternative categorization, with more accurate definitions. \\
Competition between three clones can affect adaptation for three reasons: 1) It can promote or hinder polymorphism maintenance; 2) The invasion probability of mutation 2 can be increased or decreased (relatively to the case where mutant 2 enters alone the resident population, i.e. there are only two interacting clones); 3) In cases where mutation 2 goes to fixation, its fixation time can be be shorter or longer. Hence, dynamics with three competing clones can be classified in six general cases: When clonal interaction promotes polymorphism maintenance, we call the dynamics ``clonal coexistence''. When the mutant 2 goes to fixation, we call ``clonal assistance'' when the duration of the sweep is shorter, and ``clonal interference'' (following \cite{gerrishlenski98}) when it is larger than with only two competing clones. Finally, we call ``soft'' vs. ``hard'' the dynamics depending on wheter the invasion probability of the mutant 2 is lower vs. higher than with only two competing clones. This gives 6 possible general dynamics, summarized in Table \ref{tab:cat}.\\
{\bf Rate of adaptation: fixation time vs. probability of invasion of beneficial mutants.} Clonal interference is viewed in the literature as the phenomenon of the increase in fixation time of co-occuring beneficial mutants in a population, which consequently decreases the rate of adaptation of clonal species. However, the rate of adaptation can be affected also by the rate at which new mutants invade a population, i.e. by their probability of invasion. To our knowledge, in all models dealing with clonal interference so far, derived from population genetics models, the probability of invasion of a new mutant only depends on its own features: in general, a selection coefficient $s$ is arbitrarily assigned to a mutant independently of the composition and state of the resident population when this mutant occurs. Its invasion probability and its time of fixation are approximately $2s$ and $1/s$: increasing $s$ necessarily both increases the probability of fixation and decreases the time of fixation. In our approach, we have a more general point of view: both the fixation time and probability of invasion depend on the state of the population when a mutant enters the population. We thus argue that fixation times and probabilities of invasion should be considered independently in order to evaluate to which extent interaction between various clones can affect rate of adaptation. Interestingly, depending on the state of the population when mutation occurs and on their own ecological specificities, the rate of adaptation can increase: the probability of invasion can be higher or the time to fixation can be shorter, what we propose to call ``hard clonal assistance''. \\
{\bf Clonal coexistence: cooperative interactions are not necessary.}
Several experiments of competition between clones have shown stable persistence of different strains in a single well-mixed population, which has been explained by frequency-dependent selection, niche construction or cooperative interactions. Especially, \citet{kinnersleyetal14} introduced the concept of ``clonal reinforcement'' when ``the emergence of one genotype favors the emergence and persistence of other genotypes via cooperative interactions''. Here we show that clones can favor either the emergence (``hard'' vs. ``slow'' dynamics), or the persistence (``clonal coexistence'' vs. ``clonal assistance'' or ``clonal interference'') or both (``hard clonal coexistence''), without any cooperative interactions between clones, but only competition. We do not argue that persistence observed in \citet{kinnersleyetal14} are not effectively due to cooperative interactions, rather we propose an alternative hypothesis: both facilitated emergence and stable persistence of clones can be due to non-transitive competitive interactions, or frequency-dependent selection. It would need specific experimental work to show whether or not clones effectively cooperate.
\subsection{Two specific dynamics when the second mutant lately enters the population: Rock-Paper-Scissors or annihilation of adaptation}
Two related specific dynamics are encountered only when mutant 2 enters the resident population during the second stochastic phase ($\alpha>1/S_{10}$, Tab. \ref{tableass2}): Rock-Paper-Scissor dynamics (final state J in Tab. \ref{tableass2}), or a return to the initial state, i.e. a population fixed for mutant 0 (final state G in Tab. \ref{tableass2}). These two different dynamics are illustrated in Fig. \ref{fig:phases}b and Fig. \ref{fig:clonalreinforcement}b. Both dynamics have the same parameters, except the time at which mutant 2 enters the population $\alpha \log K$. This illustrates the importance of considering stochastic dynamics: if mutant 2 enters the population early enough that mutant 1 is not extinct when mutant 2 invades, then Rock-Paper-Scissors cyclical dynamics take place, otherwise mutant 0 goes to fixation and adaptation is annihilated despite the occurrence of two beneficial mutations. In a deterministic model, for the same parameters, a mutant can not go extinct and only Rock-Paper-Scissor dynamics is possible. \\
Our results also show that Rock-Paper-Scissors dynamics can be obtained in a narrow set of parameters. In addition to specific parameters values regarding the competitive interactions between the three clones, the second mutant must occur in the population in a narrow time frame. First, it must occur after mutant 1 invaded, since mutant 2 is deleterious in a mutant 0 resident population. Second, if mutant 2 occurs too late during the second stochastic phase, then mutant 0 can be extinct before mutant 2 invades, in which case mutant 1 goes to fixation. These results have important consequences regarding our understanding of empirical Rock-Paper-Scissor dynamics observed in natural populations: either the three types of individuals involved in such stable cycles have effectively entered the population by mutation or migration in a single individual, in which case the third type of individuals has necessarily entered the population in a narrow time frame. Otherwise, the alternative explanation is that the three types of individuals went together in a single population with a sufficiently large enough population size such that the dynamics initially followed an almost-deterministic dynamics, which certainly occurred by a massive migration and mixing of three different and complementary types of individuals
\subsection{Likehood of the final states assuming prior distributions of the ecological parameters}
Figures \ref{fig:Distrib1} and \ref{fig:Distrib2} show the posterior probability of the dynamics and final states when mutation 2 enters the population during the first and second stochastic phases, assuming that the competition abilities are respectively drawn in an uniform or exponential distribution. When the variance of the distribution of the $\widetilde{C}_{ij}$ is low, all clones have similar competitive abilities ($\widetilde{C}_{ij} \simeq 1$), \emph{i.e.} invasion fitness are mostly transitive. We naturally recover predictions from population genetics models: The likeliest scenari are the fixation either of mutant 1 or 2 (Fig. \ref{fig:Distrib1}c, \ref{fig:Distrib1}d, \ref{fig:Distrib2}c, \ref{fig:Distrib2}d). Rapidly, when the variance of the uniform and exponential distributions increases, polymorphic final states become the likeliest. When the effect of mutation on competitive abilities become large, the likeliness of all dynamics rapidly reaches a plateau when competitive abilities ($\widetilde{C}_{ij}$) are drawn in an uniform distribution (Fig. \ref{fig:Distrib1}). When competitive abilities are drawn in an exponential distribution, the likeliest state is the one with three coexisting clones. Our results suggest that non-transitive fitness are mostly expected to occur when several clones are interacting as soon as mutations affect their competitive abilities. This further supports that clonal coexistence is likely to occur even when considering only competitive interactions: cooperative interactions are not necessary to explain the stable coexistence of several clones. Finally, our results show that Rock-Paper-Scissors dynamics and annihilation of adaptation are weakly probable.\\
Comparing left and right columns in Fig. \ref{fig:Distrib1} and \ref{fig:Distrib2} shows that the time at which mutation 2 enters the population only marginally affects the dynamics and the final states. Interestingly, comparing the final states between cases with two or three interacting clones (Fig.\ref{fig:Distrib2clones}) shows that more polymorphic final states are expected when three clones are interacting, even though the difference is small. Whether a further increase in the number of interacting clones could even more promote the maintenance of polymorphism because of non-transitive fitness. Fig. \ref{fig:Distrib1}, \ref{fig:Distrib2} and \ref{fig:Distrib2clones} also show that the prior distribution of the competitive abilities has important effect: polymorhic final states are more expected when the distribution is exponential.\\
Finally, Fig. \ref{fig:Distrib1}e-f and \ref{fig:Distrib2}e-f show the likeliness of clonal interference \emph{vs.} clonal assistance. When competitive abilities follow an uniform distribution, clonal interference, \emph{i.e.} the slowing down of fixation of beneficial mutation, is the most probable, especially when the competitive abilities are similar between clones (small $\mu$). However, rapidly when the difference between competitive abilities increases (large $\mu$) the likeliness of clonal assistance increases, and reaches a plateau. When mutation 2 enters the population during the second stochastic phase, clonal assistance is even likelier than clonal interference. The situation is different and more complex when competitive abilities follow an exponential distribution, especially when the second mutation enters the population during the second stochastic phase. Contrarily to the uniform distribution, the exponential distribution is asymetrical around the mean, which generates asymetrical invasion fitness distributions and explains the difference observed in our results. Indeed, when the mean $\mu$ is low, the distribution of the invasion fitness is skewed towards negative values, while it is skewed towards positive values when $\mu$ is large (not shown), which explains why polymorphic states are the likeliest for large $\mu$. Globally, our results thus suggest that clonal interference might indeed be an important factor affecting adaptation rate, but clonal assistance can be as important given non-transitive fitnesses are possible.
\section{Discussion}
Both theory and experimental observations tend to agree regarding the rate of adaptation in evolving large asexual populations: the speed at which new beneficial mutations go to fixation should decrease during the course of adaptation and reach a plateau \cite[e.g.][]{gerrishlenski98, desaifisher07}, what is effectively found in experimental evolution of bacteria \citep{devisserrozen06}. Clonal interference is believed to be the most important factor causing this limit to adaptation: different beneficial mutations occurring simultaneously in competing lineages tend to mutually decrease their probability of fixation and increase their time to fixation, what has been effectively observed in viruses \citep[e.g.][]{panditdeboer14}, bacteria \citep{devisserrozen06} or cancer \citep{greavesmaley12}. This congruence between models and data can be challenged regarding our results. Indeed, by explicitly taking into account competitive interactions between different clones, we showed that the evolutionary dynamics can be much more complex: competitive interactions between clones can either impede or foster adaptation by modifying fixation times and probabilities, and can promote coexistence of clones in the long-term. Our results are different from previous works because in our model non-transitive fitnesses are allowed, giving rise to frequency-dependent selection, non-linear and cyclical dynamics. Our model is in fact more general than previous works since they are a subcase where all competitive interaction abilities are equal ($\widetilde{C}_{ij}=1$). Most importantly, we showed that frequency-dependent selection is very likely when mutations have an effect on competitive interaction abilities (Fig. \ref{fig:Distrib1} and \ref{fig:Distrib2}), giving rise to almost equiprobable dynamics with either clonal interference or clonal assistance, i.e. respectively a decrease or an increase in the rate of adaptation. \\
We might thus wonder why a decreasing rate of adaptation is generally observed in experimental evolution. In other words, why does adaptation in clonal species mostly follow a single subcase of our model where all competitive interaction abilities are almost equal ($\widetilde{C}_{ij} \simeq 1$). Several hypotheses can be made. First, as Lenski and colleagues claimed regarding their long-term experimental evolution, the experiments were designed to avoid frequency-dependent selection: no horizontal gene transfer, a well-mixed environment with no spatial structure and a low concentration of the density-limiting resource \citep{maddamsettietal15}. One can thus imagine that these experimental conditions were effectively fulfilled and their goal is achieved. However, several experiments showed that even in ideal conditions, frequency-dependent selection occurs \citep{maddamsettietal15, kinnersleyetal14, langetal11}. Hence, even if the experimental design limits the occurrence of mutations with non-transitive effects on fitness, it does not obliterate it. A question thus remains: why is frequency-dependent selection so rare in experimental evolution? Second, we only considered three interacting clones in our model, while much more can be competing in a large clonal population. To our knowledge, it is not known whether an increasing number of competing clones decreases the plausibility of non-transitive interactions. However, we showed that non-transitive interactions are more probable with three than two interacting clones (Fig. \ref{fig:Distrib2clones}). Deterministic models with four interacting clones have also shown that even more complex dynamics can be expected, especially chaotic ones \citep{arneodoetal82, vanoetal06}. This suggests that a large number of interacting clones could not be the main cause for the rarity of non-transitive interactions. Third, our estimation of the likeliness of frequency-dependent selection are based on two strong hypotheses: the effect of mutations of competitive interaction abilities follows either an uniform or exponential distribution, and a mutation shows no trade-offs between its effect on the growth rate and the competitive abilities. To our knowledge there is neither empirical nor theoretical treatments about the distribution of the competitive interactions and the trade-offs between species or clones. \citet{maharjanetal06} estimated strength of frequency dependent selection in a limited range of frequencies in a well-mixed environment. They especially showed that non-transitive interactions can have similar extent than transitive interactions. \citet{galletetal12} detected no intransitive interactions between clones three pairs of bacterial clones in a simple experimental sets, suggesting that competitive interactions are weak. Yet, a general treatment is lacking in order to evaluate the potential importance of non-transitive interactions in adaptive populations. Whether non-transitive interactions can be expected to be frequent or not is an important question since it can challenge the generality of the evolutionary dynamics observed in controlled experimental evolution of clonal species. \\
Non-linear dynamics have been observed in various experiments and different explanations and interpretations have been proposed. \citet{langetal11} observed different dynamics in different replicates. They especially observed cases where a lineage showed two successive frequency peaks. They explained this observation by the occurrence of a third cryptic mutation affecting a preexisting lineage, what they called ``multiple mutations'' dynamics. We have shown in our model that such a dynamics can be explained with three interacting clones only: such cyclical dynamics can be observed in cases of Rock-Paper-Scissor. Importantly, we show that such cyclical dynamics can only be observed if the second mutation occurs during the second stochastic phase, i.e. when the first mutation is near fixation. It would be interesting to look thoroughly into the data to challenge this prediction. \citet{maharjanetal06}, \citet{langetal11} and \citet{maddamsettietal15} also showed the possible coexistence of clones in the long-term, which can effectively be explained by negative frequency-dependent selection, as proposed here. \citet{rosenzweigetal1994} and \citet{kinnersleyetal14} showed the long-term coexistence of three lineages derived by mutation from a single initial \emph{Escherichia coli}clone, in a long-run experimental evolution in a chemostat. They interpreted this coexistence as due to cooperative rather than competitive interactions, yet we showed that non-transitive competitive interactions can explain the emergence by mutation and stable coexistence of three clones. \citet{rosenzweigetal1994} showed that the three clones can stably coexist by pairs. Under our framework, it means that $S_{ij}>0$ for all $\{i,j\} \in \{ 0,1,2\}$, and we predict that the three clones should indeed coexist because of non-transitive competitive interactions. The three clones consume the same resources (glucose, acetate, glycerol, \cite{hellingetal87}, \cite{rosenzweigetal1994}) supporting the existence of competitive interactions. However, looking thoroughly at the data, it is not obvious that one of the clones is maintained in coexistence when in competition with another: its frequency is less than 1\% and sometimes even not detected in the experiments (see Fig. 1d in \citet{rosenzweigetal1994}). Taking an opposite point of view than the authors of the experiments, we can conclude that mutant 2 is not maintained when in competition with mutant 1 (\emph{i.e.} $S_{12}<0$). In this case, our model predicts that the three clones can not stably coexist, and consequently cooperation between the three clones can be an exclusive explanation for the observed coexistence.
\renewcommand\refname{Literature Cited}
\bibliographystyle{amnat2}
| 1d781feae6a8a59cf27da3c02181853c90ae7bb0 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
\subsection{Overview}
Distributing points on $\mathbb{S}^2:=\left\{x\in\mathbb{R}^3 : \ |x|=1\right\}$ is a classical problem arising in many settings in numerical analysis and approximation theory, in particular the study of radial basis functions, quadrature and polynomial interpolation, Quasi-Monte Carlo methods for graphics applications, finite element methods for PDE's, cosmic microwave background radiation modeling, crystallography and viral morphology, to name a few. The goal of this paper is to survey some widely used algorithms for the generation of spherical node sets. We will restrict our descriptions to ``popular" point sets, most of which can be generated ``reasonably fast." Namely, we study
\begin{itemize}
\item Fibonacci and generalized spiral nodes
\item projections of low discrepancy nodes from the unit square
\item zonal equal area nodes and HEALPix nodes
\item polygonal nodes such as icosahedral, cubed sphere, and octahedral nodes
\item minimal energy nodes
\item maximal determinant nodes
\item random nodes and
\item ``mesh icosahedral equal area nodes."
\end{itemize}
The last is a new point set devised to have many desirable properties. For each of the above configurations, we provide illustrations, and analyze several of their properties. We focus our attention primarily on
\begin{itemize}
\item equidistribution
\item separation
\item covering
\item quasi-uniformity and
\item Riesz potential energy.
\end{itemize}
For each property we provide numerical calculations, tables, and comparisons, and in some cases we prove theoretical bounds on the mesh ratio. Section 3 is devoted to asymptotic comparisons of various potential energies. We do not consider quadrature of the point sets; however, such a comparison for several of the configurations we describe here can be found in \cite{QMC}. We now formally introduce the properties we will be studying and, in Section 2, describe the point sets themselves. We leave the technical proofs to Section 4, and at the end of the document, we provide resources for Matlab source codes to generate the point sets.
\subsection{Definitions and Properties}
For low error numerical integration with respect to uniform surface area measure (\cite{ABD12} and \cite{KuipNied}) as well as in digitizing $\mathbb{S}^2$ for computer graphics purposes (\cite{signal} and \cite{FibInt}), it is important for any spherical configuration to have an approximately uniform distribution. A sequence $\left\{\omega_N\right\}_{N=1}^{\infty}$ of spherical point sets with $\omega_N$ having cardinality $N$ is called \textit{equidistributed} if the sequence of normalized counting measures,
\[\nu_N(A):=\frac{1}{N}|A\cap\omega_N|,\ \ \ \ \ \ A\ \textup{Borel set},\]
associated with the $\omega_N$'s converges in the weak-star sense to $\sigma$, the normalized surface area measure on $\mathbb{S}^2$, as $N\rightarrow \infty$. That is, for all continuous functions $f$ on $\mathbb{S}^2$,
\[\lim_{N\to\infty}\int_{\mathbb{S}^2}f\,\text{d}\nu_N = \int_{\mathbb{S}^2}f\,\text{d}\sigma.\]
An equivalent definition is that the $L_\infty$-spherical cap discrepancy
\[D_C(\omega_N):=\sup_{V\subset \mathbb{S}^2}\bigg|\frac{|V\cap \omega_N|}{N}-\sigma(V)\bigg|\to 0 ,\ \ \ \ \ \ \ N\to\infty,\]
where the supremum is taken over all spherical caps $V\subset \mathbb{S}^2$.
For the study of local statistics, separation and covering properties play an important role. The \textit{separation} of a configuration $\omega_N\subset \mathbb{S}^2$ is
\[\delta(\omega_N):=\min_{\substack{x,y\in\omega_N\\ x\neq y}} |x-y|, \]
and a sequence of spherical $N$-point configurations is said to be \textit{well-separated} if for some $c>0$ and all $N\geq 2$,
\[\delta(\omega_N)\geq cN^{-1/2}.\]
The \textit{covering radius} of $\omega_N$ with respect to $\mathbb{S}^2$ is defined to be
\[\eta(\omega_N):= \max_{y\in \mathbb{S}^2} \min_{x\in \omega_N} |x-y|,\]
and a sequence of spherical $N$-point configurations is a \textit{good-covering} if for some $C>0$ and all $N\geq 2$,
\[\eta(\omega_N)\leq CN^{-1/2}.\]
A sequence of configurations $\left \{\omega_N\right \}_{N=2}^\infty$ is said to be \textit{quasi-uniform} if the sequence
\[\bigg\{\gamma (\omega_N) := \frac{\eta (\omega_N)}{\delta (\omega_N)}\bigg\}_{N\geq 2}\]
is bounded as $N\rightarrow \infty.$ The quantity $\gamma(\omega_N)$ is called the \textit{mesh ratio} of $\omega_N$. Note that some authors define the mesh ratio as $2\gamma(\omega_N)$. A sequence of $N$-point configurations is quasi-uniform if it is well-separated and a good-covering. We remark that equidistribution does not imply quasi-uniformity or vice versa. In applications involving radial basis functions, ``1-bit" sensing, and finite element methods (\cite{RBF2}, \cite{RBF1}, \cite{RezSaff}, and \cite{FEM}), there is interest in precise bounds on $D_C(\omega_N)$, $\delta(\omega_N),\eta(\omega_N)$, and $\gamma(\omega_N)$. A trivial lower bound is $\gamma(\omega_N)\geq 1/2$ for any configuration. Asymptotically, as proved in \cite{BondHS},
\[\gamma(\omega_N) \geq \frac{1}{2\cos\pi/5} + o(1) = \frac{\sqrt{5}-1}{2} +o(1), \ \ \ \ N \rightarrow \infty,\]
for any sequence of configurations $\left \{\omega_N\right \}_{N=2}^\infty \subset \mathbb{S}^2$.
We also evaluate the potential energy of our point sets. The problem of minimizing point energies on the sphere dates to at least the beginning of the 20th century when Thomson put forth a model of the ground state configurations of electrons in \cite{Thoms}. Given a lower-semicontinuous, symmetric kernel $K:\mathbb{S}^2\times\mathbb{S}^2\rightarrow(-\infty,\infty]$, and a spherical configuration $\omega_N\subset\mathbb{S}^2$, the $K$-energy of $\omega_N$ is defined to be
\begin{equation}
\text{E}_K(\omega_N):=\sum_{\substack{x,y\in\omega_N\\ x\neq y}} K(x,y).
\label{energy_def}
\end{equation}
\noindent The infimum of $\text{E}_K(\omega_N)$ over all $N$-point configurations on $\mathbb{S}^2$ is attained and is denoted by $\mathcal{E}_K(N)$. We will restrict our attention to the class of Riesz kernels defined by
\[\ \ \ \ \ \ \ \ \ \ \
K_s(x,y) = \frac{1}{|x-y|^s},\ \ \ \ \ \ \ \ s > 0\]
\[
K_{\text{log}}(x,y) = \text{log}\frac{1}{|x-y|},\]
\[\ \ \ \ \ \ \ \ \ \ \
K_s(x,y) = -|x-y|^{-s},\ \ \ \ \ \ \ \ s < 0.\]
For brevity, the energy and minimal energy quantities for the Riesz $s$-kernel and log kernel will be denoted by $\text{E}_s(\omega_N)$, $\text{E}_{\text{log}}(\omega_N)$, $\mathcal{E}_s(N)$, and $\mathcal{E}_{\text{log}}(N)$ respectively. Determining an exact minimal configuration for a fixed $N$ and $s$ is a highly nonlinear optimization problem. In practice, gradient descent and Newton methods are used to arrive at approximate global minima \cite{Calef}; however, there is substantial interest in generating nearly optimal points more quickly~\cite{Smale}.
The \textit{Voronoi cell} of a point $x\in\omega_N\subset\mathbb{S}^2$ is the spherical polygon
\[V_x(\omega_N):= \left\{y\in\mathbb{S}^2:\ |y-x| \leq |y-z|, \forall \ z\in\omega_N\setminus \left\{x\right\}\right\}.\]
The \textit{Voronoi decomposition} of a configuration is
\[V(\omega_N):=\left\{V_x(\omega_N)\right\}_{x\in\omega_N}.\] It has been observed that the Voronoi cell decomposition of nearly optimal energy configurations appears to consist primarily of nearly regular spherical hexagons mixed with ``scars" of spherical heptagons and pentagons as shown in Figure \ref{MinEnVoronoi}. It has been conjectured that for $s>2$, the dominant term in the asymptotic expansion of $\mathcal{E}_s(N)$ is related to the Epstein-Zeta function for the hexagonal lattice (see Section 3).
\begin{figure}[h!]
\centering
\includegraphics{MinEnergyVoronoi.png}
\caption{Voronoi decomposition of approximately minimal energy nodes for $N=1000$ and $s=2$. Cells with black points are hexagonal, but not necessarily regular. Cells with red points are pentagonal and cells with cyan points are heptagonal (see online for color).}
\label{MinEnVoronoi}
\end{figure}
We conclude this section by introducing some secondary properties of spherical configurations that we will use:
A partition $P_N := \left\{W_i\right\}_{i=1}^N$ of $\mathbb{S}^2$ into $N$ cells whose pairwise intersections have $\sigma$-measure $0$ is \textit{equal area} if $\sigma(W_i) = 1/N$ for all $1\leq i\leq N$. A sequence of partitions $\left \{P_N\right \}_{N=2}^\infty$ of $\mathbb{S}^2$ such that each $P_N$ has $N$ cells is \textit{diameter bounded} if there are constants $c,C > 0$ such that for all $N\in \mathbb{N}$ and for every cell $W^N_i\in P_N$,
\begin{equation}
cN^{-1/2}\leq \text{diam }W^N_i \leq CN^{-1/2},
\label{diambdd}
\end{equation}
\noindent where diam$(A):=\sup_{x,y\in A}|x-y|$. We will call a sequence of partitions of $\mathbb{S}^2$ \textit{asymptotically equal area} if
\begin{equation}
\lim_{N\rightarrow\infty}N\max_{1\leq i\leq N} \sigma(W_i^{N}) = \lim_{N\rightarrow\infty}N\min_{1\leq i\leq N}\sigma(W_i^{N}) = 1,
\label{asympeq}
\end{equation}
\noindent and a sequence of spherical configurations $\left\{\omega_N\right\}_{N=1}^{\infty}$ will be said to be asymptotically equal area if its sequence of Voronoi decompositions is asymptotically equal area.
\section{Point Sets on $\mathbb{S}^2$}
\noindent\textbf{Generalized Spiral Points}
\
A spherical spiral on $\mathbb{S}^2$ is a path in spherical coordinates of the form
\[r = 1,\ \ \ \ \ \theta = L\phi, \ \ \ \ \ 0 \leq \phi \leq \pi,\]where $\phi$ denotes the polar angle and $\theta$ the azimuth.
Modifying a construction by Rakhmanov, Saff, and Zhou \cite{RSZ94}, Bauer \cite{Bauer} defines a sequence of $N$ points lying on a generating spherical spiral, $S_N$:
\begin{equation}
L = \sqrt{N\pi}, \ \ \ h_k = 1 - \frac{2k-1}{N}, \ \ \ \phi_k = \cos^{-1}(h_k), \ \ \ \theta_k = L\phi_k, \ \ \ k=1,...,N.
\end{equation}
\noindent The slope $L$ is chosen such that for large $N$, the distance between adjacent points on the same level of the curve is similar to the distance between adjacent levels which differ by $2\pi$ in $\theta$. Indeed, the geodesic spacing between turns of the spiral is given by $2\pi/L = \sqrt{4\pi/N}$. Meanwhile, the total arc length is
\[T = \int_{S_N} \sqrt{d\phi^2+d\theta^2\sin^2\phi} = \int_{0}^{\pi} \sqrt{1+L^2\sin^2\phi}\ d\phi = 2\sqrt{1+L^2}\textup{E}\big(L/\sqrt{1+L^2}\big),\]
where E$(\cdot)$ is the complete elliptic integral of the second kind. For large $N$,$\ \ T\approx~2L$, and the spiral is divided into nearly equal length segments of approximately $2L/N =\sqrt{4\pi/N}$. We refer to these points as the \textit{generalized spiral points}.
\
\begin{theorem}
\label{spiral distribution}
The sequence $\left\{\omega_N\right\}_{N=1}^{\infty}$ of generalized spiral point configurations is equidistributed on $\mathbb{S}^2$, quasi-uniform, and has the following asymptotic separation property:
\begin{equation}
\lim_{N\to \infty} \sqrt{N}\delta(\omega_N) = \sqrt{8-4\sqrt{3}\cos(\sqrt{2\pi}(1-\sqrt{3}))}\approx 3.131948....
\label{spiralsep}
\end{equation}
\end{theorem}
\begin{figure}
\centering
\includegraphics{Spiral.png}
\caption{Plot of $N=700$ generalized spiral points and their Voronoi decomposition.}
\label{spiral}
\end{figure}
As shown in the proof of Theorem \ref{spiral distribution}, the Voronoi cells of $\omega_n$ are asymptotically equal area, but do not approach regular hexagons. Indeed, a typical decomposition is shown in Figure \ref{spiral}. A comparison of the mesh ratios for several values of $N$ is shown in Table \ref{tab:spiral}. Numerically, the mesh ratio appears to converge to 0.8099....
\begin{table}[h!]
\begin{center}
\caption{Mesh Ratios for Generalized Spiral Nodes}
\begin{tabular}{|c|c||c|c||c|c|}
\hline
$N$ & $\gamma(\omega_N)$ & $N$ & $\gamma(\omega_N)$& $N$ & $\gamma(\omega_N)$ \\
\hline\hline
10 & 0.897131& 400& 0.816007 & 20000& 0.809510\\
20 & 0.827821& 500& 0.810128 & 30000& 0.809629\\
30 & 0.814383& 1000& 0.805465& 40000& 0.809689\\
40 & 0.826281& 2000& 0.806411& 50000& 0.809725\\
50 & 0.834799& 3000& 0.807510& 100000& 0.809797\\
100& 0.803901& 4000& 0.808077& 200000& 0.809832\\
200& 0.806020& 5000& 0.808435& 300000& 0.809844\\
300& 0.809226& 10000& 0.809151& 500000& 0.809854\\
\hline
\end{tabular}
\label{tab:spiral}
\end{center}
\end{table}
\vspace{2cm}
\noindent\textbf{Fibonacci Nodes}
\
Another set of spiral points is modeled after nodes appearing in nature such as the seed distribution on the head of a sunflower or a pine cone, a phenomenon known as spiral phyllotaxis \cite{Dixon}. Coxeter \cite{Coxeter} demonstrated these arrangements are fundamentally related to the Fibonacci sequence, $\{F_k\} = \{1,1,2,3,5,8,13,...\}$ and the golden ratio $\varphi = (1+\sqrt{5})/2$. There are two similar definitions of the spherical point set in the literature. Both are defined as lattices on the square $[0,1)^2$ and then mapped to the sphere by the Lambert cylindrical equal area projection, denoted by $\Lambda$. In Cartesian coordinates, $\Lambda$ is defined by
\[\Lambda(x,y):=(\sqrt{1-(2y-1)^2}\cos 2\pi x,\sqrt{1-(2y-1)^2}\sin 2\pi x,2y-1)\]
and, in spherical coordinates, by
\[\Lambda(x,y):=(\cos^{-1}(2y-1),2\pi x) = (\phi,\theta).\]
Define a rational lattice on $[0,1)^2$, with total points $F_k$ by
\begin{equation}
\widetilde{\omega}_{F_k} := \bigg(\bigg\{\frac{iF_{k-1}}{F_k}\bigg\},\frac{i}{F_k}\bigg),\ \ \ 0\leq i\leq F_k,
\label{FibLattice}
\end{equation}
\noindent where $\left\{x\right\} = x - \lfloor{x} \rfloor $ denotes the fractional part of $x$. On the other hand, an irrational lattice can be formed similarly for all values of total points $N$ by replacing $F_{k-1}/F_{k}$ in (\ref{FibLattice}) by $\lim_{k\rightarrow\infty}F_{k-1}/F_{k} = \varphi^{-1}$:
\[\widetilde{\omega}_{N} := \bigg(\left\{i\varphi^{-1}\right\},\frac{i}{N}\bigg),\ \ \ 0\leq i\leq N.\]
\begin{figure}
\centering
\includegraphics{Fibonacci.png}
\caption{Fibonacci nodes for $N=1001$ as defined by Swinbank and Purser and corresponding Voronoi decomposition. The visible spirals at each level are generated by a sequence of points whose index increases arithmetically by a Fibonacci number.}
\end{figure}
Swinbank and Purser \cite{SwinPurs} define a spherical point set for all odd integers $2N+1$ symmetrically across the equator derived from the irrational lattice with points shifted a half step away from the poles:
\[\ \theta_i = 2\pi i \varphi^{-1}, \ \ \sin \phi_i = \frac{2i}{2N+1},\ \ \ -N\leq i \leq N, \ \ \ -\pi/2 \leq \phi_i \leq \pi/2.\]
Denote $\omega_{2N+1}$ as the configuration generated above. Whereas for large $N$, the generalized spiral points tend towards flattening out and partitioning the sphere into distinct regions of latitude, the Fibonacci points maintain visible clockwise and counterclockwise spirals as $N$ grows. Labeling the points of $\omega_{2N+1}$ by increasing latitude, the dominant spirals emanating from $x_i\in \omega_{2N+1}$ are formed by the sequence $\left\{x_{i+jF_k}\right\}$ for some Fibonacci number $F_k$ and $j=\cdots,-2,-1,0,1,2,\cdots$.
The Fibonacci points derived from the rational lattice are studied by Aistleitner et al \cite{ABD12} and Bilyk et al \cite{Bilyk} for discrepancy estimates. In \cite{ABD12}, the spherical cap discrepancy of the points $\left\{F_k\right\}$ is bounded by
\begin{figure}
\noindent\makebox[\textwidth][c]{
\resizebox{10cm}{!}{%
\begin{tikzpicture}
\node at (3,-1.1) [font=\bf] {y$_i$};
\node at (1.5,-.15) [font=\bf] {b$_2$};
\node at (5.5,-.35) [font=\bf] {b$_1$};
\node at (3.45,0.7) [font=\bf] {b$_5$};
\node at (3.7,0) [font=\bf] {b$_3$};
\node at (2.3,0.6) [font=\bf] {b$_4$};
\node at (2.7,2.3) [font=\bf] {b$_6$};
\draw (-4,-3)--(9,-3);
\draw (-4,-2.8)--(-4,-3.2);
\node at (-4,-3.5) {$0$};
\draw (9,-2.8)--(9,-3.2);
\node at (9,-3.5) {$2\pi$};
\draw (2.96,-2)--(7.63,-2);
\node at (5.5, -2.35) {$2\pi\varphi^{-1}$};
\draw (2.96,-1.85)--(2.96,-2.15);
\draw (7.63,-1.85)--(7.63,-2.15);
\clip (-4,-2) rectangle (9cm,6cm);
\pgftransformcm{.8506}{.5257}{-.6243}{.9056}{\pgfpoint{0cm}{0cm}}
\coordinate (Bzero) at (2,-2);
\coordinate (Bone) at (6,-4);
\coordinate (Btwo) at (0,0);
\coordinate (Bthree) at (4,-2);
\coordinate (Bfour) at (2,0);
\coordinate (Bfive) at (4,0);
\coordinate (Bsix) at (4,2);
\foreach \x in {-6,-5,...,6}{
\foreach \y in {-6,-5,...,6}{
\node[draw,circle,inner sep=1.5pt] at (2*\x,2*\y) {};
}
}
\draw [ultra thick,-latex,black] (Bzero)
-- (Bone) node [above, font=\bf] {y$_{i+1}$};
\draw [ultra thick,-latex,black] (Bzero)
-- (Btwo) node [above, font=\bf] {y$_{i+2}$};
\draw [ultra thick,-latex,black] (Bzero)
-- (Bthree) node [above, font=\bf] {y$_{i+3}$};
\draw [ultra thick,-latex,black] (Bzero)
-- (Bfour) node [above, font=\bf] {y$_{i+5}$};
\draw [ultra thick,-latex,black] (Bzero)
-- (Bfive) node [above, font=\bf] {y$_{i+8}$};
\draw [ultra thick,-latex,black] (Bzero)
-- (Bsix) node [above, font=\bf] {y$_{i+13}$};
\end{tikzpicture}}}
\caption{Irrational lattice points on the square with labeled basis vectors from \cite{SwinPurs} around a node $\textbf{y}_i$. The configuration is approximately a rotation of the rectangular lattice by the angle $\tan^{-1}(\varphi)$. Each line of points along a basis vector is mapped to a spiral under the Lambert projection, $\Lambda$. The basis vector $\textbf{b}_k$ such that $\Lambda(\textbf{b}_k)$ has the shortest length forms the visible dominant spirals emanating from $\Lambda(\textbf{y}_i)$. These shortest length vectors are determined by the total number of points and zone number~$z$.}
\label{fig:Fibbasis}
\end{figure}
\[
D_C(\Lambda(\widetilde{\omega}_{F_k})) \leq \left\{
\begin{array}{lr}
44\sqrt{(2/F_k)}& \ \textup{if \textit{k} is odd,}\ \\
44\sqrt{(8/F_k)}& \ \textup{if \textit{k} is even.}
\end{array} \right.
\]
\noindent Numerical experiments in \cite{ABD12} suggest that in fact,
\[ D_C(\Lambda(\widetilde{\omega}_{F_k})) = O\bigg(\frac{(\log F_k)^c}{F_k^{3/4}}\bigg),\ \ \ \ k\rightarrow\infty \ \ \ \text{for some}\ \ 1/2\leq c \leq 1.\]
which is optimal up to a log power \cite{Beck}. Both sequences of Fibonacci configurations are equidistributed. However, since the Swinbank and Purser nodes are defined for more values of total points, we will take these to be the Fibonacci sets moving forward. In \cite{SwinPurs}, these points are also numerically shown to be asymptotically equal area.
Analyzing $\omega_{2N+1}$ as a shifted irrational lattice mapped by the Lambert projection helps to visualize the underlying spiral structure. Define a system of basis vectors
\[ \textbf{b}_k = \Lambda^{-1}(x_{i+F_k})-\Lambda^{-1}(x_i), \ \ \ \ k=1,2,3...,\]
which are independent of base point $x_i$. This is illustrated in Figure \ref{fig:Fibbasis}. Emanating from each point $x_i$, the line of points $\left\{x_i+m\textbf{b}_k\right\}_{m=\dots ,-1,0,1,\cdots}$ is mapped to a spiral on $\mathbb{S}^2$ under the Lambert projection. Like the Fibonacci sequence, the basis vectors satisfy
\[\textbf{b}_{k+1} = \textbf{b}_k + \textbf{b}_{k-1}.\]
On the sphere, the basis vectors in terms of the local Cartesian coordinate system at a point $(\phi_i,\theta_i)\in\omega_{2N+1}$ have the form
\begin{equation}
\textbf{c}_{k,i} = \bigg( (-1)^k2\pi\cos\phi_i\varphi^{-k}, \frac{2F_k}{(2N+1)\cos\phi_i}\bigg).
\label{Fib basis}
\end{equation}
For a fixed latitude $\phi$ and total number of points $2N+1$, the zone number $z$ is defined by
\[\varphi^{2z} = (2N+1)\pi\sqrt{5}\cos^2\phi.\]
Letting $d = \sqrt{4\pi/(\sqrt{5}(2N+1))}$ and using the fact that for large $k$, $F_k \approx \varphi^k/\sqrt{5}$ equation (\ref{Fib basis}) can be rewritten
\begin{equation}
\label{Fib basis2}
\textbf{c}_{k,i} \approx d((-1)^k\varphi^{z-k}, \varphi^{k-z}).
\end{equation}
\begin{figure}
\centering
\includegraphics{FibTri.png}
\caption{Triangulation of $N=3001$ nodes viewed at a slightly different angle. The enlarged box demonstrates where zone number $z$ changes with changing $\phi$. The sudden shift occurs at $z=k\pm1/2$. Along this latitude, $\textbf{c}_{k,i}$ and $\textbf{c}_{k\pm 1,i}$ have equal lengths. For $k-1/2< z< k+1/2$, the triangulation consists of basis vectors $\textbf{c}_{k,i}$ and $\textbf{c}_{k+1,i}$ as proven in \cite{SwinPurs}.}
\label{fig:Fib tri}
\end{figure}
\noindent For latitudes where $k-1/2\leq z\leq k+1/2$, $\textbf{c}_{k,i}$ has the minimum length of the basis vectors around $x_i$ and forms the dominant spiral at those latitudes. As shown in \cite{SwinPurs}, $|\textbf{c}_{k,i}|$ is also the smallest distance between points near these latitudes. Thus, the Delaunay triangulation \cite{Delaunay} of $\omega_{2N+1}$ is composed of $\textbf{c}_{k,i}, \textbf{c}_{k-1,i}$, and $\textbf{c}_{k+1,i}$ when $k-1/2\leq z \leq k+1/2$. This is shown in Figure \ref{fig:Fib tri} and allows us to prove quasi-uniformity.
\begin{proposition}
\label{Fib mesh ratio}
The sequence of Fibonacci configurations is quasi-uniform.
\end{proposition}
Numerically, the minimal separation appears to occur at the pole with value $|x_1-x_4| = |x_{2N+1}-x_{2N-2}|$ and the largest hole appears to occur in the triangles covering the poles, $\triangle x_2,x_3,x_5$ and $\triangle x_{2N},x_{2N-1},x_{2N-3}$. In a straightforward computation, it can be shown that
\[\lim_{N\to\infty} \sqrt{2N+1}|x_1-x_4| = \sqrt{16-\sqrt{112}\cos(6\pi\varphi^{-1})} = 3.09207...\]
and the circumradius $r$ of the polar triangles satisfies
\[\lim_{N\to\infty}\sqrt{2N+1}\,r = 2.72812....\]
As shown in Table \ref{tab:Fib}, the mesh ratios for Fibonacci nodes appear to converge quickly to this ratio $\approx 0.882298$.
\begin{table}[h!]
\begin{center}
\caption{Mesh Ratios for Fibonacci Nodes}
\begin{tabular}{|c|c||c|c||c|c|}
\hline
$N$ & $\gamma(\omega_N)$ & $N$ & $\gamma(\omega_N)$& $N$ & $\gamma(\omega_N)$ \\
\hline\hline
11 & 0.859197& 401& 0.881897 & 20001& 0.882289\\
21 & 0.872632& 501& 0.881978 & 30001& 0.882292\\
31 & 0.876251& 1001& 0.882139& 40001& 0.882293\\
41 & 0.877909& 2001& 0.882218& 50001& 0.882294\\
51 & 0.878857& 3001& 0.882244& 100001& 0.882296\\
101& 0.880646& 4001& 0.882258& 200001& 0.882297\\
201& 0.881489& 5001& 0.882266& 300001& 0.882297\\
301& 0.881762& 10001& 0.882282& 500001& 0.882297\\
\hline
\end{tabular}
\label{tab:Fib}
\end{center}
\end{table}
\
\noindent \textbf{Low Discrepancy Nodes}
\
Another approach for distributing points on the sphere is to minimize a suitable notion of discrepancy, such as spherical cap, $L_p$, or generalized discrepancy (cf. \cite{BD15} and \cite{CF97}). A low spherical cap discrepancy sequence $\left\{\omega_N\right\}_{N=2}^{\infty}$ satisfies \cite{Beck}
\begin{equation}
\frac{a}{N^{3/4}}\leq D_C(\omega_N)\leq A\frac{\sqrt{\log N}}{N^{3/4}}, \ \ \ \ \ \ N\geq 2,
\end{equation}
\noindent for some $a,A>0$. Low discrepancy point sets are used in Quasi-Monte Carlo methods for numerical integration and also in graphics applications in \cite{WongLH}.
One method for generating spherical nodes is to first distribute points on the square~$[0,1)^2$ with low planar discrepancy \cite{Nied92}, i.e. for some $A>0$
\begin{equation}
D(\omega_N):=\sup_{R}\bigg|\frac{R\cap\omega_N}{N}-\sigma(R)\bigg|\leq A\frac{\log N}{N},\ \ \ \ \omega_N\subset[0,1)^2,
\label{disc}
\end{equation}
\noindent where the supremum $R$ is taken over all rectangles with sides parallel to the axes. These sequences are then mapped to the sphere via the Lambert projection. While the $\log N/N$ term in (\ref{disc}) is optimal on the plane \cite{Schmidt}, it is an open problem whether the images of these sequences have optimal order spherical cap discrepancy. There are several such node distributions in the literature (cf. \cite{ABD12} and \cite{CF97}), but as their properties are similar, we only consider the following one.
\
\noindent \underline{Hammersley Nodes}
\
For an integer $p \geq 2$, the $p$-adic van der Corput sequence is defined by
\[x^{(p)}_k = \frac{a_0}{p} + \dots + \frac{a_r}{p^{r+1}},\ \ \ \ \text{where}\ \ \ \ k = a_0 + \dots +a_rp^r,\ \ \ a_i\in\left\{0,1\right\}.\]
The Hammerlsy node set on the square (\cite{CF97}, \cite{Nied92}, and \cite{WongLH}) is given by $x_k := x^{(2)}_k$ and $y_k := \frac{2k-1}{2N}$. The $N$ point spherical Hammersley node set is given by $\{\Lambda(2\pi x_k,1-2y_k)\}^N_{k=1}$. The configuration for $N=1000$ is given in Figure \ref{Hamfig}. The discrepancy of the planar Hammersley nodes is known from Niederreiter~\cite{Nied92} to satisfy (\ref{disc}) . The sequence of Hammersley configurations is equidistributed; however it is not well-separated or quasi-uniform. This makes the nodes poor candidates for energy, as shown in Section 3. Their Voronoi decompositions also exhibit no discernible geometric patterns.
\begin{figure}
\centering
\includegraphics{Hammersley.png}
\caption{Hammersley nodes for $N=1000$ and corresponding Voronoi decomposition.}
\label{Hamfig}
\end{figure}
\
\noindent\textbf{Equal Area Partitions}
\
Another class of point sets are those derived from equal area partitions of the sphere.
\vspace{2cm}
\noindent\underline{Zonal Equal Area Nodes}:
\
Rakhmanov et al \cite{RSZ94} construct a diameter bounded, equal area partition of $\mathbb{S}^2$ into rectilinear cells of the form
\[R([\tau_\phi,\nu_\phi]\times [\tau_\theta,\nu_\theta]):= \left\{(\phi,\theta)\in\mathbb{S}^2\ :\ \tau_\phi\leq\phi\leq\nu_\phi, \tau_\theta\leq\theta\leq\nu_\theta\right\}.\]
The cells are grouped by regions of equal latitude called collars that have the form $R([\tau_\phi,\nu_\phi]\times [0,2\pi])$. The cells are defined such that $\nu_\phi-\tau_\phi$ and $\nu_\theta-\tau_\theta$ approximate $\sqrt{4\pi/N}$ as $N$ grows. This ensures the correct order of the diameter bound. The cells are defined in the following way.
\
1. \textit{Determine the latitudes of the polar caps.} The first two cells are taken to be the polar caps of radius $\phi_c = \cos^{-1}(1-2/N).$
\
2. \textit{Determine the ideal collar angle and ideal number of collars.} The ideal angle between two collars is
\[\delta_I := \sqrt{4\pi/N}.\]
The ideal number of collars between the polar caps, all of which have angle $\delta_I$, is
\[n_I:=\frac{\pi-2\phi_c}{\delta_I}.\]
\begin{figure}
\label{ZonalPartition}
\centering
\includegraphics{ZonalPartition.png}
\caption{Zonal equal area partition of the sphere into 100 cells. The algorithm does not determine a unique partition as the collars can rotate past each other. Image created with the help of Paul Leopardi's equal area partitioning toolbox available at eqsp.sourceforge.net.}
\end{figure}
3. \textit{Determine the actual number of collars, $n$.} If $N = 2,$ then $n := 0$. Otherwise,
\[n := \max\left\{1,\text{round}(n_I)\right\}.\]
4. \textit{Create a list of the ideal number of cells in each collar.} The ``fitting" collar angle is
\[\delta_F := \frac{n_I}{n}\delta_I = \frac{\pi-2\phi_c}{n}.\]
Label the collars $\left\{C_j\right\}_{j=1}^{n+2}$ southward with the North polar cap as $C_1$ and the South polar cap as $C_{n+2}$. The area $A_j$ of collar $C_j$ can be written as the difference of polar cap areas:
\[A_j = 2\pi(\cos(\phi_c+(j-2)\delta_F)-\cos(\phi_c+(j-1)\delta_F)).\]
Thus the ideal number of cells $y_{j,I}$ in each collar $C_j$, $j\in\left\{2,\dots,n+1\right\}$, is given by
\[y_{j,I} = \frac{4\pi A_j}{N}.\]
5. \textit{Create a list of the actual number of cells in each collar.} We apply a cumulative rounding procedure. Letting $y_j$ be the number of cells in $C_j$, define the sequences $y$ and $a$ by $a_1:=0$, $y_1:=1$, and for $j\in \left\{2,\dots,n+1\right\}$:
\[y_j:=\text{round}(y_{j,I}+a_{j-1}),\ \ \ \ \ \ a_j:=\sum_{k=1}^{j}y_k-y_{k,I}.\]
6. \textit{Create a list of latitudes $\phi_j$ of each collar and partition each collar into cells.} We define $\phi_j$ as follows: $\phi_0 = 0$, $\phi_{n+2} = \pi$ and for $j\in \left\{1,\dots,n+1\right\}$,
\[\phi_j = \cos^{-1}(1-\frac{2}{N}\sum_{k=1}^j y_k).\]
Thus the North polar cap of radius $\phi_j$ has normalized area $\sum_{k=1}^j y_k/N$, and $C_j:=R([\phi_{j-1},\phi_j]\times[0,2\pi])$.
\
7. \textit{Partition each collar into cells.} $C_j$ has $y_j$ equal cells \[\left\{R([\phi_{j-1},\phi_j]\times[\theta_j+k\frac{y_j}{2\pi},\theta_j+(k+1)\frac{y_j}{2\pi}])\right\}_{k=0}^{y_j-1},\]
where $\theta_j\in[0,2\pi)$ can be chosen to be any starting angle. Note that because $\theta_j$ are chosen independently, the equal area partition determined by the algorithm is not unique. Indeed, the collars can rotate past each other without affecting the diameter bound or equal area property of the partition. Because the choice of the $\theta_j$'s does not strongly affect the other properties with which we are concerned, we will take them to be random.
\begin{figure}
\label{zonal}
\centering
\includegraphics{Zonal.png}
\caption{Zonal equal area points for $N=700$ and corresponding Voronoi decomposition. The configurations have a similar structure to the generalized spiral points in Figure \ref{spiral}.}
\end{figure}
The point set $\omega_N$ is defined to be the centers of the cells of the rectilinear partition. As proved by Zhou \cite{Zhou}, the cells are diameter bounded from above by $7/\sqrt{N}$; however, numerical experiments from Leopardi in \cite{Leopardi} suggest the bound to be $6.5/\sqrt{N}$. For large $N$, the zonal equal area configurations look very similar to the generalized spiral configurations. Namely they exhibit iso-latitudinal rings with separation between adjacent points equal to separation between rings and a random longitudinal shift between points in adjacent rings. As shown in Section~3, the energy computations for both point sets are nearly identical.
\begin{proposition}
The sequence of zonal equal area configurations is equidistributed and quasi-uniform.
\label{zonal quasi}
\end{proposition}
The above construction was modified by Bondarenko et al \cite{BondVia} to create a partition with geodesic boundaries for the creation of well-separated spherical designs. More details can be found in \cite{BondVia}. Table \ref{tab:Zonal} gives a comparison of the mesh ratios of the zonal points.
\begin{table}[h!]
\begin{center}
\caption{Mesh Ratios for Zonal Equal Area Nodes}
\begin{tabular}{|c|c||c|c||c|c|}
\hline
$N$ & $\gamma(\omega_N)$ & $N$ & $\gamma(\omega_N)$& $N$ & $\gamma(\omega_N)$ \\
\hline\hline
10 & 0.711934& 400& 0.769527 & 20000& 0.758100\\
20 & 0.790937& 500& 0.766808 & 30000& 0.758069\\
30 & 0.788546& 1000& 0.765356& 40000& 0.756793\\
40 & 0.843385& 2000& 0.764631& 50000& 0.756785\\
50 & 0.790252& 3000& 0.758645& 100000& 0.756770\\
100& 0.761296& 4000& 0.756510& 200000& 0.756762\\
200& 0.764846& 5000& 0.764217& 300000& 0.758015\\
300& 0.763188& 10000& 0.758192& 500000& 0.756757\\
\hline
\end{tabular}
\label{tab:Zonal}
\end{center}
\end{table}
\
\noindent\underline{HEALPix Nodes}
\
\begin{figure}
\centering
\includegraphics{HEALPixBase.png}
\caption{Base tessellation of the sphere into 12 equal area pixels. A finer mesh is created by dividing each base pixel into a $k\times k$ grid of equal area subpixels of the same shape. The HEALPix points are taken to be the centers of the pixels.}
\label{HEALPix}
\end{figure}
Developed by NASA for fast data analysis of the cosmic microwave background (CMB), the Hierarchical Equal Area iso-Latitude Pixelization (HEALPix) was designed to have three properties essential for computational efficiency in discretizing functions on $\mathbb{S}^2$ and processing large amounts of data \cite{HEALPix}:
\begin{enumerate}
\item The sphere is hierarchically tessellated into curvilinear quadrilaterals.
\item The pixelization is an equal area partition of $\mathbb{S}^2$.
\item The point sets are distributed along fixed lines of latitude.
\end{enumerate}
To create the partition of $\mathbb{S}^2$, the authors in \cite{HEALPix} first divide the sphere into~12 equal area, four sided pixels defined by the following boundaries:
\[|\cos\phi|>\frac{2}{3}, \ \ \ \ \ \theta=m\frac{\pi}{2}, \ \ \ m = 0,1,2,3\]
\[\cos\phi=\frac{-2-4m}{3}+\frac{8\theta}{3\pi},\ \ \ \ \ \frac{m\pi}{2}\leq\theta\leq\frac{(m+1)\pi}{2}, \ \ \ \ \ m=0,1,2,3\]
\[\cos\phi=\frac{2-4m}{3}-\frac{8\theta}{3\pi},\ \ \ \ \ \frac{-(m+1)\pi}{2}\leq\theta\leq\frac{-m\pi}{2}, \ \ \ \ \ m=0,1,2,3.\]
The base tessellation is shown in Figure \ref{HEALPix}. For a given $k\in \mathbb{N}$, each pixel is partitioned into a $k \times k$ grid of sub-pixels of the same shape and equal area. The HEALPix point sets are taken to be the centers of these pixels.
On the polar regions $|\cos\phi|>2/3$, the points are distributed along $k$ iso-latitudinal rings, indexed by $i$, each with $4i$ equally spaced points, indexed by~$j$:
\[|\cos\phi_i| = 1 - \frac{i^2}{3k^2}, \ \ \ \ \ \ \ \ \ \theta_j = \frac{\pi}{2i}\bigg(j-\frac{1}{2}\bigg).\]
On the equatorial region, there are $2k-1$ iso-latitudinal rings, each with $4k$ points. The rings are indexed by $k\leq |i|\leq 2k$ and the points by $1\leq j \leq 4k$:
\[|\cos\phi_i| = \frac{4}{3} - \frac{2i}{3k},\]
\[\theta_j = \frac{\pi}{2k}\bigg(j-\frac{s}{2}\bigg), \ \ \ \ \ s = (i-k+1)\mod{2}.\]
The index $s$ describes the phase shift between rings. This gives a configuration of size $N=12k^2$. The point sets are hierarchical along the subsequence $k=3^m$. Holho\c s and Ro\c sca \cite{HR15} have shown that the HEALPix points can be obtained as the image of points on a certain convex polyherdon under an area preserving mapping to the sphere.
\begin{figure}
\centering
\includegraphics{HEALPix.png}
\caption{HEALPix nodes and Voronoi decomposition for $N=1200, k=10$.}
\end{figure}
\begin{proposition}
The sequence of HEALPix configurations is equidistributed and quasi-uniform.
\label{HEALPix quasi}
\end{proposition}
\noindent Numerically, the mesh ratio appears to be bounded by 1, as shown in Table \ref{tab:HEALPix}.
\begin{table}[h!]
\begin{center}
\caption{Mesh Ratios for HEALPix nodes}
\begin{tabular}{|c|c|c||c|c|c||c|c|c|}
\hline
$k$ & $N$ & $\gamma(\omega_N)$ &$k$ & $N$ & $\gamma(\omega_N)$ &$k$ & $N$ & $\gamma(\omega_N)$\\
\hline\hline
1 & 12 & 0.864783 & 9& 972& 0.965950& 45& 24300& 0.992956\\
2& 48& 0.862243& 10& 1200& 0.969599& 50& 30000& 0.993648\\
3& 108& 0.909698 & 15& 2700& 0.979371& 60& 43200& 0.994701\\
4& 192& 0.929080& 20& 4800& 0.984328& 70& 58800& 0.995456\\
5& 300& 0.940016& 25& 7500& 0.987365& 80& 76800& 0.996020\\
6& 432& 0.951047& 30& 10800& 0.989509& 90& 97200& 0.996455\\
7& 588& 0.957584& 35& 14700& 0.990959& 100& 120000& 0.996807\\
8& 768& 0.961782& 40& 19200& 0.992082& 150& 270000& 0.997867\\
\hline
\end{tabular}
\label{tab:HEALPix}
\end{center}
\end{table}
\
\noindent\textbf{Polyhedral Nodes and Area Preserving Maps}
\
Another class of point sets are those derived from subdividing regular polyhedra and applying radial projection: $\Pi(\textbf{x}) = \textbf{x}/\|\textbf{x}\|$ or an equal area projection. These node sets are used in finite element methods to give low error solutions to boundary value problems. See, for instance, \cite{GHWIcos} and \cite{NTLCube}.
\
\noindent\underline{Radial Icosahedral Nodes}
\
This point set, as described in \cite{Teanby} and \cite{Wright} is formed by overlaying a regular triangular lattice onto each face of a regular icosahedron of circumradius $1$ and edge length $a = \csc(2\pi/5)$. Given $k\in \mathbb{N}$, for each vertex $v$, divide two adjacent edges emanating from $v$ into basis vectors of length $a/k$. For the face $\mathcal{F}$ determined by these edges and vertex, the icosahedral point set $\widetilde{\omega_{N_k}}$ on $\mathcal{F}$ is taken to be the set of lattice points generated by these basis vectors restricted to $\mathcal{F}$. The spherical points are $\omega_{N_k}:=\Pi(\widetilde{\omega_{N_k}})$. These node sets are defined for total points $N = 10k^2+2$ and hierarchical along the subsequence $k_j = k_02^j$ for any $k_0\in \mathbb{N}$.
The sequence of icosahedral configurations $\left\{\widetilde{\omega_{N_k}}\right\}_{k=1}^{\infty}$ is equidistributed. However, because radial projection is not area preserving, the sequence of spherical configurations is not equidistributed. Density is higher towards the vertices of the icosahedron and lower towards the center of the faces where the areal distortion of $\Pi$ is greatest. The Voronoi decomposition of $\omega_{N_k}$ is composed of twelve regular pentagons with all other cells regular hexagons of varying size as illustrated in Figure \ref{RadIcos}.
\begin{proposition}
\label{icosahedral distribution}
The sequence of radial icosahedral configurations is quasi-uniform.
\label{rad icos quasi}
\end{proposition}
\noindent Numerically, the mesh ratio appears to be bounded by $0.86$, as shown in Table~\ref{tab:RadIcos}.
\begin{figure}
\centering
\includegraphics{Icos.png}
\caption{Radial icosahedral odes $N=642$. The Voronoi decomposition is composed of regular hexagons of varying size and 12 pentagons at the vertices of the icosahedron.}
\label{RadIcos}
\end{figure}
\begin{table}[h!]
\begin{center}
\caption{Mesh Ratios for Radial Icosahedral Nodes}
\begin{tabular}{|c|c|c|c|c|c|}
\hline
$k$ & $N$ & $\gamma(\omega_N)$& $k$ & $N$ & $\gamma(\omega_N)$\\
\hline\hline
1& 12& 0.620429 & 20& 4002& 0.830750\\
2& 42& 0.667597 & 30& 9002& 0.838066\\
3& 92& 0.684698 & 40& 16002& 0.842358\\
4& 162& 0.745348& 50& 25002& 0.844697\\
5& 252& 0.765157 & 60 & 36002& 0.846156\\
6& 362& 0.769854& 70 & 49002 & 0.847376\\
7& 492& 0.789179& 100& 100002 & 0.849390\\
10& 1002& 0.808024& 150& 225002 & 0.850941\\
15& 2252& 0.821504& 200& 400002 & 0.851745\\
\hline
\end{tabular}
\label{tab:RadIcos}
\end{center}
\end{table}
\vspace{2cm}
\noindent\underline{Cubed Sphere Nodes}
\
A similar method as above can be applied to the cube \cite{NTLCube}. A square $k\times k$ grid is placed on each face of the cube and radially projected to the sphere. A typical point set is shown in Figure \ref{Cubed Sphere}. The configurations are defined for $N = 6k^2 -12k +8$ and are hierarchical along the subsequence $k=k_02^m$. By an argument similar to that in Proposition \ref{icosahedral distribution}, the limiting distribution is not uniform, but the sequence of configurations is quasi-uniform. Numerically, the mesh ratio seems to quickly converge to 1, as shown in Table \ref{tab:Cube}.
\begin{figure}
\centering
\includegraphics{CubedSph.png}
\caption{Plot of $N=1016$ Cubed Sphere Points and Voronoi Decomposition. The Voronoi cells tend towards regular hexagons near the vertices of the cube. Towards the middle of each face they resemble a square lattice.}
\label{Cubed Sphere}
\end{figure}
\begin{table}[h!]
\begin{center}
\caption{Mesh Ratios for Cubed Sphere Points}
\begin{tabular}{|c|c|c||c|c|c||c|c|c|}
\hline
$k$ & $N$ & $\gamma(\omega_N)$ &$k$ & $N$ & $\gamma(\omega_N)$ &$k$ & $N$ & $\gamma(\omega_N)$\\
\hline\hline
2& 8& 0.827329& 10& 488& 0.996846& 50& 14408& 0.999893\\
3& 26& 0.794265& 15& 1178& 0.994025& 60& 20888& 0.999926\\
4& 56& 0.972885& 20& 2168& 0.999289& 70& 28568& 0.999946\\
5& 98& 0.933655& 25& 3458& 0.997954& 80& 37448& 0.999959\\
6& 152& 0.989913& 30& 5048& 0.999695& 90& 47528& 0.999968\\
7& 218& 0.968757& 35& 6938& 0.998979& 100& 58808& 0.999974\\
8& 296& 0.994805& 40& 9128& 0.999831& 150& 133208& 0.999988\\
9& 386& 0.982046& 45& 11618& 0.999390 & 200& 237608& 0.999994\\
\hline
\end{tabular}
\label{tab:Cube}
\end{center}
\end{table}
\noindent\underline{Octahedral Points}
\
Unlike in the previous examples, the octahedral points, described by Holho\c s and Ro\c sca \cite{Oct}, are derived from an area preserving map $\mathcal{U}$ from the regular octahedron $\mathbb{K}$ of edge length $L = \sqrt{2\pi}/\sqrt[4]{3}$ and surface area $4\pi$ to $\mathbb{S}^2$. Let $\mathcal{U}_x, \mathcal{U}_y, $ and $\mathcal{U}_z$ denote the $x$,$y$, and $z$ components of $\mathcal{U}$ respectively. For $(X,Y,Z)\in\mathbb{K}$,
\[ \mathcal{U}_z = \frac{2Z}{L^2}(\sqrt{2}L-|Z|),\]
\[\mathcal{U}_x = \text{sgn}(X)\sqrt{1-\mathcal{U}_z^2}\cos\frac{\pi |Y|}{2(|X|+|Y|)},\]
\[\mathcal{U}_y = \text{sgn}(Y)\sqrt{1-\mathcal{U}_z^2}\sin\frac{\pi |Y|}{2(|X|+|Y|)}.\]
To produce a spherical point set, the authors form a partition $P_k$ of $k^2$ triangles on each face of the octahedron in the same manner as the radial icosahedral points and obtain an equal area spherical partition $\mathcal{P}_k = \mathcal{U}(P_k)$. The point sets~$\omega_{N_k}$ are taken to be the vertices of the triangles of $\mathcal{P}_k$. For a given~$k$, there are~$8k^2$ triangles and $N=4k^2+2$ points.
\begin{figure}
\centering
\includegraphics{Oct.png}
\caption{Equal area octahedral points for $k=15$ and $N=902$. The Voronoi decomposition of the octahedral points is composed of hexagons and 8 squares at the vertices of the octahedron. The hexagons approach regularity towards the center of the faces and deform along the edges.}
\end{figure}
The octahedral configurations have similar properties to the HEALPix node sets. They are iso-latitudinal and hierarchical along the subsequence $k=k_02^m$. The sequence of configurations is equidistributed, and in addition, the authors in \cite{Oct} compute a diameter bound for any triangular region $\mathcal{T}$ of $\mathcal{P}_k$ to be
\[\textup{diam}\ \mathcal{T}\leq \frac{2\sqrt{4+\pi^2}}{\sqrt{8k^2}}\approx \frac{7.448}{\sqrt{8k^2}}.\] Following their proof of this bound, we can calculate a lower bound on the separation and an upper bound on the mesh norm.
\begin{theorem}
\label{octahedral distribution}
The sequence of equal area octahedral configurations is quasi-uniform with
\begin{equation}
\gamma(\omega_{4k^2+2})\leq \frac{1}{4} \sqrt{\frac{4+\pi^2}{2-(k+1)^2/k^2}}\rightarrow \frac{\sqrt{4+\pi^2}}{4}\approx 0.931048..., \ \ \ \ k\rightarrow\infty.
\label {Octmesh}
\end{equation}
\label{oct quasi}
\end{theorem}
\noindent This bound seems to be near optimal. As shown in Table \ref{tab:Oct}, the mesh ratio grows to at least $0.9235$.
\
\
\begin{figure}
\noindent\makebox[\textwidth][c]{
\resizebox{10cm}{!}{%
\begin{tikzpicture}
\clip (-4.5,-2) rectangle (17cm,14cm);
\pgftransformcm{1}{0}{.5}{.866}{\pgfpoint{0cm}{0cm}}
\foreach \x in {-25,-24,...,38}{
\foreach \y in {-5,-4,...,40}{
\node[draw,circle,inner sep=1pt, thick] at (1/2*\x,1/2*\y) {};
}
}
\draw [thick] (-1,.5)--(11.5,10.5);
\draw [thick] (-5.5,3)--(7,13);
\draw [thick] (-5.5,3)--(3.5,-2);
\draw [thick] (-7.5,7.5)--(6,0);
\draw [thick] (-5,9.5)--(8.5,2);
\draw [thick] (-2.5,11.5)--(11,4);
\draw [thick] (0,13.5)--(13.5,6);
\draw [thick] (-7.5,7.5)--(-5.5,3);
\draw [thick] (-5,9.5)--(-1,.5);
\draw [thick] (-2.5,11.5)--(3.5,-2);
\draw [thick] (0,13.5)--(6,0);
\draw [thick] (2.5,15.5)--(8.5,2);
\draw [thick] (7,13)--(11,4);
\draw [thick] (11.5,10.5)--(13.5,6);
\draw [thick] (2.5,15.5)--(11.5,10.5);
\draw [red, line width = .1cm] (6,0)--(8.5,0) node [pos = 0.5, scale = 2.5, red, very thick, below]{$m$};
\draw [red, line width = .1cm] (8.5,0)--(8.5,2) node [pos = 0.3, scale = 2.5, red, very thick, right]{$n$};
\draw [red, line width = .1cm] (8,.25)--(8.5,0);
\draw [red, line width = .1cm] (8.25,-.25)--(8.5,0);
\draw [red, line width = .1cm] (8.25,1.75)--(8.5,2);
\draw [red, line width = .1cm] (8.75,1.5)--(8.5,2);
\end{tikzpicture}}}
\caption{Planar icosahedral mesh for $(m,n) = (5,4)$.}
\label{IcosMesh}
\end{figure}
\begin{table}[h!]
\begin{center}
\caption{Mesh Ratios for Octahedral Nodes}
\begin{tabular}{|c|c|c||c|c|c||c|c|c|}
\hline
$k$ & $N$ & $\gamma(\omega_N)$ &$k$ & $N$ & $\gamma(\omega_N)$ &$k$ & $N$ & $\gamma(\omega_N)$\\
\hline\hline
1 & 6 & 0.675511 & 9& 326& 0.873510& 60& 14402& 0.905758\\
2& 18& 0.872884& 10& 402& 0.875606& 70& 19602& 0.908047\\
3& 38& 0.854610& 15& 902& 0.882510& 80& 25602& 0.909875\\
4& 66& 0.856329& 20& 1602& 0.886310& 90& 32402& 0.911382\\
5& 102& 0.860536& 25& 2502& 0.888702& 100& 40002& 0.912644\\
6& 146& 0.864599& 30& 3602& 0.892884& 200& 160002& 0.919218\\
7& 198& 0.868095& 40& 6402& 0.898762& 300& 360002& 0.921947\\
8& 258& 0.871036& 50& 10002& 0.902784& 400& 640002& 0.923503\\
\hline
\end{tabular}
\label{tab:Oct}
\end{center}
\end{table}
\noindent\underline{Mesh Icosahedral Equal Area Points}
\
Improvements to the radially projected icosahedral points have been put forth by Song et al \cite{SongKim} and Tegmark \cite{Tegmark}. Here, we introduce two other improvements to these points to create new configurations. First, we generalize the icosahedral lattice structure to create configurations of more possible numbers of total points. Due to a method of Caspar and Klug \cite{CK62} derived during their investigation of the construction of viruses, we define a triangular lattice on a regular icosahedron with total points
\[N = 10(m^2+mn+n^2) + 2, \ \ \ \ \ \ \ \ (m,n)\in \mathbb{N}\times\mathbb{N}\setminus(0,0).\]
Consider the planar triangular lattice generated by $e_1 = (1,0)$ and $e_2 = (1/2,\sqrt{3}/2)$. For a given $(m,n)$, let $e_{m,n} = me_1+ne_2$ and it's rotation by~$\pi/3$ be basis vectors for an unfolded icosahedron superimposed on the lattice. This is illustrated in Figure \ref{IcosMesh}. Folding the icosahedron results in a triangular lattice~$\widetilde{\omega_{N}}$ on each face. Due to rotational symmetry of the lattice, the resulting configuration is independent of how the icosahedron is unfolded. The subsequence~$(m,0)$ produces the lattice for the radial icosahedral nodes.
\begin{figure}
\noindent\makebox[\textwidth][c]{
\begin{tikzpicture}
\draw (-0.5,-0.5)--(6.5,-0.5)--(3,5.562)--cycle;
\draw (15,0) arc (0:60:6cm);
\draw (12,5.196) arc(120:180:6cm);
\draw (9,0) arc(240:300:6cm);
\draw (5.5,3.5) [->,very thick] arc(110:70:5);
\draw [red, very thick](3,1.732)--(3,0) node [pos = 0.5, red, left]{$h$};
\draw [red, very thick](3,0)--(4.5,0) node [pos = 0.5, red, below]{$w$};
\draw [red, very thick](4.5,0)--(3,1.732);
\draw (3,.3)--(3.3,.3)--(3.3,0);
\draw (4.5,0)--(5.73,0);
\draw [dashed, very thick](3,0)--(3,-0.5);
\draw [dashed, very thick](3,1.732)--(6.5,-0.5);
\node at (4.5,0) {\textbullet};
\node at (4.5,-0.2) {$p$};
\node at (3.5,.6) [red] {$\mathcal{T}$};
\draw (12,1.732) [red, very thick] arc (150:188.5:3cm);
\draw (11.63,-0.2) [dashed, thick] arc (188.5:200:3cm);
\draw (15,0) [dashed, thick] arc (30:90:3.48cm);
\draw (11.63,-0.2) [red,very thick] arc (265:287:5cm);
\draw (13.5,0) arc (287:301.25:5cm);
\draw (13.5,0) [red, very thick] arc (30:52:6cm);
\draw (11.9,1.5) arc (269:287:1cm);
\draw (14.5,0.7) arc (150:169:1cm);
\draw (11.61,0.1)--(11.93,0.1)--(11.93,-.2);
\node at (13.5,0) {\textbullet};
\node at (3,1.732) {\textbullet};
\node at (12,1.732) {\textbullet};
\node at (3,0) {\textbullet};
\node at (11.63,-0.2) {\textbullet};
\node at (5.73,0) {\textbullet};
\node at (14.65,0.48) {\textbullet};
\node at (7.25,4.2) {\Large $\Phi$};
\node at (13.5,-0.3) {$\Phi(p)$};
\node at (12.5,0.5) [red] {$\Phi(\mathcal{T})$};
\node at (3,2) {$\mathcal{O}$};
\node at (12,2) {$\Phi(\mathcal{O})$};
\node at (12.05,1.3) {$\lambda$};
\node at (2.7,0) {$p_{\bar{A}}$};
\node at (11,-0.2) {$\Phi(p_{\bar{A}})$};
\node at (5.7,-0.2) {$p_{\bar{B}}$};
\node at (14.2,0.6) {$\psi$};
\node at (14.35,0) {$\Phi(p_{\bar{B}})$};
\end{tikzpicture}}
\caption{Illustration of area preserving map $\Phi$ defined piecewise on each triangle bounded by two altitudes of a face with relevant variables labeled.}
\label{IcosMap}
\end{figure}
Secondly, we derive an area preserving map $\Phi$ from the regular icosahedron~$\mathcal{I}$ of edge length $L = \sqrt{4\pi}/\sqrt[4]{75}$, circumradius $r = L\sin2\pi/5$, and surface area~$4\pi$ using the technique presented by Snyder \cite{Snyder} for the truncated icosahedron. We define $\Phi$ piecewise by dividing each face $\mathcal{F}\subset \mathcal{I}$ into the six triangles $\mathcal{R}_i$ partitioned by the altitudes of $\mathcal{F}$:
\
1. Parametrize each point $p\in\mathcal{R}_i$ by $h$ and $w$ as labeled in Figure \ref{IcosMap}. If $\bar{A}$ is the side of $\mathcal{R}_i$ of length $L/2\sqrt{3}$, then $w$ is the distance from $p$ to $\bar{A}$ and $h$ is the distance of $p_{\bar{A}}:= proj(p,\bar{A})$ to $\mathcal{O}$, the center of $\mathcal{F}$.
\
2. Let $\bar{B}$ be the side of $\mathcal{R}_i$ of length $L/\sqrt{3}$ and $p_{\bar{B}}$ be the intersection of the line $\overline{p_{\bar{A}}p}$ with $\bar{B}$. For the triangle $\mathcal{S} = \triangle p_{\bar{A}}p_{\bar{B}}\mathcal{O}$, find $\psi$ as in Figure \ref{IcosMap} and spherical right triangle $\Phi(\mathcal{S})$ such that Area$(\mathcal{S}) = \sigma(\Phi(\mathcal{S})) = \psi - \pi/6$. Thus,
\[\psi = \frac{h^2\sqrt{3}}{2}+\frac{\pi}{6}.\]
3. $\Phi(p)$ will lie on the great circle $\Phi(\overline{p_{\bar{A}}p_{\bar{B}}})$. Letting $\mathcal{T} = \triangle p_{\bar{A}},p,\mathcal{O}$, find $\lambda$ as in Figure \ref{IcosMap} such that Area$(\mathcal{T}) = \sigma(\Phi(\mathcal{T}))$. By the spherical law of cosines,
\[\sigma(\Phi(\mathcal{T})) = \lambda + \cos^{-1}\bigg(\frac{2\sin\lambda\cos\psi}{\sqrt{3}}\bigg)-\frac{\pi}{2}.\]
Thus
\begin{figure}
\centering
\includegraphics{MeshIcos.png}
\caption{Equal area mesh icosahedral nodes for $(m,n) = (7,2)$ and total points $N=672$. The Voronoi decomposition forms a spherical Goldberg polyhedron.}
\label{meshicos}
\end{figure}
\noindent \begin{table}[h!]
\begin{center}
\caption{Mesh Ratios for Equal Area Icosahedral Nodes}
\resizebox{\columnwidth}{!}{
\begin{tabular}{|c|c|c||c|c|c||c|c|c|}
\hline
$(m,n)$ & $N$ & $\gamma(\omega_N)$ &$(m,n)$ & $N$ & $\gamma(\omega_N)$ &$(m,n)$ & $N$ & $\gamma(\omega_N)$\\
\hline\hline
(1,0) & 12 & 0.620429 & (7,1)& 572& 0.688031& (37,27)& 30972& 0.730733\\
(1,1)& 32 & 0.617964& (7,5)& 1092& 0.707058& (40,33)& 40092& 0.732326\\
(2,0)& 42& 0.667598&(12,4)& 2082& 0.706688& (42,40)& 50442& 0.732529\\
(2,1)& 72& 0.657081&(16,3)& 3132& 0.704123& (65,50)& 99752& 0.733719\\
(3,0)& 92& 0.659610&(16,7)& 4172& 0.717067& (90,75)& 204752& 0.735013\\
(3,1)& 132& 0.668227&(19,6)& 5112& 0.712681& (100,100)& 300002& 0.735397\\
(4,1)& 212& 0.680153& (19,18)& 10272& 0.726243& (131,100)& 402612& 0.735592\\
(5,2)& 392& 0.687368& (31,21)& 20532& 0.728761& (145,115)& 509252& 0.735965\\
\hline
\end{tabular}}
\label{tab:MeshIcos}
\end{center}
\end{table}
\[\tan\lambda = \frac{\sin(\frac{hw}{2})}{\cos(\frac{hw}{2})-\frac{2\cos\psi}{\sqrt{3}}}.\]
4. Transform $(\psi,\lambda)$ into spherical coordinates.
\
\noindent The map $\Phi$ is extended to $\mathcal{I}$ by rotations and reflections. This defines the unique azimuthal equal area projection from $\mathcal{I}$ onto $\mathbb{S}^2$ \cite{Mapbook}. The spherical configurations are $\omega_N :=\Phi(\widetilde{\omega_{N}})$. A typical point set is shown in Figure \ref{meshicos}. The Voronoi cells are almost regular hexagons with 12 pentagonal cells at the vertices of $\mathcal{I}$, and the Voronoi decomposition forms a spherical Goldberg Polyhedron \cite{Goldberg}. To implement the points in Matlab, we derive explicit formulas on a triangular face. More details are provided in the companion paper \cite{Michaels}.
Unlike the radial icosahedral points, the sequence of equal area configurations is equidistributed. Regarding quasi-uniformity, the following is proved in~\cite{Michaels}.
\begin{proposition}
The sequence of equal area icosahedral configurations are quasi-uniform with
\[\gamma(\omega_N)\leq 0.798....\]
\end{proposition}
\newpage
As shown in Table \ref{tab:MeshIcos}, the mesh ratios appears to stay below $0.736$. These are the lowest mesh ratios of all point sets discussed.
\
\noindent \textbf{Coulomb Points and Log Energy Points (Elliptic Fekete Points)}
\
\begin{figure}
\centering
\includegraphics{Coulomb.png}
\caption{Coulomb points (top) and log points (bottom) and their Voronoi decomposition for $N=1024$. The structure is very similar for both.}
\label{Coulomb}
\end{figure}
For $s=1$, Riesz $s$-energy minimization is the classic Thomson problem for the Coulomb potential \cite{Thoms}. The sequence of minimal Coulomb energy configurations is known to be equidistributed, well-separated, and quasi-uniform \cite{Dahlberg}. However, no explicit bound is known for the mesh ratio. The Voronoi decomposition of these cells, as shown in Figure \ref{Coulomb}, primarily consists of close to regular hexagons with heptagons and pentagons forming scars along the sphere. For relatively small $N$, the scars grow out from the 12 vertices of the icosahedron like dislocations in a crystal due to displacement deformities. For $N >~5,000$, the scars become less fixed, spreading across the sphere. For an in-depth discussion of the scarring behavior, see Bowick et al \cite{Bow2002} and \cite{Bow2009}.
The log energy points are minimizers of the Riesz logarithmic potential. The sequence of log energy configurations is known to be equidistributed and well-separated, but covering and quasi-uniformity is an open problem. As shown in Table \ref{tab:CoulLog} below, numerically the log energy points appear to be quasi-uniform. The best known lower bound on separation is due to Dragnev \cite{logsep}:
\[\delta(\omega^{\log}_N)\geq\frac{2}{\sqrt{N}},\ \ \ N\geq 2.\]
Their geometric structure is very similar to the Coulomb points as shown in Figure \ref{Coulomb}. The energies of log and Coulomb points have the same asymptotic behavior in the dominant and second order term for many Riesz potentials. See Section 3 and Conjecture \ref{newconj}.
Generating these points is a highly nonlinear optimization problem. Unlike the configurations we have described up to now, they are not so quickly obtained. Table \ref{tab:CoulLog} displays the mesh ratios of near minimal Coulomb and log energy configurations. We remark that the sequence appears to have outliers at several values of $N$, such as $N = 20, 300,$ and $4096$. Points for $N<500$ and $N = k^2$, $k \leq 150$, were provided by Rob Womersley.
\begin{figure}
\centering
\includegraphics{MaxDet.png}
\caption{Maximal determinant nodes for $N=961$. The Voronoi decomposition is primarily composed of regular hexagons with scarring features similar to the minimal energy points.}
\label{Maxdet}
\end{figure}
\begin{table}[h!]
\begin{center}
\caption{Mesh Ratios for Coulomb and Log Energy Points}
\begin{tabular}{|c|c|c||c|c|c|}
\hline
$N$ & $\gamma(\omega^{\text{log}}_N)$& $\gamma(\omega^{\text{Coul}}_N)$ & $N$ & $\gamma(\omega^{\text{log}}_N)$ & $\gamma(\omega^{\text{Coul}}_N)$ \\
\hline\hline
10 & 0.687401& 0.689279& 500& 0.757354& 0.755834\\
20 & 0.731613& 0.733265& 1024& 0.752122& 0.755770\\
30 & 0.695481& 0.692966& 2025& 0.761261& 0.766218\\
40 & 0.669531& 0.670842& 3025& 0.765075& 0.761661\\
50 & 0.661301& 0.656591& 4096& 0.770240& 0.765712\\
100& 0.695371& 0.694604& 5041& 0.753573& 0.758457\\
200& 0.662102& 0.658561& 10000& 0.762672& 0.761964\\
300& 0.740635& 0.730182& 15129& 0.762385& 0.763398\\
400& 0.650106& 0.647351& 22500& 0.773483& 0.767096\\
\hline
\end{tabular}
\label{tab:CoulLog}
\end{center}
\end{table}
\
\noindent\textbf{Maximal Determinant Nodes (Fekete Nodes)}
\
Other node sets used in polynomial interpolation and numerical integration on the sphere are the maximal determinant nodes. Let $\phi_1,...,\phi_{(n+1)^2}$ be a basis for the space $\mathbb{P}_n(\mathbb{S}^2)$ of spherical polynomials of degree $\leq n$. The maximal determinant node set is the configuration $\omega_N := \omega_{(n+1)^2}\subset \mathbb{S}^2$ which maximizes
\[\textup{det}(\phi_i(x_j))_{i,j=1}^{(n+1)^2}\]
These points are independent of the choice of basis. The interpolatory cubature rule associated with the configuration $\omega_N$,
\[Q_n(f):=\sum_{j=1}^{N}w_jf(x_j),\]
is conjectured in \cite{Wom} to have all weights positive which is of interest in numerical integration. For more information about these points and their applications, see \cite{Reim94}, \cite{Reim99}, and \cite{Wom}. A typical node set is shown in Figure \ref{Maxdet}.
Like the minimal energy nodes, computing the maximal determinant nodes is a nonlinear optimization problem. The maximum is approximated by conjugate gradient and Newton-like methods on $\mathbb{S}^2$ \cite{Wom}. Nodes for $1\leq n\leq 165$ are available from \textit{http://web.maths.unsw.edu.au/\textasciitilde rsw/Sphere/Extremal/New/index.html}.
Berman et al \cite{BerBoucks} proved the sequence of maximal determinant configurations is equidistributed, while in \cite{Wom}, Sloan and Womersley proved it is quasi-uniform with
\[\limsup_{N\to\infty}\gamma(\omega_N) < \frac{4j_0}{\pi}\approx 3.06195,\]
where $j_0$ is the smallest positive zero of the Bessel function of the first kind, $J_0$. As shown in Table \ref{tab:MaxDet}, the mesh ratio bound appears to be much lower though it is unclear whether or not $\lim_{N\to\infty}\gamma(\omega_N)$ exists.
\begin{figure}
\centering
\includegraphics{Random.png}
\caption{Random points for $N=700$ and their Voronoi decomposition.}
\label{Random}
\end{figure}
\begin{table}[h!]
\begin{center}
\caption{Mesh Ratios for Maximal Determinant Nodes}
\begin{tabular}{|c|c|c|c|}
\hline
$N$ & $\gamma(\omega_N)$& $N$ & $\gamma(\omega_N)$\\
\hline\hline
9& 0.718884 & 625& 0.805608\\
16& 0.685587 & 1024& 0.840506\\
25& 0.768510 & 2025& 0.858874\\
36& 0.806140& 3025& 0.847347\\
49& 0.777490 & 4096& 0.859887\\
100& 0.708579& 4900 & 0.877990\\
225& 0.860728& 10201 & 0.859625\\
324& 0.799227& 15129 & 0.865695\\
400& 0.809172& 22500 & 0.881492\\
\hline
\end{tabular}
\label{tab:MaxDet}
\end{center}
\end{table}
\
\noindent \textbf{Random Points}
\
The final configurations we consider are random configurations $\omega_N^{rand}$ consisting of $N$ independent samples chosen with respect to surface area measure. Not surprisingly, these configurations do not have optimal order separation or covering and the sequence is not quasi-uniform. As proved in \cite{Randcov} and \cite{RezSaff} respectively,
\[ \lim_{N\rightarrow\infty}\mathbb{E}(\delta(\omega^{\text{rand}}_N))N = \sqrt{2\pi}, \ \ \ \ \ \lim_{N\rightarrow\infty}\mathbb{E}(\eta(\omega^{\text{rand}}_N)) \bigg(\frac{N}{\log N}\bigg)^{1/2}= 2. \]
Note that while the order of the separation of random points is off by factor of $N^{1/2}$, the covering is only off by a factor of (log$N)^{1/2}$. Figure \ref{Random} shows a realization of i.i.d. uniformly chosen random points on $\mathbb{S}^2$.
\
\noindent \textbf{Summary of Properties}
\
The following tables compare some of the properties of the point sets described above. Table \ref{tab:comp1} compares which sequences are proven to be equidistributed and well-separated, for which values of $N$ the configurations are defined, and whether a subsequence is hierarchical. Table \ref{tab:comp2} compares which sequences are quasi-uniform and the numerically determined bounds for separation and mesh ratio constants.
\
\begin{table}[h!]
\begin{center}
\caption{Summary of Point Set Properties}
\begin{tabular}{|c|c|c|c|c|}
\hline
Name & Defined for & Hier. & Equidist. & Separated \\
\hline
Gen Spiral & $N\geq 2$ & No & Yes & Yes \\
Fibonacci & Odd $N$ & No & Yes & Yes \\
Hammersley & $N\geq 2$& No& Yes& No \\
Zonal Eq. Area & $N \geq 2$ & No & Yes & Yes \\
HEALPix & $12k^2$, & Subseq.& Yes & Yes \\
Octahedral & $4k^2 + 2$& Subseq.& Yes& Yes \\
Radial Icos. & $10k^2+2$, & Subseq. & No & Yes \\
Cubed Sphere & $6k^2-12k+8$& Subseq.& No& Yes \\
Equal Area Icos. & $10(m^2+mn+n^2)+2$ & Subseq. & Yes & Yes \\
Coulomb & $N\geq 2$ & No & Yes & Yes \\
Log Energy & $N\geq 2$& No & Yes& Yes \\
Max Det. & $(1+k)^2$ & No & Yes & Yes \\
Random & $N\geq 2$& No & Yes & No\\
\hline
\end{tabular}
\label{tab:comp1}
\end{center}
\end{table}
\begin{table}[h!]
\begin{center}
\caption{Comparison of Separation and Mesh Ratio Constants}
\begin{tabular}{|c|c|c|c|}
\hline
Name & Quasi- & Numeric lower bound & Numeric upper bound \\
& Uniform& on $\liminf\delta(\omega_N)\sqrt{N}$ & on $\limsup\gamma(\omega_N)\sqrt{N}$ \\
\hline
Gen Spiral & Yes & 3.1319 & 0.8099 \\
Fibonacci & Yes & 3.0921& 0.8823 \\
Hammersley & No & N/A & N/A\\
Zonal Eq. Area & Yes & 3.3222 & 0.7568 \\
HEALPix & Yes & 2.8345 & 1.0000 \\
Radial Icos. & Yes & 2.8363 & 0.8517\\
Cubed Sphere & Yes & 2.7027 & 1.0000\\
Octahedral & Yes & 2.8284 & 0.9235\\
Equal Area Icos. & Yes & 3.1604 & 0.7360 \\
Coulomb & Yes & 3.3794 & 0.7671\\
Log Energy & Conj. & 3.3733 & 0.7735\\
Max Det. & Yes & 3.1957 & 0.8900\\
Random & No & N/A & N/A\\
\hline
\end{tabular}
\label{tab:comp2}
\end{center}
\end{table}
\section{Asymptotic Energy Considerations}
We now turn our attention to the potential energy of the above configurations, in particular, $E_{\log}(\omega_N)$ and $E_s(\omega_N)$ for $s$ = -1,1,2, and 3. We are interested in the asymptotic behavior of the energies for above configurations as $N \rightarrow \infty$ and how well the leading terms in the expansion of their energies match the known or conjectured leading terms and coefficients in the minimal energy expansion. We begin with the following well known necessary condition for asymptotically optimal point sets \cite{MEBook}.
\begin{theorem}
Let $\left\{\omega_N\right\}_{N=2}^\infty$ be a sequence of configurations that is asymptotically optimal with respect to logarithmic energy or Riesz $s$-energy for some $s>-2$, $s\neq 0$, i.e.,
\[\lim_{N\rightarrow\infty}\frac{E_{s}(\omega_N)}{\mathcal{E}_{s}(N)} = 1 \ \ \ \ \ \ \ \ \textup{or} \ \ \ \ \ \ \ \ \lim_{N\rightarrow\infty}\frac{E_{\log}(\omega_N)}{\mathcal{E}_{\log}(N)} = 1.\]
Then $\left\{\omega_N\right\}_{N=2}^\infty$ is equidistributed.
\label{optimpliesequi}
\end{theorem}
We define the energy of a probability measure $\mu$ on $\mathbb{S}^2$ with respect to the logarithmic or Riesz $s$-potential as
\[\ \ \ \ \ \ \ \ \ \ \ \mathcal{I}_{s} [\mu]:=\ \int\int \frac{1}{|\textbf{x} - \textbf{y}|^s} \text{d}\mu(\textbf{x})\text{d}\mu(\textbf{y}),\ \ \ \ \ \ s\neq 0\]
\[\mathcal{I}_{\log} [\mu]:=\ \int\int \log\frac{1}{|\textbf{x} - \textbf{y}|} \text{d}\mu(\textbf{x})\text{d}\mu(\textbf{y}).\]
The normalized surface area measure $\sigma$ is the unique minimizer of $\mathcal{I}_{\log} [\mu]$ and $\mathcal{I}_{s} [\mu]$ for $-2<s<2$, $s\neq 0$, and
\[\mathcal{I}_{\log} [\sigma] = \frac{1}{2}-\log2,\]
\[\mathcal{I}_s[\sigma] = \frac{2^{1-s}}{2-s}, \ \ \ \ -2<s<2.\]
However, for $s\geq 2$, $\mathcal{I}_{s} [\mu] = \infty$ for all $\mu$ supported on $\mathbb{S}^2$ (see for example \cite{MEBook}).
For random points, the expected value of the $s$-energy is easily computed:
\begin{equation}
\mathbb{E}[E_{s}(\omega_N^{\text{rand}})] = \mathcal{I}_{s}[\sigma](N(N-1)),\ \ \ \ \ -2<s<2.
\label{RandomEnergy}
\end{equation}
For $s\geq 2$, $\mathbb{E}[E_{s}(\omega_N^{\text{rand}})] = \infty$.
The Epstein-Zeta function for a lattice $\Lambda$ in $\mathbb{R}^2$ is given by
\[\zeta_{\Lambda}(s):= \sum\limits_{0\neq\textbf{x}\in\Lambda} |\textbf{x}|^{-s},\ \ \ \ \ \ \textup{Re}\ s>2.\]
Let $\Lambda_2$ be the regular triangular lattice in $\mathbb{R}^2$ generated by basis vectors $(1,0)$ and $(1/2,\sqrt{3}/2)$. It is known from number theory that $\zeta_{\Lambda_2}(s)$ admits the factorization
\begin{equation}
\zeta_{\Lambda_2}(s) = 6\zeta(s/2)\text{L}_{-3}(s/2),\ \ \ \ \ \ \text{Re}\ s>2,
\label{EpsZeta}
\end{equation}
where
\[\text{L}_{-3}(s) := 1-\frac{1}{2^s}+\frac{1}{4^s}-\frac{1}{5^s}+\frac{1}{7^s}-\cdots,\ \ \ \ \ \ \text{Re}\ s>1,\]
is a Dirichlet L-series. The right-hand side of (\ref{EpsZeta}) can be used to extend $\zeta_{\Lambda_2}(s)$ to $s$ with Re $s<2$.
\
\subsection{Logarithmic Potential}
The following asymptotic expansion is proven by Betermin and Sandier in \cite{Betermin}:
\begin{theorem}
\label{logexpansion}
There exists a constant $C\neq 0$, independent of $N$, such that
\[\mathcal{E}_{\log}(N) = \mathcal{I}_{\log} [\sigma] N^2-\frac{N\log N}{2} + CN + o(N), \ \ \ \ \ \ N\rightarrow \infty,\]
\[ -0.22553754 \leq C\leq \widehat{C}:=2\log2 +\frac{1}{2}\log\frac{2}{3} +3\log\frac{\sqrt{\pi}}{\Gamma(1/3)} = -0.05560530... \ \ \ \ \ \ \]
\end{theorem}
\noindent The following extension of Theorem \ref{logexpansion} is conjectured by Brauchart et al in \cite{BHS12}:
\begin{conjecture}
\label{logconjecture}
\[\mathcal{E}_{\log}(N) = \mathcal{I}_{\log} [\sigma] N^2-\frac{N\log N}{2} + \widehat{C}N +D\log N + O(1), \ \ \ \ \ \ \ N\rightarrow \infty.\]
\end{conjecture}
\noindent Beltran \cite{Bel15} provides a partial converse to Theorem \ref{optimpliesequi}:
\begin{theorem}
\label{Beltranthm}
If $\left\{\omega_N\right\}_{N=2}^\infty$ is a sequence of well-separated configurations whose spherical cap discrepancy satisfies
\[\lim_{N\rightarrow\infty}D_C(\omega_N)\log N = 0,\]
then $\left\{\omega_N\right\}_{N=2}^\infty$ are asymptotically optimal with respect to logarithmic energy.
\end{theorem}
\noindent As a corollary, Coulomb points and Fibonacci spiral points are asymptotically log optimal. We make the following natural conjecture:
\begin{conjecture}
\label{logequiconj}
The condition on the spherical cap discrepancy in Theorem \ref{Beltranthm} can be relaxed to
\[\lim_{N\rightarrow\infty}D(\omega_N) = 0.\]
Thus a sequence of configurations $\left\{\omega_N\right\}_{N=2}^{\infty}$ is equidistributed and well-separated iff it is asymptotically optimal with respect to logarithmic energy.
\end{conjecture}
\begin{figure}
\centering
\makebox[\textwidth][c]{\includegraphics[width=1.4\textwidth]{AsymptoticsLog1.png}}
\caption{First order asymptotics for log potential. The solid black line is the known first order coefficient in the minimal energy expansion as given in Theorem \ref{logexpansion}.}
\label{logasymptotics1}
\end{figure}
\begin{figure}
\centering
\makebox[\textwidth][c]{\includegraphics[width=1.4\textwidth]{AsymptoticsLog2.png}}
\caption{Second order asymptotics for log potential (with the legend of Figure \ref{logasymptotics1}). The solid black line corresponds to the known coefficient from Theorem \ref{logexpansion}.}
\label{logasymptotics2}
\end{figure}
\begin{figure}
\centering
\makebox[\textwidth][c]{\includegraphics[width=1.4\textwidth]{AsymptoticsLog3.png}}
\caption{Third order asymptotics for log energy (with the legend of Figure \ref{logasymptotics1}). The dashed line is the conjectured third order constant from \cite{BHS12}.}
\label{logasymptotics3}
\end{figure}
Figures \ref{logasymptotics1}, \ref{logasymptotics2}, and \ref{logasymptotics3} display three comparisons of the logarithmic energies of the point sets with cardinality up to 50,000. Some configurations are sampled along subsequences to avoid overcrowding the picture. Due to the computational cost of generating approximate log energy and Coulomb points, these points are only available for $N<22,500$. The energies of random configurations are plotted by expected value. We do not include the maximal determinant nodes because there is no known algorithm to generate them in polynomial time.
Figure \ref{logasymptotics1} shows $E_{\log}(\omega_N)/N^2$ for the point sets. For the equidistributed and well-separated point sets as well as for the Hammersley nodes, this ratio converges to $1/2-\log2$ supporting Conjecture \ref{logequiconj}. Though the radial icosahedral and cubed sphere points are not equidistributed and thus not asymptotically optimal, their log energy appears to be of leading order $N^2$.
Figure \ref{logasymptotics2} displays
$[E_{\log}(\omega_n) - \mathcal{I}_{\log} [\sigma] N^2]/N\log N$ for each configuration. The curves for classes of point sets begin separating. The energies of the octahedral, HEALPix, Fibonacci, generalized spiral, zonal equal area, and Coulomb points all appear to converge to the correct second order term. The energy of the Hammersley nodes appears to have $N\log N$ second term, though incorrect coefficient. While the equal area icosahedral nodes perform better than the radial icosahedral nodes, their asymptotic energy appears to have a second term different to $N\log N$. Numerical second order behavior of log energy points is also studied in \cite{NBK}.
Figure \ref{logasymptotics3} compares the energy of the configurations to the conjectured minimal log energy third order term, i.e. $[E_{\log}(\omega_n) - \mathcal{I}_{\log} [\sigma]N^2 + 1/2N\log N]/N$. Conjecture \ref{logconjecture} is supported by the behavior of the Coulomb and log points. The octahedral, HEALPix, Fibonacci, zonal equal area, and generalized spiral configurations appear to have the third term of their energy of order $N$ but the wrong coefficient. Of the algorithmically generated point sets, the generalized spiral and zonal equal area points perform the best with respect to the logarithmic energy.
\
\subsection{Riesz Potential, -2 $<\textbf{\textit{s}}<$ 2, $\textbf{\textit{s}}\neq$ 0}
\
In this range, the best known bounds on $\mathcal{E}_s(N)$ are due to Wagner (cf. \cite{WagDn} and \cite{WagUp}).
\begin{theorem}
\label{s1theorem}
For $-2<s<2$, $s\neq 0$, there exist $c_s,C_s>0$ such that for all $N\in\mathbb{N}$,
\[c_sN^{1+s/2}\leq \mathcal{E}_s(N) - \mathcal{I}_s[\sigma]N^2\leq C_sN^{1+s/2}.\]
\end{theorem}
\noindent Kuiljaars and Saff \cite{KS98} conjecture that $\lim_{N\rightarrow\infty} [\mathcal{E}_s(N) - \mathcal{I}_s[{\sigma}]N^2]/N^{1+s/2}$ exists.
\begin{conjecture}
For $-2<s<2$, $s\neq 0$,
\[\mathcal{E}_s(N) = \mathcal{I}_s[{\sigma}]N^2 + \frac{(\sqrt{3}/2)^{s/2}\zeta_{\Lambda_2}(s)}{(4\pi)^{s/2}}N^{1+s/2} + o(N^{1+s/2}),\ \ \ \ \ \ N\rightarrow\infty.\]
\label{s1conj}
\end{conjecture}
Heuristically, the second order coefficient corresponds to the Voronoi decomposition of minimal energy points being composed primarily of close to regular hexagons. A characterization of asymptotically optimal point sets is due to Leopardi \cite{s1char}.
\begin{theorem}
\label{s1char}
If a well-separated sequence of configurations $\left\{\omega_N\right\}_{N=2}^\infty$ is equidistributed, then it is asymptotically optimal for $0<s<2$,
\end{theorem}
\noindent A similar result for $s<0$ is also known (see for example \cite{MEBook}).
\begin{theorem}
A sequence of configurations $\left\{\omega_N\right\}_{N=2}^{\infty}$ is asymptotically optimal for $-2<s<0$ iff it is equidistributed.
\label{sneg1char}
\end{theorem}
\noindent Theorems \ref{s1char} and \ref{sneg1char} are the analogs of Conjecture \ref{logequiconj} for $-2<s<2$, $s\neq 0$.
In the particular case of $s=-1$, the problem of minimizing energy becomes that of maximizing sums of distances. A characterization of minimal configurations for this case is due to Stolarsky \cite{Stolarsky} (see also \cite{BDStol}) and is generalized by Brauchart and Dick in \cite{StoGen}. Let $V_t(\textbf{z}) = \left\{\textbf{x}\ |\ \textbf{x}\cdot\textbf{z}\geq t\right\}$ denote the spherical cap of height $1-t$ centered at $\textbf{z}\in\mathbb{S}^2$. The \textit{$L_2$ discrepancy} of a configuration $\omega_N$ is defined to be
\[D_{L_2}(\omega_N):= \bigg(\int_{-1}^{1}\int_{\mathbb{S}^2}\bigg(\frac{|\omega_N\cap V_t(\textbf{z})|}{N}-\sigma(V_t(\textbf{z}))\bigg)^2 d\sigma(\textbf{z})dt \bigg)^{1/2}.\]
\begin{theorem}
\textbf{Stolarsky Invariance Principle.} For any $\omega_N\subset\mathbb{S}^2$,
\[\frac{1}{N^2}\sum_{i\neq j}|\textbf{x}_i-\textbf{x}_j| + 4D_{L_2}(\omega_N)^2 = \mathcal{I}_{-1}[\sigma] = 4/3.\]
\label{Stolarsky Invariance}
\end{theorem}
\noindent This gives the immediate corollary (see also \cite{BrauStol}):
\begin{corollary}
For a configuration $\omega_N$, $E_{-1}(\omega_N) = \mathcal{E}_{-1}(N)$ iff $\omega_N$ minimizes $L_2$ discrepancy.
\end{corollary}
Figure \ref{s1asymptotics1} graphs the Coulomb energy of the point sets to first order. From Theorem \ref{s1char}, $\lim_{N\rightarrow\infty} E_1(\omega_N)/N^2 = \mathcal{I}_s[\sigma] = 1$ for all the point sets except the radial icosahedral, cubed sphere, and Hammersley nodes. As in the logarithmic case, the energies of the radial icosahedral and cubed sphere nodes appear to have $N^2$ leading order term but incorrect coefficient. The plot is inconclusive on whether the behavior of the Hammersley nodes is asymptotically optimal. Figure \ref{s1asymptotics2} shows the second order $N^{3/2}$ term from Conjecture \ref{s1conj}. The energies of the generalized spiral, Fibonacci spiral, zonal equal area, HEALPix, and octahedral nodes all appear to converge to an $N^{3/2}$ term that is different from the conjectured value. None of these point sets have a regular hexagonal lattice structure as $N$ gets large. Again the energies of equal area icosahedral points have a different second order term. The Hammersley nodes don't have low enough energy to appear on the plot. The asymptotic energy of the log nodes seems to have the same second order coefficient as the Coulomb nodes and both point sets support Conjecture \ref{s1conj}. As in the logarithmic energy case, the generalized spiral and zonal equal area points perform the best of the algorithmically generated points.
\begin{figure}
\centering
\makebox[\textwidth][c]{\includegraphics[width=1.4\textwidth]{AsymptoticsS11.png}}
\caption{First order asymptotics for $s = 1$. The solid black line is the known coefficient from Theorem \ref{s1theorem}.}
\label{s1asymptotics1}
\end{figure}
\begin{figure}
\centering
\makebox[\textwidth][c]{\includegraphics[width=1.4\textwidth]{AsymptoticsS12.png}}
\caption{Second order asymptotics for $s = 1$ (with the legend of Figure \ref{s1asymptotics1}). The dashed line is the conjectured value from \cite{KS98}.}
\label{s1asymptotics2}
\end{figure}
\begin{figure}
\centering
\makebox[\textwidth][c]{\includegraphics[width=1.4\textwidth]{AsymptoticsSNeg11.png}}
\caption{First order asymptotics for $s = -1$. The solid line follows from the Stolarsky invariance principle.}
\label{sneg1asymptotics1}
\end{figure}
\begin{figure}
\centering
\makebox[\textwidth][c]{\includegraphics[width=1.4\textwidth]{AsymptoticsSNeg12.png}}
\caption{Second order asymptotics for $s = -1$ (with the legend of Figure \ref{sneg1asymptotics1}). The dashed line is from conjecture \ref{s1conj}.}
\label{sneg1asymptotics2}
\end{figure}
Figure \ref{sneg1asymptotics1} displays $E_{-1}(\omega_N)/N^2$ for the point sets. Again the radial icosahedral and cubed sphere points have the correct order though they are not asymptotically optimal. For $s=-1$ The second term in the minimal energy expansion is of known order $\sqrt{N}$ with conjectured coefficient $\approx 0.7985$. As shown in Figure \ref{sneg1asymptotics2}, the behavior of the second order of the asymptotic energy of the point sets resembles that of their log and Coulomb energies. The minimal Coulomb and log energy points perform the best and support Conjecture \ref{s1conj}. The asymptotic energies of the octahedral, HEALPix, Fibonacci, zonal equal area, and generalized spiral appear to converge to the right second order and incorrect coefficient, and the latter two configurations have the lowest energy.
\subsection{Riesz Potential, $\textbf{\textit{s=2}}$}
\
Less is known about the case $s=2$. The first order term was proved in \cite{KS98}.
\begin{theorem}
\[\lim_{N\rightarrow\infty}\frac{\mathcal{E}_2(N)}{N^2\log N} = \frac{1}{4}.\]
\label{s2expansion}
\end{theorem}
It is not known whether the $N^2\log N$ term corresponds to any geometric property of the points. There is also no result analogous to Theorems \ref{Beltranthm} and \ref{s1char} giving a sufficient condition for a sequence of configurations to be asymptotically optimal.
Figure \ref{s2asymptotics1} shows $E_2(\omega_N)/N^2\log N$. The Hammersley points do not appear to have asymptotically minimal energy. The behavior of the octahedral, HEALPix, Fibonacci, generalized spiral, zonal equal area, and equal area icosahedral points is not clear. Their energies remain well below the known coefficient but have larger energy than the Coulomb and log energy points. Calculations beyond $N=50,000$ are needed. The generalized spiral, zonal equal area, and equal area icosahedral points perform the best.
\begin{figure}
\centering
\makebox[\textwidth][c]{\includegraphics[width=1.4\textwidth]{AsymptoticsS21.png}}
\caption{First order asymptotics for $s = 2$. The break in the y-axis values is to enhance the separation between the configurations. The solid line is the known coefficient from Theorem \ref{s2expansion}.}
\label{s2asymptotics1}
\end{figure}
\
\subsection{Riesz Potential, $\textbf{\textit{s}}>2$}
\
\begin{figure}
\centering
\makebox[\textwidth][c]{\includegraphics[width=1.4\textwidth]{AsymptoticsS31.png}}
\caption{First order asymptotics for $s = 3$. The dashed line is the conjectured coefficient.}
\label{s3asymptotics1}
\end{figure}
In the hypersingular case $s>2$ the continuous energy integral is infinite for any probability measure $\mu$. Instead, the energy is known to be dominated by the nearest neighbor interactions of points as shown in a much more general result by Hardin and Saff \cite{HS2005}.
\begin{theorem}
For $s>2$, there are constants $C_s$ such that
\[\lim_{N\rightarrow\infty}\frac{\mathcal{E}_s(N)}{N^{1+s/2}} = C_s,\ \ \ \ \ \ \ \ 0<C_s<\infty.\]
\label{hypersingular}
\end{theorem}
\noindent One can consider the analytic continuation of $\mathcal{I}_s[\sigma]$, $0<s<2$,
\[V_s: = \frac{2^{1-s}}{2-s},\ \ \ \ s\in\mathbb{C}\setminus \left\{2\right\}.\]
The following is analogous to Conjecture \ref{s1conj} (see \cite{MEBook}).
\begin{conjecture}
For $2<s<4$
\[\mathcal{E}_s(N) = \frac{(\sqrt{3}/2)^{s/2}\zeta_{\Lambda_2}(s)}{(4\pi)^{s/2}}N^{1+s/2}+V_sN^2 + o(N^2).\]
\label{s3conj}
\end{conjecture}
Figure \ref{s3asymptotics1} plots $E_3(\omega_N)/N^{5/2}$. The energies of most configurations seem to be going to the correct order but incorrect coefficient. The equal area icosahedral points outperform the spiral and zonal equal area points of the algorithmically generated configurations. This is expected because their Voronoi decomposition is closest to the regular hexagonal lattice. The log energy and Coulomb points again seem to be close to minimal and may converge to the conjectured $C_3\approx 0.199522$. The Hammersley points are not seen on the plot because their asymptotic energy does not appear to have first order $N^{5/2}$.
We conclude this section with the following observation: The apparent first and second order asymptotically minimal behavior of the energy of the Coulomb and log points for the potentials we have considered is striking and we suggest a conjecture.
\begin{conjecture}
The minimal logarithmic energy points and Coulomb points are asymptotically minimal to second term precision for all $-2<s<4$, $s\neq 0$. I.e., combined with Conjectures \ref{s1conj} and \ref{s3conj}, letting $\omega_N:=\omega_N^{log}$ or $\omega_N^{Coul}$
\[\lim_{N\rightarrow\infty}\frac{E_s(\omega_N)-\mathcal{I}_s[\sigma]N^2}{\mathcal{E}_s(N)-\mathcal{I}_s[\sigma]N^2} = 1\ \ \ or\ \ \ \lim_{N\rightarrow\infty}\frac{E_s(\omega_N)-\frac{(\sqrt{3}/2)^{s/2}\zeta_{\Lambda_2}(s)}{(4\pi)^{s/2}}N^{1+s/2}}{\mathcal{E}_s(N)-\frac{(\sqrt{3}/2)^{s/2}\zeta_{\Lambda_2}(s)}{(4\pi)^{s/2}}N^{1+s/2}}=1.\]
Furthermore, the Coulomb points are minimal in the logarithmic energy to the third order term. That is
\[\lim_{N\to\infty}\frac{E_{\log}(\omega_N^{Coul})-(1/2-\log2)N^2+1/2N\log N}{\mathcal{E}_{\log}(N)-(1/2-\log2)N^2+1/2N\log N}=1.\]
\label{newconj}
\end{conjecture}
\section{Proofs}
We begin with two auxiliary results.
\begin{proposition}
Let $\left\{P_N\right\}_{N=1}^\infty$ be a diameter bounded sequence of asymptotically equal area partitions of $\mathbb{S}^2$ such that each $P_N$ has $N$ cells. For each $P_N$, let $\omega_N$ be a configuration of points on $\mathbb{S}^2$ such that the interior of each cell of $P_N$ contains exactly one point of $\omega_N$. Then $\left\{\omega_N\right\}_{N=1}^\infty$ is equidistributed and provides a covering of $\mathbb{S}^2$ with $\eta(\omega_N) \leq CN^{-1/2}$ for all $N\in \mathbb{N}$, where $C$ is as in equation \textup{(\ref{diambdd})}.
\label{lemmaprop}
\end{proposition}
\begin{proof}
The bound on the covering radius is trivial. Let $A\subset \mathbb{S}^2$ be a spherical cap and let
\[A_\delta := \left\{x\in\mathbb{S}^2:\ \textup{dist}(x,A)\leq \delta \right\},\] where dist$(x,A) := \min_{y\in A}|x-y|$ is the standard distance function. For $x\in\omega_N$, denote by $W_x^N$ the cell of $P_N$ containing $x$. Let $\epsilon,\delta>0$ and choose $N$ large enough such that
\[N\min_x\sigma(W_x^N)\geq 1-\epsilon\]
and $x\in A\cap\omega_N$ implies $W_x^N\subset A_\delta$. Then
\[\frac{|\omega_N\cap A|}{N}\leq \frac{|\left\{x\ :\ W_x^N\subset A_\delta\right\}|}{N}\leq\frac{\sigma(A_\delta)}{N\min_x\sigma(W_x^N)}\leq \frac{\sigma(A_\delta)}{1-\epsilon}.\]
Since $\epsilon$ is arbitrary, we have
\[\limsup_N\frac{|\omega_N\cap A|}{N}\leq \sigma(A_\delta).\]
Letting $\delta\to 0$ gives
\begin{equation}
\limsup_N\frac{|\omega_N\cap A|}{N}\leq \sigma(A).
\label{lemmaeq}
\end{equation}
Applying inequality (\ref{lemmaeq}) to $\mathbb{S}^2\setminus A$, we obtain
\[\liminf_N\frac{|\omega_N\cap A|}{N}=1-\limsup_N\frac{|\omega_N\cap(\mathbb{S}^2\setminus A)|}{N}\geq 1-\sigma(\mathbb{S}^2\setminus A) = \sigma(A),\]
and thus, we have
\[\lim_{N\to\infty}\frac{|\omega_N\cap A|}{N}=\sigma(A).\]
\end{proof}
\
\noindent\textit{Proof of Theorem \ref{spiral distribution}.}
For a fixed $N$ denote in spherical coordinates $x_i:=(\phi_i,\theta_i)\in\omega_N$. We first prove the separation bound. Let $\epsilon_N = 2\sqrt{4\pi/N}$. For large $N$, by the pigeonhole principle, at least one set of adjacent points $x_k,x_{k+1}\in\omega_N$ is separated along $S_N$ by a distance less than $\sqrt{4\pi/N}$. So we can restrict our attention to $k\leq N/2$ and $\epsilon_N$-balls $B(x_k,\epsilon_N)$. For large N, if $k<8\pi+1/2$, then
\[\cos\phi_k = 1-\frac{2k-1}{N}\geq 1-\frac{16\pi}{N} \approx \cos(2\sqrt{\frac{4\pi}{N}}),\]
and $x_k$ is within the first two full longitudinal turns of $S_N$ starting from a pole. Otherwise, $B(x_k,\epsilon_N)$ contains disjoint levels of $S_N$. In this case, the minimal distance between levels in $B(x_k,\epsilon_N)$ is $\sqrt{4\pi/N}$. We compute the nearest neighbor distance between points in the same level as follows. Let
\[f_N(k):=\sqrt{N}|x_k-x_{k+1}|, \ \ \ \ x_k\in \omega_N,\ \ \ \ k<N/2.\]
Using the distance formula for spherical coordinates
\begin{equation}
|x_j-x_k|^2 = 2-2(\cos\phi_j\cos\phi_k+\sin\phi_j\sin\phi_k\cos(\theta_j-\theta_k))
\label{geodesic}
\end{equation}
and expanding $\cos^{-1}x$ around $x=1$ we have
\begin{equation}
\lim_{N\to\infty}f_N(k) = \sqrt{8k-4\sqrt{4k^2-1}\cos(\sqrt{2\pi}(\sqrt{2k-1}-\sqrt{2k+1}))},
\label{spirallimit}
\end{equation}
and $f_N(k)<f_{N+1}(k)$ for all $k<N/2-1$ and large N. Thus we have the correct order for the minimal separation between adjacent points in $\omega_N$. Furthermore, (\ref{spirallimit}) is increasing as a function of $k$ and thus
\begin{equation}
\lim_{N\to\infty} \min_k f_N(k) = \sqrt{8-4\sqrt{3}\cos(\sqrt{2\pi}(1-\sqrt{3}))}.
\label{spiralmin}
\end{equation}
Lastly, around the north pole, $k_1,k_2<8\pi+1/2$, we can again use (\ref{geodesic}) to show that
\[\lim_{N\to\infty} \sqrt{N}|x_{k_1}-x_{k_2}|\]
exists and can be computed case by case for pairs $(k_1,k_2)$. The southern hemisphere can be similarly computed. Comparing to (\ref{spiralmin}) gives the separation constant (\ref{spiralsep}).
\begin{figure}[h!]
\centering
\includegraphics{BauerVoronoiCell.png}
\caption{Limiting form of Voronoi cell for each $x_i\in K_h\cap\omega_N$ with relevant quantities labeled. We can choose $N$ large enough such that the nearest neighbors of $x_i$ lie within $\epsilon$ of the points taken along iso-latitudinal lines separated by $\sqrt{4\pi/N}$}
\label{spiralvoronoicell}
\end{figure}
For covering, given $y\in\mathbb{S}^2$,
\[\textup{dist}(S_N)\leq \sqrt{\frac{\pi}{N}}.\]
From (\ref{spirallimit}), the maximal distance from any point on $S_N$ to a point of $\omega_N$ is $O(1/\sqrt{N})$ and thus the covering radius of $\omega_N$ is also $O(1/\sqrt{N})$.
We have two additional observations. First, the Voronoi decompositions of the spiral points are diameter bounded. Secondly, the Voronoi cells are asymptotically equal area on $K_h := \left\{(x,y,z)\in \mathbb{S}^2:-h\leq z\leq h \right\}$ for any fixed $0<h<1$. By this we mean
\[\lim_{N\rightarrow\infty}N\max_{V_x(\omega_N)\subset K_h} \sigma(V_x(\omega_N)) = \lim_{N\rightarrow\infty}N\min_{V_x(\omega_N)\subset K_h} \sigma(V_x(\omega_N)) = 1.\]
Indeed, fixing $h$, for any $\epsilon>0$, we can take $N$ large enough such that given $x_i=(\phi_i,\theta_i)\in K_h\cap \omega_N$, $x_{i-1}$, and $x_{i+1}$ are almost iso-latitudinal with $x_i$ with separation $\sqrt{4\pi/N}$. I.e.,
\[|x_{i\pm 1}-(\phi_i,\theta_i\pm\csc\phi_i\sqrt{4\pi/N})|<\epsilon.\]
There exists shifts $0\leq\lambda_{i+},\lambda_{i-}\leq\sqrt{4\pi/N}$ such that the nearest points in the adjacent spiral levels are within $\epsilon$ of the points $(\phi_i\pm\sqrt{4\pi/N},\theta_i+\csc\phi_i\lambda_{i\pm})$ and $(\phi_i\pm\sqrt{4\pi/N},\theta_i+\csc\phi_i(\lambda_{i\pm}-\sqrt{4\pi/N}))$. Thus as $N\to\infty$, the Voronoi cell $V_i(\omega_N)$ approaches the form of $V$ in Figure \ref{spiralvoronoicell} and
\[\sigma(V_i(\omega_N)) = \sigma(V) + O(\epsilon^2).\]
Furthermore, we can treat $V$ as a planar polygon in Figure \ref{spiralvoronoicell}, and
\[\sigma(V)=\frac{1}{N}\]
independent of the shifts $\lambda_{i\pm}$. This we show by direct computation.
Letting $a=\sqrt{4\pi/N}$ and centering $x_i$ at $(0,0)$, the points $y_1 = (\lambda_{i+}/2,a/2)$ and $y_2 = ((\lambda_{i+}-a)/2,a/2) $ are the midpoints of the lines connecting $x_i$ to its nearest neighbors in the adjacent level which are shifted by $\lambda_{i+}$. The corresponding lines
\[L_1: y = -\frac{\lambda_{i+}}{a}(x-\frac{\lambda_{i+}}{2})+\frac{a}{2} \ \ \ \ \ \ \ L_2: y = \frac{a-\lambda_{i+}}{a}(x-\frac{\lambda_{i+}-a}{2})+\frac{a}{2}\]
form the top boundary of V and have intersection point
\[w:= (\lambda_{i+}-a/2,(\lambda_{i+}a+a^2-\lambda_{i+}^2)/2a).\]
From this we calculate the area of the top half of V to be
\[\sigma(V_{top}) = \frac{1}{4\pi}\bigg[a\bigg(\frac{-a\lambda_{i+}+\lambda_{i+}^2+a^2}{2a}\bigg)+\frac{a}{2}\bigg(\frac{a\lambda_{i+}-\lambda_{i+}^2}{a}\bigg)\bigg]=\frac{1}{2N}.\]
The same calculation holds for the bottom half of $V$ and thus (\ref{asympeq}) holds.
We now consider equidistribution. Because the height steps between points in $\omega_N$ are uniform, for a spherical cap $A$ centered at a pole,
\begin{equation}
\lim_{N\to\infty}\frac{|\omega_N\cap A|}{N}=\sigma(A).
\label{equicap}
\end{equation}
If $A$ does not contain one of the poles, then $A\subset K_h$ for some $h$ and (\ref{equicap}) holds by Proposition \ref{lemmaprop}. Finally if $A$ is a cap containing but not centered at one of the poles, $A$ can be partitioned into an open cap of height $h$ centered at the pole and $A\cap K_h$. Because (\ref{equicap}) holds on each disjoint subset, it also holds on~$A$.
\qed
\
\noindent\textit{Proof of Proposition \ref{Fib mesh ratio}.}
When $z=k,$ the basis vector $\textbf{c}_{k,i}$ has minimum length of $\sqrt{2}d$. At this latitude, $\textbf{c}_{k+1,i}$ and $\textbf{c}_{k-1,i}$ form the next most dominant spirals and have length $\sqrt{3}d$. For $z = k + 1/2$, $\textbf{c}_{k,i}$ and $\textbf{c}_{k+1,i}$ are equally dominant and have grid length $\sqrt[4]{5}d$. For a fixed latitude $\phi\neq \pm\pi/2$, $z$ increases with $N$. Points around $\phi$ will form a locally rectangular grid. Thus the separation approaches $\sqrt{2}d$ which occurs when $z=k$ and the largest hole in the triangulation around $x_i$ will be at most $\sqrt[4]{5}d/\sqrt{2}$ which occurs when $z=k\pm 1/2$. Thus off the poles, $\delta(\omega_{2N+1})\geq \sqrt{8\pi\sqrt{5}}/\sqrt{N}$ and $\eta(\omega_{2N+1})\leq \sqrt{2\pi/N}.$
Indeed, these inequalities hold for large zone numbers $z$ where $F_k\approx \varphi^k/\sqrt{5}$ holds. However, on the polar points, $x_1$ and $x_N$, the zone number
\[z=\frac{\log((2N+1)\pi\sqrt{5}(1-4N^2/(2N+1)^2))}{\log\varphi^2}\to\frac{\log(2\pi\sqrt{5})}{\log\varphi^2} = 2.75...,\ \ \ \ N\to\infty\]
and writing $\textbf{c}_{k,i}$ in the form of equation (\ref{Fib basis2}) for small $k$ overestimates the length of the vector. Using equation (\ref{Fib basis}) and noticing that $k-1/2\leq z \leq k+1/2$ implies
\[\frac{\varphi^{2k-1}}{(2N+1)\pi\sqrt{5}}\leq\cos^2\phi\leq\frac{\varphi^{2k+1}}{(2N+1)\pi\sqrt{5}},\]
we have
\[|\textbf{c}_{k,i}|^2\geq \frac{4\pi^2\varphi+20\pi F_k^2\varphi^{-2k+1}}{(2N+1)\pi\sqrt{5}}=O\bigg(\frac{1}{2N+1}\bigg), \ \ \ \ \ k-1/2\leq z \leq k+1/2.\]
Since $|\textbf{c}_{k,i}|$ is the minimal separation distance for $k-1/2\leq z\leq k+1/2$, we have the correct order of separation on $\mathbb{S}^2$. By a similar computation, we have the upper bound
\[|\textbf{c}_{k,i}|\leq O\bigg(\frac{1}{\sqrt{2N+1}}\bigg), \ \ \ \ \ \ \ \ k-3/2\leq z \leq k+3/2.\]
Since the triangulation of $\omega_{2N+1}$ in each zone consists of $\textbf{c}_{k,i}$, $\textbf{c}_{k-1,i}$, and $\textbf{c}_{k+1,i}$, the covering of $\omega_{2N+1}$ is of the correct order on all of $\mathbb{S}^2$.
\qed
\
\noindent\textit{Proof of Proposition \ref{zonal quasi}.}
The diameter boundedness of the partition and Proposition \ref{lemmaprop} gives equidistribution and covering,
\[\eta(\omega_N) \leq \frac{3.5}{\sqrt{N}}.\]
In \cite{LeopardiDiam} and \cite{Zhou}, it is established that there exists $c_1,c_2>0$ such that for all partitions $P_N$, with $\phi_j$, $n$ as defined above,
\begin{equation}
\frac{c_1}{\sqrt{N}}\leq\phi_{j+1}-\phi_j\leq \frac{c_2}{\sqrt{N}},\ \ \ \ \ \ \ \ \ 0\leq j \leq n.
\label{zonal collar order}
\end{equation}
\noindent This gives the correct order of separation between collars. For neighbors $x_1,x_2 \in \omega_N$ within collar $j$, wlog suppose $\phi_j<\pi/2$. Using the fact that the normalized area of each cell can be expressed as
\[\frac{(\cos\phi_j-\cos\phi_{j+1})}{2y_i} = \frac{1}{N},\]
we have
\[|x_1-x_2|\geq \frac{2\pi\sin\phi_j}{y_i} = \frac{4\pi\sin\phi_j}{N(\cos\phi_j-\cos\phi_{j+1})}.\]
So it suffices to show there exists $c_3>0$ such that
\begin{equation}
\frac{\sin\phi_j}{\cos\phi_j-\cos\phi_{j+1}}\geq c_3\sqrt{N}\ \ \ \ \ \ \forall N, 0\leq j\leq n.
\label{zonalsep}
\end{equation}
For a fixed $h>0$ and $\phi_j\geq h$, this follows from (\ref{zonal collar order}) and the fact that $\cos$ is Lipschitz.
On the other hand, for sufficiently small $\phi_j$, there exists $c_4>0$ such that
\[\frac{\sin\phi_j}{\cos\phi_j-\cos\phi_{j+1}}\geq c_4\frac{\phi_j}{\phi_j^2-\phi_{j+1}^2}.\]
Again applying (\ref{zonal collar order}) twice,
\[\textup{RHS}\ (\ref{zonalsep})\geq \frac{c_4}{c_1}\frac{\phi_j}{\phi_j+\phi_{j+1}}\sqrt{N}\geq \frac{c_4}{c_1}\frac{\phi_j}{2\phi_j+\frac{c_2}{\sqrt{N}}}\sqrt{N}\geq c_3\sqrt{N}.\]
In the last step we used the fact that for some $c_5>0$ and all $j$
\[\phi_j\geq\cos^{-1}(1-\frac{2}{N})\geq \frac{c_5}{\sqrt{N}}.\]
\qed
\
\noindent\textit{Proof of Proposition \ref{HEALPix quasi}.}
The pixels are diameter bounded and thus by Proposition \ref{lemmaprop} the nodes are equidistributed.
To establish separation, we examine the five cases of nearest neighbor points: The points lie in 1) the polar region or 2) the equatorial region, and the points lie in a) the same ring or b) adjacent rings, and 3) the points lie in adjacent rings at the boundary of the polar region and the equatorial region.
Case 1a: If nearest neighbor points lie in the polar region along the same ring $1\leq i\leq k$ that has radius \[r_i = \sin\phi_i= \sqrt{\frac{2i^2}{3k^2} - \frac{i^4}{9k^4}}\]
and $4i$ equally spaced points, then the separation $\delta$ satisfies
\[\delta = 2r_i\sin\frac{\pi}{4i} = 2\sqrt{\frac{2i^2}{3k^2} - \frac{i^4}{9k^4}}\sin\frac{\pi}{4i}\geq\frac{\sqrt{2}}{i}\sqrt{\frac{2i^2}{3k^2} - \frac{i^4}{9k^4}} = O\bigg(\frac{1}{k}\bigg) = O\bigg(\frac{1}{\sqrt{N}}\bigg).\]
In the middle inequality we use the fact that
\begin{equation}
\sin x \geq \frac{\sqrt{2}/2}{\pi/4}x\ \ \ \ \ \ \ \ \ 0\leq x \leq \pi/4.
\label{sineineq}
\end{equation}
Case 2a: Suppose the nearest neighbor points lie in the equatorial region along the same ring. Since each ring has $4k$ points, the smallest separation occurs at the ring farthest from the equator and closest to $z = 2/3$. Using (\ref{sineineq}) again, we have
\[ \delta = 2r_i\sin\frac{\pi}{4k}\geq \frac{2\sqrt{5}}{3}\sin\frac{\pi}{4k}\geq\frac{\sqrt{10}}{3k} = O\bigg(\frac{1}{\sqrt{N}}\bigg).\]
Case 1b: We split up the rings in the polar region into the outer half, $1\leq i \leq k/2$ and the inner half, $k/2\leq i\leq k$. On the outer half, the separation between rings is
\begin{align*}
\delta \geq r_{i+1}-r_i =& \frac{(i+1)\sqrt{6k^2-(i+1)^2}-i\sqrt{6k^2-i^2}}{3k^2}\\
\geq& \frac{(k/2+1)\sqrt{6k^2-(k/2+1)^2}-(k/2)\sqrt{6k^2-(k/2)^2}}{3k^2} = O\bigg(\frac{1}{\sqrt{N}}\bigg).\\
\end{align*}
\noindent On the inner polar rings, the separation between rings is
\[\delta \geq |\cos\phi_{i+1}-\cos\phi_i| = \frac{2i+1}{3k^2}\geq \frac{k+1}{3k^2} =O\bigg(\frac{1}{\sqrt{N}}\bigg).\]
Case 2b: In the equatorial region, the ring height $z$ increases linearly with respect to the index $i$ giving
\[\delta \geq \frac{2}{3k} =O\bigg(\frac{1}{\sqrt{N}}\bigg).\]
Case 3 follows from Case 1b and 2b.
The covering of the points follows by similar geometric arguments.
\qed
\
\noindent\textit{Proof of Proposition \ref{rad icos quasi}.}
It suffices to show $\Pi$ is locally bi-Lipschitz on each face $\mathcal{F}$ of the icosahedron. If for some $\delta$, $L_1$, $L_2 > 0$
\[L_1|x-y|\leq|\Pi(x)-\Pi(y)|\leq L_2|x-y|, \ \ \ \ \ x,y\in \mathcal{F},\ \ |x-y|<\delta,\]
then
\[\gamma(\omega_N)\leq\frac{L_2}{L_1}\gamma(\widetilde{\omega_{N_k}}) = \frac{L_2}{L_1\sqrt{3}}.\]
Let $c := \min_{x \in \text{Icos}} |x| = \sqrt{(1/3+2\sqrt{5}/15)}$. For $x,y\in \mathcal{F}$ with angle $\theta$, we have
\[|x-y|\geq 2c\sin\frac{\theta}{2}.\]
Using the fact that $\sin^{-1}x\leq \frac{\pi}{2}x$ for $x\leq 1$ and $ |x-y|< 2c$,
\[|\Pi(x)-\Pi(y)|\leq \theta\leq 2 \sin^{-1}\bigg(\frac{|x-y|}{2c}\bigg)\leq\pi c |x-y|.\]
For the other inequality, wlog suppose $c\leq|x|\leq|y|\leq 1$, and consider the line $P \subset \mathcal{F}$ connecting $x$ and $y$. Denote $z$ as the projection of $0$ onto $P$. Defining $\phi := \cos^{-1}(|z|/|x|)$, we have
\[|y|-|x| = |z|\sec(\theta+\phi)-|z|\sec\phi\leq \sec(\theta+\phi)-\sec\phi.\]
Since $\phi \leq \cos^{-1}c-\theta$ and secant is convex on $(0,\pi/2)$,
\[|y|-|x|\leq\frac{1}{c}-\sec(\cos^{-1}c-\theta)=:g(\theta).\]
Thus,
\begin{align*}
|x-y|^2 &= |x|^2+|y|^2-2|x||y|\cos \theta\leq\max_{c\leq|y|\leq 1} (|y|+g)^2+|y|^2-2|y|(|y|+g)\cos\theta\\
& = (1+g)^2+1-2(1+g)\cos\theta.\\
\end{align*}
Since
\[f(\theta):=\frac{(1+g)^2+1-2(1+g)\cos\theta}{2-2\cos\theta}\]
is continuous for $\theta\in (0,\pi/2+\cos^{-1}c)$ and $\lim_{\theta\rightarrow 0} f(\theta)$ exists, there exists $L>0$ such that
\[|x-y|\leq L|\Pi(x)-\Pi(y)|, \ \ \ \ \ \ \ \ \ \theta\in (0,\pi/2+\cos^{-1}c).\]
\qed
\
\noindent\textit{Proof of Theorem \ref{oct quasi}.}
Restricting ourselves to the face of $\mathbb{K}$ with all positive coordinates, label the vertices of the partition $\left \{A_{i,j}\right \}_{0\leq i+j\leq k}$ by
\[A_{i,j} = \bigg( \frac{iL}{k\sqrt{2}}, \frac{jL}{k\sqrt{2}}, \frac{L}{\sqrt{2}}\bigg(1 - \frac{i+j}{k}\bigg)\bigg).\]
Let $\mathcal{A}_{i,j} = \mathcal{U}(A_{i,j})$. Then
\[\mathcal{A}_{i,j} = \bigg( \frac{i+j}{k}\sqrt{2-\frac{(i+j)^2}{k^2}}\cos\frac{\pi j}{2(i+j)}, \frac{i+j}{k}\sqrt{2-\frac{(i+j)^2}{k^2}}\sin\frac{\pi j}{2(i+j)}, 1-\frac{(i+j)^2}{k^2}\bigg).\]
Then
\begin{equation}
\delta (\omega_N) = \min_{i,j}\{ \|\mathcal{A}_{i+1,j}-\mathcal{A}_{i,j}\|, \|\mathcal{A}_{i,j+1} - \mathcal{A}_{i,j}\|, \|\mathcal{A}_{i+1,j} - \mathcal{A}_{i,j+1}\|\}.
\label{OctNeighbors}
\end{equation}
Adapting \cite{Oct}, we have
\[\|\mathcal{A}_{i+1,j} - \mathcal{A}_{i,j}\|^2 = 2\frac{(i+j+1)^2}{k^2}+2\frac{(i+j)^2}{k^2} - 2\frac{(i+j+1)^2(i+j)^2}{k^4} \ \ \ \ \ \ \ \ \ \ \]
\[ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ -2\frac{(i+j)(i+j+1)}{k^2}\sqrt{2-\frac{(i+j+1)^2}{k^2}}\sqrt{2-\frac{(i+j)^2}{k^2}}\cos\frac{\pi j}{2(i+j)(i+j+1)}.\]
Along the line $i+j = c$, the minimum is obtained when the cosine term is maximized, i.e. at $j=0$. Thus
\[\min_{i,j}\|\mathcal{A}_{i+1,j} - \mathcal{A}_{i,j}\|^2 = \min_{0\leq i\leq k} \|\mathcal{A}_{i+1,0} - \mathcal{A}_{i,0}\|^2\]
\[ = \frac{2}{k^2}\min_{0\leq i \leq k}\bigg((i+1)^2 +i^2+\frac{(i+1)^2i^2}{k^2} - i(i+1)\sqrt{2-\frac{(i+1)^2}{k^2}}\sqrt{2-\frac{i^2}{k^2}}\bigg)= \frac{2}{k^2}.\]
By symmetry of the above expressions in $i$ and $j$,
\[\min_{i,j}\|\mathcal{A}_{i,j+1} - \mathcal{A}_{i,j}\|^2 = \min_{0\leq j \leq k}\|\mathcal{A}_{0,j+1} - \mathcal{A}_{0,j}\|^2 = \frac{2}{k^2}.\]
Lastly,
\[\|\mathcal{A}_{i+1,j} - \mathcal{A}_{i,j+1}\|^2 = 4\frac{(i+j+1)^2}{k^2}\bigg(2 - \frac{(i+j+1)^2}{k^2}\bigg)\sin^2\frac{\pi}{4(i+j+1)}\]
which again depends on only $i+j$. Using (\ref{sineineq}) we have
\begin{align*}
\min_{i,j}\|\mathcal{A}_{i+1,j} - \mathcal{A}_{i,j+1}\|^2 &= \min_{0\leq i \leq k} \|\mathcal{A}_{i+1,0} - \mathcal{A}_{i,1}\|^2\\
&= 4 \min_{0\leq i \leq k} \frac{(i+1)^2}{k^2}\bigg(2 - \frac{(i+1)^2}{k^2}\bigg)\sin^2\frac{\pi}{4(i+1)}\\
&\geq 4\min_{0\leq i\leq k} \frac{(i+1)^2}{k^2}\bigg(2 - \frac{(i+1)^2}{k^2}\bigg)\frac{1}{2(i+1)^2}\\
&= 4\min_{0\leq i \leq k} \frac{1}{2k^2}\bigg(2 - \frac{(i+1)^2}{k^2}\bigg) = \frac{2}{k^2}\bigg(2- \frac{(k+1)^2}{k^2}\bigg).\\
\end{align*}
Thus from (\ref{OctNeighbors})
\[\delta(\omega_{4k^2+2})^2 \geq \frac{2}{k^2}\bigg(2 - \frac{(k+1)^2}{k^2}\bigg).\]
Taking the square root and substituting $N = 4k^2 +2$ gives
\[\liminf_{N\rightarrow\infty}\delta(\omega_N)\sqrt{N} \geq \sqrt{8}.\]
Finally, the diameter bound in \cite{Oct} gives an immediate upper bound for the covering radius from which (\ref{Octmesh}) follows:
\[\eta(\omega_{4k^2+2},\mathbb{S}^2) \leq \sqrt{\frac{4+\pi^2}{8k^2}}.\]
\qed
\section{Matlab Code}
Many thanks to Grady Wright for his help in implementing some of these point sets in Matlab. Code for the Fibonacci, Hammersley, HEALPix, and cubed sphere nodes authored by Wright is available at\\
\noindent \textbf{\textit{https://github.com/gradywright/spherepts}}. The approximate Coulomb, log energy, and maximum determinant points were provided from computations by Rob Womersley. Code for the zonal equal area points is available in Paul Leopardi's Recursive Zonal Equal Area Toolbox from \textbf{\textit{eqsp.sourceforge.net.}} Code for the generalized spiral, octahedral, radial icosahedral, and equal area icosahedral nodes was created by the authors and is available upon request.
\bibliographystyle{plain}
| 368445641ac4687e3277b53dc4f78abda2dd3bbe | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
The problem of finding the best packings of lines, one-dimensional subspaces of a real or complex Euclidean space, is easy to state. Despite its simple geometric formulation, it has given rise to a surprisingly diverse literature over many years,
ranging from relatively elementary, low dimensional examples \cite{Haantjes1948} to more sophisticated constructions \cite{ConwayHardinSloane1996}, some
involving combinatorial \cite{vanLintSeidel1966,LemmensSeidel1973,Neumaier1989} or group-theoretic aspects \cite{CalderbankCameronKantorSeidel1997,XiaZhouGiannakis2005}
and results on bounds on the relationship between the number of lines and achievable angles \cite{Rankin1955,Welch1974,Koornwinder1976}. Maximal sets of
equiangular lines are known to be optimal packings, but the number of lines that can be realized is hard to determine \cite{GreavesKoolenMunemasaSzollosi2016}.
Special regard has been given to the construction of complex examples, motivated by applications in quantum information theory \cite{Zauner1999}.
Numerical searches indicate that they exist in many cases \cite{SustikTroppDhillonHeath2007,ScottGrassl2010},
but a rigorous proof of their existence is restricted to low dimensions, see \cite{ScottGrassl2010} and references therein.
Next to lines, packings of higher-dimensional subspaces have also been investigated \cite{ConwayHardinSloane1996,CalderbankHardinRainsShorSloane1999}. In this case, even less seems to be known about general construction principles that realize tight bounds
\cite{BargNogin2002,Bachoc2004,Henkel2005,Bachoc2006}.
More recently, these packing problems have been studied in the context of frame theory. Apart from geometric optimality criteria, frame design aims at
tightness, which implies that the projections onto the subspaces sum to a multiple of the identity.
The case of higher-dimensional subspaces corresponds to
fusion frames. If the number of subspaces is not too large, then in close similarity to line packings, equi-distant fusion frames
present optimal solutions \cite{KutyniokPezeshkiCalderbankLiu2009}. Examples of such constructions follow similar strategies as in the frame case \cite{Hoggar1982,CalderbankHardinRainsShorSloane1999,Appleby2007,BachocEhler2013}.
For a larger number of subspaces, such equiangular arrangements cannot be realized and one needs to find an alternative bound
for the characterization of optimal packings, for example the orthoplex bound for lines or subspaces \cite{ConwayHardinSloane1996}.
In an earlier paper, we constructed optimal line packings when the number of lines goes slightly beyond the threshold beyond which equiangular
lines are impossible to realize \cite{BodmannHaas}.
In this paper, we study the orthoplex bound for subspace packings and investigate cases in which the bound is achieved while the number of subspaces is maximal.
The main results are as follows. In order to maximize the number of subspaces while achieving the orthoplex bound, the
dimension of the subspaces is necessarily half of the dimension of the ambient space and the
chordal distance between subspaces assumes only two values. Because of the relation with the orthoplex bound, we call these subspace packings maximal orthoplectic fusion frames.
The family of examples we describe here has already appeared in the literature, either as optimal real subspace packings \cite{ShorSloane1998}, whose discovery is ascribed to a ``remarkable coincidence'', or among the more general family of quantum 2-designs \cite{Zauner1999} in complex Hilbert spaces of prime power dimensions. In the complex case, it was observed that the projections are affine, Grassmannian designs \cite{Zauner1999}, see also \cite{Roy2010}, where the construction is attributed to R{\"o}tteler.
In the present paper, we examine rigidity properties in the
construction of maximal orthoplectic fusion frames obtained from the theory of packings and designs \cite{Zauner1999, Bachoc2004,Roy2010,BachocEhler2013}.
We treat the real and complex case on the same footing, involving a new construction principle. To this end, we leverage earlier constructions of orthoplex-bound achieving, optimal line packings associated with mutually unbiased bases introduced by Schwinger \cite{Schwinger1960}. Maximal sets of mutually unbiased bases are known to exist in the complex case in prime power dimensions \cite{WoottersFields1989,BoykinSitharamTiepWocjan2007,GodsilRoy2009,Appleby2009} and in the real case if the dimension is a power of four \cite{CameronSeidel1973,LeCompteMartinOwens2010}, see also \cite{BoykinSitharamTarafiWocjan}. We obtain maximal orthoplectic fusion frames
by augmenting these maximal mutually unbiased bases with block designs, subsets of the index set that satisfy certain combinatorial conditions. The designs we construct
for our purposes are known as balanced incomplete block designs and at the same time associated with optimal constant-weight binary codes \cite{Johnson1972}, see also
\cite{Sidelnikov1975}.
The resulting families of subspaces are constructed in any real
Hilbert space whose dimension is a power of four or in any complex Hilbert space whose dimension is a power of two.
This paper is organized as follows. After the introduction, we fix notation and recall known distance and cardinality bounds on fusion frames in
Section~\ref{sec:bounds}. We relate the orthoplex bound for fusion frames with the notion of mutual unbiasedness in Section~\ref{sec:muff} and study implications
for the structure of maximal orthoplectic fusion frames as packings and as Grassmannian designs.
Finally, Section~\ref{sec:moff} presents the construction of a family of maximal orthoplectic fusion frames.
\section{Distance bounds and Grassmannian fusion frames}\label{sec:bounds}
\begin{defn}
Let $l,m,n \in \mathbb N$ and let $\BB{F}$ denote the field $\mathbb R$ or $\mathbb C$. An {\bf {\boldmath$(n,l,m)$}-fusion frame} is a set $\CAL{F}=\{ P_j\}_{j=1}^n$, where each $P_j$ is an orthogonal projection onto an $l$-dimensional subspace of $\BB{F}^m$, such that there exist positive numbers $A$ and $B$ with $0<A\leq B$ for which the chain of inequalities
$$
A\|x\|^2 \leq \sum_{j=1}^n \|P_j x\|^2 \leq B\|x\|^2
$$
holds for every $x \in \BB{F}^m$. If we can choose $A=B$, then $\CAL{F}$ is {\bf tight.} If there is $C\ge 0$ such that $\mathop{\mathrm{tr}}( P_i P_j ) = C$ for each pair $i \ne j$
in the index set $[\![ n ]\!] \equiv \{1, 2, \dots, n\}$, then
$\CAL F$ is called {\bf equiangular}.
\end{defn}
By the polarization identity, the tightness property is equivalent to the fusion frame resolving the identity $I_m$ on $\BB{F}^m$ according to
$$
\frac 1 A \sum_{j=1}^n P_j = I_m \, .
$$
More general types of fusion frames are obtained by relaxing the condition that all subspaces have the same dimension and by scaling the projections with non-negative weight factors.
For any two projections $P$ and $P'$ onto $l$-dimensional subspaces of $\BB{F}^m$, the {\it chordal distance}
is defined by
$
d_c(P,P') = \frac{1}{\sqrt 2} \| P - P'\| = (l - \mathop{\mathrm{tr}}( P P' ) )^{1/2} \, .
$
In order to
characterize optimal packings with respect to $d_c$, we use an embedding that maps the projections to vectors in a higher dimensional Hilbert space.
We denote the dimension of this space as
$$d_\BB{F} (m) = \left\{ \begin{array}{cc} \frac{(m+2)(m-1)}{2}, & \BB{F}= \BB{R} \\ m^2 -1, & \BB{F} =\BB{C} \end{array} \right. .$$
\begin{thm}[\cite{ConwayHardinSloane1996}] \label{thm:embed}
If $\CAL{F} = \{P_j\}_{j=1}^n$ is an $(n,l,m)$-fusion frame, then letting
$$
V_j = \sqrt{\frac{m}{l(m-l)}} \PARENTH{P_j - \frac{l}{m} I_m }
$$
defines a set of unit-norm vectors $\{V_j \}_{j=1}^n$ in the $d_{\BB{F}}(m)$-dimensional real Euclidean space
of symmetric/Hermitian $m \times m$ matrices with vanishing trace, equipped with the Hilbert-Schmidt norm, such that the inner products satisfy
$$
\mathop{\mathrm{tr}}\PARENTH{P_j P_k}= \frac{l^2}{m} + \frac{l (m-l)}{m} \mathop{\mathrm{tr}}( V_j V_k ),
$$
for every $j,k \in [\![ n ]\!]$. Furthermore, if $\mathcal F$ forms a tight fusion frame for $\BB{F}^m$, then
$
\sum_{j=1}^n V_j =0.
$
\end{thm}
We use Rankin's distance bound for vectors on the sphere \cite{Rankin1955} in the formulation used by Conway, Hardin and Sloane \cite{ConwayHardinSloane1996}.
\begin{thm}
Let $d$ be a positive integer.
Any $n$ vectors $\{v_1, v_2, \dots, v_n\}$
on the unit sphere in $\BB{R}^d$ have a minimum Euclidean distance
$
\min_{j, k \in [\![ n ]\!], j \ne k} \|v_j - v_k \| \le \sqrt{\frac{2n}{n-1}} \, ,
$
and if equality is achieved, then $n \le d+1$ and the vectors form a simplex.
Additionally, if $n>d+1$, then the minimum Euclidean distance is
$ \min_{j, k \in [\![ n ]\!], j \ne k} \|v_j - v_k \| \le \sqrt{2},$
and if equality holds in this case,
then $n \le 2d$. Moreover, if $n=2d$, then equality holds if and only if the vectors form an orthoplex,
the union of an orthonormal basis with the negatives of its basis vectors.
\end{thm}
In terms of the inner products between $n$ unit vectors in $\BB{R}^d$, the Rankin bound is
$$
\max_{j, k \in [\![ n ]\!], j \ne k} \langle v_j, v_k \rangle \ge - \frac{1}{n-1}, \,
$$
and if $n>d+1$, then it improves to
$$
\max_{j, k \in [\![ n ]\!], j \ne k} \langle v_j , v_k \rangle \ge 0 \, .
$$
Using the embedding from Theorem~\ref{thm:embed}, we reformulate the Rankin bound for the Hilbert-Schmidt inner products of the projections of a fusion frame.
This results in a bound that has already been derived in an alternative way before \cite{KutyniokPezeshkiCalderbankLiu2009,MasseyRuizStojanoff2010}
and in an improved bound for a larger number of subspaces, as noted in \cite{Roy2010}.
\begin{cor}\label{cor_2df}
If $\CAL{F}=\{P_j\}_{j=1}^n$ is a $(n,l,m)$-fusion frame,
then
$$
\max\limits_{j,k \in[\![ n ]\!], j \neq k} \mathop{\mathrm{tr}}(P_j P_k) \geq \frac{nl^2-ml}{m(n-1)}
$$
and if equality is achieved then the fusion frame is equiangular and $n \le d_{\BB{F}}(m)+1$.
If $n > d_{\BB{F}}(m)+1$, then
$$
\max\limits_{j,k \in[\![ n ]\!], j \neq k} \mathop{\mathrm{tr}}(P_j P_k) \geq \frac{l^2}{m}
$$
and if equality is achieved then $n \le 2 d_{\BB{F}}(m)$. Moreover, if $n=2 d_{\BB{F}}(m)$, then
equality in this bound implies that for each $j \in [\![m]\!]$, $\mathop{\mathrm{tr}}(P_j P_k) =\frac{2(2l-m)}{m}$ for exactly one $k \in [\![m]\!] \setminus \{j\}$ and $\mathop{\mathrm{tr}}(P_j P_k) = \frac{l^2}{m}$ for all other $k \in [\![m]\!] \setminus \{j\}$.
\end{cor}
\begin{defn}
Let $\CAL{F} =\CBRACK{P_j}_{j=1}^n$ be an $(n,l,m)$-fusion frame. If $\CAL{F}$ is a solution to the subspace packing problem,
that is, it minimizes $\max_{j \ne k} \mathop{\mathrm{tr}}( P_j P_k)$ among all $(n,l,m)$-fusion frames, then it is called a {\bf Grassmannian} fusion frame.
If $\CAL{F}$ is Grassmannian, then it is called {\bf orthoplex-bound achieving} if $n > d_{\BB{F}}(m)+1$ and
$$
\max\limits_{j,k \in[\![ n ]\!], j \neq k} \mathop{\mathrm{tr}}(P_j P_k) = \frac{l^2}{m}.
$$
If $\CAL{F}$ is orthoplex-bound achieving and $n=2 d_{\BB{F}}(m)$, then $\CAL{F}$ is referred to as a {\bf maximal orthoplectic fusion frame}.
\end{defn}
We wish to construct maximal orthoplectic fusion frames, which means the projections must embed exhaustively into the vertices of an orthoplex in $\BB{R}^{d_{\BB{F}}(m)}$.
The feasability of this depends on whether the embedding admits antipodal points in the higher dimensional
Euclidean sphere, which in turn depends on the relationship between $l$ and $m$.
\begin{prop}\label{prop_need_half_ss}
If $\CAL{F}=\{P_j\}_{j=1}^n$ is an $(n,l,m)$-fusion frame and $\CAL{V} = \{V_j\}_{j=1}^n \subset \BB{R}^{d_{\BB{F}}(m)}$ are the embedded vectors obtained from Theorem~\ref{thm:embed},
then
$$
\min\limits_{j,k \in[\![ n ]\!], j \neq k}\langle V_j, V_k \rangle \geq -\frac{l}{m-l}
$$
for every $j,k \in [\![ n ]\!]$. In particular, the embedding admits antipodal points only if $l \geq \frac{m}{2}$.
\end{prop}
\begin{proof}
This follows immediately from the inequality
$$
0 \leq \min\limits_{j,k \in[\![ n ]\!], j \neq k} \mathop{\mathrm{tr}}\PARENTH{P_j P_k}= \frac{l^2}{m} + \frac{l (m-l)}{m} \min\limits_{j,k \in[\![ n ]\!], j \neq k} \langle V_j, V_k\rangle.
$$
\end{proof}
Partitioning a maximal orthoplectic fusion frame into orthogonal pairs of projections shows, together with the preceding proposition,
that $m=2l$.
\begin{cor} \label{cor:half_ss}
If $\mathcal F=\{P_j\}_{j=1}^n$ is a maximal orthoplectic $(n,l,m)$-fusion frame,
then $m=2l$.
\end{cor}
Maximal orthoplectic fusion frames enjoy another property that has been studied in the literature: they are part of a family of
{\it Grassmannian 2-designs}, as shown by Zauner \cite{Zauner1999}. For our purposes, this is important because it imposes more rigidity
on their construction.
We follow Zauner's convention for the definition of these designs.
\begin{defn}
An $(n,l,m)$-fusion frame $\CAL{F}=\{P_j\}_{j=1}^n$ is a {\bf Grassmannian $t$-design} if
$$
\sum_{j=1}^n P_j^{\otimes t} = \sum_{j=1}^n (U P_j U^*)^{\otimes t}
$$
for any orthogonal matrix or unitary $U$ in the real or complex case, respectively,
where $P_j^{\otimes t} $ is the $t$-fold Kronecker/tensor product of $P_j$ with itself and $U^*$ is the adjoint of $U$ or the
transpose of $U$ in the real case.
\end{defn}
Equivalently, the
right-hand side of the defining identity can be averaged with respect to the Haar measure
$\mu$ on the group $\CAL U$ of orthogonal or unitary $m \times m$ matrices. This formulation implies a simple characterization
of the design property based on the {\it $t$-coherence tensor}
$$
K_{t,l,m} = \int_{\CAL{U}} (U P U^*)^{\otimes t} d\mu(U) \, ,
$$
where $P$ is any rank-$l$ orthogonal projection matrix. Because of the analogy
with bounds for constant-weight codes \cite{Sidelnikov1975},
Zauner calls the following estimate a generalized Sidelnikov inequality.
\begin{thm}[Zauner, Theorem 2.5 of \cite{Zauner1999} ]\label{thm:ZaunerSidelnikov}
Let $\CAL{F}=\{P_j\}_{j=1}^n$ be an
$(n,l,m)$-fusion frame, then
$$
\frac{1}{n^2} \sum_{i,j=1}^n (\mathop{\mathrm{tr}}(P_i P_j))^t \ge \mathop{\mathrm{tr}} (K_{t,l,m}^2)
$$
and equality holds if and only if $\CAL F$
is a Grassmannian $t$-design.
\end{thm}
\begin{proof}
Let $C= \sum_{j=1}^n P_j^{ \otimes t } - n K_{t,l,m}$, then
$\mathop{\mathrm{tr}}( C^2) = \sum_{i,j=1}^n (\mathop{\mathrm{tr}}(P_i P_j))^t - n^2 \mathop{\mathrm{tr}}( K_{t,l,m}^2 ) \ge 0$
and cases of equality are characterized by $C=0$, which is the Grassmannian $t$-design property.
\end{proof}
Because it is of independent interest, we study the $t$-coherence tensor and derive an alternative
proof of the Grassmannian 2-design property of maximal orthoplectic fusion frames.
\begin{prop}\label{prop_K2lm}With respect to a fixed orthonormal basis $\{e_j\}_{j=1}^m$ for $\BB F^m$, the $2$-coherence tensor for an $(n,l,m)$-fusion frame can be
reduced to the case $l=1$ by
$$
K_{2,l,m} = \PARENTH{ l - \frac{l(l-1)}{m-1}} K_{2,1,m} + \frac{l(l-1)}{m(m-1)} I_m \otimes I_m,
$$
and for $l=1$ it is given by
$$
K_{2,1,m} = a \sum_{j=1}^m E_{j,j} \otimes E_{j,j} + \sum_{\substack{j,j'=1\\ j \ne j'}}^m \PARENTH{b (E_{j,j'} \otimes E_{j',j} + E_{j,j} \otimes E_{j',j'}) + c E_{j,j'} \otimes E_{j,j'}},
$$
where we abbreviate $E_{j,j'} = e_j \otimes e_{j'}^*$ and
$a= \frac{d_{\BB{F}}(m)+ (m-1)^2}{m^2 d_{\BB{F}}(m)}$, $b=c=a/3$ if $\BB F = \BB R$ or $b=a/2$, $c=0$ if $\BB F = \BB C$.
\end{prop}
\begin{proof}
In the special case $l=1$, by the normalization $\mathop{\mathrm{tr}}( K_{2,1,m}) = 1$ and the invariance properties of $K_{2,1,m}$
under the tensor representation of the orthogonal or unitary group \cite[Ch.\ 7-9]{Ma2007},
we obtain
$$
K_{2,1,m} = a \sum_{j=1}^m E_{j,j} \otimes E_{j,j} + \sum_{\substack{j,j'=1\\ j \ne j'}}^m \PARENTH{ b (E_{j,j'} \otimes E_{j',j} + E_{j,j} \otimes E_{j',j'}) + c E_{j,j'} \otimes E_{j,j'}}.
$$
Next, we use the symmetrization identity \cite{Zauner1999,RoyScott2007}
\begin{align*}
I_m \otimes I_m & = \sum_{j,j'=1}^m \int_{\CAL U} U E_{j,j} U^* \otimes U E_{j',j'} U^* d\mu(U) \\
& = m K_{2,1,m} + m(m-1) \int_{\CAL U} U E_{1,1} U^* \otimes U E_{2,2} U^* d\mu(U) \, .
\end{align*}
Repeating this type of expansion for $K_{2,l,m}$ results in
\begin{align*}
K_{2,l,m} & = \sum_{j,j'=1}^l \int_{\CAL U} U E_{j,j} U^* \otimes U E_{j',j'} U^* d\mu(U)\\
& = l K_{2,1,m} + l(l-1) \int_{\CAL U} U E_{1,1} U^* \otimes U E_{2,2} U^* d\mu(U) \\
& = l K_{2,1,m} + \frac{l(l-1)}{m(m-1)} ( I_m \otimes I_m - m K_{2,1,m} ) \, .
\end{align*}
Rearranging terms gives the claimed expression for $K_{2,l,m}$.
\end{proof}
%
%
%
Next, we use the linear relation between the eigenvalues of $K_{2,l,m}$ and $K_{2,1,m}$ to obtain as
a corollary the expression for the squared Hilbert-Schmidt norm of the $2$-coherence tensor for general $l \in \mathbb N$,
which had been computed by Zauner without determining $K_{2,l,m}$ explicitly \cite[Lemma 2.7]{Zauner1999}.
\begin{cor}\label{cor_K2lm_val}
Given positive integers $l$ and $m$ with $l \leq m$, the squared Hilbert-Schmidt norm of the $2$-coherence tensor is
$$\displaystyle
\mathop{\mathrm{tr}}(K_{2,l,m}^2) = \frac{l^4}{m^2} + \frac{l^2(m-l)^2}{d_{\BB{F}}(m) m^2}.
%
$$
\end{cor}
\begin{proof}
Using Proposition~\ref{prop_K2lm}, a straightforward computation for the case $l=1$ shows that, for both the real and complex case, we have
$$
\mathop{\mathrm{tr}}( K_{2,1,m}^2 ) = \frac{d_{\BB{F}}(m)+ (m-1)^2}{m^2 d_{\BB{F}}(m)}.
$$
Next, the identity relating $K_{2,l,m}$ and $K_{2,1,m}$ from Proposition~\ref{prop_K2lm} gives
$$
\mathop{\mathrm{tr}}( K_{2,l,m}^2 ) = \PARENTH{l - \frac{l(l-1)}{m-1}}^2 \mathop{\mathrm{tr}}( K_{2,1,m}^2 ) + 2 \PARENTH{l - \frac{l(l-1)}{m-1}} \frac{l(l-1)}{m(m-1)} + \frac{l^2 (l-1)^2}{(m-1)^2} \, .
$$
Inserting the value of the squared Hilbert-Schmidt norm for $K_{2,1,m}$ and simplifying this expression gives
$$
\mathop{\mathrm{tr}}( K_{2,l,m}^2 ) = \frac{1}{d_{\BB{F}}(m) m^2 (m-1)} \PARENTH{d_{\BB{F}}(m) l^4(m-1) + (m-1) l^2 (m-l)^2} = \frac{l^4}{m^2} + \frac{l^2(m-l)^2}{d_{\BB{F}}(m) m^2} \, .
$$
\end{proof}
Next, we verify that maximal orthoplectic fusion frames are Grassmannian $2$-designs.
\begin{thm} Given a maximal orthoplectic $(2d_{\BB{F}}(m),m/2,m)$-fusion frame $\CAL{F} = \{P_j \}_{j=1}^{2d_{\BB{F}}(m)}$, then equality holds in the generalized Sidelnikov inequality
in Theorem~\ref{thm:ZaunerSidelnikov}
and
the fusion frame is a Grassmannian 2-design.
\end{thm}
\begin{proof}
We first compute the value of the lower bound from Theorem~\ref{thm:ZaunerSidelnikov},
$$
\mathop{\mathrm{tr}} ( K_{2,m/2,m}^2 ) = \frac{m^2(d_{\BB{F}}(m) + 1)}{16 d_{\BB{F}}(m)} \, .
$$
Next, we use Corollary~\ref{cor_2df} to compute the average squared inner product for the projection matrices,
$$
\frac{1}{4d_{\BB{F}}(m)^2} \sum_{j,j'=1}^{2d_{\BB{F}}(m)} (\mathop{\mathrm{tr}}( P_j P_{j'} ) )^2 = \frac{1}{2d_{\BB{F}}(m)} \PARENTH{ l^2 + \frac{l(2l-m)}{m} + (2d_{\BB{F}}(m) -2) \frac{l^4}{m^2} } = \mathop{\mathrm{tr}} ( K_{2,m/2,m}^2 ),
$$
where the last equality follows by $l=m/2$. As stated in Theorem~\ref{thm:ZaunerSidelnikov}, this characterizes Grassmannian 2-designs.
\end{proof}
\section{Mutually unbiased bases and fusion frames}\label{sec:muff}
\begin{defn}
If $\CAL{B}=\{b_j\}_{j=1}^m$ and $\CAL{B'}=\{b'_j\}_{j=1}^m$ are a pair of orthonormal bases for $\BB{F}^m$, then they are {\bf mutually unbiased} if
$$
|\langle b_j, b'_{j'} \rangle|^2 = \frac{1}{m}
$$
for every $j,j' \in[\![m]\!]$. A collection of orthonormal bases $\{\CAL{B}_k\}_{k \in K}$ is called a set of {\bf mutually unbiased bases} if the pair $\CAL{B}_k$ and $\CAL{B}_{k'}$
is mutually unbiased for every $k \ne k'$.
\end{defn}
The number of mutually unbiased bases is bounded in terms of the dimension $m$.
\begin{thm}[Delsarte, Goethals and Seidel \cite{DelsarteGoethalsSeidel1975}] \label{th_mub_bd}\label{thm:DGS}
Let $\BB{F}^m$ be a real or complex Hilbert space and let
$\{\CAL{B}_k \}_{k \in K}$ be a set of mutually unbiased bases for $\BB{F}^m$, where $\CAL{B}_k = \CBRACK {b_j^{(k)} }$ for each $k \in K$, and let $r=|K|$. If $\BB{F} = \BB R$, then $r \leq m/2+1$
and if $\BB F = \BB C$, then $r \le m+1$.
If equality is achieved in either case, then the real span of
the corresponding projection operators, $ \CBRACK{b_j^{(k)} \otimes \PARENTH{b_j^{(k)}}^* : j\in[\![m]\!], k \in K}$,
is that of
all symmetric or Hermitian operators on $\BB{F}^m$, respectively.
\end{thm}
\begin{proof}
After selecting an appropriate ordering on the vectors,
the Gram matrix of the corresponding rank-one projection operators
has the form
$G=I_{r}\otimes I_m+ \frac{1}{m}(J_{r}-I_{r})\otimes J_m,$ where
$I_{r}$ and $I_m$ are the $r \times r$ and $m\times m$ identity matrices, respectively, and
$J_{r}$ and $J_m$ denote the $r\times r$ and $m\times m$ matrices containing only $1$'s.
The kernel of the Gram matrix is the space of vectors $a\otimes b$ such that
$J_m b= m b$ and $J_{r} a=0$, so it is $\PARENTH{r-1}$-dimensional. Consequently, the rank of $G$
and thus the real dimension of the span of the rank one projections is $rm-r+1$.
This shows the claimed bound on $r$ and that the maximal rank is achieved if and only if
$r =m+1$ in the complex case or
$r=\frac{m}{2}+1$ in the real case, because the rank of the Gram matrix equals the dimension of the span of the underlying projections.
\end{proof}
For the real case, it is known that, for most values of $m$, the maximal number of mutually unbiased bases is less than or equal to $3$ \cite{LeCompteMartinOwens2010}; however,
if $m$ is a power of $4$, then examples exist that achieve the bound in Theorem~\ref{th_mub_bd} \cite{CameronSeidel1973}.
In the complex case, the bound is achieved if $m$ is a prime power \cite{WoottersFields1989}.
\begin{thm}[Cameron and Seidel \cite{CameronSeidel1973}, Wootters and Fields \cite{WoottersFields1989}]\label{th_prime_mubs_exist}
If $m$ is a prime power, then a family of $m+1$ mutually unbiased bases for $\BB{C}^m$ exists.
If $m$ is a power of 4, then a family of $m/2+1$ mutually unbiased bases exist for $\BB{R}^m$.
\end{thm}
Henceforth, we abbreviate $k_{\mathbb R}(m) = m/2+1$ and $k_{\mathbb C}(m) = m+1$.
The rank one orthogonal projections corresponding to maximal sets of mutually unbiased bases give rise to Grassmannian 2-designs.
\begin{prop}\label{lem_MUBs_2tight}
If $\{\CAL{B}_k\}_{k \in K}$ is a set of mutually unbiased bases for $\BB{F}^m$ with $|K| = k_{\BB{F}}(m)$, then
the family of rank-one projections
$\CAL{F} = \CBRACK{b_j^{(k)} \otimes \PARENTH{b_j^{(k)}}^* : k \in K, j\in[\![m]\!]}$ forms a
Grassmannian 2-design.
\end{prop}
\begin{proof}
We only need to compare both sides of the inequality from Theorem~\ref{thm:ZaunerSidelnikov}.
To evaluate the left-hand side, we observe that the Hilbert-Schmidt inner product
is expressed in terms of the basis vectors as $\PARENTH{\mathop{\mathrm{tr}} \PARENTH{ P_j^{(k)} P_{j'}^{(k')} }}^2 = \left| \langle b_j^{(k)}, b_{j'}^{(k')}\rangle \right|^4$.
Given any fixed basis vector $b_{j}^{(k)}$, the fourth power of the absolute value of its inner product with the other vectors in the set are $0$, which occurs $m-1$ times, $1/m^2$, which occurs $(k_{\BB{F}}(m)-1)m$ times, and $1$, which occurs once.
Averaging these gives
$$
\frac{1}{m^2 k_{\BB{F}}(m)^2} \sum_{j,j'=1}^m \sum_{k,k'=1}^{k_{\BB{F}}(m)} \PARENTH{\mathop{\mathrm{tr}}\PARENTH{P^{(k)}_j P^{(k')}_{j'}}}^2 = \frac{k_{\BB{F}}(m)+ m-1}{m^2 k_{\BB{F}}(m)} \, .
$$
Comparing with the value of $\mathop{\mathrm{tr}}( K_{2,l,m}^2 )$ in the special case $l=1$ and using $d_{\BB{F}}(m) = (m-1)k_{\BB{F}}(m) $ shows that
equality holds in the inequality in Theorem~\ref{thm:ZaunerSidelnikov}.
\end{proof}
The version of the orthoplex bound for projections motivates the notion of mutual unbiasedness for fusion frames.
\begin{defn}
If $\CAL{F}=\{P_j\}_{j=1}^n$ is an $(n,l,m)$-fusion frame and $\CAL{F'}=\{P'_j\}_{j=1}^{n'}$ is an $(n',l,m)$-fusion frame, then $\CAL F$ and $\CAL F'$ are {\bf mutually unbiased} if
$$
\mathop{\mathrm{tr}}(P_j P'_{j'}) = \frac{l^2}{m}
$$
for every $j \in[\![ n ]\!]$ and $j'\in[\![n']\!]$.
A collection of fusion frames
$\CBRACK{\CAL F_k}_{k \in K}$ for
$\BB F^m$, where each $\CAL F_k$ consists of projections onto $l$-dimensional subspaces, is a set of
{\bf mutually unbiased fusion frames} if the pair $\CAL F_k$ and $\CAL F_{k'}$ is mutually unbiased for every $k \neq k'$.
\end{defn}
Given a subset of a fixed orthonormal basis, the orthogonal projection onto the span is given by the sum of the corresponding rank one projections. Projections formed in this way are called {\em coordinate projections.}
\begin{defn}
Given an orthonormal basis $\CAL{B} = \{b_j\}_{j=1}^m$ for $\BB{F}^m$ and a subset $\CAL J \subset [\![m]\!]$, the {\bf $\CAL J$-coordinate projection} with respect to $\CAL{B}$ is
$$P_{\CAL J} =\sum\limits_{j \in \CAL J} b_j \otimes b_j^*.$$
\end{defn}
Given a pair of mutually unbiased bases, then one can select coordinate projections from the respective bases to form mutually unbiased fusion frames.
\begin{prop}\label{prop_MUBs_give_MUFFs}
If $\CAL{B} = \CBRACK{b_j}_{j=1}^m$ and ${\CAL{B}'}=\CBRACK{{b'}_j}_{j=1}^{m}$ are a pair of mutually unbiased bases for $\BB{F}^m$ and $\CAL J, \CAL{J}' \subset [\![m]\!]$ with $l=|\CAL J|=|\CAL J'|$, then
$
\mathop{\mathrm{tr}}\PARENTH{P_{\CAL J} {P'}_{\CAL J'}} = \frac{l^2}{m},$
where
$P_{\CAL J}$ is the $\CAL J$-coordinate projection with respect to $\CAL{B}$ and ${P'}_{\CAL J'}$ is $\CAL J'$-coordinate projection with respect to ${\CAL{B}'}.$ Moreover, if $\CAL{F} = \{ P_{\CAL J}\}_{\CAL J \in \BB{S}}$ is a set of coordinate projections with respect to $\CAL{B}$, $\CAL{F}' = \{ {P'}_{\CAL J'}\}_{\CAL J' \in \BB{S}'}$ is a set of coordinate projections with respect to $\CAL{B}'$, $\CAL{F}$ is an $(\left| \BB{S} \right|,l,m)$-fusion frame and $\CAL{F}'$ is a $(\left| \BB{S}'\right|,l,m)$-fusion frame, then $\CAL{F} \cup \CAL{F}'$ is mutually unbiased.
\end{prop}
\begin{proof}
We compute
$$
\mathop{\mathrm{tr}}(P_{\CAL J} {P'}_{\CAL J'})
= \sum\limits_{j \in \CAL J, j \in \CAL J'} \mathop{\mathrm{tr}}
\PARENTH{
b_j \otimes \PARENTH{b_j}^* {b'}_{j'} \otimes \PARENTH{ {b'}_{j'}}^*
}
= \sum\limits_{j \in \CAL J, j' \in \CAL J'} \left|\left\langle b_j, {b'}_{j'} \right\rangle \right|^2= \frac{l^2}{m}
.
$$
The claim about mutual unbiasedness follows directly from this computation.
\end{proof}
We repeat the embedding of fusion frames for the special case of coordinate projections. Tight fusion frames of coordinate projections have also
been investigated as commutative quantum designs by Zauner \cite{Zauner1999}. We first focus on the structure of optimal packings of coordinate projections.
\begin{thm} \label{thm:coordembed}
Let $\CAL{B} = \{b_j\}_{j=1}^m$ be an orthonormal basis for $\BB{F}^m$ and let $\BB{S} = \{\CAL{J}_j\}_{j=1}^n$ be a set of subsets of $[\![m]\!]$, each of size $|\CAL{J}_j|=l$. If $\CAL{F}=\{P_j\}_{j=1}^n$ is a family of projections
for which $P_j$ is the $\CAL{J}_j$-coordinate projection with respect to $\CAL{B}$,
then the set of unit vectors $\CAL{V} =\{V_j\}_{j=1}^n$
obtained as in Theorem~\ref{thm:embed} resides in a $m-1$-dimensional subspace
of the real Euclidean space of symmetric matrices and
$$
\mathop{\mathrm{tr}}\PARENTH{P_j P_{j'}}= \frac{l^2}{m} + \frac{l (m-l)}{m} \langle V_j, V_{j'}\rangle,
$$
for every $j,j' \in [\![ n ]\!]$.
\end{thm}
\begin{proof}
By definition, $\CAL{F}$ is a set of rank-$l$ orthogonal projections that can be regarded as diagonal matrices when represented in the basis $\CAL{B}$.
The mapping
$
P_j \mapsto V_j := \sqrt{\frac{m}{l(m-l)}} (P_j - \frac{l}{m} I_m ) $
embeds the projections into the real diagonal matrices with zero trace. This implies
$
\dim\bigl(\text{span} \CBRACK{V_j}_{j=1}^n \bigr) \leq m-1 .
$
The identity for the Hilbert-Schmidt inner products follows directly from the definition of $\{V_j\}_{j=1}^n$.
\end{proof}
In this special case, the Rankin bound can be expressed in terms of the subsets indexing the coordinate projections,
because $\mathop{\mathrm{tr}}( P_{\CAL{J}} P_{\CAL{J}'} ) = |\CAL{J} \cap \CAL{J}'|$ for any $\CAL J, \CAL{J}' \subset [\![m]\!]$.
A more general bound of this type has been derived by Johnson in the context of {\it constant-weight codes} \cite[Inequality (14)]{Johnson1972}.
\begin{cor} \label{cor:1designbd}
If $\BB{S}$ is a collection of $n$ subsets of $[\![m]\!]$ for which $n>m$ and each $\CAL J \in \BB{S}$
has size $|\CAL J|=l$, then
$$
\max_{\CAL J , \CAL{J}' \in \BB{S}, \CAL J \ne \CAL{J}'} |\CAL J \cap \CAL{J}' | \ge \frac{l^2}{m} \, .
$$
Moreover, if $n=2(m-1)$ and equality holds in this bound, then $\BB{S}$ can be partitioned into $m-1$ disjoint pairs of
subsets of size $m/2$.
\end{cor}
In light of Corollary~\ref{cor:1designbd} and Proposition~\ref{prop_MUBs_give_MUFFs}, we pursue the idea of generating tight, orthoplex-bound achieving fusion frames by using coordinate projections from {\it maximal sets of mutually unbiased bases}, which are sets of mutually unbiased bases that achieve the cardinality bound in Theorem~\ref{thm:DGS}.
In order to construct an orthoplex-bound achieving fusion frame,
we need
$n > d_{\BB{F}}(m) +1$ subspaces. Thus, given a maximal set of mutually unbiased bases, then we need a sufficient number of coordinate projections per basis with low Hilbert-Schmidt inner products.
In order to bound the inner products between coordinate projections corresponding to a given orthonormal basis, the
intersection of any two different index sets $\CAL J$ and $\CAL{J}'$
must have a small intersection, which we call a cohesiveness bound.
According to Corollary~\ref{cor:1designbd}, the maximum number of such subsets whose intersections are at most of
size $l^2/m$ is $2(m-1)$.
\begin{defn}
Let $ \BB{S}$ be a collection of subsets of $[\![m]\!]$, each $\CAL{J} \in \BB{S}$ of size $l$.
We say that $\BB{S}$ is {\bf $c$-cohesive} if there exists $c>0$ such that
$$
\max\limits_{\substack{\CAL J, \CAL{J}' \in \BB{S} \\ \CAL J \neq \CAL{J}' }} \left| \CAL J \cap \CAL{J}' \right| \leq c.
$$
If $\BB{S}$ is $l^2/m$-cohesive and $| \BB{S}| = 2(m-1)$, then it is {\bf maximally orthoplectic}.\end{defn}
\begin{thm}\label{th_costruct_OGFF}
Let $ \BB{S}$ be an $l^2/m$-cohesive collection of subsets of $[\![m]\!]$, where each $\CAL{J} \in \BB{S}$ is of size $l$, let $\{ \CAL{B}_k \}_{k \in K}$ be a set of mutually unbiased bases for $\BB{F}^m$, where $|K| | \BB{S}| > d_{\BB{F}}(m)+1$ and $\CAL{B}_k = \CBRACK{ b_j^{(k)} }_{j=1}^m$ for each $k \in K$, and let $n = | K | | \BB{S}|$. If
the set
$
\CAL{F} = \CBRACK{ P_{\CAL J}^{(k)} : k \in K, \CAL J \in \BB{S}}
$
forms an $(n,l,m)$-fusion frame, where each
$P_{\CAL J}^{(k)}$ denotes the $\CAL J$-coordinate projection with respect to $\CAL{B}_k$, then
$\CAL F$ is an orthoplex-bound achieving $(n,l,m)$-fusion frame.
Moreover, if $([\![m]\!], \BB{S})$ is maximally orthoplectic and if $|K|= k_{\BB{F}}(m)$, then the set $\CAL{F}$ is a tight, maximal orthoplectic fusion frame.
\end{thm}
\begin{proof}
The cardinality requirement in the definition of orthoplex-bound achieving fusion frames is satisfied since $n > d_{\BB{F}}(m) +1$.
Let $k, k' \in K$.
If $k \neq k'$, then
$$
\mathop{\mathrm{tr}}\PARENTH{ P_{\CAL J}^{(k)} P_{\CAL{J}'}^{(k')} } = \frac{l^2}{m}
$$
for every $\CAL J, \CAL{J}' \in \BB{S}$ by
Proposition~\ref{prop_MUBs_give_MUFFs}.
If $k = k'$, then the fact that $\BB{S}$ is $l^2/m$-cohesive yields
\begin{align*}
\max\limits_{\substack{ \CAL J, \CAL{J}' \in \BB{S} \\ \CAL J \neq \CAL{J}'}}\mathop{\mathrm{tr}}\PARENTH{ P_{\CAL{J}}^{(k)} P_{\CAL{J}'}^{(k)} }
&=
\max\limits_{\substack{ \CAL J, \CAL{J}' \in \BB{S} \\ \CAL J \neq \CAL{J}'}}
\left| \CAL J \cap \CAL{J}' \right|
\leq
\frac{l^2}{m},
\end{align*}
which shows that $\CAL{F}$ is an orthoplex-bound achieving fusion frame.
Finally, if $\BB{S}$ is maximally orthoplectic and the set of mutually unbiased bases is maximal, then
Corollary~\ref{cor:1designbd} shows that the coordinate projections belonging to each basis sum to a multiple of the identity, so the corresponding fusion frame is tight.
Hence, the union of all the coordinate projections belonging to the mutually unbiased bases
forms a set of $n = k_{\BB{F}}(m) \left(2 m-2\right)= 2d_{\BB{F}}(m)$ orthogonal projections whose pairwise inner products
are bounded by $l^2/m$, which shows that $\CAL{F}$ is a maximal orthoplectic fusion frame.
\end{proof}
Following Zauner's ideas, we repeat the study of design properties for the special case of a fusion frame formed by coordinate projections.
To this end, we define the {\it diagonal coherence tensor},
$$
D_{t,l,m} = \frac{1}{\left({m \atop l }\right)} \sum_{\CAL J \in \BB J} D_{\CAL J}^{\otimes t} \, ,
$$ where $\BB J$ is the set of all subsets of $[\![m]\!]$ of size $l$,
and, for each $\CAL{J} \in \BB J$, $D_{\CAL J}$ is the $\CAL J$-coordinate projection with respect to the canonical basis.
An elementary counting argument shows $D_{1,l,m} = \frac{l}{m} I_m$ and
$$
D_{2,l,m} = \frac{l}{m}\sum_{j=1}^m E_{j,j} \otimes E_{j,j} + \frac{l(l-1)}{m(m-1)} \sum_{\substack{j,j'=1\\j \ne j'}}^m E_{j,j} \otimes E_{j',j'}, \,
$$
where $\{e_j\}_{j=1}^m$ denotes the canonical basis for $\BB F^m$ and $E_{j,j'}=e_j \otimes e_{j'}^*$ for each $j,j' \in [\![m]\!]$.
By squaring the diagonal entries and summing, we compute
$$
\mathop{\mathrm{tr}} (D_{2,l,m}^2 ) = m \frac{l^2}{m^2} + m(m-1) \frac{l^2(l-1)^2}{m^2 (m-1)^2} = \frac{l^2}{m (m-1)}(l^2 - 2l + m) \, .
$$
With this notation, the combinatorial notion of a {\it block $t$-design} is characterized conveniently.
\begin{defn}
A {\bf $t$-$(m,l, \lambda)$ block design} $\BB{S}$ is a collection of subsets of $[\![m]\!]$ called {\bf blocks}, where each block $\CAL{J} \in \BB S$ has cardinality $l$,
such that every subset of $[\![m]\!]$ with cardinality $t$ is contained in exactly $\lambda$ blocks.
When the parameters are not important or implied by the context, then $\BB{S}$ is also referred to as a {\bf block $t$-design}.
The special case of a block $2$-design is also referred to as a {\bf balanced incomplete block design}.
\end{defn}
\begin{thm}[Zauner {\cite[Theorem 1.12]{Zauner1999}}]\label{thm:diagdesigns}
A collection $\BB{S}$ of subsets of $[\![m]\!]$, where each $\CAL J \in \BB{S}$ has size $l$, is a $t$-$(m,l, \lambda)$ block design if and only if
$$
\frac{1}{|\BB{S}|} \sum_{\CAL J \in \BB{S}} D_{\CAL J}^{\otimes t} = D_{t,l,m}, \,
$$
with
$
\lambda = |\BB{S} | \mathop{\mathrm{tr}} \PARENTH{ D_{t,l,m} \bigotimes_{s=1}^t E_{s,s}},
$
where $\{e_j\}_{j=1}^m$ denotes the canonical basis for $\BB F^m$ and $E_{s,s}=e_s \otimes e_{s}^*$ for each $s \in [\![m]\!]$.
\end{thm}
\begin{proof}
The definition of $\BB{J}$ implies that any subset of $[\![m]\!]$ of size $t$ is contained in a fixed number of sets from $\BB J$.
Since both sides of the claimed identity are diagonal in the standard basis, the block design property is a consequence of
the fact that for any subset $\{j_1, j_2, \dots, j_t\} \subset [\![m]\!]$,
$P_{\CAL J}^{\otimes t}$ has an eigenvector $ e_{j_1} \otimes e_{j_2} \otimes \cdots \otimes e_{j_t} $
corresponding to eigenvalue one
if and only if $\{j_1, j_2, \dots, j_t\} \subset \CAL J$; otherwise, it corresponds to eigenvalue zero. The claimed value for $\lambda$ follows from an elementary counting argument.
\end{proof}
In the special case where $t=1$ in Theorem~\ref{thm:diagdesigns},
we obtain the correspondence between the block 1-design property of $\BB{S}$ and tightness of the corresponding fusion frame of coordinate projections.
\begin{cor}\label{cor:prop_1designs}
If $\CAL{B}$ is any orthonormal basis for $\BB{F}^m$, then
a set of coordinate projections $\{P_{\CAL J}\}_{\CAL J \in \BB{S}}$ with respect to $\CAL B$ is a tight fusion frame if and only if $\BB{S}$
is a {\it $1$-design}.
\end{cor}
Given any positive integers $l$ and $m$ with $l \leq m$, one can choose the set of all blocks of size $l$ from $[\![m]\!]$ to form a tight fusion frame in this way.
\begin{ex}
If $ \BB{S} = \CBRACK{\CAL J : \CAL J \subset [\![m]\!], |\CAL J|=l }$, the set of all blocks of size $l$, then $\BB{S}$ forms a $t$-$(m,l, \lambda)$ block design.
Given
an orthonormal basis $\CAL{B}$ for $\BB F^m$, then the corresponding set of coordinate projections with respect to $\CAL{B}$, $\CAL{F} = \{P_{\CAL J}\}_{\CAL J \in \BB{S}}$, forms a tight $(n,l,m)$-fusion frame by Corollary~\ref{cor:prop_1designs}, where $n =\left| \BB{S} \right| = \PARENTH{{ m \atop l }}$.
\end{ex}
With the same proof as in Theorem~\ref{thm:ZaunerSidelnikov}, we obtain an analogous characterization of block $t$-designs.
\begin{cor} \label{cor:diagdesigns}
Given a collection $\BB{S}$ of subsets of $[\![m]\!]$, where each $\CAL J \in \BB{S}$ has size $l$, then
$$
\frac{1}{|\BB{S}|^2} \sum_{\CAL J, \CAL{J}' \in \BB{S}} | \CAL J \cap \CAL{J}' |^t \ge \mathop{\mathrm{tr}} ( D_{t,l,m}^2 )
$$
and equality holds if and only if $ \BB{S}$ is a $t$-$(m,l,\lambda)$ block design with
$
\lambda
$
as in Theorem~\ref{thm:diagdesigns}.
\end{cor}
We can now deduce that subsets of coordinate projections
in
a maximal orthoplectic fusion frame constructed from a
maximal set of mutually unbiased bases realize block $2$-designs. We state this more generally as a correspondence between
Grassmannian 2-designs and balanced incomplete block designs.
\begin{thm} \label{thm:GdesignBIBD}
Let $\{\CAL{B}_k \}_{k \in K}$ be a maximal set of mutually unbiased bases for $\BB F^m$, so $|K|=k_{\BB{F}}(m)$ and let $\BB{S} \subset [\![m]\!]$ be a collection of subsets, each with size $l$.
If $\CAL{F} = \CBRACK{ P_{\CAL J}^{(k)}: k \in K, \CAL J \in \BB{S} }$ is a $(2d_{\BB{F}}(m), m/2, m)$-fusion frame, where each $P^{(k)}_{\CAL J}$ is the $\CAL J$-coordinate projection with respect to $\CAL{B}_k$,
then $\CAL{F}$ is a Grassmannian 2-design if and only if $\BB{S}$ is a $2$-$(m,m/2,m/2-1)$ block design.
\end{thm}
\begin{proof}
First, let $\CAL{F}$ be a Grassmannian 2-design. By Corollary~\ref{cor_K2lm_val} and the choice of $l=m/2$,
$$
\sum_{k,k' \in K}\sum_{\CAL J,\CAL J' \in \BB{S}} \PARENTH{ \mathop{\mathrm{tr}}(P^{(k)}_{\CAL J} P^{(k')}_{\CAL J'} )}^2 = \frac{m^2 d_{\BB{F}}(m) (d_{\BB{F}}(m) +1)}{4} \, .
$$
From the assumption on the size $k_{\BB{F}}(m) = |K|$ and $d_{\BB{F}}(m)=(m-1) k_{\BB{F}}(m)$, the set $\BB{S}$ has size $|\BB{S}|=2(m-1)$.
Since the orthormal bases are unitarily equivalent, and each pair of them is mutually unbiased, we can obtain the sum
for the squared Hilbert-Schmidt inner products belonging to one basis,
\begin{align*}
\sum_{\CAL J,\CAL J' \in \BB{S}} \PARENTH{\mathop{\mathrm{tr}}(P^{(k)}_{\CAL J} P^{(k)}_{\CAL J'} )}^2 & = \frac{m^2 d_{\BB{F}}(m) (d_{\BB{F}}(m)+1)}{4 k_{\BB{F}}(m)} - (k_{\BB{F}}(m) - 1) 4 (m-1)^2 \frac{m^2}{16}\, \\
& = \frac{(m-1) m^2}{4} \bigl(k_{\BB{F}}(m) (m-1) + 1 - (k_{\BB{F}}(m) - 1)(m-1)\bigr) = \frac{(m-1) m^3}{4} \, .
\end{align*}
The average of the Hilbert-Schmidt inner products of the $|\BB{S}|=2(m-1)$ coordinate projections
belonging to each basis is then
\begin{align*}
\frac{1}{|\BB{S}|^2} \sum_{\CAL J,\CAL J' \in \BB{S}} \PARENTH{ \mathop{\mathrm{tr}}( P^{(k)}_{\CAL J} P^{(k)}_{\CAL J'} ) }^2 & = \frac{1}{|\BB{S}|} \PARENTH{l^2 + (n-2) \frac{l^4}{m^2} }
= \frac{m^3}{16(m-1)} \, .
\end{align*}
Specializing the expression $ \mathop{\mathrm{tr}} \PARENTH{D_{2,l,m}^2 }= l^2(l^2-2l+m)/m(m-1)$ to $l=m/2$ shows that equality holds in Corollary~\ref{thm:GdesignBIBD},
so $\BB{S}$ is a block $2$-design. The parameter of the design then follows from $\lambda = |\BB{S}| \frac{l(l-1)}{m(m-1)} =m/2-1$.
Conversely, if $\BB{S}$ is a $2$-$(m,m/2,m/2-1)$ block design,
then equality holds in the inequality in Corollary~\ref{cor:diagdesigns}.
Since the squared inner product between any two coordinate projections belonging to different bases equals $l^4/m^2$, the lower bound
from Corollary~\ref{cor:diagdesigns} is equivalent to a lower bound for the squared inner products among the coordinate projections
belonging to all mutually unbiased bases, and both bounds are saturated. Using the preceding two identities shows that
this implies that equality holds in the inequality in Theorem~\ref{thm:ZaunerSidelnikov}
and hence $\CAL{F}$ is a Grassmannian 2-design.
\end{proof}
\section{A family of maximal orthoplectic fusion frames}\label{sec:moff}
In this section, we construct a family of $\{0,1\}$-matrices, $\{S_r\}_{r \in \BB{N}}$, and show that they generate maximally orthoplectic block $1$-designs, and therefore generate maximal orthoplectic fusion frames by Theorem~\ref{th_costruct_OGFF}. Consequently, by Theorem~\ref{thm:GdesignBIBD} they are 2-designs. We give an independent
proof of this fact to illustrate the rigidity in the construction of these matrices.
We recall from Corollary~\ref{cor:half_ss} that a necessary condition for the existence of a maximal orthoplectic fusion frame is that the subspace dimension is
$l =\frac{m}{2}$. In order to exploit the existence of maximal sets of mutually unbiased bases in prime power dimensions, it is natural to focus on the case where $m$ is a power of two. We construct the block $1$-designs in terms of the associated {\it incidence matrices}.
\begin{defn}
The {\bf incidence matrix} $S$ associated with a sequence $ \BB{S}=\{\CAL J_1$, $\CAL J_2, \dots, \CAL J_n\}$ of subsets of $[\![m]\!]$
is an $m\times n$ matrix whose $(a,b)$-th entry is
$$
S_{a,b} = \left\{ \begin{array}{cc} 1, & a \in \CAL J_b \\ 0, & \text{otherwise}\end{array} \right.
.$$
\end{defn}
Let $S_1 =I_2$. For $r \in \BB{N}$, let $F_r = I_{(2^{r}-1)} \otimes \left( \begin{array}{cc} 0 & 1 \\ 1 & 0\end{array} \right)$ and let $1_r$ denote the $2^{r} \times 1$ matrix of all ones and let $0_r$ be the $2^{r} \times 1$ matrix of all zeros. For $r \geq 2$,
define
$S_r$ recursively and block-wise by
$$
S_r = \left(B_r^{(i)} \,\, B_r^{(ii)} \, \, B_r^{(iii)}\right),
$$
wher
$$
B_r^{(i)} = \left( \begin{array}{cc} 1_{r-1} & 0_{r-1}\\
0_{r-1} & 1_{r-1}
\end{array} \right),
B_r^{(ii)} = \left( \begin{array}{cc} S_{r-1}\\
S_{r-1}
\end{array} \right),
\text{ and }
B_r^{(iii)} = \left( \begin{array}{cc} S_{r-1}\\
S_{r-1} F_{r-1}
\end{array} \right).
$$
If $c_r$ and $\rho_r$ denote the number of columns and rows of $S_t$, respectively, then we have the recurrence relation
$$
c_1=2, c_{r+1} = 2 c_{r} +2,
$$
which has the solution $c_r = 2^{r+1}-2$. By the construction of $S_r$, $\rho_{r+1} =2 \rho_r = 2^r$, so $S_r$ is a $(2^{r+1}-2) \times 2^r$ matrix.
Furthermore, if $\tilde c^{(j)}_r$ denotes the number of ones in the $j$th column of $S_r$ and $\tilde \rho^{(j)}_r$ denotes the number of ones in the $j$th row of $S_r$, then it is straightforward to verify, by construction, that both of these values are independent of $j$. In particular, $\tilde c_r^{(j)} = 2^{r-1}$ for each $j$ and $\tilde \rho^{(j)}_r= 2^{r}-1$ for each $j$ in the index set of columns or rows, respectively.
We record this as a lemma.
\begin{lemma}\label{lem_halfones}
Each column of $S_r$ has exactly $2^{r-1}$ ones among its entries, and each row of $S_r$ has exactly $2^r -1$ ones among its entries.
\end{lemma}
Next, we examine the inner products among the columns, $\{s_j\}_{j \in [\![{c_r}]\!]}$, of $S_r$, noting that these are encoded in the matrix,
$$
S_r^* S_r = \PARENTH{\langle s_b, s_a \rangle}_{a,b =1}^{c_r}.
$$
We write ${J}_{x,y}$ for the $x\times y$ matrix whose entries all equal $1$.
\begin{lemma}\label{lem_st_gram}
For each $r \in \mathbb{N}$, the matrix $G=S_r^* S_r$ is of the form
$$
G= 2^{r-2} \cdot \left[{J}_{c_r ,c_r} + {I}_{c_r/2} \otimes \left( \begin{array}{cc}1 & -1 \\ -1 & 1 \end{array} \right) \right].
$$
\end{lemma}
\begin{proof}
We prove the claimed form of $G$ by induction. The claim is true for $r=1$, so let $r>1$ assume the claim holds for $r-1$.
Using the block structure in the definition of $S_r$, we have
\begin{align*}
{\tiny
G =
\left(
\begin{array}{ccc}
\PARENTH{B_r^{(i)} }^* B_r^{(i)} & \PARENTH{B_r^{(i)} }^* B_r^{(ii)} & \PARENTH{B_r^{(i)} }^* B_r^{(iii)}
\\
\PARENTH{B_r^{(ii)} }^* B_r^{(i)} & \PARENTH{B_r^{(ii)} }^* B_r^{(ii)} & \PARENTH{B_r^{(ii)} }^* B_r^{(iii)}
\\
\PARENTH{B_r^{(iii)} }^* B_r^{(i)} & \PARENTH{B_r^{(iii)} }^* B_r^{(ii)} & \PARENTH{B_r^{(iii)} }^* B_r^{(iii)}
\end{array}
\right).
}
\end{align*}
A direct application of the definition of $B_r^{(i)}$ and Lemma~\ref{lem_halfones} gives the values of the first row and first column of blocks in $G$,
$${\tiny G=
\left(
\begin{array}{ccc}
2^{ r-1} \cdot I_2 & 2^{r-2} \cdot J_{2 , c_{r-1} }& 2^{r-2} \cdot J_{2 ,c_{r-1}}
\\
2^{r-2} \cdot J_{ c_{r-1} , 2} & \PARENTH{B_r^{(ii)} }^* B_r^{(ii)} & \PARENTH{B_r^{(ii)} }^* B_r^{(iii)}
\\
2^{r-2} \cdot J_{ c_{r-1} , 2} & \PARENTH{B_r^{(iii)} }^* B_r^{(ii)} & \PARENTH{B_r^{(iii)} }^* B_r^{(iii)}
\end{array}
\right)}.$$
A direct application of the induction hypothesis gives us the center block,
$$
\PARENTH{B_r^{(ii)} }^* B_r^{(ii)} = S_{r-1}^* S_{r-1} + S_{r-1}^* S_{r-1} = 2^{r-2} \cdot \left[J_{c_{r-1} , c_{r-1}} + I_{(c_{r-1}/2)} \otimes \left( \begin{array}{cc}1 &-1 \\ -1 &1 \end{array} \right) \right].
$$
Next, observe that by the induction assumption and definition of $F_{r-1}$, we have
\begin{align*}
S_{r-1}^* S_{r-1}
F_{r-1} &=
2^{r-3} \cdot \left[J_{c_{r-1} ,c_{r-1}} + I_{c_r/2} \otimes \left( \begin{array}{cc}1 & -1 \\ -1 & 1 \end{array} \right) \right]
\SBRACK{
I_{c_{r-1}/2} \otimes \left( \begin{array}{cc} 0 & 1 \\ 1 & 0\end{array} \right)
}
\\
&=
2^{r-3} \cdot \left[J_{c_{r-1} , c_{r-1}} + I_{c_r/2} \otimes \left( \begin{array}{cc}-1 & 1 \\ 1& -1 \end{array} \right) \right],
\end{align*}
so it follows that
$$
\PARENTH{B_r^{(ii)} }^* B_r^{(iii)}
=
S_{r-1}^* S_{r-1} + S_{r-1}^* S_{r-1} F_{r-1} \\
=
2^{r-2} \cdot J_{c_{r-1} , c_{r-1}},
$$
and, by symmetry, we also have
$$
\PARENTH{B_r^{(iii)} }^* B_r^{(ii)}
=
2^{r-2} \cdot J_{c_{r-1} ,c_{r-1}}.
$$
Finally, observe that $F_{r-1}^* S_{r-1}^* S_{r-1} F_{r-1} = S_{r-1}^* S_{r-1}$, so
\begin{align*}
\PARENTH{B_r^{(iii)} }^* B_r^{(iii)}
&= S_{r-1}^* S_{r-1} +F_{r-1}^* S_{r-1}^* S_{r-1} F_{r-1}
\\
&= 2^{r-2} \cdot \left[J_{c_{r-1} , c_{r-1}} + I_{c_{r-1}/2} \otimes \left( \begin{array}{cc}1 &-1 \\ -1 &1 \end{array} \right) \right].
\end{align*}
This establishes that the nine blocks match the claimed form of $G$.
\end{proof}
For each $r \in \BB N$, we let $\BB{S}_r$ be the set of blocks in $[\![m]\!]$ defined in accordance with the columns of $S_r$ by
$
\CAL{J}_b = \CBRACK{a: (S_r)_{a,b}=1},
$
where $(S_r)_{a,b}$ denotes the $(a,b)$-th entry of $S_t$.
Although the block 2-design property of $\BB{S}_r$ is implicit in the result on maximal orthoplectic fusion frames stated below, we show it in a separate
proof to illustrate the additional constraints realized by the construction.
\begin{prop}
For each $r \in \BB{N}$, $S_r$
is the incidence matrix of a $2$-$(m,m/2,m/2-1)$ block design, with $m=2^r$.
\end{prop}
\begin{proof}
Since the block $2$-design property of $\BB{S}_r$ is equivalent to the statement that every row of $S_r$ has constant sum and any two distinct row vectors have a constant inner product, it is sufficient to prove the matrix identity
$$
S_ rS_r^* = 2^{r-1} I_{2^r} + (2^{r-1}-1)J_{2^r,2^r} \, .
$$
We prove this by induction jointly with an ancillary claim,
$$
S_r F_r S_r^* = 2^{r-1}J_{2^r,2^r} -2^{r-1} I_{2^r}\, .
$$
To begin,
$S_1$ satisfies $S_1 S_1^* = I_2$ and
$S_1 F_1 S_1 = F_1$, so both identities hold for $r=1$.
Assuming this is also true for $S_{r-1}$ and $F_{r-1}$, we compute
$$
S_r S_r^* = \left( \begin{array}{cc} J_{2^{r-1},2^{r-1}} + 2 S_{r-1} S_{r-1}^* & S_{r-1} S_{r-1}^*+ S_{r-1} F_{r-1} S_{r-1}^*\\
S_{r-1} S_{r-1}^*+ S_{r-1} F_{r-1} S_{r-1}^* & J_{2^{r-1},2^{r-1}} + S_{r-1} S_{r-1}^* + S_{r-1} F_{r-1}^2 S_{r-1}^*
\end{array} \right).
$$
Now using the induction assumption and $F_{r-1}^2 = I_{2^{r-1}}$, we get
$$
S_r S_r^* = 2(2^{r-2}) I_{2^r} + 2(2^{r-2}-1)J_{2^r,2^r} + J_{2^r,2^r},
$$
which simplifies to the claimed identity.
Moreover, using the fact that $F_r$ has the block form,
$$F_r = \PARENTH{\begin{array}{cc} F_1 & 0_{2 , 2 c_{r-1} } \\ 0_{ 2 c_{r-1} , 2 }& I_2 \otimes F_{r-1} \end{array}},$$ where $0_{a,b}$ is the $a \times b$ zero matrix,
a straightforward computation yields
$$
S_r F_r S_r^* = \left( \begin{array}{cc} 2 S_{r-1} F_{r-1} S_{r-1}^* & J_{2^{r-1},2^{r-1}}+ S_{r-1} F_{r-1} S_{r-1}^*+ S_{r-1} S_{r-1}^*\\
J_{2^{r-1},2^{r-1}}+ S_{r-1} F_{r-1} S_{r-1}^*+ S_{r-1} S_{r-1}^* & 2 S_{r-1} F_{r-1} S_{r-1}^* \end{array} \right)
$$
and using the induction assumption gives
$$
S_r F_r S_r^* = 2^{r-1} \left( \begin{array}{cc} J_{2^{r-1},2^{r-1}} - I_{2^{r-1}} & J_{2^{r-1},2^{r-1}}\\
J_{2^{r-1},2^{r-1}} & J_{2^{r-1},2^{r-1}} - I_{2^{r-1}} \end{array} \right) = 2^{r-1} (J_{2^r,2^r} - I_{2^r}) \, .
$$
\end{proof}
Finally, we state the main theorem of this section, which summarizes the construction of maximal orthoplectic tight fusion frames.
\begin{thm}\label{th_main}
Let $r \in \BB{N}$, where $r$ is even if $\BB{F}=\BB{R}$. If $m=2^r$ and $\{\CAL B_k\}_{k \in K}$ is a maximal collection of mutually unbiased bases
for for $\BB{F}^m$, so $|K|=k_{\BB{F}}(m)$, then
$$
\CAL{F} =
\CBRACK{
P_{\CAL{J}}^{(k)} : k \in K, \CAL{J} \in \BB{S}_{r}
}
$$
forms a tight, maximal orthoplectic fusion frame,
where each $P_\CAL{J}^{(k)}$ is the $\CAL{J}$-coordinate projection with respect to $\CAL{B}_k$.
\end{thm}
\begin{proof}
It follows directly from Lemma~\ref{lem_halfones} that $\BB{S}_r$ is a $1$-$(m,l, m-1)$ block design, where $m=2^r$, $l=2^{r-1}$ and $| \BB{S}_r| = c_r = 2^{r+1}-2$.
If $\{s_j\}_{j\in [\![{c_r}]\!]}$ denotes the columns of $S_r$, then the Gram matrix $S_r^* S_r$ encodes the intersections of the blocks by $\langle s_a,s_b\rangle~= ~\left| \CAL{J}_a \cap \CAL{J}_b\right|$, so Lemma~\ref{lem_st_gram} implies that $\max\limits_{a \ne b} \left| \CAL{J}_a \cap \CAL{J}_{b} \right| = 2^{r-2}$. This means that $\BB{S}_r$ is $c$-cohesive, where $c=\frac{l^2}{m} = \frac{m}{4}$, and since $| \BB{S}_r| = 2^{r+1}-2 = 2(m-1)$, we conclude that $\BB{S}_r$ is a maximally orthoplectic block $1$-design.
Finally, using a maximal set of mutually unbiased bases $\{\CAL B_k\}_{k \in K}$ and the
maximally orthoplectic block $1$-$(m,m/2,m-1)$ design $\BB{S}_r$, Theorem~\ref{th_costruct_OGFF} shows that the set
$
\CAL{F} =
\{
P_{\CAL{J}}^{(k)} : k \in K$, $\CAL{J} \in \BB{S}_{r}
\}
$
forms a maximal orthoplectic fusion frame,
where each $P_\CAL{J}^{(k)}$ is the $\CAL{J}$-coordinate projection with respect to $\CAL{B}_k$.
\end{proof}
| 64527c47908aa856c65efb83ac20c0c8f9ab5edd | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Error Probability Bound}\label{appendix:APPENDIX A}
In this section, we give a direct proof for \Cref{direct lemma1}. Specifically, we bound the error probability of the maximum likelihood decoder, and show that under the condition on $T$ in the lemma, the error probability is indeed approaching $0$ as $N \to \infty$.
Under the assumption that all items are equality likely to be defective, we consider the error probability given that the first $K$ items are defective, that is, \revisedSr{$S_{w=1}$} is the defective set. Denote this probability by $P_{e|1}$. We have
\begin{equation*}
P_{e|1} \leq \sum_{i=1}^{K}P(E_i),
\end{equation*}
where $E_i$ is the event of a decoding error in which the decoder declares a defective set which differs from the true one ($S_1$) in exactly $i$ items.
\revisedSr{Note that the discussion below will apply to any defective set $S_w$ of size $K$, hence the probability of error computed decays exponentially for any ``message $W$".}
In general, we follow the derivation in \cite{atia2012boolean}. However, there is a key difference. In the code construction suggested in \Cref{LowerBound}, for each item there are $M$ possible codewords (a ``bin" of size $M$). Only one of these codewords is selected by the mixer to decide in which pool tests the item will participate. Thus, when viewing this problem as a channel coding problem, if an item is defective, one and only one codeword out of its bin is actually transmitted (and summed with the codewords of the other $K-1$ defective items). Since the decoder does not know which codewords were selected in each bin (the randomness is known only to the mixer), there are multiple error events to consider. E.g., events where the decoder choose the wrong codeword for some items, yet identified parts of the bins correctly, and, of course, events where the codeword selected was from a wrong bin. This complicates the error analysis significantly. Moreover, we wish to employ the correction suggested in \cite{atia2015correction}, which results in a simpler yet stricter bound.
Consider the event $E_i$. Clearly, $E_i$ can be broken down into two disjoint events. The first is $E_i$ \emph{and} the event that the codewords selected for the correct $K-i$ items are the true transmitted ones, and the second is the event of both $E_i$ and the event that \emph{at least one of the codewords selected} for the correct items is wrong. Denote the first event as $E'_i$. Now, consider the case where we have $E_i$, that is, a correct decision on $K-i$ items, yet, out of these $K-i$ items, the decoder identified only $j$ codewords right, $0 \leq j < K-i$, and for the rest, it identified a wrong codeword in the right bin. We claim that the probability for this event is exactly $P(E'_{K-j})$, as the decoder has to mistake $\bf{X'}$ for $\bf{X}_{S_1}$, where both are collections of $K$ codewords, and $\bf{X'}$ shares exactly $j$ codewords with $\bf{X}_{S_1}$. Thus, we have:
\begin{eqnarray}\label{E and E'}
P_{e|1} &\leq& \sum_{i=1}^{K}P(E_i) \nonumber\\
&=& \sum_{i=1}^{K}\Big( P(E'_i) + \sum_{j=0}^{K-i-1} {K-i-1 \choose j}P(E'_{K-j}) \Big) \nonumber\\
&\leq&\revisedr{2^K} \sum_{i=1}^{K}P(E'_i).
\end{eqnarray}
The last inequality in \eqref{E and E'} is loose, yet it suffices for the proof in the regime where $K=O(1)$. A more delicate bounding technique might be required if $K$ might grow with $N$.
We now bound $P(E'_i)$. Particularly, we will establish the following lemma.
\begin{lemma}\label{error lemma2 with E'}
The error probability $P(E'_{i})$ is bounded by
\begin{equation*}
P(E'_{i}) \leq 2^{-T\left(E_{o}(\rho)-\rho\frac{\log\binom{N-K}{i}M^i}{T}-\frac{\log\binom{K}{i}}{T}\right)},
\end{equation*}
where the error exponent $E_{o}(\rho)$ is given by
\ifdouble
\begin{multline}
E_{o}(\rho)= -\log \sum_{Y\in \{0,1\}}\sum_{X_{\mathcal{S}^2}\in \{0,1\}}\Bigg[\sum_{X_{\mathcal{S}^1}\in \{0,1\}}P(X_{\mathcal{S}^1})\\
p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})^{\frac{1}{1+\rho}}\Bigg]^{1+\rho},\text{ } 0 \leq\rho \leq1.\nonumber
\end{multline}
\else
\begin{equation*}
E_{o}(\rho)= -\log \sum_{Y\in \{0,1\}}\sum_{X_{\mathcal{S}^2}\in \{0,1\}}\Bigg[\sum_{X_{\mathcal{S}^1}\in \{0,1\}}P(X_{\mathcal{S}^1}) p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})^{\frac{1}{1+\rho}}\Bigg]^{1+\rho},\text{ } 0 \leq\rho \leq1.
\end{equation*}
\fi
\end{lemma}
\begin{proof}
Denote by
\begin{equation*}\label{weaker_error1}
\mathcal{A} = \{w \in \revisedSr{\mathcal{W}} : |S_{1^{c},w}|=i,|S_{w}|=K\}
\end{equation*}
the set of indices corresponding to sets of $K$ items that differ from the true defective set $S_{1}$ in exactly $i$ items. $S_{1^{c},w}$ for $w \in \revisedSr{\mathcal{W}}$ denotes the set of items which are in $S_w$ but not in $S_1$.
We have
\ifdouble
\begin{eqnarray}\label{weaker_error2}
&& \Pr[E'_i|w_0=1, \textbf{X}_{\mathcal{S}_1},Y^T]\nonumber\\
& \leq & \sum_{w \in \mathcal{A}} \sum_{\textbf{X}_{\mathcal{S}_{1^{c},w}}}
Q({\scriptstyle\textbf{X}_{\mathcal{S}_{1^{c},w}}}) \frac{{\scriptstyle p_{w}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1^{c},w}})^s}} {{\scriptstyle p_{1}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1,w^{c}}})^s}}\\
& \leq & \sum_{\mathcal{S}_{1,w}} \sum_{\mathcal{S}_{1^{c},w}}\sum_{\textbf{X}_{\mathcal{S}_{1^{c},w}}}
Q({\scriptstyle \textbf{X}_{\mathcal{S}_{1^{c},w}}})\frac{{\scriptstyle p_{w}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1^{c},w}})^s}} {{\scriptstyle p_{1}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1,w^{c}}})^s}}\nonumber,
\end{eqnarray}
\else
\begin{eqnarray}\label{weaker_error2}
\Pr[E'_i|w_0=1, \textbf{X}_{\mathcal{S}_1},Y^T] & \leq & \sum_{w \in \mathcal{A}} \sum_{\textbf{X}_{\mathcal{S}_{1^{c},w}}}
Q({\scriptstyle\textbf{X}_{\mathcal{S}_{1^{c},w}}}) \frac{{\scriptstyle p_{w}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1^{c},w}})^s}} {{\scriptstyle p_{1}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1,w^{c}}})^s}}\\
& \leq & \sum_{\mathcal{S}_{1,w}} \sum_{\mathcal{S}_{1^{c},w}}\sum_{\textbf{X}_{\mathcal{S}_{1^{c},w}}}
Q({\scriptstyle \textbf{X}_{\mathcal{S}_{1^{c},w}}})\frac{{\scriptstyle p_{w}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1^{c},w}})^s}} {{\scriptstyle p_{1}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1,w^{c}}})^s}}\nonumber,
\end{eqnarray}
\fi
where \eqref{weaker_error2} is exactly \cite[eq. (25)]{atia2012boolean}, as when considering $E'_i$ we assume the decoder not only got $K-i$ items right, but also the correct codeword in each such bin. Hence
\ifdouble
\begin{eqnarray}
&&\hspace{-0.5cm} \Pr[E_i|w_0=1, \textbf{X}_{\mathcal{S}_1},Y^T]\nonumber\\
&\hspace{-0.5cm} \stackrel{(a)}{\leq} &\hspace{-0.4cm} \Big(\sum_{\mathcal{S}_{1,w}} \sum_{\mathcal{S}_{1^{c},w}}\sum_{\textbf{X}_{\mathcal{S}_{1^{c},w}}}
Q({\scriptstyle\textbf{X}_{\mathcal{S}_{1^{c},w}}}) \frac{{\scriptstyle p_{w}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1^{c},w}})^s}} {{\scriptstyle p_{1}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1,w^{c}}})^s}}\Big)^{\rho}\nonumber\\
&\hspace{-0.5cm} \stackrel{(b)}{\leq} &\hspace{-0.4cm} \Big(\sum_{\mathcal{S}_{1,w}} {\scriptstyle\binom{N-K}{i}M^{i}}\sum_{\textbf{X}_{\mathcal{S}_{1^{c},w}}}
Q({\scriptstyle\textbf{X}_{\mathcal{S}_{1^{c},w}}})\frac{{\scriptstyle p_{w}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1^{c},w}})^s}} {{\scriptstyle p_{1}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1,w^{c}}})^s}}\Big)^{\rho}\nonumber\\
&\hspace{-0.5cm} \stackrel{(c)}{\leq} &\hspace{-0.4cm} {\scriptstyle\binom{N-K}{i}^\rho M^{i\rho}}\sum_{\mathcal{S}_{1,w}}\Big(\sum_{\textbf{X}_{\mathcal{S}_{1^{c},w}}}
Q({\scriptstyle\textbf{X}_{\mathcal{S}_{1^{c},w}}})\frac{{\scriptstyle p_{w}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1^{c},w}})^s}} {{\scriptstyle p_{1}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1,w^{c}}})^s}}\Big)^{\rho}\nonumber
\end{eqnarray}
\else
\begin{eqnarray}
\Pr[E_i|w_0=1, \textbf{X}_{\mathcal{S}_1},Y^T]&\stackrel{(a)}{\leq} & \Big(\sum_{\mathcal{S}_{1,w}} \sum_{\mathcal{S}_{1^{c},w}}\sum_{\textbf{X}_{\mathcal{S}_{1^{c},w}}}
Q({\scriptstyle\textbf{X}_{\mathcal{S}_{1^{c},w}}}) \frac{{\scriptstyle p_{w}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1^{c},w}})^s}} {{\scriptstyle p_{1}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1,w^{c}}})^s}}\Big)^{\rho}\nonumber\\
& \stackrel{(b)}{\leq} & \Big(\sum_{\mathcal{S}_{1,w}} {\scriptstyle\binom{N-K}{i}M^{i}}\sum_{\textbf{X}_{\mathcal{S}_{1^{c},w}}}
Q({\scriptstyle\textbf{X}_{\mathcal{S}_{1^{c},w}}})\frac{{\scriptstyle p_{w}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1^{c},w}})^s}} {{\scriptstyle p_{1}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1,w^{c}}})^s}}\Big)^{\rho}\nonumber\\
&\stackrel{(c)}{\leq} & {\scriptstyle\binom{N-K}{i}^\rho M^{i\rho}}\sum_{\mathcal{S}_{1,w}}\Big(\sum_{\textbf{X}_{\mathcal{S}_{1^{c},w}}}
Q({\scriptstyle\textbf{X}_{\mathcal{S}_{1^{c},w}}})\frac{{\scriptstyle p_{w}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1^{c},w}})^s}} {{\scriptstyle p_{1}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1,w^{c}}})^s}}\Big)^{\rho}\nonumber
\end{eqnarray}
\fi
for all $s>0$ and $0\leq\rho\leq1$. Note that (a) is since the probability is less than $1$ and can be raised to the power of $\rho$. (b) is critical. It follows from the symmetry of codebook construction, namely, the inner summation depends only on the codewords in $\textbf{X}_{\mathcal{S}_{1^{c},w}}$ but not the ones in $\mathcal{S}_{1^{c},w}$.
Due to the code's construction, i.e., its binning structure, there are exactly $\binom{N-K}{i}M^{i}$ possible sets of codewords to consider for $\mathcal{S}_{1^{c},w}$. (c) follows as the sum of positive numbers raised to the $\rho$-th power is smaller than the sum of the $\rho$-th powers.
We now continue similar to \cite{atia2012boolean}, substituting the conditional error probability just derived in a summation over all codewords and output vectors. We have
\ifdouble
\begin{eqnarray}
P(E'_i) & = & \sum_{\textbf{X}_{\mathcal{S}_{1}}}\sum_{Y^T}p_1(\textbf{X}_{\mathcal{S}^1},Y^T)\Pr[E'_i|w_0=1, \textbf{X}_{\mathcal{S}_1},Y^T]\nonumber\\
& \leq & {\scriptstyle\binom{N-K}{i}^\rho M^{i\rho}}\sum_{\mathcal{S}_{1,w}}\sum_{Y^T}\sum_{\textbf{X}_{\mathcal{S}_{1}}}p_1({\scriptstyle \textbf{X}_{\mathcal{S}^1},Y^T})\nonumber\\
&& \left(\sum_{\textbf{X}_{\mathcal{S}_{1^{c},w}}}Q({\scriptstyle\textbf{X}_{\mathcal{S}_{1^{c},w}}})
\frac{{\scriptstyle p_{w}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1^{c},w}}})^s} {{\scriptstyle p_{1}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1,w^{c}}}})^s}\right)^{\rho}\nonumber.
\end{eqnarray}
\else
\begin{eqnarray}
P(E'_i) & = & \sum_{\textbf{X}_{\mathcal{S}_{1}}}\sum_{Y^T}p_1(\textbf{X}_{\mathcal{S}^1},Y^T)\Pr[E'_i|w_0=1, \textbf{X}_{\mathcal{S}_1},Y^T]\nonumber\\
& \leq & {\scriptstyle\binom{N-K}{i}^\rho M^{i\rho}}\sum_{\mathcal{S}_{1,w}}\sum_{Y^T}\sum_{\textbf{X}_{\mathcal{S}_{1}}}p_1({\scriptstyle \textbf{X}_{\mathcal{S}^1},Y^T}) \left(\sum_{\textbf{X}_{\mathcal{S}_{1^{c},w}}}Q({\scriptstyle\textbf{X}_{\mathcal{S}_{1^{c},w}}})
\frac{{\scriptstyle p_{w}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1^{c},w}}})^s} {{\scriptstyle p_{1}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1,w^{c}}}})^s}\right)^{\rho}\nonumber.
\end{eqnarray}
\fi
There are $\binom{K}{K-i}$ sets $\mathcal{S}_{1,w}$, and the summation does not depend on which set is it, hence, we get
\ifdouble
\begin{multline*}
P(E'_i) \leq {\scriptstyle\binom{N-K}{i}^\rho M^{i\rho}\binom{K}{i}}\sum_{Y^T}\sum_{\textbf{X}_{\mathcal{S}_{1}}}p_1({\scriptstyle\textbf{X}_{\mathcal{S}^1},Y^T})
\\ \left(\sum_{\textbf{X}_{\mathcal{S}_{1^{c},w}}}Q({\scriptstyle{\scriptstyle\textbf{X}_{\mathcal{S}_{1^{c},w}}}}) \frac{{\scriptstyle p_{w}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1^{c},w}})^s}} {{\scriptstyle p_{1}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1,w^{c}}})^s}}\right)^{\rho}
\end{multline*}
\else
\begin{equation*}
P(E'_i) \leq {\scriptstyle\binom{N-K}{i}^\rho M^{i\rho}\binom{K}{i}}\sum_{Y^T}\sum_{\textbf{X}_{\mathcal{S}_{1}}}p_1({\scriptstyle\textbf{X}_{\mathcal{S}^1},Y^T})
\left(\sum_{\textbf{X}_{\mathcal{S}_{1^{c},w}}}Q({\scriptstyle{\scriptstyle\textbf{X}_{\mathcal{S}_{1^{c},w}}}}) \frac{{\scriptstyle p_{w}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1^{c},w}})^s}} {{\scriptstyle p_{1}(Y^T,\textbf{X}_{\mathcal{S}_{1,w}}|\textbf{X}_{\mathcal{S}_{1,w^{c}}})^s}}\right)^{\rho}
\end{equation*}
\fi
which continue similar to \cite{atia2012boolean}, results in
\ifdouble
\begin{multline*}
P(E'_i) \leq
{\scriptstyle\binom{N-K}{i}^\rho M^{i\rho}\binom{K}{i}}\Bigg[\sum_{Y}\sum_{X_{\mathcal{S}_{1,w}}}
\\
\Big(\sum_{X_{\mathcal{S}_{1,w^{c}}}}Q({\scriptstyle X_{\mathcal{S}_{1^{c},w}}}) p_1^{1/(1+\rho)}({\scriptstyle X_{\mathcal{S}_{1,w}},Y|X_{\mathcal{S}_{1,w^{c}}}})\Big)^{1+\rho}\Bigg]^T.
\end{multline*}
\else
\begin{equation*}
P(E'_i) \leq
{\scriptstyle\binom{N-K}{i}^\rho M^{i\rho}\binom{K}{i}}\Bigg[\sum_{Y}\sum_{X_{\mathcal{S}_{1,w}}}
\Big(\sum_{X_{\mathcal{S}_{1,w^{c}}}}Q({\scriptstyle X_{\mathcal{S}_{1^{c},w}}}) p_1^{1/(1+\rho)}({\scriptstyle X_{\mathcal{S}_{1,w}},Y|X_{\mathcal{S}_{1,w^{c}}}})\Big)^{1+\rho}\Bigg]^T.
\end{equation*}
\fi
Thus, we have
\begin{equation*}
P(E'_i) \leq 2^{-T\left(E_0(\rho) -\rho\left(\frac{\log{N-K \choose i}}{T}+\frac{i\log M}{T} \right) -\frac{\log{K \choose i}}{T}\right)}.
\end{equation*}
\end{proof}
\section{Problem Formulation}\label{formulation}
In \emph{SGT}, a legitimate user desires to identify a small unknown subset $\mathcal{K}$ of defective items from a larger set $\mathcal{N}$, while \revised{minimizing} the number of measurements $T$ and keeping the eavesdropper, which is able to observe a subset of the tests results, ignorant regarding the status of the $\mathcal{N}$ items.
\revised{Let} $N=|\mathcal{N}|$, $K=|\mathcal{K}|$ denote the total number of items, and the number of defective items, respectively.
\revised{As formally defined below, the legitimate user should (with high probability) be able to correctly estimate the set $\mathcal{K}$; on the other hand, from the eavesdropper’s perspective, this set should be ``almost" uniformly distributed over all possible ${N}\choose{K}$ sets. We assume that the number $K$ of defective items in $\mathcal{K}$ is known {\it a priori} to all parties - this is a common assumption in the GT literature \cite{macula1999probabilistic}.}\footnote{\revised{If this is not the case, \cite{damaschke2010competitive,damaschke2010bounds} give methods/bounds on how to ``probably approximately" correctly learn the value of $K$ in a single stage with $O(\log N)$ tests.}}
\off{The status of the items, defective or not, should be kept secure from the eavesdropper, but detectable to the legitimate user. We assume that the number $K$ of defective items in $\mathcal{K}$ is known a priori. This is a common assumption in the GT literature \cite{macula1999probabilistic}.}
Throughout the paper, we use boldface to denote matrices, capital letters to denote random variables, lower case letters to denote their realizations, and calligraphic letters to denote the alphabet. Logarithms are in base $2$ and $h_b(\cdot)$ denotes the binary entropy function.
\Cref{figure:secure-group-testing} gives a graphical representation of the model.
In general, and regardless of security constraints, \revised{non-adaptive} GT is defined by a testing matrix
\begin{equation*}
\textbf{X}=[X_{1}^{T};X_{2}^{T};\ldots; X_{N}^{T}] \in \{0,1\}^{N\times T},
\end{equation*}
where each row corresponds to a separate item $j\in\{1,\ldots,N\}$, and each column corresponds to a separate pool test $t\in\{1,\ldots,T\}$.
For the $j$-th item,
\begin{equation*}
X_{j}^{T}=\{X_{j}(1),\ldots, X_{j}(T)\}
\end{equation*}
is a binary row vector, with the $t$-th entry $X_{j}(t)=1$ if and only if item $j$ participates in the $t$-th test.
\ifdouble
\begin{figure*}
\centering
\includegraphics[trim=0cm 0cm 0cm 0cm,clip,scale=0.8]{secure-group-testing4.png}
\caption{Noiseless non-adaptive secure group-testing setup.}
\label{figure:secure-group-testing}
\end{figure*}
\else
\begin{figure*}
\centering
\includegraphics[trim=0cm 0cm 0cm 0cm,clip,scale=0.78]{secure-group-testing4.png}
\caption{Noiseless non-adaptive secure group-testing setup.}
\label{figure:secure-group-testing}
\end{figure*}
\fi
If $A_j\in \{0,1\}$ \revised{denotes} an indicator function for the $j$-th item, determining whether it belongs to the defective set, i.e., $A_j=1$ if $j \in \mathcal{K}$ and $A_j = 0$ otherwise, \revised{the (binary) outcome of the $t \in \{1,\ldots,T\}$ pool test $Y(t)$ equals}
\begin{equation*}
\revised{Y(t)=\bigvee_{j=1}^{N}X_{j}(t)A_j=\bigvee_{d\in \mathcal{K}}X_{d}(t)},
\end{equation*}
where $\bigvee$ is used to denote the boolean OR operation. \off{\revised{The second equality follows since only defective items influence whether or not test outcome is positive.}}
\off{\revised{In our SGT model, as in the Wyer wiretap-$II$ model, the eavesdropper is allowed to observe {\it any} $\delta T$ test outcomes of its choice.
The adversary’s observations are denoted by $Z^{T}=\{Z(1),\ldots, Z(T)\}$, of which $(1-\delta)T$ are {\it erasures}. Specifically, $Z(t) = Y(t)$ if the $t$-th test is not erased, and $Z(t) = ?$ if it is erased.}}
In SGT, we assume an eavesdropper \revisedSr{who} observes a noisy vector $Z^{T}=\{Z(1),\ldots, Z(T)\}$, generated from the outcome vector $Y^{T}$. In the erasure case considered in the work, the probability of erasure is $1-\delta$, i.i.d.\ for each test.
That is, on average, $T\delta$ outcomes are not erased and are accessible to the eavesdropper via $Z^{T}$. Therefore, in the erasure case, if $B_t\in \{1,?\}$ is an erasure indicator function for the $t$-th pool test, i.e., $B_t=1$ with probability $\delta$, and $B_t=?$ with probability $1-\delta$, the eavesdropper observes
\begin{equation*}
Z(t)=Y(t)B_t= \revised{\left(\bigvee_{j=1}^{N}X_{j}(t)A_j\right)B_t},\text{ }\text{ } t=1,\ldots,T.
\end{equation*}
Denote by \revisedSr{$W \in \mathcal{W} \triangleq \{1,\ldots, {N \choose K}\}$} the index of the \emph{subset of defective items}. We assume $W$ is uniformly distributed, that is, there is no \emph{a priori} bias \revised{towards} any specific subset.\footnote{\revised{This is a common probabilistic model for the set of defectives in group-testing. Another model, called {\it Probabilistic Group Testing}, assumes that items are defective with probability $K/N$. Yet another model assumes that any set of size at most K (rather than exactly $K$) instantiates with equal probability. In many group-testing scenarios results for one probabilistic model for the set of defectives can be translated over to other scenarios, so we focus on the model presented above, where {\it exactly} $K$ items are defective.}}
Further, denote by $\hat{W}(Y^T)$ the index recovered by the legitimate decoder, after observing $Y^T$. In this work, we assume that the mixer may use a \emph{randomized} testing matrix. In this case, the random bits used are know only to the mixer, and are not assumed to be shared with the decoder. In other words, the ``codebook" which consists of all possible testing matrices is known to all parties, Alice, Bob and Eve. \revisedSr{However, if the mixer choose a specific} $\textbf{X}$, the random value is not shared with Bob or Eve. We refer to the codebook consisting of all possible matrices, together with the decoder at Bob's side as SGT algorithm. \off{We refer to the testing matrix, together with the decoder as a \emph{SGT algorithm}.
\revised{Note that all parties (Alice, Bob, and Eve) know the testing matrix {\it a priori}.}}
As we are interested in the asymptotic behavior, \revised{i.e., in ``capacity style" results, with a focus on the number of tests $T$ (as a function of $N$ and $K$) required to guarantee high probability of recovery as the number of items $N$ grows without bound. For simplicity, in the first part of this work, we focus primarily on the regime where $K$ is a constant independent of $N$.} In \Cref{efficient_algorithms}, we give an algorithm which applies to any $K$.\footnote{\revised{Following the lead of~\cite{aldridge2017capacity}, in principle, many of our results in this section as well can be extended to the regime where $K = o(N^{1/3})$, but for ease of presentation we do not do so here).}} The following definition lays out the goals of SGT algorithms.
\begin{definition}
A sequence of SGT algorithms with parameters $N,K$ and $T$ is asymptotically \revised{(in N)} \emph{reliable} and \emph{weakly} or \emph{strongly} secure if,
\hspace{-0.3cm}(1) \emph{Reliable}: \revised{The probability (over the index $W$) of incorrect reconstruction of $W$ at the legitimate receiver converges to zero. That is,}
\begin{equation*}
\lim_{N \to \infty} P(\hat{W}(Y^T) \ne W) = 0.
\end{equation*}
(2) \emph{Weakly secure}: \revised{One potential security goal is so-called {\it weak information-theoretic security} against eavesdropping. Specifically, if the eavesdropper observes $Z^T$, a scheme is said to be weakly secure if}
\begin{equation*}
\lim_{T \to \infty} \frac{1}{T}I(W;Z^T) = 0.
\end{equation*}
(3) \emph{Strongly secure}: \revised{A stronger notion of security is so-called {\it strong information-theoretic security} against eavesdropping. Specifically, if the eavesdropper observes $Z^T$, a scheme is said to be strongly secure if}
\begin{equation*}
\lim_{T \to \infty}I(W;Z^T) = 0.
\end{equation*}
\end{definition}
\revised{
\begin{remark}
Note that strong security implies that in the limit the distribution over $Z^T$ is essentially statistically independent of the distribution over $W$. Specifically, the {\it KL divergence} between $p_{Z^T,W}$ and $p_{Z^T}p_{W}$ converges to $0$.
\end{remark}
\begin{remark}
While weak security is a much weaker notation of security against eavesdropping than strong security, and indeed is implied by strong security, nonetheless we consider it in this work for the following reason. Our impossibility result will show that even guaranteeing weak security requires at least a certain number of tests, and our achievability results will show that essentially the same number of tests suffices to guarantee strong security. Hence both our impossibility and achievability results are with regard to the corresponding ``harder to prove" notion of security.
\end{remark}
To conclude, the goal in this work is to design (for parameters $N$ and $K$) an $N \times T$ measurement matrix (which is possibly randomized) and a decoding algorithm $Wˆ(Y^T)$, such that on observing $Y^T$, the legitimate decoder can (with high probability over $W$) identify the subset of defective items, and yet, on observing $Z^T$, the eavesdropper learns essentially nothing about the set of defective items.
}
\off{
To conclude, The goal is to design (for parameters $N$ and $K$) an $N\times T$ measurement matrix and a decoding algorithm $\hat{W}(Y^T)$, such that observing $Y^{T}$, the legitimate user will identify the subset of defective items (with high probability), yet, observing $Z^{T}$, the eavesdropper cannot (asymptotically) identify the status of the items, defective or not.
}
\section{Conclusions}\label{conc}
In this paper, we proposed a novel non-adaptive SGT algorithm, which with parameters $N,K$ and $T$ is asymptotically \emph{reliable} and \emph{secure}. Specifically, when the fraction of tests observed by Eve is $0 \leq \delta <1$, we prove that the number of tests required for both correct reconstruction at the legitimate user (with high probability) and negligible mutual information at Eve's side is $\frac{1}{1-\delta}$ times the number of tests required with no secrecy constraint.
We further provide sufficiency and necessity bounds on the number of tests required in the SGT model to obtains both, \emph{reliability} and \emph{secrecy} constraints.
Moreover, we analyze in the proposed secure model, computationally efficient algorithms at the legitimate decoder, previously considered for the non-secure GT in the literature which identify the definitely non-defective items.
\section{Efficient Algorithms}\label{efficient_algorithms}
The achievability result given in \Cref{direct theorem1} uses a random codebook and ML decoding at the legitimate party. The complexity burden in ML, however, prohibits the use of this result for large $N$. In this section, we derive and analyze an efficient decoding algorithm, which maintains the reliability result using a much simpler decoding rule, at the price of only slightly more tests. The secrecy constraint, as will be clear, is maintained by construction, as the codebook and mixing process do not change compared to the achievability result given before. Moreover, the result in this section will hold for any $K$, including even the case were $K$ grows linearly with $N$.
Specifically, we assume the same codebook generation and the testing procedure given in \Cref{LowerBound}, and analyze the \emph{Definite Non-Defective} (DND) algorithm, previously considered for the non-secure GT in the literature \cite{chan2014non,aldridge2014group}.
The decoding algorithm at the legitimate user is as follows. Bob attempts to match the rows of \textbf{X} with the outcome vector $Y^T$. If a particular row $j$ of \textbf{X} has the property that all locations $t$ where it has $1$, also corresponds to a $1$ in $Y(t)$, then that row \emph{can correspond to a defective item}. If, however, the row has $1$ at a location $t$ where the output has $0$, then it is not possible that the row corresponds to a defective item. The problem, however, when considering the code construction in this paper for SGT, is that the decoder does not know which row from each bin was selected for any given item. Thus, it takes a conservative approach, and declares an item as defective if at least one of the rows in its bin signals it may be so. An item is not defective only if \emph{all the rows in its bin prevent it from being so}.
It is clear that this decoding procedure has no false negatives, as a defective item will always be detected. It may have, though, false positives. A false positive may occur if all locations with ones in a row corresponding to a non-defective item are hidden by the ones of other rows corresponding to defective items and selected by the mixer. To calculate the error probability, fix a row of $\textbf{X}$ corresponding to a non-defective item (a row in its bin). Let $j_1;\ldots;j_k$ index the rows of $\textbf{X}$ corresponding to the $K$ defective items, and selected by the mixer for these items (that is, the rows which were actually added by the Boolean channel).
An error event associated with the fixed row occurs if at any test where that row has a $1$, at least one of the entries $X_{j_1}(t), \ldots, X_{j_k}(t)$ also has a $1$. The probability for this to happen, per column, is $p(1-(1-p)^K)$. Hence, the probability that a test result in a fixed row is hidden from the decoder, in the sense that it cannot be declared as non defective due to a specific column, is
\[
p(1-(1-p)^K)+(1-p) = 1-p(1-p)^K.
\]
Since this should happen for all $T$ columns, the error probability for a fixed row is $\left(1-p(1-p)^K\right)^T$. Now, to compute the error probability for the entire procedure we take a union bound over all $M(N-K)$ rows corresponding to non-defective items. As a result, we have
\ifdouble
\begin{eqnarray}\label{eq:PeComp}
P_e &\leq& M(N-K)\left(1-p(1-p)^{K}\right)^T
\nonumber\\
& \stackrel{(a)}{\leq}& MN\left(1-\frac{y}{K}\left(1-\frac{y}{K}\right)^{K}\right)^{\beta K\log N}
\nonumber\\
&=& MN\left(1-\frac{y}{K}\left(1-\frac{y}{K}\right)^{K-1}\left(1-\frac{y}{K}\right)\right)^{\beta K\log N}
\nonumber\\
& \stackrel{(b)}{\leq}& MN\left(\left(1-\frac{y\left(1-\frac{y}{K}\right)}{Ke^{y}}\right)^{K}\right)^{\beta\log N}
\nonumber\\
& \stackrel{(c)}{\leq}& MNe^{-y\left(1-\frac{y}{K}\right)e^{-y}\beta\log N}\nonumber\\
&\leq& MN^{1-y\left(1-\frac{y}{K}\right)e^{-y}\beta\frac{1}{\ln 2}}
\nonumber\\
& \stackrel{(d)}{=}& MN^{1-\frac{1}{2}\beta (1-\frac{\ln2}{K})}
\nonumber\\
&\stackrel{(e)}{=}& 2^{\beta (\delta-\epsilon) \log N}N^{1-\frac{1}{2}\beta (1-\frac{\ln2}{K})}
\nonumber\\
&=& N^{\beta (\delta-\epsilon)}N^{1-\frac{1}{2}\beta (1-\frac{\ln2}{K})}
\nonumber\\
&\leq& N^{1-\beta\left(\frac{1}{2}(1-\frac{\ln 2}{K})-\delta \right)}.
\end{eqnarray}
\else
\begin{eqnarray}\label{eq:PeComp}
P_e &\leq& M(N-K)\left(1-p(1-p)^{K}\right)^T
\nonumber\\
& \stackrel{(a)}{\leq}& MN\left(1-\frac{y}{K}\left(1-\frac{y}{K}\right)^{K}\right)^{\beta K\log N}
\nonumber\\
&=& MN\left(1-\frac{y}{K}\left(1-\frac{y}{K}\right)^{K-1}\left(1-\frac{y}{K}\right)\right)^{\beta K\log N}
\nonumber\\
& \stackrel{(b)}{\leq}& MN\left(\left(1-\frac{y\left(1-\frac{y}{K}\right)}{Ke^{y}}\right)^{K}\right)^{\beta\log N}
\nonumber\\
& \stackrel{(c)}{\leq}& MNe^{-y\left(1-\frac{y}{K}\right)e^{-y}\beta\log N}\nonumber\\
&\leq& MN^{1-y\left(1-\frac{y}{K}\right)e^{-y}\beta\frac{1}{\ln 2}}
\nonumber\\
& \stackrel{(d)}{=}& MN^{1-\frac{1}{2}\beta (1-\frac{\ln2}{K})}
\nonumber
\end{eqnarray}
\begin{eqnarray}
&\hspace{-3.4cm} \stackrel{(e)}{=}& 2^{\beta (\delta-\epsilon) \log N}N^{1-\frac{1}{2}\beta (1-\frac{\ln2}{K})}
\nonumber\\
&\hspace{-3.4cm}=&N^{\beta (\delta-\epsilon)}N^{1-\frac{1}{2}\beta (1-\frac{\ln2}{K})}
\nonumber\\
&\hspace{-3.4cm}\leq&N^{1-\beta\left(\frac{1}{2}(1-\frac{\ln 2}{K})-\delta \right)}.
\end{eqnarray}
\fi
In the above, (a) follows by taking $p=y/K$ and setting $T$ as $\beta K\log N$, for some positive $y$ and $\beta$, to be defined. (b) follows since $e^{-y} \leq (1-y/n)^{n-1}$ for small $y>0$ and any integer $n >0$. In the sequence below, we will use it with $y=\ln 2$, for which it is true. (c) follows since $e^{-x} \ge (1-x/n)^{n}$ for $x>0$ and any integer $n >0$. (d) follows by choosing $y=\ln 2$. (e) is by setting $M=2^{T\frac{\delta-\epsilon}{K}}$ and substituting the value for $T$.
The result in \eqref{eq:PeComp} can be interpreted as follows. As long as $\delta$, the leakage probability at the eavesdropper, is smaller than $\frac{1}{2}(1-\frac{\ln 2}{K})$, choosing $T=\beta K \log N$ with a large enough $\beta$ results in an exponentially small error probability. For example, for large enough $K$ and $\delta=0.25$, one needs $\beta > 4$, that is, about $4K\log N$ tests to have an exponentially small (with $N$) error probability while using an efficient decoding algorithm. To see the dependence of the error probability on the number of tests, denote
\[
\epsilon = \beta\left(\frac{1}{2}\left(1-\frac{\ln 2}{K}\right)-\delta\right) - 1.
\]
Then, if the number of tests satisfies
\[
T \ge \frac{1+\epsilon}{\frac{1}{2}(1-\frac{\ln 2}{K})-\delta} K \log N
\]
one has
\[
P_e \leq N^{-\epsilon}.
\]
Thus, while the results under ML decoding (\Cref{direct theorem1}) show that any value of $\delta <1$ is possible (with a $\frac{1}{1-\delta}$ toll on $T$ compared to non-secure GT), the analysis herein suggests that using the efficient algorithm, one can have a small error probability only for $\delta < 1/2$, and the toll on $T$ is greater than $\frac{1}{\frac{1}{2}-\delta}$. This is consistent with the fact that this algorithm is known to achieve only half of the capacity for non-secure GT \cite{aldridge2014group}. However, both these results may be due to coarse analysis, and not necessarily due to an inherent deficiency in the algorithm.
\begin{remark}[Complexity]
It is easy to see that the algorithm runs over all rows in the codebook, and compares each one to the vector of tests' results. The length of each row is $T$. There are $N$ items, each having about $2^{\frac{\delta}{K}T}$ rows in its bin. Since $T=O(K \log N)$, we have $O(N^2)$ rows in total. Thus, the number of operations is $O(N^2T)=O(KN^2 \log N)$. This should be compared to $O(KN\log N)$ without any secrecy constraint.
\end{remark}
\Cref{fig:DND_simulation} includes simulation results of the secure DND GT algorithm proposed, compared with ML decoding and the upper and lower bounds on the performance of ML.
\ifdouble
\begin{figure}
\centering
\includegraphics[trim= 1.6cm 0.0cm 0cm 0.5cm,clip,scale=0.228]{Success_Prob_of_all_the_DND_ML_items_N_500_K_3_Delta_01_NumIterations_8000_date_2017_3_11_6_54_v3-eps-converted-to.pdf}
\caption{\emph{Definite Non-Defective} and ML simulation results.}
\label{fig:DND_simulation}
\end{figure}
\else
\begin{figure}
\centering
\includegraphics[trim= 0cm 0.0cm 0cm 0.5cm,clip,scale=0.4]{Success_Prob_of_all_the_DND_ML_items_N_500_K_3_Delta_01_NumIterations_8000_date_2017_3_11_6_54_v3.eps}
\caption{\emph{Definite Non-Defective} and ML simulation results.}
\label{fig:DND_simulation}
\end{figure}
\fi
\section{Introduction}\label{intro}
The classical version of Group Testing (GT) was suggested during World War II in order to identify syphilis\revised{-}infected draftees while dramatically reducing the number of required tests \cite{dorfman1943detection}. Specifically, when the number of infected draftees, $K$, is much smaller than the population size, $N$, instead of examining each blood sample individually, one can conduct a small number of of \emph{pooled samples}. Each pool outcome is negative if it contains no infected sample, and positive if it contains at least one infected sample. The problem is thus to identify the infected draftees via as few pooled tests as possible. \Cref{basic_testing} (a)-(c) depicts a small example.
Since its exploitation in WWII, GT \revised{has been} utilized in numerous fields, including biology and chemistry \cite{du1999combinatorial,macula1999probabilistic}, communications \cite{varanasi1995group,cheraghchi2012graph,wu2015partition,wu2014achievable}, sensor networks \cite{bajwa2007joint}, pattern matching \cite{clifford2007k} and web services \cite{tsai2004testing}. GT \revised{has} also found applications in the emerging field of Cyber Security, e.g., detection of significant changes in network traffic \cite{cormode2005s}, Denial of Service attacks \cite{xuan2010detecting} and indexing information for data forensics \cite{goodrich2005indexing}.
Many \revised{scenarios} which utilize GT \revised{involve} sensitive information which should not be revealed if some of the tests leak \revised{(for instance, if one of the several labs to which tests have been distributed for parallel processing is compromised). However, in GT, leakage of even a single pool-test outcome may reveal significant information about the tested items. If the test outcome is negative it indicates that none of the items in the pool is defective;} if it is positive, at least one of the items in the pool is defective (see \Cref{basic_testing} (d) for a short example). Accordingly, \revised{it is critical to} \revisedSr{ensure} that a leakage of a fraction of the pool-tests \revised{outcomes} to undesirable or malicious eavesdroppers does not give them any useful information on the status of the items.
It is very important to note that \emph{protecting GT is different from protecting the communication between the parties}.
\revisedSr{To protect GT, one should make sure that information about the status of individual items is not revealed if a fraction of the test outcomes leaks.}
\off{To protect the communication, one should merely protect the bits transmitted, and can do so by encoding them before the transmission.} However, in GT, we do not want to assume one entity has access to all pool-tests, and can apply some encoding function before they are exposed. We also do not want to assume a mixer can add a certain substance that will prevent a third party from testing the sample. To protect GT, one should make sure that without altering mixed samples, if a fraction of them leaks, either already tested or not, information is not revealed.
While the current literature includes several works on the privacy in GT algorithms for digital objects \cite{atallah2008private,goodrich2005indexing,freedman2004efficient,rachlin2008secrecy}, these works are based on cryptographic schemes, assume the testing matrix is not known to all parties, \revisedSr{impose} a high computational burden, and, last but not least, assume the computational power of the eavesdropper is limited \cite{tagkey2004391,C13}. \emph{Information theoretic security} \revised{considered for secure communication} \cite{C2,C13}, on the other hand, if applied appropriately to GT, can offer privacy at the price of \emph{additional tests}, without keys, obfuscation or assumptions on limited power. \revised{Due to the analogy between channel coding and group-testing regardless of security constraints, \cite{atia2012boolean,baldassini2013capacity}, in \Cref{BooleanCompressed} we present an extensive survey of the literature on secure communication as well.}
\subsection*{Main Contribution}
In this work, we formally define Secure Group Testing (SGT), suggest SGT algorithms based on information\revised{-}theoretic principles and analyse their performance. In the considered model, there is an eavesdropper \revised{Eve who might} observe part of the \revised{vector of pool-tests outcomes}. The goal of the test designer is to design the tests \revised{in a manner} such that a legitimate decoder can decode the status of the items (\revised{whether the items are} defective or not) with an arbitrarily small error probability. \revised{It should {\it also} be the case that as} long as \revised{Eve} the eavesdropper gains only part of the output vector (a fraction $\delta$ \revised{- a bound on the value of $\delta$ is known {\it a priori} to the test designer, but which specific items are observed is not), Eve} cannot (asymptotically, \revised{as the number of items being tested grows without bound}) gain any significant information on the status of \revised{any of} the items.
We propose a SGT code and \revised{corresponding decoding} algorithms which \revised{ensure high of reliability (with high probability over the test design, the legitimate decoder should be able to estimate the status of each item correctly)}, as well as \revised{\emph{strong secrecy}} conditions \revised{(as formally defined in \Cref{formulation}) - which ensures that essentially no information about the status of individual items’ leaks to Eve.}
\revised{Our first SGT code and corresponding decoding algorithm (based on Maximum Likelihood (ML) decoding) requires a number of tests that is essentially information-theoretically optimal in N, K and $\delta$ (as demonstrated in \Cref{converse} by corresponding information-theoretic converse that we also show for the problem).}
\revised{The second code and corresponding decoding algorithm, while requiring a constant factor larger number of tests than is information-theoretically necessary (by a factor that is a function of $\delta$), is computationally efficient. It maintains the reliability and secrecy guarantees, yet requires only O($N^2T$) decoding time, where $T$ is the number of tests.}
\ifdouble
\begin{figure}
\centering
\begin{subfigure}[b]{0.24\textwidth}
\includegraphics[scale=0.7]{LabPoolTesting2.png}
\caption{}
\end{subfigure}
\begin{subfigure}[b]{0.24\textwidth}
\includegraphics[scale=0.7]{LabPoolTesting3.png}
\caption{}
\end{subfigure}
\begin{subfigure}[b]{0.24\textwidth}
\includegraphics[scale=0.7]{LabPoolTesting4.png}
\caption{}
\end{subfigure}
\begin{subfigure}[b]{0.24\textwidth}
\includegraphics[scale=0.7]{SLabPoolTesting21.png}
\caption{}
\end{subfigure}
\else
\begin{figure}
\begin{flushleft}
\begin{subfigure}[b]{0.232\textwidth}
\includegraphics[scale=0.7]{LabPoolTesting2.pdf}
\caption{}
\end{subfigure}
\begin{subfigure}[b]{0.232\textwidth}
\includegraphics[scale=0.7]{LabPoolTesting3.pdf}
\caption{}
\end{subfigure}
\begin{subfigure}[b]{0.232\textwidth}
\includegraphics[scale=0.7]{LabPoolTesting4.pdf}
\caption{}
\end{subfigure}
\begin{subfigure}[b]{0.232\textwidth}
\includegraphics[scale=0.7]{SLabPoolTesting21.pdf}
\caption{}
\end{subfigure}
\end{flushleft}
\fi
\caption[]{Classical group testing: \textit{An example of test results, a simple decoding procedure at the legitimate decoder and the risk of leakage. The example includes 7 items, out of which at most one defective (the second one in this case; unknown to the decoder). Three pooled tests are conducted. Each row dictates in which pooled tests the corresponding item participates. (a) Since the first result is negative, items 1 and 6 are not defective. (b) The second result is positive, hence at least one of items 2 and 4 is defective. (c) Based on the last result, as item 4 cannot be defective, it is clear that 2 is defective. Note that decoding in this case is simple: any algorithm which will simply rule out each item whose row in the matrix is not compatible with the result will rule out all but the second item, due to the first and last test results being negative, thus identifying the defective item easily. (d) An eavesdropper who has access to part of the results (the first two) can still infer useful information. Our goal is construct a testing matrix such that such an eavesdropper remains ignorant.}}
\label{basic_testing}
\end{figure}
\off{
We further provide sufficiency and necessity bounds on the number of tests required, and show that the code construction, together with Maximum Likelihood (ML) decoding is order-optimal (in $N$, $K$ and $\delta$). The second algorithm, while missing the exact behaviour with respect to $\delta$, is highly efficient, maintains the reliability and secrecy guarantees, yet requires only $O(N^2T)$ operations, where $T$ is the number of tests.}
We do so by proposing a model, which is, in a sense, analogous to a \revised{{\it wiretap channel model}}, as depicted in \Cref{figure:group_testing_model}.
\revised{In this analogy the subset of defective items (unknown {\it a priori} to all parties) takes the place of a confidential message. The testing matrix (representing the design of the pools - each row corresponds to the tests participated in by an item, and each column corresponds to a potential test) is a succinct representation of the \revisedSr{encoder's} codebook. Rows or disjunctive unions of rows of this testing matrix can be considered as codewords.} The decoding algorithm is analogous to a channel decoding process, and the eavesdropped signal is the output of \textit{an erasure channel}, namely, having only any part of the transmitted signal from the legitimate source to the legitimate receiver.
\revised{In classical non-adaptive group-testing, each row of the testing matrix comprises of a length-$T$ binary vector which determines which pool-tests the item is tested in.
In the SGT code constructions proposed in this work, each item instead corresponds to a vector chosen uniformly at random from a pre-specified set of random and independent vectors.}
Namely, we use \emph{stochastic encoding}, and each vector \revised{corresponds to different sets of} pool-tests \revised{an item may participate in}. For each item the lab picks one of the vectors in its set \revised{(we term the set associated with item $j$ as \revisedSr{``Bin j"})} \revised{uniformly} at random, and the item participates in the pool-tests according to this randomly chosen vector.
\revised{The set \revised{(\revisedSr{``}Bin")} is known {\it a priori} to all parties, but the specific vector chosen by the encoder/mixer is only known to the encoder/mixer, and hence is {\it not a shared key/common randomness} in any sense.} A schematic description of our procedure is depicted in \Cref{fig:WiretapCoding}. \off{It is important to note, though, that the randomness is required only at the lab (the ``mixer"), and it is \emph{not a shared key} in any sense.}
\ifdouble
\begin{figure}
\centering
\includegraphics[trim=9cm 9.2cm 1cm 9.05cm,clip,scale=0.85]{SecureGroupTestModelE.pdf}
\caption{\revised{An analogy between a wiretap erasure channel and the corresponding SGT model.}}
\label{figure:group_testing_model}
\end{figure}
\else
\begin{figure}
\centering
\includegraphics[trim=6cm 9.2cm 0cm 9.05cm,clip,scale=1]{SecureGroupTestModelE.pdf}
\caption{\revised{An analogy between a wiretap erasure channel and the corresponding SGT model.}}
\label{figure:group_testing_model}
\end{figure}
\fi
Accordingly, by obtaining a pool-test result, without knowing the specific vectors chosen by the lab for each item, the eavesdropper may gain only negligible information regarding the items themselves. Specifically, we show that \revised{by careful design of the testing procedure,} even though the pool-tests in which each item participated \revised{are} chosen randomly and even though the legitimate user does not know \emph{a-priori} in which pool-tests each item has participated, the legitimate user \revised{will, with high probability over the testing procedure, be able to correctly identify the set of defective items,} while the eavesdropper, observing only a subset of the pool-test results, will have no significant information regarding the status of the items.
The structure of this work is as follows. In \Cref{BooleanCompressed}, we present an extensive survey and summarize the related work. In \Cref{formulation}, a SGT model is formally described. \Cref{main results} includes our main results, with the direct proved in \Cref{LowerBound} and converse proved in \Cref{converse}. \Cref{efficient_algorithms} describes \revised{a computationally} efficient algorithm, and proves an upper bound on its error probability. \Cref{conc} concludes the paper.
\section{Code Construction and a Proof for \Cref{direct theorem1}} \label{LowerBound}
In order to keep the eavesdropper, which obtains only a fraction $\delta$ of the outcomes, ignorant regarding the status of the items, we \emph{randomly} map the items to the tests. Specifically, as depicted in \Cref{fig:WiretapCoding}, for each item we generate a bin, containing several rows. The number of such rows \emph{corresponds} to the number of tests that the eavesdropper can obtain, yet, unlike wiretap channels, \emph{it is not identical} to Eve's capacity, and should be normalized by the number of defective items.
Then, for the $j$-th item, we randomly select a row from the $j$-th bin. This row will determine in which tests the item will participate.
In order to rigorously describe the construction of the matrices and bins, determine the exact values of the parameters (e.g., bin size), and analyze the reliability and secrecy, we first briefly review the representation of the GT problem as a channel coding problem \cite{atia2012boolean}, together with the components required for SGT.
A SGT code consists of an index set \revisedSr{$\mathcal{W} =\{1,2,\ldots \binom{N}{K}\}$}, its $w$-th item corresponding to the $w$-th subset $\mathcal{K}\subset \{1,\ldots,N\}$; A discrete memoryless source of randomness $(\mathcal{R},p_R)$, with known alphabet $\mathcal{R}$ and known statistics $p_R$; An encoder,
\begin{equation*}
f : \revisedSr{\mathcal{W}} \times \mathcal{R} \rightarrow \mathcal{X}_{S_w}\in\{0,1\}^{K\times T}
\end{equation*}
which maps the index $W$ of the defective items to a matrix $\textbf{X}^{T}_{S_{w}}$ of codewords, each of its rows corresponding to a different item in the index set $S_w$\revisedSr{, $w\in \mathcal{W}$, $|S_w|=K$}.
The need for a \emph{stochastic encoder} is similar to most encoders ensuring information theoretic security, as randomness is required to confuse the eavesdropper about the actual information \cite{C13}. Hence, we define by $R_K$ the random variable encompassing the randomness required \emph{for the $K$ defective items}, and by $M$ the number of rows in each bin. Clearly, $M^K=H(R_K)$.
At this point, \revisedSr{an} important clarification is in order. The lab, of course, does not know which items are defective. Thus, operationally, it needs to select a row for each item. However, in \emph{the analysis}, since only the defective items affect the output (that is, only their rows are \revisedSr{ORed together} to give $Y^T$), we refer to the ``message" as the index of the defective set $w$ and refer only to the random variable $R_K$ required to choose the rows in their bins.
In other words, unlike the analogous communication problem, in GT, \emph{nature} performs the actual mapping from $W$ to $\textbf{X}^{T}_{S_{w}}$. The mixer only mixes the blood samples according to the (random in this case) testing matrix it has.
A decoder at the legitimate user is a map
\begin{equation*}
\hat{W} : \revisedSr{\mathcal{Y}}^{T} \rightarrow \revisedSr{\mathcal{W}}.
\end{equation*}
The probability of error is $P(\hat{W}(Y^{T})\neq W)$. The probability that an outcome test leaks to the eavesdropper is $\delta$. We assume a memoryless model, i.e., each outcome $Y(t)$ depends only on the corresponding input $X_{S_w}(t)$, and the eavesdropper observes $Z(t)$, generated from $Y(t)$ according to
\begin{equation*}
p(Y^T,Z^T|X_{S_w})=\prod_{t=1}^{T}p(Y(t)|X_{S_w}(t))p(Z(t)|Y(t)).
\end{equation*}
We may now turn to the detailed construction and analysis.
\ifdouble
\begin{figure}
\centering
\includegraphics[trim= 9.3cm 8cm 2cm 7.6cm,clip,scale=0.9]{Wiretap_coding5_one_col.pdf}
\caption{Binning and encoding process for a SGT code.}
\label{fig:WiretapCoding}
\end{figure}
\else
\begin{figure}
\centering
\includegraphics[trim= 6.3cm 8cm 2cm 7.6cm,clip,scale=1]{Wiretap_coding5_one_col.pdf}
\caption{Binning and encoding process for a SGT code.}
\label{fig:WiretapCoding}
\end{figure}
\fi
\subsubsection{Codebook Generation}
\revisedSr{Choose M such that $$\log_2(M) = T(\delta-\epsilon^{\prime})/(K)$$}\off{Set $M = 2^{T\frac{\delta-\revised{\epsilon^{\prime}}}{K}}$.}\revisedSr{for some $\epsilon^{\prime}>0$. $\epsilon^{\prime}$ will affect the equivocation.}
Using a distribution $P(X^{T})=\prod^{T}_{i=1}P(x_i)$, for each item generate $M$ independent and identically distributed codewords $x^{T}(m)$, $1 \leq m \leq M$\off{, where $\epsilon$ can be chosen arbitrarily small}.
The codebook is depicted in the left hand side of \Cref{fig:WiretapCoding}.
Reveal the codebook to Alice and Bob. We assume Eve may have the codebook as well.
\subsubsection{Testing}
For each item $j$, the mixer/lab selects uniformly at random one codeword $x^{T}(m)$ from the $j$-th bin.
Therefore, the SGT matrix contains $N$ randomly selected codewords of length $T$, one for each item, defective or not. Amongst is an unknown subset $X^{T}_{S_{w}}$, with the index $w$ representing the true \revisedSr{defective items}. An entry of the $j$-th random codeword is $1$ if the $j$-item is a member of the designated pool test and $0$ otherwise.
\subsubsection{Decoding at the Legitimate Receiver}
The decoder looks for a collection of $K$ codewords $X_{S_{\hat{w}}}^{T}$, \textit{one from each bin}, for which $Y^T$ is most likely. Namely,
\begin{equation*}
P(Y^{T}|X_{S_{\hat{w}}}^{T})>P(Y^{T}|X_{S_{w}}^{T}), \forall w \neq \hat{w}.
\end{equation*}
Then, the legitimate user (Bob) declares $\hat{W}(Y^T)$ as the set of bins in which the rows $\hat{w}$ reside.
\subsection{Reliability}
Let $(\mathcal{S}^{1},\mathcal{S}^{2})$ denote a partition of the defective set $S$ into disjoint sets $\mathcal{S}^{1}$ and $\mathcal{S}^{2}$, with cardinalities $i$ and $K-i$, respectively.\footnote{\revisedSr{This partition helps decompose the error events into classes, where in class $i$ one already knows $K-i$ defective items, and the dominant error event corresponds to missing the other $i$. Thus, it is easier to present the error event as one ``codeword" against another. See Appendix~\ref{appendix:APPENDIX A} for the details.}}
Let $I(X_{\mathcal{S}^1};X_{\mathcal{S}^2},Y)$ denote the mutual information between $X_{\mathcal{S}^1}$ and $X_{\mathcal{S}^2},Y$, under the i.i.d.\ distribution with which the codebook was generated and remembering that $Y$ is the output of a Boolean channel.
The following lemma is a key step in proving the reliability of the decoding algorithm.
\begin{lemma}\label{direct lemma1}
If the number of tests satisfies
\begin{eqnarray*}
T \geq (1+\varepsilon)\cdot\max_{i=1,\ldots ,K} \frac{\log\binom{N-K}{i}M^i}{I(X_{\mathcal{S}^1};X_{\mathcal{S}^2},Y)},
\end{eqnarray*}
then, under the codebook above, as $N\rightarrow \infty$ the average error probability approaches zero.
\end{lemma}
Next, we \revisedn{prove} \Cref{direct lemma1}, which extends the results in \cite{atia2012boolean} to the codebook required for SGT. Specifically, to obtain a bound on the required number of tests as given in \Cref{direct lemma1}, we first state Lemma 2, which bounds the error probability of the ML decoder using a Gallager-type bound \cite{gallager1968information}.
\revisedn{
\begin{definition}
The probability of error event $E_{i}$ in the ML decoder defined, as the event of mistaking the true set for a set which differs from it in exactly $i$ items.
\end{definition}
\begin{lemma}\label{error lemma2}
The error probability $P(E_{i})$ is bounded by
\begin{equation*}
P(E_{i}) \leq 2^{-T\left(E_{o}(\rho)-\rho\frac{\log\binom{N-K}{i}M^i}{T}-\frac{\log\binom{K}{i}}{T}\revisedr{-\frac{K}{T}}\right)},
\end{equation*}
where the error exponent $E_{o}(\rho)$ is given by
\ifdouble
\begin{multline}
E_{o}(\rho)= -\log \sum_{Y\in \{0,1\}}\sum_{X_{\mathcal{S}^2}\in \{0,1\}}\Bigg[\sum_{X_{\mathcal{S}^1}\in \{0,1\}}P(X_{\mathcal{S}^1})\\
p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})^{\frac{1}{1+\rho}}\Bigg]^{1+\rho},\text{ } 0 \leq\rho \leq1.\nonumber
\end{multline}
\else
\begin{equation*}
E_{o}(\rho)= -\log \sum_{Y\in \{0,1\}}\sum_{X_{\mathcal{S}^2}\in \{0,1\}}\Bigg[\sum_{X_{\mathcal{S}^1}\in \{0,1\}}P(X_{\mathcal{S}^1})
p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})^{\frac{1}{1+\rho}}\Bigg]^{1+\rho},\text{ } 0 \leq\rho \leq1.
\end{equation*}
\fi
\end{lemma}
In Appendix \ref{appendix:APPENDIX A} we analyze the bound provided in \Cref{error lemma2}.
Note that there are two main differences compared to non-secured GT. First, the decoder has $\binom{N}{K}M^K$ possible subsets of codewords to choose from, $\binom{N}{K}$ for the number of possible bins and $M^K$ for the number of possible rows to take in each bin. Thus, when fixing the error event, there are $\binom{N-K}{i}M^i$ subsets to confuse the decoder.
Moreover, due to the bin structure of the code, there are also many ``wrong" codewords which are not the one transmitted on the channel, hence create a decoding error codeword-wise, yet the actual bin decoded may still be the right one.}
\begin{proof}[Proof of \Cref{direct lemma1}]
\revisedn{
For this lemma, we follow the derivation in \cite{atia2012boolean}. However, due the different code construction, the details are different.
Specially, for each item there is a \emph{bin} of codewords, from which the decoder has to choose. Define
\begin{equation*}
\emph{f}(\rho) = E_{o}(\rho)-\rho\frac{\log\binom{N-K}{i}M^i}{T}-\frac{\log\binom{K}{i}}{T}\revisedr{-\frac{K}{T}}.
\end{equation*}
We wish to show that $T\emph{f}(\rho)\rightarrow \infty$ as $N \rightarrow \infty$.
Note that $\log\binom{K}{i}$ is a constant for the fixed $K$ regime. Thus for large $T$ we have $\lim_{T \rightarrow \infty} \emph{f}(0) = 0$.}
Since the function $\emph{f}(\rho)$ is differentiable and has a power series expansion, for a sufficiently small $\alpha$, by Taylor series expansion in the neighborhood of $\rho \in [0,\alpha]$ \revisedn{we have}
\begin{equation*}
\emph{f}(\rho) = \emph{f}(0) + \rho \frac{d\emph{f}}{d\rho} \mid_{\rho = 0} + \Theta(\rho^2).
\end{equation*}
\revisedn{Now,}
\ifdouble
\begin{eqnarray}
&& \hspace{-0.85cm} \frac{\partial E_{o}}{\partial\rho} |_{\rho = 0}\nonumber\\
&\hspace{-0.7cm} = &\hspace{-0.5cm} \sum_Y \sum_{X_{\mathcal{S}^2}}[\sum_{X_{\mathcal{S}^1}}P(X_{\mathcal{S}^1})p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})\log p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})\nonumber\\
&\hspace{-0.7cm} - &\hspace{-0.5cm} \sum_{X_{\mathcal{S}^1}}P(X_{\mathcal{S}^1})p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1}) \sum_{X_{\mathcal{S}^1}}P(X_{\mathcal{S}^1})p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})]\nonumber\\
&\hspace{-0.7cm} = &\hspace{-0.5cm} \sum_Y \sum_{X_{\mathcal{S}^2}}\sum_{X_{\mathcal{S}^1}}P(X_{\mathcal{S}^1})p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})\nonumber\\
&& \hspace{-0.85cm} \log\frac{p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})}{\sum_{X_{\mathcal{S}^1}}P(X_{\mathcal{S}^1})p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})
= I(X_{\mathcal{S}^1};X_{\mathcal{S}^2},Y).\nonumber
\end{eqnarray}
\else
\begin{eqnarray}
&& \hspace{-0.85cm} \frac{\partial E_{o}}{\partial\rho} |_{\rho = 0}\nonumber\\
&\hspace{-0.7cm} = &\hspace{-0.5cm} \sum_Y \sum_{X_{\mathcal{S}^2}}[\sum_{X_{\mathcal{S}^1}}P(X_{\mathcal{S}^1})p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})\log p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})\nonumber\\
&\hspace{-0.7cm} - &\hspace{-0.5cm} \sum_{X_{\mathcal{S}^1}}P(X_{\mathcal{S}^1})p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1}) \sum_{X_{\mathcal{S}^1}}P(X_{\mathcal{S}^1})p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})]\nonumber\\
&\hspace{-0.7cm} = &\hspace{-0.5cm} \sum_Y \sum_{X_{\mathcal{S}^2}}\sum_{X_{\mathcal{S}^1}}P(X_{\mathcal{S}^1})p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1}) \log\frac{p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})}{\sum_{X_{\mathcal{S}^1}}P(X_{\mathcal{S}^1})p(Y,X_{\mathcal{S}^2}|X_{\mathcal{S}^1})
= I(X_{\mathcal{S}^1};X_{\mathcal{S}^2},Y).\nonumber
\end{eqnarray}
\fi
\revisedn{
Hence, if
\begin{equation*}
(1+\varepsilon)\frac{\log\binom{N-K}{i}M^i}{T}<I(X_{\mathcal{S}^1};X_{\mathcal{S}^2},Y)
\end{equation*}
for some constant $\varepsilon > 0$, the exponent is positive for large enough $T$ and we have $P(E_{i})\rightarrow 0$ as $T\rightarrow \infty$ for $\rho > 0$.
Using a union bound one can show that taking the maximum over $i$ will ensure a small error probability in total.}
\end{proof}
\revisedn{The expression $I(X_{\mathcal{S}^1};X_{\mathcal{S}^2},Y)$ in \Cref{direct lemma1} is critical to understand how many tests are required, yet it is not a function of the problem parameters in any straight forward mannar. We now bound it to get a better handle on $T$.}
\begin{claim}\label{InformationClaim}
For large $K$, and under a fixed input distribution for the testing matrix $(\frac{\ln(2)}{K},1-\frac{\ln(2)}{K})$, the mutual information between $X_{\mathcal{S}^1}$ and $(X_{\mathcal{S}^2},Y)$ is lower bounded by
\begin{equation*}
I(X_{\mathcal{S}^1};X_{\mathcal{S}^2},Y)\geq \frac{i}{K}.
\end{equation*}
\end{claim}
\begin{proof}[Proof of \Cref{InformationClaim}]\revisedn{First, note that}
\begin{eqnarray*}\label{eq:MutualInfB}
I(X_{\mathcal{S}^1};X_{\mathcal{S}^2},Y)
&&\hspace{-0.6cm} \stackrel{(a)}{=} I(X_{\mathcal{S}^1};X_{\mathcal{S}^2})+I(X_{\mathcal{S}^1};Y|X_{\mathcal{S}^2})\nonumber\\
&&\hspace{-0.6cm} = H(Y|X_{\mathcal{S}^2})-H(Y|X_{\mathcal{S}})\nonumber\\
&&\hspace{-0.6cm} \stackrel{(b)}{=} q^{K-i}H(q^{i})\nonumber\\
&&\hspace{-0.6cm} = q^{K-i}\left[q^{i}\log\frac{1}{q^{i}}
+ \left( 1 - q^{i}\right)\log\frac{1}{\left(1- q^{i}\right)}\right],
\end{eqnarray*}
\revisedn{where equality (a) follows since the rows of the testing matrix are independent, and (b) follows since $H(Y|X_{\mathcal{S}})$ is the uncertainty of the legitimate receiver given $X_{\mathcal{S}}$, thus when observing the noiseless outcomes of all pool tests, this uncertainty is zero. Also, note that the testing matrix is random and i.i.d.\ with distribution $(1-q,q)$, hence the probability for $i$ zeros is $q^i$.}
\revisedn{Then, under a fixed input distribution for the testing matrix $(p=\frac{\ln(2)}{K},q=1-\frac{\ln(2)}{K})$ and large $K$ it is easy to verify that the bounds meet at the two endpoint of $i=1$ and $i=K$, yet the mutual information is concave in $i$ thus the bound is obtained. This is demonstrated graphically in \Cref{figure:Mutual_Information_Bound}.}
\end{proof}
\ifdouble
\begin{figure}
\centering
\includegraphics[trim=2.2cm 14.4cm 1.5cm 7.2cm,clip,scale=0.52]{MATLAB.pdf}
\caption{Mutual Information Bound}
\label{figure:Mutual_Information_Bound}
\end{figure}
\else
\begin{figure}
\centering
\includegraphics[trim=2.2cm 14.4cm 1.5cm 7.2cm,clip,scale=0.8]{MATLAB.pdf}
\caption{Mutual Information Bound}
\label{figure:Mutual_Information_Bound}
\end{figure}
\fi
Applying \Cref{InformationClaim} to the expression in \Cref{direct lemma1}, we have
\begin{equation*}
\frac{\log\binom{N-K}{i}M^i}{I(X_{\mathcal{S}^1};X_{\mathcal{S}^2},Y)} \leq \frac{\log\binom{N-K}{i}M^i}{\frac{i}{K}}.
\end{equation*}
Hence, substituting $M=2^{T\frac{\delta-\revised{\epsilon_K}}{K}}$, a sufficient condition for reliability is
\begin{eqnarray*}
T &\ge& \max_{1 \leq i \leq K}\frac{1+\varepsilon}{\frac{i}{K}}\left[ \log\binom{N-K}{i} + \frac{i}{K}T(\delta-\revised{\epsilon_K}) \right]
\end{eqnarray*}
Rearranging terms results in
\begin{eqnarray}\label{eq:reduce_h}
T & \ge & \max_{1 \leq i \leq K} \frac{1}{1-\revised{\delta +\epsilon_K-\varepsilon\delta+\varepsilon\epsilon_K}}\frac{1+\varepsilon}{i/K}\log\binom{N-K}{i},\nonumber
\end{eqnarray}
\revised{where by reducing $\epsilon_K$ and $\varepsilon\epsilon_K$ we increase the bound on $T$, and with some constant $\varepsilon > 0$.}
Noting that this is for large $K$ and $N$, and that $\varepsilon$ is independent of them, achieves the bound on $T$ provided in \Cref{direct theorem1} and reliability is established.
\off{
\subsection{Scaling Regime Of $K=o(N)$} \label{K=o(N)}
Now, we point out that the result in \Cref{direct lemma1} can be extended to the more general case where both $N$ and $K$ are allowed to scale simultaneously such that $K=o(N)$.
From the Lagrange form of the Taylor Series expansion, we can write $E_{o}(\rho)$ in terms of its first derivative evaluated at zero and a remainder term which depends on the second derivative.
We have already shown that $E_{o}(0)=0$ and $E_{o}^{\prime}(0)=I(X_{\mathcal{S}^1};Y|X_{\mathcal{S}^2})>0$.
Consequently, we need to lower bound $E_{o}(\rho)$ by taking the worst-case second derivative.
Establishing this result requires a more careful analysis, and hence, we present it in Appendix \ref{appendix:APPENDIX D}.
It is shown that the same sufficient condition on the number of tests in \Cref{direct lemma1} holds up to an extra $polylog(K)$ factor for an arbitrarily small average error probability, thus,
\begin{equation*}
\lim_{\substack{N \rightarrow \infty\\K=o(N)}}P_e\rightarrow 0.
\end{equation*}
Namely,
\begin{lemma}\label{direct lemma3}
If the number of tests satisfies
\begin{eqnarray*}
T & \geq & \max_{1 \leq i \leq K} \frac{\log\binom{N-K}{i}+(\log\binom{K}{i}K\revisedn{+ K})\log^2(K/i)}{(1/2-\delta)i/K},
\end{eqnarray*}
then, under the codebook above, as $N\rightarrow \infty$ and $K=o(N)$ the average error probability approaches zero.
\end{lemma}
\begin{proof}
From the expression of $P(E_i)$ we have that
\begin{multline*}
\hspace{-0.4cm} \sum_{i}P(E_i) \leq K \max_{i} P(E_i)\leq\max_{i}K \revisedn{\exp K}\\
\exp\left(-T\left(E_{o}(\rho)-\rho\frac{\log\binom{N-K}{i}M^{i}}{T}-\frac{\log\binom{K}{i}}{T}\right)\right).
\end{multline*}
Consequently, we need to ensure that
\begin{equation*}
TE_{o}(\rho) \geq \rho\log\binom{N-K}{i}M^{i}+\log\binom{K}{i}+\log K \revisedn{+ K}.
\end{equation*}
Now using the lower bound for $E_{o}(\rho)$ , we have
\begin{eqnarray*}
E_{o}(\rho) &\geq& \rho I(X_{\mathcal{S}^1};X_{\mathcal{S}^2},Y)-\frac{\rho^2}{2}|(E_{o}(0))^{\prime\prime}|\\
& \geq & \rho I(X_{\mathcal{S}^1};X_{\mathcal{S}^2},Y)-\frac{\rho^2}{2}\frac{i}{K}\log^2\frac{K}{i}\\
& \geq & \rho \frac{i}{K}-\frac{\rho^2}{2}\frac{i}{K}\log^2\frac{K}{i}
\end{eqnarray*}
where the last inequality follows from the lower bound for the mutual information in the noiseless case provided in \Cref{InformationClaim}.
Substituting this lower bound, we obtain
\begin{eqnarray*}
T\rho\frac{i}{K}&&\hspace{-0.7cm} \left( 1 -\frac{\rho}{2}\log^2\frac{K}{i} \right)\\
&& \geq \rho\log\binom{N-K}{i}M^{i}+\log\binom{K}{i}+\log K \revisedn{+ K}.
\end{eqnarray*}
Hence, with $M=2^{T\frac{\delta}{K}}$, since $\epsilon$ can be chosen arbitrarily small, we have
\begin{eqnarray*}
T\rho\frac{i}{K}&&\hspace{-0.7cm} \left( 1-\frac{\rho}{2}\log^2\frac{K}{i} \right)\\
&& \geq \rho\log\binom{N-K}{i}2^{T\frac{i\delta}{K}}+\log\binom{K}{i}+\log K \revisedn{+ K}.
\end{eqnarray*}
By choosing $\rho=\frac{1}{\log^2(K/i)}$ and
\begin{eqnarray*}
T & \geq & \max_{1 \leq i \leq K} \frac{\log\binom{N-K}{i}+(\log\binom{K}{i}K\revisedn{+ K})\log^2(K/i)}{(1-1/2-\delta)i/K},
\end{eqnarray*}
the inequality is satisfied.
Noting that this is for $K = o(N)$, achieves the bound on $T$ provided in \Cref{direct lemma3} and the reliability is established.
\end{proof}
}
\subsection{Information Leakage at the Eavesdropper}\label{LowerBoundLeakage}
We now prove the security constraint is met. Hence, we wish to show that $I(W;Z^{T})/T\rightarrow 0$, as $T\rightarrow \infty$.
Denote by $\mathcal{C}_T$ the random codebook and by $X_{\mathcal{S}}^T$ the set of codewords corresponding to the true defective items. We have,
\begin{eqnarray}
&& \hspace{-0.8cm}\textstyle\textstyle \frac{1}{T}I(W;Z^T|\mathcal{C}_{T}) = \frac{1}{T}\left(I(W,R_K;Z^T|\mathcal{C}_{T})-I(R_K;Z^T|W,\mathcal{C}_{T})\right)\nonumber\\
&\hspace{-0.6cm} \stackrel{(a)}{=} &\hspace{-0.5cm}\textstyle \frac{1}{T}\left(I(X_{\mathcal{S}}^T;Z^T|\mathcal{C}_{T})-I(R_K;Z^T|W,\mathcal{C}_{T})\right)\nonumber\\
&\hspace{-0.6cm} = &\hspace{-0.5cm}\textstyle \frac{1}{T}(I(X_{\mathcal{S}}^T;Z^T|\mathcal{C}_{T})-H(R_K|W,\mathcal{C}_{T})+H(R_K|Z^T,W,\mathcal{C}_{T}))\nonumber\\
&\hspace{-0.6cm} \stackrel{(b)}{=} & \hspace{-0.5cm}\textstyle \frac{1}{T}\left(I(X_{\mathcal{S}}^T;Z^T|\mathcal{C}_{T})-H(R_K)+H(R_K|Z^T,W,\mathcal{C}_{T})\right)\nonumber\\
&\hspace{-0.6cm} \stackrel{(c)}{=}&\hspace{-0.5cm} I(X_{\mathcal{S}};Z|\mathcal{C}_{T}) - \frac{1}{T}K\log M + \frac{1}{T}H(R_K|Z^T,W,\mathcal{C}_{T})\nonumber\\
&\hspace{-0.6cm} \stackrel{(d)}{\leq}&\hspace{-0.5cm} \delta - \frac{1}{T}K\left(T \frac{\delta-\epsilon^{\prime}}{K}\right) + \frac{1}{T}H(R_K|Z^T,W,\mathcal{C}_{T})\stackrel{(e)}{\leq} \epsilon_T + \revisedSr{\epsilon^{\prime}},\nonumber
\end{eqnarray}
where $\epsilon_T \to 0$ as $T \to \infty$\off{and \revised{$\epsilon_K \to 0$ as $K \to \infty$}}. (a) is since there is a $1:1$ correspondence between $(W,R_K)$ and $X_{\mathcal{S}}^T$; (b) is since $R_K$ is independent of $W$ and $\mathcal{C}_T$; (c) is since in this \emph{direct result} the codebook is defined by the construction and is memoryless, as well as the channel; (d) is since by choosing an i.i.d.\ distribution for the codebook one easily observes that $I(X_{\mathcal{S}};Z| \mathcal{C}_T) \leq \delta$. Finally, (e) is for the following reason: Given $W$ and the codebook, Eve has a perfect knowledge regarding the \emph{bins from which the codewords were selected}. It requires to see whether she can indeed estimate $R_K$. Note that the channel Eve sees in this case is the following \emph{multiple access channel}: each of the defective items can be considered as a ``user" with $M$ messages to transmit. Eve's goal is to decode the messages from all users. This is possible if the rates at which the users transmit are within the capacity region of this MAC. Indeed, this is a (binary) Boolean MAC channel, followed by a simple erasure channel. The sum capacity cannot be larger than $\delta$, and this sum capacity is easily achieved by letting one user transmit at a time, or, in our case, \revised{where the codebook is randomly i.i.d.\ distributed, under a fixed input distribution $p=1-2^{\frac{1}{K}}$. Since actually, we use input distribution for the testing matrix of $(\frac{\ln(2)}{K},1-\frac{\ln(2)}{K})$, and $\lim_{K \to \infty} 1-2^{\frac{1}{K}}/\frac{\ln(2)}{K}=1$, for large $K$, each user obtain the same capacity; namely, each user sees a capacity of $\delta/K$.\off{ Hence Eve may decode if $\frac{1}{T}\log M$ is smaller than that value.}}
\off{
\revised{since the codebook is randomly i.i.d.\ distributed, under a fixed input distribution for the testing matrix $(\frac{\ln(2)}{K},1-\frac{\ln(2)}{K})$ and for large $K$, each user obtain the same capacity, namely, each user sees a capacity of $\delta/K$.} Hence Eve may decode if $\frac{1}{T}\log M$ is smaller than that value.}
\begin{remark}
Under non-secure GT, it is clear that simply adding tests to a given GT code (increasing $T$) can only improve the performance of the code (in terms of reliability). A legitimate decoder can always disregard the added tests. For SGT, however, the situation is different. Simply adding tests to a given code, \emph{while fixing the bin sizes}, might make the vector of results vulnerable to eavesdropping. In order to increase reliability, one should, of course, increase $T$, but also increase the bin sizes proportionally, so the secrecy result above will still hold. This will be true for the efficient algorithm suggested in \Cref{efficient_algorithms} as well.
\end{remark}
\begin{remark}
\revised{To establish the weak secrecy \revisedSr{constraint} we set $M$ to be $2^{T\frac{\delta-\revisedSr{\epsilon^{\prime}}}{K}}$, where the readability is archived without any \revisedSr{constraint} on $\revisedSr{\epsilon^{\prime}}$. However, in Appendix~\ref{strong_secrecy} to establish the strong secrecy \revisedSr{constraint} we require $M$ to be $2^{T\frac{\delta+\revisedSr{\epsilon^{\prime}}}{K}}$.}
\end{remark}
\begin{remark}
\revisedSr{Note that since
\begin{eqnarray*}
\frac{1}{T}H(R_K|Z^T,W,\mathcal{C}_{T}) \hspace{-0.2cm} &=\hspace{-0.2cm}& \frac{1}{T}H(R_K)-\frac{1}{T}I(R_K;Z^T,W,\mathcal{C}_{T}) \\
\hspace{-0.2cm}&=\hspace{-0.2cm}& \frac{1}{T}\log M - \frac{1}{T}I(R_K;Z^T,W,\mathcal{C}_{T}),
\end{eqnarray*}
any finite-length approximation for $I(R_K;Z^T,W,\mathcal{C}_{T})$ will give a finite length approximation for the leakage at the eavesdropper.
For example, one can use the results in \cite{polyanskiy2010channel}, to show that the leakage can be approximated as $\frac{1}{\sqrt{T}}+\epsilon^{\prime}$}.
\end{remark}
\section{Main Results}\label{main results}
Under the model definition given in \Cref{formulation}, our main results are the following sufficiency (direct) and necessity (converse) conditions, characterizing the maximal number of tests required to guarantee both reliability and security. The proofs are deferred to \Cref{LowerBound} and \Cref{converse}.
\subsection{Direct (Sufficiency)}
The sufficiency part is given by the following theorem.
\begin{theorem}\label{direct theorem1}
Assume a SGT model with $N$ items, out of which $K=O(1)$ are defective. For any $0 \leq \delta < 1$, if
\begin{equation}\label{main_result_eq}
T \geq \max_{i=1,\ldots ,K} \frac{1+\varepsilon}{1-\delta} \frac{K}{i}\log\binom{N-K}{i},
\end{equation}
for some $\varepsilon \revisedSr{>} 0$ independent of $N$ and $K$, then there exists a sequence of SGT algorithms which are reliable and secure. That is, as $N\rightarrow \infty$, both the average error probability approaches zero \revisedSr{exponentially} and an eavesdropper with leakage probability $\delta$ is kept ignorant.
\end{theorem}
The construction of the SGT algorithm, together with the proofs of reliability and secrecy are deferred to Section \ref{LowerBound}.
\revisedSr{In fact, in \Cref{LowerBound} we actually prove that the maximal error probability decays to $0$.} However, a few important remarks are in order now.
First, rearranging terms in \cref{main_result_eq}, we have
\begin{equation*}\label{main_result}
T \geq \frac{1}{1-\delta} \max_{i=1,\ldots ,K}\frac{(1+\varepsilon)K}{i}\log\binom{N-K}{i}.
\end{equation*}
That is, compared to only a reliability constraint, the number of tests required for \emph{both reliability and secrecy} is increased by the multiplicative factor $\frac{1}{1-\delta}$, where, again, $\delta$ is the leakage probability at the eavesdropper.
\revisedSr{The result given in \Cref{direct theorem1} uses an ML decoding at the legitimate receiver. The complexity burden in ML, however, prohibits the use of this result for large N. In \Cref{direct theorem5}, we suggest an efficient decoding algorithm, which maintains the reliability and the secrecy results using a much simpler decoding rule, at the price of only slightly more tests.}
\revisedSr{Using an upper bound on $\log\binom{N-K}{i}$, the maximization in \Cref{direct theorem1} can be solved easily, leading to simple bound on $T$ with tight scaling and only a moderate constant.
\begin{corollary}\label{corollary}
For SGT with parameters $K << N$ and $T$, reliability and secrecy can be maintained if
\begin{equation*}
T \geq \frac{1+\varepsilon}{1-\delta} K \log (N-K)e.
\end{equation*}
\end{corollary}
\begin{proof}
Substituting $\log\binom{N-K}{i} \leq i\log \frac{(N-K)e}{i}$, the maximum over $i$ is easily solved.
\end{proof}
Note that together with the converse below, this suggests $T= \Theta \left(\frac{K\log N}{1-\delta}\right)$, and, a $\Theta \left(K\log N\right)$ result for $\delta$ bounded away from $1$.
\off{Finally, we note that while the results in \Cref{direct theorem1} are only for the constant $K$ regime, in principle, using the techniques in \cite{aldridge2017capacity}, can be extended to a larger regime, but for ease of presentation, we focus on the constant regime in this paper.}}
\subsection{Converse (Necessity)}
The necessity part is given by the following theorem.
\begin{theorem}\label{converse theorem}
Let $T$ be the minimum number of tests necessary to identify a defective set of cardinality $K$ among population of size $N$ while keeping an eavesdropper, with a leakage probability $\delta < 1$, ignorant regarding the status of the items. Then, \off{for large enough $N$ one must have}\revisedSr{if $\frac{1}{T} I(W;Z^T)<\epsilon$, one must have:}
\begin{eqnarray*}
T \ge \frac{1-\epsilon_T}{1-\delta}\log\binom{N}{K},
\end{eqnarray*}
\revisedSr{where $\epsilon_T = \epsilon +\tilde{\epsilon}_T$, with $\tilde{\epsilon}_T\rightarrow 0$ as $T\rightarrow \infty$.}
\end{theorem}
\revisedn{The lower bound is derived using Fano's inequality to address reliability, assuming a negligible mutual information at the eavesdropper, thus keeping an eavesdropper with leakage probability $\delta$ ignorant, and information inequalities bounding the rate of the message on the one hand, and the data Eve \emph{does not see} on the other. Compared with the lower bound without security constraints, it is increased by the multiplicative factor $\frac{1}{1-\delta}$.}
\subsection{Secrecy capacity in SGT}
Returning to the analogy in \cite{baldassini2013capacity} between channel capacity and group testing, one might define by $C_s$ the (asymptotic) minimal threshold value for $\log\binom{N}{K}/T$, above which no reliable and secure scheme is possible. Under this definition, the result in this paper show that $C_s = (1-\delta)C $, where $C$ is the capacity without the security constraint. Clearly, this can be written as
\begin{equation*}
C_s = C - \delta C,
\end{equation*}
raising the usual interpretation as the \emph{difference} between the capacity to the legitimate decoder and that to the eavesdropper \cite{C13}. Note that as the effective number of tests Eve sees is $T_e = \delta T$, her GT capacity is $\delta C$.
\subsection{Efficient Algorithms}
Under the SGT model definition given in \Cref{formulation}, we further consider a computationally efficient algorithm at the legitimate decoder. Specifically, we analyze the \emph{Definite Non-Defective} (DND) algorithm (originally called \emph{Combinatorial Orthogonal Matching Pursuit} (COMP)), considered for the non-secure GT model in the literature \cite{chan2014non,aldridge2014group}. The theorem below states that indeed efficient decoding (with arbitrarily small error probability) and secrecy are possible, at the price of a higher $T$. Interestingly, the theorem applies to any $K$, and not necessarily only to $K = O(1)$. This is, on top of the reduced complexity, an important benefit of the suggested algorithm.
\begin{theorem}\label{direct theorem5}
Assume a SGT model with $N$ items, out of which $K$ are defective. Then, for any $\delta < \frac{1}{2}\left(1-\frac{\ln 2}{K}\right)$, there exists an efficient decoding algorithm, requiring $O(N^2T)$ operations, such that if the number of tests satisfies
\[
T \ge \frac{1+\epsilon}{\frac{1}{2}(1-\frac{\ln 2}{K})-\delta} K \log N
\]
its error probability is upper bounded by
\[
P_e \leq N^{-\epsilon}.
\]
\end{theorem}
The construction of the DND GT algorithm, together with the proofs of reliability and secrecy are deferred to \Cref{efficient_algorithms}.
\off{\revisedSr{The efficiency of the result given in \Cref{direct theorem5} gain/larger range of $K$ comes at a cost of an increased number of tests/smaller parameter range of $\delta$ where the results are valid.}}
Clearly, the benefits of the algorithm above come at the price of additional tests and a smaller range of $\delta$ it can handle.
\off{
We will note that in the same way that we analyze the DND algorithm in \Cref{efficient_algorithms}, where the decoder consider each of the $MN$ codewords separably, as in the non-secure case for the $N$ codewords, without to take into account the binning structure of the code, we suppose that it is possible to analyze the \emph{Definite Defective} (DD) algorithm proposed in \cite{aldridge2014group} for the case where there is eavesdropper which we want to kept ignorant.
\subsubsection{Secure Definite defective algorithm}
The construction of the Secure-DD GT algorithm, together with the proofs of reliability and secrecy are deferred to Section \ref{DD}.
}
\section{Background and Related Work}\label{BooleanCompressed}
\subsection{Group-testing}
\revised{Group-testing comes in various flavours, and the literature on these is vast. At the risk of leaving out much, we reference here just some of the models that have been considered in the literature, and specify our focus in this work.}
\subsubsection{Performance Bounds}
GT can be \emph{non-adaptive}, where the testing matrix is designed beforehand, \revisedSr{{\it adaptive}}, where each new test can be designed while taking into account previous test \emph{results}, or a combination of the two, where testing is adaptive, yet with batches of non-adaptive tests. It is also important to distinguish between \emph{exact recovery} and a \emph{vanishing probability of error}.
\off{\revisedSid{Adaptive versus non-adaptive, exact recovery versus approximate recovery, zero-error reconstruction versus vanishing error reconstruction, noiseless versus noisy test outcomes, binary versus non-binary test outcomes.}
\revisedSid{2. Adaptive versus non-adaptive group-testing: The items to be tested in a pool may be chosen as a function of previous test outcomes (this model corresponds to {\it adaptive group-testing}), or chosen independently of prior test outcomes (this model corresponds to {\it adaptive group-testing}). Various computationally efficient algorithms for adaptive group-testing algorithms (see for example \cite{aldridge2014group, damaschke2010competitive})}}
To date, the best known lower bound on the number of tests required (non-adaptive, exact recovery) is $\Omega(\frac{K^2}{\log K}\log N)$ \cite{furedi1996onr}. The best known explicit constructions were given in \cite{porat2011explicit}, resulting in $O(K^2 \log N)$. However, focusing on exact recovery requires more tests, and forces a combinatorial nature on the problem. Settling for high probability reconstructions allows one to reduce the number of tests to the order of $K\log N$.\footnote{A simple information theoretic argument explains a lower bound. There are $K$ defectives out of $N$ items, hence $N \choose K$ possibilities to cover: $\log {N \choose K}$ bits of information. Since each test carries at most one bit, this is the amount of tests required. Stirling's approximation easily shows that for $K \ll N$, the leading factor of that is \revisedSr{$K\log(N/K)$}.} For example, see the channel-coding analogy given in \cite{atia2012boolean}. A similar analogy to wiretap channels will be at the basis of this work as well. In fact, probabilistic methods with an error probability guarantee appeared in \cite{sebHo1985two}, without explicitly mentioning GT, yet showed the $O(K \log N)$ bound. Additional probabilistic methods can be found in \cite{scarlett2015limits} for \emph{support recovery}, or in \cite{scarlett2016phase}, when an interesting \emph{phase transition} phenomenon was observed, yielding tight results on the threshold (in terms of the number of tests) between the error probability approaching one or vanishing. \off{Finally, note that the significant difference between exact reconstruction and negligible error probability is also apparent in adaptive GT, e.g., \cite{aldridge2012adaptive}.}
\subsubsection{A Channel Coding Interpretation}
As mentioned, the analogy to channel coding \revisedSr{has} proved useful \cite{atia2012boolean}. \cite{baldassini2013capacity} defined the notion of \emph{group testing capacity}, that is, the value of $\lim_{N \to \infty} \frac{\log{{N}\choose{K}}}{T}$ under which reliable algorithms exist, yet, over which, no reliable reconstruction is possible. A converse result for the Bernoulli, non-adaptive case was given in \cite{aldridge2017capacity}. Strong converse results were given in \cite{tan2014strong,johnson2015strong}, again, building on the channel coding analogy, as well as converses for noisy GT \cite{scarlett2016converse}. In \cite{aldridge2012adaptive}, adaptive GT was analyzed as a channel coding \emph{with feedback} problem.
\subsubsection{Efficient Algorithms}
A wide variety of techniques were used to design efficient GT decoders. Results and surveys for early non-adaptive decoding algorithms were given in \cite{chen2008survey,de2005optimal,indyk2010efficiently}. Moreover, although most of the works described above mainly targeted fundamental limits, some give efficient algorithms as well. In the context of this work, it is important to mention the recent COMP \cite{chan2014non}, DD and SCOMP \cite{aldridge2014group} algorithms, concepts from which we will use herein.
\off{
\subsubsection*{Noisy Group Testing\off{ and Graph Models}}
Since GT has numerous applications, several different models were suggested, with diverse tools being applied to tackle them. Perhaps the most studied extension is that of \emph{noisy} GT. Noisy models can include cases of \emph{false positives}, where with some probability a negative pool is still marked as positive, or \emph{false negatives} if defective items get \emph{diluted}. \cite{atia2012boolean} includes several results in this context. Additional converse results for noisy GT are also in \cite{scarlett2016converse}.
Noise level can also depend on the number of items tested in each pool \cite{Goparaju2016polar}. Analysis of noisy GT from a Signal to Noise ratio perspective is given in \cite{malloy2014near}.}
\off{
In \emph{threshold group testing} \cite{chen2009nonadaptive,cheraghchi2010improved,chan2013near}, the test results are sensitive to the number of defective items within a test, hence, are not necessarily a Boolean sum. A recent work also allowed the presence of \emph{inhibitors}, which can mask test results \cite{de2005optimal,ganesan2015non}.
The capacity of Semi-quantitative group testing was discussed in \cite{6283599}, where the tests' output is not necessarily binary (positive or negative). In a similar context, \cite{sparseGT} considered the case where the testing matrix has to be sparse (either because an item cannot be tested too many times, or because a test cannot include too many items). \cite{damaschke2010competitive} considered the case where the number of defectives, $K$, is unknown in advance. \cite{colbourn1999group} considered the case where the items are ordered, and the defective set is a consecutive subset of the items.
Finally, since at the heart of GT stands a binary testing matrix, dictating in which test each item participates, GT also has a clean bipartite graph representation, when the Left set of vertices may stand for the items, and the Right for the tests. Indeed, such a representation gives a very elegant combinatorial result regarding the number of errors a certain testing matrix can withstand \cite{5707017}. Results on \emph{cover-free families} yield lower bounds for GT as well \cite{stinson2000some}. These definitions also lead to a natural definition for \emph{list GT} \cite{cheraghchi2009noise}, where the goal is to output a list containing all defective items, and perhaps additional ones up to the size of the list. Bipartite graphs for GT and noisy GT where also used in \cite{cai2013grotesque,lee2015saffron}. \cite{wadayama2013analysis} considered a more structured graph model, i.e., when the bipartite graph is regular, with items having degree $l$ and tests having degree $r$.}
\revised{\subsection{Secure communication}
\revised{It is very important to note that {\it making GT secure is different from making communication secure, as \revisedSr{remarked} in \Cref{intro}}. Now, we briefly survey the literature in secure communication, since many of the ideas/models/primitives in secure communication will have analogues in secure group-testing.}
\subsubsection{Information-theoretic secrecy}
In a secure communication setting, transmitter Alice wishes to send a message $m$ to receiver Bob. To do so, she is allowed to encode $m$ into a (potentially random) function $x = f(m)$, and transmit $x$ over a medium. It is desired that the eavesdropper Eve should glean no information about $m$ from its (potentially noisy) observation $z$. This information leakage is typically measured via the mutual information between $m$ and $z$.\off{(we defer a more detailed discussion on secrecy metrics to Section II).} The receiver Bob should be able to reconstruct $m$ based on its (also potentially noisy) observation of $x$ (and, potentially, a shared secret that both Bob and Alice know, but Eve is ignorant of).}
\revised{There are a variety of schemes in the literature for information-theoretically secure communications.}\footnote{\revised{Security in general has many connotations — for instance, in the information-theory literature it can also mean a scheme that is resilient to an {\it active adversary}, for instance a communication scheme that is resilient to jamming against a malicious jammer. In this work we focus our attention on {\it passive eavesdropping adversaries}, and aim to ensure secrecy of communications vis-a-vis such adversaries. We shall thus henceforth use the terms security and secrecy interchangeably.}}
\revised{Such schemes typically make one of several assumptions (or combinations of these):}
\begin{itemize}
\item \revised{{\it Shared secrets/Common randomness/Symmetric-key encryption:} The first scheme guaranteed to provide information-theoretic secrecy was by \cite{C1}, who analyzed the secrecy of one-time pad schemes and showed that they ensure perfect secrecy (no leakage of transmitted message). He also provided lower bounds on the size of this shared key. The primary disadvantage of such schemes is that they require a shared key that is essentially as large as the amount of information to be conveyed, and it be continually refreshed for each new communication. These requirements typically make such schemes untenable in practice.}
\item \revised{{\it Wiretap secrecy/Physical-layer secrecy:} Wyner \emph{et al.} \cite{C2,ozarow1984wire} first considered certain communication models in which the communication channel from Alice to Eve is a degraded (noisier) version of the channel from Alice to Bob, and derived the information-theoretic capacity for communication in such settings. These results have been generalized in a variety of directions. See \cite{csiszar2004secrecy,csiszar2008secrecy,C13} for (relatively) recent results. The primary disadvantage of such schemes is that they require that it be possible to instantiate communication channels from Alice to Bob that are better than the communication channel from Alice to Eve. Further, they require that the channel parameters of both channels be relatively well known to Alice and Bob, since the choice of communication rate depends on these parameters. These assumptions make such schemes also untenable in practice, since on one hand Eve may deliberately situate herself to have a relatively clear view of Alice's transmission than Bob, and on the other hand there are often no clear physically-motivated reasons for Alice and Bob to know the channel parameters of the channel to Eve.}
\item \revised{{\it Public discussion/Public feedback:} A very nice result by Maurer (~\cite{maurer1993secret} and subsequent work - see \cite{C13} for details) significantly alleviated at least one of the charges level against physical-layer security systems, that they required the channel to Bob to be \revisedSr{``}better" than the channel to Eve. Maurer demonstrated that feedback (even public feedback that is noiselessly observable by Eve) and multi-round communication schemes can allow for information-theoretically secure communication from Alice to Bob even if the channel from Alice to Bob is worse than the channel from Alice to Eve. Nonetheless, such public discussion schemes {\it still} require some level of knowledge of the channel parameters of the channel to Eve.}
\end{itemize}
\revised{\subsubsection{Cryptographic security}
Due to the shortcomings highlighted above, modern communication systems usually back off from demanding information-theoretic security, and instead attempt to instantiate {\it computational security}. In these settings, instead of demanding small information leakage to arbitrary eavesdroppers, one instead assumes bounds on the computational power of the eavesdropper (for instance, that it cannot computationally efficiently invert \revisedSr{``}one-way functions"). Under such assumptions one is then often able to provide conditional security, for instance with a public-key infrastructure ~\cite{kocher1996timing,stinson2005cryptography}.}
\revised{Such schemes have their own challenges to instantiate. For one, the computational assumptions they rely on are sometimes unfounded and hence sometimes turn out to be prone to attack \cite{tagkey2004391,C13,dent2006fundamental}. For another, the computational burden of implementing cryptographic primitives with strong guarantees can be somewhat high for Alice and Bob~\cite{bernstein2008attacking}.}
\revised{\subsection{Secure Group-Testing}
On the face of it, the connection between secure communication and secure group-testing is perhaps not obvious. We highlight below scenarios that make these connections explicit.
Paralleling the classification of secure communication schemes above, one can also conceive of a corresponding classification of secure GT schemes.}
\subsubsection{\revised{Information-theoretic schemes}}
\begin{itemize}
\item \revised{{\it Shared secrets/Common randomness/Symmetric-key encryption:} A possible scheme to achieve secure group testing, is to utilize a shared key between Alice and Bob. For example, consider a scenario in which Alice the nurse has a large number of blood samples that need to be tested for the presence of a disease. She sends them to a lab named Eve to be tested. To minimize the number of tests done via the lab, she pools blood samples appropriately. However, while the lab itself will perform the tests honestly, it can't be trusted to keep medical records secure, and so Alice keeps secret the identity of the people tested in each pool.
\footnote{Even in this setting, it can be seen that the {\it number} of diseased individuals can still be inferred by Eve. However, this is assumed to be a publicly known/estimable parameter.}}
\revised{Given the test outcomes, doctor Bob now desires to identify the set of diseased people. To be able to reconstruct this mapping, a relatively large amount of information (the mapping between individuals' identities and pools tested) needs to be securely communicated from Alice to Bob. As in the one-time pad secure communication setting, this need for a large amount of common randomness makes such schemes unattractive in practice. Nonetheless, the question is theoretically interesting, \revisedSr{and} some interesting results have been recently reported in this direction by \cite{atallah2008private,goodrich2005indexing,freedman2004efficient,rachlin2008secrecy}.}
\item \revised{{\it Wiretap secrecy/Physical-layer secrecy:} This is the setting of this paper. Alice does not desire to communicate a large shared key to Bob, and still wishes to maintain secrecy of the identities of the diseased people from \revisedSr{``}honest but curious" Eve. Alice therefore does the following two things: (i) For some $\delta \in (0,1)$, she chooses a $1/\delta$ number of independent labs, and divides the T pools to be tested into $1/\delta$ {\it pool sets} of $T\delta$ pools each, and sends each set to a distinct lab. (ii) For each blood pool, she {\it publicly} reveals to {\it all} parties (Bob, Eve, and anyone else who's interested) a set ${\cal S}(t)$ of {\it possible} combinations of individuals whose blood could constitute that specific pool $t$. As to which specific combination from ${\cal S}(t)$ of individuals the pool actually comprises of, only Alice knows {\it a priori} - Alice generates this private randomness by herself, and does not leak it to anyone (perhaps by destroying all trace of it from her records).}
\revised{The twin-fold goal is now for Alice to choose pool-sets and set of ${\cal S}(t)$ for each $t \in [T]$ to ensure that as long as no more than one lab leaks information, there is sufficient randomness in the set of ${\cal S}(t)$ so that essentially no information about the diseased individuals identities leaks, but Bob (who has access to the test reports from all the $1/\delta$ labs) can still accurately estimate (using the publicly available information on ${\cal S}(t)$ for each test $t$) the disease status of each individual.
\revisedSr{This scenario closely parallels the scenario in Wyner's Wiretap channel\off{ \revisedSr{II}}.
Specifically, this corresponds to Alice communicating a sequence of $T$ test outcomes to Bob, whereas Eve can see only a $\delta$ fraction of test outcomes. To ensure secrecy, Alice injects private randomness (corresponding to which set from ${\cal S}(t)$ corresponds to the combination of individuals that was tested in test $t$) into each test - this is the analogue of the \off{coset-}coding schemes often used for Wyner's wiretap channels.}}
\revised{
\begin{remark}
It is a natural theoretical question to consider corresponding generalizations of this scenario with other types of broadcast channels from Alice to Bob/Eve (not just degraded erasure channels), especially since such problems are well-understood in a wiretap security context. However, the physical motivation of such generalizations is not as clear as in the scenario outlined above. So, even though in principle the schemes we present in \Cref{formulation} can be generalized to other broadcast channels, to keep the presentation in this paper clean we do not pursue these generalizations here.
\end{remark}
\begin{remark}
Note that there are other mechanisms via which Alice could use her private randomness. For instance, she could {\it deliberately} contaminate some fraction of the tests she sends to each lab with blood from diseased individuals. Doing so might reduce the amount of private randomness required to instantiate secrecy. While this is an intriguing direction for future work, we do not pursue such ideas here.
\end{remark}
}
\item \revised{{\it Public discussion/Public feedback:} The analogue of a public discussion communication scheme in the secure group-testing context is perhaps a setting in which Alice sends blood pools to labs in {\it multiple rounds}, also known as {\it adaptive group testing} in the GT literature. Bob, on observing the set of test outcomes in round $i$, then publicly broadcasts (to Alice, Eve, and any other interested parties) some (possibly randomized) function of his observations thus far. This has several potential advantages. Firstly, adaptive group-testing schemes (e.g.\cite{aldridge2014group}) significantly outperform the best-known non-adaptive group-testing schemes (in terms of smaller number of tests required to identify diseased individuals) in regimes where $K=\omega(N^{1/3})$}. \revised{One can hope for similar gains here. Secondly, as in secure communication with public discussion, one can hope that multi-round GT schemes would enable information-theoretic secrecy even in situations where Eve may potentially have access to {\it more} test outcomes than Bob. Finally, such schemes may offer storage/computational complexity advantages over non-adaptive GT schemes. Hence this is an ongoing area of research, but outside the scope of this paper.}
\end{itemize}
\revised{\subsubsection{Cryptographic secrecy}
As in the context of secure communication, the use of cryptographic primitives to keep information about the items being tested secure has also been explored in sparse recovery problems - see, for instance \cite{atallah2008private,goodrich2005indexing,freedman2004efficient,rachlin2008secrecy}. Schemes based on cryptographic primitives have similar weaknesses in the secure GT context as they do in the communication context, and we do not explore them here.}
\section{Information Leakage at the Eavesdropper\\ (Strong Secrecy)}\label{strong_secrecy}
We wish to show that $I(W;Z^{T})\rightarrow 0$.
Denote by $\mathcal{C}_T$ the random codebook and by $\textbf{X}_{\mathcal{S}_w}^T$ the set of codewords corresponding to the true defective items.
We assumed $W \in \{1,\ldots, {N \choose K}\}$ is uniformly distributed, that is, there is no \emph{a-priori} bias to any specific subset.
Further, the codebook includes independent and identically distributed codewords. The eavesdropper, observing $Z^T$, wish to decode the true $K$ independent and identically distributed codewords, which correspond to the defective items, one of ${N \choose K}$ subsets.
To analyze the information leakage at the eavesdropper, we note that the channel Eve sees in this case is the following \emph{multiple access channel}. Each of the items can be considered as a \revisedSr{``}user" with $M$ specific codewords.
Eve's goal is to identify the active users.
Note the Eve's channel can be viewed as (binary) Boolean MAC, followed by a BEC$(1-\delta)$.
\off{by decode all the transmitted codewords.
This is possible if the rates at which the users transmit are within the capacity region of this MAC. Indeed, this is a (binary) Boolean MAC channel, followed by a simple erasure channel.}
The sum capacity to Eve cannot be larger than $\delta$.
\revised{In fact, since the codebook is randomly i.i.d.\ distributed, under a fixed input distribution for the testing matrix $(\frac{\ln(2)}{K},1-\frac{\ln(2)}{K})$, Eve can obtain from each active user a rate of at most $\delta/K$}.
Consequently, Eve may obtain a sub-matrix $\tilde{\textbf{Z}}$ of possibly transmitted codewords, where from each codeword Eve sees at most a capacity of $\delta/K$.
However, in this analysis, we help Eve even further to get the true sub-matrix $\tilde{\textbf{Z}}$, with the maximum rate $\delta/K$.
That is, we assume Eve gets the full rate, identifies the codewords of the users except for the erasures. We give Eve the power not to be confused from the existence of other, $N-K$ users which did not transmit eventually.
\off{that she can obtain in the Boolean MAC channel followed by erasure channel with erasure probability of $1-\delta$.}
Providing this information to Eve only makes her stronger and thus a coding scheme that succeeds to keep Eve ignorant, will also succeed against the original Eve.
Now, once Eve obtains the true sub-matrix $\tilde{\textbf{Z}}$, since the codebook and the subsets of the items was generated independently and identically, we will analyze the information that leaks at the eavesdropper from each possibly transmitted codeword in $\tilde{\textbf{Z}}$ separately.
Namely, per original codeword $X_j^T$, on average, $\frac{\delta}{K}T $ from the $T$ outcomes are not erased and are accessible to the eavesdropper via $\tilde{Z}_{\tilde{j}}^{T}$.
Thus, out of the $M$ independent and identically distributed codewords, there is an exponential number of codewords, that from eavesdropper perspective, may have participated in $\textbf{X}_{\mathcal{S}_w}^T$ and could have resulted in the same $\tilde{Z}_{\tilde{j}}^{T}$.
These consistent codewords are exactly those that lie in a ball around $\tilde{Z}_{\tilde{j}}^{T}$ of radius $d \approx (1-\frac{\delta}{K}) T$ as depicted in \Cref{figure:StrongSecrecy}.
\ifdouble
\begin{figure}
\centering
\includegraphics[trim=0cm 0cm 0cm 0cm,clip,scale=1]{StrongSecrecy3.png}
\caption{Codewords exactly lie in a ball around $\tilde{Z}_{\tilde{j}}^{T}$ of radius $d \approx (1-\frac{\delta}{K})T$), where $j$ is the codeword index and $1\leq i \leq N$ is the bin index.}
\label{figure:StrongSecrecy}
\end{figure}
\else
\begin{figure}
\centering
\includegraphics[trim=0cm 0cm 0cm 0cm,clip,scale=1.5]{StrongSecrecy3.png}
\caption{Codewords exactly lie in a ball around $\tilde{Z}_{\tilde{j}}^{T}$ of radius $d \approx (1-\frac{\delta}{K})T$), where $j$ is the codeword index and $1\leq i \leq N$ is the bin index.}
\label{figure:StrongSecrecy}
\end{figure}
\fi
The eavesdropper does not know what is the codeword $X^T_j$ in $\textbf{X}_{\mathcal{S}_w}^T$, which was selected by the mixer and she even does not know what $d_H = \|X^T_j - \tilde{Z}_{\tilde{j}}^{T}\|$ is exactly (where $\|\cdot\|$ is the hamming distance), other than the fact that $d_H \approx (1-\frac{\delta}{K})T$.
However, we help Eve by providing $d$ and by choosing a small yet exponential set of codewords (of size $2^{T\frac{\epsilon}{K}}$, for an appropriately small $\epsilon$) from the codebook $\mathcal{C}$, chosen from the set of all codewords at distance $d$ from $\tilde{Z}_{\tilde{j}}^{T}$, with the additional guarantee that it also contains the true $X^T_j$.
We refer to this set as the oracle-given set $\textbf{X}_{\mathcal{O}racel}^T$.
Again, providing this information to Eve only makes her stronger and thus a coding scheme that succeeds to keep Eve ignorant, will also succeed against the original Eve.
Conditioned on Eve’s view $\tilde{Z}_{\tilde{j}}^{T}$ and $\textbf{X}_{\mathcal{O}racel}^T$, each of the codewords in $\textbf{X}_{\mathcal{O}racel}^T$ is equally likely to have been participated in the pool tests.
Nonetheless, Eve still has a reasonable amount of uncertainty about which of the $2^{T\frac{\epsilon}{K}}$ codewords was actually participated indeed, this is the uncertainty that we leverage in our analysis.
We define, for any $\tilde{Z}_{\tilde{j}}^{T}$ and $d$, a ball and a shell as
\begin{equation*}
\mathcal{V}ol(\tilde{Z}_{\tilde{j}}^{T},d)=\{X^T_j:d_H(X^T_j,\tilde{Z}_{\tilde{j}}^{T})\leq d\},
\end{equation*}
\begin{equation*}
\mathcal{S}h(\tilde{Z}_{\tilde{j}}^{T},d)=\{X^T_j:d_H(X^T_j,\tilde{Z}_{\tilde{j}}^{T})= d\}.
\end{equation*}
Hence, we define the probability for a codeword to fall in shell as
\ifdouble
\begin{multline*}
Pr(X^T_j \in\mathcal{C} \cap X^T_j\in \mathcal{S}h(\tilde{Z}_{\tilde{j}}^{T},d))\\
= \frac{\mathcal{V}ol(\mathcal{S}h(\tilde{Z}_{\tilde{j}}^{T},d))}{2^T} = \frac{2^{\left(1-\frac{\delta}{K}\right)T}}{2^{T}}.
\end{multline*}
\else
\begin{equation*}
Pr(X^T_j \in\mathcal{C} \cap X^T_j\in \mathcal{S}h(\tilde{Z}_{\tilde{j}}^{T},d))
= \frac{\mathcal{V}ol(\mathcal{S}h(\tilde{Z}_{\tilde{j}}^{T},d))}{2^T} = \frac{2^{\left(1-\frac{\delta}{K}\right)T}}{2^{T}}.
\end{equation*}
\fi
For each item we have $M =2^{\left(\frac{\delta+\revised{\revisedSr{\epsilon^{\prime}}}}{K}\right)T}$ codewords. Thus, \emph{on average}, the number of codewords Eve sees on a shell, per defective item is
\ifdouble
\begin{multline*}
|\{w:X^T_j(w)\in \mathcal{S}h(\tilde{Z}_{\tilde{j}}^{T},d)\}|\\
= \frac{2^{\left(\frac{\delta+\revised{\revisedSr{\epsilon^{\prime}}}}{K}\right) T}\cdot 2^{\left(1-\frac{\delta}{K}\right)T}}{2^T} = 2^{T\frac{\revised{\revisedSr{\epsilon^{\prime}}}}{K}}.
\end{multline*}
\else
\begin{equation*}
|\{w:X^T_j(w)\in \mathcal{S}h(\tilde{Z}_{\tilde{j}}^{T},d)\}|
= \frac{2^{\left(\frac{\delta+\revised{\revisedSr{\epsilon^{\prime}}}}{K}\right) T}*2^{\left(1-\frac{\delta}{K}\right)T}}{2^T} = 2^{T\frac{\revised{\revisedSr{\epsilon^{\prime}}}}{K}}.
\end{equation*}
\fi
Hence, we can conclude that, on average, for every item, Eve has quite a few options in a particular shell.
Now that we established that the average number of codewords per item $|\{w:X^T_j(w)\in \mathcal{S}h(\tilde{Z}_{\tilde{j}}^{T},d)\}|$ is $2^{T\frac{\revised{\revisedSr{\epsilon^{\prime}}}}{K}}$, we wish to calculate the probability that the \emph{actual number of options} deviates from the average by more than $\varepsilon$.\off{ is
\begin{equation*}
(1-\varepsilon)2^{T\frac{\revised{\revisedSr{\epsilon^{\prime}}}}{K}}\leq |\{w:X^T_j(w)\in \mathcal{S}h(\tilde{Z}_{\tilde{j}}^{T},d)\}| \leq (1+\varepsilon)2^{T\frac{\revised{\revisedSr{\epsilon^{\prime}}}}{K}},
\end{equation*}
which is the probability that the number of the codewords per item is not in the range.}
We define
\ifdouble
\begin{multline*}
\mathcal{E}_{C_1}(\tilde{Z}_{\tilde{j}}^{T},d):= Pr\{ (1-\varepsilon)2^{T\frac{\revised{\revisedSr{\epsilon^{\prime}}}}{K}}\leq \\
|\{w:X^T_j(w)\in \mathcal{S}h(\tilde{Z}_{\tilde{j}}^{T},d)\}| \leq (1+\varepsilon)2^{T\frac{\revised{\revisedSr{\epsilon^{\prime}}}}{K}}\}
\end{multline*}
\else
\begin{equation*}
\mathcal{E}_{C_1}(\tilde{Z}_{\tilde{j}}^{T},d):= Pr\{ (1-\varepsilon)2^{T\frac{\revised{\revisedSr{\epsilon^{\prime}}}}{K}}\leq
|\{w:X^T_j(w)\in \mathcal{S}h(\tilde{Z}_{\tilde{j}}^{T},d)\}| \leq (1+\varepsilon)2^{T\frac{\revised{\revisedSr{\epsilon^{\prime}}}}{K}}\}.
\end{equation*}
\fi
Let us fix a pair $\tilde{Z}_{\tilde{j}}^{T}$ and $d$ for large enough $T$. By the Chernoff bound, and taking union bound over $\tilde{Z}_{\tilde{j}}^{T}$ and $d$, we have
\begin{equation*}
Pr(\mathcal{E}_{C_1}(\tilde{Z}_{\tilde{j}}^{T},d)) \geq 1- 2^{-\varepsilon^{\prime}2^{T\frac{\revised{\revisedSr{\epsilon^{\prime}}}}{K}}}.
\end{equation*}
Due to the super exponential decay in $T$, even when we take a union bound over all the codewords and all shells, we get that the probability that a codeword will have $|\{w:X^T_j(w)\in \mathcal{S}h(\tilde{Z}_{\tilde{j}}^{T},d)\}|$ options significantly different than $2^{T\frac{\revised{\epsilon^{\prime}}}{K}}$ is very small.
Actually, with high probability, Eve has almost the same number of options per item. Hence, for Eve\off{, per item $|\{w:X^T_j(w)\in \mathcal{S}h(\tilde{Z}_{\tilde{j}}^{T},d)\}|$,} all the codewords are almost equiprobable.\off{ To conclude, for all the defective items, by the chain rule for entropies, since all the codewords in the codebook are independent:
$H(\textbf{X}_{\mathcal{S}_w}^{T}) = \sum_{j \in\mathcal{K}} H(X^T_j(w))$
and $I(W;Z^{T})\rightarrow 0$.}
In other words, Eve distribution on the items converges super-exponentially fast to a uniform one, hence $H(W|Z^T)\rightarrow H(W)$ and we have $I(W;Z^{T})\rightarrow 0$.
\section{Converse (Necessity)}\label{converse}
In this section, we derive the necessity bound on the required number of tests.
Let $\bar{Z}$ denote the random variable corresponding to the tests which are not available to the eavesdropper. Hence, $Y = (Z,\bar{Z})$.
By Fano's inequality, if $P_e\rightarrow 0$, we have
\begin{equation*}
H(W|Y) \leq T\epsilon'_T,
\end{equation*}
where $\epsilon'_T \rightarrow 0$ as $T\rightarrow\infty$. Moreover, the secrecy constraint implies
\begin{equation}\label{eq:Necessity19}
I(W;Z)\leq T\epsilon''_T,
\end{equation}
where $\epsilon''_T\rightarrow 0$ as $T\rightarrow \infty$. Consequently,
\begin{eqnarray}\label{eq:R_s}
\log \binom{N}{K} &=& H(W) \nonumber\\
&=& I(W;Y) + H(W|Y) \nonumber\\
&\stackrel{(a)}{\leq}& I(W;Z,\bar{Z}) + T\epsilon'_T \nonumber\\
&=& I(W;Z) + I(W;\bar{Z}|Z) + T\epsilon'_T \nonumber\\
&\stackrel{(b)}{\leq} & I(W;\bar{Z}|Z) + T\epsilon'_T + T\epsilon''_T \nonumber\\
&\stackrel{(c)}{\leq} & I(\textbf{X}_{\mathcal{S}_w};\bar{Z}|Z) + T(\epsilon'_T + \epsilon''_T) \nonumber\\
&=& H(\bar{Z}|Z) - H(\bar{Z}|\textbf{X}_{\mathcal{S}_w},Z) + T(\epsilon'_T + \epsilon''_T) \nonumber\\
&\stackrel{(d)}{\leq} & H(\bar{Z}) + T(\epsilon'_T + \epsilon''_T)
\nonumber
\end{eqnarray}
where (a) follows from Fano's inequality and since $Y = (Z,\bar{Z})$, (b) follows from \eqref{eq:Necessity19}, (c) follows from the Markov chain $W\rightarrow X_{\mathcal{S}_w}^{T} \rightarrow Y \rightarrow Z$ and (d) is since conditioning reduces entropy.
We now evaluate $H(\bar{Z})$. Denote by $\mathcal{\bar{E}}$ the set of tests which are not available to Eve and by $\bar{E}_\gamma$ the event $\{|\mathcal{\bar{E}}| \leq T(1-\delta)(1+\gamma)\}$ for some $\gamma >0$. We have
\begin{eqnarray*}
H(\bar{Z}) &=& P(\bar{E}_\gamma)H(\bar{Z}| \bar{E}_\gamma) + P(\bar{E}^c_\gamma)H(\bar{Z}| \bar{E}^c_\gamma)
\\
&\leq& T(1-\delta)(1+\gamma) + TP(\bar{E}^c_\gamma)
\\
& \leq& T(1-\delta)(1+\gamma) + T2^{-T(1-\delta)f(\gamma)},
\end{eqnarray*}
where the last inequality follows from the Chernoff bound for i.i.d.\ Bernoulli random variables with parameter $(1-\delta)$ and is true for some $f(\gamma)$ such that $f(\gamma)>0$ for any $\gamma > 0$.
Thus, we have
\[
\log \binom{N}{K} \leq T(1-\delta)(1+\gamma) + T2^{-T(1-\delta)f(\gamma)} +
T(\epsilon'_T + \epsilon''_T).
\]
That is,
\[
T \ge \frac{1-\epsilon_T}{1-\delta}\log \binom{N}{K},
\]
for some $\epsilon_T$ such that $\epsilon_T \to 0$ as $T \to \infty$. This completes the converse proof.
\off{(e) is assuming $L$ denotes the set of test which leak to Eve, (f) is since each test has a binary output and (g) is by taking a ... where we denote $q_e=1-\delta$, as the erasure probability at the eavesdropper per outcome. Furthermore, we define the event
\begin{eqnarray*}
&&\hspace{-0.75cm} \mathcal{E}_{\bar{Z}(t)} := \Bigg\{\left(q_e-\epsilon^{\prime}_{T}\right)T \leq \sum_{t=0}^{T} H(\bar{Z}(t)) \leq \left(q_e+\epsilon^{\prime}_{T}\right)T \Bigg\},
\end{eqnarray*}
as the probability that the actual erasure probability per outcome deviates from the average by more than $\epsilon^{\prime}$.
Thus, by Chernoff bound and taking the union bound over $\sum_{t=0}^{T} H(\bar{Z}(t))$ for large enough $T$, we have,
$P_r(\mathcal{E}_{\bar{Z}(t)})\geq 1-2^{-T\epsilon^{\prime}_{T}}$.
Due to the exponential decay in $T$, even where we take a union bound over all the outcomes the provability of deviates from the average is very small.
Namely, is necessity
\begin{equation*}\label{eq:Necessity8a}
T \ge \frac{1}{(1-\delta)}\log\binom{N}{K},
\end{equation*}
where $\epsilon^{\prime}_{T}\rightarrow 0$, $\varsigma_T\rightarrow 0$ and $\epsilon_T \rightarrow 0$ as $T\rightarrow \infty$.
Noting that this achieves the secrecy and the necessity bound on the number of pool tests provided in \Cref{converse theorem}.}
| 2f82fba56e2343bdc0e0090786e9c344c20fc91a | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Distinct from conventional phase transitions driven by spontaneous symmetry breaking,
Lifshitz phase transition is driven by an abrupt change of the topology
of the Fermi surface~\cite{Lifshitz}. Recently, it has been predicted in three-dimensional (3D)
condensed matter materials that type-I Weyl fermions (i.e., Weyl points)~\cite{volovik,Murakami2007NJP,Wan2011prb,Vafek2014}
can transition to type-II~\cite{Soluyanov2015Nature} (or called structured Weyl points~\cite{Yong2015PRL}) through Lifshitz transition~\cite{Soluyanov2015Nature,Yong2015PRL}.
Although type-I Weyl fermions~\cite{Weyl} were initially predicted in particle physics,
type-II Weyl fermions may not be allowed to exist there due to Lorentz symmetry,
which is absent in condensed matter materials.
Such Lifshitz transition is perceived as the change of a type-I Weyl point's single point Fermi surface
to a type-II's open Fermi surface consisting of particle and hole pockets along
with a touching point~\cite{Yong2015PRL,Soluyanov2015Nature}.
While remarkable experimental and theoretical progress has been reported recently on
type-II Weyl semimetals and their properties~\cite{Yong2015PRL,Soluyanov2015Nature,Bergholtz2015PRL,
Volovik2016,Yang2016,Udagawa2016,Tiwari2016,Beenakker2016,
Vernevig2016M, Chang2016NC, Koepernik2016PRB,
Liang2016, Kaminskyi2016, Xu2016, Deng2016, Chen2016, Zhou2016,
Shi2016,Bruno2016}, it remains a significant
challenge in solid-state materials to observe the characteristic Lifshitz
phase transition between type-I and type-II Weyl points.
Ultracold atomic gases provide an ideal platform to observe the Lifshitz
transition because of their high controllability. And recent experiments on
one-dimensional and two-dimensional (2D) spin-orbit
coupling~\cite{Lin2011Nature, Jing2012PRL, Zwierlen2012PRL, PanJian2012PRL,
Qu2013PRA, Jing2015Nature, Jing2015,Shuai2015,Lev2016,Ketterle2016} in ultracold atomic gases
have further paved the way for discovering novel
topological quantum states \cite{Bloch2012,SpielmanReview,ZollerReview}. Although type-I
and type-II Weyl points were proposed in quasiparticle spectra of spin-orbit
coupled Fermi superfluids~\cite{Gong2011prl,
Sumanta2013PRA,Yong2014PRL,Dong2014arXiv,Yong2015PRL,Xu2015Review}, realization of such
superfluids is still a big challenge with current experimental technology
\cite{Jing2012PRL,Zwierlen2012PRL,Jing2015Nature,Jing2015}. A more feasible
experimental scheme is to realize Weyl points in the single-particle spectra
without the need of low-temperature Fermi superfluids. While some proposals have
been made concerning type-I Weyl points \cite{Anderson2012PRL,
Jiang2012PRA,Ganeshan2015PRB,Tena2015RPL,Law2015}, the scheme to realize
both type-I and type-II Weyl points and their Lifshitz transition in
the single-particle spectra of cold atoms is still lacking and highly desired.
In this paper, we propose a novel 3D model to realize both type-I and type-II Weyl
points in the single-particle spectra of cold atoms in 2D optical lattices.
In this model, we find two Weyl points carrying a Chern number -1 and a four-fold
degenerate point carrying a Chern number 2,
protected by 2D pseudo-time-reversal (2D-PTRS), 2D inversion (2D-IS), and combined rotational
symmetries. By continuous tuning of a Zeeman field, all these degenerate points
can experience the Lifshitz transition from type-I to type-II.
Furthermore, we find a laser-atom coupling configuration to implement the model
based on the current experimental technology that realizes the required spin-orbit
coupling. Due to the controllability of the Zeeman field through continuous
tuning of a convenient experimental knob, this system offers a unique
opportunity to observe the characteristic topological Lifshitz quantum
phase transition between type-I and type-II Weyl points.
\section{Model Hamiltonian}
We start by briefly reviewing the concept of type-I and type-II Weyl points
that are described by an effective Hamiltonian $H_{W}=v_{0}k_{z}+\sum_{\mu
=x,y,z}v_{\mu }k_{\mu }\sigma _{\mu }$ \cite{Yong2015PRL,Soluyanov2015Nature
, where $k_{\mu }$ ($\sigma _{\mu }$) denote momenta (Pauli matrices) and
v_{0},v_{\mu }$ are real parameters. Its energy spectrum is given by $E_{\pm
}(\mathbf{k})=v_{0}k_{z}\pm \sqrt{\sum_{\mu =x,y,z}v_{\mu }^{2}k_{\mu }^{2}}$
with $\pm $ labeling particle and hole bands. When $|v_{0}|<|v_{z}|$, the
energy of the particle (hole) band is positive (negative), except at the
touching point where the energy vanishes. This touching point is dubbed
type-I Weyl point. When $|v_{0}|>|v_{z}|$, in certain regions, the energy
goes negative for the particle band and positive for the hole band, leading
to an open Fermi surface besides a touching point at $E_{\pm }(\mathbf{k}=0)=0
. This structure is dubbed type-II Weyl point.
To realize both type-I and type-II Weyl points with cold atoms, we consider the
following Hamiltonian that describes atoms in 2D optical lattices
\begin{equation}
H^{\prime }=\frac{\mathbf{p}^{2}}{2m}+\sum_{\nu =x,y}V_{\nu }\sin
^{2}(k_{L\nu }r_{\nu })+h_{z}\sigma _{z}+V_{SO},
\label{HC}
\end{equation
where $\mathbf{p}=-i\hbar \nabla $ is the momentum operator, $m$ is the mass
of atoms, $V_{\nu }$ ($\nu =x,y$) denote the strength of optical
lattices with the period being $a_{\nu }=\pi /k_{L\nu }$ along the $\nu $
direction, $h_{z}$ is the Zeeman field, $\sigma _{\nu }$ are Pauli matrices
for spins, and $V_{SO}$ is a laser-induced spin-orbit coupling term taking
the form
\begin{equation}
V_{SO}=\Omega _{SO}(M_{x}+iM_{y})e^{ik_{Lz}r_{z}}|\uparrow \rangle
\left\langle \downarrow \right\vert +\text{H.c}.
\end{equation
with $M_{x}=\sin (k_{Lx}r_{x})\cos (k_{Ly}r_{y})$, $M_{y}=\sin
(k_{Ly}r_{y})\cos (k_{Lx}r_{x})$, $\Omega _{SO}$ proportional to the
laser strength, and $k_{L\nu }$ with $\nu=x,y,z$ determined by lasers'
wave vector along the $\nu$ direction. Later, we will describe the laser configuration that
directly realizes the Hamiltonian (1), in particular the spin-orbit coupling
term $V_{SO}$. Employing a unitary transformation with
$U=e^{-ik_{Lz}r_{z}/2}|\uparrow\rangle\langle\uparrow|+e^{ik_{Lz}r_{z}/2}|\downarrow\rangle\langle\downarrow|$
yields $H=UH^\prime U^{-1}$ with
\begin{equation}
H=\frac{\hbar ^{2}k_{z}^{2}}{2m}+\tilde{h}_{z}\sigma _{z}+H_{2D},
\end{equation
where $k_{z}=p_{z}/\hbar $, $\tilde{h}_{z}=\hbar ^{2}k_{L{z}}k_{z}/(2m)+h_{z}
$, and the 2D\ Hamiltonian $H_{2D}$ in the $(x,y)$ plane is expressed as
\begin{eqnarray}
H_{2D} &=&\sum_{\nu =x,y}\left[ \frac{p_{\nu }^{2}}{2m}+V_{\nu }\sin
^{2}(k_{L\nu }r_{\nu })\right] \notag \\
&+&\left[ \Omega _{SO}(M_{x}+iM_{y})|\uparrow \rangle \left\langle
\downarrow \right\vert +\text{H.c.}\right] .
\end{eqnarray}
\begin{figure}[t]
\includegraphics[width=3.4in]{Fig1.eps}
\caption{(Color online) (a) Lattice structure in the $(x,y)$ plane, where
each unit cell is made up of A and B sites and $\mathbf{a}_1=a\mathbf{e}_x-
\mathbf{e}_y$ with $a=a_x=a_y$ and $\mathbf{a}_2=a\mathbf{e}_x+a\mathbf{e}_y$
are unit vectors. (b) The first brillouin zone denoted by the dashed box.
\mathbf{b}_1=\protect\pi(\mathbf{e}_x-\mathbf{e}_y)/a$ and $\mathbf{b}_2
\protect\pi(\mathbf{e}_x+\mathbf{e}_y)/a$ are reciprocal unit vectors. Green
[${\bf k}^{W \pm}=(k_x^{W \pm}a,k_y^{W \pm}a,k_z^{W \pm}a_z)=(0,0,2m\pi(\pm 4\bar{t}-h_z)/(\hbar^2 k_{Lz}^2)$]
and red solid [${\bf k}^{W0}=(k_x^{W0}a,k_y^{W0}a,k_z^{W0}a_z)=(0,\pi,-2m\pi h_z/(\hbar^2 k_{Lz}^2))$]
circles denote the position where Weyl points exist in the momentum space.
Here, $a_z=\protect\pi/k_{Lz}$. (c)(d) Spectra
near a type-I and type-II four-fold degenerate point with respect to $(k_ya,k_za_z)$ for
k_xa=0$ (see the spectra of the continuous model in Appendix A
). In (d), the gray plane (i.e., $E=E_W$ with $E_W$ being the energy at the
degenerate point) intersects both the particle and hole bands in addition to
the degenerate point, implying it is type-II. Here,
$E_R=\hbar^2k_R^2/2m$ with $k_R$ being the
absolute value of lasers' wave vector. }
\label{Fig1}
\end{figure}
To see how the Weyl points emerge in this model, we discretize $H_{2D}$ and
study its physics in the tight-binding model (see Appendix A
for details of discretization). The tight-binding form of $H$ can be written
as [let us first neglect $\hbar ^{2}k_{z}^{2}/(2m)$ term and focus on type-I
Weyl points]
\begin{eqnarray}
H_{TB}&&=\sum_{k_{z}}\sum_{{\bf x}}[ \tilde{h}_{z}\hat{c}_{k_{z},{\bf x}}^{\dagger }\sigma_z\hat{c}_{k_{z},{\bf x}}
+\sum_{\nu=x,y} (-t_{\nu}\hat{c}_{k_{z},{\bf x}}^{\dagger }\hat{c}_{k_{z},{\bf x}+{\bf g}_\nu}
\nonumber \\
&&+(-1)^{j_x+j_y}t_{SO\nu}\hat{c}_{k_{z},{\bf x}}^{\dagger }\sigma_\nu\hat{c}_{k_{z},{\bf x}+{\bf g}_\nu}+\text{H.c.}) ],
\end{eqnarray}
where ${\bf g}_\nu=a_\nu{\bf e}_\nu$; $\hat{c}^\dagger_{k_{z},{\bf x}}=(
\begin{array}{cc} \hat{c}^\dagger_{k_{z},{\bf x},\uparrow} & \hat{c}^\dagger_{k_{z},{\bf x},\downarrow}
\end{array}
)$ with $\hat{c}_{k_{z},{\bf x},\sigma }^{\dagger }$ ($\hat{c}_{k_{z},{\bf x},\sigma }$) being the
creating (annihilating) operator and ${\bf x}=j_{x}a_x{\bf e}_x+j_{y}a_y{\bf e}_y$;
$t_\nu$ and $t_{SO\nu}$ denote the tunneling and spin-orbit coupling
strength along the $\nu$ direction.
Different from the well-known 2D Chern insulator~\cite{Qi2008,Huges2016},
this Hamiltonian involves the position dependent spin-orbit coupling, and we thus
need to choose a unit cell consisting of two sites: A and B [as shown in
Fig.~\ref{Fig1}(a)] and the Hamiltonian in the momentum space in
the new basis $\Psi (\mathbf{k})^{T}$ with $\Psi (\mathbf{k})=
\begin{array}{cccc}
e^{ik_{x}a_{x}}\hat{A}_{\mathbf{k}\uparrow } & e^{ik_{x}a_{x}}\hat{A}_
\mathbf{k}\downarrow } & \hat{B}_{\mathbf{k}\uparrow } & \hat{B}_{\mathbf{k
\downarrow
\end{array
)$
reads
\begin{equation}
H_{TB}(\mathbf{k})=\tilde{h}_{z}\sigma _{z}-h_{t}\tau _{x}+\tau
_{y}(-d_{x}\sigma _{x}+d_{y}\sigma _{y}), \label{HTBk}
\end{equation
where $h_{t}=2\sum_{\nu =x,y}t_{\nu }\cos (k_{\nu }a_{\nu })$,
d_{x}=2t_{SOx}\sin (k_{x}a_{x})$ and $d_{y}=-2t_{SOy}\sin (k_{y}a_{y})$;
\tau $ are Pauli matrices acting on A, B sublattices. In the $(k_{x},k_{y})$
plane, $H_{TB}(\mathbf{k}_{\perp },k_{z})$ ($\mathbf{k}_{\perp }=k_{x}\mathbf{e
_{x}+k_{y}\mathbf{e}_{y}$) respects 2D-IS: $\tau _{x}H_{TB}(\mathbf{k})\tau _{x}=H_{TB}(
\mathbf{k}_{\perp },k_{z})$ and when $\tilde{h}_{z}=0$, 2D-PTRS:
\mathcal{T}H_{TB}(\mathbf{k})\mathcal{T}^{-1}=H_{TB}(-\mathbf{k}_{\perp },k_{z})$ with
$\mathcal{T}=i\tau _{x}\sigma _{y}\mathcal{K}$ and $\mathcal{K}$ being the
complex conjugate operator. These two symmetries guarantee that the spectrum
in this specific plane ($\tilde{h}_{z}=0$) is at least doubly degenerate,
implying that the touching point, if exists, is four-fold degenerate. We note
that in the continuous model 2D-IS corresponds to $\mathcal{I}_{C}
\mathcal{I}_{C}^{-1}=H(-r_x+\pi /k_{L{x}},-r_y,r_z)=H$ with the
inversion center located at $(\pi /(2k_{L{x}}),0)$ and 2D-PTRS corresponds to $\mathcal
T}_{C}H_{2D}\mathcal{T}_{C}^{-1}=H_{2D}$ with $\mathcal{T}_{C}\equiv
\mathcal{I}_C\mathcal{P}\sigma _{y}\mathcal{K}$ and $\mathcal{P}H_{2D
\mathcal{P}^{-1}=H_{2D}(-r_x,-r_y)$.
Specifically, the eigenvalues of $H_{TB}(\mathbf{k})$ read $E_{\mathbf{k
}=\pm \sqrt{d_{\perp }^{2}+(h_{t}\pm \tilde{h}_{z})^{2}}$ with $d_{\perp
}^{2}=d_{x}^{2}+d_{y}^{2}$, which supports the above symmetry analysis
that the energy band at each $\mathbf{k}$ is doubly degenerate without
\tilde{h}_{z}$. Clearly, when $d_{x}=d_{y}=0$ and $h_{t}\pm \tilde{h}_{z}=0
, there emerge degenerate points. This requires $(k_{x}a_{x},k_{y}a_{y})=(0,\pi )$
or $(0,0)$. In the former case, a single degenerate point appears at
k_{z}^{W0}a_{z}=-2m\pi h_{z}/(\hbar ^{2}k_{L{z}}^{2})$ if $t_{x}=t_{y}$
(thus $h_{t}=0$ at this point) as a result of a combined rotational symmetry
(i.e., $U_{4}HU_{4}^{-1}=H$ where $U_{4}=S_{4}C_{4}$ with $S_{4}=e^{i\pi
\sigma _{z}/4}$ and $C_{4}$ being the four-fold rotational operator along $z$
when $V_{x}=V_{y}$ and $k_{L{x}}=k_{L{y}}$) readily achievable in
experiments; in this plane, both 2D-PTRS and 2D-IS are preserved and
this point's degeneracy is therefore four-fold (also seen from the eigenvalues).
The dispersion is linear along all three momenta directions as visually
shown in Fig.~\ref{Fig1}(c). Compared with a Dirac point in a
Hamiltonian with both TRS and IS in 3D~\cite{Huges2016}, in our case,
two similar symmetries (2D-PTRS and 2D-IS) are both respected only in the plane $\tilde{h}_{z}=0$. In
fact, the four-fold degenerate point can be viewed as consisting of two Weyl points with the same Chern
number. To demonstrate this, let us write down the effective Hamiltonian near the point,
which, after a unitary transformation, reads
\begin{equation}
H(\mathbf{q})\sim (v_{z}q_{z}\sigma _{z}+v_{x}q_{x}\sigma
_{x}+v_{y}q_{y}\sigma _{y})\tau _{0},
\end{equation
where $v_{z}=\hbar ^{2}k_{L{z}}/2m$, $v_{x}=2t_{SOx}a_{x}$, and
v_{y}=-2t_{SOy}a_{y}$; $\tau _{0}$ is a $2\times 2$ identity matrix; the
momenta $\mathbf{q}$ are measured with respect to the degenerate point. It is
clear that each spinor corresponds to a Weyl point with the same chirality~\cite{Yang2015Nano}.
To characterize the topological charge of a degenerate point, we define the first
Chern number
\begin{equation}
C=\frac{1}{2\pi }\sum_{n=1,2}\oint_{\mathcal{S}}{\bm\Omega }_{n}(\bm k)\cdot
d{\bm S}, \label{ChernEq}
\end{equation
where the surface $\mathcal{S}$ encloses a considered degenerate point, and ${\bm\Omega
_{n}(\bm k)=i\langle \nabla _{\bm k}u_{n}(\bm k)|\times |\nabla _{\bm
k}u_{n}(\bm k)\rangle $ is the Berry curvature~\cite{XiaoRMP} for the $n$-th
band with $|u_{n}(\bm k)\rangle $ being its wave function. For our
parameters, a direct calculation yields $C=2$ for the above discussed four-fold degenerate point.
In the latter case where $(k_{x}^{W\pm}a_{x},k_{y}^{W\pm}a_{y})=(0,0)$,
two degenerate points occur at $k_{z}^{W\pm
}a_{z}=2m\pi (\pm 4\bar{t}-h_{z})/(\hbar ^{2}k_{L{z}}^{2})$ with $\bar{t
=(t_{x}+t_{y})/2$. Their degeneracy is double instead of four-fold (owing to the breaking of
2D-PTRS), and their dispersion is also linear in all three momentum directions. Our
calculation demonstrates $C=-1$ for either of them.
Due to the existence of the energy $\hbar ^{2}
k_{z}}^{2}/(2m)$, the energy at these two Weyl points is different except when
h_{z}=0$. Moreover, while we consider single-particle physics, a crude estimate using
mean-field analysis suggests that weak repulsive short-range interactions may
shift the locations of Weyl and four-fold degenerate points along $z$ and cause
the transition between the type-I and type-II but not destroy them (see Appendix B).
To illustrate that these type-I Weyl and four-fold degenerate points can transition to
type-II, we include $\hbar ^{2}k_{z}^{2}/(2m)$ and expand the
Hamiltonian near such points, e.g., $(0,\pi /a_{y},k_{z}^{W0})$,
\begin{equation}
H(\mathbf{q})\sim (v_{0}q_{z}+v_{z}q_{z}\sigma _{z}+v_{x}q_{x}\sigma
_{x}+v_{y}q_{y}\sigma _{y})\tau _{0}
\end{equation
with $v_{0}=-2h_{z}/k_{L{z}}$. Remarkably, when $|h_{z}|>\hbar ^{2}k_{L
z}}^{2}/(4m)$ (i.e., $|v_{0}|>v_{z}$), at certain regions, the energy of
both particle bands goes negative while that of both hole bands goes
positive [as visually displayed in Fig.~\ref{Fig1}(d)], indicating that the
four-fold degenerate point becomes type-II.
Similarly, the type-I Weyl point at ${\bf k}^{W+}$ becomes type-II when
h_{z}<4\bar{t}-\hbar ^{2}k_{L{z}}^{2}/(4m)$ or $h_{z}>4\bar{t}+\hbar ^{2}k_{
{z}}^{2}/(4m)$, and the point at ${\bf k}^{W-}$ becomes type-II when
h_{z}>-4\bar{t}+\hbar ^{2}k_{L{z}}^{2}/(4m)$ or $h_{z}<-4\bar{t}-\hbar
^{2}k_{L{z}}^{2}/(4m)$. In Fig.~\ref{Fig2}(a), we map out the phase diagram
displaying the following phases: all degenerate points are type-I,
all are type-II, and partial type-I and partial type-II. In cold atoms,
$h_z$ can be easily tuned by changing two-photon detuning to
continuously drive the transition among different phases, thereby providing an
ideal platform to directly observe the Lifshitz transition.
\begin{figure}[t]
\includegraphics[width=3.4in]{Fig2.eps}
\caption{(Color online) (a) Phase diagram with respect to $h_z$ and $t$ ($t=t_x=t_y$),
in which $W_\beta {\text{I}}$ ($W_\beta \text{II}$) with $\beta=0,+,-$ represent
the phase with a type-I (type-II) Weyl or four-fold degenerate point located at ${\bf k}^{W\beta}$.
The dashed line shows the case for $t=0.058E_R$.
(a)(b) Spectra as a function of $k_za_z$ for $k_y=0$ under
open boundary condition along the $x$ direction. In (a), $h_z=0$ and the
Weyl points are all type-I; in (b), $h_z=0.4E_R$ and a Weyl point on the
left side is type-II and all others are type-I. The black lines denote
surface states.}
\label{Fig2}
\end{figure}
We now turn to the study of surface states in this system.
In Fig.~\ref{Fig2}, the energy spectra with respect to $k_za_z$
for $k_y=0$ are plotted: (a) $h_z=0$ with type-I degenerate points
and (b) $h_z=0.4E_R$ with a type-II point.
It shows that in both cases there emerge
surface states (called Fermi arc) connecting the four-fold degenerate point
at the center with two other Weyl points on two sides. The spectra in Fig.~\ref{Fig2}(b) also
illustrate the feature of the type-II Weyl point at
k_za_z/\pi=-0.62$ that all the particle and hole bands near the point
at each $k_z$ are positive or negative with respect to $E_{W-}$, the energy
at the point.
Apart from the model that we have discussed, if we choose a simplified scheme
with $M_{x}=\sin (k_{L{x}}r_{x})e^{ik_{L{y}}r_{y}}$ and $M_{y}=\sin (k_{L{y
}r_{y})e^{-ik_{L{x}}r_{x}}$, we can still obtain both type-I and type-II
Weyl points. However, while it still respects 2D-PTRS, it breaks 2D-IS, splitting
the four-fold degenerate point into two doubly degenerate ones (see
Appendix A). To satisfy these symmetry requirements, we need to
add an additional term into the model (\ref{HTBk}) : $\tau _{z}(\alpha
_{1}\sigma _{y}-\alpha _{2}\sigma _{x})$ which respects 2D-PTRS but breaks 2D-IS.
If $V_{x}=V_{y}$ and $k_{L{x}}=k_{L{y}}$, we have $\alpha _{1}=\alpha
_{2}=\alpha $ due to a symmetry requirement (see Appendix A),
and the spectrum is $E_{\mathbf{k}}=\pm \sqrt{(d_{x}+\lambda \alpha
)^{2}+(d_{y}-\lambda \alpha )^{2}+(h_{t}-\lambda \tilde{h}_{z})^{2}}$ with
\lambda =\pm 1$, and Weyl points occur at
[k_{x}^Wa,k_{y}^Wa,k_{z}^Wa_{z}]=[\lambda \delta \theta ,\pi +\lambda \delta
\theta ,-2m\pi h_{z}/(\hbar ^{2}k_{L{z}}^{2})]$ (or $[\lambda \delta \theta
,-\lambda \delta \theta ,2m\pi (4\lambda {\bar{t}}\cos \delta \theta
-h_{z})/(\hbar ^{2}k_{L{z}}^{2})]$) with $\delta \theta =-\sin ^{-1}(\alpha
/2t_{SOx})$. Moreover, when $|h_{z}|>\hbar ^{2}k_{L{z}}^{2}/(4m) $ (
|h_{z}-4\lambda {\bar{t}}\cos \delta \theta |>\hbar ^{2}k_{L{z}}^{2}/(4m)$),
the former (latter) Weyl points become type-II.
\section{Realization of type-II Weyl points}
To realize the Hamiltonian~(\ref{HC}), we propose an experimental scheme (as
shown in Fig.~\ref{Fig3}) that is based on a modification of the
experimental configuration in Ref.~\cite{Shuai2015}. We consider two
hyperfine states of alkali atoms such as $^{40}$K and $^{87}$Rb and employ
two independent Raman processes to create the spin-dependent optical
lattices. Each Raman process involves two linearly polarized blue-detuned
Raman laser beams with the polarization being along $y$ (parallel to the
magnetic field) and $x$, respectively. Each pair of Raman laser beams is
characterized by a pair of Rabi frequencies $[\Omega _{1}=\Omega _{10}\sin
(k_{Lx}r_{x})e^{-ik_{L{z}}r_{z}/2},\Omega _{2}=\Omega _{20}\cos (k_{L{y
}r_{y})e^{ik_{L{z}}r_{z}/2}]$ and $[\Omega _{1}^{\prime }=\Omega _{10}\sin
(k_{L{y}}r_{y})e^{-ik_{L{z}}r_{z}/2},\Omega _{2}^{\prime }=i\Omega _{20}\cos
(k_{L{x}}r_{x})e^{ik_{L{z}}r_{z}/2}]$, respectively. They form a standing
wave along $x$ or $y$ but remain a plane wave along $z$. Since the laser
beam $\Omega _{1}^{\prime }$ ($\Omega _{2}^{\prime }$) is obtained by
reflecting the beam $\Omega _{1}$ ($\Omega _{2}$) by mirrors, they possess
the same frequency $\omega _{1}$ ($\omega_2$), and no phase locking is
required~\cite{Xiongjun2014PRL,Yong2016PRA}. The laser beam $\Omega _{2}$
with a different frequency $\omega _{2}$ is generated by applying an
acoustic-optical modulator on a beam split from the first laser beam. Each
pair of Raman laser beams couple two hyperfine states independently, leading
to the spin-orbit coupling term $V_{SO}$ with $\Omega _{SO}=\Omega
_{10}^*\Omega _{20}/\Delta _{e}$. Moreover, owing to the stark effects,
these lasers also create optical lattices along the $x$ and $y$ directions:
V_{x}\sin ^{2}(k_{L{x}}r_{x})$ and $V_{y}\sin ^{2}(k_{L{y}}r_{y})$ with
V_{x}=V_{y}=2(|\Omega _{10}|^{2}-|\Omega _{20}|^{2})/\Delta _{e}$. The
Zeeman field $h_{z}$ is generated by the two-photon detuning $\delta $
through $h_{z}=\delta /2$ as shown in Fig. \ref{Fig3}. In comparison, a
similar laser configuration can also realize the simplified spin-orbit
coupling scheme that we mentioned before. In this case, we further simplify
the laser configuration [as shown in Fig.~\ref{Fig3}(c)] so that the Rabi
frequencies take the form $\Omega _{2}=\Omega _{20}e^{ik_{L{z}}r_{z}/2+ik_{L
y}}r_{y}}$ and $\Omega _{2}^{\prime }=i\Omega _{20}e^{ik{L{z}}r_{z}/2-ik_{L{
}}r_{x}}$, corresponding to plane waves for the second set of laser beams.
\begin{figure}[t]
\includegraphics[width=3.4in]{Fig3.eps}
\caption{(Color online) Schematics of laser configurations to realize the
Hamiltonian (\protect\ref{HC}) [(b)] and its simplified one [(c)]. $\Omega
_{1}$ ($\Omega _{2}$) possess the same frequency as $\Omega _{1}^\prime$ (
\Omega _{2}^\prime$) generated by reflecting the laser beam
\Omega_{1}$ ($\Omega_{2}$) by mirrors. The magnetic field is along the $y$
direction, and $\protect\delta$ is the two-photon detuning. The double
arrows denote the orientation of linear polarization of laser beams.}
\label{Fig3}
\end{figure}
We may choose either $^{40}$K (fermions) or $^{87}$Rb (bosons) atoms for
observation of Weyl points in experiments. Here we take $^{40}$K as an
example. With a blue-detuned laser beam at wavelength $764$ nm
(corresponding to $\Delta _{e}=2\pi \times 1.38$ THz)~\cite{Jing2012PRL},
the recoil energy $E_{R}/\hbar =2\pi \times 8.5$ kHz. If we choose a
geometry with $k_{L{x}}=k_{L{y}}=k_{R}\sin \theta $ and $k_{L{z}}=2k_{R}\cos
\theta $ with $\theta =60^{\circ }$ (the angle between laser
beams and $z$ axis), $\Omega _{10}=2\pi \times 0.15$ GHz,
\Omega _{20}=\Omega _{10}/3$, we have $V_{x}=V_{y}=3.7E_{R}$ and $\Omega
_{SO}=0.7E_{R}$. With these experimental parameters, the tight-binding
parameters are $t_{x}=t_{y}=0.058E_{R}$, and $t_{SOx}=-t_{SOy}=0.028E_{R}$.
\delta $ can be readily tuned from zero and when $\delta $ crosses $0.53E_{R}
$, we will observe a Lifshitz-type quantum phase transition from type-I to
type-II Weyl points.
To detect the Weyl points of fermionic atoms and their Lifshitz phase
transition, one can measure their linear spectra along all three momenta
directions by momentum resolved radio-frequency spectroscopy, which has been
utilized for observation of a 2D Dirac cone in spin-orbit-coupled atomic
gases~\cite{Jing2015Nature,Jing2015}. The Lifshitz transition is directly
reflected by a sharp change of the dispersion of particle or hole bands
along $k_z$ near a Weyl point so that the slopes of their spectra have the
same sign. For bosonic atoms such as $^{87}$Rb, although the Fermi surface
does not exist, there is still the band structure with a touching point. One
may consider driving a BEC across a Weyl point by a constant force $F$ and
measuring the Landau-Zener tunneling probability~\cit
{Esslinger2012Nature,Lim2012PRL,Law2015,Xiongjun2016}, which is
P_{LZ}=e^{-\pi E_g^2/(4vF)}$ with $E_g$ being the gap between the considered
particle and hole bands and $v$ being the velocity of the BEC~\cite{LandZene
. Therefore, the gap closing at a Weyl point is signalled by a peak of the
Landau-Zener tunneling probability. When the BEC bypasses a type-I (type-II)
Weyl point, a finite fraction of atoms remains in the hole band and these
atoms move in the opposite (same) direction along $z$ compared with those
tunneling into a higher band. This different behavior can be utilized to
measure the Lifshitz transition.
In summary, we have proposed a scheme well based on the current experimental
technology to realize both type-I and type-II Weyl points in the
single-particle spectra with cold atoms in an optical lattice. The proposed
system offers a unique opportunity to observe and study the topological
Lifshitz-type quantum phase transition from type-I to type-II Weyl points by
continuously tuning one of the experimental knobs.
\begin{acknowledgments}
We thank S.-T. Wang and S. A. Yang for helpful discussions. This work was supported by the
ARL, the IARPA LogiQ program, and the AFOSR MURI program.
\end{acknowledgments}
\section*{Appendix A: DERIVATION OF TIGHT-BINDING MODEL}
\setcounter{equation}{0} \setcounter{table}{0}
\renewcommand{\theequation}{A\arabic{equation}}
In this appendix, we derive the tight-binding model from the continuous model
$H$ in Eq. (3) in the main text and compare their spectra to verity the tight-binding
model's reliability.
Let us first focus on the discretization of $H_{2D}$, which can
be written as in the second quantization language
\begin{equation}
H_{II}=\int d\mathbf{r}\hat{\psi}^{\dagger }(\mathbf{r})H_{2D}\hat{\psi}(\mathbf{
}),
\end{equation
where $\bf r$ is restricted to the $(x,y)$ plane and
\hat{\psi}(\mathbf{r})=
\begin{array}{cc}
\hat{\psi}_{\uparrow }(\mathbf{r}) & \hat{\psi}_{\downarrow }(\mathbf{r}
\end{array
]^{T}$ with $\hat{\psi}_{\sigma }(\mathbf{r})$ [$\hat{\psi}_{\sigma
}^{\dagger }(\mathbf{r})$] annihilating (creating) an atom with spin $\sigma $
($\sigma =\uparrow ,\downarrow $) located at $\mathbf{r}$. They satisfy the
anti-commutation or commutation relation $[\hat{\psi}_{\sigma }(\mathbf{r})
\hat{\psi}_{\sigma ^{\prime }}^{\dagger }(\mathbf{r}^{\prime })]_{\pm
}=\delta _{\sigma \sigma ^{\prime }}\delta (\mathbf{r}-\mathbf{r}^{\prime })$
for fermionic atoms ($+$) or bosonic atoms ($-$), respectively. We expand the field
operator using local Wannier functions
\begin{equation}
\hat{\psi}_{\sigma }(\mathbf{r})=\sum_{n,j_x,j_y,\sigma }W_{n,j_x,j_y }\hat{c
_{n,j_x,j_y,\sigma },
\end{equation
where $W_{n,j_x,j_y}$ is the Wannier function for $\Omega_{SO}=0$ located at the site $(j_x,j_y)$ for
the $n$-th band, and $\hat{c}_{n,j_x,j_y,\sigma }$ ($\hat{c}_{n,j_x,j_y,\sigma }^\dagger$)
annihilates (creates) an atom located at the state $W_{n,j_x,j_y}$ with spin $\sigma $.
Let us focus on the physics in the lowest band and thus assume $n=1$, thereby
simplifying the above expression
\begin{equation}
\hat{\psi}_{\sigma }(\mathbf{r})\approx \sum_{j_x,j_y}W_{j_x,j_y}\hat{c}_{j_x,j_y,\sigma },
\end{equation
where $W_{j}=W_{j_{x}}^{x}(r_{x})W_{j_{y}}^{y}(r_{y})$
with $W_{j_{\nu }}^{\nu }(r_{\nu })=W^{\nu }(r_{\nu }-j_{\nu }a_{\nu })$
being the Wannier function along $\nu $. We note that although this is
an approximation, it proves to be qualitatively correct and we will verify it by comparing the spectra obtained by
solving the continuous model and the tight-binding one. Using this expansion, we obtain
the following tight-binding model without $V_{SO}$
\begin{eqnarray}
H_{t}=&&-\sum_{j_x,j_y,\sigma}\left(t_{x }\hat{c}_{j_x,j_y,\sigma
}^{\dagger }\hat{c}_{j_{x}+1,j_y,\sigma }
+t_{y}\hat{c}_{j_x,j_y,\sigma
}^{\dagger }\hat{c}_{j_x,j_{y}+1,\sigma } \right) \nonumber \\
&&+\text{H.c.},
\end{eqnarray
where $t_\nu$ with $\nu=x,y$ denote the hopping amplitudes defined as
\begin{equation}
t_{\nu } =-\int dr_{\nu }W_{j_\nu}^{\nu}\left[ \frac{p_{\nu }^{2}}{2m}+V_{\nu }\sin
^{2}(k_{L\nu }r_{\nu })\right] W_{j_{\nu }+1}^\nu. \label{ET}
\end{equation
We approximately derive the tight-binding term contributed by the spin-dependent
lattice as follows
\begin{widetext}
\begin{eqnarray}
H_{SO} &=&\Omega _{SO}\int d\mathbf{r}\hat{\psi}_{\uparrow }^{\dagger }
\mathbf{r})\left(M_x+iM_y\right)\hat{\psi}_{\downarrow }(\mathbf{r})+\text{H.c.}, \nonumber \\
&\approx &\Omega _{SO}\sum_{j_x,j_y}\sum_{j_x^\prime,j_y^\prime}\Big [\hat{c}_{j_x,j_y,\uparrow }^{\dagger
\hat{c}_{j^\prime_x,j^\prime_y,\downarrow }t_{SOx}^{(j_x,j_y),(j^\prime_x,j^\prime_y)}
-i
\hat{c}_{j_x,j_y,\uparrow }^{\dagger
\hat{c}_{j^\prime_x,j^\prime_y,\downarrow }t_{SOy}^{(j_x,j_y),(j^\prime_x,j^\prime_y)}
\Big ]+\text{H.c.},
\end{eqnarray
where
\begin{eqnarray}
&&t_{SOx}^{(j_x,j_y),(j^\prime_x,j^\prime_y)}=\int d\mathbf{r}W_{j_x,j_y}\sin(k_{Lx}r_{x})\cos(k_{Ly}r_{y})W_{j^{\prime }_x,j^\prime_y}, \\
&&t_{SOy}^{(j_x,j_y),(j^\prime_x,j^\prime_y)}=-\int d\mathbf{r}W_{j_x,j_y}\cos(k_{Lx}r_{x})\sin(k_{Ly}r_{y})W_{j^{\prime }_x,j^\prime_y}.
\end{eqnarray
Note that we have added a minus sign in the definition of $t_{SOy}$ in order to write the
Hamiltonian in a compact form.
\begin{figure*}[t]
\includegraphics[width=5.2in]{SupFig1.eps}
\caption{(Color online) Single-particle spectra in the ($k_y,k_z$) plane for $k_x=0$ obtained
by ab initio theory (i.e., diagonalizing the continuous model) in (a1) and (b1) [the spectra
of tight-binding model are plotted in Fig. 1(c) and (d) in the main text].
To compare these spectra in detail, we further plot them (with solid black and dashed cyan
lines denoting the spectra of continuous and tight-binding model, respectively) around a degenerate
point [$(k_y a/\pi=1,k_z=0)$ for the first row panel and $(k_y a/\pi=1,k_z a_z/\pi=-0.7)$
for the second row panel] as a function of $k_z$ and $k_y$ in (a2,b2) and (a3,b3), respectively.
Here, we choose $V_x=V_y=3.7E_{R}$ and $\Omega_{SO}=0.7E_{R}$ corresponding to
the tight-binding model with $t_x=t_y=0.058E_{R_x}$ and $t_{SO}=0.028E_{R}$.
For (a1-a3), $h_z=0$, while for (b1-b3), $h_z=0.7E_R$. For the
spectra of the continuous model, we have shifted the energy at the degenerate
point to the energy at the same point of the spectra of the tight-binding model.}
\label{SIfig1}
\end{figure*}
\begin{figure*}[t]
\includegraphics[width=7in]{SupFig2.eps}
\caption{(Color online)
Single-particle spectra in the ($k_x,k_y$) plane for $k_z=0$, obtained
by ab initio theory in (a1) and (b1) and
by diagonalizing the tight-binding model in (a2) and (b2).
(a1-a4) correspond to the model
with $M_x=\sin(k_{Lx}r_{x})\cos(k_{Ly}r_{y})$,
$M_y=\sin(k_{Ly}r_{y})\cos (k_{Lx}r_{x})$, and (b1-b4)
to the simplified model with $M_x=\sin(k_{Lx}r_x)e^{ik_{Ly}r_y}$
and $M_y=\sin(k_{Ly}r_y)e^{-ik_{Lx}r_x}$.
To compare these spectra in detail, we further plot them (with solid black and dashed cyan
lines denoting the spectra of continuous and tight-binding model, respectively) around a degenerate
point [$(k_x=0,k_y a/\pi=1)$ for the first row panel and $(k_x a/\pi=-0.024,k_y a/\pi=0.976)$ denoted by the red square
in (b1) and (b2) for the second row panel] as a function of $k_x$ and $k_y$ in (a3,b3) and (a4,b4), respectively.
Here, we choose $V_x=V_y=3.7E_{R}$, $\Omega_{SO}=0.7E_{R}$, and $h_z=0$, corresponding to the tight-binding model
with $t_x=t_y=0.058E_{R_x}$ and $t_{SO}=0.028E_{R}$. For comparison, we have shifted the
spectra of the continuous model so that the energy at the degenerate point is zero.}
\label{SIfig2}
\end{figure*}
Employing the condition $W_{0}^{\nu }(r_{\nu })=W_{0}^{\nu }(-r_{\nu })$
given that one of the optical wells is located at $\mathbf{r}=(0,0)$ when $\Omega_{SO}=0$, we get
\begin{eqnarray}
&&t_{SOx}^{(j_x,j_y),(j_x,j_y^\prime)}=t_{SOy}^{(j_x,j_y),(j_x^\prime,j_y)}=0, \\
&&t_{SOx}^{(j_x,j_y),(j_x+1,j_y)}=-t_{SOx}^{(j_x,j_y),(j_x-1,j_y)}=(-1)^{j_x+j_y}t_{SOx}^{(0,0),(1,0)}, \\
&&t_{SOy}^{(j_x,j_y),(j_x,j_y+1)}=-t_{SOy}^{(j_x,j_y),(j_x,j_y-1)}=(-1)^{j_x+j_y}t_{SOy}^{(0,0),(0,1)},
\end{eqnarray
where the last two relations are obtained because $\sin(k_{L\nu}(r_{\nu}+a_\nu))=-\sin(k_{L\nu}r_{\nu})$ and $\cos(k_{L\nu}(r_{\nu}+a_\nu))=-\cos(k_{L\nu}r_{\nu})$. Therefore, if we only consider the nearest-neighbor
hopping, we obtain the following spin-orbit coupling term of the tight-binding model:
\begin{eqnarray}
H_{SO} &\approx &\Omega _{SO}\sum_{j_x,j_y}\left(\hat{c}_{j_x,j_y,\uparrow }^{\dagger
\hat{c}_{j_x+1,j_y,\downarrow }t_{SOx}^{(j_x,j_y),(j_x+1,j_y)}
+\hat{c}_{j_x,j_y,\uparrow }^{\dagger
\hat{c}_{j_x-1,j_y,\downarrow }t_{SOx}^{(j_x,j_y),(j_x-1,j_y)} \right. \nonumber \\
&&\left.-i\hat{c}_{j_x,j_y,\uparrow }^{\dagger
\hat{c}_{j_x,j_y+1,\downarrow }t_{SOy}^{(j_x,j_y),(j_x,j_y+1)}
-i\hat{c}_{j_x,j_y,\uparrow }^{\dagger
\hat{c}_{j_x,j_y-1,\downarrow }t_{SOy}^{(j_x,j_y),(j_x,j_y-1)}
\right)+\text{H.c.},\\
&=&\Omega _{SO}\sum_{j_x,j_y}(-1)^{j_x+j_y}\left[t_{SOx}^{(0,0),(1,0)}(\hat{c}_{j_x,j_y,\uparrow }^{\dagger
\hat{c}_{j_x+1,j_y,\downarrow }
-\hat{c}_{j_x,j_y,\uparrow }^{\dagger
\hat{c}_{j_x-1,j_y,\downarrow })\right. \nonumber \\
&&\left.-it_{SOy}^{(0,0),(0,1)}(\hat{c}_{j_x,j_y,\uparrow }^{\dagger
\hat{c}_{j_x,j_y+1,\downarrow }
-\hat{c}_{j_x,j_y,\uparrow }^{\dagger
\hat{c}_{j_x,j_y-1,\downarrow }
)\right]+\text{H.c.},\\
&=&\sum_{\bf x}\sum_{\nu=x,y}(-1)^{j_x+j_y}t_{SO\nu}\hat{c}_{{\bf x}}^{\dagger }\sigma_\nu\hat{c}_{{\bf x}+{\bf g}_\nu}+\text{H.c.},
\end{eqnarray}
where in the last step we have recast the Hamiltonian into a compact form by defining
$t_{SO\nu}=\Omega_{SO}t_{SO\nu}^{(0,0),(0,1)}$, ${\bf g}_\nu=a_\nu{\bf e}_\nu$, $\hat{c}^\dagger_{{\bf x}}=(
\begin{array}{cc} \hat{c}^\dagger_{{\bf x},\uparrow} & \hat{c}^\dagger_{{\bf x},\downarrow}
\end{array})$ with $\hat{c}^\dagger_{{\bf x},\sigma}\equiv\hat{c}^\dagger_{j_x,j_y,\sigma}$ and
${\bf x}=j_{x}a_x{\bf e}_x+j_{y}a_y{\bf e}_y$. When $V_x=V_y$ and $k_{Lx}=k_{Ly}$, $t_{SOx}=-t_{SOy}$.
In 3D, after replacing $\hat{c}_{{\bf x},\sigma}$ with $\hat{c}_{k_z,{\bf x},\sigma}$ and
including $H_t$ and the Zeeman field term, we obtain $H_{TB}$ in Eq.(5) in the main text.
\end{widetext}
To verify the reliability of the tight-binding model, in Fig.~\ref{SIfig1} and Fig.~\ref{SIfig2},
we compare the energy spectra of the tight-binding model with those
of the continuous model, which are numerically calculated using Fourier
series expansion of a Bloch function. In Fig.~\ref{SIfig1} (a1) and (b1),
we present the spectra of the continuous model, which qualitatively agree with
their tight-binding counterparts in Fig. 1(c) and (d) of the main text. To see their
difference more quantitatively, we plot
their spectra around a degenerate point with respect to $k_z$ and $k_y$
in Fig.~\ref{SIfig1}(a2,b2) and (a3,b3), respectively, illustrating excellent agreement between
these two approaches in the vicinity of a degenerate point; this implies that the tight-binding
model can well characterize these degenerate points. As momenta move far away from
these points, there appears a slight discrepancy, which might be reduced when
the next-nearest-neighbor hopping is included. In Fig.~\ref{SIfig2}(a1-a2),
we further compare their spectra in the $(k_x,k_y)$ plane; the doubly degenerate spectra
in Fig.~\ref{SIfig2}(a) and those in Fig.~\ref{SIfig2}(b) are in qualitative
agreement with each other and both exhibit a four-fold degenerate point. In addition,
a more quantitative comparison in Fig.~\ref{SIfig2}(a3) and (a4) demonstrates the reliability
of the tight-binding model.
For the
simplified model with $M_x=\sin(k_{Lx}r_x)e^{ik_{Ly}r_y}$
and $M_y=\sin(k_{Ly}r_y)e^{-ik_{Lx}r_x}$,
it breaks 2D-IS but still respects 2D-PTRS and therefore in the tight-binding model
we need to add an additional term $\tau_z(\alpha_1\sigma_y-\alpha_2\sigma_x)$ that
preserves 2D-PTRS but lacks 2D-IS. When $V_x=V_y$ and $k_{Lx}=k_{Ly}$,
we have $\alpha_1=\alpha_2=\alpha$ because the 2D continuous Hamiltonian
respects a $\Pi_C$ symmetry, i.e., $\Pi_CH_{2D}\Pi_C^{-1}=H_{2D}$, where $\Pi_C=M\sigma_xS_4^{-1}$ with $MH_{2D}(x,y)M^{-1}=H_{2D}(y,x)$;
the representation of this symmetry in the momentum space corresponds to $\Pi H_{TB}^{2D}({\bf k})\Pi^{-1}=H_{TB}^{2D}(k_y,k_x)$
[$H_{TB}^{2D}\equiv H_{TB}(\tilde{h}_z=0)$] with $\Pi=(\sigma_x-\sigma_y)/\sqrt{2}$, indicating that the additional term must take the form of $\sigma_x-\sigma_y$. In Fig.~\ref{SIfig2}(b1-b2), we plot the spectra of the continuous and tight-binding models and both figures illustrate
that a four-fold degenerate touching point splits into two doubly degenerate ones. Our further comparison around a touching point
in Fig.~\ref{SIfig2}(b3-b4) shows the quantitative agreement of the latter with the former.
\section*{Appendix B: ANALYSIS OF MANY-BODY EFFECTS}
In this section, we make a crude estimate of many-body effects on the degenerate points in the presence of
weak repulsive atom-atom interactions for fermionic atoms. For alkali atoms, the interaction readily tuned
by Feshbach resonances is short-range and can be written
as (we only consider dominant on-site interactions)
\begin{equation}
H_{Int} =g\sum_{k_{z}k_{z}^{\prime}Q}\sum_{{\bf x}}\hat{c}_{k_{z}{\bf x}\uparrow}^{\dagger}\hat{c}_{-k_{z}+Q{\bf x}\downarrow}^{\dagger}\hat{c}_{-k_{z}^{\prime}+Q{\bf x}\downarrow}\hat{c}_{k_{z}^{\prime}{\bf x}\uparrow},
\end{equation}
where $g$ denotes the strength of interactions proportional to the $s$-wave scattering length. For weak
interactions, using mean-field approximations yields
\begin{eqnarray}
H_{Int}\approx &&\sum_{{\bf k}}{\Psi}({\bf k})^{\dagger}H_{Int}^M({\bf k}){\Psi}({\bf k})+N\sum_{D=A,B}\frac{|m_{\parallel,D}|^{2}}{g} \nonumber \\
&&-N\sum_{D=A,B}\frac{m_{z,\uparrow,D}m_{z,\downarrow,D}}{g},
\end{eqnarray}
where $N$ is the number of sites in the $(x,y)$ plane,
\begin{eqnarray}
m_{z,\sigma,D} &=&g\sum_{k_{z}}\langle\hat{D}_{k_{z}{\bf x}\sigma}^{\dagger}\hat{D}_{k_{z}{\bf x}\sigma}\rangle, \\
m_{\parallel,D} &=&m_{x,D}+im_{y,D} \nonumber \\
&=&g\sum_{k_{z}}\langle\hat{D}_{k_{z}{\bf x},\uparrow}^{\dagger}\hat{D}_{k_{z}{\bf x},\downarrow}\rangle
\end{eqnarray}
and
\begin{eqnarray}
H_{Int}^{M}({\bf k})=&&m_{z,1}+m_{z,2}\sigma_{z}+m_{x,1}\sigma_{x}+m_{y,1}\sigma_{y}+m_{z,3}\tau_{z} \nonumber \\
&&+m_{z,4}\sigma_{z}\tau_{z}+m_{x,2}\sigma_{x}\tau_{z}+m_{y,2}\sigma_{y}\tau_{z},
\end{eqnarray}
with
\begin{eqnarray}
m_{z,1} =\frac{1}{4}(m_{z,\downarrow,A}+m_{z,\uparrow,A}+m_{z,\downarrow,B}+m_{z,\uparrow,B}), \\
m_{z,2} =\frac{1}{4}(m_{z,\downarrow,A}-m_{z,\uparrow,A}+m_{z,\downarrow,B}-m_{z,\uparrow,B}), \\
m_{z,3} =\frac{1}{4}(m_{z,\downarrow,A}+m_{z,\uparrow,A}-m_{z,\downarrow,B}-m_{z,\uparrow,B}), \\
m_{z,4} =\frac{1}{4}(m_{z,\downarrow,A}-m_{z,\uparrow,A}-m_{z,\downarrow,B}+m_{z,\uparrow,B}),
\end{eqnarray}
and
\begin{eqnarray}
m_{x,1} =-\frac{1}{2}(m_{x,A}+m_{x,B}), \\
m_{x,2} =-\frac{1}{2}(m_{x,A}-m_{x,B}), \\
m_{y,1} =-\frac{1}{2}(m_{y,A}+m_{y,B}), \\
m_{y,2} =-\frac{1}{2}(m_{y,A}-m_{y,B}).
\end{eqnarray}
If we consider using the ground state of non-interacting fermionic
atoms as the initial state for iteration while searching for the many-body
ground state, we have $m_{z,\sigma,A}=m_{z,\sigma,B}$ and $m_{x,A}=m_{x,B}=m_{y,A}=m_{y,B}=0$,
so that $m_{z,3}=m_{z,4}=m_{x,1}=m_{y,1}=m_{x,2}=m_{y,2}=0$.
Based on this argument, we have
\begin{equation}
H_{Int}({\bf k})=m_{z,1}+m_{z,2}\sigma_{z}.
\end{equation}
Clearly, the presence of interactions may induce an effective Zeeman field,
which will shift the locations of degenerate points along the z direction and
may cause the transition between the type-I and type-II, but will neither destroy
Weyl nor four-fold degenerate points.
| 05c9323d04aaf310a99b2ab5ce1c6c8a16178640 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction \label{introd}}
Strong lensing
provides a useful tool to uncover the mass distribution in
galaxies \citep[e.g.,][]{0004-637X-575-1-87,2002MNRAS.337L...6T,2006ApJ...649..599K}
and galaxy clusters \citep[e.g.,][]{1989ApJ...337..621K,1998MNRAS.294..734A,2007MNRAS.376..180N,2010AdAst2010E...9Z,2010ApJ...715L.160C,2010ApJ...723.1678C}.
Gravitational arcs have also been used to constrain the background cosmological model
\citep[e.g.,][]{1998A&A...330....1B,1999A&A...341..653C,2002A&A...387..788G,2002MNRAS.337L...6T,2001PThPh.106..917Y,2004MPLA...19.1083M, 2005MNRAS.362.1301M,2010Sci...329..924J,Magana2015,0004-637X-806-2-185,2016A&A...587A..80C}.
More recently, arcs and Einstein rings, in combination with kinematic information for the lenses, have been employed for testing modified gravity \citep[e.g.,][]{Schwab2010,Enander2013,2016arXiv160203385P}.
Furthermore,
strong lenses are also being exploited as cosmic telescopes, enabling spectroscopic and spatially resolved studies of high-redshift sources, such as
dwarf galaxies at distant redshifts \citep{Marshal2007}, star-forming galaxies \citep{Stark2008}, quasar accretion disks \citep{2008ApJ...673...34P}, and faint Lyman-alpha blobs \citep{2015arXiv151205655C}.
The many applications of gravitational arcs in astrophysics and cosmology have spurred the search for these objects in both space-based and ground-based observations.
This includes searches in Hubble Space Telescope (HST) mosaics, such as the {\it Hubble Deep Field} \citep[HDF;][]{1996ApJ...467L..73H}, {\it HST Medium Deep Survey} \citep{1999AJ....117.2010R}, {\it Great Observatories Origins Deep Survey} \citep[GOODS;][]{2004ApJ...600L.155F},
{\it Extended Groth Strip} \citep[EGS;][]{Marshall2009robot}, {\it HST Cosmic Evolution survey} \citep[COSMOS;][]{2008ApJS..176...19F,2008MNRAS.389.1311J}
and in targeted observations of galaxies
\citep{2006ApJ...638..703B,2012ApJ...744...41B}
and clusters \citep{2005Smith,2005ApJ...627...32S,2010MNRAS.406.1318H,2016ApJ...817...85X}.
Investigations from the ground include
follow-ups of clusters
\citep{1999A&AS..136..117L, 2003ApJ...584..691Z,2008AJ....135..664H,2010A&A...513A...8K,2013MNRAS.432...73F}
and galaxies \citep{2006MNRAS.369.1521W},
and searches in
wide-field surveys, such as the {\it Red-Sequence Cluster Survey} \citep[RCS;][]{2003ApJ...593...48G,2012ApJ...744..156B},
{\it Sloan Digital Sky Survey} \citep[SDSS;][]{2007Estrada,2009MNRAS.392..104B, 2010ApJ...724L.137K, 2011RAA....11.1185W,2012ApJ...744..156B}, {\it Deep Lens Survey} \citep[DLS;][]{Kubo2008},
{\it Canada-France-Hawaii Telescope (CFHT) Legacy Survey}
\citep[CFHTLS;][]{2007A&A...461..813C, 2012More,Maturi2014,RINGFINDER,SPACEWARPSII,ParaficzCFHTLS},
{\it CFHT Stripe 82 Survey} (CS82; Caminha, More et al., in prep.), and Dark Energy Survey\footnote{\texttt{http://www.darkenergysurvey.org}} \citep[DES;][]{DES01} Science Verification data \citep{DESSL1Nord}.
As of now, the largest homogeneous samples of gravitational arcs have on the order of a hundred systems. These numbers will increase by one order of magnitude with the close completion of the Kilo Degree Survey\footnote{\texttt{http://kids.strw.leidenuniv.nl/}} \citep[KiDS;][]{KiDSr2} and DES \citep{DESnonDE}, which will cover, respectively, 1000 and 5000 square degrees with sub-arcsecond seeing.
Comparable numbers are expected from the ongoing Hyper Suprime-Cam\footnote{\texttt{http://www.naoj.org/Projects/HSC/surveyplan.html}} (HSC) and the forthcoming Javalambre Physics of the Accelerating Universe Astrophysical Survey \citep[J-PAS;][]{JPASredBook} projects.
These numbers are expected to increase even further in the near future, with the operation of the Large Synoptic Survey Telescope \citep[LSST;][]{LSST20} and
Euclid\footnote{\texttt{http://www.euclid-ec.org/}}
\citep{EuclidSciBook},
which are both expected to detect $\mathcal{O}\left(10^5\right)$ systems with arcs \citep{Collet2015}.
The vast majority of the current samples of arc systems involve a visual search and classification. This is true for the targeted surveys and also for the wide-field imaging surveys, where either the full footprint or cutouts
around potential lenses (e.g., luminous red-galaxies, galaxy clusters) are visually inspected.
This manual procedure is still possible for current surveys with good image quality,
which cover at most few hundred square degrees. However, it will become prohibitive for DES and KiDS, and especially for LSST and Euclid. Therefore, the development of automated arc finding methods is absolutely needed for the scrutiny of these surveys in the quest for gravitational arcs.
Regardless of the size of the survey, automated arc detection is important for an objective and reproducible definition of arc samples, which often includes the determination of arc properties.
This is of course critical for arc statistics \citep[see, e.g.,][]{2013SSRv..177...31M,2016ApJ...817...85X} and for any comparison of real and simulated data \citep[e.g.,][]{2005Horesh,2011MNRAS.418...54H}
and among different data sets \citep[e.g.,][]{2010MNRAS.406.1318H}.
Motivated by these needs, several automated methods to find gravitational arcs have been proposed in recent years.
Most of these methods focus on pattern recognition, i.e., on identifying shapes that look like gravitation arcs, in particular, thin and elongated structures \citep[e.g.,][]{2004Lenzen,2005Horesh, 2006Alard,2007Seidel, 2012More};
in some cases, these methods require a degree of curvature
\citep[e.g.,][]{2007Estrada,Kubo2008}.
\citet{Maturi2014} combine this approach with a multicolor selection of the sources.
\citet{Marshall2009robot} use lens inverse modeling to find strong lenses, i.e., assuming that a given object in an image is a consequence of lensing and determining whether the lensing solution is favored by the data.
More recently, new arc finders have been proposed that subtract the lens candidate (usually early-type galaxies) light distribution, either using two bands, as in \citet{RINGFINDER}, or by modeling the lens in a single band, as in \citet{JosephPCA2014} and \citet{Brault2015}.
The residuals are then investigated, using their shapes \citep{JosephPCA2014}, by color selection \citep{RINGFINDER}, or with inverse modeling \citep{Brault2015}.
The inverse modeling approach is particularly interesting as it uses the physics of lensing to find candidates, however its applicability has been restricted to galaxy-scale lenses; this is because the automated modeling is much more tractable in this case owing both to the simplicity of the lens model and the identification of the images.
The lens subtraction approach is a necessity for galaxy-scale lenses, especially when observed from the ground, as the arcs can be embedded in the galaxy's light.
On the other hand it is less critical for arcs on cluster scales, which span larger angular sizes than the galaxies and the PSF.
Our main interest here is to look for arcs in galaxy groups and clusters that can be found even with a single band survey (as in CS82). Therefore, in this paper we focus only on the development of a pattern-recognition based arc finder.
Most arc finders in this category use sets of measurements of the objects, such as ellipticity, length, $L$, width $W$, and {\it arcness,} to determine whether they are arc candidates or not. They usually employ hard (i.e., fixed and mutually independent) cuts, whose values may be arbitrarily assigned or tuned using data or simulations. However, given the diversity of arc properties (shapes, sizes, $S/N$ ratios, etc.) and their physical origin, different
cuts could perform better in different regions of the multidimensional space of arc parameters.
For example, arcs may be very elongated and not necessarily curved for galaxy cluster lenses, while arcs are not as drastically elongated for galaxy lenses but exhibit a clear curvature.
Therefore, a flexible criterium based on a combination of parameters may be more efficient than applying hard cuts.
This is a typical situation in which machine learning methods can be extremely helpful. A suitably trained algorithm can then classify the objects into arcs or not, given a set of input values for the object features. Such training can be carried out either on real data (on objects previously known to be arcs) or using simulations, by feeding the algorithm with a large set of arc and nonarc samples. This process is
characteristic of supervised learning methods, of which the most well known is the back-propagation artificial neural network \citep[ANN;][]{williams1986learning,rumelhart1988learning}.
The choice of the set of input parameters is as important as the choice of the classification method and its configurations. In this work we adopt measurements derived from the Mediatrix filamentation method \citep{2012Bom,BomMediatrix}, a novel iterative technique that decomposes elongated objects into segments along their intensity ridge line. This method provides several morphological parameters that are well suited to characterize arcs, including the length along the ridge line, $L$, the width $W$, and, most notably, estimates of the object center of curvature and its significance.
Therefore, the purpose of this work is to construct an ANN gravitational arc finder based on the Mediatrix filamentation method, or {\it ANN Mediatrix Arcfinder} (AMA) for short. We use a sample of simulated gravitational arcs and a sample of nonarcs from HST images to train and validate the ANN. This sample is used to
pin down a few configurations among the many possible choices involved in the ANN detection process: the types of images used for the training, the selection of inputs given to the ANN, the number of neurons, and the final threshold for classification.
As an illustration of the application of the method to real data, we consider four galaxy cluster images from HST and run the AMA on them, comparing the results with the training and validation.
The paper is organized as follows. Next section provides an overview of the AMA algorithm: the processes to detect and select the objects, the measurements carried out on them, in particular the Mediatrix filamentation, and the ANN used in the arc identification.
Section \ref{train_and_validation} describes the training and validation process, the samples used, and the tuning of the ANN for arc detection. Section \ref{aplication} shows an example of application of the method on real data.
Final discussions and concluding remarks are presented in section
\ref{discusao}.
\section{\label{apresentacao} The Mediatrix arcfinder algorithm}
Most gravitational arc finding methods based on pattern recognition techniques can be divided into four steps: object
segmentation, preselection, measurement, and final classification.
In the segmentation phase, sets of pixels above the background are grouped into objects (as discussed in Sec. \ref{identificacao}).
In the preselection phase, we define a sample of objects to be analyzed, performing cuts to eliminate those that can be readily discarded as not being arcs (Sec. \ref{preselection}).
The measurements are carried out through
the Mediatrix filamentation method \citep[][see Sec. \ref{mediatrix}]{2012Bom}, which provides a number of parameters that characterize the objects.
For the final classification
we use an ANN trained to identify arc candidates (see Sec. \ref{RNA}).
\subsection{ \label{identificacao} Object segmentation}
The first step is to identify the objects in the image, separating them from the background and defining which set of pixels belong to a given object. To this end, we use the \texttt{SExtractor} \citep{bertin96} software, which has several parameters controlling the object identification, deblending, and measurement process, including the minimum signal-to-noise ratio for a given pixel to be considered, minimum number of pixels, and deblending thresholds.
Tuning these parameters is a key step in arc identification, especially as arcs are low surface brightness objects, are often close to brighter sources, and can thus be easily missed and/or blended with other sources.
However, in this paper we do not perform a systematic optimization of these parameters. We rather use a set of manually tuned values that provided good results on a visual inspection, as our main focus is in the measurement and final classification phases.
\citet{2005Horesh}, \citet{2007Estrada}, and \citet{Kubo2008} also use \texttt{SExtractor} in the object segmentation phase of their arc finding methods.
\texttt{SExtractor} provides an output catalog containing measurements on the objects identified and several image outputs with the same dimensions as the original one. Here we use two such output images, namely \texttt{OBJECTS}, containing the pixel values of all objects identified, and \texttt{SEGMENTATION}, in which all pixels belonging to the same object have the same value (corresponding to the object ID in the catalog). From these two images we produce a single array per object, which contains only the pixels belonging to that object and the respective pixel value. These arrays are called {\it postage stamps} and are kept in the memory for the next steps. From this point on the AMA algorithm will work separately on each object.
\subsection{ \label{preselection} Preselection}
Among the measurements provided by \texttt{SExtractor} are the object semimajor axis $A$ and
semiminor axis $B$ derived from the weighted central second moments of the pixels \citep{bertin96}.
From them we define the
ellipticity $e$ as
\begin{equation}
e=1-\frac{B}{A},
\end{equation}
which is used to eliminate from the sample objects with ellipticities below some threshold $e_{\rm th}$.
For the images used in this paper, we set $e_{\rm th} = 0.4$.
We also add a cut on the maximum number of pixels to exclude objects that are too large and are definitely nonarcs.
To avoid spurious detections we remove objects that are close to the image borders.
We do not make cuts in the object signal-to-noise ratio or magnitude, as this could remove some of the faint arcs.
We do not apply any star-galaxy separation either, as the cut on $e$ already removes the stars.
\subsection{\label{mediatrix} Measurements with the Mediatrix filamentation method}
In this section we review the Mediatrix filamentation method \citep{2012Bom,BomMediatrix} and introduce the outputs of this process that are used to produce the inputs for the ANN to identify the arcs.
The Mediatrix filamentation method is an iterative algorithm to assign $N$ filaments along the intensity ridgeline of a given object and derive morphological measurements such as length, width, curvature, and curvature center.
The method operates on a set of points with a given intensity, i.e., in the postage stamps of the objects after the preselection phase. The first step is to find the two most distant set of points $E_1$ and $E_2$. The first Mediatrix point $M^1_1$ is defined by the highest intensity pixel of the object along the perpendicular bisector to the segment $\overline{E_1E_2}$ (see Fig. \ref{d_definition}). In practice this is chosen as the highest intensity pixel from all pixels with positions $\vec{x}_i$ whose distance $s_{ij}$ to the bisector $y_{j}$ satisfies the condition: $s_{ij}(\vec{x}_i,y_j) \leq \frac{\sqrt{2}}{2}dpix$, where $dpix$ is the pixel scale.
The same procedure is repeated between $E_1$ and $M^1_1$ and $E_2$ and $M^1_1$, defining a new pair of Mediatrix points, $M^2_1$ and $M^2_2$, where the superscript defines the iteration level and the subscript denotes each point. The whole process is repeated $n$ times. For each step $k$, the pair of neighboring points from all previous steps defines a new Mediatrix point, $M^{k}_{l}$, the $l$-$est$ point in the $k$-$est$ iteration level.
The output of the Mediatrix method is a set of
$N$ vectors $\vec{n_i}$, with $|\vec{n_i}|=l_i$, where $l_i$ is the distance between a given pair of neighboring Mediatrix points. The orientation is
perpendicular to $l_i$ and its origin in the middle point of $l_i$. An example of the output of the Mediatrix filamentation on a simulated gravitational arc is shown in Fig. \ref{mediatrixexample}.
\begin{figure}[!ht]
\begin{center}
\sidecaption
\resizebox{\hsize}{!}{\includegraphics{d_definition-v3.png}}
\caption{\label{d_definition}
Illustration of the Mediatrix filamentation points, corresponding to an iteration level of $n=3$. Only one vector and a few points and segments are shown for clarity. The object chosen is an ArcEllipse \citep{arcellipse} for illustration purposes. Also shown is the distance $\vec{d}_j$ from a point $\vec{r}$ on the plane to the line spanned the vector $\vec{n_i}$. }
\end{center}
\end{figure}
\begin{figure}[!ht]
\begin{center}
\sidecaption
\resizebox{\hsize}{!}{\includegraphics{20150201teste1.png}}
\caption{\label{mediatrixexample}
Example of the Mediatrix method applied to a simulated arc for $n=3$. The dots show the Mediatrix points and the arrows the $\vec{n_i}$ vectors. The solid line is the circle that contains points $E_1$, $M^1_1$, and $E_2$.
}
\end{center}
\end{figure}
For an iteration level $k$ we define the object length $L_k$ as
\begin{equation}
\label{length} L_k= \sum^{N}_{i=1} l_i.
\end{equation}
and width $W_k$ as
\begin{equation}
\label{width}
W_k=\frac{4A}{\pi L_k},
\end{equation}
where $A$ is the object area in pixels.
This relation between area, length and width is exact for arc-shaped objects known as ArcEllipses \citep{arcellipse}, as used to illustrate Fig. \ref{d_definition},
and is an excellent approximation for a specific solution for gravitational arcs (Pacheco et al. in preparation).
When the segments $l_i$ are too small, there is no point in continuing the Mediatrix iteration. On the contrary, the decomposition starts to be dominated by noise (or by the finite pixel size). We expect that when $l_i \lesssim W_k$ the directional information is lost and no further division is useful. Therefore we impose the following condition as a criterium to stop the iteration:
\begin{equation}
\label{limite}
l_i \leq \alpha W_k,
\end{equation}
where
$\alpha$ is a parameter that
is set to $\alpha=1$ in this work.
The decision to continue with the Mediatrix decomposition is taken independently for each segment following Eq. (\ref{limite}). The iteration generally stops for regular objects at the same level for all segments, such that the total number of segments (and oriented vectors) is $N=2^n$, where $n$ is the last iteration level. However, if the shape is irregular, the iteration can be carried out to different levels for different regions of the object.
Also, if the object is composed by noncontiguous sets of pixels, the code may not find any pixel along the perpendicular bisector of two given Mediatrix points. In this case, the iteration is stopped so that no further division happens between those two points.
After the last iteration, the sum in Eq. (\ref{length}) is carried out for all segments defining the final length $L$.
The final width $W$ is defined using Eq. (\ref{width}).
We may define a center of curvature by determining the circle that passes through the points $E_1$, $M^1_1$, and $E_2$ (as shown in Fig. \ref{mediatrixexample}). The center of this circle, $\vec{r}_c$, is defined as a center of curvature and the curvature radius is the circle radius, $R_c$. Using these points we may also define a length as the circle arc length, $L_c$, between the two extreme points, $E_1$ and $E_2$.
Another possibility is to define a curvature center based on all points from the Mediatrix filamentation: the point on the plane that minimizes the distances to the perpendicular bisectors of all Mediatrix segments $l_i$ (i.e., the distance to the lines defined by the vectors $\vec{n}_i$). For this sake we define the function $M( \vec{r})$ such that
\begin{equation}
\label{m_de_r} M( \vec{r}) := \frac{1}{N} \sum^N_{i=1}|\vec{d_i}(\vec{r},\vec{n}_i)|^2,
\end{equation}
where $|\vec{d}|$ is the distance of point $\vec{r}$ to the line defined by $\vec{n}_i$ (see Fig. \ref{d_definition}). The center of curvature is defined as the point $\vec{r}_0$ that minimizes the function $M( \vec{r})$.
If the object has a well-defined curvature center, the function $ M( \vec{r})$ not only has a clear global minimum, but small deviations from $r_0$ also lead to a large increase in $ M( \vec{r})$.
We may treat $ M( \vec{r})$ analogously to a chi-squared function and define a confidence region (CR) such that
\begin{equation}
\label{RC} M( \vec{r})-M_0 \leq \sigma_m ,
\end{equation} where $M_0 = M( \vec{r}_0)$ and $\sigma_m$ is an arbitrary parameter. In this paper we use $\sigma_m=1$.
We expect that curved arcs have a small CR as compared to the arc size. We noticed by visual inspection of the CR and the objects, that for arcs, the CR is usually elongated along the radial direction and does not in general intersect the arc.
We provide, as inputs to the ANN, combinations of the parameters described above derived from the Mediatrix filamentation. These combinations are defined so as to be scale invariant, such that they depend mostly on the object shape and are weakly sensitive to the pixel scale. In some cases we normalize the output by the appropriate power of $L$ to produce the scale invariant quantities.
In particular we tested the ANN with the following set of parameters:
$i$) The length-to-width ratio $L/W$.
$ii$) The mean of the scalar products of each unitary vector $\vec{n}_i/|\vec{n}_i|$ with its neighbor, $\vec{n}_{i+1}/|\vec{n_{i+1}}|$, i.e.,
\begin{equation}
\label{s} s:= \frac{1}{N} \sum^{N-1}_{i=1} \frac{\vec{n}_i}{|\vec{n}_i|} \centerdot \frac{\vec{n}_{i+1}}{|\vec{n}_{i+1}|}.
\end{equation}
This quantity provides a measurement of the coherence of the shape. For very irregular objects, its value is low, while for long and smooth objects (curved or not), its value should be close to 1.
$iii$) The minimum value of the function $M( \vec{r})$ divided by the arc length squared (for dimensional reasons), $M_0/L^2$.
$iv$)
The arc aperture $\Delta \theta:=L/R$, where the radius $R$ is the distance from $\vec{r}_0$ to $M^1_1$.
$v$) The ratio between the arc aperture defined above and the one constructed from the circle that contains points $E_1$, $M^1_1$, and $E_2$,
$\Delta \theta/\Delta \theta_c := \left(L/R\right)/\left(L_c/R_c\right)$.
$vi$) The distance between the center of the circle $\vec{r}_c$ and the minimum of $M(\vec{r})$ normalized by the arc length,
$\delta r: = |\vec{r}_0-\vec{r}_c|/L.$
$vii$) The ratio between the major axis of the CR, $L_{CR}$, and the arc length: $L_{CR}/L$.
$viii$) The eccentricity of the CR, $e_{\rm CR}$.
$ix$) The ratio between the number of pixels from the arc enclosed by the CR, $A_{CR}$, and total number of pixels in the object, $A$, i.e.,
$A_{CR}/A$.
The choice of parameters above is somewhat arbitrary, but is inspired by the visual assessment of these quantities on samples of arcs and objects that are clearly nonarcs. An important component of this paper is to
obtain a set that is at the same time good for discriminating arcs from nonarcs and is less time consuming.
\subsection{\label{RNA} Arc identification with an artificial neural network}
The arc identification process through the ANN can be subdivided into two parts: the training process and the actual classification. In the current implementation
we use a standard back-propagation and fully connected ANN~\citep{williams1986learning,rumelhart1988learning}. The ANN has the following structure: {\it a)} an input layer with $i$ neurons, where $i$ is the number of inputs used in the specific ANN configuration, which in this case is
a subset of the parameters described in section \ref{mediatrix}; {\it b)} a second layer with $j$ hidden neurons;
and {\it c)} the output layer with one neuron.
The ANN activation function is linear and the output is a floating point number $R$ in the range $-1$ to $1$.
The AMA code was developed using the python binding for the Fast Artificial Neural Network (FANN) library\footnote{For further information see \texttt{http://leenissen.dk/fann/wp/}}.
In order to recognize the arc shape using this type of ANN, it is necessary to train this neural network on a group of objects, which were previously classified as arcs and nonarcs.
A successful training process is determinant to reach acceptable results in any back-propagation ANN code. The training requires presenting to the ANN two groups: the arcs group (AG), with desired output $+1$ and the nonarc group (NAG), with desired output $-1$. The two groups also need to have the same order of number of objects, otherwise the ANN may just output as the result a number that represents the larger group.
The AG and NAG are split into a training group and a validation group. In this work we used $80\%$ from all objects chosen randomly from the sample of AG and NAG for training and $20\%$ for validation.
After the training, the ANN is applied to the validation group, yielding an output value $R$ for each object.
We have thus to set
a threshold $t$ on this output such that the code finally
classifies each object (i.e., each set of input measurements on the object) as an arc or not.
After running in the validation group
the code computes the completeness, $c$, defined as the fraction of arcs recovered (i.e., {\it the ratio of the number of detected arcs and the total number of arcs} in the validation group) and the fraction of false positives, $f$, defined as the fraction of nonarcs that are classified as arcs (i.e., the ratio of the number of nonarcs detected as arcs and the total number of nonarcs in the validation group).
After the validation test,
the training and validation groups are redefined randomly and the whole process is repeated $k$ times. In this work we retrained the ANN for a single set of input parameters $40$ times. The validation code outputs the mean completeness $\bar{c}$, mean false positive fraction $\bar{f}$, and their standard deviation for the $k=40$ validation groups.
\section{\label{train_and_validation} Training and validation of the ANN}
In this work we use the training and validation steps to characterize the behavior of the AMA with respect to all aspects of the ANN identification process mentioned in Sec. \ref{RNA} above, including the types of images used in the training, sets of inputs, number of hidden neurons, and threshold for classification.
The goal is to define a good set of configurations for practical applications of the AMA.
In particular we seek to have a high completeness $c$ at the same time limiting the fraction of false positives $f$.
This search for the best parameters for the ANN detection is described in Sec. \ref{training}
The arcfinder method was trained using a sample of $175$ simulated arcs (AG) described in Sec. \ref{addarcs} and $437$ nonarcs (NAG) taken from HST images, as described in Sec. \ref{aplication}.
These numbers are the result of steps 1 (object identification) and 2 (preselection), and thus all objects from the two samples already pass the preselection criteria described in Sec. \ref{identificacao}. In particular all have ellipticities above $e_{\rm th}=0.4$.
Therefore, in all comparisons and tests described in this paper we are really testing the measurements + ANN steps of the whole arc finding process, which is the aim of this contribution.
\subsection{\label{paramsinputs} ANN inputs from the Mediatrix filamentation}
We performed the training with ten different subsets of the parameters {\it i} to {\it ix} described in Sec. \ref{mediatrix} to determine the best combination of Mediatrix parameters to be used as input
for the ANN.
Each subset is labeled with a letter from A to J.
The input configurations are presented in Table \ref{combinations}.
We divided the input parameter sets into two groups depending on whether the CR evaluation is necessary or not for a given configuration. Group $1$ includes only measurements derived directly from the Mediatrix filamentation process ({\it i} and {\it ii}) and from the minimization of $M(\vec r)$ ({\it iii} to {\it vi}), while group $2$
contains measurements that depend on the CR (parameters {\it vii} to {\it ix}).
The total time to run the AMA varies only slightly within each group, but changes considerably between the two groups, as the process to obtain the CR is currently the most time consuming step of the AMA.
From sets A to J, the number of inputs (i.e., the number of parameters in the input vectors) is systematically decreased (except for I and J, which have only one input each).
The three sets of configurations in Group 2 (A, B, C) include the determination of the CR and are thus those that take more computational time. From configuration D downward the time drops substantially. In all cases but one (J), we keep the parameter ($i$), i.e., $L/W$, which is historically the primer arc indicator.
In Sec. \ref{training} we test the AMA for each set and compare the results for $c$ and $f$ to define the best set for practical applications, both in terms of maximizing completeness and minimizing contamination, also accounting for the computational time.
\begin {table*}[htp]%
\caption{Combinations of inputs used for the neural network training (A to J).}
\label{combinations}
\begin {tabular}{c|ccc|cccccc}
\toprule
& Group & 2 & & Group & 1 \\
\hline
& $L_{CR}/L$ & $e_{\rm CR}$ & $A_{CR}/A$ & $L/W$ & $s$ & $M_0/L^2$ & $\Delta \theta$ & $\Delta \theta/\Delta \theta_c$ & $\delta r$ \\
\hline
A &X &X &X &X &X &X &X &X &X \\
B & &X &X &X &X & X &X &X &X \\
C & & &X &X &X &X & X &X &X \\
D & & & &X &X & X & X & X &X \\
E & & & & X & & X & X & X & X \\
F & & & & X & & & X& X& X \\
G & & & & X & & & & X& X \\
H & & & & X & & & & & X \\
I & & & & X & & & & & \\
J & & & & & & & & & X \\
\bottomrule
\end {tabular}
\end {table*}
\subsection{\label{addarcs} Arc simulations with AddArcs}
Given the intrinsic variation in gravitational arc shapes it is important to have a large enough training sample so as to encompass some of their diversity and, at the same time, have sufficient statistics to train the ANN. However, the current samples of arcs taken under uniform observing conditions are still substantially small.
Also, we need a truth table of arcs in the AG and not all known arcs have spectroscopic confirmation. Moreover, we want to be able to control some observational and instrumental parameters, such as the background and noise, point spread function, and pixel size so as to test the arc finder under different conditions. For this sake, we use simulated gravitational arcs for the training and validation phases.
We use exactly the same sample of simulated arcs as in \citet{BomMediatrix}. That paper provides a more detailed description of the simulation process and the selection of this specific sample. Below we provide an outline of the simulation procedure for completeness.
The simulated sample was created using the \texttt{AddArcs} pipeline (Brandt et al., in preparation), which uses two input catalogs: one with the properties of the lenses (such as mass, ellipticity, and redshift) and one with the properties of the sources (such as magnitude, size, ellipticity, and redshift), plus a number of configurations that can be set, such as observational and instrumental parameters. The code distributes the sources in random positions for each lens in the catalog, following the specified surface number density, and then it randomly chooses the source parameters from the source catalog. Given the input models for the source and the lens, from their respective catalogs, the pipeline uses the \texttt{gravlens} code \citep{gravlens}
recursively to perform the projection of the sources onto the image plane. It then identifies which images correspond to arcs and generates postage stamps from them, providing as one of its outputs a pixelized surface brightness distribution of these objects, i.e., a simulated image of a gravitational arc.
For our simulated arc sample the input catalog contains galaxy cluster scale halos from $N$-body simulations\footnote{We use a catalog from the {\it Las Damas/Carmen} $N$-body simulation, \texttt{http://lss.phy.vanderbilt.edu/lasdamas/}.} and we assume a Navarro--Frenk--White density profile \citep[NFW;][]{nfw1,nfw2}, with elliptical surface mass density \citep[see, e.g.,][]{2013caminha}, a given mass--concentration parameter relation \citep{2008MNRAS.387..536G,2007MNRAS.381.1450N}, and fixed ellipticity. The sources are given by a S\'ersic surface brightness distribution \citep{sersic} with parameters derived from the Hubble Ultra Deep Field Survey \citep[UDF;][]{2006AJ....132.1729B,2006AJ....132..926C}.
In Sec. \ref{aplication} we apply the trained AMA to real images of systems containing arcs taken with the WFPC2 instrument on HST \citep{2005Smith}. As in our training the AG is given by the simulated arc sample, we set the observational conditions in the simulations to mimic these HST images. In particular, we use the same pixel scale as WFPC2 and convert the counts in each pixel on the simulated images to data numbers using the properties of these HST images \citep[for details, see Appendix A of][]{BomMediatrix}. At this point the simulated arcs are smooth, i.e., the pixels have no fluctuations from noise and the simulated images have no background. We refer to this calibrated set of arcs as pure arcs.
However, real astronomical images have noise (including Poisson noise from the counts in pixels) and background. Even though it is common to work on background-subtracted images, of course the background noise remains. Therefore, for a proper test of the arc finding process we need to include at least these two effects on the simulated images, as they are of fundamental importance for object detection and measurement.
The (constant) background is added to all pixels as measured from the HST images. Then each pixel is assigned a new value sampled from a Poisson distribution with the mean given by that pixel value in count units (including object plus background). Finally, the new image with background and noise is converted again to the data units. For details of these processes, see Appendix A of \citet{BomMediatrix}. The derived images form the background and noise sample of arcs.
Both samples of simulated arcs go through the object identification and preselection phases, as described in section \ref{apresentacao}. In particular, \texttt{SExtractor} is run on each simulated image containing one arc and a postage stamp is created for that object. The Mediatrix method is then applied and the derived parameters are input to the ANN.
The validation of the trained ANN is performed using the background and noise sample as it is the more realistic sample. Nevertheless, it is interesting to test the results of the training carried out
using each type of image as entries. These tests are discussed in the next section.
\subsection{\label{training} The training and validation results: Determining good ANN configurations}
In this section we present the training and validation results and use them to select an optimal set of configurations for the ANN arc finding process. We start by looking at the dependence of $f$ and $c$ as a function of the number of hidden neurons ($N_h$). We considered all configurations from A to J described in Table \ref{combinations} and varied the number of neurons from $2$ to $15$ fixing the threshold for the ANN output to $t=0$ (i.e., objects with the ANN response function $R>0 $ are defined as arcs). First we considered the training carried out on the pure arcs set and then on the background and noise images as the AG (the NAG is the same in both cases). In the first case, we do not notice any significant variation of the completeness and false positives with $N_h$ for any input configuration. On the other hand, when the training is performed using the images with background and noise, we do see a dependence on the hidden neurons for some of the configurations. This is shown in Fig. \ref{neurongra} for only three configurations:
A (representative of group 2), D (representative of group 1 with several inputs), and J (with only one input).
We see that the dependence of the completeness on $N_h$ is only significant for the A configuration,
for which
we can observe a significant increase in $c$ up to $N_h \sim 6$.
In all configurations that we have investigated, there is no gain in increasing $N_h$ above this value.
Thus, only for the configuration with the highest number of inputs and using the training set with more variation among the systems (due to the noise in the AG) does the ANN require more complexity than two hidden neurons.
On the other hand, the false positive fraction does increase with $N_h$ in most cases. This is less visible in configuration $A$ because of the large variance and could be due to overtraining, when the number of neurons is large.
As we see below, we end up choosing the pure arcs as a training sample
and therefore we could be tempted to choose a very small number for $N_h$, as this also significantly decreases the computational time. On the other hand, to be on the safe side for real world applications, we still want to have a larger number of neurons
for dealing with the diversity of real arcs.Therefore, we set our final number of hidden neurons to Nh = 4 in all tests to reach a balance between computational time, false positives, and completeness.
Therefore, we set our final number of hidden neurons to $N_h = 4$ in all tests to reach a balance between computational time, false positives, and completeness.
\begin{figure}[!ht]
\begin{center}
\sidecaption
\resizebox{\hsize}{!}{\includegraphics{Neurons_Treinada_BKG_NOISE.png}}
\caption{\label{neurongra} Mean completeness $\bar{c}$ and fraction of false positives $\bar{f}$ as function of hidden neurons, using background and noise arcs in the AG for training and a threshold $t=0$. The error bars are the standard deviation from the 40 training plus validation subsamples. We shifted the symbols horizontally for clarity.}
\end{center}
\end{figure}
Now we turn to the choice of the final set of configurations and the types of images for the training. In Fig. \ref{configgra} we show the results for $c$ and $f$ for the ten configurations in Table \ref{combinations} using pure arcs (in large blue circles) and those with background and noise (in small green dots) for the training. Clearly, configurations A to D have the best performance, both for completeness and for false detections. The results from training on pure arcs in general have a smaller variance than using background and noise, in particular for configurations A and B. In addition, for lower thresholds, A and B have much more contamination when trained on arcs with background and noise than with pure arcs.
At first sight it could seem surprising that training with the more realistic set of arcs in general gives worse results. For the ANN, it is better to learn with a more consistent and well-defined set of parameters from the arcs, than having a larger variation on these parameters owing to noise, even if the validation is carried out on the images that do have noise and background. It is important to mention that the noise is added only once to the arc sample, i.e., the variance among the 40 subsamples is not due to different realizations of the noise, but rather to the spread in the parameters caused by the noise in each subsample of arcs. For now on, we choose to carry out the training process using only the pure arc sample.
As for the inputs, we see from Fig. \ref{configgra} that the combination that gives the highest mean completeness and a low fraction of false detections is configuration A, from group 2. However, D also gives a good performance (as good as the other configurations in group 2) but is in group 1, i.e., is computationally faster, as it does not require the computation of the CR. We therefore keep these two sets of inputs, A (the best) and D (almost as good as A, for both $c$ and $f$, but faster), for the next test and for applications to real data.
\begin{figure}[!ht]
\begin{center}
\sidecaption
\resizebox{\hsize}{!}{\includegraphics{DiferentesTreinamentos-slegenda.png}}
\caption{\label{configgra} Mean completeness $\bar{c}$ and false positive fraction $\bar{f}$ for the different sets of inputs (for threshold $t=0$). The results from the training in pure arcs are shown as large blue dots, while those using images with background and noise are shown as small green dots. The error bars are the standard deviation from the sets of training plus validation subsamples. We shifted the symbols horizontally for the two types of input images for clarity.}
\end{center}
\end{figure}
Finally we look at the dependence of $c$ and $f$ on the threshold $t$ for these two selected configurations.
We vary $t$ in steps of $0.25$. The cases $t=-1$ and $t=+1$ are trivial as all objects are classified as arcs and nonarcs, respectively. In Fig. \ref{thrgra} we show the results for $t$ in the range $[-0.75,0.75]$.
As expected, both the completeness and the false detections decrease as $t$ is increased. However $c$ has a softer dependence with $t$ than $f$.
Two possible choices for $t$ are in order. If we want to have a higher completeness, even at the expense of a higher percentage of false detections, then $t=-0.75$ is a good choice. This threshold would be preferred, for example, in targeted surveys, where a visual inspection to discard false positives is feasible even if the false detections outnumber the real arcs.
In this case we obtain $c \sim 95 \%$ and $ 90 \% $ and $ f \sim 10\%$ and $25\%$ for configurations A and D, respectively.
On the other hand, if we seek a purer sample of arcs, a good choice is $t=+0.25$. After this value $c$ drops considerably, while $f$ does not vary much. This choice could typically be adopted for a wide-field survey, where we need to minimize the fraction of the objects to be inspected for a final selection. In this case $c$ decreases a bit to $\sim 90\%$ and $80\%$, but $f$ drops substantially to $\sim 3 \%$ and $2 \%$, respectively, for A and D.
We recall that $f$ is defined as the fraction of false positives with respect to the total number of nonarcs, i.e., it is essentially the number of false detections over the total number of objects that pass the preselection cuts. In the training and validation process the numbers of arcs and nonarcs are of the same order of magnitude. However, for wide-field surveys, the number of arcs is roughly five orders of magnitude less than the total number of objects detected. Therefore, even if the preselection phase filters out $90\%$ of the objects and $f$ is as low as $1\%$, the false detections would still outnumber the real arcs by large amounts. Thus, even a low contamination as currently achieved with $t=+0.25$ would still require a further step of visual inspection when applied to large surveys, as happens with most arc finders proposed so far.
It is worth pointing out that $t$ can always be set a posteriori in the sense that the ANN is specified without the need to define a threshold. Once the inputs and hidden neurons are defined and the training is carried out on a given sample, the ANN is fully determined. When the ANN is applied to the data, the result is an output value of the response function for each object. Therefore we can vary the value of $t$ after the ANN is run and choose a suitable balance between $c$ and $f$ to set the threshold.
We see from Fig. \ref{thrgra} that the completenesses for configurations A and D are compatible with each other
within their standard deviations for the whole range of $t$ (except for $t =0.75$),
showing that both configurations are comparable for arc detection (although $\bar{c}$ is systematically higher for A).
Regarding the false detection fraction, it is clearly higher for configuration D and $t<0$. The highest difference with A occurs for $t = -0.75$ and is smaller than two standard deviations. It is not clear whether this is a real difference between the two configurations or if it is just a fluctuation.
In the next section we apply the two ANN (i.e., with configurations A and D) trained as described in this section to objects from real HST images.
\begin{figure}[!ht]
\begin{center}
\sidecaption
\resizebox{\hsize}{!}{\includegraphics{Threshold_Treinada_Pure_Arcs.png}}
\caption{\label{thrgra} Mean completeness $\bar{c}$ and mean false detection fraction $\bar{f}$ for the input configurations A and D as function of threshold. The results are obtained from the training on pure arcs with an ANN with four hidden neurons. Symbols are shifted in the horizontal direction for clarity.}
\end{center}
\end{figure}
\section{\label{aplication} Application on HST cluster images}
In this section we show an example of application of the AMA to real images taken with the HST. In particular, we consider a well-known sample of massive clusters observed with
the Wide Field and Planetary Camera 2 (WFPC2) instrument on HST from \citet{2005Smith}.
Images from this camera have been used in other exploratory studies of arc finders \citep{2004Lenzen,2005Horesh,2007Seidel}, and, in particular, the same \citet{2005Smith} sample was used in \citet{2005Horesh}.
The WFPC2 instrument has a mosaic of three wide-field (WF) CCDs (forming an L pattern) and a smaller CCD with finer pixel scale close to the center of the field.
The exposures of each cluster are centered on one of the WF CCDs. Several exposures are taken with a dither pattern, so as to combine all CCDs into a single image with no gaps. The HST server%
\footnote{The HST data products can be downloaded from the European HST Archive at ESA/ESAC:
\texttt{http://archives.esac.esa.int/ehst/}}
provides both the combined image with all CCDs and a combined image of all exposures for each single CCD.
For the purposes of this paper it is better to work on the single images per CCD, as the mosaic images have strong $S/N$ variations and artifacts
in the regions between CCDs and close to the edges. This can produce spurious detections and affect the background estimation and it is beyond the scope of this paper to deal with them. For each CCD image we remove the regions near the image borders to avoid the spurious detections.
We use the images from \citet{2005Smith}
to apply the AMA to find arcs in the images, but also to provide the sample of nonarcs for the training of the ANN. This is so that the NAG have exactly the same observational and instrumental conditions as the images in which we look for arcs. We mimic those same conditions in one of the simulated arc samples as discussed in\footnote{ For details on the noise and background evaluation in the HST images
see \citet{BomMediatrix}.}
Sec.\ref{addarcs}. In particular, we consider CCDs that do not contain the cluster center and have no apparent arcs to provide the NAG. We use seven such CCDs,
carry out the detection and preselection steps and end up with the sample of 437 nonarcs used in the training of the ANN discussed in the previous section.
We apply the AMA to four clusters in the \citet{2005Smith} sample that have giant and clearly visible arcs, namely Abell 68, Abell
383, Abell 773, and Abell 963.
The images of the CCD with more arcs for each cluster were visually inspected and the arc candidates were classified in three categories:
A for the best candidates, objects with curved shapes close to the cluster central regions or galaxy cluster members; B for intermediate candidates, which are curved but do not have a cluster center or galaxy as a center of curvature or that are close to the cluster center but are not curved; and C for more ambiguous candidates that do not fall in the previous categories. This classification is somewhat arbitrary
but is useful for a first assessment of the ability of the ANN to recover the arcs as a function of their quality/likelihood. For each cluster we label the arcs in each category with a number (e.g. a4, b2, and c1). In Figs. \ref{A68_and_A383} and \ref{A773_and_A963} we show cutouts of the images encompassing the regions of each selected CCD where arcs were visually selected and marked.
In cluster A68 we marked 9, 2, and 1 arcs in the A, B, and C categories, respectively. For A383 we labeled 7, 4, and 2 arcs in these categories, while for A773 the numbers are 1, 1, 1, and for A963 7, 3, 1. This gives a total of 39 arc candidates, 24, 10, and 5 respectively in the A, B, and C, categories. We use these identifications as a truth table for testing the AMA.
The aim here is only to have a set of objects with a morphology visually associated with gravitational arcs.
It is hard to compare the numbers above with the other identifications for the same clusters in the literature. On one hand, we only consider the arcs from a single CCD, do not include radial arcs, and do not impose an $L/W$ cut for arc selection. On the other hand the visual identification is rather ambiguous anyway.
In any case the orders of magnitude are compatible with the visual searches in
\citet{2005ApJ...627...32S}, which identified $27$ arc systems in these four clusters and \citet{2005Smith}, which found $33$ multiple images in the four CCD chips we consider in this work.
\begin{figure}[!ht]
\begin{center}
\sidecaption
\resizebox{\hsize}{!}{\includegraphics{A68_and_A383.png}}
\caption{Arc candidates in Abell 68 and Abell 383. The candidates were classified in $3$ categories: A, best candidates indicated with blue circles; B, intermediate candidates indicated with green rectangles; and C, ambiguous candidates indicated with magenta ellipses.}
\label{A68_and_A383}
\end{center}
\end{figure}
\begin{figure}[!ht]
\begin{center}
\sidecaption
\resizebox{\hsize}{!}{\includegraphics{A773_and_A963.png}}
\caption{Arc candidates in Abell773 and Abell963. The arc candidates in the $3$ categories are indicated following the same convention as in Fig. \ref{A68_and_A383}.}
\label{A773_and_A963}
\end{center}
\end{figure}
Now we apply the AMA algorithm, following the steps described in Sec. \ref{apresentacao}, to the four selected clusters. First we run
\texttt{SExtractor} (step 1), with the same configurations used in the training and validation phases, obtaining a total of 1378 detections. Applying the ellipticity and other cuts on the preselection (step 2) leaves us with 304 objects, on which we run the Mediatrix filamentation (step 3), providing the inputs for the ANN.
It must be pointed out that not all arcs visually identified comprising our truth table are found to be objects in the \texttt{SExtractor} run (likely because of their low surface brightness). Moreover, many of the arcs found end up blended with other objects in the image and therefore the morphology of the generated postage stamp does not represent an arc anymore. The total number of arcs that are either not detected or blended is 15.
For a fair comparison with the results of Sec. \ref{training} these objects should not be considered in the denominator for computing the completeness $c$, as we do not expect the ANN to identify them as objects; the ANN was not trained on blended objects and obviously the arcs that are not detected cannot be classified by the ANN.
Finally, we apply the ANN, trained as described in Sec. \ref{training}, using either the sets of inputs A or D, which are selected from our exploratory runs on the simulated arcs (using four hidden neurons and trained using the pure arcs as AG). The resulting number of arcs detected is of course a function of the threshold $t$. In Table \ref{resultsreal}, as an
example we show the results using configuration D and $t=-0.75$. In that case, a total of 16 arcs were identified and there were 43 false detections.
We also show the number of arcs visually identified in each category along with the arcs that were not detected by \texttt{SExtractor} and those that were blended with other objects. We note that in A68 four out of six detectable arcs were found, while for A383 six out of seven were identified by the ANN, in A773 one out of two and in A963 five out of eight. The overall $c$ in this case is $16/24\simeq 67 \%$.
Considering each subclass, we have a completeness of $71\%$, $67\%$, and $50\%$
for the A, B and C categories respectively. Although these numbers show the expected trend with arc quality, they are all mutually consistent taking Poisson statistics into account, and we cannot conclude whether the arc finder performs better or not with arc quality. Therefore, in the remaining of this paper we consider all categories together for evaluating $c$.
\begin{center}
\begin {table*}[htp]%
\caption{Arc detections in the 4 selected HST clusters. Columns 2 to 6: number of arc candidates visually identified in each category (A, B, and C), objects that are not detected by \texttt{SExtractor,} and arcs that are blended with other objects. Column 7 shows the arcs detected by the ANN for configuration D and $t=-0.75$. The last column shows the number of false positives (i.e., objects classified as arcs but not on the arc truth table) for the same configurations.}
\label{resultsreal}
\begin {tabular}{|c |c | c | c | c | c | | c |c |}
\hline \hline
& \multicolumn{3}{c|}{Category} & \multicolumn{2}{c||}{SExtractor} & \multicolumn{2}{c|}{ANN}\\
\hline \hline
Cluster & A &B & C & Not detected & Blended & Arc detections & $N_f$ \\
\hline
Abell 68 & 9 & 2 & 1 & a9 & a2, a3, a4, b1 & a6, a7, a8, b2 &$12$ \\
\hline
Abell 383 & 7 & 4 & 2 & a5, b1, b4 & a7, b2, c2 & a1, a2, a5, a6, b4, c1 & $3$ \\
\hline
Abell 773 & 1 & 1 & 1& --- & a1 & c1 & $13$ \\
\hline
Abell 963 & 7 & 3 & 1 & --- & a1, a3, a6 & a2, a5, a7, b2, b3 & $15$ \\
\hline \hline
\end {tabular}
\end {table*}
\end{center}
As a comparison, we point out that the arc finder run presented in \citet{2005Horesh} found 16 arcs in the four clusters under consideration. Restricting to the areas where we denoted arcs for our truth table (see Figs. \ref{A68_and_A383} and \ref{A773_and_A963}) these authors found a total of $9$ arcs, $5$ of which are in common with the sample of 16 arcs detected with the AMA, and $3$ are blended in our detections.
On the other hand, only objects with $L/W > 7$ are selected by \citet{2005Horesh}, while we make no cuts in this quantity. In any case, the focus in \citet{2005Horesh} is not on completeness, but on a comparison between a real and a simulated arc sample.
A more detailed comparison of the AMA with this and other arc finders is outside the scope of this paper
\citep[see][for preliminary results]{2015mgm..conf.2088D}.
As for the false detection fraction $f$, it is computed as the ratio of the false positives
to the total number of nonarcs given as input to the ANN.
For the configuration in Table \ref{resultsreal} we have $f \sim 15\%$. With respect to the total number of objects detected, the fraction of false detections is $3\%$.
The false positives are objects classified as arcs, but that are not on the truth table. In principle, some of these objects could be real arcs that are missed by visual inspection. However, we did look at all false positives and only two of them could be associated with arcs; they were in fact pieces of arcs with multiple peaks, other pieces of which have been identified by the arc finder. Thus these cases are negligible for the purposes of this paper.
In Fig. \ref{cfHST} we show $c$ and $f$ as a function of the threshold $t$ for configurations A and D combining the arc detections for the four cluster images considered in this section. Poisson error bars are indicated. We see that the ANN achieves a reasonable completeness, $\sim 70\%$ for configuration D, and $\sim 50\%$ for A, for a low threshold ($t= -0.75$), while still keeping a low contamination rate, $\sim 15\%$ and $\sim 9\% $ for configurations D and A, respectively.
As in the simulations, configurations A and D yield values of $c$ that are compatible with each other in the whole range of $t$, taking their Poisson errors into account.
However, $c$ differs considerably between the real and simulated data. The completeness is significantly lower on the data and decreases more abruptly with the threshold than what was seen in the simulations.
We interpret this result as an indication that the simulations are not realistic enough for a proper quantitative comparison with the real data. In fact,
the simulated arcs are very diverse, but do not include some relevant degrees of realism, such as surface brightness variations in the sources and lenses with substructure. In addition, the real arcs are often close to bright galaxy cluster members, which can affect their segmentation and deblending (both due to contamination from the galaxies and background misestimation on the crowded field) and thus affect their shape. The strong effect of blending with other objects is already accounted for in our comparisons in the sense that these objects are removed from the denominator of $c$. However, a less significant contamination from close-by objects or a breaking into smaller objects affects the Mediatrix measurements and therefore the arc detection with the ANN, which was trained in simulations that do not include these effects.
In the validation $c$ is a bit higher for A than D in the whole range of $t$.
On the other hand, in the real data there is an apparent trend for D to have a higher completeness than configuration A for the smallest threshold. For the remaining interval the performance is very similar among configurations A and D.
\begin{figure}[!ht]
\begin{center}
\sidecaption
\resizebox{\hsize}{!}{\includegraphics{RealArcsThresholdAnalisys.png}}
\caption{\label{cfHST} Completeness $c$ and fraction of false detections $f$, including Poisson error bars, for the arcs in the 4 selected HST Abell clusters, for configurations A and D, as a function of threshold.
Data points for configuration A are shifted horizontally for clarity.}
\end{center}
\end{figure}
Regarding $f$, the results are also compatible between A and D within the error bars, and the difference is again higher for $t=-0.75$. The false positive fraction drops substantially for higher $t$, reaching $f \lesssim 5\%$ for $t>-0.25$.
Comparing the false positive fraction obtained during the validation process with those from the runs on the HST images (bottom pannels of Figs. \ref{thrgra} and \ref{cfHST}), we see that the results are consistent for configuration A in the whole range of $t$.
However, for configuration D and $t<0$ the false detection fraction is clearly higher in the validation than in the real data. The highest discrepancy occurs for $t = -0.75$, but is smaller than twice the standard deviation within the training plus validation samples.
This difference could be just a consequence of the difference in $f$ between A and D pointed out in the previous section or may be a difference in the behavior of $f$ in the real data with respect to the validation.
The agreement between the validation set and the real data for $f$ is expected;
the NAG is obtained from the same set of images from WFPC2 for the training plus validation, although the CCDs are different and include other clusters. Thus we would expect a similar behavior between the bottom panels of Figs. \ref{thrgra} and \ref{cfHST} for $f$. This is indeed the case for configuration A and strengthens the case for a fluctuation in the false positives obtained in the simulations for configuration D.
\section{Discussion} \label{discusao}
The purpose of the paper is to present the AMA%
\footnote{The method is implemented in the python language and the source code is available upon request to the authors.
The training sample is also available upon request.}
and provide a simple example of application, first illustrating the training and validation process on simulated arcs and then the application to a real, albeit small, data set.
The major novel aspects of the present work are the use of the Mediatrix method in step $3$ of the arc finding process and the use of simulations. The simulated arcs are used not only to train, but also to find a good set of ANN configurations for step $4$.
There is room for many improvements in the processes described in this paper, most notably in the use of more realistic simulations and increasing the number of systems in the application to real data, but also in other aspects arc detection.
On the object identification
and segmentation side, betterments can be implemented in order to detect faint sources, avoid the breaking of large arcs, improve on the deblending with nearby objects, and to find and segment sources in high background regions, which are known issues for arc identification.
Although \texttt{SExtractor} does not deal with all these issues in an optimal way for arc detection, it has been used in many arc finders for their object identification to some degree \citep[e.g.,][]{2005Horesh,2007Estrada,Kubo2008,Marshall2009robot,JosephPCA2014, Maturi2014}.
A better performance than running \texttt{SExtractor} in a straight way, as in the current work,
has been obtained by carrying out multiple runs of this code with different thresholds \citep{2005Horesh}, or by using \texttt{SExtractor} alone for pixel thresholding \citep{Kubo2008}.
Other arc finders use different approaches for object detection and segmentation, which are specifically oriented toward identifying arcs \citep[e.g.,][]{2004Lenzen,2006Alard,2007Seidel,2012More,2016ApJ...817...85X}.
A key issue for detection and segmentation for arcs is that these objects are often embedded in the haloes of bright galaxies (especially for galaxy-scale arcs and radial arcs) or blended with foreground galaxies (especially for arcs in clusters). One approach that has been implemented to address this issue is to fit and subtract the light profile of galaxies, as in \citet{2005ApJ...627...32S,Brault2015}. Several
codes have been proposed in the literature to this end \citep[e.g.,][]{PyMorph2010,GALPHAT2011,GALAPAGOS2012}, often running \texttt{galfit}~\citep{galfit32010} recursively to fit each galaxy by a combination of elliptical brightness distributions with \citet{sersic} profiles. Advanced versions of \texttt{SExtractor} also fit and subtract all identified objects in a field \citep[e.g.,][Moraes et al., in prep.]{SEMF2012ApJ...757...83D,SEMF2015A&A...578A..79D}. Other schemes to subtract objects from images, which could be useful for arc finding have also been proposed \citep[e.g.,][]{JimenezTeja2012}.
A different approach has been carried out by \citet{2016ApJ...817...85X}, who propose a new detection and segmentation scheme, working in intensity difference space, which performs well in bright halos without the need of subtraction.
Regarding the preselection, for applications to wide-field surveys this phase must also include the removal of image artifacts such as satellite tracks, star spikes, and regions with a large amount of noise or with a steeply varying background. In the current example we remove noisy regions by simply cutting off objects that are close to the CCD borders. However the AMA already includes a proper handling of survey masks, which are produced to avoid bright star halos and spikes, satellites, and other image features. The approach of \citet{2016ApJ...817...85X} is also well suited to remove diffraction spikes without the need to use masks and may be useful for less bright nonmasked stars whose spikes could contaminate the arc detection.
For object measurement we propose the use of the Mediatrix filamentation method, and several parameters derived from it, as it was designed for elongated and curved objects. Most arc finders end up using less parameters and simpler measurement schemes to characterize the arc candidates, such as $L$ and $L/W$,
and only in a few cases include estimates of the curvature \citep{2007Estrada,Kubo2008}. However, other sets of inputs in addition to the Mediatrix inputs could be given to the ANN, such as higher order moments of the brightness distribution, including the arcness \citep{Kubo2008}.
We argue that by using a machine learning algorithm for the final candidate selection (in this case a back-propagation ANN) one may achieve a better efficiency in finding arcs than using hard cuts on a few variables. By working on a multidimensional parameter space, it is possible to deal with correlations among the variables and empirically obtain combinations that represent gravitational arcs. For example, arcs tend to be more curved and smaller at galaxy scales than in massive clusters, such that a single cut in $L/W$ or arcness would not be optimal for finding arcs at both scales.
Artificial neural networks were first used by \citet{2007Estrada} to search for arcs. In their case the simulated arcs are simply sections of a circle with a
surface brightness profile that is uniform along the tangential direction and is convolved with a Gaussian with FWHM similar to the typical seeing of the images. The ANN is trained using a hundred such simulated arcs covering a range of sizes and brightnesses, which are added to SDSS images.
The objects are also identified with \texttt{SExtractor} and the preselection is also carried out using an estimate of the object's elongation. Finally, four inputs are given to the ANN, based on a fit of the object by a circle and on a determination of the object's length (using its furthest pixels).
\citet{2007Estrada} study the efficiency for recovering the simulated arcs both for a visual inspection and for the automated process as a function of peak surface brightness and $L/W$.
A maximum efficiency of $40\%$ (with respect to the number of simulated arcs) is achieved in the automated search. The present work can be seen as an improvement on \citet{2007Estrada} in the sense that we use more realistic simulated arcs and a wider set of input measurements
well suited to characterize the arcs, in addition to tuning the ANN configurations to improve the completeness.
Of course the key to a good performance in a learning algorithm is the realism of the training sample. Many improvements can be incorporated into the simple \texttt{AddArcs} simulations described in this work, such as considering more realistic lenses \citep[e.g.,][]{2011MNRAS.418...54H,2016ApJ...817...85X} and sources \citep{Kubo2008,Marshall2009robot, 2011MNRAS.418...54H}.
Moreover, in addition to having a realistic sample of isolated arcs, those have to be added to real images, for example, to address the issue of blending with other sources and of embedding in the halo of bright galaxies in cluster cores.
Other works have used simulations to test arc finders, define their parameters, or train their methods, and in some cases determine the selection functions \citep[e.g.,][]{2005Horesh,Kubo2008,Marshall2009robot, 2011MNRAS.418...54H, RINGFINDER,JosephPCA2014,Brault2015,2016ApJ...817...85X}.
Another possibility is to use the growing number of strong lensing systems detected in wide-field surveys and HST images to perform the training on real data sets. For example, over 600 candidate systems have been detected in recent studies using CFHTLS data
\citep{Maturi2014,RINGFINDER,Brault2015,SPACEWARPSII,ParaficzCFHTLS}, which could be used to train and better characterize the AMA and other arc finders.
By training and validating the ANN with more realistic simulated arcs or with real data, we expect to
reach a better agreement for $c$ in comparison to applications to other data sets (and therefore to achieve a higher completeness), which is different from what we found when applying our trained ANN to the HST data.
In general, the several arc finders proposed in the literature carry out an end-to-end approach from the science image to a list of arc candidates, implementing all four steps that we refer to in this paper. However, they have their own solutions for each step with different degrees of sophistication and specificity for finding arcs. For example, in this work we focus on the third and fourth steps, respectively, by using a set of object measurement parameters that are well suited to arcs and a trained ANN, while most methods use simple cuts on a few parameters for the final classification. On the other hand, we use a generic object segmentation code that is not optimal for arcs.
For most methods, these four steps could be performed interchangeably. Therefore, if they are presented in a modular way, we would be able to test the performance of each step independently.
Moreover, new arc finders could be created
combining the solutions for each step from those available that work better in specific situations.
Several possible concatenations of the arc finder modules could also be compared using their end-to-end performance.
Given that about a dozen finders have already been proposed, and that we are on the verge of applying them on large data sets, it would be very useful to carry out such a comparison.
This is one of the avenues we will pursue for future work
\citep[see,][for preliminary results with a few arc finders]{2015mgm..conf.2088D}.
After a decade of progress in the development of gravitational arc finders, several codes
are ready for exploring the new generation of wide-field surveys in the quest for gravitational arcs.
However more progress is still needed for fully automated runs so as to produce samples that can be readily exploited for their applications.
Besides improving the efficiency in some situations, the most important is to limit the false positives to a level low enough to be corrected for and certainly less than the number of real systems, as thousands to hundreds of thousands strong lenses are expected
in the forthcoming data.
Different strategies have been proposed and implemented to address these issues.
Combining aspects of these solutions, which are more suited to each step of arc detection, seems a natural way to proceed. We believe that using neural networks or other machine learning methods may provide an important contribution to the task of selecting more complete and pure samples of gravitational arcs for a broad range of deflector scales and backgrounds.
\begin{acknowledgements}
C.~R.~Bom is partially funded by the Brazilian agency CNPq of the Ministry of Science and Technology and Innovation (MCTI).
M.~Makler is partially supported by CNPq (grant 312353/2015-4) and FAPERJ (grant E-26/110.516/2-2012).
C.~H.~Brandt is supported by the Brazilian agency CAPES, grant 15113/13-2 under the CAPES/ICRAnet agreement.
We thank J. Estrada, H. Lin, C. Furlanetto, B. Rossetto, A. More, B. Santiago,
and the DES Strong Lensing Working Group members for useful comments along the development of this work.
This paper makes use of observations made with the NASA/ESA Hubble Space Telescope, obtained from the ESO/ST-ECF Science Archive Facility.
\end{acknowledgements}
\bibliographystyle{aa}
| f1b4f07909b994657851c9ef78989b16a5985a47 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section*{\bf $\Psi_k$ Scientific Highlight Of The Month}
\newcommand{\ensuremath\mathrm{d}}{\ensuremath\mathrm{d}}
\begin{center}
{\large \bf Towards a Common Format for Computational Material Science Data}
{Luca M. Ghiringhelli$^{1,}$\footnote{ghiringhelli@fhi-berlin.mpg.de},
Christian Carbogno$^{1}$,
Sergey Levchenko$^{1}$,
Fawzi Mohamed$^{1}$,
Georg Huhs$^{2,3}$,
Martin L\"{u}ders$^{4}$,
Micael Oliveira$^{5,6,}$\footnote{mjt.oliveira@ulg.ac.be},
and Matthias Scheffler$^{1,7}
}
{\small $^1$Fritz Haber Institute of the Max Planck Society, Berlin, Germany}\\
{\small $^2$Barcelona Supercomputing Center, Spain}
{\small $^3$Humboldt-Universit\"{a}t zu Berlin}\\
{\small $^4$Daresbury Laboratory, UK}
{\small $^5$University of Li\`{e}ge, Belgium}\\
{\small $^6$Max Planck Institute for the Structure and Dynamics of Matter, Hamburg, Germany}\\
{\small $^7$Department of Chemistry and Biochemistry and Materials Department, University of California--Santa Barbara, Santa Barbara, CA 93106-5050, USA}\\
\end{center}
\begin{abstract}
Information and data exchange is an important aspect of scientific progress. In computational materials science, a prerequisite for smooth data exchange is standardization, which means using agreed conventions for, e.g., units, zero base lines, and file formats. There are two main strategies to achieve this goal.
One accepts the heterogeneous nature of the community which comprises scientists from physics, chemistry, bio-physics, and materials science, by complying with the diverse ecosystem of computer codes and thus develops ``converters'' for the input and output files of all important codes. These converters then translate the data of all important codes into a standardized, code-independent format.
The other strategy is to provide standardized open libraries that code developers can adopt for shaping their inputs, outputs, and restart files, directly into the same code-independent format. We like to emphasize in this paper that these two strategies can and should be regarded as complementary, if not even synergetic.\\
The main concepts and software developments of both strategies are very much identical, and, obviously, both approaches should give the same final result.
In this paper, we present the appropriate format and conventions that were agreed upon by two teams, the Electronic Structure Library (ESL) of CECAM and the NOMAD (NOvel MAterials Discovery) Laboratory, a European Centre of Excellence (CoE). This discussion includes also the definition of hierarchical metadata describing state-of-the-art electronic-structure calculations.
\end{abstract}
\newpage
\section{Introduction}
To aid and guide the search for new and improved materials, computational materials science is increasingly employing
``high-throughput screening'' calculations \cite{Ceder09,Wood09,Wolverton10,Ceder11,Rajan11,Jakobsen12,Curtarolo13}.
In practice, this means that computational material scientists produce a huge amount of materials data on their local workstations, computer clusters, and supercomputers using a variety of very different computer programs, in this domain usually called ``codes''. Though being extremely valuable, this information is hardly available to the community, since most of the data is stored locally or even deleted right away. In publications, typically, only a small subset of the calculated results is reported, namely that which is directly relevant for the very specific topic addressed in the actual manuscript. Although several repositories have been created and maintained in the past for domain-specific applications, these typically do not store the full inputs and outputs of all calculations. And, typically, they are restricted to results obtained by just one or very few codes. Since 2014, this situation is changing: the NOMAD Repository (\texttt{https://NOMAD-Repository.eu}) \footnote{The NOMAD Repository is a joint activity of the groups of Matthias Scheffler (FHI), Claudia Draxl (HU) and the Max Planck Computing and Data Facility: \texttt{https://NOMAD-Repository.eu}. Data is stored for at least 10 years and a DOI is provided so that data is citable. \texttt{http://NOMAD-repository.eu} is the only repository for materials science accepted by Nature Scientific Data (\texttt{http://www.nature.com/sdata/})} is designed to meet the increasing demand for storing scientific data and making them available to the community. It is a unique facility, as it accepts (and requests) input and output files of all important computer codes used in computational materials science, and it guarantees to store these data for at least 10 years.\\
Currently, the repository holds the information of over 3 million calculations which corresponds to more than a billion CPU-core hours. While this has been and is useful for its purpose of data sharing via a repository (enabling the confirmatory analysis of materials data, their reuse, and repurposing), the data is very heterogeneous, just as they are provided by the different codes.
Thus, they are not useful for data analytics and extensive comparisons, because the input and output files of different codes typically use different units, representations, and file formats. \\
Advancing the sharing and comparison of such data is a pressing issue that needs to be addressed, as exemplified in recent initiatives.
For instance, the NOMAD Laboratory CoE \cite{NOMAD} is presently building a unified, code-independent database, so that big-data analytics techniques can be applied to obtain unprecedented insight from the vast amount of already existing calculations. \\
In a similar spirit, the E-CAM CoE \cite{ECAM}, which was recently established by CECAM to build an e-infrastructure for software, training, and consultancy in simulation and modeling, is committed to actively support the development and adoption of software libraries and standards within the electronic structure community. One measure aiming at this is CECAM's Electronic Structure Library (ESL) initiative \cite{ESL}, which includes in its plans the development of an Electronic Structure Common Data Format (ESCDF). The ESCDF provides a standardized data format and an Application Programming Interface (API) that every code can use.
Code interoperability is also strongly driven by communities, using the results for further analysis, such as spectroscopy calculations, based on converged and relaxed ground state results. These applications have been at the heart of the European Theoretical Spectroscopy Facility (ETSF) \cite{Gonze08} activities, and are now a central part of the efforts of the COST network EUSpec \cite{EUSPEC}.\\
As mentioned above, the $\Psi_k$ community is using many different computer codes.
A list of the most important electronic-structure codes in the materials science community is given in Table 1. The codes are ranked by the number of citations of the last five years and their main features are summarized.
\begin{table}[b!]
\footnotesize
\begin{tabular}{l|l|r|l|l|l}
Code & Search Name & Citations & Description & License\tabularnewline
& & (2011-15) & & \tabularnewline
Gaussian & Frisch \cite{notetable}& 18\,700 & AE, PSP, GTO & C\tabularnewline
VASP & Kresse & 16\,500 & PSP, PW, PAW & C(O)$^\textrm{(a)}$\tabularnewline
CASTEP & Payne & 5\,970 & PSP, PW, PAW & C(O)$^\textrm{(b)}$\tabularnewline
GAMESS & Gordon & 5\,360 & AE, PSP, GTO & F\tabularnewline
WIEN2k & Blaha & 4\,880 & AE, (L)APW & C\tabularnewline
Quantum ESPRESSO & Giannozzi \cite{notetable} & 4\,810 & PSP, PW & G\tabularnewline
Molpro & Werner & 4\,190 & AE, PSP, GTO & C\tabularnewline
SIESTA & Soler & 4\,040 & PSP, NAO & G\tabularnewline
TURBOMOLE & Ahlrichs & 3\,730 & AE, PSP, GTO & C\tabularnewline
ADF & Baerends & 2\,860 & AE, STO & C\tabularnewline
DMol$^3$ & Delley \cite{notetable}& 2\,893 & AE, PSP, NAO & C\tabularnewline
ORCA & Neese & 2\,540 & AE, PSP, GTO & F\tabularnewline
CRYSTAL & Dovesi \cite{notetable}& 2\,225 & AE, PSP, GTO & C$^\textrm{(b,c)}$\tabularnewline
MOPAC & Stewart & 2\,070 & PSP, STO & F(O)$^\textrm{(d)}$\tabularnewline
ABINIT & Gonze & 1\,870 & PSP, PW, PAW, WLT & G\tabularnewline
Q-Chem & Shao \cite{notetable} & 1\,740& AE, PSP, GTO & C\tabularnewline
Jaguar & {\em Schr\"{o}dinger} & 1\,670 & AE, PSP, GTO, STO & C\tabularnewline
Dalton & \AA{}gren & 1\,580 & PSP, GTO & F\tabularnewline
NWChem & Valiev & 1\,410 & AE, PSP, GTO, PW, PAW & G\tabularnewline
MOLCAS & Lindh & 1\,200 & AE, PSP, GTO & C$^\textrm{(e)}$\tabularnewline
CP2K & Hutter \cite{notetable}& 1\,190 & AE, PSP, GTO, PW & G\tabularnewline
ACES III & Bartlett \cite{notetable} & 1\,150 & PSP, GTO & G\tabularnewline
CPMD & Hutter & 938 & PW, PSP & F\tabularnewline
TB-LMTO-ASA & Andersen \cite{notetable}& 840 & AE, LMTO, $N$MTO & G\tabularnewline
octopus & Rubio & 792 & PSP, RS & G\tabularnewline
CFOUR & Stanton \cite{notetable}& 738 & AE, PSP, GTO & F\tabularnewline
GPAW & Mortensen & 624 & PSP, PW, PAW, RS, NAO & G\tabularnewline
DIRAC & Saue \cite{notetable}& 579 & AE, LAO & F(O)\tabularnewline
CASINO & Needs \cite{notetable}& 475 & AE, PSP, PW, GTO, STO, NAO & F$^\textrm{(d)}$\tabularnewline
FPLO & Koepernik & 459 & AE, NAO & C\tabularnewline
FHI-aims & Blum & 407 & AE, NAO & C(O)\tabularnewline
OpenMX & Ozaki & 359 & PSP, NAO & G\tabularnewline
COLUMBUS & Lischka & 333 & AE, GTO & F\tabularnewline
Smeagol & Lambert & 302 & PSP, NAO & F\tabularnewline
Elk & Dewhurst \cite{notetable}& 225 & AE, LAPW & G\tabularnewline
\end{tabular}
\end{table}
\begin{table}[t!]
\footnotesize
\begin{tabular}{l|l|r|l|l|l}
Code & Name & Citations & Description & License\tabularnewline
& & (2011-15) & & \tabularnewline
Yambo & Marini & 220 & PSP, PW & G\tabularnewline
FLEUR & Bl\"{u}gel & 212 & AE, LAPW & F(O)\tabularnewline
ONETEP & Skylaris & 204 & PSP, PW/RS, PAW, NAO & C(O)\tabularnewline
CONQUEST O(N) & Bowler \cite{notetable}& 201 & AE, PSP, NAO, PW & I\tabularnewline
Psi4 & Sherrill & 187 & AE, GTO & F(O)\tabularnewline
TeraChem & Martinez \cite{notetable}& 175 & AE, PSP, GTO & C\tabularnewline
exciting & Draxl \cite{notetable}& 171 & AE, LAPW & G\tabularnewline
BigDFT & Genovese & 141 & PSP, WLT & G\tabularnewline
BerkeleyGW & Louie & 140 & AE, PSP, PW, NAO, RS & F(O)\tabularnewline
PARSEC & Saad & 135 & PSP, RS & G\tabularnewline
ATOMPAW & Holzwarth & 105 & AE, PSP, PW, PAW & F(O)\tabularnewline
\end{tabular}
\caption{\footnotesize List of the electronic-structure codes with more than 100 citations in the 2011-2015 period. The number of citations is determined via Google Scholar, with a search performed in April 2016, by searching for the name of the code (as reported in the first column) {\em and} the name of one of the main developers (or {\em company} that develops and commercializes the product, as reported in the second column). The reason of the second search criterion is that the name of several codes have different meanings. We found that the combination of the two column gave very few, if any, false positive results. Nonetheless, in some cases we had to further adjust the search criterion to eliminate the false positive results; the actual search strings are listed in note \cite{notetable}. The significance of the citation numbers should not be overrated (as in any citation analysis). For example, young codes that were only developed in the last 7-8 years may have a high gradient of their employment in the community but still have a rather low citation index in the table. Nevertheless, we believe that the general impression provided by the table is correct. The fourth column lists the main features of each code and the fifth the type of license. A list including also force-field based codes is maintained at: \texttt{https://www.nomad-coe.eu/index.php?page=codes}. The meaning of the abbreviations is the following. For ``Description'', AE: All Electron, PW: Plane Wave, GTO: Gaussian-Type atomic Orbitals, NAO: Numeric Atomic Orbitals, STO: Slater-Type Orbitals, LAPW: Linearized Augmented Plane Wave, PAW: Projector Augmented wave, RS: Real Space, PSP: pseudopotential (including also ECP, effective core potential), LAO: London Atomic Orbitals, LMTO/$N$MTO: (Linear) Muffin-Tin Orbitals / $N^\textrm{th}$-order MTO, WLT: Wavelet; for ``License'', G: GPL or LGPL (includes also GPL compatible licenses, such as Educational Community License or Apache 2), F: Free (other licenses), C: Commercial / charged (usually, with a smaller cost for academics compared to non-academics), (O): Open source, I: Individual-basis license (via contacting the authors), $^\textrm{(a)}$: Free for academic use in Austria, $^\textrm{(b)}$: Free for academic use in the UK, $^\textrm{(c)}$: Free demo serial version (max 20 atoms/cell), $^\textrm{(d)}$: Commercial for non-academic use, $^\textrm{(e)}$: version 8.2 will have license G.}
\end{table}
In this highlight paper, we will discuss the challenges of electronic-structure codes, but note that the NOMAD Laboratory CoE also includes force-field based codes. The various
electronic-structure codes differ by many conceptual and numerical aspects, first of all by the type of basis sets that are employed.
Besides, some codes describe all electrons, others use a simplified description of the core electrons. On one side, this results in very different data representations; on the other side, it is not trivial to compare certain quantities like energies and wavefunctions.\\
In the remainder of this paper, we will define a common code-independent representation for all relevant quantities (e.g., structure, energy, electronic wave functions, trajectories of the atoms, etc.).
The ideas and concepts presented here are also the result of discussions that took place at a CECAM/Psi-k workshop, held in Lausanne in January 2016, that was attended by experts and key developers of more than 20 of the most important electronic structure codes.
\section{The conversion layer}
In this section, we discuss the key issues for converting the information present in the inputs and outputs of electronic-structure codes into a common format. More specifically, this discussion addresses: i) The metadata infrastructure for storage and retrieval of the code-independent quantities; ii) the uniform file format as defined by the ESCDF team; iii) the zero-level reference for energy-related quantities; iv) the representation for the electronic and vibrational band structures and density of states; v) the compact representation of scalar fields such as wave functions, electron densities, and exchange-correlation potentials; vi) the unified representation of quantities related to excited-state calculations ($GW$, Bethe-Salpeter).
Furthermore, we discuss the general challenge, touching all the above points, of establishing error bars and confidence levels, with respect to the adopted numerical settings, for each stored/retrieved calculation.
\subsection{Metadata for the code-independent format}
Metadata is the name or label that characterizes corresponding values.
For example ``XC\_method'' is a metadata name and ``LDA'' may be the corresponding value. Thus, if one thinks of storing data as {\em key}--{\em value} pairs (as in a dictionary), the {\em key} is the metadata. \\
There are several examples of metadata approaches in computational materials science, the most prominent being ChemML, the Chemical Markup Language \cite{CML}, CIF, the Crystallography Information File \cite{CIF}, code specific implementations such as the VASP \cite{VASP} or Molpro \cite{MOLPRO} XML outputs, or very simple and intuitive file formats like XYZ \cite{XYZ} for the storage of the configuration of a system (atomic species, coordinates, also allowing for comments).\\
In general, a metadata structure is built {\em a priori}, by starting from a list of names that identify the needed concepts and quantities. Code outputs and file formats are then designed to reflect the {\em a priori} metadata structure. In principle, {\em a priori} metadata approaches aim to be as exhaustive as possible; in practice, they are typically designed with a specific scientific field, application and/or code in mind, as the examples above show. Conversely, the ``NOMAD Meta Info'' \cite{metainfo}, i.e., the metadata defined and used for the NOMAD code-independent data format, is defined {\em a posteriori}, by starting from the existing inputs and outputs of many different codes stemming from different scientific fields and thus with quite diverse feature sets.
From a computer science point of view, the {\em a posteriori} approach is much more challenging than the {\em a priori} one, since it
requires a more flexible, extensive, and generalized infrastructure~\cite{metaweb}: On the one hand, the diverse quantities stored by each code need to find the proper place in the global hierarchy. On the other hand, not all codes' inputs and outputs contain all the defined metadata; therefore, having unassigned metadata when sorting a given input/output in the code-independent format must not create problems. From a physics, chemistry, and materials science point of view, the {\em a posteriori} approach has the critical advantage that essentially all the information contained in any input/output file of any considered code is recognized and stored into the properly identified {\em key}--{\em value} pair. Intrinsically, this guarantees a truly exhaustive coverage of all properties, even those that currently might not appear of interest. In turn, this lays the founding for authentic big data mining approaches that unveil hitherto unexpected correlation and relationships.\\
In NOMAD Meta Info \cite{metainfo}, the {\em key} is not a simple string but a more structured object, with several attributes. The {\em name} is an attribute, a string that must be unique, well defined, intuitive and as short as possible\footnote{As the {\em names} will be the string frequently present in parser scripts when reading inputs/outputs and assigning a value to a given metadata, as well as used in query scripts in order to locate a particular metadata, it is convenient to keep these strings as short as possible.}. It is used to identify the metadata and therefore used to associate {\em values} with their metadata.
The {\em description} is another attribute and contains an extensive human-readable text that clarifies the meaning of the metadata. \\
Another important attribute of a metadata is its {\em type}. Concrete {\em values}, scalars, strings, or (multidimensional) arrays, that are extracted by the parsers that read input and output files, have an associated {\em concrete-type} metadata.
These {\em values} are organized in groups to which {\em section-type} metadata are associated. In computational informatics, these {\em sections} would correspond to tables of a relational database model \footnote{In a relational database, data are organized in tables, where rows represent instances of some entity (e.g., a customer, a product) and the columns values attributed to that instance (e.g., the address, the price). Rows are identified by unique keys.}, where the {\em values} would be the rows, and the {\em concrete} metadata the columns.\\
By considering the different steps that define an electronic-structure calculation run (i.e., from the invocation of the code to the completion of the task described in the input files, or the interruption due to several reasons), the following main {\em section-type} metadata are defined:
\begin{itemize}[noitemsep,nolistsep]
\item ``section\_run'': represents a single run of the program,
\item ``section\_method'': contains the information defining the theory level and convergence parameters,
\item ``section\_system'': contains the specifics of the system configuration (atom species, coordinates, lattice vectors),
\item ``section\_single\_configuration\_calculation'': contains the results for a physical system as defined in a single ``section\_method'' and a single ``section\_system'',
\item ``section\_scf\_iteration'': contains the results of a single self-consistency iteration.
\end{itemize}
The chosen hierarchy reflects that {\em sections} can be nested, meaning that each outer one can contain one or more inner {\em sections},~e.g.,~each ``section\_single\_configuration\_calculation'' typically contains multiple iterations of ``section\_scf\_iteration''. Also, the different {\em sections} depend on each other,~e.g.,~the outcome in ``section\_single\_configuration\_calculation'' depends on the system, method, and program defined in the higher layers of the structure.
\begin{figure}[h!]
\begin{minipage}[c]{0.70\textwidth}
\includegraphics[width=0.9\textwidth,clip]{metadata_concept_sectioning_only_highres.png}
\end{minipage}\hfill
\begin{minipage}[c]{0.30\textwidth}
\caption{
A simplified metadata structure, according to NOMAD Meta Info \cite{metainfo}, for a simple electronic-structure calculation is shown.
The metadata indicated in black are of {\em concrete} type. The {\em names} are self explaining (the full {\em description} is given in \cite{metaweb}). The metadata indicated in red are of type {\em section} and the content of the {\em section} defined by them is inside the red box. {\em Sections} can also have references, indicated by the dashed arrows, to other {\em sections} besides the parent {\em section}.
The layout of the metadata in the figure is reminiscent of the JSON file, that is one of the file formats in which they are effectively stored. } \label{metafig}
\end{minipage}
\end{figure}
The power and flexibility of this approach is schematically displayed in Fig. \ref{metafig}: We show the (simplified) representation following the metadata structure (according to NOMAD Meta Info \cite{metainfo}) of a simple calculation (a ``section\_run'') where one structure (defined in ``section\_system'') is evaluated with one electronic-structure method (defined in ``section\_method''). The results are reported in a single ``section\_single\_configuration\_calculation'', that in this example contains, besides the final results, also results from various scf iterations (``section\_scf\_iteration''). The metadata of {\em concrete} type (black font) have also {\em units} (in NOMAD Meta Info, we always use SI units) as attribute. In practice, ``energy\_total'' could have {\em value} ``$-1.344 \cdot 10^{-20}$'', with {\em units} ``J'', for joule. {\em Concrete} metadata with associated values are contained in {\em sections} (red font) and {\em sections} can also be contained in parent {\em sections}, as graphically indicated by the boxes. {\em Sections} can also have references (dashed arrows) to other {\em sections}. In the present case, this is needed to relate the results contained in ``section\_single\_configuration\_calculation'' with the physical model used to obtained them (``section\_method'') and the geometrical structure used as input (``section\_system''). In more complex cases, several ``section\_single\_configuration\_calculation'' can have references to the same ``section\_method'' (e.g., a geometry optimization) and/or to the same ``section\_system'' (e.g. the same system calculated with a self-consistent electronic-structure method which is used as a starting point for a many-body perturbation method), see \cite{metadata}.
The standard definition of NOMAD Meta Info is maintained in a git repository \cite{metaweb} and contributions are welcome. The current metadata structure can be browsed at \cite{metainfo}.\\
NOMAD Meta Info \cite{metainfo} is kept independent of the actual storage format as much as possible; it is a conceptual model, and in principle not bound to any specific storage method.
In previous efforts, the metadata structure was normally seen as keys without any structure, consequently the relationships between various key--value pairs was expressed using the means of the specific storage format that was chosen.
To actually make the data usable, we do express the structure of the values in the metadata using sections and references, and define the types used to store a concrete value.
We knowingly limited the allowed types and the kind of structuring we support, to strike a balance between being able to represent any quantity in a logical way, and being able to use this representation in multiple places and formats.
Some things might be represented slightly better for example in JSON, XML, HDF5, CIF,... using some special features of each specific format, but then it is difficult to transfer those features in a natural way to another representation, especially if one does not know automatically which structuring of a specific format should be preserved, and what can be ignored with little loss.\\
We feel that a simplified view supporting multiple representations allows one to choose different formats depending on what needs to be optimized in a specific application, and simplifies what the user has to understand.
We officially support two formats to store data using the NOMAD Meta Info: the human readable JSON and HDF5, a binary format that can store efficiently large arrays and higher dimensional objects.
This covers already quite different trade-offs with respect to human readability and storage efficiency, and has influenced and validated the choices done in the NOMAD Meta Info.
The data could also be represented in other formats (CIF, XML, Relational Databases,...), and an official mapping to some of them could be added in the future.\\
More details, including details on the practical implementation are given in Refs. \cite{metaweb} and \cite{metadata}.
\subsection{The electronic structure common data format (ESCDF).}
Whenever code developers are willing to design their inputs and outputs in a standardized format, the ESCDF library will
provide the necessary common ground.\\
In particular, the ESCDF is focused on giving tools for developers to save the data produced by their code for the
purpose of restarting a calculation, sharing the data with other codes, or further processing.
This is achieved by standardizing the quantities to be put in an output file and via adopting self-describing formats
like HDF5 or NetCDF, which are extendable and allow the inclusion of metadata needed to interpret them.
The synergy between the NOMAD Laboratory CoE and ESCDF lies in the fact that, in the conversion layer, the NOMAD Laboratory will use
the ESCDF file format, while the ESCDF will use the NOMAD metadata.
For a given code to be able to use the information produced by another code, the mere capability of reading it from a file is not sufficient.
Indeed, as explained in the Introduction, the various electronic-structure codes use very different data representations,
of which the type of basis set is the most paradigmatic example. This means that, after reading the data, it might be necessary to
perform some conversions. Some are quite straightforward, like changing units, but other conversions can be quite involved,
either requiring complex algorithms and/or be computationally demanding. For example, to be able to use wavefunctions that were written
as a linear combination of Gaussian-type atomic orbitals, a plane wave code would have to perform a change of basis.
Such transformations are beyond the scope of the ESCDF, but are an essential component of the NOMAD converters.
From here the synergies between the two projects become clear: the ESCDF provides the tools for a standardized access
to the data stored within the files, while the NOMAD conversion layer converts the data to a code-independent representation.\\
The first version of the ESCDF will include specifications to read/write the following type of data: geometry/structure
of the system, basis sets, densities, potentials, and wavefunctions. At this point, the specifications do not aim at covering
exhaustively all the quantities that an electronic code might need to read/write. Instead, the focus is on making sure the specifications
are flexible and extendable. Because of its hierarchical structure and greater
flexibility, HDF5 was chosen over NetCDF as the underlying file format. Each type of data is stored in an HDF5 group, which
can be arranged in a way that is similar to a file system. This allows to store data for different use cases. For example,
the most common use case is performing one calculation for one system, but the format should also be able deal with several
systems that are calculated simultaneously. Using the NOMAD metadata and the experience accumulated by a previous standardization
effort led by the European Theoretical Spectroscopy Facility (ETSF)~\cite{Gonze08}, a set of specifications have already been agreed for the
geometry/structure and work is currently underway regarding the basis sets and scalar fields.\\
The associated software library and corresponding API will focus on flexibility, extensibility, and performance in order to maximize its usefulness and adoption by the
community of code developers. In particular, we aim at providing an API that does not force code developers to change the way
how they store their data in memory, even in the case of parallel applications where the data is distributed among different processors.
This is technically challenging, but essential if one wants to allow the code developers to focus on implementing new features and exploring new ideas
instead of spending time porting and optimizing their codes to specific computer architectures.
\subsection{A common energy zero for total energies.}
Many physical properties, e.g. forces, elastic constants, energy barriers, depend on total-energy differences. Therefore, they are well defined and the comparison of results from different codes is readily possible.
To make also total energies stemming from different codes comparable, it is necessary to define a reference energy scale. To achieve this goal, a simple, pragmatic computational prescription viable for all codes is necessary.\\
Our idea is to define relative energies, where the energy of conveniently defined reference atoms is subtracted from the total energies calculated by each code. Ideally, the reference atoms would be isolated neutral atoms and we would then have formation/atomization energies as code-independent energies. However,
it is well known that calculations for isolated atoms can be problematic for solid-state electronic-structure codes that use plane-wave basis sets (including augmented planewave ((L)APW) codes and alike). These codes are designed to study periodic systems, and the description of isolated atoms then requires large unit cells to ensure that the atoms do not interact. This makes the calculations expensive, and it may even prevent a systematic convergence study of all numerical settings, such as basis sets and grids, etc., for some atoms.
To bridge the gap between periodic and non-periodic codes, both free atoms and simple bulk systems should be used as reference systems. \\
The coexistence of several reference-energy definitions is not a limit in the comparison as long as at least one code can encompass all definitions and evaluate all the reference values. Such code (or group of codes), would serve as a ``Rosetta stone'': This 2200 years old stele
enabled the comparison and identification of Ancient Egyptian hieroglyphs, Demotic script, and Ancient Greek.
Here is a brief summary of the adopted strategy, which is thoroughly discussed in Ref. \cite{rosetta}.\\
When using free atoms as reference, the simplest choice is to define fully converged atoms (with respect to basis sets and integration grids). Their energies are calculated once for each physical model (exchange-correlation -- xc -- treatment) and, for pseudopotential-based codes, for each pseudopotential. In practice, free atoms are evaluated as spin unpolarized non-relativistic, in order to allow for a safe comparison among different codes where the implementation of the spin and relativistic treatments may be very different. However, other sets with spin-polarized atoms and selected relativistic treatments could be also stored, in order to obtain (atomization) energies that are closer to a physical meaning.
Atomic energies evaluated at the same numerical (besides the physical) settings as the calculation of interest can also be considered. On one side this strategy allows for a well known partial cancellation of numerical errors, on the other side the number of stored entries can potentially grow uncontrolled. To avoid this, a machine learning strategy that predicts the best atomic energy to be subtracted on the basis of a minimal, informative amount of stored information could be designed.\\
When using period bulk systems as reference, in a so-called ``thermodynamic approach'', the practical choice is to use the same crystal structures as used in K. Lejaeghere \emph{et al.} \cite{Cottenier, Cottenier2}. The few gaps in these publications (lanthanides and actinides) can be filled by using the structures reported in the CRC Handbook \cite{CRC}. For species like O and N that form a molecular solid, where the combination of covalent and weak interaction may add numerical noise, one could also consider to use a binary compound where the other species of the binary compound forms a covalent or metallic crystal. Also in this case the practical choice is to use {\em fully converged} (with respect to basis sets, integration grids, and $k$ grids) reference calculations, one for each physical model (including pseudopotential). Considering to subtract reference solids at the same numerical settings as the calculation of interest implies the same pros and cons as for the free atoms.
\subsection{Electronic and vibrational properties of solids}
Electronic band structures are typically represented along high-symmetry paths in the first Brillouin zone.
In literature one can find a large variety of such paths, differing in directions and sequence of the path-segments.
For an easy comparison between different calculations and codes, a practical choice is to represent all band structures along the paths defined in the paper by W. Setyawan and S. Curtarolo\cite{Bands}, if these were calculated.
Other properties of general interest are the density of electronic states and the effective masses.
For electronic band structures, density of states (DOS), and related quantities, the energy zero can be conveniently set to the highest occupied Kohn-Sham level.\\
Calculations of harmonic vibrations in bulk materials (phonons) store all relevant information either in real space (force-constant matrix) or in reciprocal space (appropriate set of dynamical matrices), whereby it is important to note that, for non-polar materials, these two quantities are unambiguously related to each other via Fourier transformations. For polar materials, this data might be augmented by the dielectric tensor and the Born effective charges, which affect the long-wavelength vibrations in such materials.
For calculations that contain them, force-constant matrices or dynamical matrices can be efficiently stored and they allow to compute other vibrational properties with negligible computational effort. For instance, vibrational band structures along the exact same paths used for the electronic band structures, but also densities of vibrational states and thermodynamic properties such as specific heats can be easily derived. Obviously, an identifier of the original calculations used to obtain the vibrational properties should be stored along with the discussed vibrational properties. This ensures that the employed computational and physical settings can be retrieved, if needed.
\subsection{Compact representation of scalar fields: density, wavefunction, xc potentials, etc.}
The comparison of scalar fields across methodologies and codes requires to translate the internal, code and basis set specific representation of these fields into a common format. For such a representation, an all-electron formalism is desirable, since it allows to evaluate additional properties such as electric field gradients and NMR shifts.\\
Different codes use different basis functions which can be divided into two subclasses: localized functions and periodic functions. Popular choices for localized functions are Gaussian-type orbitals and numeric atomic orbitals (NAOs), while for periodic functions plane-waves or augmented plane waves are often used. Plane-wave basis sets are usually combined with pseudopotentials. Approximate all-electron wavefunctions can be straightforwardly restored from such pseudopotential calculations\cite{psptoae}.\\
To represent wavefunctions in a code-independent format, conversion to a universal basis set can be used. The conversion from the original basis $\phi_\alpha$ to the universal basis $\eta_\beta$ is performed by solving:
\begin{equation}
\sum_\gamma S_{\beta\gamma}\tilde{C}_\lambda^i = \sum_\alpha \langle \eta_\beta |\phi_\alpha \rangle C_\alpha^i \, ,
\end{equation}
where $\tilde{C}_\lambda^i$ and $C_\alpha^i$ are the coefficients of the expansion in the new and original basis, respectively, and ${\bf S}$ is the overlap matrix for the universal basis functions. Additional constraints can be taken into account when minimizing differences between the original and the universal representation, such as strict orthonormalization of the transformed wavefunctions.\\
The following two choices are most promising candidates for the universal basis:
\begin{itemize}[noitemsep,nolistsep]
\item Gaussian basis functions. Online libraries of Gaussian basis sets are available (e.g., \\ \texttt{https://bse.pnl.gov/bse/portal}).
\item NAOs constructed once and for all for each species. A hierarchical construction (similar to Gaussian basis sets) of highly optimized NAO basis sets of various sizes is implemented in some codes, such as DMol$^3$ and FHI-aims.
\end{itemize}
Which specific all-electron basis set is best suited for this purpose will be evaluated in detail in an upcoming publication\cite{wavefunc}.
In addition to the code-independent representation, a common storage format that allows for a quick restoration of scalar fields in the native, code-specific representation is useful, for example for restarts from previous calculations. Such representation can be very compact because often only part of the information needs to be stored. The missing part can then be quickly obtained on the fly by running the code. For example, storing only plane-wave coefficients is sufficient for VASP to restore localized functions describing the wavefunctions near a nucleus, whereas an explicit storage of these localized functions would be very demanding. Since the concept of representing scalar fields in a basis-set expansion is common for all electronic-structure codes, defining a common, code-independent file format for storing the restart information for different codes is possible. Namely, one can store information identifying the functional form of a basis function (plane wave, contracted Gaussian, etc.), and the corresponding expansion coefficient. Obviously, such a representation cannot be used for code-independent data analysis, but it is beneficial from the practical point of view.
\subsection{Quantities related to excited-state calculations.}
Advanced many-body perturbation theory (MBPT) calculations ($GW$, Bethe-Salpeter equation, etc.) currently output only few properties (spectra, self-energies, etc.) that need to be parsed and stored. To facilitate the analysis of this kind of calculations, it is essential to develop and store a detailed classification of all approximations used in the MBPT calculation in the metadata, given that many different numerical formalisms are implemented in different MBPT codes.\\
The $GW$ approach is nowadays considered a routine approach for computing quasi-particle band structures. However, what is generically called $GW$ comprises a lot of different approximations (besides those of the underlying ground state calculations). Presently, the situation is less transparent than for ground-state. For this reason, it is necessary to store the following information:
\begin{itemize}[noitemsep,nolistsep]
\item starting point (xc functional of the underlying DFT calculation),
\item whether the calculation has been carried out in a perturbative manner or self-consistently (in fact, several types of self-consistency have been developed),
\item further approximations, like plasmon-pole models,
\item auxiliary basis sets used for non-local operators,
\item numerical approximations, such as size of reciprocal space meshes, basis set size, frequency/time grid settings, etc.,
\item whether the sum over unoccupied states is avoided / truncated / approximated,
\item whether involved quantities are represented in real or reciprocal space,
\item whether the Coulomb potential is truncated.
\end{itemize}
Obviously, all these approximations must be labelled by appropriate metadata (see also Section 2.1).
$GW$-related quantities of interest to be stored are:
\begin{itemize}[noitemsep,nolistsep]
\item matrix elements of the exchange and correlation contributions to the self-energies evaluated at the Kohn-Sham states in case of $G_0W_0$. For self-consistent $GW$ calculations, the matrix elements are evaluated with quasi-particle states.
\item matrix elements of the xc potential,
\item quasi-particle energies,
\item spectral functions (if calculated).
\end{itemize}
For optical spectra determined by TDDFT or the solution of the Bethe-Salpeter equation the situation is similar to what was described above. The quantities of interest are:
\begin{itemize}[noitemsep,nolistsep]
\item excitation spectra, energies, and oscillator strengths,
\item exciton binding energies (if available),
\item spectra obtained by the independent-particle approximation and/or KS response function.
\end{itemize}
\subsection{Establishing error bars, uncertainties, and confidence levels.}
Quantifying the errors and uncertainties of the data included in computational materials' databases is an essential step to make this data useful. Challenges in this field arise, since the errors are code, material, and even property specific. Also, the dependence of different errors on each other needs to be taken into account. A first step in this direction is to establish unique identifiers for structures through ``similarity recognition''. In this way, group of calculations performed on atomic structures that are not identical but obtained by small distortions of the coordinates and/or the unit cell can be automatically recognized as similar and an error analysis can be performed. Furthermore, a systematic investigation of numerical errors is required across codes for both simple and complex properties, which also requires a clear definition of errors/deviances, e.g., for continuous functions. With respect to errors arising from the use of approximated xc-functionals, (more) test sets are required as a reliable, high-level reference.
The use of an experimental benchmark is avoided, for the moment, as temperature and pressure conditions, intrinsic defects and impurities, surfaces, or
dislocations can make it difficult to obtain unambiguous (test) sets of reference values from experiments.\\
For a given atomic configuration, there are different error bars corresponding to the different approximations:
\begin{itemize}[noitemsep,nolistsep]
\item basis set,
\item choice of pseudopotential (if employed),
\item grids and other numerical approximations,
\item $k$-mesh,
\item treatment of relativity,
\item xc functional.
\end{itemize}
Thus, every calculated result stored in a code-independent format should be connected using the method-related metadata with six numbers that refer to the mentioned (mostly code-specific) approximations.
Obviously, energies and energy differences will be associated with different error bars, just to mention one example. In general, the importance of these error bars depends on the material's property of interest. \\
The mentioned error-bar contributions may be evaluated from any dataset that contains results corresponding to the same material, but using different approximations and/or different codes. However, it will also be necessary to evaluate relevant quantities at different levels of approximations systematically. In fact, this will also yield a ``test set for materials'' which is a well-known and most useful concept in quantum chemistry but largely absent, so far, in materials science. The build-up of a ``test set for materials science and engineering'' has been initiated.
Some such studies are being performed by the groups of G. Kresse in Vienna, and others by the NOMAD team \cite{errors,MSE}.\\
In addition to the numerical errors discussed above, it is essential to develop means to assess the possible error coming from the actual implementation (coding) of the various atomic-scale calculations. This requires the establishment of high-quality benchmark calculations for various materials and properties. For ground-state calculations, the work by Cottenier and co-workers \cite{Cottenier, Cottenier2} for bulk materials represents a first and important step in this direction. The next step, which considers low symmetry situations, e.g. defects and surfaces is just initiated. For excited-state codes the GW100 paper\cite{GW100} comparing TURBOMOL, FHI-aims, and BerkleyGW for quasi-particle energies of molecules, represents a first step towards this goal.
\section{Conclusions}
We have presented challenges and practical strategies for achieving a common format for the representation of computational materials science data.
The strategy goes through the definition of a metadata infrastructure that can be used for writing a conversion layer from existing codes to a code-independent format (the NOMAD Laboratory CoE), as well as to provide a library for the direct output of present and future electronic structure codes into a standard format (the ESCDF initiative). For several crucial topics, like the common energy zero-level, electronic and vibrational properties, the scalar-field representation, and excited states calculations, we present practical choices for achieving our goal of a code-independent representation.
We also present the challenge of establishing reliable error bars and uncertainties for each calculation stored in a database, in terms of the adopted numerical settings.
\section{Acknowledgements}
This project has received funding from the European Union's Horizon 2020 research and innovation program under grant agreement No 676580, The NOMAD Laboratory, a European Center of Excellence, and the BBDC (contract 01IS14013E).\\
We thank James Kermode and Saulius Gra\v{z}ulis for their contribution to the discussion on the metadata, and Pasquale Pavone for precious suggestions on the metadata structure and names.\\
We thank Patrick Rinke for carefully reading the manuscript.\\
We thank Claudia Draxl and Kristian Thygesen for their contribution to the discussions on the necessary information to be stored for excited-state calculations and on the error bars and uncertainties.\\
We gratefully acknowledge Damien Caliste, Fabiano Corsetti, Hubert Ebert, Jan Minar, Yann Pouillon, Thomas Ruh, David Strubbe, and Marc Torrent for their contributions to the ESCDF specifications.\\
We acknowledge inspiring discussions with Georg Kresse, Peter Blaha, Xavier Gonze, Bernard Delley, and J\"{o}rg Hutter on the energy-zero definition and scalar-field representation.\\
We thank Ole Andersen, Evert Jan Baerends, Peter Blaha, Lambert Colin, Bernard Delley, Thierry Deutsch, Claudia Draxl, John Kay Dewhurst, Roberto Dovesi, Paolo Giannozzi, Mike Gillan, Xavier Gonze, Michael Frisch, Martin Head-Gordon, Juerg Hutter, Klaus Koepernik, Georg Kresse, Roland Lindh, Hans Lischka, Andrea Marini, Todd Martinez, Jens J\o{}rgen Mortensen, Frank Neese, Richard Needs, Taisuke Ozaki, Mike Payne, Angel Rubio, Trond Saue, Chris Skylaris, Jose Soler, John Stanton, James Stewart, Marat Valiev for checking the information provided in Table 1 and for useful suggestions.
\bibliographystyle{unsrt}
| f5165e921a216187c57053a865f6922c34b909a3 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Let $T$ be a bounded operator on a complex Hilbert space $\mathcal{H}$. Then $T$ is called complex symmetric if there exists a conjugation $C$ such that $T=CT^*C$. Here a conjugation is a conjugate-linear, isometric involution on $\mathcal{H}$. The operator $T$ may also be called $C$-symmetric if $T$ is complex symmetric with respect to a specifical conjugation $C$. For more details about complex symmetric operators one may turn to \cite{CS1} and \cite{Cs2}.
In this paper, we are particularly interested in the complex symmetry of composition operators induced by analytic self-maps of $D$. This subject was started by Garcia and Hammond in \cite{GH}.
Recall that for each analytic self-map $\varphi$ of the unit disk $D$, the composition operator given by
$$C_\varphi f=f\comp\varphi$$
is always bounded on $H^2(D)$. Here, the Hardy-Hilbert space $H^2(D)$ is the set of analytic functions on $D$ such that
$$||f||_{H^2}^2=\sup_{0<r<1}\int_{0}^{2\pi}|f(re^{i\theta})|^2\frac{d\theta}{2\pi}<\infty.$$
Several simple examples of complex symmetric composition operators on $H^2(D)$ arise immediately. For example, every normal operator is complex symmetric (see \cite{CS1}), so when $\varphi(z)=sz$ with $|s|\leqslant 1$, $C_\varphi$ is normal hence complex symmetric on $H^2(D)$. Also, Theorem 2 in \cite{Cs3} states that
each operator satisfying a polynomial equation of order $2$ is complex symmetric. So when $\varphi$ is an elliptic automorphism of order $2$, then $C_\varphi^2=I$, thus $C_\varphi$ is complex symmetric on $H^2(D)$. In \cite{Invo} Waleed Noor find the conjugation $C$ such that $C_\varphi$ is $C$-symmetric when $\varphi$ is an elliptic automorphism of order $2$.
However, we are still far away from our final destination: finding out all composition operators which are complex symmetric on $H^2(D)$. The first step is, of course, to determine the complex symmetric composition operators induced by automorphisms. Bourdon and Waleed Noor considered this problem in \cite{BN}. The next Proposition is Proposition 2.1 in \cite{BN}
\begin{proposition}\label{propa}
Let $\varphi$ be a self-map of $D$. If $C_\varphi$ is complex symmetric on $H^2(D)$, then $\varphi$ has a fixed point in $D$.
Particularly, if $\varphi$ is either a parabolic or a hyperbolic automorphism, then $C_\varphi$ cannot be complex symmetric on $H^2(D)$.
\end{proposition}
Thanks to this proposition, we only need to investigate the elliptic automorphisms of the unit disk $D$. It turns out that things depend much on the orders of the automorphisms. The next proposition is one of the main result in \cite{BN}.
\begin{proposition}\label{propb}
Let $\varphi$ be an elliptic automorphism of order $q$. If $q=2$, then $C_\varphi$ is always complex symmetric on $H^2(D)$. If $4\leqslant q\leqslant\infty$, then $C_\varphi$ is complex symmetric on $H^2(D)$ only if $\varphi$ is a rotation.
\end{proposition}
However, the order 3 elliptic case remains as an open question, which is posed by Bourdon and Waleed Noor in \cite{BN}:
\begin{question}
Is $C_\varphi$ complex symmetric on $H^2(D)$ when $\varphi$ is an elliptic automorphism of order $3$?
\end{question}
The aim of this paper is to solve this problem. By dong this we complete the project of finding out all invertible composition operators which are complex symmetric $H^2(D)$.
In our main result Theorem \ref{main}, we prove that if $\varphi$ is an elliptic automorphism of order $3$ and not a rotation, then the composition operator $C_\varphi$ cannot be complex symmetric on $H^2(D)$. Then we can come to our final result as follows. It will be given as Corollary \ref{last} in the third section.
\begin{main}\label{fin}
Suppose $\varphi$ is an automorphism of $D$. Then $C_\varphi$ is complex symmetric on $H^2(D)$ if and only if $\varphi$ is either a rotation or an elliptic automorphism of order two.
\end{main}
\section{Preliminary}
The Hardy space $H^2(D)$ is naturally a Hilbert space, with the inner product
$$\langle f,g\rangle=\sup_{0<r<1}\int_{0}^{2\pi}f(re^{i\theta})\overline{g(re^{i\theta})}\frac{d\theta}{2\pi}.$$
For each $w\in D$, let
$$K_w(z)=\frac{1}{1-\overline{w}z}.$$
Then $K_w\in H^2(D)$ is the reproducing kernel at the point $w$, i.e., $$\langle f,K_w\rangle=f(w)$$ for all $f\in H^2(D)$.
It is well known that the automorphisms of the unit disk $D$ fall into three categories: parabolic and hyperbolic automorphisms have not fixed point in $D$, and besides them are the elliptic automorphisms who have a unique fixed point in $D$.
\begin{definition}
The order of an elliptic automorphism $\varphi$ is the smallest positive integer such that $\varphi_n(z)=z$ for all $z\in D$. Here $\varphi_n=\varphi\comp\varphi\comp...\comp\varphi$ denotes the $n$-th iterate of $\varphi$. If no such positive integer exists, then $\varphi$ is said to have order $\infty$.
\end{definition}
Note that if the order of an automorphism $\varphi$ is one, then $\varphi$ is identity on $D$. If $\varphi$ has order two, then $\varphi$ is of the form
$$\varphi(z)=\varphi_a(z)=\frac{a-z}{1-\overline{a}z}$$
for some $a\in D$.
\begin{remark}
The notation $\varphi_a$ will be used throughout this paper. $\varphi_a$ is the involution automorphism exchanges $0$ and $a$.
\end{remark}
By Propositions \ref{propa} and \ref{propb}, we only need to be concerned with the elliptic automorphisms that have order $3$. Moreover, if the fixed point of a automorphism is $0$, then it is a rotation and of course complex symmetric. So throughout this paper we will always assume that $\varphi$ is an elliptic automorphism of order $3$ with fixed point $a\in D\backslash\{0\}$.
\\
For a complex symmetric operator $T$, one should keep the following simple result in mind.
\begin{lemma}
Suppose $T$ is $C$-symmetric on $\mathcal{H}$. Then $\lambda\in\mathbb{C}$ is a eigenvalue of $T$ if and only if $\overline{\lambda}$ is a eigenvalue of $T^*$. Moreover, the conjugation $C$ maps the eigenvectors subspace $Ker(T-\lambda)$ onto $Ker(T^*-\overline{\lambda})$.
\end{lemma}
\begin{proof}
One only need to note that $T=CT^*C$ implies $T-\lambda=C(T^*-\overline{\lambda})C$.
\end{proof}
The next lemma gives the eigenvectors of $C_\varphi$ on $H^2(D)$ when $\varphi$ is an elliptic automorphism of order $3$. The main calculation of this lemma is done in \cite{BN}, with the help of Theorem 9.2 in Cowen and MacCluer's book \cite{CC}.
\begin{lemma}\label{le}
Suppose $\varphi$ is an elliptic automorphism of order $3$ with fixed point $a\in D$. Let $\Lambda_m=Ker(C_\varphi-\varphi'(a)^m)$ and $\Lambda^*_m=Ker(C_\varphi^*-\overline{\varphi'(a)}^m)$ for $m=0,1,2$, then
$$\Lambda_m=\overline{ \mathrm{span}}\{\varphi_a^{3j+m}; j=0,1,2,...\}$$ and $$\Lambda_m^*=\overline{\mathrm{span}}\{e_{3j+m}-ae_{3j+m-1}; j=0,1,2,...\},$$
where $e_{-1}=0$ and $e_k=K_a\varphi_a^k$ for $k=0,1 ,2,...$
\end{lemma}
\begin{proof}
Let $\tau=\varphi_a\comp\varphi\comp\varphi_a$. Then $\tau$ is a rotation of order $3$, that is, $\tau(z)=\varphi'(a)z$. So $C_\tau z^k=\varphi'(a)^kz^k$ for $k=0,1,2,...$ Moreover, since $C_\tau^*=C_{\tau^{-1}}$ where $\tau^{-1}(z)=\overline{\varphi'(a)}z$, we also have $C_\tau^* z^k=\overline{\varphi'(a)}^kz^k$. Thus
$$z^k\in Ker(C_\tau-\varphi'(a)^k)\cap Ker(C_\tau^*-\overline{\varphi'(a)}^k)$$
for $k=0,1,2,...$
Now by the definition of $\tau$ we have $C_\varphi C_{\varphi_a}=C_\tau C_{\varphi_a}$ and $C_\varphi^* C_{\varphi_a}^*=C_\tau^* C_{\varphi_a}^*$, so
$C_{\varphi_a}z^k\in Ker(C_\varphi-\varphi'(a)^k)$ and $C_{\varphi_a}^*z^k\in Ker(C_\varphi^*-\overline{\varphi'(a)}^k)$. Since $\varphi'(a)^3=1$ one can get
$$\overline{span}\{C_{\varphi_a}z^{3j+m}; j=0,1,2,...\}\subset\Lambda_m$$
and
$$\overline{span}\{C_{\varphi_a}^*z^{3j+m}; j=0,1,2,...\}\subset\Lambda_m^*.$$
On the other hand, both $C_{\varphi_a}$ and $C_{\varphi_a}^*$ are invertible on $H^2(D)$, then
$$\overline{span}\{C_{\varphi_a}z^k; k=0,1,2,...\}=\overline{span}\{C_{\varphi_a}^*z^k; k=0,1,2,...\}=H^2(D).$$
so we have
$$\Lambda_m=\overline{span}\{C_{\varphi_a}z^{3j+m}; j=0,1,2,...\}$$
and
$$\Lambda_m^*=\overline{span}\{C_{\varphi_a}^*z^{3j+m}; j=0,1,2,...\}.$$
Finally, $C_{\varphi_a}z^k=\varphi_a^k$ and the proof of Lemma 2.2 in \cite{BN} shows that $C_{\varphi_a}z^k=e_k-ae_{k-1}$, hence the proof is done.
\end{proof}
\begin{remark}
Note that $||e_j||^2=(1-|a|^2)^{-1}$ for $j=0,1,2,...$ and $\langle e_j,e_k\rangle=0$ whenever $j\ne k$.
\end{remark}
\begin{remark}\label{re2}
If $C_\varphi$ is $C$-symmetric with respect to a conjugation $C$, then $C\Lambda_m=\Lambda_m^*$ for $m=0,1,2$.
\end{remark}
\section{Proof of the main result}
In this section, we will focus on the proof of our main result, i.e., Theorem \ref{main}, which assert that no elliptic automorphism of order $3$ except for rotations can induce a complex symmetric composition operator on $H^2(D)$.
We would like to point out here that throughout the rest of this paper, each notation will always represent the same thing as it
did initially. For example, $\varphi$ is always a elliptic automorphism of order $3$ in what follows, $a$ is always the fixed point of $\varphi$ in $D\backslash\{0\}$, and $\rho$ always represents the same constant $-\frac{\overline{a}^2}{a}\cdot\frac{1-|a|^2}{1-|a|^4}$.
We will assume that $C_\varphi$ is $C$-symmetric with respect to some conjugation $C$, and finally we will show this assumption leads to a contradiction. Now we start by determining the image of a certain vector under the conjugation $C$. The notation $\{e_j\}_{j=0}^\infty$ in Lemma \ref{le} is still valid in this section.
\begin{claim}\label{1}
Let $\varphi$ be an elliptic automorphism of order $3$ with fixed point $a\in D\backslash\{0\}$. If $C_\varphi$ if $C$-symmetric on $H^2(D)$ with respect to a conjugation $C$, then we have
$$Ce_0=c_0\frac{1-\overline{a}^3\varphi_a^3}{1-\rho\varphi_a^3},$$
where $c_0$ is a constant and $\rho=-\frac{\overline{a}^2}{a}\cdot\frac{1-|a|^2}{1-|a|^4}$.
\end{claim}
\begin{proof}
Let $h_0=Ce_0$. Since $e_0\in\Lambda_0^*$, we have $h_0\in\Lambda_0$. So we can suppose that
$$h_0=\sum_{j=0}^\infty c_j\varphi_a^{3j}.$$
It is obvious that $e_0$ is orthogonal to $\Lambda_2^*$. So by Remark \ref{re2} and the fact that $C$ is an isometry, $h_0$ is orthogonal to $\Lambda_2$, which means that $\langle h_0,\varphi_a^{3k+2}\rangle=0$ for $k=0,1,2,...$ So we have the following equations,
\begin{align}\label{3.1}
\sum_{j=0}^{k}c_j\overline{a}^{3k+2-3j}+\sum_{j=k+1}^{\infty}c_ja^{3j-3k-2}=0
\end{align}
for $k=0,1,2...$ Replace $k$ by $k+1$ in (\ref{3.1}) we get
$$\sum_{j=0}^{k+1}c_j\overline{a}^{3k+5-3j}+\sum_{j=k+2}^{\infty}c_ja^{3j-3k-5}=0,$$
hence
\begin{align}\label{3.2}
\sum_{j=0}^{k+1}c_j\overline{a}^{3k+5-3j}a^3+\sum_{j=k+2}^{\infty}c_ja^{3j-3k-2}=0.
\end{align}
Combining (\ref{3.1}) and (\ref{3.2}), we have
\begin{align}\label{3.3}
\sum_{j=0}^kc_j\overline{a}^{3k+2-3j}+c_{k+1}a\frac{1-|a|^4}{1-|a|^6}=0
\end{align}
for $k=0,1,2...$
Thus, we have $c_1=\tilde\rho c_0$ and $c_{j+1}=\rho c_j$ for $j=1,2,3...$, where
$\tilde\rho=-\frac{\overline{a}^2}{a}\cdot\frac{1-|a|^6}{1-|a|^4}$ and $\rho=-\frac{\overline{a}^2}{a}\cdot\frac{1-|a|^2}{1-|a|^4}$.
Therefore,
\begin{align*}
h_0&=c_0+c_1\sum_{j=1}^\infty\rho^{j-1}\varphi_a^{3j}
\\&=c_0+c_0\frac{\tilde\rho\varphi_a^3}{1-\rho\varphi_a^3}
\\&=c_0\frac{1-\overline{a}^3\varphi_a^3}{1-\rho\varphi_a^3}.
\end{align*}
\end{proof}
\begin{claim}
For the constant $c_0$ in Claim \ref{1}, we have
$$|c_0|=\frac{1}{1-|a|^4}.$$
\end{claim}
\begin{proof}
Let
$$g=\frac{\overline{\rho}-\varphi_a^3}{1-\rho\varphi_a^3},$$
then $g$ is an inner function and $g(0)=-\frac{a^2}{\overline{a}}$.
A easy calculation shows that $h_0=\gamma_1g+\gamma_2$, where
$$\gamma_1=c_0\frac{\overline{a}^2}{a}(1+|a|^2) , \gamma_2=c_0(1+|a|^2).$$
So we have
\begin{align*}
||h_0||^2&=\langle\gamma_1g+\gamma_2,\gamma_1g+\gamma_2\rangle
\\&=|\gamma_1|^2+|\gamma_2|^2+2\Re\{\gamma_1\overline{\gamma_2}g(0)\}
\\&=|c_0|^2(1-|a|^2)(1+|a|^2)^2.
\end{align*}
Since $C$ is isometric, we can know that
$$||h_0||^2=||e_0||^2=\frac{1}{1-|a|^2},$$
thus $|c_0|=(1-|a|^4)^{-1}$.
\end{proof}
\begin{claim}
For the function $h_0=Ce_0$ in the proof of Claim \ref{1}, we have
$$\langle h_0,\varphi^{3k}\rangle=c_0(1-|a|^4)\rho^k$$\
for $k=0,1,2...$
\end{claim}
\begin{proof}
\begin{align}\label{3.4}
\langle h_0,\varphi^{3k}\rangle=\sum_{j=0}^{k}c_j\overline{a}^{3k-3j}+\sum_{j=k+1}^{\infty}c_ja^{3j-3k}.
\end{align}
Comparing (\ref{3.4}) with (\ref{3.1}), we can get that
\begin{align}\label{3.5}
\langle h_0,\varphi^{3k}\rangle=(1-|a|^4)\sum_{j=0}^{k}c_j\overline{a}^{3k-3j}.
\end{align}
So (\ref{3.5}), along with (\ref{3.3}), shows that
\begin{align*}
\langle h_0,\varphi^{3k}\rangle&=(1-|a|^4)\frac{c_{k+1}}{\tilde\rho}
\\&=c_0(1-|a|^4)\rho^k.
\end{align*}
\end{proof}
\begin{claim}\label{4}
Under the assumption of Claim \ref{1}, we have
$$Ce_1=-c_0\frac{\overline{a}(1-|a|^6)}{a(1-|a|^4)}\cdot\frac{\varphi_a(1-\overline{a}^3\varphi_a^3)}{(1-\rho\varphi_a^3)^2}+\overline{a}h_0.$$
\end{claim}
\begin{proof}
Let $h_1=Ce_1$. Since $e_1-ae_0\in\Lambda_1^*$, we have $h_1-\overline{a}h_0\in\Lambda_1$. So we can assume that
$$h_1=\sum_{j=0}^\infty b_j\varphi_a^{3j+1}+\overline{a}h_0.$$
It is obvious that $e_1$ is orthogonal to $\Lambda_0^*$, so $h_1$ is orthogonal to $\Lambda_0$, which means that $\langle h_1,\varphi_a^{3k}\rangle=0$ for $k=0,1,2...$ So we have the following equations,
$$\sum_{j=0}^{\infty}b_ja^{3j+1}+c_0\overline{a}(1-|a|^4)=0,$$
and
$$\sum_{j=0}^{k-1}b_j\overline{a}^{3k-3j-1}+\sum_{j=k}^{\infty}b_ja^{3j+1-3k}+c_0\overline{a}(1-|a|^4)\rho^k=0$$
for $k=1,2,3...$
So
$$b_0a\frac{1-|a|^4}{1-|a|^6}+c_0\overline{a}=0,$$
and
$$\sum_{j=0}^{k-1}b_j\overline{a}^{3k-3j-1}+b_ka\frac{1-|a|^4}{1-|a|^6}+c_0\overline{a}\rho^k=0$$
for $k=1,2,3...$
Now let
$$\delta_j=-\frac{b_ja(1-|a|^4)}{c_0\overline{a}(1-|a|^6)},$$
then $\delta_0=1$, $\delta_1=\rho+\tilde\rho$, and
$$\delta_{k+1}=\rho\delta_k+\tilde\rho\rho^k$$
for $k=1,2,3...$ Hence we can get that
$$\delta_k=\rho^k+k\tilde\rho\rho^{k-1}$$
for $k=0,1,2...$
Thus we have
\begin{align*}
h_1-\overline{a}h_0&=\sum_{j=0}^\infty b_j\varphi_a^{3j+1}
\\&=-c_0\frac{\overline{a}(1-|a|^6)}{a(1-|a|^4)}\sum_{j=0}^\infty \delta_j\varphi_a^{3j+1}
\\&=-c_0\frac{\overline{a}(1-|a|^6)}{a(1-|a|^4)}\left(\sum_{j=0}^\infty\rho^j\varphi_a^{3j+1}+\sum_{j=0}^\infty j\tilde\rho\rho^{j-1}\varphi_a^{3j+1}\right)
\\&=-c_0\frac{\overline{a}(1-|a|^6)}{a(1-|a|^4)}\left(\frac{\varphi_a}{1-\rho\varphi_a^3}+\frac{\tilde\rho\varphi_a^4}{(1-\rho\varphi_a^3)^2}\right)
\\&=-c_0\frac{\overline{a}(1-|a|^6)}{a(1-|a|^4)}\cdot\frac{\varphi_a(1-\overline{a}^3\varphi_a^3)}{(1-\rho\varphi_a^3)^2}.
\end{align*}
\end{proof}
Now we can prove our final result as follows.
\begin{theorem}\label{main}
If $\varphi$ is an elliptic automorphism of order 3 with fixed point $a\in D\backslash \{0\}$, then $C_\varphi$ is not complex symmetric on $H^2(D)$.
\end{theorem}
\begin{proof}
Suppose that $C_\varphi$ is $C$-symmetric with respect to conjugation $C$. Then Claim \ref{1} to \ref{4} hold.
Let
$$f=\frac{1-|a|^6}{1-|a|^4}\cdot\frac{1-\overline{a}^3\varphi_a^3}{(1-\rho\varphi_a^3)^2},$$
Then $||f||=|c_0|^{-1}\cdot||h_1-\overline{a}h_0||=(1-|a|^4)||h_1-\overline{a}h_0||$.
A tedious calculation shows that $f=\beta_1g^2+\beta_2g+\beta_3$, where $g=\frac{\overline{\rho}-\varphi_a^3}{1-\rho\varphi_a^3}$ and
\begin{align*}
\beta_1&=\frac{(1-|a|^4)(1+|a|^2)^2}{1-|a|^6}(\rho^2-\overline{a}^3\rho)=\frac{\overline{a}^4}{a^2}(1+|a|^2);
\\\beta_2&=\frac{(1-|a|^4)(1+|a|^2)^2}{1-|a|^6}(-2\rho+\overline{a}^3+\overline{a}^3|\rho|^2)=\frac{\overline{a}^2}{a}(1+|a|^2)(2+|a|^2);
\\\beta_3&=\frac{(1-|a|^4)(1+|a|^2)^2}{1-|a|^6}(1-\overline{a}^3\overline{\rho})=(1+|a|^2)^2.
\end{align*}
So
\begin{align*}
||f||^2&=\langle\beta_1g^2+\beta_2g+\beta_3,\beta_1g^2+\beta_2g+\beta_3\rangle
\\&=|\beta_1|^2+|\beta_2|^2+|\beta_3|^2+2\Re\left(\beta_1\overline{\beta_2}g(0)+\beta_2\overline{\beta_3}g(0)+\beta_1\overline{\beta_3}g(0)^2\right)
\\&=(1+2|a|^2-2|a|^4-|a|^6)(1+|a|^2)^2.
\end{align*}
However, since $C$ is isometric,
\begin{align*}
||f||^2&=(1-|a|^4)^2||h_1-\overline{a}h_0||^2
\\&=(1-|a|^4)^2||e_1-\overline{a}e_0||^2
\\&=(1-|a|^4)^2(1+|a|^2)(1-|a|^2)
\\&=(1-|a|^4)(1+|a|^2)^2.
\end{align*}
So we have
\begin{align*}
(1+2|a|^2-2|a|^4-|a|^6)(1+|a|^2)^2&=(1-|a|^4)(1+|a|^2)^2
\\2|a|^2-|a|^4-|a|^6&=0
\\|a|^2+|a|^4&=2,
\end{align*}
which is impossible since $a\in D$.
\end{proof}
At last we can get our main result as a corollary. It gives a complete description of the automorphisms which can induce complex symmetric operators on $H^2(D)$.
\begin{corollary}\label{last}
Suppose $\varphi$ is an automorphism of $D$. Then $C_\varphi$ is complex symmetric on $H^2(D)$ if and only if $\varphi$ is either a rotation or an elliptic automorphism of order two.
\end{corollary}
| 00e9b59ae196e05d550e526ca7d514b33e3b35c6 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
This paper explores Tur\'{a}n-type problems for a class of relational structures that can each be thought of as generalized directed hypergraphs. This class includes the standard undirected $r$-uniform hypergraphs that have been extensively studied in combinatorics as well as totally directed $r$-uniform hypergraphs where each edge is a set of $r$ vertices under a linear ordering. Instances of this latter structure have been studied in the extremal setting by Erd\H{o}s, Brown, Simonovits, Harary, and others \cite{brown1973, brown1969, brown2002, brown1984}. Other instances of this class are uniform versions of the model used to represent definite Horn formulas in the study of propositional logic and knowledge representation \cite{angluin1992, russell2002}. The combinatorial properties of this model have been recently studied by Langlois, Mubayi, Sloan, and Gy. Tur\'{a}n in \cite{langlois2009} and by this author in \cite{cameron2015} and \cite{cameron2015deg}. Other structures in this class are slight variations on the $d$-simplex structures studied by Leader and Tan in \cite{leader2010}.
Tur\'{a}n-type extremal problems for uniform hypergraphs make up a large and well-known area of research in combinatorics that ask the following: ``Given a family of forbidden $r$-uniform hypergraphs $\mathcal{F}$ what is the maximum number of edges an $r$-uniform hypergraph on $n$ vertices can have without containing any member of $\mathcal{F}$ as a (not necessarily induced) subgraph?" Such problems were named after Paul Tur\'{a}n due to his important early results and conjectures concerning forbidden complete $r$-graphs \cite{turan1941, turan1954, turan1961}.
A related question for undirected hypergraphs was proposed by Erd\H{o}s known as the jumping constant conjecture. A real number $\alpha \in [0,1)$ is called a jump for an integer $r \geq 2$ if there exists some positive constant $c$ which depends only on $\alpha$ such that for any $\epsilon > 0$ and positive integer $l$ there exists a positive integer $N$ for which any $r$-uniform hypergraph on $n \geq N$ vertices which has edge density at least $\alpha + \epsilon$ contains a subgraph on $l$ vertices with edge density at least $\alpha + c$. It is well-known that when $r=2$, every $\alpha \in [0,1)$ is a jump \cite{erdos1966, erdos1946}. Moreover, every $\alpha \in \left[0, \frac{r!}{r^r}\right)$ is a jump for $r \geq 3$ \cite{erdos1971}. In 1984, Frankl and R\"{o}dl disproved the jumping constant conjecture when they found the first instance of a nonjump for each $r \geq 3$\cite{frankl1984}. Since then many infinite sequences of nonjumps have been found, but the smallest known nonjump to date is $\frac{5r!}{2r^r}$ for each $r \geq 3$ determined by Frankl, Peng, R\"{o}dl, and Talbot in \cite{frankl2007}. The only additional jumps that have been found are all $\alpha \in [0.2299,0.2316), \left[0.2871,\frac{8}{27} \right)$ for $r=3$ found by Baber and Talbot in \cite{baber2011} using Razborov's flag algebra method \cite{razborov2007}.
Extremal issues of these kinds have also been considered for digraphs and multigraphs (with bounded multiplicity) in \cite{brown1973} and \cite{brown1969} and for the more general directed multi-hypergraphs in \cite{brown1984}. In \cite{brown1969}, Brown and Harary determined the extremal numbers for several types of specific directed graphs including all tournaments - that is, a digraph with one edge in some orientation between every pair of vertices. In \cite{brown1973}, Brown, Erd\H{o}s, and Simonovits determined the general structure of extremal sequences for every forbidden family of digraphs analogous to the Tur\'{a}n graphs for simple graphs.
In \cite{brown1984}, Brown and Simonovits proved several general extremal results about $r$-uniform directed $q$-hypergraphs. In this model the edges are ordered $r$-tuples of vertices with multiplicity up to $q$ for some fixed positive integer $q$. Among their results on this model are three that will be reproduced in this paper in a more general setting: Supersaturation, Continuity, and Approximation. Roughly speaking supersaturation implies that a large graph with an edge-density more than the Tur\'{a}n density for a particular forbidden family must contain many copies of members of that family. Continuity shows that given an infinite forbidden family, we can get arbitrarily close to its extremal number with a finite subfamily. Approximation is a structural result that shows that given a forbidden family, we can approximate an extremal sequence to an arbitrarily small difference by taking some sequence of graphs that all exclude this family and which all fall into some ``nice" form. All of these notions will be made rigorous in the paper.
In \cite{langlois2010} and \cite{langlois2009}, Langlois, Mubayi, Sloan, and Gy. Tur\'{a}n studied extremal properties of certain small configurations in a directed hypergraph model. This model can be thought of as a $2 \rightarrow 1$ directed hypergraph where each edge has three verticies, two of which are ``tails" and the third is a ``head." They determined the extremal number for one such subgraph with two edges, and found the extremal number of a second configuration with two edges up to asymptotic equivalence. In \cite{cameron2015} and \cite{cameron2015deg}, this author followed up this work and found the exact extremal numbers for every $2 \rightarrow 1$ directed hypergraph with exactly two edges. This $2 \rightarrow 1$ model is one instance of the class of models discussed in this paper.
The graph theoretic properties of a more general definition of a directed hypergraph were studied by Gallo, Longo, Pallottino, and Nguyen in \cite{gallo1993}. There a directed hyperedge was defined to be some subset of vertices with a partition into head vertices and tail vertices. This is a nonuniform version of models considered in this paper.
The totally directed hypergraph model considered in \cite{brown1984} and the $r \rightarrow 1$ directed hypergraph model resulting from the study of Horn clauses both lead to the natural question of all possible ways to define a directed hypergraph. The definition in this paper of the class of general directed hypergraph models attempts to unify all of the possible ``natural" ways one could define a directed hypergraph so that certain extremal questions can be answered about all of them at once. Adding to the motivation of considering more general structures is the recent interest in Razborov's flag algebra method which applies to all relational theories and not just undirected hypergraphs. The fact that the $d$-simplex model studied by Leader as well as many other somewhat geometric models come out of the class defined in this paper was a very interesting accident.
This paper is organized as follows. In Section 2, we define the class of generalized directed hypergraphs and extend the concepts of Tur\'{a}n density, blowups, and supersaturation to this setting. In Section 3, we define the idea of a jump for a given model of directed hypergraphs and prove several results about these jumps and how the jumps from one instance of the class relate to jumps in another. In Section 4, we adapt a couple of results proved in \cite{brown1984} for totally directed hypergraphs with multiplicity to any GDH. In Section 5, we ask some questions that arose from studying these structures and discuss alternate definitions that would generalize the concept further.
\section{Basic Definitions and Results}
The following definition for a generalized directed hypergraph is intended to include most uniform models that could reasonably be called uniform directed hypergraphs. This includes models where the edges are $r$-sets each under some partition into $k$ parts of fixed sizes $r_1,\ldots,r_k$ with some linear ordering on the $k$ parts. The definition only includes structures where an $r$-set could include multiple edges up to the number of possible orientations allowed. That is, we do not consider the ``oriented" versions of the models where only one edge is allowed per $r$-set. The definition is given in terms of logic and model theory for convenience only. No deep results from those subfields are used. The use of this notation also makes further generalizations like nonuniform directed hypergraphs or oriented directed hypergraphs easy.
\begin{definition}
\label{GDHdef}
Let $\mathcal{L}=\{E\}$, a language with one $r$-ary relation symbol $E$. Let $T$ be an $\mathcal{L}$-theory that consists of a single sentence of the form \[\forall x_1 \cdots x_r E(x_1,\ldots,x_r) \implies \bigwedge_{i \neq j} x_i \neq x_j \land \bigwedge_{\pi \in J_T} E(x_{\pi(1)},\ldots,x_{\pi(r)})\] for some subgroup of the group of permutations on $r$ elements, $J_T \subseteq S_r$. Call such a theory a \emph{generalized directed hypergraph theory} and any finite model of $T$ is a \emph{generalized directed hypergraph (GDH)}.
\end{definition}
Note that this definition includes graphs, hypergraphs, and $r \rightarrow 1$ directed hypergraphs. For example, the theory for a $2 \rightarrow 1$ directed hypergraph is \[T = \{\forall xyz E(x,y,z) \implies x \neq y \land x \neq z \land y \neq z \land E(y,x,z)\}.\] It is easy to see that when $r=2$ we have only two GDH theories. The theory associated with the group $S_2$ is the theory of graphs, and the theory associated with the trivial group is the theory of directed graphs.
When $r=3$ there are six subgroups of $S_3$. Three of these are all isomorphic to $\mathbb{Z}_2$ with each generated by a permutation that swaps two elements. The corresponding GDH theory for any of these can be thought of as having pointed $3$-sets for edges or as being ($2 \rightarrow 1$)-graphs. Of the other subgroups, $S_3$ itself gives the theory of undirected $3$-uniform hypergraphs, the trivial group gives totally directed $3$-edges, and the subgroup generated by a three-cycle isomorphic to $\mathbb{Z}_3$ yields a GDH theory where the edges can be thought of as $3$-sets that have some kind of cyclic orientation - either clockwise or counter-clockwise. Figure~\ref{lattice} summarizes the models of GDHs when $r=3$. Note that in general, $S_r$ always corresponds to the normal undirected $r$-graph model and the trivial group always corresponds to totally directed hypergraphs.
\begin{figure}
\centering
\begin{tikzpicture}
\node[above] at (0,3) {$S_3$};
\node[left] at (-1,2) {$\mathbb{Z}_3 $};
\node[right] at (1,1) {$\mathbb{Z}_2$};
\node[below] at (0,0) {$<i>$};
\draw[->] (-1,2) -- (0,3);
\draw[->] (1,1) -- (0,3);
\draw[->] (0,0) -- (-1,2);
\draw[->] (0,0) -- (1,1);
\node at (2.5, 1.5) {$\implies$};
\filldraw [black] (5.5,3.25) circle (1pt);
\filldraw [black] (6,3.25) circle (1pt);
\filldraw [black] (6.5,3.25) circle (1pt);
\draw[thick] (5.5,3.25) -- (6.5,3.25);
\filldraw [black] (4.25,2.25) circle (1pt);
\filldraw [black] (4.75,1.75) circle (1pt);
\filldraw [black] (3.75,1.75) circle (1pt);
\draw[thick,->] (4.25,2.25) -- (4.75,1.75);
\draw[thick,->] (4.75,1.75) -- (3.75,1.75);
\draw[thick,->] (3.75,1.75) -- (4.25,2.25);
\filldraw [black] (7.25,1.25) circle (1pt);
\filldraw [black] (7.25,0.75) circle (1pt);
\filldraw [black] (7.75,1) circle (1pt);
\draw[thick] (7.25,1.25) -- (7.25,0.75);
\draw[thick,->] (7.25,1) -- (7.75,1);
\filldraw [black] (5.5,-0.25) circle (1pt);
\filldraw [black] (6,-0.25) circle (1pt);
\filldraw [black] (6.5,-0.25) circle (1pt);
\draw[thick, ->] (5.5,-0.25) -- (6.5,-0.25);
\draw[->] (6,0) -- (5,2);
\draw[->] (6,0) -- (7,1);
\draw[->] (7,1) -- (6,3);
\draw[->] (5,2) -- (6,3);
\end{tikzpicture}
\caption{The subgroup lattice of $S_3$ and the corresponding lattice of directed hypergraphs.}
\label{lattice}
\end{figure}
A fun thought experiment is to consider the kinds of edges that arise when $r=4$. Many of them are geometric in nature. For instance, the alternating group $A_4$ gives a theory where edges can be thought of tetrahedrons (at least in an abstract sense). In fact, in \cite{leader2010} Leader and Tan study the ``oriented" versions of the models that come from the alternating groups for any $r \geq 3$.
In this paper when the theory is not specified we are simply discussing GDHs that are all models of the same fixed theory. When discussing multiple theories we will often refer to $T$-graphs to mean models of a GDH theory $T$. Throughout the paper, $J_T$ will always stand for the subgroup $J_T \subseteq S_r$ that determines the GDH theory $T$ and $m_T$ will always be the order of this subgroup, $m_T = |J_T|$. Also, $V_G$ and $E_G$ will be used to denote the underlying set of elements of a model $G$ and its relation set respectively.
The following basic propositions are given without proof. The first is a simple consequence thatwe are working in a relational language, and the second results from the fact that $J_T$ is a group.
\begin{proposition}
For any GDH theory $T$ and any nonnegative integer $n$, there exists a GDH $G \models T$ on $n$ elements. Moreover, for any nonnegative integer $k<n$, the substructure of $G$ induced on any $k$-subset of the elements of $G$ is also a $T$-graph.
\end{proposition}
\begin{proposition}
Given a GDH $G$ with $r$-ary relation set $E_G$, there exists an equivalence relation $\sim$ on $E_G$ defined by \[(a_1,\ldots,a_r) \sim (b_1,\ldots,b_r)\] if and only if for each $i$, $b_i = a_{\pi(i)}$ for some $\pi \in J_T$.
\end{proposition}
We can now use these propositions to extend the concepts of extremal graph theory to GDHs in a natural way.
\begin{definition}
For any GDH $G$, an \emph{edge} of $G$ will always refer to an equivalence class of $[E_G]_{\sim}$.
\end{definition}
\begin{definition}
Given a GDH $G$ on $n$ elements, denote the number of edges of $G$ by $e_T(G)$ and let the \emph{edge density} of $G$ be defined as \[d_T(G) := \frac{e_T(G)}{\frac{r!}{m_T} {n \choose r}}.\]
\end{definition}
Note that since \[e_T(G) = \frac{|E_G|}{m_T},\] then the density is \[d_T(G) = \frac{(n-r)!|E_G|}{n!}\] and could have been defined this way while mostly avoiding talk of edges as equivalence classes of $E_G$. However, the above definition makes the following extremal concepts reduce to their standard definitions in the undirected case.
\begin{definition}
Given two GDHs $G$ and $H$ and a function $\psi: V_H \rightarrow V_G$, we say that $\psi$ is a \emph{homomorphism} if for all $(a_1,\ldots,a_r) \in E_H$, $(\psi(a_1),\ldots,\psi(a_r)) \in E_G$.
We say that $G$ contains a copy of $H$ if there exists some injective homomorphism, $\psi:V_H \rightarrow V_G$. Otherwise, we say that $G$ is $H$-free. Similarly, we would say that a GDH $G$ is $\mathcal{F}$-free for some family $\mathcal{F}$ of GDHs if $G$ is $F$-free for all $F \in \mathcal{F}$.
\end{definition}
\begin{definition}
Given a family of GDHs $\mathcal{F}$ and a positive integer $n$, let the $n$th \emph{extremal number}, $\text{ex}_T(n,\mathcal{F})$, be defined as the maximum number of edges over all $\mathcal{F}$-free GDHs on $n$ elements, \[\text{ex}_T(n,\mathcal{F}) := \max_{\mathcal{F}\text{-free } G_n} \{e_T(G_n) \}.\] The \emph{Tur\'{a}n density} of $\mathcal{F}$ is defined as \[\pi_T(\mathcal{F}) := \lim_{n \rightarrow \infty} \frac{\text{ex}_T(n,\mathcal{F})}{\frac{r!}{m_T} {n \choose r}}.\]
\end{definition}
Our first main result is to show that these Tur\'{a}n densities exist for any GDH theory. The proof is the standard averaging argument used to show that these limiting densities exist for families of undirected hypergraphs \cite{keevash2011}.
\begin{theorem}
For any GDH family $\mathcal{F}$ the Tur\'{a}n density exists.
\end{theorem}
\begin{proof}
Let $G$ be an $\mathcal{F}$-free GDH on $n$ elements with $\text{ex}_T(n,\mathcal{F})$ edges. For each $i=1,\ldots,n$ let $G^i$ be the subGDH of $G$ induced by removing the $i$th vertex. Each edge of $G$ appears in exactly $n-r$ of these subGDHs. Therefore, \[(n-r)e_T(G) = e_T(G^1)+ \cdots e_T(G^n).\] Moreover, $e_T(G) = \text{ex}_T(n,\mathcal{F})$ and each $G^i$ is also $\mathcal{F}$-free so $e_T(G^i) \leq \text{ex}_T(n-1,\mathcal{F})$. Therefore, \[\text{ex}_T(n,G) \leq \frac{n}{n-r} \text{ex}_T(n-1,\mathcal{F}).\] So \[\frac{\text{ex}_T(n,G)}{\frac{r!}{m_T} {n \choose r}} \leq \frac{n}{n-r} \frac{\text{ex}_T(n-1,\mathcal{F})}{\frac{r!}{m_T} {n \choose r}} = \frac{\text{ex}_T(n-1,\mathcal{F})}{\frac{r!}{m_T} {n-1 \choose r}}.\] Therefore, the sequence of these extremal densities is monotone decreasing as a function of $n$ in the range $[0,1]$. Hence, the limit exists.
\end{proof}
\subsection{Blowups and Blowup Density}
We'll now extend the concept of the blowup of uniform hypergraphs to the more general setting of GDHs and define the corresponding notion of the blowup density. As with hypergraphs, the blowup of a GDH can be thought of as the replacement of each vertex with many copies and taking all of the resulting edges. Formally,
\begin{definition}
Let $G$ be a GDH with $V_G = \{x_1,\ldots,x_n\}$, and let $t = (t_1,\ldots, t_n)$ be a tuple of positive integers. Define the \emph{$t$-blowup} of $G$ to be the $\mathcal{L}$-structure $G(t)$ where \[V_{G(t)} = \{x_{11},\ldots,x_{1t_1},\ldots,x_{n1},\ldots,x_{nt_n}\}\] and \[(x_{i_1j_1},\ldots,x_{i_rj_r}) \in E_{G(t)} \iff (x_{i_1},\ldots,x_{i_r}) \in E_G.\]
\end{definition}
\begin{proposition}
Let $G$ be a GDH on $n$ vertices, and let $t=(t_1,\ldots,t_n)$ be a tuple of positive integers. Then the $t$-blowup of $G$ is also a GDH.
\end{proposition}
\begin{proof}
We need only show that the $\mathcal{L}$-structure $G(t)$ models $T$. So let \[(x_{i_1j_1},\ldots,x_{i_rj_r}) \in E_{G(t)}.\] Then $(x_{i_1},\ldots,x_{i_r}) \in E_G$. Since $G \models T$ this implies that $i_a \neq i_b$ whenever $a \neq b$. Hence, the elements $x_{i_aj_a} \neq x_{i_bj_b}$ whenever $a \neq b$. It also implies that $(x_{i_{\pi(1)}},\ldots,x_{i_{\pi(r)}}) \in E_G$ for any $\pi \in J_T$. Hence, \[(x_{i_{\pi(1)}j_{\pi(1)}},\ldots,x_{i_{\pi(r)}j_{\pi(r)}}) \in E_{G(t)}\] for any $\pi \in J_T$. Therefore, $G(t) \models T$.
\end{proof}
Next, we consider the edge density of a given blowup by defining the edge polynomial for a GDH.
\begin{definition}
Let $G$ be a GDH on $n$ vertices. For each $r$-set $R \in {V_G \choose r}$, let $e_R$ be the number of edges of $G$ in $R$. Then let the \emph{edge polynomial} be \[p_G(x) := \sum_{R \in {V_G \choose r}} e_R\prod_{i \in R} x_i.\]
\end{definition}
This polynomial is a simple generalization of the standard edge polynomial for undirected hypergraphs. To see this more easily note that for a given GDH $G$, the edges of $G$ are in bijection with the monomials the sum $p_G$ were we to write the sum out with no coefficients greater than one.
From this we see that the edge density of the $(t_1,\ldots,t_n)$-blowup of $G$ is \[\frac{p_G(t_1,\ldots,t_n)}{\frac{r!}{m_T}{t \choose r}} = m_T\frac{p_G(t_1,\ldots,t_n)}{t(t-1)\cdots(t-r+1)}\] where $t= \sum t_i$. Let $t$ increase to infinity and for each $t$ pick a vector $(t_1,\ldots,t_n)$ that maximizes this edge density. Then this sequence of densities is asymptotically equivalent to the sequence of numbers \[m_T p_G\left(\frac{t_1}{t},\ldots,\frac{t_n}{t}\right).\] This motivates the following definition.
\begin{definition}
Let $G$ be a GDH on $n$ vertices. Let \[S^n = \left\{(x_1,\ldots,x_n)|x_i \geq 0 \land \sum_{i=1}^nx_i = 1\right\},\] the standard $(n-1)$-dimensional simplex. Define the \emph{blowup density} of $G$ as \[b_T(G) = m_T \max_{x \in S^n} \{p_G(x)\}.\]
\end{definition}
Since any $x \in S^n$ is the limit of some sequence $\left\{\left(\frac{t_1}{t},\ldots,\frac{t_n}{t}\right)\right\}$ with positive $t_i$ as $t \rightarrow \infty$, then the blowup density of a GDH $G$ is the best limiting density of any sequence of blowups of $G$.
The remaining definition and basic result about blowups given in this subsection will be useful when extending results about jumps and nonjumps from undirected hypergraphs to GDHs generally in Section 3.
\begin{definition}
Let $T'$ and $T$ be GDH theories such that $J_{T'} \subseteq J_{T} \subseteq S_r$. For a $T$-graph $F$ and a $T'$-graph $F'$ we say that $F$ \emph{contains} $F'$ if $V_F = V_{F'}$ and every edge of $F'$ is contained in some edge of $F$ (where the edges are considered under their equivalence class definition as subsets of $E_F$ and $E_{F'}$). We say that $F$ is the \emph{minimum $T$-container} of $F'$ if $F$ has no edges that do not contain edges of $F'$.
\end{definition}
\begin{proposition}
\label{containing}
Let $T'$ and $T$ be GDH theories such that $J_{T'} \subseteq J_{T} \subseteq S_r$. Let $F'$ be a $T'$-graph and let $F$ be the minimum $T$-container of $F'$. Then \[\frac{m_{T'}}{m_T}b_T(F) \leq b_{T'}(F') \leq b_T(F)\] with equality on the left if $F'$ has exactly one edge contained in each edge of $F$ and equality on the right if each edge of $F$ contains all $\frac{m_T}{m_{T'}}$ possible edges of $F'$.
Moreover, if $F'$ has exactly $k$ edges contained in each edge of $F$, then \[b_{T'}(F') =\frac{km_{T'}}{m_T}b_T(F).\]
\end{proposition}
\begin{proof}
Let $|V_{F'}| = |V_{F}|=v$, then for any $x \in S^v$, \[p_F(x) \leq p_{F'}(x) \leq \frac{m_T}{m_{T'}}p_F(x)\] with equality on the left if $F'$ has exactly one edge contained in each edge of $F$ and equality on the right if each edge of $F$ contains all $\frac{m_T}{m_{T'}}$ possible edges of $F'$. Hence, \[\max_{x \in S^v}{p_F(x)} \leq \max_{x \in S^v}{p_{F'}(x)} \leq \max_{x \in S^v}{\frac{m_T}{m_{T'}}p_F(x)}.\] This implies that \[\frac{m_{T'}}{m_T}b_T(F) \leq b_{T'}(F') \leq b_T(F).\] In particular, if $F'$ has exactly $k$ edges contained in each edge of $F$, then for any $x \in S^v$, \[p_{F'}(x) = kp_F(x)\] which implies the result.
\end{proof}
\subsection{Supersaturation and Related Results}
Supersaturation holds for GDHs as it does for undirected hypergraphs, and the proof of this result is the same as the one for hypergraphs found in \cite{keevash2011} with only minor differences.
\begin{theorem}[Supersaturation]
\label{supersaturation}
Let $F$ be a GDH on $k$ elements. Let $\epsilon > 0$. For sufficiently large $n \geq n_0(F,\epsilon)$, any GDH $G$ on $n$ elements with density $d(G) \geq \pi_T(F) + \epsilon$ will contain at least $c {n \choose k}$ copies of $F$ for some constant $c = c(F,\epsilon)$.
\end{theorem}
\begin{proof}
Fix some positive integer $l$ so that \[\text{ex}_T(l,F) < \left(\pi_T(F) + \frac{\epsilon}{2}\right) \frac{r!}{m_T} {l \choose r}.\] Let $G$ be a GDH on $n > l$ elements with edge density $d_T(G) \geq \pi(F) + \epsilon$. Then $G$ must contain more than $\frac{\epsilon}{2}{n \choose l}$ $l$-sets with density at least $\pi_T(F) + \frac{\epsilon}{2}$. Otherwise, at most $\frac{\epsilon}{2}{n \choose l}$ $l$-sets contain more than $\left(\pi_T(F) + \frac{\epsilon}{2}\right){l \choose r}$ edges. Therefore, we can count the number of edges in $G$ by $l$-sets and get an upper bound of \[{n-r \choose l-r}e_T(G) \leq \frac{\epsilon}{2} {n \choose l} {l \choose r} \frac{r!}{m_T} + \left(1-\frac{\epsilon}{2}\right){n \choose l}\left(\pi_T(F)+\frac{\epsilon}{2}\right){l \choose r}\frac{r!}{m_T}.\] We can now replace $e_T(G)$ since \[e_T(G) \geq \left(\pi_T(F)+\epsilon\right){n \choose r} \frac{r!}{m_T}.\] This is enough to get the contradiction.
Since $G$ contains more than $\frac{\epsilon}{2}{n \choose l}$ $l$-sets with density at least $\pi_T(F) + \frac{\epsilon}{2}$, then it contains a copy of $F$ in each. A given copy of $F$ appears in ${n-k \choose l-k}$ $l$-sets of $G$. Therefore, there are more than \[\frac{\epsilon}{2}{n \choose l}{n-k \choose l-k}^{-1} = c {n \choose k}\] distinct copies of $F$ in $G$ where \[c = \frac{\epsilon}{2}{l \choose k}^{-1}.\]
\end{proof}
Similarly, the following theorem is an extension from the same result for undirected hypergraphs, and the proof is an adaptation of the one found in \cite{keevash2011}.
\begin{theorem}
\label{blowup}
Let $F$ be a GDH on $k$ vertices and let $t=(t_1,\ldots,t_k)$ be an $k$-tuple of positive integers. Then $\pi_T(F) = \pi_T(F(t))$.
\end{theorem}
\begin{proof}
That $\pi_T(F) \leq \pi_T(F(t))$ is trivial since $F(t)$ contains a copy of $F$ so any $F$-free GDH is automatically $F(t)$-free.
Therefore, we only need to show that $\pi_T(F) \geq \pi_T(F(t))$. Suppose not, then for sufficiently large $n$ there exists some $F(t)$-free GDH $G$ on $n$ elements with edge density strictly greater than $\pi_T(F)$. By supersaturation this implies that $G$ contains $c{n \choose k}$ copies of $F$.
Define $G^*$ to be the $k$-uniform hypergraph where $V_{G^*} = V_{G}$ and $\{a_1,\ldots,a_k\} \in E_{G^*}$ iff and only if $\{a_1,\ldots,a_k\}$ contains a copy of $F$ in $G$. Since the edge density of $G^*$ is $c>0$, then for large enough $n$, $G^*$ must contain an arbitrarily large complete $k$-partite subgraph.
For each edge $F$ maps to the vertices in at least one out of $k!$ total possible ways to make an injective homomorphism in $G$. Therefore, by Ramsey Theory, if we take the parts of this complete $k$-partite subgraph large enough and color the edges by the finite number of non-isomorphic ways that $F$ could possibly map to the $k$ vertices, we will get an arbitrarily large monochromatic $k$-partite subgraph where each part has $t$ vertices. This must have been a copy of $F(t)$ in $G$, a contradiction.
\end{proof}
The fact that the Tur\'{a}n density of a blowup equals the Tur\'{a}n density of the original GDH leads to the following nice characterization of degenerate families of GDH - those families with Tur\'{a}n density zero.
\begin{theorem}[Characterization of Degenerate GDH]
\label{degenerate}
Let $\mathcal{F}$ be some family of GDHs, then $\pi_T(\mathcal{F}) = 0$ if and only if some member $F \in \mathcal{F}$ is a subGDH of the $t$-blowup of a single edge for some vector, $t=(t_1,\ldots,t_r)$, of positive integers. Otherwise, $\pi(\mathcal{F}) \geq \frac{m_T}{r^r}$.
\end{theorem}
\begin{proof}
Suppose that no member of $\mathcal{F}$ is such a blowup. Then no member is contained in the $(t,t,\ldots,t)$-blowup of $S$. Let $S(t)$ stand for this blowup, then the sequence of GDHs, $\{S(t)\}_{t=1}^{\infty}$, is an $\mathcal{F}$-free sequence. The density of any such $S(t)$ is \[d_T(S(t)) = \frac{t^r}{\frac{r!}{m_T} {tr \choose r}} = \frac{m_T t^r (tr-r)!}{(tr)!}.\] These densities tend to $\frac{m_T}{r^r}$ as $t$ increases. Therefore, \[\pi_T(\mathcal{F}) \geq \frac{m_T}{r^r} > 0.\]
Conversely, suppose some $F \in \mathcal{F}$ is a $(t_1,\ldots,t_r)$-blowup of a single edge. By Theorem~\ref{blowup}, $\pi_T(F) = \pi_T(S) = 0$ since $\text{ex}_T(n,S)$=0 for all $n$. Therefore, $\pi_T(\mathcal{F})=0$.
\end{proof}
\section{Jumps}
Now we turn to the issue of finding jumps and nonjumps for GDH theories. The definition of a jump for undirected hypergraphs extends naturally to this setting as does the important connection between jumps and blowup densities.
\begin{definition}
Let $T$ be a GDH theory, then $\alpha \in [0,1)$ is a \emph{jump} for $T$ if there exists a $c >0$ such that for any $\epsilon > 0$ and any positive integer $l$, there exists a positive integer $n_0(\alpha, \epsilon, l)$ such that any GDH $G$ on $n \geq n_0$ elements that has at least $(\alpha + \epsilon) \frac{r!}{m_T} {n \choose r}$ edges contains a subGDH on $l$ elements with at least $(\alpha + c) \frac{r!}{m_T} {l \choose r}$ edges.
\end{definition}
Note that by Theorem~\ref{degenerate} every $\alpha \in \left[0,\frac{m_T}{r^r}\right)$ is a jump for any $r$-ary GDH theory $T$. This generalizes the well-known result of Erd\H{o}s \cite{erdos1964} that every $\alpha \in [0,\frac{r!}{r^r})$ is a jump for $r$-graphs. The following important theorem on jumps for GDH theories was originally shown by Frankl and R\"{o}dl \cite{frankl1984} for undirected hypergraphs. Their proof works equally well in this setting so the differences here are in name only.
\begin{theorem}
\label{mainguy}
The GDH theory $T$ has a jump $\alpha$ if and only if there exists a finite family $\mathcal{F}$ of GDHs such that $\pi_T(\mathcal{F}) \leq \alpha$ and $b_T(F) > \alpha$ for each $F \in \mathcal{F}$.
\end{theorem}
\begin{proof}
Let $\alpha$ be a jump and let $c$ be the supremum of all corresponding ``lengths" $c$ to the jump. Fix a positive integer $k$ so that \[{k \choose r} \left(\alpha + \frac{c}{2}\right) > \alpha \frac{k^r}{r!}.\] Let $\mathcal{F}$ be the family of all GDHs on $k$ elements with at least $\left(\alpha + \frac{c}{2}\right) {k \choose r} \frac{r!}{m_T}$ edges. Then $\pi_T(\mathcal{F}) \leq \alpha$ since any slightly larger density implies arbitrarily large subsets with density $\alpha + c$. This in turn would imply the existence of a $k$-subset with density at least $\alpha + c$. This $k$-subset would include some member of $\mathcal{F}$. On the other hand, a given $F \in \mathcal{F}$ will have blowup density \[b_T(F) \geq m_T p_F\left(\frac{1}{k},\ldots,\frac{1}{k}\right)>\alpha.\]
Conversely, suppose that such a finite family $\mathcal{F}=\{F_1,\ldots,F_k\}$ exists. Let $\epsilon >0$ and let $\{G_n\}$ be an infinite sequence of GDHs with density that tends to $\alpha + \epsilon$. As in the proof of Theorem~\ref{supersaturation}, for any positive integer $l$, $G_n$ must contain at least $\frac{\epsilon}{2}{n \choose l}$ $l$-subsets with density at least $\alpha + \frac{\epsilon}{2}$.
Let $l$ be large enough so that any GDH on $l$ vertices with density at least $\alpha + \frac{\epsilon}{2}$ contains some $F_i$ from $\mathcal{F}$. Therefore, any $G_n$ with $n > l$ contains $\frac{\epsilon}{2}{n \choose l}$ $l$-sets each with some $F_i$. Since there are only $k$ members of $\mathcal{F}$, then this implies that at least $\frac{\epsilon}{2k}{n \choose l}$ $l$-sets contain the same $F_i$.
Let $|V(F_i)|=v_i$. By the proof of Theorem~\ref{supersaturation} this implies that there is some positive constant $b$ such that $G_n$ contains at least $b{n \choose v_i}$ distinct copies of $F_i$. By the proof of Theorem~\ref{blowup} this shows that if $n$ is large enough, then we get a copy of an arbitrarily large $t$-blowup of $F_i$.
Let $c = \min_{F_i \in \mathcal{F}}{b_T(F_i)}$. For some subset $\mathcal{F}' \subseteq \mathcal{F}$, each $F_i \in \mathcal{F}'$ yields an infinite subsequence of $\{G_n\}$ which contains arbitrarily large $t$-blowups of $F_i$. The densities of these blowups all tend to at least $c$. Therefore, for any positive integer $m$, there exists an $m$-set of each $\{G_n\}$ for sufficiently large $n$ with density at least $\alpha + c$. Hence, $\alpha$ is a jump.
\end{proof}
The following proposition is needed to compare jumps between different GDH theories.
\begin{proposition}
\label{thmB}
The GDH theory $T$ has a jump $\alpha$ if and only if there exists some $c > 0$ such that for all families $\mathcal{F}$ of GDHs, either $\pi_T(\mathcal{F}) \leq \alpha$ or $\pi_T(\mathcal{F}) \geq \alpha + c$.
\end{proposition}
\begin{proof}
Let $\alpha$ be a jump for $T$ and let $c>0$ be some corresponding ``length" to the jump. Suppose that $\mathcal{F}$ is a finite family of GDHs of type $T$ for which $\alpha < \pi_T(\mathcal{F}) < \alpha + c$. Let $\{G_n\}$ be a sequence of extremal $\mathcal{F}$-free GDHs. For each positive integer $k$ there exists some $G_n$ that contains a $k$-subset with at least $(\alpha + c){k \choose r} \frac{r!}{m_T}$ edges. Take the sequence of these subsets. They are all $\mathcal{F}$-free by assumption, and the limit of their densities is at least $\alpha + c$. Therefore, $\pi_T(\mathcal{F}) \geq \alpha + c$, a contradiction.
Conversely, assume that $\alpha$ is not a jump. Let $c>0$, then for some $0<\epsilon<c$ and some positive integer $l$, there exists an infinite sequence of GDHs, $\{G_n\}$ for which each GDH has density at least $\alpha + \epsilon$ and all $l$-sets have strictly less than $(\epsilon + c){l \choose r}\frac{r!}{m_T}$ edges. Hence, $\{G_n\}$ is $\mathcal{F}$-free where $\mathcal{F}$ is the set of all $l$-GHDs with at least $(\alpha + c){l \choose r} \frac{r!}{m_T}$ edges. So $\pi_T(\mathcal{F}) \geq \alpha + \epsilon$. Since any GDH with density at least $\alpha + c$ must have an $l$-set with density at least $\alpha + c$, then $\pi_T(\mathcal{F}) < \alpha + c$.
\end{proof}
We will now look at how jumps are related between two different GDH theories for some fixed edge size $r$. We will see that in general jumps always ``pass up" the subgroup lattice. That is, if $J_{T'} \subseteq J_T$ for GDH theories $T'$ and $T$, then a jump for $T'$ is a jump for $T$. The converse is not true in general. In fact, for any GDH theories $T'$ and $T$ with $J_{T'} \subseteq J_T$ such that the order of $J_T$ is at least three times that of $J_{T'}$ we will show that the set of jumps for $T'$ is not equal to the set of jumps for $T$. The case where $m_T = 2m_{T'}$ is open.
\subsection{Jumps pass up the lattice}
First, we will show that for GDH theories $T$ and $T'$ with $J_{T'} \subseteq J_T$ the set of Tur\'{a}n densities of forbidden families of $T$-graphs is a subset of the set of Tur\'{a}n densities for $T'$.
\begin{theorem}
\label{thmC}
Let $T$ and $T'$ be two GDH theories such that $J_{T'} \subseteq J_{T}$. Then for any family $\mathcal{F}$ of $T$-graphs there exists a family $\mathcal{F}'$ of $T'$-graphs for which $\pi_{T'}(\mathcal{F}') = \pi_{T}(\mathcal{F})$. Moreover, if $\mathcal{F}$ is a finite family, then $\mathcal{F}'$ is also finite.
\end{theorem}
\begin{proof}
For each $F \in \mathcal{F}$ let $F_{T'}$ be the set of all $T'$-graphs that have exactly one edge contained in every edge of $F$. That is, since $J_{T'} \subseteq J_{T}$, then there are $\frac{m_T}{m_{T'}}$ possible $T'$ edges contained within one $T$ edge. So $F_{T'}$ is a finite set with at most $\left(\frac{m_T}{m_{T'}}\right)^{e_T(F)}$ members. Let \[\mathcal{F}' = \bigcup_{F \in \mathcal{F}} F_{T'}.\] Then $\mathcal{F}'$ is a family of $T'$-graphs. Moreover, $\mathcal{F}'$ is finite if $\mathcal{F}$ is finite. We want to show that $\pi_{T'}(\mathcal{F}') = \pi_{T}(\mathcal{F})$.
First, let $\{G_n'\}$ be an extremal $\mathcal{F}'$-free sequence of $T'$-graphs. For each $G_n'$ let $G_n$ be the $T$-graph constructed by replacing each $T'$-edge of $G_n'$ with its containing $T$-edge (multiple $T'$-edges could correspond to the same $T$-edge but each $T$-edge can only be added once).
The sequence $\{G_n\}$ is $\mathcal{F}$-free since otherwise some $G_n$ contains some $F \in \mathcal{F}$ which means that $G_n'$ must have contained at least one member of $F_{T'}$. Therefore, \[\pi_{T}(\mathcal{F}) \geq \lim_{n \rightarrow \infty} d_T(G_n) \geq \lim_{n \rightarrow \infty} \frac{\frac{m_{T'}}{m_T} e_{T'}(G_n')}{\frac{r!}{m_T}{n \choose r}} = \lim_{n \rightarrow \infty} \frac{ex_{T'}(n,\mathcal{F}')}{\frac{r!}{m_{T'}}{n \choose r}} = \pi_{T'}(\mathcal{F}').\]
Conversely, now let $\{G_n\}$ be an extremal $\mathcal{F}$-free sequence of $T$-graphs. For each $G_n$ construct a $T'$-graph $G_n'$ by replacing each $T$-edge with all $\frac{m_T}{m_{T'}}$ $T'$-edges contained in it. The sequence $\{G_n'\}$ is $\mathcal{F}'$-free with $\frac{m_T}{m_{T'}}\text{ex}_T(n,\mathcal{F})$ edges. Therefore, \[\pi_{T'}(\mathcal{F}') \geq \lim_{n \rightarrow \infty} \frac{\frac{m_T}{m_{T'}}\text{ex}_T(n,\mathcal{F})}{\frac{r!}{m_{T'}} {n \choose r}} = \pi_T(\mathcal{F}).\] So $\pi_{T'}(\mathcal{F}') = \pi_{T}(\mathcal{F})$.
\end{proof}
The converse of Theorem~\ref{thmC} is false in general. For example, the permutation subgroup for the theory $T'$ of ($2 \rightarrow 1$)-uniform directed hypergraphs is a subgroup of the permutation group for the theory $T$ of undirected $3$-graphs, $S_3$. The extremal number for the directed hypergraph is $F = \{ab \rightarrow c, cd \rightarrow e\}$ (see Figure~\ref{F}) is \[ex_{T'}(n,F) = \left\lfloor \frac{n}{3} \right\rfloor {\left\lceil \frac{2n}{3} \right\rceil \choose 2}\] as shown in \cite{langlois2010}. Therefore, the Tur\'{a}n density is $\pi_{T'}(F) = \frac{4}{27}$. However, it is well-known that no Tur\'{a}n densities exist for $3$-graphs in the interval $\left(0,\frac{6}{27}\right)$.
\begin{figure}
\centering
\begin{tikzpicture}
\filldraw [black] (-2,2) circle (1pt);
\filldraw [black] (-2,0) circle (1pt);
\draw[thick] (-2,2) -- (-2,0);
\filldraw [black] (0,1) circle (1pt);
\draw[thick, ->] (-2,1) -- (0,1);
\filldraw [black] (0,-1) circle (1pt);
\draw[thick] (0,1) -- (0,-1);
\draw[thick,->] (0,0) -- (2,0);
\filldraw [black] (2,0) circle (1pt);
\end{tikzpicture}
\caption{$\pi(F) = \frac{4}{27}$}
\label{F}
\end{figure}
\begin{corollary}
\label{jumpsgoup}
Let $T$ and $T'$ be two GDH theories such that $J_{T'} \subseteq J_{T}$. If $\alpha$ is a jump for $T'$, then it is also a jump for $T$.
\end{corollary}
\begin{proof}
If $\alpha$ is not a jump for $T$, then for any $c > 0$ there exists by Proposition~\ref{thmB} a family $\mathcal{F}$ such that $\alpha < \pi_T(\mathcal{F}) < \alpha + c$. So by Theorem~\ref{thmC} there exists a family $\mathcal{F}'$ of $T'$-graphs with $\alpha < \pi_{T'}(\mathcal{F}') < \alpha + c$. So $\alpha$ is not a jump for $T'$.
\end{proof}
Corollary~\ref{jumpsgoup} immediately implies that all nonjumps found for $r$-uniform undirected hypergraphs must also be non-jumps for any GDH with an $r$-ary relation. However, the converse is not true in general.
\subsection{Jumps do not pass down the lattice}
Roughly speaking, the current best method of demonstrating that a particular $\alpha$ is not a jump for $r$-uniform hypergraphs is to construct a sequence of hypergraphs each with blowup densities that are strictly larger than $\alpha$ but for which any relatively small subgraph has blowup density at most $\alpha$. This method originated in \cite{frankl1984} and generalizes to GDHs as the following definition and lemma demonstrate.
\begin{definition}
Let $\alpha \in [0,1)$. Call $\alpha$ a \emph{demonstrated nonjump} for a GDH theory $T$ if there exists an infinite sequence of GDHs, $\{G_n\}$, such that $b_T(G_n) > \alpha$ for each $G_n$ in the sequence and for any positive integer $l$ there exists a positive integer $n_0$ such that whenever $n \geq n_0$ then any subGDH $H \subseteq G_n$ on $l$ or fewer vertices has blowup density $b_T(H) \leq \alpha$.
\end{definition}
\begin{lemma}
Every demonstrated nonjump is a nonjump.
\end{lemma}
\begin{proof}
Suppose not. Assume that $\alpha$ is a demonstrated nonjump but is a jump. Then there exists a finite family of GDHs $\mathcal{F}$ such that $\pi_T(\mathcal{F}) \leq \alpha$ and $b_T(F) > \alpha$ for each $F \in \mathcal{F}$. Let $l$ be the maximum number of vertices over the members of $\mathcal{F}$. Let $n$ be large enough so that any subGDH on $l$ or fewer vertices has blowup density at most $\alpha$. Then some large enough blowup of $G_n$ contains some $F \in \mathcal{F}$ as a subGDH since the blowup density of each $G_n$ tends to something strictly greater than $\alpha$. Let $H$ be the minimal subGDH of $G_n$ for which the corresponding blowup contains this copy of $F$. Since $H$ has at most $l$ vertices, then it has a blowup density at most $\alpha$. Hence, \[b_T(F) \leq b_T(H(t)) \leq b_T(H) \leq \alpha,\] a contradiction.
\end{proof}
We can now show that a demonstrated nonjump for a GDH theory $T$ yields multiple nonjumps of equal and lesser values down the lattice to GDH theories $T'$ for which $J_{T'} \subseteq J_T$.
\begin{theorem}
Let $T$ and $T'$ be GDH theories such that $J_{T'} \subseteq J_{T}$. Let $\alpha$ be a demonstrated nonjump for $T$. Then $\frac{km_{T'}}{m_T} \alpha$ is a demonstrated nonjump for $T'$ for $k = 1, \ldots, \frac{m_T}{m_{T'}}$.
\end{theorem}
\begin{proof}
Let $\alpha$ be a demonstrated nonjump for $T$. Let $\{G_n\}$ be the corresponding infinite sequence of GDHs. Fix some $k \in \{1,\ldots,\frac{m_T}{m_{T'}}\}$. For each $n$ let $G_n'$ be a $T'$-graph constructed from $G_n$ by replacing each $T$-edge with $k$ $T'$-edges in any orientation. Then by Proposition~\ref{containing} we know that \[b_{T'} (G_n') = \frac{km_{T'}}{m_T}b_T(G_n)\] and any $H' \subseteq G_n'$ corresponding to $H \subseteq G_n$ also gives: \[b_{T'} (H') = \frac{km_{T'}}{m_T}b_T(H).\] Therefore, $b_{T'}(G_n') > \frac{km_{T'}}{m_T} \alpha$ for each $n$ and for any positive integer $l$, there exists a $n_0$ such that $b_{T'}(H) \leq \alpha$ for any subGDH $H \subseteq G_n$ for all $n \geq n_0$.
\end{proof}
Constructions of sequences of undirected $r$-graphs which show that $\frac{5r!}{2r^r}$ is a demonstrated nonjump for each $r \geq 3$ were given in \cite{frankl2007}. This gives the following corollary.
\begin{corollary}
Let $T$ be an $r$-ary GDH theory for $r \geq 3$. Then $\frac{5m_Tk}{2r^r}$ is a nonjump for $T$ for $k=1,\ldots,\frac{r!}{m_T}$.
\end{corollary}
This in turn shows that the set of jumps for a theory $T'$ is a proper subset of the set of jumps for $T$ for any $T$ such that $J_{T'} \subseteq J_T$ and $m_T \geq 3m_{T'}$.
\begin{corollary}
Let $T$ and $T'$ be $r$-ary GDH theories such that $J_{T'} \subseteq J_{T}$ and $m_T \geq 3m_{T'}$. Then there exists an $\alpha$ that is a nonjump for $T'$ and a jump for $T$.
\end{corollary}
\begin{proof}
Take $k=1$, then $\frac{5m_{T'}}{2r^r}$ is a nonjump for $T'$. Since $m_T \geq 3m_{T'}$, then $m_T > 2.5 m_{T'}$. So \[\frac{5m_{T'}}{2r^r} < \frac{m_T}{r^r}.\] Therefore, $\frac{5m_{T'}}{2r^r}$ is a jump for $T$ since every $\alpha \in \left[0,\frac{m_T}{r^r}\right)$ is a jump for $T$.
\end{proof}
\section{Continuity and Approximation}
The following two results are direct adaptations of two theorems from \cite{brown1984}. They are both general extremal results related to everything discussed in this paper but did not fit nicely into the other sections. The first result, Continuity, relates extremal numbers of any infinite family of GDHs to the extremal numbers of its finite subfamilies. The second, Approximation, discusses structural aspects of (nearly) extremal sequences for any forbidden family.
\begin{theorem}[Continuity]
Let $\mathcal{F}$ be an infinite family of $T$-graphs. For each $\epsilon > 0$ there exists a finite subfamily $\mathcal{F}_{\epsilon} \subset \mathcal{F}$ such that \[\text{ex}_T\left(n,\mathcal{F}\right) \leq \text{ex}_T\left(n,\mathcal{F}_{\epsilon}\right) < \text{ex}_T\left(n,\mathcal{F}\right) + \epsilon n^r\] for sufficiently large $n$.
\end{theorem}
\begin{proof}
Let $\mathcal{F}$ be the infinite family of GDHs. For each positive integer $k$ let $\mathcal{F}_k$ be the subfamily of $\mathcal{F}$ where each member has at most $k$ vertices. Let \[\gamma_k = \lim_{n \rightarrow \infty} \frac{\text{ex}_T\left(n, \mathcal{F}_k\right)}{\frac{r!}{m_T} {n \choose r}}\] and let \[\gamma = \lim_{n \rightarrow \infty} \frac{\text{ex}_T\left(n, \mathcal{F}\right)}{\frac{r!}{m_T} {n \choose r}}.\] Since $\mathcal{F}_k \subset \mathcal{F}$, then $\{\gamma_k\}_{k=1}^{\infty}$ is a monotone decreasing sequence and $\gamma_k \geq \gamma$ for all $k$.
Assume for some $\epsilon > 0$ that $\gamma_k > \gamma + \epsilon$ for all $k$. Note that \[\frac{\text{ex}_T\left(n,\mathcal{F}_k\right)}{\frac{r!}{m_T} {n \choose r}} \geq \gamma_k\] is true for all $n$. In particular, when $n=k$ there is an $\mathcal{F}_n$-free GDH on $n$ vertices with strictly more than $\left( \gamma + \epsilon \right) \frac{r!}{m_T} {n \choose r}$ edges. Since an $\mathcal{F}_n$-free GDH on $n$ vertices is also necessarily $\mathcal{F}$-free, then this implies that \[\text{ex}_T\left(n, \mathcal{F} \right) > \left(\gamma + \epsilon\right) \frac{r!}{m_T} {n \choose r},\] a contradiction.
\end{proof}
Theorem 6 in \cite{brown1984} is the Approximation Theorem for totally directed $r$-uniform hypergraphs with bounded multiplicity. We will use the following equivalent statement (in the case of multiplicity one) written in terms of Tur\'{a}n densities as a lemma to prove that this approximation result holds for all GDHs.
\begin{lemma}
\label{approx}
Let $\mathcal{F}'$ be a family of forbidden totally directed $r$-graphs ($r$-GDHs under the trivial group), and let $\epsilon>0$. Then there exists some totally directed $r$-graph $G'$ such that every blowup of $G'$ is $\mathcal{F}'$-free and \[\pi(\mathcal{F}') \geq b(G) > \pi(\mathcal{F}') - \epsilon.\]
\end{lemma}
\begin{theorem}[Approximation]
Let $\mathcal{F}$ be a family of forbidden $T$-graphs, and let $\epsilon > 0$, then there exists some $T$-graph $G$ for which all blowups of $G$ are $\mathcal{F}$-free and \[\pi_T(\mathcal{F}) \geq b_T(G) > \pi_T(\mathcal{F}) - \epsilon.\]
\end{theorem}
\begin{proof}
Let $\mathcal{F}'$ be the family of totally directed $r$-graphs as defined in the proof of Theorem~\ref{thmC}. That is, the family of directed hypergraphs for which we know that $\pi(\mathcal{F}') = \pi_T(\mathcal{F})$. We know from the proof of that theorem that any $T$-graph that is the minimal container for an $\mathcal{F}'$-free graph is $\mathcal{F}$-free. By Lemma~\ref{approx} there exists some totally directed $\mathcal{F}'$-free $r$-graph , $G'$, such that \[\pi(\mathcal{F}') \geq b(G) > \pi(\mathcal{F}') - \epsilon.\] By Proposition~\ref{containing} we know that if $G'$ is the minimal containing $T$-graph of $G$, then $b_T(G) \geq b(G')$. Hence, \[\pi_T(\mathcal{F}) \geq b_T(G) \geq b(G') > \pi(\mathcal{F}') - \epsilon = \pi_T(\mathcal{F}) - \epsilon.\]
\end{proof}
\section{Conclusion}
Some questions naturally come up in studying GDHs. Most notably it would be nice to show that the set of jumps for some GDH theory $T'$ is a proper subset of the set of jumps of any theory $T'$ up the lattice including those for which $m_T = 2m_{T'}$. Or on the other hand it would be very interesting to learn that this is not true in certain cases for $r \geq 3$!
\begin{conjecture}
Let $T'$ and $T$ be $r$-ary GDH theories for $r \geq 3$ such that $J_{T'} \subseteq J_T$ and $m_T = 2m_{T'}$. Then there exists some $\alpha \in [0,1)$ for which $\alpha$ is a jump for $T$ but not for $T'$.
\end{conjecture}
It is known by a result in \cite{brown1984} that every $\alpha \in [0,1)$ is a jump for digraphs. Therefore, the conjecture is not true when $r=2$. On a related note, is it always true that when $J_{T'} \subset J_T$, there always exists a family $\mathcal{F}'$ of $T'$-graphs such that $\pi_{T'}(\mathcal{F}')$ is not contained in the set of Tur\'{a}n densities for $T$?
\begin{conjecture}
Let $T'$ and $T$ be theories such that $J_{T'} \subseteq J_T$. Then there exists some family $\mathcal{F}'$ of $T'$-graphs such that $\pi_{T'}(\mathcal{F}')$ is not contained in the set of Tur\'{a}n densities for $T$.
\end{conjecture}
Finally, it would be nice to generalize the definition of a GDH to include other combinatorial structures. For instance we could easily change the current formulation to include multiple relations in order to capture nonuniform GDHs and those with edges that have bounded multiplicity like the structures studied in \cite{brown1984}. We could even allow these theories to contain general statements that relate the different relations. An example of this might be the theory of some kind of GDH with an edge-coloring that behaves in a certain way (at least locally). In another direction we could take away the requirement that all vertices of an edge be distinct to allow for kinds of generalized loops or add a condition that the existence of certain edges preclude the existence of others such as in the oriented cases studied in \cite{leader2010}, \cite{cameron2015}, and \cite{cameron2015deg}.
| 492771c29347aa952f13e923bb808937d1a93847 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Motivation}
Understanding of the analytic structure of the contribution of a graph to a Feynman amplitude, a time-honored problem \cite{Pham},
is related to an analysis of
its reduced graphs and the graphs in which internal edges are on the mass-shell. The former case relates to graphs in which internal edges shrink. The latter case relates to graphs with cut edges. The set of cut edges is uniquely determined by the choice of a spanning forest for the graph:
such a spanning forest defines a unique set of edges connecting distinct components of the forest. It is those edges we will put on the mass-shell.
Pairs of graphs and a chosen ordered spanning tree or forest deliver the cubical chain complex \cite{HV}.
A given ordering of the edges of the spanning tree $T$ defines a sequence of spanning forests $F$, and to any pair $(\Gamma,F)$ for fixed $\Gamma$
we can associate:\\
-a reduced graph $\Gamma_F$ obtained by shrinking all edges of $\Gamma$ to length zero which do not connect different components of the spanning forest\\
-a cut graph $\Gamma^F$ where all those edges connecting different components are put on-shell, so are marked by a Cutkosky cut,\\
-the set of graphs $G^F=\Gamma-E_{\Gamma_F}$ obtained from $\Gamma$ by removing the edges which connect distinct components of the spanning forest.
Such data define a cell-complex. With it they define a set of lower triangular matrices, one for each ordering of the edges in $T$, which allow to analyse a graph amplitude from its reduced graphs and the variations obtained by putting internal edges on-shell.
\subsection{Results}
A sequence of cuts (edge sets $\epsilon_i$ determines from $i$-component forests, $i\geq 2$)
\[\epsilon_2\to \epsilon_3\to\cdots\to \epsilon_{v_\Gamma}\]
will shift the normal threshold $s_0(\epsilon_2)$ associated with a chosen cut $\epsilon_2$ to anomalous thresholds
\[
s_0(\epsilon_2)\to s_1(\epsilon_3)\to\cdots\to s_{v_\Gamma-2}(\epsilon_{v_\Gamma}).
\]
The resulting sequence of anomalous thresholds $s_i(\epsilon_{i+2})$, $i>0$ is a sequence of values for a channel variable $s$ defined by $\epsilon_2$.
They are computed from the divisors associated to $\epsilon_{i+2}$. The latter are functions of all kinematical variables.
For example, for the one-loop triangle the divisor in $\mathbb{C}^3$ associated to $\epsilon_3$ is a simple function of
\[
\lambda(p_1^2,p_2^2,p_3^2)=p_1.p_2^2-p_1^2p_2^2=p_2.p_3^2-p_2^2p_3^2=p_3.p_1^2-p_3^2p_1^2,\,p_1+p_2+p_3=0.
\]
The three representations of $\lambda(p_1^2,p_2^2,p_3^2)$ allow to compute $s_1(\epsilon_3)$ for the three choices of a channel variable $s=p_3^2$ or $s=p_1^2$ or $s=p_2^2$ respectively.
As a result, to a graph $\Gamma$ we can assign a collection of lower triangular matrices $M_i^\Gamma$ with the following properties:\\
i) All entries in the matrix correspond to well-defined integrable forms under on-shell renormalization conditions.\\
ii) Anomalous thresholds $s_i$ are determined from properties of graph polynomials. They provide lower boundaries for dispersion integrals associated to these integrable forms.\\
iii) Along the diagonal in the matrices $M_i^\Gamma$ we find leading threshold entries: all quadrics for all edges in a graph are on the mass-shell.\\
iv) The variation of a column in $M_i^\Gamma$ wrt to a given channel is given by the column to the right.\\
v) The subdiagonal entries $(M_i^\Gamma)_{k,k-1}$ are determined from the diagonal entries $(M_i^\Gamma)_{k-1,k-1}$
and $(M_i^\Gamma)_{k,k}$ via a dispersion integral. This gives $(k-1)$ two-by-two matrices each of which has an interpretation via the optical theorem.
This hence determines the first subdiagonal.\\
vi) Continuing, all subdiagonals and hence the whole matrix $(M_i^\Gamma)_{r,s}$ is determined via iterated dispersion.
This answers the question how to continue the optical theorem beyond two-point functions.
\section{The cubical chain complex}
We follow \cite{HV}.
Consider a pair $(\Gamma,T)$ of a bridge free graph $\Gamma$ and a chosen spanning tree $T$ for it.
Assume $T$ has $k$ edges. Consider the $k$-dimensional unit cube. It has origin $(0,\cdots,0)$ and $k$ unit vectors
$(1,0,\cdots,0),\ldots$, $(0,\cdots,0,1)$ form its edges regarded as $1$-cells. A change of ordering of the edges of $T$ permutes those edges.
The origin is decorated by a rose on $|\Gamma|$ petals, and the corner $(1,1,\cdots,1)$ decorated by $(\Gamma,V_\Gamma)$, with $k=|V_\Gamma|-1$,
and we regard $V_\Gamma$ as a spanning forest consisting of $k+1$ distinct vertices.
The complex is best explained by assigning graphs as in the following example.
$${\;\raisebox{10mm}{\epsfysize=80mm\epsfbox{cubical.eps}}\;}$$
The cell is two-dimensional as each of the five spanning trees of the graph $\Gamma$, the dunce's cap graph, in the middle of the cell has length two.
We have chosen a spanning tree $T$ provided by the edges $e_1$ and $e_3$, indicated in red. The boundary of our two-dimensional cell has four one-dimensional edges, bounded by two of the four 0-dimensional corners each.
To these lower dimensional cells we assign graphs as well as indicated.
The spanning tree has length two and so there are $2=2!$ orderings of its edges, and hence two lower triangular $3\times 3$ matrices $M^\Gamma_i$ which we can assign to this cell.
They look as follows:
$${\;\raisebox{10mm}{\epsfxsize=140mm\epsfbox{cubicalM.eps}}\;} $$
These square matrices are lower triangular.
Note that we have cuts which separate the graph into two components determining a normal threshold which appears already in a reduced graph on the diagonal, and in the lower right corner a cut into three components, which determines an anomalous threshold. All these cuts determine variations, as stated in Cutkosky's theorem \cite{BK}.
\section{Cutkosky's theorem}
We quote from \cite{BK} where you find details. For a graph $\Gamma$ and a choosen spanning forest $F$ we let the quotient graph $\Gamma''$ -the reduced graph-
be the graph obtained by shrinking all edges $e\in E'$ of $\Gamma$ which do not connect distinct components of $F$, so $E'=E_\Gamma-E''$,
and $E''$ all edges of $\Gamma$ which do connect distinct components of $F$.
Assume the reducedgraph $\Gamma''$ has a physical singularity at an external momentum point $p''$, i.e. the intersection $\bigcap_{e\in E''}Q_e$ of the propagator quadrics associated to edges in $E''$ has such a singularity at a point lying over $p''$. Let $p$ be an external momentum point for $\Gamma$ lying over $p''$. Then the variation of the amplitude $I(\Gamma)$ around $p$ is given by Cutkosky's formula
\[\text{var}(I(\Gamma)) = (-2\pi i)^{\# E''}\int\frac{\prod_{e\in E''} \delta^+(\ell_e)}{\prod_{e\in E'} \ell_e}.
\]
\section{Anomalous thresholds}
Let us come back to a generic graph $\Gamma$. We want to determine anomalous thresholds. With their help, dispersion relations can be established when real analycity in kinematical variables can be established.
We analyse the Landau singularities of $\Gamma$ in terms of $\Gamma/e$, where
$e$ is such an edge. To completely analyse the graph, we have to consider all possibilities to shrink it edge after edge (the generalization to multiple edges is in \cite{BK}).
We have for the second Symanzik polynomial $\Phi$
\[
\Phi(\Gamma)=\overbrace{\Phi(\Gamma/e)}^{=:X}+A_e\overbrace{\left\{ \Phi(\Gamma-e)-m_e^2\psi(\Gamma/e) \right\} }^{=:Y}-A_e^2\overbrace{m_e^2\psi(\Gamma-e)}^{Z}.
\]
Solving $\Phi(\Gamma)=0$ for $A_e$ is a quadratic equation with coefficients $X,Y,Z$. Note that $Z>0$ is independent of kinematical variables, while $X,Y$ depend on momenta and masses. In particular, for a chosen channel variable $s$ we can write $X=sX_s+N$, with $X_s$ independent of kinematics and $N$ a constant in the channel variabel $s$. It depends on other kinematic variables though.
In terms of parametric variables, $X,Y$ are functions of the parametric variables of the reduced graph.
The above quadratic equation has a discriminant $D=Y^2+4XZ$, and we find a physical Landau singularity for positive $Y$ and vanishing discriminant $D=0$. Define $Y_0:=Y(p_A^{\Gamma/e},\{Q,M\})$ to be the evaluation of $Y$ by evaluating parametric variables at the point of the Landau singularity for the reduced graph.
The condition $D=0$ allows to determine the anomalous threshold from
\[
s(\{A\},\{Q,M\})=\frac{Y^2-4ZN}{4ZX_s},
\]
minimizing over parametric variables $A_e\geq 0$.
Let $\mathit{T}_s^\Gamma$ be the set of all ordered spanning trees $T$ of a fixed graph $\Gamma$ which allow for the same associated channel variable $s$.
We have the following result.\\
i) A necessary and sufficient condition for a physical Landau singularity is $Y_0>0$ with $D=0$.\\
ii) The corresponding anomalous threshold $s_F$ for fixed masses and momenta $\{M,Q\}$ is given as the minimum of $s(\{a,b\},\{Q,M\})$ varied over edge variables $\{a,b\}$. It is finite ($s_F>-\infty$) if the minimum is a point inside $p\in\mathbb{P}^{e_\Gamma-1}$ in the interior of the
integration domain $A_i>0$. If it is on the boundary
of that simplex, $s_F=-\infty$.\\
iii) If for all $T\in \mathit{T}_s^\Gamma$ and for all their forests $(\Gamma,F)$ we have $s_F>-\infty$, the Feynman integral $\Phi_R(\Gamma)(s)$ is real analytic as a function of $s$
for $s<\min_F\{s_F\} $.\\
iv) For $Y>0$ and $X<0$, both zeroes of $\Phi(\Gamma)=0$ appear for $A_e>0$. For $Y>0$ and $X>0$, only one zero is inside the domain of integration. As a result for $X=0$ corresponding to the threshold provided by the reduced $\Gamma/e$ we have a discontinuity.
\section{Example}
We consider the triangle graph $\Delta$.
In fact, we augment it with one of its three possible spanning trees, say on edges $e_2,e_3$, so $E_T=\{e_2,e_3\}$. The corresponding cell in the cubical chain complex is
\begin{equation} \label{trianglecell}
{\;\raisebox{-40mm}{\epsfxsize=80mm\epsfbox{trianglecubical.eps}}\;}
\end{equation}
For the Cutkosky cut we choose two of the three edges, say $\epsilon_2=\{e_1,e_2\}$. This defines the channel $s=p_a^2$ and the matrix $M^\Delta_1$.
The other cut in that matrix is the full cut separating all three vertices.
We give $M^\Delta_1$ in the following figure:
$$
M^\Delta_1={\;\raisebox{-40mm}{\epsfxsize=80mm\epsfbox{trianglematrix.eps}}\;}
$$
We now calculate:
\begin{eqnarray*}
\Phi_\Delta & = & \overbrace{p_a^2A_1A_2-(m_1^2A_1+m_2^2A_1)(A_1+A_2)}^{=\Phi_{\Gamma/e_3}}\\
& & +A_3((p_b^2-m_3^2-m_1^2)A_1+(p_c^2-m_1^2-m_3^2)A_2)-A_3^2m_3^2,
\end{eqnarray*}
so
\begin{eqnarray*}
\Phi_\Delta & = & \Phi_{\Delta/e_3}
+A_3Y-A_3^2 m_3^2\overbrace{\psi_{\Delta-e_1}}^{=1},
\end{eqnarray*}
as announced:
\[
X=\Phi_{\Delta/e_3},\,Y=\overbrace{(p_b^2-m_3^2-m_1^2)}^{=:l_1}A_1+\overbrace{(p_c^2-m_1^2-m_3^2)}^{=:l_2}A_2,\,Z=m_3^2.
\]
We have $Y_0=m_2 l_1+m_1 l_2$, and need $Y_0>0$ for a Landau singularity.
Solving $\Phi(\Delta/e_3)=0$ for a Landau singularity determines
the familiar physical threshold in the $s=p_a^2$ channel, leading for the reduced graph to
\begin{equation}
p_Q: s_0=(m_2+m_3)^2,\, p_A: A_1m_1=A_2m_2.
\end{equation}
We let $D=Y^2+4XZ$ be the discriminant. For a Landau singularity we need
\[
D=0.
\]
We have
\begin{equation}
\Phi_\Delta=-m_3^2\left(A_3-\frac{Y+\sqrt{D}}{2m_3^2}\right)\left(A_3-\frac{Y-\sqrt{D}}{2m_3^2}\right),
\end{equation}
where $Y,D$ are functions of $A_1,A_2$ and $m_1^2,m_2^2,m_3^2,s,p_b^2,p_c^2$. Note that at $D=0$ we have
\[
2m_3^2A_3=A_1l_1+A_2l_2,
\]
which determines a co-dimension one (a line) hypersurface of $\mathbb{P}^2$. Finding the anomalous thresold determines a point on this line (it fixes the ratio $A_1/A_2$), and hence the anomalous threshold determines a point in $\mathbb{P}^2$.
We can write
\[
0=D=Y^2+4Z(sA_1A_2-N),
\]
with $N=(A_1m_1^2+A_2m_2^2)(A_1+A_2)$ $s$-independent.
This gives
\begin{equation}
s(A_1,A_2)=\frac{4ZN-(A_1l_1+A_2l_2)^2}{4ZA_1A_2}=:\frac{A_1}{A_2}\rho_1+\rho_0+\frac{A_2}{A_1}\rho_2.
\end{equation}
Define two Kallen functions $\rho_1=-\lambda_1=-\lambda(p_b^2,m_1^2,m_3^2)$ and
$\rho_2=-\lambda_2=-\lambda(p_c^2,m_2^2,m_3^2)$. Both are real and non-zero off their threshold or pseudo-threshold.
Then, for
\[
\rho_1>0,\,\rho_2>0,
\]
we find the threshold $s_1$ at
\begin{equation}
s_1=(m_1+m_2)^2+\frac{4m_3^2(\sqrt{\lambda_2}m_1-\sqrt{\lambda_1}m_2)^2-(\sqrt{\lambda_1}l_2+\sqrt{\lambda_2}l_1)^2}{4m_3^2\sqrt{\lambda_1}\sqrt{\lambda_2}}.
\end{equation}
On the other hand for the coefficients of $\rho_1<0$ and/or $\rho_2<0$ we find a minimum
\begin{equation} s_1=-\infty,
\end{equation}
along the boundaries $A_1=0$ or $A_2=0$.
The domains $Y>0,X<0$ and $Y>0,X>0$ determine the domains of parametric integration for the variation prescribed by Cutkosky's theorem,
whilst the normal and anomalous threshold (when finite) determine the lower boundaries of the dispersion integrals needed to reconstruct the function from its variation.
Let us now discuss the triangle in more detail.
It allows three spanning trees on two edges each, so we get six matrices $M^\Delta_i$, $i=1,\ldots,6$ altogether,
by having two possibilities to order the two edges for each spanning tree.
The six matrices $M^\Delta_i$ come in groups of two for each spanning tree.
For each of the three spanning trees we get a cell as in (\ref{trianglecell}).
The boundary operator for such a cell in the cubical cell complex of \cite{HV} is the obvious one stemming from co-dimension one hypersurfaces at $0$ or $1$
with suitable signs. So the square populated by the triangle $\Delta$ in (\ref{trianglecell}) has four boundary components, the edges populated by the four graphs as indicated. Those four edges are the obvious boundary of the square.
If we now consider all graphs in (\ref{trianglecell}) as evaluated by the Feynman rules, we can consider for a given cell a boundary operator which replaces evaluation at
the $x_e=0$-hypersurface by shrinking edge $e$, and evaluation at the $x_e=1$-hypersurface by setting edge $e$ on the mass-shell.
Then, to check that this is a boundary operator for the amplitudes defined by the graphs in (\ref{trianglecell}) we need to check that the amplitudes for the four graphs at the four corners are uniquely defined from the amplitudes of the graphs at the adjacent edges: for example, the imaginary part of the amplitude of the graph on the left vertical edge is related to the amplitude of the graph at the upper left corner:
This imaginary part must be also obtained from shinking edge $e_3$ in the graph on the upper horizontal edge by setting $A_3$ to zero in the integrand and integrating over the hypersurface $A_3=0$ of the integration simplex $\sigma_\Delta$. This is indeed the case, and similar checks work for all other corners.
In summary, the analytic structure of Feynman amplitudes realized the structure of the cubical chain complex. The latter is highly non-trivial.
Its further study in the conext needed for physics will inform our understanding of amplitudes considerably. Future work will be dedicated in understanding the relation between the monodromy of physical singularities and the fundamental group underlying Outer Space as used in \cite{HV}.
| 798eb17894125d70c3c5974449f44faf3c669178 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Let $\lambda_1,\ldots,\lambda_n$ be $n$ distinct values. The Lagrange interpolation formula says that a polynomial $P(x)$ of degree not greater than $n-1$ in one variable can be written in the following:
$$P(x) = \sum_{i=1}^nP(\lambda_i)L_i(x),$$
where
$$L_i(x) = \prod_{j\neq i}\frac{x-\lambda_j}{\lambda_i-\lambda_j}.$$
This formula not only provides a means of polynomial approximation, but also plays a significant role in establishing combinatorial identities in many variables including identities on symmetric polynomials. For instance, given $n$ distinct values $\lambda_1,\lambda_2,\ldots,\lambda_n$ $(n \geq 2)$ and an integer $m \geq 0$, we have the following identity:
\begin{equation}\label{id1}
\sum_{i=1}^n\frac{\lambda_i^m}{\displaystyle\prod_{j\neq i}(\lambda_i-\lambda_j)}=h_{m-n+1}(\lambda_1,\ldots,\lambda_n),
\end{equation}
where $h_k$ is the $k$-th homogeneous symmetric polynomial in $n$ variables, which is defined to be zero for $k<0$. The identity can be easily derived from the Lagrange interpolation formula (see \cite[Theorem 2.2]{CL} for more details). In particular, for $m = 0, 1, \ldots,n-1$, the left hand side of (\ref{id1}) is independent to $\lambda_1,\ldots,\lambda_n$. It is equal to zero if $m < n-1$ and one if $m=n-1$. This implies the following result.
\begin{proposition}\label{pro1}
Let $P(x)$ be a polynomial of degree not greater than $n-1$ in one variable. Then the sum
$$\sum_{i=1}^n\frac{P(\lambda_i)}{\displaystyle\prod_{j\neq i}(\lambda_i-\lambda_j)}= c_n,$$
where $c_n$ is the coefficient of $x^{n-1}$ in the polynomial $P(x)$.
\end{proposition}
The proof of Proposition \ref{pro1} is elementary. It is a direct corollary of the Lagrange interpolation formula.
The first goal of this paper is to give a generalization of Proposition \ref{pro1} for multivariate symmetric polynomials. Throughout we always assume that all polynomials are over the field of rational numbers or any field of characteristic zero. For convenience, we shall write $[n]$ for the set $\{1,2,\ldots,n\}$. Let $\lambda_1,\ldots,\lambda_n$ be $n$ distinct values. For each subset $I = \{i_1,\ldots,i_k\} \subset [n]$, we denote by $\lambda_I = (\lambda_{i_1},\ldots,\lambda_{i_k})$ and $I^c = [n]\setminus I$. Recall that a polynomial $P(x_1,\ldots,x_k)$ is said to be symmetric if it is invariant under permutations of $x_1,\ldots,x_k$. We obtain the following result.
\begin{theorem}\label{main}
Let $P(x_1,\ldots,x_k)$ be a symmetric polynomial of degree not greater than $k(n-k)$ in $k$ variables $(k < n)$. Then the sum
$$\sum_{I\subset [n],|I|=k}\frac{P(\lambda_I)}{\displaystyle\prod_{i\in I}\prod_{j\in I^c}(\lambda_i-\lambda_j)} = \frac{c(k,n)}{k!},$$
where $c(k,n)$ is the coefficient of $x_1^{n-1}\ldots x_k^{n-1}$ in the polynomial
$$P(x_1,\ldots,x_k)\prod_{i=1}^k\prod_{j \neq i}(x_i-x_j).$$
\end{theorem}
More generally, we also obtain an identity involving doubly symmetric polynomials. Recall that a polynomial $P(x_1,\ldots,x_k,y_1,\ldots,y_{n-k})$ is said to be doubly symmetric if it is invariant under permutations of $x_1,\ldots,x_k$ and permutations of $y_1,\ldots,y_{n-k}$ respectively. We obtain the following result.
\begin{theorem}\label{thm3}
Let $P(x_1,\ldots,x_k, y_1,\ldots,y_{n-k})$ be a doubly symmetric polynomial of degree not greater than $k(n-k)$. Then the sum
$$\sum_{I\subset [n],|I|=k}\frac{P(\lambda_I,\lambda_{I^c})}{\displaystyle\prod_{i\in I}\prod_{j\in I^c}(\lambda_i-\lambda_j)} = \frac{d(k,n)}{k!(n-k)!},$$
where $d(k,n)$ is the coefficient of $x_1^{n-1}\ldots x_k^{n-1}y_1^{n-1}\ldots y_{n-k}^{n-1}$ in the polynomial
$$P(x_1,\ldots,x_k, y_1,\ldots,y_{n-k})\prod_{i=1}^k\prod_{j \neq i}(x_i-x_j)\prod_{i=1}^{n-k}\prod_{j \neq i}(y_i-y_j)\prod_{i=1}^{n-k}\prod_{j=1}^k(y_i - x_j).$$
\end{theorem}
The second goal of this paper is to give a way of dealing with integrals over Grassmannians. The idea is as follows. Localization in equivariant cohomology allows us to express integrals in terms of some data attached to the fixed points of a torus action. In particular, for Grassmannians, we obtain interesting formulas with nontrivial relations involving rational functions. Let $G(k,n)$ be the Grassmannian of $k$-dimensional linear spaces in $\mathbb C^n$. Consider the following integrals:
$$\int_{G(k,n)}\Phi(\mathcal S)\quad ,\quad \int_{G(k,n)}\Psi(\mathcal Q)\quad,\quad \int_{G(k,n)}\Delta(\mathcal S,\mathcal Q),$$
where $\Phi(\mathcal S), \Psi(\mathcal Q)$ are respectively characteristic classes of the tautological sub-bundle $\mathcal S$ and quotient bundle $\mathcal Q$ on the Grassmannian $G(k,n)$, and $\Delta(\mathcal S,\mathcal Q)$ is a characteristic class of both $\mathcal S$ and $\mathcal Q$.
Using localization in equivariant cohomology, Weber \cite{We} and Zielenkiewicz \cite{Zi} presented a way of expressing the integrals as iterated residues at infinity of holomorphic functions. However, there is no guarantee expressions appearing in the Atiyah-Bott-Berline-Vergne formula for Grassmannians are in fact constants rather than rational functions. Our identities provide a way of handling such expressions.
\begin{theorem}\label{integral1}
Suppose that $\Phi(\mathcal S)$ is represented by a symmetric polynomial $P(x_1,\ldots,x_k)$ of degree not greater than $k(n-k)$ in $k$ variables $x_1,\ldots,x_k$ which are the Chern roots of $\mathcal S$ and $\Psi(\mathcal Q)$ is represented by a symmetric polynomial $Q(y_1,\ldots,y_{n-k})$ of degree not greater than $k(n-k)$ in $n-k$ variables $y_1,\ldots,y_{n-k}$ which are the Chern roots of $\mathcal Q$. We then have the following statements:
\begin{enumerate}
\item The integral
$$\int_{G(k,n)}\Phi(\mathcal S) = (-1)^{k(n-k)}\frac{c(k,n)}{k!},$$
where $c(k,n)$ is the coefficient of $x_1^{n-1}\ldots x_k^{n-1}$ in the polynomial
$$P(x_1,\ldots,x_k)\prod_{i=1}^k\prod_{j \neq i}(x_i-x_j).$$
\item The integral
$$\int_{G(k,n)}\Psi(\mathcal Q) = \frac{c(k,n)}{(n-k)!},$$
where $c(k,n)$ is the coefficient of $y_1^{n-1}\ldots y_{n-k}^{n-1}$ in the polynomial
$$Q(y_1,\ldots,y_{n-k})\prod_{i=1}^{n-k}\prod_{j \neq i}(y_i-y_j).$$
\end{enumerate}
\end{theorem}
\begin{theorem}\label{integral2}
Suppose that $\Delta(\mathcal S,\mathcal Q)$ is represented by a doubly symmetric polynomial
$$P(x_1,\ldots,x_k,y_1,\ldots,y_{n-k})$$
of degree not greater than $k(n-k)$ in $n$ variables which are the Chern roots of $\mathcal S$ and $\mathcal Q$ respectively. We then have the integral
$$\int_{G(k,n)}\Delta(\mathcal S,\mathcal Q) = (-1)^{k(n-k)}\frac{d(k,n)}{k!(n-k)!},$$
where $d(k,n)$ is the coefficient of $x_1^{n-1}\ldots x_k^{n-1}y_1^{n-1}\ldots y_{n-k}^{n-1}$ in the polynomial
$$P(x_1,\ldots,x_k, y_1,\ldots,y_{n-k})\prod_{i=1}^k\prod_{j \neq i}(x_i-x_j)\prod_{i=1}^{n-k}\prod_{j \neq i}(y_i-y_j)\prod_{i=1}^{n-k}\prod_{j=1}^k(y_i - x_j).$$
\end{theorem}
As a special case, if $\Delta(\mathcal S,\mathcal Q) = \Phi(\mathcal S)\Psi(\mathcal Q)$ is represented by the product of two symmetric polynomials $P(x_1,\ldots,x_k)$ and $Q(y_1,\ldots,y_{n-k})$, then we have the following corollary.
\begin{corollary}
Suppose that $\deg(P) + \deg(Q) \leq k(n-k)$. The integral
$$\int_{G(k,n)}\Phi(\mathcal S)\Psi(\mathcal Q) = (-1)^{k(n-k)}\frac{d(k,n)}{k!(n-k)!},$$
where $d(k,n)$ is the coefficient of $x_1^{n-1}\ldots x_k^{n-1}y_1^{n-1}\ldots y_{n-k}^{n-1}$ in the polynomial
$$P(x_1,\ldots,x_k)Q(y_1,\ldots,y_{n-k})\prod_{i=1}^k\prod_{j \neq i}(x_i-x_j)\prod_{i=1}^{n-k}\prod_{j \neq i}(y_i-y_j)\prod_{i=1}^{n-k}\prod_{j=1}^k(y_i - x_j).$$
\end{corollary}
The rest of the paper is organized as follows: The identities are proved in Section 2. Section 3 is to give a review of equivariant cohomology and the proof of Theorem \ref{integral1}.
\section{Proof of the identities}
We shall use $X$ to denote the set $\{x_1,\ldots,x_k\}$ or the $n$-tuple $(x_1,\ldots,x_k)$, and $Y$ to denote the set $\{y_1,\ldots,y_{n-k}\}$ or the $(n-k)$-tuple $(y_1,\ldots,y_{n-k})$. We denote by
$$Y-X = \prod_{i=1}^{n-k}\prod_{j=1}^k(y_i - x_j).$$
For $I=\{i_1,\ldots,i_k\}$ and $J=\{j_1,\ldots,j_{n-k}\}$, we denote by $\lambda_I = (\lambda_{i_1},\ldots,\lambda_{i_k})$, $\lambda_J = (\lambda_{j_1},\ldots,\lambda_{j_{n-k}})$, and
$$\lambda_I - \lambda_J = \prod_{i\in I}\prod_{j\in J}(\lambda_i-\lambda_j).$$
\begin{proof}[Proof of Theorem \ref{main}]
Set
$$F(X) = P(X)\prod_{i=1}^k\prod_{j\neq i}(x_i-x_j).$$
By the assumption, the degree of $F(X)$ is not greater than $k(n-k) + k(k-1) = k(n-1)$. Since $P$ is symmetric, so $F$ is also.
By the division algorithm for multivariate polynomials (see \cite[Theorem 3]{CLO}), there exist the polynomials $F_i(X), i = 1, \ldots,k$ and $R(X)$ such that
$$R(X) = F(X) - \sum_{i=1}^kF_i(X)\prod_{j=1}^n(x_i-\lambda_j),$$
and all partial degrees of $R(X)$ are not greater that $n-1$.
By the Lagrange interpolation formula, we have
$$R(X) = \sum_{i_1=1}^nR(\lambda_{i_1},x_2,\ldots,x_k)L_{i_1}(x_1).$$
By the Lagrange interpolation formula for the polynomials $R(\lambda_{i_1},x_2,\ldots,x_k)$, we have
$$R(X) = \sum_{i_1=1}^n\sum_{i_2=1}^nR(\lambda_{i_1},\lambda_{i_2},x_3,\ldots,x_k)L_{i_1}(x_1)L_{i_2}(x_2).$$
So on, we have
$$R(X) = \sum_{i_1,\ldots,i_k=1}^nR(\lambda_I)\prod_{l=1}^kL_{i_l}(x_l).$$
For each $I=\{i_1,\ldots,i_k\}$, we have $R(\lambda_I) = F(\lambda_I)$, and if $i_s = i_t$ for some $s\neq t$, then $R(\lambda_I)=0$. Since the degree of $F(X)$ is not greater than $k(n-1)$, so the coefficient of $x_1^{n-1}\ldots x_k^{n-1}$ in $R(X)$ is equal to that in $F(X)$. Thus the coefficient of $x_1^{n-1}\ldots x_k^{n-1}$ in $F(X)$ is equal to
$$k!\sum_{I\subset[n],|I|=k}\frac{F(\lambda_I)}{\displaystyle\prod_{i\in I}\prod_{j\neq i}(\lambda_i-\lambda_j)}.$$
For each $I=\{i_1,\ldots,i_k\} \subset [n]$, we have
$$F(\lambda_I) = P(\lambda_I)\prod_{i\in I}\prod_{j\in I,j\neq i}(\lambda_i-\lambda_j),$$
and
$$\prod_{i\in I}\prod_{j\neq i}(\lambda_i-\lambda_j) = \prod_{i\in I}\prod_{j\in I^c}(\lambda_i-\lambda_j)\prod_{i\in I}\prod_{j\in I,j\neq i}(\lambda_i-\lambda_j).$$
This implies that the coefficient of $x_1^{n-1}\ldots x_k^{n-1}$ in $F(X)$ is equal to
$$k!\sum_{I\subset [n],|I|=k}\frac{P(\lambda_I)}{\displaystyle\prod_{i\in I}\prod_{j\in I^c}(\lambda_i-\lambda_j)}.$$
Theorem \ref{main} is proved as desired.
\end{proof}
\begin{remark}
If $P(x_1,\ldots,x_k)$ is a symmetric polynomial whose partial degrees are not greater than $n-k$, then we have the following formula, which is proved by Chen and Louck \cite[Theorem 2.1]{CL},
$$P(X) = \sum_{I\subset [n],|I|=k}P(\lambda_I)\frac{\displaystyle\prod_{x_i\in X}\prod_{j\in I^c}(x_i-\lambda_j)}{\displaystyle\prod_{i\in I}\prod_{j\in I^c}(\lambda_i-\lambda_j)}.$$
Using this interpolation formula, we get
$$\sum_{I\subset [n],|I|=k}\frac{P(\lambda_I)}{\displaystyle\prod_{i\in I}\prod_{j\in I^c}(\lambda_i-\lambda_j)} = d(k,n),$$
where $d(k,n)$ is the coefficient of $x_1^{n-k}\ldots x_k^{n-k}$ in $P(x_1,\ldots,x_k)$. Indeed, this is a special case of Theorem \ref{main}. It is known that $k!$ is the coefficient of $x_1^{k-1}\ldots x_k^{k-1}$ in the polynomial
$$\prod_{i=1}^k\prod_{j \neq i}(x_i-x_j).$$
This is proved in \cite{Ze}. If the partial degrees of $P(x_1,\ldots,x_k)$ are not greater than $n-k$, then we get
$$c(n,k) = d(k,n)k!.$$
\end{remark}
\begin{proof}[Proof of Theorem \ref{thm3}] Set $$F(X,Y) = P(X,Y)\prod_{i=1}^k\prod_{j \neq i}(x_i-x_j)\prod_{i=1}^{n-k}\prod_{j \neq i}(y_i-y_j)(Y-X).$$
By the assumption, the degree of $F(X,Y)$ is not greater than
$$k(n-k) + k(k-1) + (n-k)(n-k-1) + k(n-k) = n(n-1).$$
By the division algorithm for multivariate polynomials (see \cite[Theorem 3]{CLO}), there exist the polynomials $F_i(X,Y), i = 1, \ldots,k, G_j(X,Y), j = 1,\ldots,n-k$, and $R(X,Y)$ such that
$$R(X,Y) = F(X,Y) - \sum_{i=1}^kF_i(X,Y)\prod_{l=1}^n(x_i-\lambda_l) - \sum_{j=1}^{n-k}G_j(X,Y)\prod_{l=1}^n(y_j-\lambda_l),$$
and all partial degrees of $R(X,Y)$ are not greater that $n-1$. By the Lagrange interpolation formula as in the proof of Theorem \ref{main}, we have
$$R(X,Y) = \sum_{i_1,\ldots,i_k,j_1,\ldots,j_{n-k}=1}^nR(\lambda_I,\lambda_J)\prod_{l=1}^kL_{i_l}(x_l)\prod_{l=1}^{n-k}L_{j_l}(y_l).$$
For each $I=\{i_1,\ldots,i_k\}$ and $J=\{j_1,\ldots,j_{n-k}\}$, we have
$$R(\lambda_I,\lambda_J) = F(\lambda_I,\lambda_J),$$
and if $i_s = i_t$ or $j_s=j_t$ or $i_s = j_t$ for some $s\neq t$, then $R(\lambda_I,\lambda_J)=0$. Since the degree of $F(X,Y)$ is not greater than $n(n-1)$, so the coefficient of $x_1^{n-1}\ldots x_k^{n-1}y_1^{n-1}\ldots y_{n-k}^{n-1}$ in $R(X,Y)$ is equal to that in $F(X,Y)$. Thus the coefficient of $x_1^{n-1}\ldots x_k^{n-1}y_1^{n-1}\ldots y_{n-k}^{n-1}$ in $F(X,Y)$ is equal to
$$k!(n-k)!\sum_{I\subset [n],|I|=k}\frac{F(\lambda_I,\lambda_{I^c})}{\displaystyle\prod_{i\in I}\prod_{j\neq i}(\lambda_i-\lambda_j)\prod_{i\in I^c}\prod_{j\neq i}(\lambda_i-\lambda_j)}.$$
For each $I=\{i_1,\ldots,i_k\} \subset [n]$, we have
\begin{align*}
F(\lambda_I,\lambda_{I^c}) & = P(\lambda_I,\lambda_{I^c})\prod_{i\in I}\prod_{j\in I,j\neq i}(\lambda_i-\lambda_j)\prod_{i\in I^c}\prod_{j\in I^c,j\neq i}(\lambda_i-\lambda_j)(\lambda_{I^c} - \lambda_I)\\
& = P(\lambda_I, \lambda_{I^c})\prod_{i\in I}\prod_{j\in I,j\neq i}(\lambda_i-\lambda_j)\prod_{i\in I^c}\prod_{j\neq i}(\lambda_i-\lambda_j),
\end{align*}
and
$$\prod_{i\in I}\prod_{j\neq i}(\lambda_i-\lambda_j) = \prod_{i\in I}\prod_{j\in I^c}(\lambda_i-\lambda_j)\prod_{i\in I}\prod_{j\in I,j\neq i}(\lambda_i-\lambda_j).$$
This implies that the coefficient of $x_1^{n-1}\ldots x_k^{n-1}y_1^{n-1}\ldots y_{n-k}^{n-1}$ in $F(X,Y)$ is equal to
$$k!\sum_{I\subset [n],|I|=k}\frac{P(\lambda_I,\lambda_{I^c})}{\displaystyle\prod_{i\in I}\prod_{j\in I^c}(\lambda_i-\lambda_j)}.$$
Theorem \ref{thm3} is proved as desired.
\end{proof}
\begin{remark}
Theorem \ref{thm3} is more general than Theorem \ref{main}. Indeed, if $P(x_1\ldots,x_k)$ is a symmetric polynomial, then it is also a doubly symmetric polynomial. Theorem \ref{thm3} says that the sum
$$\sum_{I\subset [n],|I|=k}\frac{P(\lambda_I)}{\displaystyle\prod_{i\in I}\prod_{j\in I^c}(\lambda_i-\lambda_j)} = \frac{d(k,n)}{k!(n-k)!},$$
where $d(k,n)$ is the coefficient of $x_1^{n-1}\ldots x_k^{n-1}y_1^{n-1}\ldots y_{n-k}^{n-1}$ in the polynomial
$$P(x_1,\ldots,x_k)\prod_{i=1}^k\prod_{j \neq i}(x_i-x_j)\prod_{i=1}^{n-k}\prod_{j \neq i}(y_i-y_j)\prod_{i=1}^{n-k}\prod_{j=1}^k(y_i - x_j).$$
It is known that $(n-k)!$ is the coefficient of $y_1^{n-k-1}\ldots y_{n-k}^{n-k-1}$ in the polynomial
$$\prod_{i=1}^{n-k}\prod_{j \neq i}(y_i-y_j).$$
This is proved in \cite{Ze}. Thus $(n-k)!$ is also the coefficient of $y_1^{n-1}\ldots y_{n-k}^{n-1}$ in the polynomial
$$\prod_{i=1}^{n-k}\prod_{j \neq i}(y_i-y_j)\prod_{i=1}^{n-k}\prod_{j=1}^k(y_i - x_j).$$
This means that
$$\frac{d(k,n)}{(n-k)!} = c(k,n),$$
which is the coefficient of $x_1^{n-1}\ldots x_k^{n-1}$ in the polynomial
$$P(x_1,\ldots,x_k)\prod_{i=1}^k\prod_{j \neq i}(x_i-x_j).$$
The statement of Theorem \ref{main} is obtained as desired.
\end{remark}
\section{Localization in equivariant cohomology}
In this section, we recall some basic definitions and results in the theory of equivariant cohomology. For more details on this theory, we refer to \cite{AB, BV, Bo, Br, CK, EG1, EG2}. Throughout we consider all cohomologies with coefficients in the complex field $\mathbb C$.
Let $T = (\mathbb C^*)^n$ be an algebraic torus of dimension $n$, classified by the principal $T$-bundle $ET \to BT$, whose total space $ET$ is contractible. Let $X$ be a compact space endowed with a $T$-action. Put $X_T = X\times_T ET$, which is itself a bundle over $BT$ with fiber $X$. Recall that the $T$-equivariant cohomology of $X$ is defined to be $H^*_{T}(X) = H^*(X_T)$, where $H^*(X_T)$ is the ordinary cohomology of $X_T$. Note that $H^*_T(\pt) = H^*(BT)$. By pullback via the map $X\to \pt$, we see that $H^*_T(X)$ is an $H^*(BT)$-module. Thus we may consider $H^*(BT)$ as the coefficient ring for equivariant cohomology.
A $T$-equivariant vector bundle is a vector bundle $E$ on $X$ together with a lifting of the action on $X$ to an action on $E$ which is linear on fibers. Note that $E_T$ is a vector bundle over $X_T$.
The $T$-equivariant Chern classes $c_i^T(E) \in H^*_T(X)$ are defined to be the Chern classes $c_i(E_T)$. If $E$ has rank $r$, then the top Chern class $c_r^T(E)$ is called the $T$-equivariant Euler class of $E$ and is denoted $e^T(E) \in H^*_T(X)$. More generally, the $T$-equivariant characteristic class $c^T(E) \in H^*_T(X)$ is defined to be the characteristic class $c(E_T)$.
Let $\chi(T)$ be the character group of the torus $T$. For each $\rho \in \chi(T)$, let $\mathbb C_{\rho}$ denote the one-dimensional representation of $T$ determined by $\rho$. Then $L_{\rho} = (\mathbb C_{\rho})_T$ is a line bundle over $BT$, and the assignment $\rho \mapsto -c_1(L_{\rho})$ defines an group isomorphism $f: \chi(T) \simeq H^2(BT)$, which induces a ring isomorphism $\Sym(\chi(T)) \simeq H^*(BT)$. We call $f(\rho)$ the weight of $\rho$. In particular, we denote by $\lambda_i$ the weight of $\rho_i$ defined by $\rho_i(x_1,\ldots,x_n) = x_i$. We thus obtain an isomorphism
$$H^*_T(\pt) = H^*(BT) \simeq \mathbb C[\lambda_1,\ldots,\lambda_n].$$
Let $\mathcal R_T \simeq \mathbb C(\lambda_1,\ldots,\lambda_n)$ be the field of fractions of $\mathbb C[\lambda_1,\ldots,\lambda_n]$. An important result in equivariant cohomology is the localization theorem. Historically, localization in equivariant cohomology was studied by Borel \cite{Bo} and then further investigated by Quillen \cite{Q}, Atiyah-Bott \cite{AB}, and Berline-Vergne \cite{BV}. Among many versions of the formulation of the localization theorem, we choose the one by Atiyah and Bott \cite{AB}.
\begin{theorem} [Atiyah-Bott \cite{AB}]
Let $X^T$ be the fixed point locus of the torus action. Then the inclusion $i : X^T \hookrightarrow X$ induces an isomorphism
$$i^* : H^*_T(X) \otimes \mathcal R_T \simeq H^*_T(X^T)\otimes \mathcal R_T.$$
\end{theorem}
Moreover, Atiyah and Bott \cite{AB} also gave an explicit formula for the inverse isomorphism. If $X$ is a compact manifold and $X^T$ is finite, then the localization theorem can be rephrased as follows:
\begin{theorem} [Atiyah-Bott \cite{AB}, Berline-Vergne \cite{BV}]
Suppose that $X$ is a compact manifold endowed with a torus action and the fixed point locus $X^T$ is finite. For $\alpha \in H^*_T(X)$, we have
\begin{equation}\label{ABBV}
\int_X\alpha = \sum_{p\in X^T}\frac{\alpha|_p}{e_p},
\end{equation}
where $e_p$ is the $T$-equivariant Euler class of the tangent bundle at the fixed point $p$, and $\alpha|_p$ is the restriction of $\alpha$ to the point $p$.
\end{theorem}
For many applications, the Atiyah-Bott-Berline-Vergne formula can be formulated in more down-to-earth terms. We are mainly interested in the computation of integrals over Grassmannians.
\begin{proof}[Proof of Theorem \ref{integral1}]
Consider the action of $T = (\mathbb C^*)^n$ on $\mathbb C^n$ given in coordinates by
$$(a_1,\ldots,a_n) \cdot (x_1, \ldots , x_n) = (a_1x_1, \ldots , a_nx_n).$$
This induces a torus action on the Grassmannian $G(k,n)$ with isolated fixed points $p_I$ corresponding to coordinate $k$-planes in $\mathbb C^n$. Each fixed point $p_I$ is indexed by a subset $I\subset [n]$ of size $k$.
For each $p_I$, the torus action on the fibers $\mathcal S|_{p_I}$ and $\mathcal Q|_{p_I}$ have the characters $\rho_i$ for $i\in I$ and $\rho_j$ for $j\in I^c$ respectively. Since the tangent bundle is isomorphic to $\mathcal S^\vee \otimes \mathcal Q$, the characters of the torus action on the tangent bundle at $p_I$ are
$$\{\rho_j - \rho_i \mid i \in I, j \in I^c \}.$$
Thus the $T$-equivariant Euler class of the tangent bundle at $p_I$ is
$$e_{p_I} = \prod_{i\in I}\prod_{j\in I^c}(\lambda_j-\lambda_i) = (-1)^{k(n-k)}\prod_{i\in I}\prod_{j\in I^c}(\lambda_i-\lambda_j).$$
By the assumption, the $T$-equivariant characteristic classes at $p_I$ are
$$\Phi^T(\mathcal S|_{p_I}) = P(\lambda_I) \texttt{ and } \Psi^T(\mathcal Q|_{p_I}) = Q(\lambda_{I^c}).$$
By the Atiyah-Bott-Berline-Vergne formula, we have
\begin{align*}
\int_{G(k,n)}\Phi(\mathcal S) & = \sum_{p_I}\frac{\Phi^T(\mathcal S|_{p_I})}{e_{p_I}}\\
& = (-1)^{k(n-k)}\sum_{I\subset[n],|I|=k}\frac{P(\lambda_I)}{\displaystyle\prod_{i\in I}\prod_{j\in I^c}(\lambda_i-\lambda_j)},
\end{align*}
\begin{align*}
\int_{G(k,n)}\Psi(\mathcal Q) & = \sum_{p_I}\frac{\Psi^T(\mathcal Q|_{p_I})}{e_{p_I}}\\
& = \sum_{I\subset[n],|I|=k}\frac{Q(\lambda_{I^c})}{\displaystyle\prod_{i\in I}\prod_{j\in I^c}(\lambda_j-\lambda_i)}\\
& = \sum_{I\subset[n],|I|=n-k}\frac{Q(\lambda_{I})}{\displaystyle\prod_{i\in I}\prod_{j\in I^c}(\lambda_i-\lambda_j)},
\end{align*}
Combining with Theorem \ref{main}, Theorem \ref{integral1} is proved as desired.
\end{proof}
\begin{proof}[Proof of Theorem \ref{integral2}]
By the assumption, the $T$-equivariant characteristic class at $p_I$ is
$$\Delta^T(\mathcal S|_{p_I},\mathcal Q|_{p_I}) = P(\lambda_I,\lambda_{I^c}).$$
By the Atiyah-Bott-Berline-Vergne formula, we have
\begin{align*}
\int_{G(k,n)}\Delta(\mathcal S,\mathcal Q) & = \sum_{p_I}\frac{\Delta^T(\mathcal S|_{p_I},\mathcal Q|_{p_I})}{e_{p_I}}\\
& = (-1)^{k(n-k)}\sum_{I\subset[n],|I|=k}\frac{P(\lambda_I,\lambda_{I^c})}{\displaystyle\prod_{i\in I}\prod_{j\in I^c}(\lambda_i-\lambda_j)}.
\end{align*}
Combining with Theorem \ref{thm3}, Theorem \ref{integral2} is proved as desired.
\end{proof}
\subsection*{Acknowledgements}
Part of this work was done while the author was visiting the Korea Institute for Advanced Study (KIAS). This work is finished during the author's postdoctoral fellowship at the National Center for Theoretical Sciences (NCTS), Taipei, Taiwan. He thanks all for the financial support and hospitality.
| b5098d949a782cf58171d6a3778593ad0ca0f95c | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section*{Abstract}
We present a spectrally accurate method for the rapid evaluation of free-space Stokes
potentials, i.e. sums involving a large number of free space Green's
functions. We consider sums involving stokeslets, stresslets and
rotlets that appear in boundary integral methods and potential methods for solving Stokes
equations.
The method combines the framework of the Spectral Ewald method for
periodic problems \cite{Lindbo2010}, with a very recent approach to solving the
free-space harmonic and biharmonic equations using fast Fourier
transforms (FFTs) on a uniform grid \cite{Vico2016}.
Convolution with a truncated Gaussian function is used to place point
sources on a grid.
With precomputation of a scalar grid quantity that does not depend on these sources, the amount of
oversampling of the grids with Gaussians can be kept at a factor of two,
the minimum for aperiodic convolutions by FFTs.
The resulting algorithm has a computational complexity of $O(N \log
N)$ for problems with $N$ sources and targets.
Comparison is made with a fast multipole method (FMM) to show that the
performance of the new method is competitive.
\noindent\makebox[\linewidth]{\rule{\textwidth}{1pt}}
\setcounter{tocdepth}{2}
\section{Introduction}
In this paper, we consider the evaluation of free-space
potentials of Stokes flow, i.e. vector fields defined by sums involving a large
number of free space Green's functions such as the so-called stokeslet, stresslet or rotlet.
The stokeslet is the free space Green's function for velocity, and is given by
\[
S(\vec{r})=\frac{1}{r} \vec{I} + \frac{1}{r^3} \vec{r} \vec{r}, \mbox{ or }
S_{jl}(\vec{r})=\frac{\delta_{jl}}{r} + \frac{r_j r_l}{r^3}, \quad j,l=1,2,3,
\]
with $r=|\vec{r}|$ and where $\delta_{jl}$ is the Kronecker delta.
The stresslet and rotlet will be introduced in the following.
The discrete sums are on the form
\begin{equation}
\vec{u}(\vec{x}_{\texttt{m}})=
\sum_{\substack{\ni=1 \\\ni \ne \texttt{m}}}^{N}
S(\vec{x}_{\texttt{m}}-\vec{x}_{\ni}) \vec{f}(\vec{x}_{\ni}), \quad \texttt{m}=1,\ldots,N.
\label{eqn:usumfree}
\end{equation}
and appear in boundary integral methods and potential methods for solving Stokes
equations.
These sums have the same structure as the classical Coulombic or
gravitational $N$-body problems that involve the harmonic kernel, and
the direct evaluation of such a sum for $\texttt{m}=1,\ldots,N$ requires
$O(N^2)$ work. The Fast Multipole Method (FMM) can reduce that cost
to $O(N)$ work, where the constant multiplying $N$ will depend on the
required accuracy. FMM was first introduced by Greengard and Rokhlin
for the harmonic kernel in 2D and later in 3D
\cite{Greengard1987,Cheng1999} and has since been extended to other
kernels, including the fundamental solutions of Stokes flow considered
here \cite{Rodin2000,Duraiswami2006,Tornberg2008,Wang2007,Ying2004}. Related
is also the development of a so called pre-corrected FFT method based
on fast Fourier transforms. This method has been applied to the rapid
evaluation of stokeslet sums for panel-based discretizations of
surfaces \cite{White2006}.
For periodic problems, FFT-based fast methods built on the foundation
of so-called Ewald summation have been succesful. Also here,
development started for the harmonic potential, specifically for
evaluation of the electrostatic potential and force in connection to
molecular dynamic simulations, see e.g the survey by Deserno and Holm
\cite{Deserno1998}. One early method was the {\em Particle Mesh Ewald
} (PME) method by Darden et al. \cite{Darden1993}, later refined to
the {\em Smooth Particle Mesh Ewald } (SPME) method by Essman et
al. \cite{Essmann1995}. The SPME method was extended to the fast
evaluation of the stokeslet sum by Saintillan et
al. \cite{Saintillan2005}. To recover the exponentially fast
convergence of the Ewald sums that is lost when such a traditional PME
approach is used, the present authors have developed a spectrally
accurate PME-type method, the Spectral Ewald (SE) method both for the
sum of stokeslets \cite{Lindbo2010}, and stresslets
\cite{AfKlinteberg2014a}. It has also been implemented for the sum of
rotlets \cite{AfKlinteberg2016rot}, and the source code is available online \cite{se_github}.
The Spectral Ewald method was recently used to accelerate the
Stokesian Dynamics simulations in \cite{Wang2016}.
The present work deals with the efficient and fast summation of free
space Green's functions for Stokes flow (stokeslets, stresslets and
rotlets), as exemplified by the sum of stokeslets in
\eqref{eqn:usumfree}. The problem has no periodicity, but the
approach will still be based on Ewald summation and fast Fourier
transforms (FFTs), using ideas from \cite{Vico2016} to extend the
Fourier treatment to the free-space case. Before we
explain this further, we will introduce the idea behind Ewald
summation.
\subsection{Triply periodic Ewald summation}
Consider the Stokes equations in $\mathbb R^3$, singularly forced at
arbitrary locations $\vec{x}_{\ni}$, $\ni=1,\ldots,N$, with strengths $8
\pi \mu \vec{f}(\vec{x}_{\ni})\in \mathbb R^3$ (with the $8 \pi \mu$ scaling for
convenience). Introduce the three dimensional delta function
$\delta(\vec{x}-\vec{x}_0)$, and write
\begin{align}
-\nabla p + \mu \nabla^2 \vec{u} + \vec{g}(\vec{x}) & =0, \quad
\vec{g}(\vec{x})= 8 \pi \mu \sum_{\ni=1}^N \vec{f}(\vec{x}_{\ni}) \, \delta(\vec{x}-\vec{x}_{\ni}),
\quad \quad
\quad \quad \quad \quad
\label{eq:Stokes_sing} \\
\nabla \cdot \vec{u} & = 0,
\notag
\end{align}
where $\vec{u}$ is the velocity, $p$ is the pressure and $\mu$ is the
viscosity.
The free-space solution to this problem, evaluated at the source
locations, is given by \eqref{eqn:usumfree}.
The classical Ewald summation formulas were derived for the triply
periodic problem for the electrostatic potential by Ewald in 1921
\cite{Ewald1921} and for the stokeslet by Hasimoto in 1959
\cite{Hasimoto1959}. Here, assume that all the point forces are
located within a box $\mathcal{D}=[-L_1 /2, L_1 /2] \times [-L_2 /2, L_2
/2] \times [-L_3 /2, L_3 /2]$, and that we impose periodic boundary
conditions. The solution to this problem is a sum not only over all
the point forces, but also over all their periodic replicas,
\begin{equation}
\vec{u}^{3P}(\vec{x}_{\texttt{m}})= \sum_{\vec{p} \in P_3} \sum_{\ni=1}^{N*}
S(\vec{x}_{\texttt{m}}-\vec{x}_{\ni}+\vec{p} ) \vec{f}(\vec{x}_{\ni}), \quad \texttt{m}=1,\ldots,N.
\label{eqn:usumper}
\end{equation}
Here, the sum over $\vec{p}$ formalizes the periodic replication of the
point forces with
\begin{equation}
P_3 =\{ (jL_1,lL_2,pL_3\}: (j,l,p) \in \mathbb{Z}^3 \}. \quad
\label{eqn:perdef}
\end{equation}
The ${N*}$ indicates that the term ($\ni=\texttt{m}$, $\vec{p}={\bf 0}$) is excluded
from the sum.
The slow decay of the stokeslet however makes this infinite sum
divergent.
To make sense of this summation, one usually assumes that the point
forces are balanced by a mean pressure gradient, such that the
velocity integrates to zero over the periodic box.
Under these assumptions, Hasimoto \cite{Hasimoto1959} derived the following Ewald
summation formula
\begin{align}
\vec{u}^{3P}(\vec{x}_{\texttt{m}}) = &
\sum_{\vec{p} \in P_3}\ \sum_{\ni=1}^{N*} S^R(\vec{x}_{\texttt{m}} - \vec{x}_{\ni} + \vec{p},\xi) \vec{f}(\vec{x}_{\ni}) +
\frac{1}{V} \sum_{\vec{k} \neq 0} \hat{S}^F(\vec{k},\xi)
\sum_{\ni=1}^N \vec{f}(\vec{x}_{\ni}) \expk{(\vec{x}_{\texttt{m}} - \vec{x}_{\ni})}
\notag \\
& + \lim_{|\vec{r}|\rightarrow 0} \left( S^{R}(\vec{r},\xi)-S(\vec{r}) \right)\vec{f}(\vec{x}_{\texttt{m}}),
\label{eq:usum_SR_SF_limit}
\end{align}
where the $\ni=\texttt{m}$, $\vec{p}=0$ term is excluded from the real space
sum, $V=L_1 L_2 L_3$, and
\begin{equation}
S^R(\vec{r}, \xi) = 2\left( \frac{\xi e^{-\xi^2 r^2} }{\sqrt{\pi} r^2} +
\frac{ \operatorname{erfc}{(\xi r)} }{2 r^3} \right) (r^2 \vec{I} + \vec{r}\rb) -
\frac{4\xi}{\sqrt{\pi}} e^{-\xi^2 r^2} \vec{I}, \quad ,
\label{eq:SR_Hasimoto}
\end{equation}
\begin{equation}
\hat{S}^F(\vec{k},\xi)=
8\pi \left(1 + \frac{k^2}{4\xi^2} \right)\frac{1}{k^4}(\vec{I} k^2 - \vec{k}
\vec{k}) e^{-k^2/4\xi^2} ,
\label{eq:ShatH}
\end{equation}
with $r = | \vec{r}|, \ k=|\vec{k}|$,
\begin{align}
\vec{k} \in \mathbb K = \{ 2\pi(j_1/L_1, j_2/L_2, j_3/L_3) : \v j \in \mathbb{Z}^3 \}
\end{align}
and
\begin{equation}
\lim_{|\vec{r}|\rightarrow 0} \left( S^{R}(\vec{r},\xi)-S(\vec{r})
\right)=-\frac{4 \xi}{\sqrt{\pi}} \vec{I}.
\label{eq:Slimit}
\end{equation}
The last term in \eqref{eq:usum_SR_SF_limit} is commonly referred to as the self-interaction
term. When evaluating the potential at $\vec{x}_{\texttt{m}}$, we should exclude
the contribution from the point force at that same location.
For the real space part, we can directly
skip the term in the summation when $\vec{p}=0$ and $\ni=\texttt{m}$.
We however need to subtract the contribution from this point that has
been included in the Fourier sum.
We can use that $S^F=S-S^R$, and subtract the
limit as $|\vec{r}| \rightarrow 0$ \eqref{eq:usum_SR_SF_limit}. Both $S$ and $S^R$ are singular, but the limit
of the difference is finite \eqref{eq:Slimit}.
Both sums now decay exponentially, one in real space and one in
Fourier space. The parameter $\xi>0$ is a decomposition parameter that controls the
decay of the terms in the two sums. The sum in real space can naturally be truncated to
exclude interactions that are now negligible. The sum in $k$-space
however, is still a sum of complexity $O(N^2)$, now with a very large
constant introduced by the sum over $\vec{k}$.
Methods in the PME family make use of FFTs to evaluate the $k$-space
sum, accelerating the evaluation such that $\xi$ can be chosen
larger to push more work into the $k$-space sum, allowing for tighter
truncation of the real space sum, and in total an $O(N \log N)$
method. This procedure introduces approximations since a grid must be
used and, as with the FMM, the constant multiplying $N \log
N$ will depend on the accuracy requirements.
\subsection{The free-space problem and this contribution}
Considering the free space problem, we can introduce the same kind of
decomposition as in \eqref{eq:usum_SR_SF_limit}.
The real space sum stays the same, with the minor
change that the sum over $\vec{p}$ is removed, and the self interaction
term does not change.
However,
the discrete sum in Fourier space is replaced by the inverse
Fourier transform,
\begin{align}
\v u^F(\vec{x},\xi) = \frac{1}{(2\pi)^3} \int_{\mathbb R^3}
\hat{S}^F(\v k, \xi)
\cdot \sum_{\ni=1}^N \vec{f}(\vec{x}_{\ni}) e^{i\v k\cdot (\vec{x}-\v
x_{\ni})} \dif \v k .
\label{eq:invtransf}
\end{align}
Here, note the $1/k^2$ singularity in $\hat{S}^F(\v k, \xi)$ as
defined in \eqref{eq:ShatH}. The integral is well defined, and
integration can be performed e.g. in spherical coordinates. A numerical
quadrature method in spherical coordinates would however require
non-uniform FFTs for non-rectangular grids in $k$-space.
Instead, we will use a very recent idea introduced by Vico et
al. \cite{Vico2016} to solve free space problems by FFTs on uniform
grids.
The method by Vico et al. \cite{Vico2016} is based on the idea to use
a modified Green's function. With a right hand side of compact
support, and a given domain inside which the solution is to be found,
a truncated Green's function can be defined that coincides with the
original one for a large enough domain (and is zero elsewhere), such
that the analytical solution defined through a convolution of the
Green's function with the right hand side remains unchanged.
The gain is that the Fourier transform of this truncated Green's
function will have a finite limit at $\vec{k}=0$. A length scale related
to the truncation will however be introduced, introducing oscillations
in Fourier space which will require some upsampling to resolve.
The authors of \cite{Vico2016} present this approach for radial Green's functions, e.g.
the harmonic and biharmonic kernels. In the present work, we are
considering kernels that are not radial. We will however use this idea
in a substep of our method, defining the Fourier
transform of the biharmonic (for stokeslet and stresslet) or harmonic
(for rotlet) kernels, and define our non-radial kernels from these.
The need of upsampling that the truncation brings can be taken care of
in a precomputation step, and hence for a scalar quantity only. What remains
is an aperiodic discrete convolution that requires an upsampling of a
factor of two.
The key ingredients in our method for the rapid summation of
kernels of Stokes flow (stokeslet, stresslet and rotlet) in free
space will hence be the following.
We make use of the framework of Ewald summation, to split the
sums in two parts - one that decays rapidly in real space, and one
in Fourier space. The Fourier space treatment is based on the
Spectral Ewald method for triply and doubly periodic problems that has
been developed previously \cite{AfKlinteberg2014a,Lindbo2010,
Lindbo2011c,Lindbo2011e}. This means that point forces will be
interpolated to a uniform grid using truncated Gaussian functions that
are scaled to allow for best possible accuracy given the size of the
support. The implementation of the gridding is made efficient by the
means of Fast Gaussian Gridding (FGG) \cite{Greengard2004,Lindbo2011c}.
In the periodic problem, an FFT of each component of the grid function
is computed, a scaling is done in Fourier space, and after inverse
FFTs, truncated Gaussians are again used to evaluate the result at any
evaluation point. The new development in this paper is to extend this
treatment to the free space case, when periodic sums are replaced by
discretized Fourier integrals. As mentioned above, a precomputation
will be made to compute a modified free-space harmonic or biharmonic
kernel that will be used to define the scaling in Fourier space.
The details are yet to be explained, but as we hope to convey in the
following, the method that we develop here for potentials of Stokes
flow, can easily be extended to other kernels.
For any kernel that can be expressed as a differentiation of the harmonic and/or biharmonic
kernel, the Ewald summation formulas can easily be derived and only
minor changes in the implementation of the method will be needed.
Any method based on Ewald summation and acceleration by FFTs will be
most efficient in the triply periodic case. As soon as there is one or
more directions that are not periodic, there will be a need of some
oversampling of FFTs, which will increase the computational cost. For
the FMM, the opposite is true. The free space problem is the fastest
to compute, any periodicity will invoke an additional cost, which will
become substantial or even overwhelming if the base periodic box has a
large aspect ratio. Hence, implementing the FFT-based Spectral Ewald
method for a free-space problem and comparing it to an FMM method will
be the worst possible case for the SE method.
Still, as we will show in the results section, using an open source
implementation of the FMM \cite{fmm_webpage}, our new
method is competitive and often performs better than that implementation of the FMM for
uniform point distributions (one can however expect this adaptive FMM to
perform better for highly non-uniform distributions).
There is an additional value in having a method that can be used for
different periodicities, thereby keeping the structure intact and
easing the integration with the rest of the simulation code,
concerning e.g. modifications of quadrature methods in a boundary
integral method to handle near interactions. A three dimensional
adaptive FMM is also much more intricate to implement than the SE
method. Open source software for the Stokes FMM does exist for the
free space problem (as the one used here), but we are not
aware of any software for the periodic problem.
\subsection{Outline of paper}
The outline of the paper is as follows. In section
\ref{sect:freeGreen} we start by introducing the stokeslet, stresslet
and rotlet, and write them on the operator form that we will later
use.
In section \ref{sect:Ewaldsum} we introduce the ideas behind
Ewald decomposition, and establish a framework for straight forward
derivation of decompositions of different kernels.
The new approach to solve free-space problems by
FFTs introduced by Vico et al. \cite{Vico2016} is presented
in the following section, together with a
detailed discussion on oversampling needs and precomputation. The new
method for evaluating the Fourier space component is described in
section \ref{sect:kspace}, while the evaluation of the real space sum
is briefly commented on in section \ref{sec:eval-real-space}.
New truncation error estimates are derived in section
\ref{sec:truncation-errors}, and in section \ref{sect:summary} we
summarize the full method. Numerical results are presented in section
\ref{sect:numres}, where the performance of the method is discussed
and comparison to an open source implementation of the FMM
\cite{fmm_webpage} is made.
\section{Green's functions of free-space Stokes flow}
\label{sect:freeGreen}
We will consider three different Green's functions of free-space
Stokes flow, the stokeslet $S$, the stresslet $T$
and the rotlet $\Omega$. They are defined as
\begin{align}
S_{jl}(\v r) &= \frac{\delta_{jl}}{r} + \frac{r_jr_l}{r^3},
\label{eq:def_stokeslet} \\
T_{jlm}(\v r) &= -6 \frac{r_jr_lr_m}{r^5}
\label{eq:def_stresslet}, \\
\Omega_{jl}(\v r) &= \epsilon_{jlm}\frac{r_m}{r^3},
\label{eq:def_rotlet}
\end{align}
where $r = |\v r|$. They can equivalently be
formulated as operators acting on $r=|\vec{r}|$ \cite{Pozrikidis1996,Fan1998},
and in the case of the rotlet also on $1/r$,
\begin{align}
S_{jl} (\v r)&= \left(\delta_{jl}\nabla^2 - \nabla_j\nabla_l\right) r,
\label{eq:stokeslet_op}\\
T_{jlm} (\v r)&= \left[ \left(
\delta_{jl}\nabla_m+\delta_{lm}\nabla_j+\delta_{mj}\nabla_l
\right) \nabla^2 - 2\nabla_j\nabla_l\nabla_m \right] r,
\label{eq:stresslet_op}\\
\Omega_{jl}(\v r) &= \left(-\epsilon_{jlm}\nabla_m \nabla^2 \right)
r =\left(-2 \epsilon_{jlm}\nabla_m \right) \frac{1}{r}.
\label{eq:rotlet_op}
\end{align}
For a single forcing term, $8\pi \mu \vec{f}$ at a source location
$\vec{x}_0$, we write the solution for velocity as
\begin{equation}
\vec{u}(\vec{x})= S(\vec{x}-\vec{x}_0) \vec{f}, \mbox{ or }
u_j(\vec{x})= S_{jl}(\vec{x}-\vec{x}_0) f_l, \quad j=1,2,3,
\label{eq:vel_source_pt}
\end{equation}
where the repeated index is summed over according to Einstein summation convention.
Similarly to the velocity, the stress field and vorticity
associated with this velocity can be written,
\[
\sigma_{jl}(\vec{x})= T_{jlm}(\vec{x}-\vec{x}_0) f_m,
\quad
\omega_j(\vec{x})=\Omega_{jl}(\vec{x}-\vec{x}_0) f_l.
\]
In integral equations, the stresslet often appears instead multiplying
sources with two indices, also producing a velocity,
\[
\vec{u}_{j}(\vec{x})= T_{jlm}(\vec{x}-\vec{x}_0) f_{lm},
\]
and this is the case that we will consider here. (The typical form is
$T_{jlm} n_{l} q_m$, where ${\vec n}$ is a vector normal to a surface).
We want to rapidly evaluate discrete-sum potentials of the type given
in (\ref{eqn:usumfree}), either at the source locations as indicated
in that sum, or at any other arbitrary points, and we want to do so
for the three different Green's functions. To allow for a generic
notation in the following despite the differences, we introduce the
unconventional notation
\begin{align}
\v u(\vec{x}) = \sum_{\ni=1}^N G(\vec{x} - \vec{x}_{\ni}) \cdot \vec{f}(\vec{x}_{\ni}),
\label{eq:usum}
\end{align}
where $G$ can denote either the stokeslet $S$, the stresslet
$T$ and the rotlet $\Omega$, and the dot-notation
$\vec{u}=G(\vec{r}) \cdot \vec{f}$ will be understood to mean
\[
u_j(\vec{x})= S_{jl}(\vec{r}) f_l, \quad
u_j(\vec{x})= T_{jlm}(\vec{r}) f_{lm}, \quad
u_j(\vec{x})= \Omega_{jl}(\vec{r}) f_l, \quad j=1,2,3,
\]
in the three different cases.
\section{Ewald summation}
\label{sect:Ewaldsum}
\subsection{Decomposing the Green's function}
In Ewald summation we take a non-smooth and long-range Green's
function $G$, such as
(\ref{eq:def_stokeslet}--\ref{eq:def_rotlet}), and decompose it into two parts,
\begin{align}
G(\v r) = G^R(\v r) + G^F(\v r).
\end{align}
This is done such that $G^R$, called the real space component,
decays exponentially in $r=|\vec{r}|$. At the same time $G^F$, called Fourier
space component, decays exponentially in Fourier space. The original
example of this, derived by Ewald \cite{Ewald1921}, decomposes the
Laplace Green's function as
\begin{align}
\frac{1}{r} = \frac{\operatorname{erfc}(\xi r)}{r} + \frac{\operatorname{erf}(\xi r)}{r},
\label{eq:laplace_decomp}
\end{align}
where $\xi$ is a parameter that controls the decay rates in the real
and Fourier spaces. Here the real space component decays like
$e^{-\xi^2r^2}$, while the Fourier space component decays like
$k^{-2}e^{-k^2/4\xi^2}$. The rapid decay rates allow truncation of the
components; the real space component is reduced to local interactions
between near neighbors, while the Fourier space component is truncated
at some maximum wave number $k_{\infty}$.
There are two different ways of deriving an Ewald decomposition, which
we shall refer to as screening and splitting. In screening, one
introduces a screening function $\gamma(\vec{r}, \xi)$, $\int_{\mathbb
R^3}\gamma(\vec{r}, \xi) \dif \v r = 1$, that decays smoothly away from
zero. The Green's function is then decomposed using its convolution
with $\gamma$,
\begin{align}
G(\v r) = G(\v r) - (G * \gamma)(\v r, \xi) + (G * \gamma)(\v r, \xi),
\end{align}
such that
\begin{align}
G^R(\v r, \xi) &= G(\v r) - (G * \gamma)(\v r, \xi), \\
G^F(\v r, \xi) &= (G * \gamma)(\v r, \xi),
\end{align}
and (by the convolution theorem)
\begin{align}
\widehat{G}^F(\v k, \xi) &= \widehat{G}(\v k)\widehat\gamma(\v
k, \xi),
\label{eq:GF_Ggamma}
\end{align}
where $\widehat f$ denotes the Fourier transform of $f$,
\begin{align}
\widehat f (\v k) = {\mathcal F}[f](\v k) = \int_{\mathbb R^3} f(\vec{x})
e^{-i \v k \cdot \vec{x}} \dif \vec{x} .
\end{align}
The original Ewald decomposition \eqref{eq:laplace_decomp} can be
derived in this fashion, using the screening function
\begin{align}
\gamma_E(\v r,\xi)=\xi^3\pi^{-3/2}e^{-\xi^2r^2} \quad
\rightleftharpoons \quad \widehat \gamma_E(\v k,\xi) =
e^{-k^2/4\xi^2} ,
\end{align}
where $r=|\vec{r}|$, $k=|\vec{k}|$.
For the stokeslet \eqref{eq:def_stokeslet} an Ewald decomposition was
derived by Hasimoto \cite{Hasimoto1959}, which was later shown
\cite{Hernandez-Ortiz2007} to be equivalent to using the screening
function
\begin{align}
\gamma_H(\v r,\xi)=\xi^3\pi^{-3/2}e^{-\xi^2r^2} \left(\frac{5}{2}-\xi
^2 r^2\right) \quad \rightleftharpoons \quad \widehat
\gamma_H(\v k,\xi) = e^{-k^2/4\xi^2}
\left(1+\frac{1}{4}\frac{k^2}{\xi^2}\right) .
\label{eq:hasimoto}
\end{align}
In splitting, one starts with the operator form of the Green's
function (\ref{eq:stokeslet_op}--\ref{eq:rotlet_op}), $G = \operatorname{K} r$,
and splits the Green's function using a splitting function
$\Phi$,
\begin{align}
G(\v r) = \operatorname{K} [r-\Phi(r, \xi)] + \operatorname{K}\Phi(r, \xi),
\end{align}
such that
\begin{align}
G^R(\v r,\xi) & = \operatorname{K}[r-\Phi(r,\xi)], \\
\widehat{G}^F(\v k, \xi) &= \widehat\operatorname{K}(\v k)\widehat\Phi(k,
\xi),
\label{eq:GF_Khat_Phihat}
\end{align}
where $\widehat\operatorname{K}(\v k)$ denotes $\operatorname{K}$ applied to $e^{-i\v k \cdot \vec{x}}$
(e.g. if $\operatorname{K}=\Delta$ then $\widehat\operatorname{K}=-|\v k|^2=-k^2$). The splitting method
was invented by Beenakker \cite{Beenakker1986}, who used
\begin{align}
\Phi_B(r,\xi) = r \operatorname{erf}(\xi r) \quad \rightleftharpoons \quad
\widehat\Phi_B(k, \xi) = -\frac{8\pi}{k^4}\left( 1 +
\frac{1}{4}\frac{k^2}{\xi^2} + \frac{1}{8}\frac{k^4}{\xi^4}
\right) e^{-k^2/4\xi^2}.
\end{align}
We have now defined $ \widehat{G}^F(\v k, \xi)$ in two different
ways in (\ref{eq:GF_Ggamma}) and (\ref{eq:GF_Khat_Phihat}), and can
equate the two. We have that $G=K|\vec{r}|=Kr$, where
$B(\vec{r})=|\vec{r}|=r$ is the fundamental solution of the biharmonic
equation, i.e.
\[
\nabla^4 B(\vec{r})=- 8\pi \delta(\vec{x}).
\]
From this, we get
\begin{equation}
\hat{G}(\vec{k}) =\widehat{B}(\vec{k})
\hat{K}(\vec{k})=-\frac{8\pi}{k^4}\hat{K}(\vec{k}).
\label{eq:Ghat_wBhat}
\end{equation}
Hence, the screening and splitting methods can be shown
\cite{afKlinteberg2016phd} to be related to each other as
\begin{align}
\widehat\Phi = -\frac{8\pi}{k^4}\widehat\gamma
\quad \rightleftharpoons \quad
\gamma = -\frac{1}{8\pi}\nabla^4\Phi .
\end{align}
Using this, one can derive the screening and splitting functions
related to the Ewald, Hasimoto and Beenakker decompositions, shown in
Table \ref{tab:decompositions}.
\begin{table}[htbp]
\resizebox{\textwidth}{!}{
\centering
\begin{tabular}{l|l|l|l}
& $\gamma$ & $\widehat\gamma$ & $\Phi$ \\
\hline
Ewald &
$\alpha e^{-\xi^2r^2}$
& $e^{-k^2/4\xi^2}$
& $r\operatorname{erf}(\xi r) + \frac{r\operatorname{erf}(\xi r)}{2\xi^2r^2} + \frac{e^{-\xi^2r^2}}{\sqrt{\pi } \xi }$ \\
Hasimoto &
$\alpha e^{-\xi ^2r^2} \left(\frac{5}{2}-\xi ^2 r^2\right)$
& $e^{-k^2/4 \xi^2} \left(1+\frac{1}{4}\frac{k^2}{\xi^2}\right)$
& $r\operatorname{erf}(\xi r) + \frac{e^{-\xi^2r^2}}{\sqrt{\pi } \xi }$ \\
Beenakker &
$ \alpha e^{-\xi^2 r^2} \left(10-11 \xi ^2
r^2+2 \xi ^4 r^4\right) $
& $ e^{-k^2/4\xi^2}\left( 1 + \frac{1}{4}\frac{k^2}{\xi^2} +
\frac{1}{8}\frac{k^4}{\xi^4} \right)$
& $r\operatorname{erf}(\xi r)$
\end{tabular}}
\caption{Summary of the screening and splitting functions related to the Ewald, Hasimoto and Beenakker decompositions (in that order). The constant $\alpha=\xi^3\pi^{-3/2}$, and $\xi$ is the decomposition parameter.}
\label{tab:decompositions}
\end{table}
The relations listed in the table are very useful in derivation of
Ewald summation formulas. Finding the real space part $G^R(\vec{r},\xi)$
is easiest using the splitting approach, since this only involves differentiation.
The $k$-space term is however simpler to derive with the screening
approach. Combining \eqref{eq:GF_Ggamma} and \eqref{eq:Ghat_wBhat}
it directly follows
\begin{align}
\widehat{G}^F(\v k, \xi) &= \hat{K}(\vec{k}) \widehat{B}(\vec{k})
\widehat\gamma(\v k, \xi).
\label{eq:GF_KBGhat}
\end{align}
where $\widehat\operatorname{K}(\v k)$ denotes the prefactor that is produced as
$\operatorname{K}$ is applied to $e^{-i\v k \cdot \vec{x}}$.
Considering the information in the table, we can see that the Ewald
decomposition yields the fastest decay in Fourier space. However, this
screening function can only be used if the Green's function can be
written as an operator acting on $1/r$, like the rotlet
\eqref{eq:rotlet_op}.
In this case, we can think about the splitting approach as if
splitting $1/r$ such that $\Phi$ is $\operatorname{erf}(\xi r)/r$ as in
\eqref{eq:laplace_decomp}.
If we attempt to use the Ewald screening function for the stokeslet or
stresslet, this will not produce a useful decomposition since this
screening function does not ``screen'' the point forces. The field
produced by a point force convolved with the screening function does
not converge rapidly (with distance from the source location) to the
field produced by that point force. If we were to do the calculation,
this manifests itself in slowly decaying terms in the real space sum.
Both the Hasimoto and Beenacker screening functions work for the
stokeslet and stresslet. The Hasimoto decomposition will yield
somewhat faster decaying terms in both real and Fourier space, and
will henceforth be the one that we will use.
\subsection{Ewald free-space formulas}
\label{sec:ewald-free-space}
In the triply periodic setting, the Ewald summation formula as derived
by Hasimoto was given in \eqref{eq:usum_SR_SF_limit}. As given in
\eqref{eq:invtransf}, for the free-space problem the discrete sum in
Fourier space is replaced by the inverse Fourier transform. With our
generic notation, we can evaluate the discrete-sum potential
\eqref{eq:usum} as
\begin{align}
\v u(\vec{x}) = \sum_{\ni=1}^N G^R(\vec{x} - \vec{x}_{\ni},\xi) \cdot \vec{f}(\vec{x}_{\ni}) +
\frac{1}{(2\pi)^3} \int_{\mathbb R^3} \widehat{
G}^F(\v k,\xi) \cdot \sum_{n=1}^N \vec{f}(\vec{x}_{\ni}) e^{i\v k\cdot (\vec{x}-\v
x_{\ni})} \dif \v k .
\label{eq:usumEwald}
\end{align}
We now apply the screening approach to derive the real space formulas,
\begin{align}
S_{jl}^R(\vec{r}, \xi) & = \left(\delta_{jl}\nabla^2 -
\nabla_j\nabla_l\right) \left[ r-\Phi^H(r,\xi) \right]
\notag \\
T^R_{jlm} (\v r) & = \left[ \left(
\delta_{jl}\nabla_m+\delta_{lm}\nabla_j+\delta_{mj}\nabla_l
\right) \nabla^2 - 2\nabla_j\nabla_l\nabla_m \right] \left[
r-\Phi^H(r,\xi) \right]
\notag \\
\Omega^R_{jl}(\v r) & = - \epsilon_{jlm}\nabla_m \nabla^2 \left[ r-\Phi^E(r,\xi) \right]=
-2 \epsilon_{jlm}\nabla_m
\frac{\operatorname{erfc}(\xi \vec{r})}{r}
\notag
\end{align}
where the splitting functions $\Phi^H$ and $\Phi^E$ are
found in the first and second lines of Table \ref{tab:decompositions},
and we get
\begin{align}
S_{jl}^R(\vec{r}, \xi) & = 2\left( \frac{\xi e^{-\xi^2 r^2} }{\sqrt{\pi}} +
\frac{ \operatorname{erfc}{(\xi r)} }{2 r} \right) (\delta_{jl} + \hat{r}_j \hat{r}_l) -
\frac{4\xi}{\sqrt{\pi}} e^{-\xi^2 r^2} \delta_{jl} , \quad ,
\label{eq:SR_Op} \\
T^R_{jlm} (\v r) & = - \frac{2}{r} \left[ \frac{3 \, \operatorname{erfc}(\xi r)}{r} +
\frac{2 \xi}{\sqrt{\pi}} \left( 3+2\xi^2r^2\right)
e^{-\xi^2 r^2} \right] \hat{r}_j \hat{r}_l \hat{r}_m
\notag \\
& \qquad \qquad \qquad \qquad +
\frac{4 \xi^3}{\sqrt{\pi}} e^{-\xi^2 r^2} (\delta_{jl}\hat{r}_m + \delta_{lm} \hat{r}_j
+\delta_{mj}\hat{r}_l),
\label{eq:TR_Op} \\
\Omega^R_{jl}(\v r) & = 2 \varepsilon_{jlm} \hat{r}_m
\left( \frac{\operatorname{erfc}(\xi r)}{r^2} + \frac{2 \xi}{\sqrt{\pi}}
\frac{1}{r} e^{-\xi^2 r^2} \right),
\label{eq:RR_Op}
\end{align}
where $\hat{\vec{r}}=\vec{r}/|\vec{r}|$.
Only the stokeslet has a non-zero limit as given in
\eqref{eq:Slimit}, that must be included to remove the self-interaction
when evaluating at a source point location.
Let us now turn to the Fourier space terms. We will use the screening
approach, with the Hasimoto screening for the stokeslet and stresslet,
and the Ewald screening for the rotlet. From \eqref{eq:GF_KBGhat},
for each Green's function using the specific form of the differential
operator in \eqref{eq:stokeslet_op}-\eqref{eq:rotlet_op} to define
$\widehat\operatorname{K}(\v k)$, with $\widehat\gamma_E(\v k, \xi)$ and
$\widehat\gamma_H(\v k, \xi)$ as given in the first and second line of
Table \ref{tab:decompositions}, we obtain
\begin{align*}\small
\widehat{S}^F(\v k,\xi) = A^{S}(\v k,\xi)
e^{-k^2/4 \xi^2}, \quad
\widehat{T}^F(\v k,\xi) = A^{T}(\v k,\xi)
e^{-k^2/4 \xi^2}, \quad
\widehat{\Omega}^F(\v k,\xi) = A^{\Omega}(\v k,\xi) e^{-k^2/4 \xi^2},
\end{align*}
where
\begin{align}
A^{S}_{jl}(\v k,\xi) &=
-\left(k^2\delta_{jl} - k_jk_l
\right)
\left(1+k^2/(4\xi^2)\right) \widehat B(|\vec{k}|),
\label{eq:AS}\\
A^{T}_{jlm}(\vec{k},\xi) &=
i \left[(k_m \delta_{jl} +k_j \delta_{lm} +k_l \delta_{mj}) k^2 -2
k_j k_l k_m \right]
\left(1+k^2/(4\xi^2)\right) \widehat B(|\v k|) ,
\label{eq:AT}\\
A^{\Omega}_{jl} (\vec{k},\xi) &=
- i \varepsilon_{jlm} k_m k^2 \widehat B(|\v k|) =
2 i \varepsilon_{jlm} k_m \widehat H(|\v k|).
\label{eq:AR}
\end{align}
Here, $\widehat{H}(k)$ and $\widehat{B}(k)$ are
the Fourier transforms of the harmonic Green's function
$H(r)=1/r$ and the biharmonic Green's function $B(r)=r$,
respectively,
\begin{equation}
\widehat{H}(k)
=\frac{4\pi}{k^2}, \quad \quad
\widehat{B}(k)
=-\frac{8\pi}{k^4}.
\label{eq:def_harm_biharm}
\end{equation}
For a smooth compactly supported function $\widehat{G}^F(\v
k,\xi)$, the Fourier integral in \eqref{eq:usumEwald} can be
approximated to spectral accuracy with a trapezoidal rule in each
coordinate direction, allowing for the use of FFTs for the
evaluation. Inserting the definitions in \eqref{eq:def_harm_biharm}
into (\ref{eq:AS}--\ref{eq:AR}), we can however note that the Fourier
component has a singularity at $k=0$ for all three Green's functions.
We will introduce modified Green's functions for the harmonic
and biharmonic equations, that will still yield the exact same result
as the original ones in the solution domain, and where the Fourier
transforms of these functions have no singularity for $k=0$.
The necessary ideas will be introduced in the next session, following
the recent work by Vico et al. \cite{Vico2016}.
\section{Free-space solution of the harmonic and biharmonic equations}
\label{sec:free-space-solution}
Consider the Poisson equation
\[
-\Delta \varphi(\vec{x})=4 \pi f(\vec{x})
\] with free space boundary conditions. The solution is given by
\begin{equation} \varphi(\vec{x})=\int_{\mathbb R^3} H(|\vec{x}-\vec{y}|)
f(\vec{y}) \, d\vec{y}=\frac{1}{(2\pi)^3} \int_{\mathbb R^3}
\widehat{H}(|\vec{k}|) \hat{f}(\vec{k}) e^{i \vec{k} \cdot \vec{x}},
\label{eq:conv_harmonic}
\end{equation} where $H(r)=1/r$ is the harmonic Green's
function and $\widehat{H}(k)=4\pi/k^2$ its Fourier
transform. Note that they are both radial.
Assume now that $f$ is compactly supported within a domain
$\tilde{\mathcal{D}}$; a box with sides $\v \tilde{L}$,
\begin{align}
\tilde{\mathcal{D}} = \{\vec{x} \mid x_i \in [0, \tilde{L}_i]\,\},
\end{align}
and that we seek the solution $\varphi(\vec{x})$ for $\vec{x} \in
\tilde{\mathcal{D}}$. The largest point to point distance in the domain is
$|\v\tilde{L}|$. Let $\mathcal{R}\ge|\v\tilde{L}|$. Without changing the
solution we can then replace $H$ with a truncated version,
\begin{align}
H^\mathcal{R}(r) = H(r) \operatorname{rect}\left(\frac{r}{2\mathcal{R}}\right),
\end{align}
where
\begin{align}
\operatorname{rect}(x) =
\begin{cases}
1 &\text{for } |x| \le 1/2,\\
0 &\text{for } |x| > 1/2.
\end{cases}
\end{align}
The Fourier transform of this truncated Green's function is
\cite{Vico2016}
\begin{align}
\widehat H^\mathcal{R}(k) = 8 \pi
\left( \frac{\sin(\mathcal{R} k/2)}{k}\right)^2.
\label{eq:HhatR}
\end{align}
This function has a well-defined limit at $k=0$,
\begin{align}
\widehat H^\mathcal{R}(0) = \lim_{k \rightarrow 0} \widehat
H^\mathcal{R}(k) = 2 \pi \mathcal{R}^2.
\end{align}
Similarly, to solve the biharmonic equation on the same size domain,
we can define $B^\mathcal{R}(r) = B(r)
\operatorname{rect}\left(\frac{r}{2 \mathcal{R}}\right)$, which has the Fourier transform
\cite{Vico2016}
\begin{align}
\widehat B^\mathcal{R}(k) = 4 \pi \frac{
(2-\mathcal{R}^2k^2)\cos(\mathcal{R} k) + 2 \mathcal{R} k \sin(\mathcal{R} k) - 2
}{k^4} ,
\label{eq:BhatR}
\end{align} with the limit value
\begin{align} \widehatB^\mathcal{R}(0) = \lim_{k \rightarrow 0}
\widehat B^\mathcal{R}(k) = \pi \mathcal{R}^4.
\end{align}
\subsection{Solving the harmonic and biharmonic equations using FFTs}
\label{sec:FFTsolve} We will now describe how to solve the harmonic
and biharmonic equations using FFTs. In doing so, we will for
simplicity of notation assume that the domain is a cube,
i.e. $\tilde{L}_1=\tilde{L}_2=\tilde{L}_3=\tilde{L}$.
Then do the following
\begin{enumerate}
\item Introduce a grid of size $\tilde{M}^3$ with grid size
$h=\tilde{L}/\tilde{M}$, and evaluate $f(\vec{x})$ on that grid.
\item Define an oversampling factor ${s_f}$, and zero-pad to do a 3D
FFT of size $({s_f} \tilde{M})^3$, defining $\hat{f}(\vec{k})$ for
\[ \vec{k}=\frac{2\pi}{\tilde{L}} \frac{1}{{s_f}} (k_1,k_2,k_3), \quad
\quad k_i \in \left\{ -\frac{{s_f} \tilde{M}}{2},\ldots, \frac{{s_f}
\tilde{M}}{2}-1 \right\}
\]
\item Set $\mathcal{R}=\sqrt{3}\tilde{L}$ and evaluate $\widehat H^\mathcal{R}(k)$
\eqref{eq:HhatR}, with $k=|\vec{k}|$ for the set of $\vec{k}$-vectors defined
above.
\item Multiply $\hat{f}(\vec{k})$ and $\widehat H^\mathcal{R}(k)$ for each $\vec{k}$. Do
a 3D IFFT and truncate the result to keep the $\tilde{M}^3$ values
defining the approximation of the solution $\varphi(\vec{x})$ on the grid.
\end{enumerate} To solve the biharmonic equation instead, replace
$\widehat H^\mathcal{R}(k)$ by $\widehat B^\mathcal{R}(k)$ as given in
\eqref{eq:BhatR}.
With this, we have for all $\vec{x}_j$ in the grid computed the approximation
\[
\varphi(\vec{x}_j)
\approx \frac{(\Delta k)^3}{(2\pi)^3} \sum_{k_1,k_2,k_3=-{s_f} \frac{\tilde{M}}{2}}^{{s_f} \frac{\tilde{M}}{2}-1}
\widehat{H}^\mathcal{R}(|\vec{k}|) \hat{f}(\vec{k}) e^{i \vec{k} \cdot \vec{x}_j}
\]
where $\Delta k=\frac{2\pi}{\tilde{L}} \frac{1}{{s_f}}$.
Note that we at no occasion explicitly multiply with a
prefactor, assuming there is a built-in scaling of $1/({s_f} \tilde{M})^3$ in
the 3D inverse FFT. There should be a multiplication with $h^3$ in step 2,
and with $(\Delta k/2\pi)^3$ above, but that cancels such that only
the built-in scaling remains.
Since the convolution is aperiodic, we need to oversample by at least
a factor of two. In Vico et al. \cite{Vico2016} they advise that we
need an additional factor of two to resolve the oscillatory behavior
of the Fourier transform of the truncated kernel, which would yield
${s_f}=4$. It does however turn out that the need of oversampling is
less than this, as we will discuss in the next section.
If we oversample sufficiently, the error will decay spectrally with
$\tilde{M}$ given that the right hand side $f$ is smooth.
\subsection{Zero-padding/oversampling}
\label{sec:oversampling}
Consider the first integral in \eqref{eq:conv_harmonic}. With $f$
compactly supported on a cube with size $\tilde{L}^3$, $H$ must be
defined on a cube with size $(2\tilde{L})^3$ to be able to compute the
convolution. $H^\mathcal{R}$, with $\mathcal{R}=|\tilde{\v L}|=\sqrt{3}\tilde{L}$
coincides with $H$ inside the sphere of radius $\mathcal{R}$, the smallest
sphere with the cube inscribed. When we use the FFT, we ``periodize''
the computations. We hence need to zero pad the data so that this
periodization interval is large enough to make sure that $H^\mathcal{R}$ is
not polluted within the cube of size $(2\tilde{L})^3$.
Assume that we zero pad the data up to a domain size of
$2\tilde{L}+\delta$. If the sphere of radius $\mathcal{R}$ is to fit within
this domain, we would have $\delta=2(\mathcal{R}-\tilde{L})$. However, as is
illustrated in Figure \ref{fig:oversampling}, since it is enough that
$H^\mathcal{R}$ is not polluted within the cube, it is sufficient with
$\delta=\mathcal{R}-\tilde{L}$. In terms of an oversampling factor ${s_f}$,
this corresponds to
\begin{align}
{s_f} \tilde{L} \ge 2 \tilde{L} + \delta=
\tilde{L}+\mathcal{R},
\end{align}
and the necessary condition becomes
\begin{align}
{s_f} \ge \frac{\tilde{L}+ \mathcal{R}}{\tilde{L}},
\end{align}
such that with $\mathcal{R}=\sqrt{3} \tilde{L}$, we get
${s_f} \ge 1 + \sqrt{3} \approx 2.8$.
Note, that an argument based instead on a large enough sampling ratio in the
Fourier domain to resolve the oscillatory truncated Green's function
would yield the smallest oversampling rate to be that with
$\delta=2(\mathcal{R}-\tilde{L})$, where the Green's function is without
pollution in the full sphere, and hence ${s_f} \ge 2\sqrt{3} \approx 3.5$.
For non-cubic domains, we will have a larger oversampling requirement,
\begin{align}
{s_f} \ge 1+ \frac{\mathcal{R}}{\min_i \tilde{L}_i},
\end{align}
which is ${s_f} \ge 1 + |\widetilde{\v L}|/(\min_i \tilde{L}_i)$ with the smallest possible $\mathcal{R}$.
This additional cost can however be limited to a
precomputation step, through the scheme suggested in \cite{Vico2016}
as discussed in the next section.
\begin{figure}[htbp
\centering
\includegraphics{oversampling}
\caption{Illustration of the minimum zero-padding $\delta$ required
to accurately represent the $\mathcal{R}$-truncated Green's function inside
the domain of dimensions $2\widetilde{\v L}$, when using a
periodic Fourier transform. The condition $\delta \ge \mathcal{R}-\tilde{L}$ must be satisfied to avoid pollution from neighboring Green's
functions inside the domain of interest.}
\label{fig:oversampling}
\end{figure}
\subsection{Precomputation}
\label{sec:precomputation}
We will now further discuss step 4 in the algorithm introduced in
section \ref{sec:FFTsolve}.
For ease of notation we do so in one
dimension. Each dimension will be treated the same way, so the
extension is simple to make.
Let $M_g={s_f} \tilde{M}$ be the number of grid points, such that
$h=\tilde{L}/\tilde{M}=({s_f} \tilde{L})/M_g$, and let
$k=(2\pi/({s_f} \tilde{L})) \bar{k}$, where $\bar{k}$ is an integer.
By the means of an IFFT, we can compute
\[
\varphi_j=\frac{1}{M_g} \sum_{\bar{k}=-M_g/2}^{M_g/2-1}
\widehat{G}(k) \hat{f}_{\bar{k}} e^{i
\frac{2\pi}{M_g} \bar{k} j}, \quad \quad j=0,\ldots,M_g-1,
\]
where $\widehat{G}(k)$ could be either
$\widehat{H}^\mathcal{R}(k)$ or $\widehat{B}^\mathcal{R}(k)$,
and the Fourier coefficients $\hat{f}_{\bar{k}}$ have been computed by
an FFT,
\[
\hat{f}_{\bar{k}}=\sum_{l=0}^{M_g-1}
f(lh) e^{-i \frac{2\pi}{M_g} \bar{k} l} .
\]
Inserting how the $\hat{f}_{\bar{k}}$:s are actually computed, and
rearranging the order of the sums, we get
\[
\varphi_j=\sum_{l=0}^{M_g-1}
\left[ \frac{1}{M_g} \sum_{\bar{k}=-M_g/2}^{M_g/2-1}
\widehat{G}(k) e^{i \frac{2\pi}{M_g} \bar{k} (j-l)} \right]
f(lh)
=\sum_{l=0}^{M_g-1} G_{j-l} f(lh),
\]
where $G_{j-l}$, $j=0,\ldots,M_g-1$ will define the effective Green's function
on the grid, centered at grid point $l$.
Note here, that $f$ has compact support and $f(lh)=0$ for
$l>\tilde{M}-1$, and even though $\varphi_j$ is computed on the large grid, we
will truncate and keep only the $\tilde{M}$ first values.
Hence, for each $l$, only $\tilde{M}$ values of $G_{j-l}$ is actually
needed to produce our result, and since $G_{(j+1)-(l+1)}=G_{j-l}$ a total
of $2\tilde{M}$ grid values of the Green's function are used in the
calculation.
Hence, it is possible to precompute an effective Green's function on
the grid , without knowing $f$, and truncate to save it centered on a
grid with $2\tilde{M}$ values. Let us denote by $\tilde{G}$ the
mollified Green's function that is the result of this procedure,
defined on a grid of size $2\tilde{M}$, or in 3D $(2\tilde{M})^3$. An
example of the mollified harmonic Green's function is shown in Figure
\ref{fig:mollified}.
\begin{figure}[htbp]
\centering
\begin{subfigure}[t]{0.49\textwidth}
\centering
\includegraphics[width=\textwidth,height=0.6\textwidth]{mollified_k}
\caption{Fourier space representations of the original
\eqref{eq:def_harm_biharm}, truncated \eqref{eq:HhatR} and
mollified harmonic Green's functions. The latter is computed
using an FFT.}
\end{subfigure}
\hfill
\begin{subfigure}[t]{0.49\textwidth}
\centering
\includegraphics[width=\textwidth,height=0.6\textwidth]{mollified_r}
\caption{The harmonic Green's function and its mollified counterpart.}
\end{subfigure}
\caption{Example of the mollified harmonic Green's function
generated by inverse transform of $\hat{H}^\mathcal{R}$ using a
finite-size IFFT.}
\label{fig:mollified}
\end{figure}
Once $f$ is given, we can now compute
our $\varphi_j$ values using an aperiodic convolution.
This requires the following steps:
\begin{enumerate}
\item Do a 3D FFT of $\tilde{G}$.
\item Zero-pad and do a 3D FFT of size $(2\tilde{M})^3$ of $f$.
\item Multiply the two, do an IFFT, and save the last $\tilde{M}$ values
in each dimension, i.e. $(\tilde{M})^3$ values as the result.
\end{enumerate}
In practice, the FFT of $\tilde{G}$ is only done once if the equation
is to be solved with several right hand sides $f$. When precomputing
$\tilde{G}$, an oversampling factor ${s_f} \ge 1+\mathcal{R}/ \tilde{L}$ is
needed as discussed in the previous section. What remains is then an
aperiodic convolution with the right hand side as described above, and
an oversampling of $f$ with a factor of $2$ in this step is
sufficient.
\section{Evaluating the Fourier space component}
\label{sect:kspace}
Let us now go back to the Fourier space component in the Ewald
decomposition \eqref{eq:usumEwald}. We will use the notation
\begin{align}
\widehat{G}^{F,\mathcal{R}}(\v k, \xi) = A^{G,\mathcal{R}}(\v k, \xi)
e^{-k^2/4\xi^2},
\label{eq:Gwoexp}
\end{align}
where $G=S$, $T$, and $\Omega$, and were the
superindex $\mathcal{R}$ indicates that $\widehat{H}(k)$ and
$\widehat{B}(k)$ are replaced by $\widehat{H}^\mathcal{R}(k)$
and $\widehat{B}^\mathcal{R}(k)$ in the definitions \eqref{eq:AS},
\eqref{eq:AT} and \eqref{eq:AR}. This means that the modified Green's
functions $\widehat{S}^{F,\mathcal{R}}$, $\widehat{T}^{F,\mathcal{R}}$
and $\widehat{\Omega}^{F,\mathcal{R}}$, will have no singularity at $k=0$.
The task is now to compute
\begin{align}
\v u^F(\vec{x},\xi) = \frac{1}{(2\pi)^3} \int_{\mathbb R^3} e^{i\v
k\cdot\vec{x}} e^{-k^2/4\xi^2}
A^{G,\mathcal{R}}(\v k, \xi) \cdot \sum_{n=1}^N \vec{f}(\vec{x}_{\ni}) e^{-i\v
k \cdot \vec{x}_{\ni}} \dif \v k
\label{eq:uf_sum}
\end{align}
for a given set of target points. The integrand of the inverse
transform is now smooth and can after truncation be easily evaluated
using the trapezoidal rule, but evaluation is still costly ---
$\mathcal{O}(N^2)$ if evaluating at $N$ target points. We will now outline
the spectral Ewald method, which uses the fast Fourier transform to
reduce the cost of this evaluation, yielding a method with a total
cost (including the real space sum) of $\mathcal{O}(N \log N)$. Before we
discuss the actual discretization and implementation details, we start
by describing the mathematical foundation of the method.
\subsection{Foundations}
First we introduce a scalar parameter $\eta>0$ and define
\begin{align}
\widehat{\gb}(\v k, \xi) = \sum_{n=1}^N \vec{f}(\vec{x}_{\ni}) e^{-i\v k\cdot
\vec{x}_{\ni}} e^{-\eta k^2/8\xi^2},
\label{eq:gridfcnhat_def}
\end{align}
which is the Fourier transform of the smooth function
\begin{align}
\gb(\vec{x}, \xi) = \sum_{n=1}^N \vec{f}(\vec{x}_{\ni}) \left(\frac{2\xi^2}{\pi\eta}\right)^{3/2}
e^{-2\xi^2|\vec{x}-\vec{x}_{\ni}|^2/\eta}.
\label{eq:Hx}
\end{align}
Hence, instead of using \eqref{eq:gridfcnhat_def} to directly evaluate
$\widehat{\gb}(\v k, \xi)$, we can evaluate $\gb(\vec{x}, \xi)$
and use an actual computation of the Fourier transform,
\begin{equation}
\widehat{\gb}(\v k, \xi) =\int_{\mathbb R^3} \gb(\vec{x}, \xi)
e^{-i \vec{k} \cdot \vec{x}} \, \dif \vec{x}.
\end{equation}
We furthermore define
\begin{align}
\widehat{\vec{w}}(\v k, \xi) = e^{-(1-\eta)k^2/4\xi^2}
A^{G,\mathcal{R}}(\v k, \xi) \cdot \widehat{\gb}(\v k,
\xi),
\label{eq:gridfcnsc}
\end{align}
such that \eqref{eq:uf_sum} can be written
\begin{align}
\v u^F(\vec{x}, \xi) = \frac{1}{(2\pi)^3} \int_{\mathbb R^3}
\widehat{\vec{w}}(\v k, \xi) e^{-\eta k^2/8\xi^2} e^{ i\v
k\cdot\vec{x} } \dif \v k .
\label{eq:Fx}
\end{align}
Using the convolution theorem we can write this as
\begin{align}
\v u^F(\vec{x}, \xi) = \int_{\mathbb R^3} \vec{w}(\v y,\xi)
\left(\frac{2\xi^2}{\pi\eta}\right)^{3/2} e^{-2\xi^2|\vec{x}-\v
y|^2/\eta} \dif \v y,
\label{eq:u_integr}
\end{align}
where
\begin{equation}
\vec{w}(\vec{x},\xi)
= \frac{1}{(2\pi)^3} \int_{\mathbb R^3} \widehat{\vec{w}}(\v k, \xi)
e^{i \vec{k} \cdot \vec{x}} \, \dif \vec{k}.
\end{equation}
\subsection{Discretization}
\label{sec:discretization}
Assume that we are to evaluate \eqref{eq:uf_sum} for $\vec{x}=\vec{x}_{\texttt{m}}$,
$\texttt{m}=1\ldots N$, and for simplicity of notation that all points are contained in a
cube with equal sides $L$,
\[
\vec{x}_{\ni} \in \mathcal{D} = [0, L]^3, \quad \ni = 1 \dots N.
\]
The choice of $\eta$ will be discussed shortly, in section
\ref{sec:approximation-errors}. At this point, assume that the
Gaussians $e^{-2\xi^2|\cdot|^2/\eta}$ in \eqref{eq:Hx} and
\eqref{eq:u_integr} decay rapidly, and will be truncated outside a
diameter $\supp$. Then $\gb$ becomes compactly supported, such
that we can compute $\widehat{\gb}$ using an FFT, and the
integral in \eqref{eq:u_integr} becomes a local operation around each
target point $\vec{x}_{\texttt{m}}$. To accomodate the support of the truncated
Gaussians, we must extend the domain by some length $\deltaL$. We will
discuss the choice of this length in the discussion on $\eta$. For
now, we consider the extended domain with sides $\tilde{L} = L +
\deltaL$,
\begin{align}
\tilde{\mathcal{D}} = [-\deltaL/2, L + \deltaL/2]^3.
\end{align}
This domain is discretized using a uniform grid with $\tilde{M}^3$
points and grid spacing $h = \tilde{L} / \tilde{M}$.
To initialize our calculations, we precompute
$\widehat{H}^\mathcal{R}(k)$ in case of the rotlet, and
$\widehat{B}^\mathcal{R}(k)$ in case of the stokeslet or
stresslet, as described in section \ref{sec:precomputation}. They need
to be precomputed on a domain of size $2\tilde{L}$, with
$\mathcal{R} = \sqrt{3}\tilde{L}$.
The first step of our computations is to evaluate $\gb$ on the
grid as in \eqref{eq:Hx}. After that we zero-pad the FFT by a a
factor of $2$, to have an oversampled representation of
$\widehat{\gb}$, before we scale it to define
$\widehat{\vec{w}}$ as in \eqref{eq:gridfcnsc}. We will then
multiply by the precomputed fundamental solution
($\widehat{H}^\mathcal{R}(k)$ or $\widehat{B}^\mathcal{R}(k)$)
and the additional scaling factors as given in \eqref{eq:AS},
\eqref{eq:AT} and \eqref{eq:AR}, and apply an inverse FFT to perform a
discrete convolution.
The computation of $\v u^F(\vec{x}_{\texttt{m}},\xi)$, $\texttt{m}=1,\ldots,N$ can hence
be broken down into the following steps:
\begin{enumerate}
\item \emph{Spreading}: Compute $\gb$ on the grid using
\eqref{eq:Hx} and truncated Gaussians.
\item \emph{FFT}: Compute $\widehat{\gb}$ using the
three-dimensional FFT, zero-padded to the double size.
\item \emph{Scaling}: Compute $\widehat{\vec{w}}$ using
\eqref{eq:gridfcnsc} and precomputed $\widehat{H}^\mathcal{R}(k)$
or $\widehat{B}^\mathcal{R}(k)$.
\item \emph{IFFT}: Apply the inverse three-dimensional FFT to
$\widehat{\vec{w}}$. Truncate the result to have $\vec{w}$
defined on the original grid.
\item \emph{Quadrature}: For each $\vec{x}_{\texttt{m}}$, $\texttt{m}=1,\ldots,N$
evaluate $\v u^F$ using \eqref{eq:u_integr} and the trapezoidal rule,
with the Gaussian truncated outside the sphere of diameter $\supp$ centered at
$\vec{x}_{\texttt{m}}$.
\end{enumerate} This is the spectral Ewald method. A major cost of the
method is the large number of exponential function evaluations in
steps 1 and 5. This can be accelerated through the method of fast
Gaussian gridding (FGG) \cite{Greengard2004,Lindbo2011c}. It is then
natural to truncate the Gaussians outside a cube of $P^3$ grid points,
in which case
\begin{align}
\supp = hP .
\end{align}
The computational cost of the FGG in steps 1 and 5 is then $\mathcal
O(NP^3)$, while the cost of the FFTs in steps 2 and 4 is
$\mathcal{O}(\tilde{M}^3 \log \tilde{M} )$. The cost of the scaling in step 3 is
$\mathcal{O}(\tilde{M}^3)$, and negligible in this context.
\subsection{Errors in the spectral Ewald method}
\label{sec:approximation-errors}
The use of the spectral Ewald method for computing the Fourier space
component introduces approximation errors in the solution, which are
separate from the Fourier integral truncation error (further discussed
in section \ref{sec:fourier-space-error}). The approximation errors
stem from the use of a discrete quadrature rule in the quadrature
step, and from the truncation and discretization of the Gaussians
$e^{-2\xi^2|\cdot|^2 / \eta}$ in the spreading and quadrature
steps. The Ewald parameter $\xi$ should be regarded as free, since it
is used for work and error balancing between the real and Fourier
space sums (more on this in section \ref{sec:parameter-choice}). This
leaves two variables for controlling the approximation errors: the
scalar parameter $\eta$ and the Gaussian truncation width
$\supp=hP$. Following \cite{Lindbo2011c}, we write $\eta$ as
\begin{align}
\eta = \left( \frac{\xi d}{m} \right)^2,
\end{align}
where $m$ is a shape parameter controlling how fast the Gaussian
decays within the support $\supp$. It can be shown \cite{Lindbo2011c}
that the approximation errors decay exponentially in $P$ with the
choice $m(P) = C \sqrt{\pi P}$, and that the constant $C$ should be
taken slightly below unity for optimal results (we use the value
$C=0.976$ suggested in \cite{Lindbo2011c}). With these choices, the
approximation errors of the method are controlled through a single
parameter $P$, and they furthermore decay exponentially in that
parameter.
It is evident from the algorithm that $\deltaL$ must be chosen such
that the support of the truncated Gaussians in \eqref{eq:Hx} and
\eqref{eq:u_integr} is included, i.e. $\deltaL \ge \supp$. However, it
turns out that this is not always enough. In the spectral Ewald method
we have taken the Gaussian $e^{-\xi^2 r^2}$ of the screening function
(Table \ref{tab:decompositions}) and separated it into a series of
convolutions of Gaussians, through the factorization
\begin{align}
e^{-k^2/4\xi^2}=e^{-\eta k^2/8\xi^2} \cdot e^{-(1-\eta)k^2/4\xi^2}
\cdot e^{-\eta k^2/8\xi^2} .
\end{align}
The first and last factors correspond to the Gaussian
$e^{-2\xi^2|\cdot|^2 / \eta}$ in the gridding and quadrature steps,
and is already properly resolved and truncated by our choices of
$\eta$ and $d$. For $\eta \ge 1$ the entire ''original'' Gaussian is
contained in these two factors, and the middle factor can be viewed as
a deconvolution of the type used in the nonuniform FFT
\cite{Lee2005}. However, for $\eta < 1$ the middle factor corresponds
to the ''remainder'' Gaussian $e^{-\xi^2|\cdot|^2 / (1 - \eta)}$, and
\eqref{eq:gridfcnsc} corresponds to a convolution with that Gaussian,
carried out in Fourier space. For the convolution to be properly
represented, we must make sure that the domain $\tilde{L}$ includes the
support of $e^{-\xi^2|\cdot|^2 / (1 - \eta)}$ to the desired
truncation level. The original Gaussians are truncated at the level
$e^{-2\xi^2(d/2)^2/\eta} = e^{-m^2/2}$. For the remainder Gaussians to
be truncated at the same level, we need that
\begin{align}
e^{-\xi^2(\deltaL/2)^2/(1-\eta)} \le e^{-m^2/2},
\end{align}
i.e.
\begin{align}
\deltaL \ge \sqrt{2 (1-\eta) m^2/\xi^2} .
\end{align}
To guarantee that both Gaussians have proper support, we thus need
\begin{align}
\deltaL \ge
\begin{cases}
d & \text{ if } \eta \ge 1 ,\\
\max\left(d, \sqrt{2 (1-\eta) m^2/\xi^2} \right) & \text{ if } \eta < 1 .
\end{cases}
\label{eq:deltaL_cond}
\end{align}
With this extra support for $\eta < 1$ the approximation errors are
decoupled from the Fourier space truncation errors, which are further
discussed in section \ref{sec:fourier-space-error}. An example of this
decoupling is shown in Figure \ref{fig:P_errors}, where it can be seen
that the larger choice of $\deltaL$ is actually only needed if the grid size $\tilde{M}$ is picked
larger than necessary for a given error tolerance.
\begin{figure}[htbp]
\centering
\begin{subfigure}[b]{0.49\textwidth}
\centering
\includegraphics[width=\textwidth,height=0.6\textwidth]{stokeslet_P_err_1}
\caption{$\deltaL$ set through \eqref{eq:deltaL_cond}.}
\end{subfigure}
\hfill
\begin{subfigure}[b]{0.49\textwidth}
\centering
\includegraphics[width=\textwidth,height=0.6\textwidth]{stokeslet_P_err_2}
\caption{$\deltaL=d$.}
\end{subfigure}
\caption{Error in the stokeslet Fourier space component for various
values of the discrete Gaussian support $P$. The system is a unit
cube with 1000 sources, $\tilde{M} \in [2,40]$, $\xi=2\pi$ and
$k_{\infty} = \pi \tilde{M} / L=\pi \tilde{M}$. To the left the domain is extended to
include the support of the remainder Gaussians, while to the right
the domain is only extended to cover the support of the gridding
and quadrature Gaussians. Evidently, the extra support is only
needed if $\tilde{M}$ is picked larger than necessary for a given
error tolerance.}
\label{fig:P_errors}
\end{figure}
\section{Evaluating the real space component}
\label{sec:eval-real-space}
The real space part of the free-space Ewald sum \eqref{eq:usumEwald}
has the general form
\begin{align}
\v u^R(\vec{x}) = \sum_{\ni=1}^N G^R(\vec{x} - \vec{x}_\ni) \cdot
\vec{f}(\vec{x}_\ni).
\label{eq:usum_r}
\end{align}
Since $G^R(r)$ decays rapidly (roughly as $e^{-\xi^2 r^2}$), the
sum can be truncated outside some truncation radius $r_c$. Assume that
we wish to evaluate the potential at points $\vec{x}_\texttt{m}$, $\texttt{m}=1\dots
N$. The expression that we need to evaluate is then
\begin{align}
\v u^R(\vec{x}_\texttt{m}) = \sum_{\substack{\ni=1\\|\vec{x}_\texttt{m} - \vec{x}_\ni| \le r_c}}^N
G^R(\vec{x}_\texttt{m} - \vec{x}_\ni) \cdot \vec{f}(\vec{x}_\ni), \quad \texttt{m}=1\dots N.
\end{align}
Naively implemented, this has an $\mathcal{O}(N^2)$ computational cost. It
is however straightforward to find the interaction list of each target
point $\vec{x}_\texttt{m}$ by first creating a cell list \cite{Frenkel2001}. This
reduces the real space cost to $\mathcal{O}(N)$, under the assumption that
the average number of interactions of each target point stays
constant when $N$ changes.
\section{Truncation errors}
\label{sec:truncation-errors}
Truncation errors are introduced when we cut off the real space
interactions outside a radius $r_c$, and when we truncate the Fourier
space integral outside a maximum wave number $k_{\infty}$. The magnitudes
of these errors can be accurately estimated through the analysis
methodology introduced by Kolafa \& Perram \cite{Kolafa1992} for
periodic electrostatic force computations. Denoting by $\Delta \v u(\v
x)$ the pointwise truncation error in the solution, one can then
derive statistical error estimates for the root mean square (RMS)
truncation error, defined as
\begin{align}
\delta \v u = \sqrt{ \frac{1}{N} \sum_{\ni=1}^N \left|\Delta \v u(\vec{x}_{\ni})\right|^2 }.
\end{align}
The analyses for both the real and Fourier space components rely on
the following property:
\begin{lemma}{(Kolafa \& Perram \cite[appx.A]{Kolafa1992})}
Let $(\vec{x}_{\ni}, q_n)$ be a configuration of point sources, and let
\begin{align}
\Delta F(\vec{x}) = \sum_{\ni=1}^N q_n \Delta f(\vec{x}-\vec{x}_{\ni}),
\end{align}
be an error measure due to a set of pointwise errors, $\Delta f =
f_{\text{approx}}-f_{\text{exact}}$. Assuming that the points are
randomly distributed, and that $\Delta F$ has a Gaussian
distribution, then the root mean square (RMS) error
\begin{align}
\delta F = \sqrt{\frac{1}{N} \sum_{\ni=1}^N \left(\Delta F(\vec{x}_{\ni})\right)^2}
\end{align}
can be approximated as
\begin{align}
\delta F^2 \approx \frac{1}{|V|} \sum_i q_i^2 \int_V (\Delta f(\v r))^2 \dif \v r,
\end{align}
where $V$ is the volume enclosing all point-to-point vectors $\v
r_{ij} = \vec{x}_i - \vec{x}_j$.
\label{lemma:rms}
\end{lemma}
\subsection{Fourier space truncation error}
\label{sec:fourier-space-error}
The Fourier space error comes from truncating the integral of the
Fourier transform outside a maximum wave number $k_{\infty}$,
\begin{align}
\Delta\v u^F(\vec{x}) = \frac{1}{(2\pi)^3} \int_{k > k_{\infty}}
\widehat G^F(\v k, \xi) \cdot \sum_{\ni=1}^N \v f(\vec{x}_{\ni})
e^{i\v k\cdot(\vec{x} - \vec{x}_{\ni})} \dif \v k,
\end{align}
where all points $\vec{x}_{\ni}$ are contained in a cube of size $L$.
In our case the integral is approximated using an FFT over an $M^3$
grid covering an $L^3$ domain, such that
\begin{align}
k_{\infty} = \frac{2\pi}{\tilde{L}} \frac{\tilde{M}}{2} .
\end{align}
The RMS of the truncation error is given by
\begin{align}
\delta\v u^F =\sqrt{\frac{1}{N} \sum_{\ni=1}^N \left| \Delta\v u^F(\vec{x}_{\ni}) \right|^2 },
\end{align}
and can be estimated using the method of Kolafa \& Perram. Such
estimates already exist for the periodic stokeslet \cite{Lindbo2010}
and rotlet \cite{AfKlinteberg2016rot} potentials, as well as for the
Beenakker decomposition of the stresslet
\cite{AfKlinteberg2014a}. However, it turns out that periodic
estimates fail for free-space potentials that are based on the
truncated biharmonic potential $B^\mathcal{R}$. This is because
the dominating term of $\widehatB^\mathcal{R}$ for large $k$ is a
factor $\mathcal{R}^2 k^2/2$ larger than $\widehatB$. For
potentials based on the truncated harmonic potential
$H^\mathcal{R}$ one can use the periodic estimates, as the
difference in magnitude between $\widehatH^\mathcal{R}$ and
$\widehatH$ is negligible. We can thus use existing estimates
for the rotlet, while we need to derive new ones for the stokeslet and
stresslet. The final set of estimates is shown in Table \ref{tab:est_fourier}.
\subsubsection{Stokeslet}
Beginning with the stokeslet potential, we consider the truncation
error contribution from a single source located at the origin,
\begin{align}
\Delta u^F_j(\vec{x}) = e_{jl}(\vec{x})f_l,
\end{align}
where
\begin{align}
e_{jl}(\v r) &= \frac{1}{(2\pi)^3} \int_{k > k_{\infty}} \left(1 +
\frac{k^2}{4\xi^2}\right) \widehatB^\mathcal{R}(\v k) e^{-(k/
2\xi)^2} e^{i \v k\cdot \v r} k^2 \left(\delta_{jl} - \hat k_j
\hat k_l \right) \dif \v k,
\end{align}
and
\begin{align}
\widehat B^\mathcal{R}(\v k) = 4 \pi \frac{
(2-\mathcal{R}^2k^2)\cos(\mathcal{R} k) + 2 \mathcal{R} k \sin(R k) - 2
}{k^4} .
\end{align}
We now keep only the highest order term in $k$, which dominates the
error for large $k_{\infty}$,
\begin{align}
e_{jl} &\approx -\frac{4 \pi \mathcal{R}^2}{4\xi^2(2\pi)^3} \int_{k >
k_{\infty}} k^2\cos(\mathcal{R} k) e^{-(k/ 2\xi)^2} e^{i \v k\cdot \v r}
\left(\delta_{jl} - \hat k_j \hat k_l \right) \dif \v k.
\end{align}
The error should be independent of the coordinate system orientation,
and since we are deriving a statistical error measure we approximate
the directional component by its root mean square (computed using
spherical coordinates),
\begin{align}
\left(\delta_{jl} - \hat k_j \hat k_l \right) \approx
\sqrt{\frac{1}{9}\sum_{j,l=1}^3 \left(\delta_{jl} - \hat k_j \hat
k_l \right)^2} = \frac{\sqrt{2}}{3}.
\end{align}
We now integrate in spherical coordinates, choosing a system such that
$\hat k_3 \parallel \hat z$. Integration in $\theta\in[0,\pi]$ then gives us
\begin{align}
e_{jl} &\approx
-\frac{\sqrt{2}}{3} \frac{\mathcal{R}^2}{2\xi^2\pi r}
\int_{k > k_{\infty}} k^3 \cos(\mathcal{R} k) \sin(rk)
e^{-(k/ 2\xi)^2}
\dif k.
\end{align}
To get a good approximation of $e_{jl}$, we need to approximate the
remaining integral.
First, we have from the exponential decay that
the dominating contribution will come from the beginning of the
interval, where $k \approx k_{\infty}$. This allows us to approximate
\begin{align}
I &= \int_{k > k_{\infty}} k^3 \cos(\mathcal{R} k) \sin(rk) e^{-(k/ 2\xi)^2}
\dif k \approx k_{\infty}^3 \int_{k > k_{\infty}} \cos(\mathcal{R} k) \sin(rk) e^{-(k/ 2\xi)^2} \dif k .
\end{align}
Next, we have that $\mathcal{R} \gg r$, so $\mathcal{R}$ is the dominating
frequency in the integrand, such that we can write
$\cos(\mathcal{R} k)\sin(rk) \approx \cos(\mathcal{R} k) \sin(rk_{\infty})$,
and
\begin{align}
I \approx k_{\infty}^3 \sin(k_{\infty} r) \int_{k > k_{\infty}} e^{i\mathcal{R} k-(k/ 2\xi)^2} \dif k,
\end{align}
where we implicitly assume that the real part of the complex
exponential is our quantity of interest. A final approximation (again
assuming $k\approx k_{\infty}$) makes this integrable, and we get
\begin{align}
| I | &\approx \left| \frac{k_{\infty}^3 \sin(k_{\infty} r)}{i\mathcal{R} - k_{\infty}/ 2\xi^2}
\int_{k > k_{\infty}} (i\mathcal{R} - k/ 2\xi^2)
e^{iRk-(k/ 2\xi)^2} \dif k \right| \\
&= \left| \frac{k_{\infty}^3 \sin(k_{\infty} r)}{i\mathcal{R} - K/ 2\xi^2}
e^{i\trunck_{\infty}-(k_{\infty}/ 2\xi)^2} \right|
\le \frac{k_{\infty}^3 |\sin(k_{\infty} r)|}{|i\mathcal{R} - k_{\infty}/ 2\xi^2|}
e^{-(k_{\infty}/ 2\xi)^2}.
\end{align}
We have (for a cube) that $\mathcal{R} \ge \sqrt{3} \tilde{M} h$, while
$k_{\infty} = \pi \tilde{M}/\tilde{L} = \pi / h$. Typical parameter values are
around $k_{\infty}/\xi = \mathcal O(10)$ and $\tilde{M} = \mathcal{O}(50)$, so
$k_{\infty}/2\xi^2 = \mathcal O(50 h/\pi) \ll \mathcal{R}$ and $|i\mathcal{R} -
k_{\infty}/ 2\xi^2| \approx \mathcal{R}$. We can therefore write
\begin{align}
| I | \approx \frac{k_{\infty}^3 |\sin(k_{\infty} r)|}{\mathcal{R}} e^{-(k_{\infty}/ 2\xi)^2},
\end{align}
and
\begin{align}
|e_{jl}| \approx \frac{\sqrt{2}}{6} \frac{\mathcal{R} k_{\infty}^3 |\sin(k_{\infty}
r)|}{\xi^2\pi r} e^{-(k_{\infty}/ 2\xi)^2} .
\end{align}
We can now use Lemma \ref{lemma:rms} to estimate the statistical error
by integrating over a sphere of radius $L/2$ (which then contains all
point sources),
\begin{align}
\left( \delta\v u^F \right)^2
&\approx \sum_{\ni=1}^N \sum_{j=1}^3 f_l^2(\vec{x}_{\ni}) \frac{1}{|V|} \int_V e_{jl}^2(r) \dif \v r \\
&\approx Q\frac{6}{\pi L^3} 3 \left( \frac{\sqrt{2}}{6} \frac{\mathcal{R}
k_{\infty}^3}{\xi^2\pi} e^{-(k_{\infty}/ 2\xi)^2} \right)^2 \int_0^{L/2}
\sin^2(k_{\infty} r) 4\pi \dif r,
\end{align}
where
\begin{align}
Q = \sum_{\ni=1}^N |\v f(\vec{x}_{\ni})|^2 .
\label{eq:Q_stokeslet}
\end{align}
Assuming that $\sin^2(k_{\infty} r)$ has many oscillations in the interval
$[0,L/2]$, we replace it by its average value $1/2$, such that
$\int_0^{L/2} \sin^2(k_{\infty} r) 4\pi \dif r \approx \pi L$.
Finally, we can write the stokeslet truncation error estimate as
\begin{align}
\delta\v u^F \approx \sqrt{Q}\frac{\mathcal{R} k_{\infty}^3}{\xi^2 \pi L} e^{-(k_{\infty}/
2\xi)^2} .
\end{align}
\subsubsection{Stresslet}
For the stresslet, the derivation for the error estimate is completely
analogous to the one for the stokeslet. The difference is that the
leading order term is $k^3$ instead of $k^2$, and that the RMS of the
directional component is
\begin{align}
\sqrt{
\frac{1}{27} \sum_{j,l,m=1}^3 \left(
\left(\delta_{jl}\hat k_m+\delta_{lm}\hat
k_j+\delta_{mj}\hat k_l\right) - 2\hat k_j\hat k_l\hat k_m
\right)^2}
= \frac{7}{27} .
\end{align}
This allows us to directly write the stresslet truncation error
estimate as
\begin{align}
\delta\v u^F \approx \sqrt{\frac{7Q}{6}}\frac{\mathcal{R} k_{\infty}^4}{\xi^2 \pi L} e^{-(k_{\infty}/
2\xi)^2},
\end{align}
where
\begin{align}
Q = \sum_{\ni=1}^N \sum_{l,m=1}^3 q_l^2(\vec{x}_{\ni}) n_m^2(\vec{x}_{\ni}).
\label{eq:Q_stresslet}
\end{align}
\begin{table}[htbp]
\centering
\begin{tabular}{c|c|c|c}
& Stokeslet, $S^F$
& Stresslet, $T^F$
& Rotlet, $\Omega^F$
\\ \hline
&&&\\
$\delta\v u^F$ &
$\displaystyle \sqrt{Q}\frac{\mathcal{R} k_{\infty}^3}{\xi^2 \pi L} e^{-k_{\infty}^2/4\xi^2}$ &
$\displaystyle \sqrt{\frac{7Q}{6}}\frac{\mathcal{R} k_{\infty}^4}{\xi^2 \pi L} e^{-k_{\infty}^2/4\xi^2}$ &
$\displaystyle \sqrt{\frac{8\xi^2Q}{3\pi L^3 k_{\infty}}} e^{-k_{\infty}^2/4\xi^2}$
\\
&&&\\
\end{tabular}
\caption{Fourier space truncation errors for the stokeslet, stresslet, and rotlet \cite{AfKlinteberg2016rot}. The quantity $Q$ is defined as in \eqref{eq:Q_stokeslet} for the stokeslet and rotlet, and as in \eqref{eq:Q_stresslet} for the stresslet.}
\label{tab:est_fourier}
\end{table}
\begin{figure}[htbp]
\centering
\begin{subfigure}[b]{0.32\textwidth}
\centering
\includegraphics[width=\textwidth]{fourier_err_stokeslet}
\caption{Stokeslet}
\label{fig:fourier_err:stokeslet}
\end{subfigure}
\begin{subfigure}[b]{0.32\textwidth}
\centering
\includegraphics[width=\textwidth]{fourier_err_stresslet}
\caption{Stresslet}
\label{fig:fourier_err:stresslet}
\end{subfigure}
\begin{subfigure}[b]{0.32\textwidth}
\centering
\includegraphics[width=\textwidth]{fourier_err_rotlet}
\caption{Rotlet}
\label{fig:fourier_err:rotlet}
\end{subfigure}
\caption{RMS of relative Fourier space truncation errors for the
stokeslet, stresslet and rotlet. Dots are measured value, solid
lines are computed using the estimates of Table
\ref{tab:est_fourier}. The system is $N=10^4$ randomly distributed
point sources in a cube with sides $L=3$, with $k_{\infty}=\pi \tilde{M}
/\tilde{L}$, $\xi=3.49$, $M=1\dots50$, and $P=32$.}
\label{fig:fourier_err}
\end{figure}
\subsection{Real space truncation error}
The real space truncation error is due to neglecting interactions in
the real space sum for which $r > r_c$,
\begin{align}
\Delta\v u^R(\vec{x}) = \sum_{|\vec{x} - \vec{x}_{\ni}| > r_c} %
G^R(\vec{x} - \vec{x}_{\ni}) \cdot \v f(\vec{x}_{\ni}) .
\end{align}
The RMS of this error is given by
\begin{align}
\delta\v u^R =\sqrt{\frac{1}{N} \sum_{\ni=1}^N \left| \Delta\v u^R(\vec{x}_{\ni}) \right|^2 }
\end{align}
Following the analysis by Kolafa \& Perram, we can use Lemma
\ref{lemma:rms} to estimate $\delta\v u^R$ as
\begin{align}
\left( \delta\v u^R \right)^2 \approx \frac{1}{L^3}\sum_{\ni=1}^N
\left( \v f(\vec{x}_{\ni}) \right)^2
\cdot \int_{r>r_c} \left(G^R(\v r)\right)^2 \dif \v r .
\end{align}
Estimates based on this approximation are already available in the
literature for the stokeslet \cite{Lindbo2011e} and rotlet
\cite{AfKlinteberg2016rot} decompositions used in this paper, and are
shown in the summary in Table \ref{tab:est_real}. We will here derive
a similar estimate for the Hasimoto decomposition of the stresslet,
essentially by repeating the derivation of \cite{AfKlinteberg2014a}
for the Beenakker decomposition.
The real space component of the stresslet has the form
\begin{align}
T^R_{jlm}(\xi, \v r) = A_1(\xi, r) \hat r_j \hat r_l \hat r_m +
A_1(\xi, r) (\delta_{jl}\hat r_m + \delta_{lm}\hat r_j + \delta_{mj}\hat r_l ),
\end{align}
and the RMS error is approximated as
\begin{align}
\left( \delta\v u^R \right)^2 \approx \frac{1}{L^3}\sum_{\ni=1}^N
\sum_{j=1}^3
q_l^2(\vec{x}_{\ni}) q_m^2(\vec{x}_{\ni}) \int_{r>r_c} \left(T^R_{jlm}(\v
r)\right)^2 \dif \v r .
\label{eq:TRrms}
\end{align}
Arguing that the error should be independent of the coordinate system
orientation, we replace $(T^R_{jlm})^2$ by its average value
over the tensor components, computed using spherical coordinates
\begin{align}
\sum_{j=1}^3 \left(T^R_{jlm}(\v r)\right)^2 \approx%
3 \overline{\left(T^R\right)^2} = \frac{3}{27}
\sum_{j,l,m=1}^3 \left( T_{jlm}^R \right)^2 = \frac{1}{9}
\left(A_1^2+6 A_1 A_2+15 A_2^2\right) .
\label{eq:TR2avg}
\end{align}
This quantity has only radial dependence and is integrable,
\begin{align}
\begin{split}
\int_{r > r_c} 3 \overline{\left(T^R\right)^2} 4\pi r^2 \dif r =&
-\frac{32}{3} \sqrt{\pi } \xi e^{-\xi ^2 r_c^2} \text{erfc}\left(\xi
r_c\right)+21 \sqrt{2 \pi } \xi \text{erfc}\left(\sqrt{2} \xi
r_c\right)\\&+\frac{16 \pi \text{erfc}\left(\xi
r_c\right){}^2}{r_c}+\frac{4}{9} \xi ^2 r_c e^{-2 \xi ^2 r_c^2}
\left(28 \xi ^2 r_c^2-3\right)
\\ \approx &
\frac{112}{9} \xi ^4 r_c^3 e^{-2 \xi ^2 r_c^2},
\end{split}
\label{eq:TR2int}
\end{align}
where we have kept only the dominating term (for large $\xi r_c$) in
the last step. Combining \eqref{eq:TRrms}, \eqref{eq:TR2avg} and
\eqref{eq:TR2int} gives us the estimate for the stresslet shown in
Table \ref{tab:est_real}.
\begin{table}[htbp]
\centering
\begin{tabular}{c|c|c|c}
& Stokeslet, $S^R$
& Stresslet, $T^R$
& Rotlet, $\Omega^R$
\\
\hline
&&&\\
$\delta\v u^R$
& $\displaystyle \sqrt{\frac{4 Q r_c}{L^3}} e^{-\xi^2 r_c^2}$
& $\displaystyle \sqrt{\frac{112 Q \xi^4 r_c^3}{ 9 L^3 }} e^{-\xi^2 r_c^2}$
& $\displaystyle \sqrt{\frac{8 Q}{3 L^3 r_c}} e^{-\xi^2 r_c^2}$
\\
&&&\\
\end{tabular}
\caption{Real space truncation errors for the stokeslet \cite{Lindbo2011e}, stresslet, and rotlet \cite{AfKlinteberg2016rot}. The quantity $Q$ is defined in the same way as for the Fourier component.}
\label{tab:est_real}
\end{table}
\begin{figure}[htbp]
\centering
\begin{subfigure}[b]{0.32\textwidth}
\centering
\includegraphics[width=\textwidth]{real_err_stokeslet}
\caption{Stokeslet}
\label{fig:real_err:stokeslet}
\end{subfigure}
\begin{subfigure}[b]{0.32\textwidth}
\centering
\includegraphics[width=\textwidth]{real_err_stresslet}
\caption{Stresslet}
\label{fig:real_err:stresslet}
\end{subfigure}
\begin{subfigure}[b]{0.32\textwidth}
\centering
\includegraphics[width=\textwidth]{real_err_rotlet}
\caption{Rotlet}
\label{fig:real_err:rotlet}
\end{subfigure}
\caption{RMS of relative real space truncation errors for the
stokeslet, stresslet and rotlet. Dots are measured value, solid
lines are computed using the estimates of Table
\ref{tab:est_real}. The system is $N=2000$ randomly distributed
point sources in a cube with sides $L=3$, with $\xi=4.67$, and
$r_c \in [0, L/2]$.}
\label{fig:real_err}
\end{figure}
\section{Summary of method}
\label{sect:summary}
We now summarize the free-space fast Ewald method for Stokes
potentials. Our goal is to compute a discrete-sum potential of the
type \eqref{eq:usum},
\begin{align}
\v u(\vec{x}) = \sum_{\ni=1}^N G(\vec{x} - \vec{x}_{\ni}) \cdot
\vec{f}(\vec{x}_{\ni}),
\end{align}
for a set of $N$ target point $\vec{x}$, with $G$ being the stokeslet
\eqref{eq:def_stokeslet}, stresslet \eqref{eq:def_stresslet} or rotlet
\eqref{eq:def_rotlet}. We assume that all target and source points are
contained in the cubic domain $\mathcal{D} = [0,L]^3$.
Using an Ewald decomposition (sec. \ref{sec:ewald-free-space}) and
an Ewald parameter $\xi > 0$, we split the potential into a
short-range part $\v u^R$ acting locally, and a long-range part $\v
u^F$ computed in Fourier space,
\begin{align}
\v u(\vec{x}) = \v u^R(\vec{x}, \xi)+ \v u^F(\vec{x}, \xi) + \v u^{\text{self}}(\vec{x}, \xi),
\end{align}
where $\v u^{\text{self}}$ refers to the self-interaction term
\eqref{eq:Slimit}, which has to be taken into account only for the
stokeslet potential.
The real space component is truncated outside an interaction radius
$r_c$,
\begin{align}
\v u^R(\vec{x}) \approx \sum_{\substack{\ni=1\\|\vec{x} - \vec{x}_\ni| \le r_c}}^N
G^R(\vec{x} - \vec{x}_\ni) \cdot \vec{f}(\vec{x}_\ni).
\end{align}
This is a local operation in the neighborhood of each target point
$\vec{x}$, and can be efficiently evaluated using e.g. a cell list
(sec. \ref{sec:eval-real-space}).
The Fourier space component is evaluated through a Fourier integral,
truncated at a maximum wave number $k_{\infty}$,
\begin{align}
\v u^F(\vec{x},\xi) \approx \frac{1}{(2\pi)^3} \int_{|\vec{k}| \le k_{\infty}} e^{i\v
k\cdot\vec{x}}
\widehat{G}^{F,\mathcal{R}}(\v k, \xi)
\cdot \sum_{\ni=1}^N \vec{f}(\vec{x}_{\ni}) e^{-i\v
k \cdot \vec{x}_{\ni}} \dif \v k.
\end{align}
The superscript $\mathcal{R}$ denotes that we have removed the singularity
in the integrand (at $k=0$) by truncating the original Green's
function outside a maximum interaction radius $\mathcal{R}$
(eq. (\ref{eq:AS}--\ref{eq:AR}), eq. \eqref{eq:Gwoexp}, sec.
\ref{sec:free-space-solution}). The integral is evaluated using the
spectral Ewald method (sec. \ref{sec:discretization}), which uses FFTs
on an $\tilde{M}^3$ grid to efficiently compute the long-range
interactions. The method requires the domain $\mathcal{D}$ to be extended
by a length $\deltaL$ \eqref{eq:deltaL_cond} for accurate function
support, and then zero-padded by a factor 2 for the convolution to be
aperiodic when using FFTs. In fact, an oversampling factor ${s_f} \ge
1+\sqrt{3} \approx 2.8$ is required to accurately resolve the
truncated Green's function (sec. \ref{sec:oversampling}), but the cost
of that can be reduced to a precomputation step involving the
fundamental solution to the harmonic or biharmonic equation
(sec. \ref{sec:precomputation}).
\subsection{Computational complexity}
\label{sec:comp-complexity}
We wish to express how the computational cost of the method scales
with an increased number of sources and targets $N$, which we assume
to be evenly distributed in the domain $\mathcal{D}$. The system can be
scaled up in two different ways: by increasing the point density in a
fixed domain, or by increasing the domain size $L$ with a fixed point
density. Either way, the scaling arguments have as their starting
point that the real space sum be $\mathcal{O}(N)$. This is achieved by
keeping a constant number of near neighbors (within $r_c$) for each
target under scaling. Additionally, we want the level of the
truncation errors to be constant, which is achieved by keeping $\xi
r_c$ and $\tilde{M} \xi^{-1} L^{-1}$ constant.
If $N$ increases with $\mathcal{D}$ fixed, then $r_c \propto N^{-1/3}$ is
required for an $\mathcal{O}(N)$ real space sum. If the accuracy is to
remain constant, then $\xi \propto r_c^{-1} \propto N^{1/3}$ and the
grid size is scaled as $\tilde{M} \propto \xi \propto N^{1/3}$. This puts
the Fourier space cost at $\mathcal{O}(\tilde{M}^3 \log \tilde{M}) \propto
\mathcal{O}(N \log N)$.
If the domain size $L$ increases with a fixed point density, then $N
\propto L^{3}$ and the real space sum is $\mathcal{O}(N)$ if we keep $r_c$
and $\xi$ constant. Then $\tilde{M} \propto L \propto N^{1/3}$, such
that the Fourier space cost is $\mathcal{O}(\tilde{M}^3 \log \tilde{M}) \propto
\mathcal{O}(N \log N)$.
\subsection{Parameter selection}
\label{sec:parameter-choice}
For a given system ($N$ charges in a domain of size $L$), the required
parameters for our free-space Ewald method are the Ewald parameter
$\xi$, the real space truncation radius $r_c$, the number of grid
points $M$ covering the original domain, and the Gaussian support
width $P$. Based on these parameters one can then set $\deltaL$ using
\eqref{eq:deltaL_cond}, which then gives $\tilde{L} = L + \deltaL$. This
in turn gives $\tilde{M}$, by satisfying $h=L/M=\tilde{L}/\tilde{M}$. We
will here draft a strategy for optimizing $\xi$, $r_c$, $M$ and $P$ in
a large-scale numerical computation.
For a given value of $\xi$ and absolute error tolerance $\epsilon$,
close-to optimal values for $M$ and $r_c$ can be computed using the
estimates in Tables \ref{tab:est_fourier} and \ref{tab:est_real}. The
support width $P$ affects the error in the Fourier space component,
and we have in practice observed that for the relative error, $P=16$
gives 8 digits of accuracy, $P=24$ gives 12 digits, and $P=32$ is
enough to guarantee that the approximation errors are at roundoff.
Which value of $\xi$ to choose is highly implementation dependent, as
the variable is used to shift the workload between the real and
Fourier space components. A straightforward strategy for finding an
optimal value is to start with a small but representative subset of
the original system, and compute a reference solution for that
subset. Picking a starting value for $\xi$, one then sets $P=32$, and
adjusts $M$ and $r_c$ until the error tolerance is strictly met. Then
$P$ can be decreased in steps of two\footnote{$P$ a multiple of two is
favorable for code optimization using vector instructions.} until
the tolerance is reached again. Using this starting point for $(\xi,
r_c, M, P)$, one then does a parameter sweep in $\xi$ for finding the
configuration with the smallest runtime, while keeping $\xi r_c$ and
$M/\xi$ constant during the sweep. Once an optimal setup is found, the
original (large) system can be computed using the same set of
parameters, except $M$ which is scaled such that $L/M$ remains
constant for both systems.
\section{Results}
\label{sect:numres}
We consider systems of $N$ random point sources drawn from a
uniform distribution in a box of size $L^3$.
We evaluate the sum \eqref{eq:usum} with \textit{stokeslets}
\eqref{eq:def_stokeslet}, \textit{stresslets} \eqref{eq:def_stresslet}
and \textit{rotlets} \eqref{eq:def_rotlet} using our free space
Spectral Ewald (FSE) method, at the same $N$ target locations.
All components of the force/source strengths are random numbers from a
uniform distribution on $[-1,1]$.
All computationally intensive routines are written in C and are called
from Matlab using MEX interfaces. The results are obtained on a
desktop workstation with an Intel CoreTM i7-3770 Processor (3.40 GHz)
with four cores and 8 GB of main memory.
To measure the actual errors, we compare to the result from evaluating
the sum by direct summation.
In the left plot of figure \ref{fig:runtime_all_kernels}, the computing time
for evaluation of the sums is plotted versus $N$, for all three
kernels and for both the Spectral Ewald (FSE) method and direct
summation. The parameters in the Spectral Ewald method have been set
to keep the relative rms error below $0.5 \cdot 10^{-8}$.
The optimal value of $\xi$ cannot be determined theoretically, since
it is implementation and hardware dependent. When we vary $N$ in
figure \ref{fig:runtime_all_kernels}, we change the size of the box,
to keep a constant number density $N/L^3=2500$. If an optimal value of
$\xi$ is determined for one system (see discussion in section
\ref{sec:parameter-choice}), the same value can be kept as the system
is scaled up or down in this manner.
The parameters $r_c$, $P$ and the grid resolution $L/M$ are kept
constant as $N$ and hence $L$ is increased, yielding an increase in
the grid size.
We have used $\xi=7$ for all three kernels, $r_c=0.63$, $0.63$ and
$0.58$ for the stokeslet, stresslet and rotlet, respectively, and
$P=16$ for all kernels. For $L=2$, $M$ is set to $48$, $50$ and $38$
for the three kernels, and is then scaled with $L$.
The precomputation step does not depend on the location
of the sources, and can be performed once the size of the domain is
set.
The precomputation cost can therefore
usually be amortized over many calls to the method, as a simulation
code is run for many time steps and possibly iterations within time steps.
Despite this, we have chosen to plot the runtimes including the
precomputation cost, and later discuss it in more detail.
\begin{figure}[htbp
\centering
\includegraphics[width=0.49\textwidth,height=0.5\textwidth]{time_N_updated}
\includegraphics[width=0.49\textwidth,height=0.5\textwidth]{runtime_err_updated}
\caption{(left) Comparison of direct and fast evaluation of the sum in
\eqref{eq:usum} for the Stokeslet, Stresslet, and Rotlet
including the precomputation step.
The system grows at constant density ($N/L^3$ is constant), with
$\xi=7$ and $r_c$ constant for all values of $N$.
The relative rms error is less than $0.5\times10^{-8}$.
(right) Runtime of computing \eqref{eq:usum} using FSE for all kernels as a function of the relative rms error. $N=20 \, 000$, $L=2$ and $\xi=7$.
}
\label{fig:runtime_all_kernels}
\end{figure}
From these data (figure \ref{fig:runtime_all_kernels} left), we can
find the approximate breakeven points, i.e. the values of $N$ for
which any larger system will benefit from using the fast method. We
find it to be approximately $N=20 \, 000$ for the stokeslet,
$30 \, 000$ for the stresslet and $19 \, 000$ for the rotlet with
precomputation, which is reduced to $18 \, 000$, $25 \, 000$ and
$15 \, 000$ without the precomputation step. If the precomputation
step is to be done only once, the decomposition parameter $\xi$ should
however be chosen differently for optimal performance, which would
bring down the break even point further. Note that this is a strict
error tolerance. For lower accuracy requirements, the cross over
occurs at lower values of $N$. These are higher values than have
previously been reported in the literature, e.g. in
\cite{Tornberg2008}, where $N=5000$ was reported as the breakeven for
the Stokelet. There are two factors affecting these numbers, one is
that these results are run on multiple cores for which the direct sum
parallelizes better than the FFTs involved in the fast method. The
other factor is that the direct sums relatively speaking have become
faster to evaluate also on a single core, where compilers can speed up
the code significantly using vector instructions, while the more
complicated algorithms cannot benefit from this as extensively.
To make sure that our method is competitive, we have compared to a
fast multipole implementation available as free software
\cite{fmm_webpage}, running both codes on a single core and comparing
timings for the Stokeslet. We set the accuracy level to six digits
in the FMM. For $N=20 \, 000$, this yields a relative rms error of
about $5.6 \cdot 10^{-9}$, and we set the parameters for the FSE
method to obtain a similar error level (for this case we get
$4.3 \cdot 10^{-9}$). For $N=20 \, 000$, the FSE code (including
precomputation) and the FFM code both use about 3 seconds. The direct
evaluation takes 3.6s with our code and 6.7s with the code provided
with the FMM package. It should however be noted, that the FMM as
well as the direct code from that package, returns not only the three
vector components produced by the Stokeslet, but also the associated
(scalar) pressure, which increases the cost somewhat. The breakeven
point for both FSE and FMM is about $N=17 \, 000$ when comparing to
our direct code. If we instead compare to the direct code in the FMM
package, the break even point for the FMM decreases to $N=10 \, 000$.
Most fair would be to compare the FMM to a direct sum written as the
faster one, but including also the pressure component, which should
place the break even point between the two numbers above. For the FSE
code, assuming that the precomputation will be done only once, and
choosing $\xi$ instead to optimize the runtime without precomputation,
the break even point drops from $N=17 \, 000$ to $11 \, 000$.
Let us consider also a larger system with $N=400 \, 000$, with
$L=5.4288...$ such that $N/L^3=2500$. For the stokeslet summation by
FSE, we pick the parameters $\xi=8$, $r_c=0.5651$, $M=144$ and $P=16$
to obtain a relative rms error of $5 \cdot 10^{-8}$. This means that
the FFTs are computed for grids with $\tilde{M}=2(M+P)$. The time for
evaluation is about 64 seconds (including the precomputation), and the
speed-up compared to our direct evaluation of the sum is a factor of
about $23$. Excluding the precomputation cost, the computing time is
reduced by 15 s, and this factor increases to $29$. For the FMM, the
evaluation time is about 180 seconds, yielding a speed-up of a factor
of about $8$ compared to our direct sum or a factor of $15$ as
compared to the one provided with the FMM code \cite{fmm_webpage}.
Checking the relative rms error from both the FSE and FMM
computations, they are similar, around $0.5 \cdot 10^{-8}$ for FSE and
$10^{-8}$ for the FMM. Hence, for this example on a single core, the
FSE method including the precomputation is almost three times as fast
as the FMM method, but the difference would be reduced somewhat if the
time for computing the extra pressure component was excluded.
In the adaptive FMM code, a box is split into 8 children boxes if the
number of sources is larger than a set value. If any of the children
boxes still have too many source points, it is split again. With a
uniform distribution of points, most leaf boxes are on the same level
of refinement, which in this case will be four divisions. The curve
for computational cost versus N will not be smooth, since this is a
discrete process (either you keep the box as one, or you split into
eight), which changes the cost balance between different parts of the
algorithm. This is why the larger computational cost for the FMM
method in this case could not be predicted considering the timing for
$N=20 \, 000$, where the timing of the FMM and FSE methods were
similar.
We did not set out to make a thorough comparison of the two
methods. All results are for uniform distributions of source points.
Typically the FSE performs better compared to the FMM for higher
accuracies. Moving towards an increasingly non-uniform point
distribution, the adaptivity of the FMM will at some point pay off.
With this, we have however showed that the FSE method is competitive with
the FMM.
To show how the computational cost depends on the accuracy
requirements, we now consider a fixed system with $N=20 \, 000$ sources in
a box with $L=2$ and vary the error tolerance.
In the right plot of figure \ref{fig:runtime_all_kernels}, we plot the runtime for
summing the stokeslet, stresslet and rotlet kernels as a function of
the relative rms error in the result. Computing the $k$-space
contribution for the stokeslet and rotlet involves gridding of three
vector components, three FFTs, a scaling in Fourier-space, three
inverse FFTs and the quadrature step for the three components of the
solution, see the algorithm in section \ref{sec:discretization}. The
stresslet instead requires the gridding of 9 components and hence 9
FFTs. After the scaling step, there are three resulting vector
components, as for the other kernels. All three kernels require the
same amount of precomputing. Hence, it is not surprising to see that
the stresslet is the most expensive kernel to compute.
We expect a higher cost of the stokeslet as compared to the rotlet due
to the slower decay of the Fourier space part, as given
in table \ref{tab:est_fourier}. This means that larger FFT grids are
needed to obtain the same accuracy. See e.g. the discussion in
connection to the left plot in figure \ref{fig:runtime_all_kernels}
where the choice of $M$ for the box $L=2$ is $48$ for the stokeslet
and $38$ for the rotlet.
For the same system as in figure \ref{fig:runtime_all_kernels} (left), we now study the
computational cost for the different part of the calculations for the
stokeslet. In the left plot of figure \eqref{fig:stokeslet_detail} we
show the total evaluation runtime for the stokeslet sum together with
the three parts that makes up this total cost: the real space and
Fourier space evaluations plus the precomputation in Fourier space.
We use the choice of $\deltaL=d$ in \eqref{eq:deltaL_cond}, such that
$\tilde{L}=L+d$. With this, $\tilde{M}=M+P$,
and the FFTs in the Fourier space evaluations will be of size
$(2\tilde{M})^3$. For the precomputation, the size of the FFT grids in
each dimension will be taken as the smallest even number that is
greater than $(1+\sqrt{3}) \tilde{M}$. The plot shows that the
computational cost is very similar for each part.
As discussed above, the precomputation does not depend on the sources
and can be done only once as long as the domain size does not change.
Excluding the precomputation cost from the timing of the
stokeslet, the runtime is reduced somewhere between a quarter and one
third. Readjustment of $\xi$ to instead balance computational costs
excluding the precomputation, would yield a further reduced run time.
In the right plot of figure \ref{fig:stokeslet_detail}, we further
break down the cost of evaluating the Fourier space sum into three
parts: {\em Grid} (the to and from grid operations with Gaussians),
{\em FFT} (the
total of $6$ FFTs) and {\em Scale}, the multiplication in step 3 of the
algorithm in section \ref{sec:discretization}.
Note here that the oscillations in the FFT curve
are due to the fact that the FFT is more efficient for some grid sizes.
The scaling step is clearly the cheapest of the three parts.
The cost of the gridding step is $O(P^3N)$, where $P^3$ are the number
of grid points in the support of a Gaussian, and the cost of each
FFT of size $(2\tilde{M})^3$ is $O(\tilde{M}^3 \log \tilde{M})$.
Due to the connection to the real space sum, the choice of $\tilde{M}$
will be such that this cost scales as $O(N \log N)$, as discussed in
section \ref{sec:comp-complexity}.
\begin{figure}[htbp
\centering
\includegraphics[width=0.49\textwidth,height=0.5\textwidth]{stokeslet_fse_updated}
\includegraphics[width=0.49\textwidth,height=0.49\textwidth]{stokeslet_fourier_updated}
\caption{Breakdown of runtimes (left) and Fourier space runtime
(right) for evaluating the Stokeslet as a function of number of
particles. The full run-time was also shown in the left plot of
figure \ref{fig:runtime_all_kernels}. }
\label{fig:stokeslet_detail}
\end{figure}
\section{Conclusions}
We have presented a new fast summation method for free space Green's
functions of Stokes flow. The method is based on an Ewald
decomposition to split the sum in two parts, one in real space and
one in Fourier space. The real space sum can simply be truncated
outside of some radius of interaction that depends on the choice of
decomposition parameter and the required accuracy. The focus of this
paper is on the Fourier space sum, the treatment of which is set in
the framework of the Spectral Ewald method, previously developed for
periodic problems \cite{Lindbo2010,AfKlinteberg2014a}. The adaptation
to the free space problem involves a very recent approach to solving
the free-space harmonic and biharmonic equations using FFTs on a
uniform grid \cite{Vico2016}. The Ewald Fourier space kernels for the
stokeslet, stresslet and rotlet are defined from the precomputed
Fourier representation of mollified harmonic (rotlet) and biharmonic
(stokeslet and stresslet) kernels, and the method can easily be
extended to any kernel that can be expressed as a differentiation of
the harmonic and/or biharmonic kernel.
New truncation error estimates have been derived for the free space
kernels.
The extension of the FFT based Spectral Ewald method to the free space
problem incurs an additional computational cost compared to the
periodic problem.
This is essentially due to the computation of larger FFTs, as
computational grids are zero-padded to the double size before the FFTs
are computed. There is also an additional cost of two oversampled
FFTs for precomputing the Fourier representation of the mollified
harmonic or biharmonic kernel. This precomputation do not depend on
the sources, and the cost can often be amortized over many sum
evaluations.
Truncation error estimates have been derived for the kernels for
which they did not already exist, such that precise estimates of the
errors introduced by truncating the real and Fourier space sums are
available for all three kernels, the stokeslet, stresslet and rotlet.
Errors decay exponentially in the physical distance and wave mode
number used for cut-off.
Approximation errors in the evaluation of the Fourier sum decays
exponentially with the support of the Gaussians.
An intricate detail needed to preserve the decoupling between
truncation and approximation errors that is not relevant for the periodic
Spectral Ewald method was discussed in section
\ref{sec:approximation-errors}.
Numerical results are presented for the evaluation of the stokeslet,
stresslet and rotlet sums. They show the expected $O(N \log N)$
computational cost of the method.
We have compared to an open source implementation of the FMM method
\cite{fmm_webpage}, and have shown that our method is competitive, as
it performs better for the uniform source distributions and high
accuracies considered here.
With this, we have developed a new FFT based method for the fast
evaluation of free space Green’s functions for Stokes flow
(stokeslets, stresslets and rotlets) in a free space setting. This
free space Spectral Ewald method
allows the use of the same framework as the periodic one, which makes
it easy to swap methods depending on the problem under
consideration. The source code for the triply periodic SE method is
available online \cite{se_github}, and we plan to shortly release also the code for
this free space implementation.
\section*{Acknowledgements}
This work has been supported by the G\"{o}ran Gustafsson Foundation
for Research in Natural Sciences and Medicine and by the Swedish
Research Council under grant no. 2011-3178. The authors gratefully
acknowledge this support.
| 83f47d85985b8a6a79fbb413e67ed51a6f42f291 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
The Aubry-Andr\' e-Harper (AAH) model has been extensively used as a quasiperiodic model to theoretically study the phase transition between extended, critical and localized phases \cite{Harper, Aubry, Hiramoto1, Ostlund, Kohmoto1, Kohmoto2, Thouless, Hiramoto2, Geisel, Han, Chang, Takada, Liu, Wang}. With the realization of the AAH model in photonic crystals \cite{Negro, Lahini, Kraus1} and ultracold atoms \cite{Roati, Modugno}, this model has gained attention in recent years. The abundant phenomena revealed by the AAH model make it an ideal test field for topological phases and the transitions between them both in the incommensurate and commensurate cases \cite{Kraus2, Lang1, Lang2, Cai, DeGottardi, Ganeshan, Satija, Barnett, Deng, Grusdt, Zhu}. When the lattice is incommensurate, the system will go through a transition from an extended phase to a localized phase due to the disordered on-site potential \cite{Cai, DeGottardi}. If the lattice is commensurate, the AAH model can be used to explore emerging topological states of matter. In addition, quantum many-body localization phenomena are also widely explored in this model \cite{Iyer, Schreiber}. Recently, the connection between the AAH model, the one-dimensional Kitaev chain \cite{Kitaev}, and topological superconductivity are investigated. In Ref. \cite{Ganeshan}, a generalized AAH model which has modulations both in the on-site potentials and the hopping amplitudes between the neighboring lattice sites is discussed, and the topological states are attributed to the topological properties of the one-dimensional Majorana chain. This model is also connected to the Su-Schrieffer-Heeger (SSH) \cite{Su} model when the lattice is commensurate. In Ref. \cite{Wang}, the phase diagram of a non-Abelian AAH model with p-wave superfluidity is discussed and the phase transition from a metallic phase to an insulator phase is checked. However, most AAH models explored before have only modulations in the on-site potential or do not include the superconducting pairing, thus limiting the usage of these models. If we introduce superconducting pairing into the generalized AAH models discussed in Ref. \cite{Ganeshan}, more interesting phenomena can be expected. With p-wave superconducting pairing and modulations in the model, there would be both SSH-like and Kitaev-like topologically nontrivial phases, and thus so-called electron fractionalization to the Majorana fermions can be realized, as discussed in Ref. \cite{Wakatsuki}.
In this paper, we consider a generalized AAH model with a superconducting pairing potential. Both the hopping amplitudes between the nearest neighboring lattice sites and the on-site potentials are modulated by a cosine function which has a periodicity of $1/\alpha$. When $\alpha$ is irrational, the lattice is incommensurate, and we find that the system will go through a phase transition from the extended states to localized states when the disordered on-site potential increases beyond some critical value. This critical value as well as the critical region of the system will be significantly reduced due to the modulation in the hopping amplitude and thus makes the system much easier to be localized. On the other hand, in the commensurate lattice when $\alpha$ is rational, this model shows different topological phases as we tune the system parameters. We mainly discuss the case with $\alpha=1/2$ and find that there are three phases presented in the system: a Su-Schrieffer-Heeger (SSH)-like trivial phase, a SSH-like topological phase, and a Kitaev-like topological phase. By calculating the topological numbers of the system, we present the phase diagrams and phase boundaries for these phases. This generalized AAH model provides us with a useful test field for the study of phase transitions between extended states and Anderson localized states, or the transitions between different topological phases.
The rest of the paper is organized as follows. In Sec. \ref{sec2}, we introduce the generalized AAH model and write down the Hamiltonian. Then we consider the incommensurate case in Sec. \ref{sec3}, where we will discuss the phase transition from the extended phase to the localized phase, and especially the influences of the modulations in the hopping amplitude on this transition. In Sec. \ref{sec4}, we check the commensurate lattice with $\alpha=1/2$ and investigate those different phases presented in this system. The last section (Sec. \ref{sec5}) is dedicated to a brief summary.
\section{Model Hamiltonian}\label{sec2}
The Hamiltonian of the generalized one-dimensional (1D) Aubry-Andr\' e-Harper model we consider in this paper is described by the following Hamiltonian
\begin{equation}\label{Eq1}
H= \sum_{j=1}^{N} V_j c_j^\dagger c_j + \sum_{j=1}^{N-1} [ -t_j c_{j+1}^\dagger c_j +\Delta c_{j+1}^\dagger c_j^\dagger + H.c. ]
\end{equation}
where $c_j^\dagger$ ($c_j$) is the creation (annihilation) operator at site $j$, $V_{j} = V \cos(2\pi \alpha j + \varphi_V)$ is the on-site potential, $t_j = t [1+\lambda \cos (2\pi \alpha j + \varphi _\lambda)]$ is the hopping amplitude between the nearest neighboring lattice sites, and $\Delta$ is the superconducting pairing gap which is taken to be real. This one-dimensional chain has $N$ sites and both the on-site potential and the hopping amplitude are modulated by a cosine function with periodicity $1/\alpha$ but with phase factor $\varphi_v$ and $\varphi_\lambda$ , respectively. If $\Delta=0$, this model will reduce to the generalized AAH model introduced in Ref. \cite{Ganeshan}, which covers both the diagonal and off-diagonal AAH model. If $\lambda=0$, then the Hamiltonian is the same as the 1D Kitaev chain with on-site potential modulations which are investigated in detail in Refs. \cite{Cai, DeGottardi}. It has been predicted that if $\alpha$ is irrational, the Kitaev model $(\lambda=0, V \neq 0)$ or the diagonal AAH model $(\lambda=0, \Delta=0, V \neq 0)$ will go through a localization transition where all extended states become localized as $V$ is increased beyond some critical value. However, if $\alpha$ is rational, it is known that there is no such phase transition in these two models. For simplicity, we will set $\varphi_V = \varphi_\lambda = \varphi$ in this paper but we emphasize that all the results obtained in this paper can also be extended to the $\varphi_V \neq \varphi_\lambda$ case.
By introducing Majorana operators $c_j = (\gamma_{j,1} + i \gamma_{j,2})/2$, the Hamiltonian could be rewritten as
\begin{equation}\label{Eq2}
\begin{split}
H &= \frac{i}{2} \sum_{j} \{ V_{2j-1} \gamma_{2j-1,1} \gamma_{2j-1,2} + V_{2j} \gamma_{2j,1} \gamma_{2j,2} \\
&+ ( -t_{2j-1} - \Delta ) \gamma_{2j,1} \gamma_{2j-1,2} + ( -t_{2j} + \Delta ) \gamma_{2j,1} \gamma_{2j+1,2} \\
&+ ( t_{2j-1} - \Delta ) \gamma_{2j,2} \gamma_{2j-1,1} + ( t_{2j} + \Delta ) \gamma_{2j,2} \gamma_{2j+1,1} \}.
\end{split}
\end{equation}
Thus the system can be taken as two 1D Majorana chains coupled by potential $V_j$ and we may expect that there will be different topological phases depending on whether or not the Majorana fermions at the ends of the Majorana chains are paired. The phase transition from the extended state to the localized state could also be expected according to previous works on the quasiperiodic 1D Kitaev chain \cite{Cai, DeGottardi}. In the following, we will discuss the properties of this generalized AAH model both in the incommensurate case and the commensurate case.
\section{Incommensurate case}\label{sec3}
When $\alpha$ is irrational, the model system becomes quasiperiodic. Since the model is similar to a 1D Kitaev model with modulated on-site potentials which shows an Anderson localization transition when $V$ becomes large enough [$V > 2(t+\Delta)$], we may expect that the same transition would also show up in our system. We take $\alpha = (\sqrt{5}-1)/2$ as an example, but the results can also be generalized to other incommensurate situations. The Hamiltonian can be diagonalized by using the Bogoliubov-de Gennes (BdG) transformation \cite{Gnnes, Lieb},
\begin{equation}\label{}
\eta_n^\dagger = \sum_{j=1}^{N} [u_{n,j} c_{n,j}^\dagger + v_{n,j} c_{n,j}],
\end{equation}
where $u_{n,j}$ and $v_{n,j}$ are chosen to be real and $n$ is the energy band index. Then the wave function of the Hamiltonian is
\begin{equation}\label{}
| \Psi_n \rangle = \eta_n^\dagger | 0 \rangle = \sum_{j=1}^{N} [u_{n,j} c_{n,j}^\dagger + v_{n,j} c_{n,j}] | 0 \rangle.
\end{equation}
From the Schr\"odinger equation $\mathcal{H} | \Psi_n \rangle = \epsilon_n | \Psi_n \rangle$, we have
\begin{widetext}
\begin{equation}
\left\{
\begin{aligned}
-t_{j-1} u_{n,j-1} + \Delta v_{n,j-1} + V_j u_{n,j} - t_j u_{n,j+1} - \Delta v_{n,j+1} &= \epsilon_n u_{n,j},\\
-\Delta u_{n,j-1} + t_{j-1} v_{n,j-1} - V_j v_{n,j} + \Delta u_{n,j+1} + t_j v_{n,j+1} &= \epsilon_n v_{n,j}.
\end{aligned}
\right.
\end{equation}
By representing the wave function as
\begin{equation}\label{}
| \Psi_n \rangle = [u_{n,1}, v_{n,1}, u_{n,2}, v_{n,2}, \cdots, u_{n,N}, v_{n,N}]^T,
\end{equation}
the Hamiltonian can be written as a $2N \times 2N$ matrix,
\begin{equation}\label{}
\mathcal{H}_n=
\begin{pmatrix}
A_1 & B & 0 & \cdots & \cdots & \cdots & C \\
B^\dagger & A_2 & B & 0 & \cdots & \cdots & 0 \\
0 & B^\dagger & A_3 & B & 0 & \cdots & 0 \\
\vdots & \ddots & \ddots & \ddots & \ddots & \ddots & \vdots \\
0 & \cdots & 0 & B^\dagger & A_{N-2} & B & 0 \\
0 & \cdots & \cdots & 0 & B^\dagger & A_{N-1} & B \\
C^\dagger & \cdots & \cdots & \cdots & 0 & B^\dagger & A_N
\end{pmatrix},
\end{equation}
\end{widetext}
where
\begin{equation}\label{}
A_j =
\begin{pmatrix}
V_j & 0 \\
0 & -V_j
\end{pmatrix},
\end{equation}
\begin{equation}\label{}
B=\begin{pmatrix}
-t_j & -\Delta \\
\Delta & t_j
\end{pmatrix},
\end{equation}
and
\begin{equation}\label{}
C=\begin{pmatrix}
-t_{j+1} & \Delta \\
-\Delta & t_{j+1}.
\end{pmatrix}
\end{equation}
Here we have assumed a periodic boundary condition which implies that $c_{j+N}=c_j$. This is legitimate when $N$ is large enough since the irrational number $\alpha$ can be approximated by rational numbers \cite{Wang}. From this Hamiltonian, we can get $u_{n,j}$ and $v_{n,j}$ and thus the wave function of the system.
In order to investigate the phase transition of this incommensurate 1D chain, we need to calculate the inverse participation ratio (IPR) which for a normalized wave function $\Psi_n$ is defined as $IPR=\sum_j (u_{n,j}^4 + v_{n,j}^4)$. The IPR measures the inverse of the number of occupied lattice sites and is a very useful quantity in characterizing the localization transitions of quasiperiodic systems. We further define the mean inverse participation ratio (MIPR) as $MIPR = \frac{1}{2N} \sum_{n=1}^{2N} \sum_{j=1}^{N}(u_{n,j}^4 + v_{n,j}^4)$. MIPR is close to zero for extended states of the system, however, it will tend to a finite value of $O(1)$ for localized states.
\begin{figure}[!ht]
\centering
\subfigure[MIPR vs V]{
\label{fig1a}
\includegraphics[width=3.1in]{fig1a.png}}
\subfigure[MIPR vs $\lambda$]{
\label{fig1b}
\includegraphics[width=3.1in]{fig1b.png}}
\caption{(Color online) (a) MIPR vs V under different $\lambda$; (b) MIPR vs $\lambda$ under different V. Here we set $\Delta=0.2t$, $\varphi=0$, and the number of lattice sites $N=377$, which is the 14th Fibonacci number.}
\label{fig1}
\end{figure}
Figure \ref{fig1} shows the variation of the MIPR as the parameter of the system changes. Throughout this paper, we take $t=1$ as the energy unit. Other parameters in Fig. \ref{fig1} are $\Delta=0.2t$, $\varphi=0$, and the number of lattice sites $N=377$. The solid blue dots in Fig.\ref{fig1a} corresponds to the usual 1D Kitaev chain with a quasiperiodic disordered on-site potential ($\lambda=0, V \neq 0$). The phase transition takes place when $V_c>2(t+\Delta)$, as was reported in earlier work \cite{Cai}. There is a plateau in the MIPR before it increases sharply when $\lambda=0$, which corresponds to the critical phase before the system enters the localized phase, which is discussed in Ref. \cite{Wang}. If $\lambda$ becomes nonzero and gets stronger, as is shown in Fig. \ref{fig1a}, the critical value of the phase transition $V_c$ is reduced and the critical region will disappear gradually. It becomes more significant when $\lambda$ is close to the value of $t$, as shown by the triangular green dots and the open purple dots in Fig. \ref{fig1a}, where the MIPR increases rapidly at a much smaller value of $V$. This is understandable since as $\lambda$ increases, especially when it reaches the strength of $t$, the modulation in the hopping amplitude will become increasingly stronger, the coupling strength between some lattice sites then will be weakened and the system thus can be localized more easily and more quickly. In Fig. \ref{fig1b}, we present the evolution of the MIPR as a function of $\lambda$ with different $V$. From this figure, we can see that when $V=0$, the MIPR will increase from zero to a finite though small value, which means that even without the disorders coming from the on-site potentials, the modulation in the hopping amplitude alone can also influence the extended state of the system. When $V \neq 0$, it is clear that the MIPR increases sharply when $\lambda$ is close to $1$ and reaches to a maximum, then drops again when $\lambda$ becomes stronger. This signifies that the system is more likely to be localized when the hopping amplitudes are also modulated, which is consistent with the results we get in Fig. \ref{fig1a}.
So due to the modulation in the hopping amplitude between the nearest neighboring lattice sites, we find that the incommensurate generalized AAH model with p-wave superconducting pairing is easier to be localized when tuning the disordered on-site potential. The critical value of the potential beyond which the phase transition happens is significantly reduced especially when the modulation in the hopping amplitude is strong. These results can also be generalized to situations with other irrational $\alpha$ values.
\section{Commensurate case}\label{sec4}
When $\alpha$ is rational, the lattice is commensurate. It is well known that the system will not undergo a localization transition as that in the incommensurate case. In this section, we mainly discuss the $\alpha=1/2$ case. The generalized AAH model with p-wave superconducting pairing in this case can be tuned between different topological phases including the Su-Schrieffer-Heeger-like (SSH-like) topological phase and the Kitaev-like topological phase. When $\alpha=1/2$, the Hamiltonian in Eq. (\ref{Eq1}) becomes
\begin{widetext}
\begin{equation}\label{}
\begin{split}
H= &-V \cos \varphi \sum_j ( c_{2j-1}^\dagger c_{2j-1} - c_{2j}^\dagger c_{2j} ) -t \sum_j [(1-\lambda \cos \varphi) c_{2j}^\dagger c_{2j-1} + (1+\lambda \cos \varphi) c_{2j+1}^\dagger c_{2j} + H.c. ] \\
& +\Delta \sum_j ( c_{2j}^\dagger c_{2j-1}^\dagger + c_{2j+1}^\dagger c_{2j}^\dagger + H.c. ) \\
=& -V \cos \varphi \sum_j ( c_{A,j}^\dagger c_{A,j} - c_{B,j}^\dagger c_{B,j} ) - t \sum_j [ (1-\lambda \cos \varphi) c_{B,j}^\dagger c_{A,j} + (1+\lambda \cos \varphi ) c_{A,j+1}^\dagger c_{B,j} + H.c. ]\\
&+\Delta \sum_j ( c_{B,j}^\dagger c_{A,j}^\dagger + c_{A,j+1}^\dagger c_{B,j}^\dagger + H.c. ).
\end{split}
\end{equation}
\end{widetext}
This Hamiltonian is very similar to the one discussed in Ref. \cite{Wakatsuki}, except that here the modulations are added to the on-site potential and the hopping amplitude by the function $\lambda \cos \varphi$. It is reduced to the SSH model when $\mu=\Delta=0$ and to the 1D Kitaev model when $\lambda=0$ except that the chemical potential is positive or negative alternatively. Since both SSH-like and Kitaev-like models are included in this generalized system, we may expect that with appropriate system parameters, different topological phases would show up. To verify this we can check the energy spectrum of the system, which can be reached by diagonalizing the Hamiltonian rewritten in the Majorana fermions representation [see Eq. (\ref{Eq2}))].
\begin{figure}[!ht]
\centering
\includegraphics[width=3.1in]{fig2.png}
\caption{(Color online) Energy vs $\varphi$ with different on-site chemical potential $V$ and different site numbers. In (a) and (b), the site number is chosen to be even, $N=100$, while in (c) and (d), it is chosen to be odd, $N=101$. Other parameters are $\lambda=0.4t$, $\Delta=0.2t$. The zero energy modes in the gap are doubly degenerate when $V$ is zero. However, if $V$ takes nonzero values, some of the zero modes split and become nondegenerate. The regions with zero energy modes unchanged when $V$ is nonzero correspond to the Kitaev-like topological phase, while those regions with zero energy modes splitting in the finite $V$ situation correspond to the SSH-like topological phase.}
\label{fig2}
\end{figure}
In Fig. \ref{fig2}, we present the spectra of the system with an even site number ($N=100$) and an odd site number ($N=101$). From Fig. \ref{fig2}(a), we can see that there are gapped topologically trivial regimes and nontrivial regimes with zero energy modes. However, comparing with Fig. \ref{fig2}(b), it is clear that when $V$ becomes nonzero, parts of the zero modes in the topologically nontrivial regime split into two non zero energy modes, but the rest of the zero modes remain unchanged. More interestingly, when the site number is odd, there are also zero modes in the trivial regime when the site number is even and those zero modes also split when $V$ becomes nonzero [see Fig. \ref{fig2}(c) and \ref{fig2}(d)]. This even-odd effect in the energy spectrum is a feature of the SSH model. The regime with zero energy modes which will not split when $V$ is finite is the Kitaev topological regime. So there are three phases in this generalized system: a SSH-like trivial phase, a SSH-like topological phase, and a Kitaev-like topological phase. In the SSH-like topological phase, there will be one Dirac fermion at each end of the chain, while in the Kitaev-like topological phase, there will be one Majorana fermion at each end of the chain. The zero energy modes originate from the particle-hole symmetry which has two sources: In the SSH-like phase, it is the sublattice symmetry (when $V=0$) while in the Kitaev-like phase, it is the superconductivity. As the superconducting pairing strength becomes stronger, the superconductivity will dominate and there will be no SSH-like phase in the system. Figure \ref{fig3} shows the energy spectra of the system with $\Delta =0.4t$ and $0.5t$ both in the even and odd site number situations. When $\Delta=0.4t$, there is only a very small region which shows the even-odd effect, as shown in Figs. \ref{fig3}(a) and \ref{fig3}(b). When $\Delta$ becomes larger, there are always zero modes as long as the system is in the topological nontrivial phase [Fig. \ref{fig3}(c) and \ref{fig3}(d)]. So the Kitaev-like phase dominates and the SSH-like phase disappears. As the parameters change, there would be a phenomenon of electron fractionization to the Majorana fermions \cite{Wakatsuki}.
\begin{figure}[!ht]
\centering
\includegraphics[width=3.1in]{fig3.png}
\caption{(Color online) Energy vs $\varphi$ with different on-site chemical potential $V$ and different site numbers. $\lambda$ is set to be $0.4t$. (a) $\Delta=0.4t$, $V=0.0$, $N=100$; (b) $\Delta=0.4t$, $V=0.1t$, $N=101$; (c) $\Delta=0.5t$, $V=0.0$, $N=100$; and (d) $\Delta=0.5t$, $V=0.1t$, $N=101$.}
\label{fig3}
\end{figure}
To give a more intuitive picture of the different topological phases in this commensurate situation, we illustrate the schematic diagrams of the AAH model in Fig. \ref{fig4} by setting $\alpha=1/2$ in Eq. (\ref{Eq2}). As shown by Fig. \ref{fig4}(a), the AAH model can be taken as two Majorana fermion chains which are coupled by the on-site chemical potentials. The different bonding strengths between the Majorana fermions will lead to different topological phases. When the system is in the SSH-like topological phase, there is a fermion at each end of the one-dimensional chain, [see Fig. \ref{fig4}(b)]. If the system is in the Kitaev-like topological phase, there will be an unpaired Majorana fermion at each end of the chain, which is shown in Fig. \ref{fig4}(c).
\begin{figure}[!ht]
\centering
\includegraphics[width=3.2in]{fig4.png}
\caption{(Color online) Illustrations of the AAH model and the different topological phases in the commensurate case. (a) The AAH model can be taken as two coupling Majorana fermion chains according to Eq. (\ref{Eq2}) (here we set $\alpha=1/2$). $\gamma_1$ and $\gamma
_2$ are the Majorana fermions composing the fermionic particle at each lattice site. The bonds with different colors and thicknesses represent the coupling strength between the Majorana fermions. (b) SSH-like topological phase with an unpaired fermion at each end of the chain. (c) In the Kitaev-like topological phase, there is an unpaired Majorana fermion at each end of the chain.}
\label{fig4}
\end{figure}
In order to characterize these different topological phases and determine the phase boundaries, we need to calculate the topological numbers. First, we transform the Hamiltonian into the momentum representation. By defining
\begin{equation*}
\Psi = [c_{qA}^\dagger \ c_{qB}^\dagger \ c_{-qA} \ c_{-qB}]^T,
\end{equation*}
the Hamiltonian can be written as $\mathcal{H}=\frac{1}{2} \sum_q \Psi^\dagger \mathcal{H} (q) \Psi$, where
\begin{equation}\label{}
\mathcal{H} (q)=
\begin{pmatrix}
-V\cos \varphi & g(q) & 0 & h(q) \\
g^*(q) & V\cos \varphi & -h^*(q) & 0 \\
0 & -h(q) & V\cos \varphi & -g(q) \\
h^*(q) & 0 & -g^*(q) & -V\cos \varphi
\end{pmatrix}
\end{equation}
with $g(q) = -t [ (1-\lambda \cos \varphi) + (1+ \lambda \cos \varphi) e^{-iq} ]$ and $h(q) = -\Delta (1 - e^{-iq})$. After diagonalizing $\mathcal{H} (q)$ the eigenvalues can be obtained as
\begin{equation}\label{}
\begin{split}
E^2&(q) = (V \cos \varphi)^2 + |g(q)|^2 + |h(q)|^2 \\
& \pm 2\sqrt{(V\cos \varphi)^2 |h(q)|^2 + 4\Delta^2 t^2 (\lambda \cos \varphi)^2 (\cos q -1)^2}.
\end{split}
\end{equation}
Thus we have
\begin{equation}\label{eigenvalues}
\begin{split}
E(0)=&\pm \sqrt{(V\cos \varphi)^2 + 4t^2}, \\
E(\frac{\pi}{2}) =& ( \sqrt{(V\cos \varphi)^2 + 2t^2 (\lambda \cos \varphi)^2} \pm \sqrt{2} \Delta)^2 + 2t^2, \\
E(\pi) =& \pm \sqrt{(V\cos \varphi)^2 + 4t^2 (\lambda \cos \varphi)^2} \pm 2\Delta.
\end{split}
\end{equation}
The gap will close at $q=\pi$ when $(V\cos \varphi)^2 = 4[\Delta^2 - t^2 (\lambda \cos \varphi)^2]$. For $\Delta=0$, the spectrum is reduced to
\begin{equation}\label{}
\begin{split}
&E(q)=\\
&\pm \sqrt{V^2 \cos^2 \varphi + 2t^2 [ 1+\lambda^2 \cos^2 \varphi + (1-\lambda^2 \cos^2 \varphi) \cos q ]}
\end{split}
\end{equation}
which is similar to the SSH model spectrum. If $\lambda=0$, then the spectrum is reduced to
\begin{equation}\label{}
E(q) = \pm \sqrt{(V\cos \varphi \pm 2\Delta \sin \frac{q}{2})^2 + 4t^2 \cos^2 \frac{q}{2}},
\end{equation}
which is similar to the Kitaev model spectrum. The difference here is that the system is topological for $|V \cos \varphi| <2|\Delta|$ and trivial for $|V \cos \varphi|>2|\Delta|$. However, the spectrum for a usual Kitaev model is $E(q)=\pm \sqrt{(2t\cos \frac{q}{2} - \mu)^2 + 4 \Delta^2 \sin^2 \frac{q}{2}}$, where the system is topological for $|\mu|<2t$ and trivial for $|\mu|>2t$. In order to make further investigations about the topological phases of the commensurate lattice, we need to check the topological quantum numbers of the system with different parameters.
Following the discussions in Ref. \cite{Wakatsuki}, we find that the topological class of our model is also BDI and we can characterize our system by the $\mathbb{Z}$ index. Now we calculate the topological numbers of the system. Firstly, we consider the sublattice symmetric case ($V=0$) whose topological number can be calculated as (check the Appendix for details)
\begin{equation}\label{}
N_1 = \Theta (t \lambda \cos \varphi -\Delta) + \Theta (t\lambda \cos \varphi +\Delta)
\end{equation}
with $\Theta(x)$ being the step function. Figure \ref{fig5}(a) illustrates the phase diagram of the system according the value of topological number $N_1$. There are three phases: (i) When $|\Delta|< t \lambda |\cos \varphi|$, $\cos \varphi < 0$, $N_1=0$, the system is in the SSH-like trivial phase; (ii) when $|\Delta| < t\lambda |\cos \varphi|$, $\cos \varphi >0$, $N_1 =2$, the system is in the SSH-like topological phase; (iii) when $|\Delta|>t\lambda |\cos \varphi|$, $N_1=1$, the system is in the Kitaev-like topological phase. As shown in this figure, when $|\Delta| > \lambda$ (here $\lambda=0.4t$), the system will be totally in the Kitaev-like topological phase, which is consistent with the numerical results for the energy spectra we presented in Fig. \ref{fig3}. The topological number $N_1$ can be interpreted as the number of Majorana fermion pairs at the ends of the chain. So when $N_1=2$, there will be two Majorana fermions at each end of the chain which will pair up with each other and become a Dirac fermion, which is the feature of the SSH model. If $N_1=1$, there will be one single unpaired Majorana fermion at each end of the chain, so the system is in the Kitaev-like topological phase.
\begin{figure}[!ht]
\centering
\includegraphics[width=3.1in]{fig5.png}
\caption{(Color online) (a) Phase diagram of the system in the $V=0$ case; (b) phase boundary of the system in the $V\neq 0$ case; (c) and (d) show the cross sections of (b) at $V=0.1t$ and $V=0.5t$, respectively. The numbers in the figures denote the topological number $N_1$ or $N_2$. Here $\lambda$ is also set to be $0.4t$.}
\label{fig5}
\end{figure}
Next we check the $V\neq 0$ case. Using similar techniques, the topological number $N_2$ for this sublattice asymmetric case is (see the Appendix)
\begin{equation}\label{}
N_2 = - \int_{-\pi}^{\pi} \frac{dq}{2\pi i} \partial_q \ln Z(q),
\end{equation}
where
\begin{equation}\label{}
\begin{split}
Z(q) =& Det M_2(q) = (V \cos \varphi)^2 + (g-h)(g^* + h^*) \\
=& (V \cos \varphi)^2 + 2[ t^2(1+\lambda^2 \cos^2 \varphi) - \Delta^2 ] \\
& + 2[ t^2(1-\lambda^2 \cos^2 \varphi) + \Delta^2 ]\cos q - 4i t\Delta \sin q.
\end{split}
\end{equation}
Here, $N_2$ is the winding number of $Z(q)$, and is determined by the cross points of the real axis at $q=0$ and $\pi$. When $\Delta > 0$, we have
\begin{equation}\label{}
\begin{split}
Z(0)Z(\pi) < 0 &\Rightarrow N_2=1, \\
Z(0)Z(\pi) > 0 &\Rightarrow N_2=0,
\end{split}
\end{equation}
with
\begin{equation}\label{}
\begin{split}
Z(0) =& (V \cos \varphi)^2 + 4 t^2, \\
Z(\pi) =& (V\cos \varphi)^2 + 4(t^2 \lambda^2 \cos^2 \varphi - \Delta^2).
\end{split}
\end{equation}
When $\Delta<0$, we have $N_2=-1$ in the topological regime but it is equivalent to the $N_2=1$ phase. In Figs. \ref{fig5}(b) - Fig. \ref{fig5}(d), we present the phase diagrams of the system according to the topological number $N_2$. There are two phases when $V \neq 0$: $N_2 = \pm1$ is the Kitaev-like topological phase, and $N_2=0$ is the trivial phase. The phase boundary between these two phases is $(V\cos \varphi)^2 = 4[\Delta^2 - t^2 (\lambda \cos \varphi)^2]$, which is the gap closing condition we got earlier [see Eq. (\ref{eigenvalues})].
So, by checking the energy spectra and the topological numbers of the commensurate AAH model, we can completely describe the phases of the system. In addition, we can also tune the system into different topological phases by changing the system parameters such as the superconducting pairing amplitude and the modulation of the hopping amplitude, etc.
\section{Summary}\label{sec5}
In this paper, we have investigated a generalized Aubry-Andr\' e-Harper model with superconducting pairing both in the incommensurate and commensurate case. Due to the modulations in the hopping amplitude and the on-site potential, this model shows aboundant physical phenomena as we tune the system parameters. When the lattice is incommensurate [$\alpha=(\sqrt{5}-1)/2$], the system shows a phase transition from extended states to localized states when the disordered potential becomes larger than some critical value. This critical value as well as the critical region will be reduced due to the modulation in the hopping amplitude especially when this modulation become strong. On the other hand, in the commensurate lattice case ($\alpha=1/2$), the system presents three different phases in different parameter regimes: a SSH-like trivial phase, a SSH-like topological phase, and a Kitaev-like topological phase. The topological numbers and the phase diagram are illustrated in this work. This generalized AAH model combined with superconducting pairing unifies those different phases and can provide an ideal test field for different kinds of phenomena. All these results can be generalized to discuss other cases with different $\alpha$ values, both rational and irrational.
As proposed in Ref. \cite{Ganeshan} (see the Supplemental Material there), a generalized AAH model with modulations in the hopping amplitude could be realized in cold atom systems. The generalized Aubry-Andr\'e-Harper model we propose here, though more complicated, might also be realized in cold atom optical lattices \cite{Billy, Roati}, or semiconductor structures \cite{Sarma, Merlin} in the future. This generalized AAH model may also have potential applications in understanding the topological properties of other similar systems, as discussed in Refs. \cite{Li, Ganeshan2}.
\section*{Acknowledgments}
This work has been supported by the NSFC under Grant No. 11274195 and the National Basic Research Program of China (973 Program) Grants No. 2011CB606405 and No. 2013CB922000. S. C. is supported by NSFC under Grants No. 11425419, No. 11374354 and No. 11174360, and the Strategic Priority Research Program (B) of the Chinese Academy of Sciences (No. XDB07020000).
\section*{Appendix}
\setcounter{equation}{0}
\renewcommand{\theequation}{{A}.\arabic{equation}}
In this Appendix, we investigate the topological class and calculate the topological numbers of the system. This part is mainly following the discussions in Ref. \cite{Wakatsuki}. Our model is time-reversal symmetric, $T \mathcal{H} (q) T^{-1} = \mathcal{H} (-k)$, since the coefficients $V$, $t$, $\lambda$ and $\Delta$ are all real. The time-reversal operator is defined as $T=K$, which takes the complex conjugate. When $V=0$, the system has sublattice symmetry. The operator for sublattice symmetry is defined by
\begin{equation}\label{}
C_1 = \sigma_z =
\begin{pmatrix}
1 & 0 & 0 & 0 \\
0 & -1 & 0 & 0 \\
0 & 0 & 1 & 0 \\
0 & 0 & 0 & -1
\end{pmatrix},
\end{equation}
where $\sigma_z$ is the Pauli matrix acting on the sublattice degree of freedom. It can be shown that $C_1 \mathcal{H} (q) C_1^{-1} = -\mathcal{H}(q)$. Since we have $T^2=1$ and $C_1^2 = 1$, the topological class of the system is BDI.
If $V \neq 0$, the system will no longer have sublattice symmetry anymore. However, due to the superconducting pairing, particle-hole symmetry can still be guaranteed. The particle-hole operator is defined by $P=\tau_x K$, with $\tau_x$ being the Pauli matrix acting on the particle-hole space. The Hamiltonian satisfies $P \mathcal{H} (q) P^{-1} = -\mathcal{H} (-q)$. We can introduce the chiral operator as
\begin{equation}\label{}
C_2 = TP = \tau_x =
\begin{pmatrix}
0 & 0 & 1 & 0 \\
0 & 0 & 0 & 1 \\
1 & 0 & 0 & 0 \\
0 & 1 & 0 & 0
\end{pmatrix}.
\end{equation}
Then we have $C_2 \mathcal{H}(q) C_2^{-1} = -\mathcal{H} (q)$ and $C_2^2 = 1$, so the topological class is still BDI. The 1D system in the BDI class is characterized by the $\mathbb{Z}$ index.
Now we calculate the topological numbers of the system. Firstly, we consider the sublattice symmetric case ($V=0$) whose topological number is defined as
\begin{equation}\label{}
N_1 = Tr \int_{-\pi}^{\pi} \frac{dq}{4\pi i} C_1 g^{-1} \partial_q g,
\end{equation}
where $g(q)=-\mathcal{H}^{-1} (q)$ is the Green's function at zero energy. $N_1$ is equivalent to the chiral index. After introducing a unitary transformation
\begin{equation}\label{}
U_1 =
\begin{pmatrix}
1 & 0 & 0 & 0 \\
0 & 0 & 1 & 0 \\
0 & 1 & 0 & 0 \\
0 & 0 & 0 & 1
\end{pmatrix},
\end{equation}
we have
\begin{equation}\label{}
U_1 C_1 U_1^\dagger = \tau_z, \hspace{1cm} U_1 \mathcal{H} U_1^\dagger =
\begin{pmatrix}
0 & M_1 \\
M_1^\dagger & 0
\end{pmatrix},
\end{equation}
with
\begin{equation}\label{}
M_1=
\begin{pmatrix}
g & h \\
-h & -g
\end{pmatrix}.
\end{equation}
where $g(q)$ and $h(q)$ are defined earlier. Then the chiral index is given by
\begin{equation}\label{}
\begin{split}
N_1= & -Tr \int_{-\pi}^{\pi} \frac{dq}{2\pi i} M_1^{-1} \partial_q M_1 \\
=& - \int_{-\pi}^{\pi} \frac{dq}{2\pi i} \partial_q \ln Det[M_1]\\
=& - \int_{-\pi}^{\pi} \frac{dq}{2\pi i} \partial_q \ln [(h-g)(h+g)]
\end{split}
\end{equation}
with
\begin{equation}\label{}
\begin{split}
h -g =& [t (1- \lambda \cos \varphi) - \Delta] + [ t(1+\lambda \cos \varphi) + \Delta ]e^{-iq}, \\
h+g =& [-t(1-\lambda \cos \varphi) - \Delta] - [t(1+\lambda \cos \varphi) - \Delta] e^{-iq}.
\end{split}
\end{equation}
So the topological number $N_1$ is
\begin{equation}\label{}
N_1 = \Theta ( t \lambda \cos \varphi - \Delta) + \Theta (t\lambda \cos \varphi + \Delta )
\end{equation}
with $\Theta(x)$ being the step function.
Next we check the $V\neq 0$ case. The topological number $N_2$ associated with the chiral operator $C_2$ is
\begin{equation}\label{}
N_2 = Tr \int_{-\pi}^{\pi} \frac{dq}{4\pi i} C_2 g^{-1} \partial_q g.
\end{equation}
Considering a unitary transformation
\begin{equation}\label{}
U_2= \frac{1}{\sqrt{2}}
\begin{pmatrix}
1 & 0 & 1 & 0 \\
0 & 1 & 0 & 1 \\
-i & 0 & i & 0 \\
0 & -i & 0 & i
\end{pmatrix},
\end{equation}
which corresponds to the Majorana fermion representation
\begin{equation}\label{}
c_j = \frac{1}{2}(\gamma_{j,1} + i \gamma_{j,2}), \hspace{1cm} c_j^\dagger = \frac{1}{2}(\gamma_{j,1} - i \gamma_{j,2}).
\end{equation}
It is easy to check that
\begin{equation}\label{}
U_2 C_2 U_2^\dagger = \tau_z, \hspace{1cm} U_2 \mathcal{H}U_2^\dagger =
\begin{pmatrix}
0 & M_2 \\
M_2^\dagger & 0
\end{pmatrix},
\end{equation}
with
\begin{equation}\label{}
M_2=
\begin{pmatrix}
-i V\cos \varphi & i(g-h) \\
i(g^* + h^*) & i V\cos \varphi
\end{pmatrix}.
\end{equation}
Thus we have
\begin{equation}\label{}
N_2 = -Tr \int_{-\pi}^{\pi} \frac{dq}{2\pi i} M_2^{-1} \partial_q M_2 = - \int_{-\pi}^{\pi} \frac{dq}{2\pi i} \partial_q \ln Z(q),
\end{equation}
where
\begin{equation}\label{}
\begin{split}
Z(q) =& Det M_2(q) = (V \cos \varphi)^2 + (g-h)(g^* + h^*) \\
=& (V \cos \varphi)^2 + 2[ t^2(1+\lambda^2 \cos^2 \varphi) - \Delta^2 ] \\
& + 2[ t^2(1-\lambda^2 \cos^2 \varphi) + \Delta^2 ]\cos q - 4i t\Delta \sin q.
\end{split}
\end{equation}
$N_2$ is the winding number of $Z(q)$, and is determined by the cross points of the real axis at $q=0$ and $\pi$. For $\Delta > 0$, we find
\begin{equation}\label{}
\begin{split}
Z(0)Z(\pi) < 0 &\Rightarrow N_2=1, \\
Z(0)Z(\pi) > 0 &\Rightarrow N_2=0,
\end{split}
\end{equation}
with
\begin{equation}\label{}
\begin{split}
Z(0) =& (V \cos \varphi)^2 + 4 t^2, \\
Z(\pi) =& (V\cos \varphi)^2 + 4(t^2 \lambda^2 \cos^2 \varphi - \Delta^2).
\end{split}
\end{equation}
For $\Delta<0$, we have $N_2=-1$ in the topological regime but it is equivalent to the $N_2=1$ phase.
| 7923d57751f0940ffe5657f50eeeaf8ba5dca26c | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section*{Results}
\begin{figure}
\includegraphics[width=80mm]{figures/fig2_simulated_errors.eps}
\caption{\label{fig:simErr} Actual and estimated errors in simulated
data. (a) Root mean square (RMS) error
$\sqrt{\mean{(\mu_\text{est.}-\mu_\text{true})^2}}$ vs.~root mean
estimated error variance $\sqrt{\mean{\varepsilon^2}}$ for MLE
localization from the CRLB and Laplace approximations. (b) PSF
width prior and fitted parameter distributions with (MAP) and
without (MLE) prior. (c) Actual and estimated errors for MAP
localization. (d) Probability plot of point-wise errors normalized
by point-wise Laplace uncertainty estimates,
$(\mu_\text{est.}-\mu_\text{true})/\sqrt{\varepsilon^2}$. The dashed
line indicates the reference standard Gaussian distribution.}
\end{figure}
\subsection*{Estimating point-wise uncertainty}
Estimating uncertainty is closely related to estimating positions,
where the maximum likelihood estimate (MLE) is generally considered
the optimal method. A maximum likelihood method starts with a
likelihood function, i.e., the probability density function of a
probabilistic model for generating images of spots (pixel counts in a
small region around a spot) with the emitter position among the
adjustable parameters. The MLE are the parameters that maximize the
likelihood function for a particular spot image. Following common
practice, we model EMCCD camera noise with the high-gain approximation
and Gaussian readout noise \cite{chao2013,mortensen2010}, and the spot
shape by a symmetric Gaussian intensity profile plus a constant
background intensity. The fit parameters are thus spot position
$(\mu_x,\mu_y)$, background intensity $b$, PSF width $\sigma$, and
spot amplitude $N$ (see Methods, \Eq{eq:gaussPSF}), while the camera
noise is assumed known from calibration.
What is localization uncertainty? The localization error is the difference
$\mu_\text{est.}-\mu_\text{true}$ between estimated and true
positions. By localization uncertainty, we seek the distribution of
the error, either in a Bayesian posterior sense, or in the sense of
repeated localizations of spots with the same uncertainty. The
uncertainty of the localization is related to the shape of the
likelihood maximum: a sharply peaked maximum means that only a narrow
set of parameters are likely, while a shallow maximum means greater
uncertainty.
The Cram\'er-Rao lower bound (CRLB) is the smallest possible
variance of an unbiased estimator for a given set of model
parameters, and is related to the expected sharpness of the
likelihood maximum (see Methods, \Eq{eq:CRLB}).
While this is strictly speaking not a statement about a single image,
but rather about the average information content of data generated by
a model, it is often used to estimate localization uncertainty. A
Bayesian alternative is the Laplace approximation \cite{mackay2003},
which derives an approximate Gaussian posterior distribution for the
fit parameters in terms of the sharpness of the likelihood maximum for
each particular image (see Methods, \Eq{eq:laplace}). Both estimators
supply an estimated variance $\varepsilon^2$, but none of them are
well characterized as estimators of localization uncertainty.
To test these estimators, we analyzed a large set of simulated movies
of a fluorescent particle diffusing at D=1
\si{\square\micro\metre\per\second} in an \textit{E. coli}-like
geometry. The movies cover a broad range of experimentally relevant
imaging conditions (see Methods) and include realistic EMCCD noise,
background fluorescence, a non-Gaussian and space-dependent PSF
\cite{kirshner2013}, and motion blur \cite{deschout2012}.
A basic consistency check is that the average estimated error
variance, $\mean{\varepsilon^2}$, agrees with the variance of the
actual errors, $\mean{(\mu_\text{est.}-\mu_\text{true})^2}$. In
Fig.~\ref{fig:simErr}a, we compare the square root of these quantities
for different imaging conditions, based on MLE localizations combined
with either CRLB or Laplace uncertainty estimators. We obtain
consistency under good imaging conditions, where the spots are bright
and the average errors low. However, as conditions worsen and the
errors increase, the uncertainty is underestimated, especially by the
CRLB.
There are several possible reasons for this behavior. The Laplace
approximation is based on a truncated Taylor expansion of the log
likelihood (see Methods). The CRLB is strictly not a point-wise error
estimator at all, and is further defined in terms of the true
parameter values, although by necessity evaluated with the fitted
ones. The simplified Gaussian PSF model performs well for localization
\cite{mortensen2010}, but this does not guarantee good uncertainty
estimates. Any of these approximations might fail in noisy or low
light conditions. Another possibility is that localizations under
poor imaging conditions are corrupted by a sub-population of fits that
converge to a local minimum that does not reflect the underlying spot
shape, e.g., fitting a single bright pixel as a very narrow spot, or
misinterpreting the PSF shoulders as background or vice versa.
To address the latter complication, we constrained the localizations
using prior distributions on selected parameters, thus replacing MLE
with maximum a aposteriori estimation (MAP). Fixing parameter values
is not suitable here, since both the size and shape of spots fluctuate
due to fluorophore motion and varying imaging conditions. Furthermore,
it is experimentally easier to obtain independent information about
the background and PSF width than about the spot intensity. We
therefore limit our attention to background and PSF width, and found
the following priors to perform well:
a log-normal prior centered on the true value for the background
intensity, and a skewed log-normal \cite{azzalini1985} prior for the
PSF width to penalize fits with unphysical widths below that of an
in-focus spot.
The PSF width prior together with the distribution of fitted PSF
widths are shown in Fig.~\ref{fig:simErr}b. PSF widths below 1 pixel
(80 \si{\nano\metre}) are virtually eliminated in the MAP
fits. Background and spot amplitudes (see Fig.~\ref{SI:fig:parPrior})
are shifted somewhat downwards and upwards, respectively. As seen in
Fig.~\ref{fig:simErr}c, this substantially improves the agreement
between true and estimated errors under all imaging conditions. The
Laplace estimator still outperforms the CRLB by a small margin, and
numerical experiments on a wider range of priors (see SI
Fig.~\ref{SI:fig:priors_SI}) further confirm that the Laplace
estimator is more robust to non-optimal priors than the CRLB.
How does the prior help? A direct comparison of the true errors for
the MLE and MAP fits (SI Fig.~\ref{SI:fig:MLE_vs_MAPerrors}) reveals
very small differences, indicating that the improvement is mainly due
to improved uncertainty estimates. This is consistent with the
theoretical observation that the Fisher information matrix for the
localization problem is nearly block-diagonal \citep{mortensen2010}
with very weak coupling between two groups of fitting parameters:
positions ($\mu_x,\mu_y$) and shape parameters ($N,b,\sigma$). Thus,
additional information about one of these groups does not reduce the
errors of the other significantly. For uncertainty estimation, this
is not the case: information about the background and PSF shape does
help in estimation position uncertainty.
While mean-square errors are useful, we are ultimately interested in
the full distribution of errors. In particular, most
\cite{berglund2010,calderon2016,relich2016,bernstein2016,koo2015,slator2015,calderon2014,vestergaard2014}
(but not all \cite{ashley2015}) statistical models of SPT data assume
Gaussian errors, but this assumption has not been tested. The Laplace
approximation (\Eq{eq:laplace}) is a Gaussian approximation. If it was
exact, the errors normalized by the estimated standard deviation would
be Gaussian with unit variance, and produce a straight line in the
Gaussian probability plot of Fig.~\ref{fig:simErr}d. The actual
normalized errors based on MAP localizations is more Gaussian than the
MLE ones, and follow the straight line for almost four standard
deviations. This shows that the improved consistency of the MAP
estimates also translates to more Gaussian error distributions.
Overall, these results show that point-wise uncertainty estimation
using the Laplace approximation works well in a wide range of
experimentally relevant conditions, but that experimentally accessible
additional information about background and PSF shape is necessary to
get consistent results in low light conditions. Moreover, we see good
support for the common assumption of Gaussian errors.
\begin{figure*}
\centering
\includegraphics[width=160mm]{figures/fig3_realData.eps}
\caption{\label{fig:realData} Validating estimators of localization
uncertainty using real data. (a) Intensity in different frames for
two different beads and the background. Excitation intensity
indicated by yellow background. (b) Image examples. Top: background
in high intensity frames. Middle: Bead 1 in a high intensity
frame. Bottom: Bead 1 in a low intensity frame. (c) Actual and
estimated RMS errors for different beads.}
\end{figure*}
\subsection*{Validation on real data}
To test the above conclusions on real data, we imaged immobilized
fluorescent beads, alternating strong and weak excitation as shown in
Fig. \ref{fig:realData}a. We used images under strong excitation
conditions to extract a high accuracy ground truth for testing the
uncertainty estimates in the dim images. We estimated the position
and uncertainty of spots using the MAP estimation described above,
with a background prior centered around the mean background (0.8
photons/pixel) seen in dim frames. A drift-corrected ground truth was
estimated by cubic spline interpolation between the mean positions
obtained from each block of 10 consecutive bright images. Since
the intensity differs by about a factor 10 between bright and dim
frames, the RMS errors of the ground truth should be approximately
10-fold lower than that of a single dim spot. Fig.~\ref{fig:realData}c
shows the resulting error-uncertainty comparison, with every point
corresponding to a single bead. It reproduces the behavior on
simulated images in Fig.~\ref{fig:simErr}b, thus confirming our
conclusion that the Laplace approximation is preferable to CRLB for
uncertainty estimators, and that the good performance of our proposed
PSF width prior is not limited to that particular simulated data set.
\begin{figure}[t]
\centering
\includegraphics[width=80mm]{figures/fig4_Destimation.eps}
\caption{\label{fig:Dest} Estimated diffusion constants
vs.~signal-to-noise ratio $SNR=\sqrt{D\Delta
t/\mean{\varepsilon^2}}$ \cite{vestergaard2014} in simulated
10-step trajectories, mean value and 1\% quantiles.}
\end{figure}
\subsection*{Diffusion constants}
Next, we consider how estimated localization uncertainties leads to
better estimates of diffusion constants, arguably the most common
analysis of single particle tracking data. We divided the synthetic
data shown in Fig.~\ref{fig:frameExamples} into trajectories of length
10, estimated positions and uncertainties with the MAP and Laplace
estimators described above, and estimated diffusion constants using
the covariance-based estimators of Ref.~\cite{vestergaard2014} with
and without the use of uncertainty estimates. Fig.~\ref{fig:Dest}
shows the mean value and 1\% quantiles of the two estimators applied
to every imaging condition, plotted against the signal-to-noise ratio.
As expected, the use of estimated uncertainties improves the
variability of the diffusion estimates substantially. The
covariance-based estimators only use the average uncertainty in each
trajectory (see Methods). We also implemented a maximum likelihood
estimator for the diffusion constant \cite{relich2016} which makes
explicit use of the point-wise uncertainties (see SI text
\ref{SI:sec:diffusionMLE}), but found no further improvement.
\begin{figure}
\includegraphics[width=80mm]{figures/fig5_HMM.eps}
\caption{\label{fig:HMM} Detecting binding events in and out of focus.
(a) Input $z$-coordinates and RMS errors, and (b) representative
spot images along simulated trajectories, from time points indicated
by + in (a). Gray areas indicate binding events. (d) Average state
occupancy for the different analysis algorithms ( ``HMM n.b.'' =
no-blur HMM). (e) Distribution of estimated diffusion constants for
the two states, with true values indicated by dashed lines. (f)
Corrected diffusion constant estimates. (g) Spot images and HMM
occupancies for trajectories simulated without defocus effects.}
\end{figure}
\subsection*{Analysis of multi-state data}
We now turn to a more challenging problem where point-wise errors do
matter: data where both the diffusion constant and localization error
change significantly on similar time scales. In single particle
tracking, changes in diffusion constant can be used as a non-invasive
reporter on intracellular binding and unbinding events
\cite{persson2013}. However, diffusive motion and localization errors
contribute additively to the observed step length statistics (see
Eq.~\ref{MT:eq:dxcov}), and thus changes in diffusion constants and
localization errors cannot be reliable distinguished.
As an example, we consider a protein that alternates between free
diffusion ($D=1$ \si{\square\micro\metre\per\second}) and a bound
state simulated by slow diffusion ($D=0.1$
\si{\square\micro\metre\per\second}). We focus on a single trajectory
with 4 binding/unbinding events, two of which occur about 400 nm out
of focus, and thus are accompanied by substantial broadening of the
PSF and accompanying increases in localization errors. This defocus
matches roughly the radius of an \textit{E.~coli} cell, and the
scenario could model tracking experiments with cytoplasmic proteins
that can bind to the inner cell membrane.
Using SMeagol \cite{linden2016}, we simulated 12 000 replicas of the
above set of events, at a camera frame rate of 200 \si{\hertz},
continuous illumination, and 300 photons/spot on
average. Fig.~\ref{fig:HMM}a shows the $z$ coordinates in the input
trajectory, and the frame-wise RMS errors produced by the MAP
localization algorithm described above. The input points are sparse
(tens of \si{\milli\second} apart), which allows SMeagol to produce
simulated movies that contain the same binding events and defocus
trends, but vary in the detailed diffusion trajectory as well as in
noise realizations. Examples of simulated spots along a trajectory are
shown in Fig.~\ref{fig:HMM}b.
To analyze this challenging data set, we extended the maximum
likelihood diffusion constant estimator with explicit point-wise
errors to include multiple diffusion states governed by an hidden
Markov model (HMM), for which we derived a variational EM algorithm
(see SI text \ref{SI:sec:HMM}). We then analyzed each simulated
trajectory with three different 2-state HMMs: (i) vbSPT, which models
the observed positions as pure diffusion and neglects blur and
localization errors \cite{persson2013}, (ii) the above-mention HMM
that explicitly models these effects, and (iii) the Kalman filter
limit of the HMM in (ii), which models localization errors but not
blur effects. Since multi-state Kalman-type algorithms have been
studied previously \cite{bernstein2016,slator2015,calderon2014}, it is
interesting to compare models with and without blur.
Fig.~\ref{fig:HMM}c shows the inferred average state from the three
different models. As expected, the HMMs that include localization
errors outperform vbSPT at detecting the strongly defocused first and
third binding events. The full HMM does not give the best
classification of the two short binding events, but it does give the
lowest overall misclassification rate, 9\% versus 9.6\% and 17\% for
the Kalman and vbSPT models, respectively, so the apparent worse time
resolution might reflect an overall tendency to invent spurious
transitions by the Kalman and vbSPT models.
Next, we look at estimated diffusion constants. Here, the Kalman and
vbSPT models make systematic errors as seen in the bare parameters in
Fig.~\ref{fig:HMM}d. However, by comparing the step length statistics
between the full and simplified models, one can derive heuristic
correction factors for these models (see Methods, \Eq{eq:Dcorr}). As
shown in Fig.~\ref{fig:HMM}e, this reduces the bias substantially,
especially for the high diffusion constant.
To finally compare the different HMMs on more well-behaved data, we
reran the same experiment but with all $z$ coordinates rescaled by a
factor 1/5 in the PSF model, which essentially removes the
$z$-dependent defocus effects. On this less challenging data set,
event detection is much improved and the differences between the three
HMMs are much less pronounced. However, with misclassification rates
of 4.8\%, 6.6\%, and 7.6\% for the HMM, Kalman model, and vbSPT,
respectively, the full HMM still does overall better.
\subsection*{Position refinement}
Since the new HMM includes the true trajectory as a hidden variable
and performs a global analysis, it can be used to refine individual
localized positions, and in principle beat the Cramer-Rao lower bound
for single image localizations. Fig.~\ref{fig:pos}a shows the true,
measured, and refined positions for part of a two-state trajectory,
and Fig.~\ref{fig:pos}b the relative change of the RMS error for each
frame in Fig.~\ref{fig:HMM}a after refinement. The refinement improves
the RMS errors with up to 50\%. Large localization errors and small
diffusion constant leads to larger relative improvement, which is
intuitive since those factors both make a single point less
informative relative its neighbors. A few points show a small error
increase. On closer inspection, these turn out to be situated near
hard-to-detect state-changes, and therefore tend to be refined using an
incorrect diffusion constant.
\begin{figure}
\includegraphics[width=80mm]{figures/fig6_refinement.eps}
\caption{\label{fig:pos}
Improved localization accuracy by modeling particle motion. (a) True,
measured, and HMM-refined positions from part of a 2-state
trajectory. (b) Relative change of RMS localization error after
HMM-based refinement, for every frame in Fig.~\ref{fig:HMM}a.}
\end{figure}
\section*{Discussion}
Fluorophore positions are not the only useful kind of information in
super-resolution microscopy images. Here, we have shown that
point-wise position uncertainty can also be extracted and used to
improve quantitative data analysis. This is particularly important
for live cell data where dynamic phenomena can be studied, and one may
expect more heterogeneous imaging conditions where many of the
theoretical estimates of localization errors, that may apply for cells
with immobilized internal structure (``fixed''), will have little
relevance.
In general, our results indicate that estimation of position
uncertainty is more sensitive to the fit model than estimation of
position. For parctical use, we find that an estimate based on the
Laplace approximation combined with external information about the
fluorescent background and PSF shape performs well in a wide range of
experimentally relevant conditions.
An intuitive reason why the CRLB performs worse may be that it makes
explicit use of the fit model twice, for both fitting and predicting
the model-dependent uncertainty, while the Laplace approximation only
requires the first step.
Theoretically, the posterior density formalism of the Laplace
approximation also fits more naturally with model-based time-series
analysis where the particle trajectory is treated as a latent variable
to be integrated out.
While we have focused on 2D localization using conventional optics,
the extension to e.g., three dimensions using dual plane imaging
\cite{liu2013} or engineered PSFs \cite{pavani2009} present no
principal difficulties, as long as appropriate PSF models for
localization can be formulated. Note however that even a perfect
characterization of an experimental PSF \cite{liu2013} does not
constitute a perfect localization model, since it does not describe
random PSF shape fluctuations due to motion blur
\cite{deschout2012}. Current techniques for 3D localization are
inherently asymmetric and yield different in-plane and axial accuracy
\cite{rieger2014}, which further underscores the need for downstream
analysis methods to incorporate heterogeneous localization
uncertainty.
Most super-resolution microscopy applications are however not aimed at
particle tracking, but imaging. For PALM/STORM type imaging of fixed
samples, the ability to estimate the uncertainty of individual spots
does not improve the localizations themselves, but it may still
improve the final resolution since uncertain points can be omitted on
a quantitative basis. However, the consequences for live cell imaging
are more interesting, since the same fluorophore may be detected in
different positions over different frames if the target is moving. For
this case, we have shown that the combination of estimating
uncertainty and modeling the fluorophore motion can produce refined
position estimates, in principle pushing the localization errors below
the single-spot Cramer Rao lower bound, by merging information from
consecutive frames in an optimal way.
\nocite{chow2009,ghahramani1997,bishop2006,rabiner1986,bronson2009,johnson2014,burnham}
\subsection*{Synthetic data}
We generated synthetic microscopy data using SMeagol, a software for
accurate simulations of dynamic fluorescence microscopy at the single
molecule level \cite{linden2016}. We modeled the optics using a
Gibson-Lanni PSF model with 584 nm wavelength and NA=1.4
\cite{kirshner2013}, and an EMCCD camera with the high gain
approximation \cite{chao2013,mortensen2010} plus Gaussian readout
noise.
For localization and diffusion estimation tests, we simulated simple
diffusion ($D=1$\si{\square\micro\metre\per\second}) in a cylinder of
length 14.4 \si{\micro\metre} and diameter 2 \si{\micro\metre},
similar to long \textit{E.~coli} cells, in order to avoid confinement
artifacts in the longitudinal direction. We generated 48 data sets
spanning a wide range of imaging conditions by varying frame duration
(8 \si{\milli\second} or 10 \si{\milli\second}, exposure time 2
\si{\milli\second}, 4 \si{\milli\second}, 6 \si{\milli\second}, and 3
\si{\milli\second}, 5 \si{\milli\second}, 8 \si{\milli\second}
respectively, background intensity (1 or 2 photons per pixel), average
spot brightness (75-300 photons/spot), EM gain 20 or 30, and readout
noise level 4 or 8. Each data set contained $10^4$ individual spots.
For the simulated multi-state data, we hand-modified a single SMeagol
input trajectory from a simulated 2-state model to contain 4 binding
events with different durations and z-coordinates as seen
Fig.~\ref{fig:HMM}a, and also thinned out the input trajectory to
create more variability in the particle paths between different
realizations. We then simulated many realizations from this input
trajectory, using the same PSF model as above, continuous illumination
with a sample time of 5 \si{\milli\second}, EMCCD gain 90, an average
spot intensity of 300 photons/spot, and a time-dependent background
that decays exponentially from 0.95 to 0.75 background photons per
pixel with a time-constant of 0.75 s.
\subsection*{Real data}
For estimating localization errors in the real imaging conditions, we
use immobilized fluorescent beads with the diameter of 0.1
\si{\micro\metre} (TetraSpeck Fluorescent Microspheres, ThermoFischer
T7284). The beads where diluted in ethanol and then placed on a
coverslip where we let them dry in before adding water as a mounting
medium.
Imaging was done with a Nikon Ti-E microscope which was configured for
EPI-illumination with a 514 \si{\nano\metre} excitation laser
(Coherent Genesis MX STM) together with matching filters (Semrock
dichroic mirror Di02-R514 with emission filter Chroma HQ545/50M-2P
70351). Intensity modulation was made possible by an acousto-optic
tunable filter (AOTF) (AA Opto Electronics, AOTFnC) which was
triggered by a waveform generator (Tektronix, AFG3021B). The waveform
used was a sequence of square pulses, high for 200 \si{\milli\second}
and low for 1800 \si{\milli\second}. The two illumination intensities,
high and low, corresponds to 10.7 \si{\kilo\watt\per\square\centi\metre}
and 0.63 \si{\kilo\watt\per\square\centi\metre},
respectively.
Fluorescent beads where viewed through a 100x (CFI Apo TIRF 100x oil,
Na=1.49) objective with a 2X (Diagnostic instruments DD20NLT)
extension in front an Andor Ultra 897 EMCCD camera (Andor Technology
Ltd.).
This configuration puts the pixel size to 80 \si{\nano\metre} which is
the same pixel size set in the simulated data. The data set
constituted of 1000 frames (Fig. \ref{fig:realData}a) with an exposure
time of 30 \si{\milli\second}. EMCCD noise characteristics (gain,
offset, readout noise) were determined by analyzing a “dark” movie
obtained with the shutter closed.
\subsection*{Localization}
We perform maximum likelihood (MLE) localization using a high-gain
EMCCD noise model \cite{chao2013,mortensen2010}, which relates the
probability $q(c_i|E_i)$ of the offset-subtracted pixel count $c_i$
for a given pixel intensity $E_i$ (expected number of photons/frame)
in pixel $i$. For the intensity $E(x,y)$, we model the PSF with a
symmetric Gaussian,
\begin{equation}\label{eq:gaussPSF}
E(x,y)=\frac{b}{a^2}+\frac{N}{\sqrt{2\pi\sigma^2}}
\exp\Big(-\frac{(x-\mu_x)^2+(y-\mu_y)^2}{2\sigma^2}\Big),
\end{equation}
with pixel size $a$, background $b$ (expected number of
photons/pixel), spot width $\sigma$, amplitude $N$ (expected number of
photons/spot), and spot position $(\mu_x,\mu_y)$, and approximate the
pixel intensity
\begin{equation}
E_i=\int_\text{px. $i$} E(x,y)dxdy
\end{equation}
by numerical quadrature \cite{chao2015}. The log likelihood of an
image containing a single spot is then given by
\begin{equation}
\ln L(\theta) = \ln q_0(\theta)+\sum_{i\in ROI} \ln q(c_i|E_i(\theta)),
\end{equation}
where $\theta=(\mu_x,\mu_y,b,N,\sigma)$ are fit parameters, and $q_0$
is a prior distribution (we set $\ln q_0=0$ for MLE fitting). To
avoid the complications of spot identification, we use known positions
to determine the ROI and initial guess for $(\mu_x,\mu_y)$. All
localizations are performed using a $9\times 9$ ROI. Fits that failed
to converge or estimated uncertainties larger than 2 pixels where
discarded.
\subsection*{Cramer-Rao lower bound}
The CRLB is a lower bound on the variance of an unbiased estimator
\cite{cramer1945,rao1945}.
We use an accurate approximation to the CRLB for a symmetric Gaussian
PSF from Ref.~\cite{rieger2014},
\begin{equation}\label{eq:CRLB}
\varepsilon^2_\text{CRLB}=2\frac{\sigma_a^2}{N}\left(
1+4\tau+\sqrt{\frac{2\tau}{1+4\tau}}
\right),
\end{equation}
with $\tau=2\pi\sigma_a^2 b/(Na^2)$, $\sigma_a^2=\sigma^2+a^2/12$, and
the prefactor 2 accounts for EMCCD excess noise \cite{mortensen2010}.
\subsection*{Laplace approximation}
An alternative way to approximate the uncertainty of the fit
parameters is to Taylor expand the likelihood around the
maximum-likelihood parameters $\theta^*$ to second order, that is
\begin{equation}\label{eq:laplace}
\ln L(\theta)\approx \ln L(\theta^*)+
\frac{\partial \ln L}{\partial\theta}\Big|_{\theta^*}(\theta-\theta^*)
+\frac12(\theta-\theta^*)^T
\frac{\partial^2\ln L}{\partial \theta^2}\Big|_{\theta^*}
(\theta-\theta^*).
\end{equation}
The first-order term is zero, since $\theta^*$ is a local maximum.
This approximates the likelihood by a Gaussian with covariance matrix
given by the inverse Hessian, i.e., $\Sigma=[\partial^2\ln L/\partial
\theta^2]^{-1}$. In a Bayesian setting, this expresses the
(approximate) posterior uncertainty about the fit parameters. The
estimated uncertainties (posterior variances) are given by the
diagonal entries of the covariance matrix, e.g.,
$\varepsilon^2_\text{Lap.}(\mu_x)=\Sigma_{\mu_x,\mu_x}$. We compute
the Hessian numerically using Matlab's built-in optimization routines,
and use the log of the scale parameters $b,N,\sigma$ for fitting,
since they are likely to have a more Gaussian-like posterior
\cite{mackay1998}.
\subsection*{Prior distributions}
The prior distributions for localization used in the main text are:
normal priors with mean value $\ln b_\text{true}$ and std.~0.2 for the
log background intensity, resulting in the log-normal prior $b\in\ln
N(\ln b_\text{true},0.2^2)$. For the PSF width $\sigma$, we use a
skewed normal prior for $\ln \sigma$ to penalize fits with $\sigma$
below the minimum in-focus width of the PSF. The skew normal density
is given by \cite{azzalini1985}
\begin{equation}
\label{eq:skewed_prior}
f(x,x_0,w,\alpha) =
\frac{2}{w}
\phi\left(\frac{x-x_0}{w}\right)\Phi\left(\alpha\frac{x-x_0}{w}\right),
\end{equation}
where $\phi$ is the probability distribution function and $\Phi$ is
the cumulative distribution function for the unit normal distribution
$N(0,1)$. We used $x_0=\log(1.5)$, $w=1$, and $\alpha = 5$, as
sketched in Fig.~\ref{fig:simErr}b, with the pixel width $a=80$
\si{\nano\metre} as the length unit. These priors are shown in SI
Fig.~\ref{SI:fig:parPrior}.
\subsection*{Covariance-based diffusion estimator}
If $x_k$ ($k=0,1,\ldots$) is the measured trajectory of a freely
diffusing particle with diffusion constant $D$, the widely used model
for camera-based tracking by Berglund \cite{berglund2010} predicts
that the measured step lengths $\Delta_k=x_{k+1}-x_k$ are zero-mean
Gaussian variables with covariances given by
\begin{equation}\label{eq:dxcov}
\mean{\Delta_k^2}=2D\Delta t(1-2R)+2\varepsilon^2,\quad
\mean{\Delta_k\Delta_{k\pm1}}=2D\Delta tR-\varepsilon^2,
\end{equation}
and uncorrelated otherwise. Here, $0\le R\le 1/4$ is a blur
coefficient that depends on how the images are acquired (e.g., $R=1/6$
for continuous illumination), $\Delta t$ is the measurement time-step,
and $\varepsilon^2$ is the variance of the localization errors.
Substituting sample averages for $\mean{\Delta_k^2}$ and
$\mean{\Delta_k\Delta_{k+1}}$ and solving for $D$ yields a
covariance-based estimator (CVE) with good performance
\cite{vestergaard2014}. If $\varepsilon^2$ is known or can be
estimated independently, the first relation in \Eq{eq:dxcov} alone
yields a further improved estimate of $D$. As we argue in
Sec.~\ref{SI:sec:cve}, these estimators apply also for variable
localization errors if $\varepsilon^2$ is replaced by the average
$\mean{\varepsilon^2}$.
\subsection*{Maximum likelihood and multi-state diffusion}
The Berglund model \cite{berglund2010} can also be used directly for
maximum likelihood inference, which has the potential advantage that
point-wise errors can be modeled \cite{relich2016}. The basic
assumption is to model the observed positions $x_k$ as averages of the
true diffusive particle path $y(t)$ during the camera exposure, plus a
Gaussian localization error, i.e.,
\begin{equation}\label{eq:blurdef}
x_k=\int_{0}^{\dt}y(k\dt+t)f(t)dt+\varepsilon_k\xi_k,
\end{equation}
where $f(t)$ is the normalized shutter function \cite{berglund2010},
$\varepsilon_k$ is the localization uncertainty (standard deviation)
at time $k$, and $\xi_k$ are independent Gaussians random numbers with
unit variance. Continuous illumination is described by a constant
shutter function, $f(t)=1/\dt$. The limit where blur effects are
neglected can be described by setting $f(t)$ to a delta function,
which corresponds to instantaneous position measurement. This reduces
\Eq{eq:blurdef} to a standard Kalman filter \cite{calderon2016}, and
leads to $R=0$ in \Eq{eq:dxcov}.
In SI text \ref{SI:sec:diffusionMLE}, we derive a maximum likelihood
estimator that learns both $D$ and $y(t)$. In SI
Sec.~\ref{SI:sec:HMM}, we extend the model to multi-state diffusion,
by letting the diffusion constant switch randomly between different
values corresponding to different hidden states in an HMM, and derive
a variational EM algorithm for maximum likelihood inference of model
parameters, hidden states, and refined estimates of the measured
positions.
To interpret estimated diffusion constants from simplified models, one
may ``derive'' corrected diffusion estimates $D^*$ by equating
expressions for the step length variance $\mean{\Delta_k^2}$ from
\Eq{eq:dxcov} with and without those effects present. For the Kalman
($R=0$) and vbSPT ($R=\varepsilon=0$) models, we get
\begin{equation}\label{eq:Dcorr}
D^*=\frac{D_\text{Kalman}}{1-2R},\quad \text{and }
D^*=\frac{D_\text{vbSPT}-\mean{\varepsilon^2}/\dt}{1-2R},
\end{equation}
respectively, which is what we use in Fig.~\ref{fig:HMM}e.
\subsection*{Software}
Matlab open source code for the EM and localization algorithms are
available at
\href{https://github.com/bmelinden/uncertainSPT}{github.com/bmelinden/uncertainSPT}.
\section*{Methods}
\input{methods.tex}
\subsection*{Acknowledgements}
\input{acknowledgements.tex}
\subsection*{Author contributions}
\input{contributions.tex}
\input{osak_arxiv.bbl}
\clearpage
\onecolumngrid
\section{Localization error estimates with different priors}\label{sec:locpriors}
In this section we present how different prior functions assigned to
the PSF width $\sigma$ and the background $b$ influence the accuracy
of the computed RMS errors and RMS uncertainties for the synthetic
data set considered in this paper.
We considered log-normal priors for the background intensity, either
centered around the true (simulated) background in each data set, or
around the average background intensity in all data sets. For the PSF
width we tested two types of priors: (i) a log-normal prior around the
expected value for the PSF width; (ii) a skew-normal prior on
$\ln(\sigma)$, which assigns very low probability to PSF widths below
$\approx 0.8$ pixels, as well as low probabilities for very wide PSFs.
We do not impose any prior on the spot amplitude $N$.
The probability density function for the log-normal prior, denoted as
$\ln N(\mu,\sigma^2)$, is defined by
\begin{equation}
\label{eq:logNormal_prior}
f_{ln}(x,\sigma,\mu) =
\frac{1}{x\sqrt{2\pi\sigma^2}}e^{-\frac{(ln x-\mu)^2}{2\sigma^2}},
\end{equation}
where $\mu$ and $\sigma$ are the mean and standard deviation
parameters. The skew-normal distribution \cite{azzalini1985} has a
density function given by
\begin{equation}
f_{sn}(x,x_0,\omega,\alpha) =
\frac{2}{\omega}\phi\left(\frac{x-x_0}{\omega}\right)\Phi\left(\alpha\frac{x-x_0}{\omega}\right),
\end{equation}
where $\phi$ is the probability density function and $\Phi$ is the
cumulative distribution function for the unit normal distribution
$N(0,1)$, and the parameter $\alpha$ determines the degree of
asymmetry; with $\alpha$ positive (negative) increasing the weight
above (below) the mean value parameter $x_0$.
For our investigations on the effects of different prior
distributions, we used two background priors:
\begin{itemize}
\item
Log-normal ``correct'' prior on $b$: $\ln N(\ln b_\text{true},0.2^2)$,
where $b_\text{true}$ is either 1 or 2 photons/pixel, which is the
background level used for simulating that particular data set.
\item
Log-normal ``average'' prior on $b$: $\ln N(\ln 1.5,0.2^2)$, centered
around the average background intensity in all data sets, that is, 1.5
photons/pixel.
\end{itemize}
For the PSF width, we considered the two following priors:
\begin{itemize}
\item
Log-normal prior on $\sigma$: $\ln N(\ln 1.5,0.2^2)$, since the
half-width of the focused PSF width is around 1.5 pixels wide.
\item
Skew-normal prior on $\ln(\sigma)$: $f_{sn}(x,x_0,\omega,\alpha)$, where
$x_0=\ln(1.5)$, $\alpha = 5$ and $\omega=1$.
\end{itemize}
\begin{figure*}
\centering
\includegraphics{figures/figS3_err_vs_uncert_priors.eps}
\caption{\label{fig:priors_SI} Error-uncertainty plots for different
combinations of prior distributions on $b$ and $\sigma$.}
\end{figure*}
Error-uncertainty comparisons for various combinations of the above
priors are shown in Fig.~\ref{fig:priors_SI}. As seen in
Fig.~\ref{fig:priors_SI}a (which reproduces Fig.~\ref{MT:fig:simErr}a
in the main text), imposing a flat prior (no prior) leads to estimated
uncertainties that substantially underestimate the true errors,
especially for points with low photon count. In comparison, all other
prior combinations present an average improvement. Overall, the
Laplace estimator seems more consistent (RMS uncertainty mostly closer
to RMS error), and also less sensitive to the choice of priors. The
latter trend is especially clear in the examples using the ``average''
background prior, which is always either a systematic over- or
underestimate, which results in substantially larger variability or
the CRLB results.
\begin{figure}
\centering
\includegraphics{figures/SI_MLE_MAP_errors.eps}
\caption{\label{fig:MLE_vs_MAPerrors} Localization errors for MLE vs
MAP localizations, using the prior described in the main text and
Fig.~\ref{fig:priors_SI}f.}
\end{figure}
\section{Diffusion with time-varying localization errors}\label{sec:blurmodel}
Here, we give a detailed derivation of the various estimators and
models mentioned in the main text methods section, starting from
\Eq{MT:eq:blurdef}, the assumption that localizing a moving object
amounts to detecting the time-averaged position with some
(independent) localization error, which we will assume to be Gaussian.
Parts of these derivations have been given elsewhere
\cite{berglund2010,relich2016}, but are restated here in a different
form that facilitates a generalization to multi-state models.
\subsection{Diffusive camera-based tracking with blur and localization errors}
To streamline the presentation, we will use units where $\dt=1$, use
the subscript $t=0,1,2,\ldots$ to denote discrete time dependence, and
use the step length and localization variances, $\lambda_t=2D_t\dt$
and $v_t=\varepsilon_t^2$, respectively. We allow both of them to be
time-dependent, but assume them to be statistically independent and
restrict the diffusion constant to be constant throughout each
frame. Then, \Eq{MT:eq:blurdef} reads
\begin{equation}\label{eq:localization}
x_t=\int_0^1f(t') y(t+t')dt'+\sqrt{v_t}\xi_t,
\end{equation}
where $\xi_t$ are independent identically distributed (iid) N(0,1)
variables, and $y(t)$ is the true trajectory of the particle being
localized. The shutter distribution $f(t)$ is a probability density on
$[0,1]$, which describes the image acquisition process (e.g., $f(t)=1$
for continuous acquisition), but neglects stochastic elements such as
fluorophore blinking. It has the distribution function
\begin{equation}
F(t)=\int_0^t f(t')dt'.
\end{equation}
We divide $y(t)$ in two parts, the true positions $y_t$ at the
beginning of each frame, which evolve according to
\begin{equation}\label{eq:ydiffusion}
y_{t+1}=y_t+\sqrt{\lambda_t}\eta_t,
\end{equation}
where $\eta_t$ are again iid N(0,1), and a conditional interpolating
process between them, described by Brownian bridges
\cite{chow2009}. Thus, for $0\le t'\le 1$, we write
\begin{equation}\label{eq:interp}
y(t+t')=y_t + t'(y_{t+1}-y_t)+\sqrt{\lambda_t}B_t(t'),
\end{equation}
where $B_t$ are a set of iid standard Brownian bridges. These are
Gaussian processes on the interval [0,1], defined by
\begin{equation}\label{eq:bridges}
B_t(0)=B_t(1)=0,\quad
\mean{B_t(t')}=0,\quad
\mean{B_t(t')B_t(t'')}=t'(1-t'')\text{, for $t'\le t''$},
\end{equation}
and also independent on different intervals, so that
$\mean{B_r(t')B_v(t'')}=0$ if $r\ne v$. Substituting the interpolation
formula \Eq{eq:interp} in the localization model \Eq{eq:localization},
we get
\begin{equation}\label{eq:interp2}
x_t=y_t(1-\tau)+ y_{t+1}\tau+\sqrt{v_t}\xi_t
+\sqrt{\lambda_t}\int_0^1f(t')B_t(t')dt',
\end{equation}
where we have introduced the shutter average, given by
\begin{equation}
\tau=\int_0^1tf(t)dt.
\end{equation}
Using the properties of Brownian bridges, \Eq{eq:bridges}, one can
show that the last integral in \Eq{eq:interp2} is a Gaussian random
variable with mean zero and variance
\begin{equation}
\beta\equiv\Var\left[\int_0^1f(t')B_t(t')dt'\right]=\tau(1-\tau)-R,
\end{equation}
where $R$ is the blur coefficient of Ref.~\cite{berglund2010}, given
by
\begin{equation}
R=\int_0^1F(t)\big(1-F(t)\big)dt.
\end{equation}
By assumption, $v_t$ and $B_t$ are statistically independent, and thus
one can add up the noise in the measurement model and arrive at
\begin{equation}\label{eq:interp3}
x_t=y_t(1-\tau)+ y_{t+1}\tau+\sqrt{v_t+\beta\lambda_t}\zeta_t,
\end{equation}
where $\zeta_t$ are again iid N(0,1).
\subsection{Constant exposure}
An important class of shutter distribution are those that are constant
during some fraction $t_E$ of the each frame, and then zero, i.e.,
\begin{equation}
f(t)=\left\{\begin{array}{ll}
\frac{1}{t_E},&t\le t_E,\\
0,&t> t_E,\\
\end{array}\right.
\end{equation}
which leads to
\begin{equation}
\tau=\frac{t_E}{2},\quad
R=\frac{t_E}{6},\qua
\beta=\frac14 t_E(\frac43-t_E)
\end{equation}
We see that $R\le \frac16$ and $\tau\le\frac12$, with maxima at
continuous exposure ($t_E=1$).
On the other hand, $\beta$ has a maximum of $\frac19$ at
$t_E=\frac23$, and the value of $\frac{1}{12}$ at continuous exposure
can only be further lowered when $t_E<\frac13$.
It is unclear if this is significant, since with non-constant exposure
there is some freedom in how the shutter distribution is defined, and
one could also place it symmetrically in the interval and get
$\tau=0.5$ for all exposure times. We defer further investigations of
this issue to future work.
\subsection{Covariance relations}\label{sec:cve}
The covariance matrix for the steps $\Delta_t=x_{t+1}-x_t$ can be
found from Eqs.~(\ref{eq:ydiffusion},\ref{eq:interp3}). With some
manipulations, we get
\begin{align}
\mean{\Delta_t^2}= &(1-\tau)\lambda_t+\tau\lambda_{t+1}
-(\lambda_{t+1}+\lambda_t)R+v_t+v_{t+1},\\
\mean{\Delta_t\Delta_{t+1}}= &\lambda_{t+1}R-v_{t+1},\\
\mean{\Delta_t\Delta_{t+t'}}= &0\text{, if $|t'|>1$},
\end{align}
where the expectations $\mean{\cdot}$ are understood to be over the
noise distributions only. If we further assume simple diffusion,
$\lambda_t=2D\dt=const.$, and average over time as well, we recover
covariance relations of the same form as \Eq{MT:eq:dxcov},
\begin{equation}
\mean{\Delta_t^2}=2D\dt(1-2R)+2\mean{v_t},\quad
\mean{\Delta_t\Delta_{t\pm 1}}=2D\dt R-\mean{v_t},
\end{equation}
where the averages are now over time as well, and $\varepsilon^2$ is
identified as the time-average $\mean{v_t}$. Thus, the
covariance-based estimators of Ref.~\cite{vestergaard2014} should apply
also to non-constant localization errors.
\section{Maximum likelihood estimator}\label{sec:diffusionMLE}
For the case of a single trajectory of simple diffusion in one
dimension, the likelihood for the diffusion constant follows from
Eqs.~\ref{eq:ydiffusion} and \ref{eq:interp3},
\begin{eqnarray}
p(x|\lambda) &=& \int dy p(x|y,\lambda)p(y|\lambda),\label{eq:lamlik}\\
p(y|\lambda) &=&\prod_{t=1}^T \big(2\pi\lambda\big)^{-\frac12}
\exp\Big[-\frac{(y_{t+1}-y_t)^2}{2\lambda}\Big],\\
p(x|y,\lambda)&=& \prod_{t=1}^T \big(2\pi(v_t+\beta\lambda)\big)^{-\frac12}
\exp\Big[-\frac{1}{2}(v_t+\beta\lambda)^{-1}
\big(x_t-(1-\tau)y_t-\tau y_{t+1}\big)^2\Big],
\end{eqnarray}
where we neglected to supply a starting density for $y_1$, since the
problem is translation invariant. The integral over the hidden path in
\Eq{eq:lamlik} is a multivariate Gaussian and can be solved exactly in
several ways \cite{relich2016}. Defining
\begin{equation}
\vec{y}=[y_1,\;y_2,\ldots,y_{T+1}]^\dagger,\quad
\vec{x}=[x_1,\;x_2,\ldots,x_{T}]^\dagger,
\end{equation}
where $^\dagger$ denotes matrix transpose, we get
\begin{multline}\label{eq:pxlam1}
p(x|\lambda)=\int d\vec{y} \exp\bigg[
-T\ln(2\pi)-\frac{T}{2}\ln(\lambda)
-\frac12\sum_{t=1}^T\ln(v_t+\beta\lambda)
-\frac12\Big(\vec{y}^\dagger\Lambda\vec{y}-2\vec{y}^\dagger W\vec{x}
+\vec{x}^\dagger V\vec{x}
\Big)\bigg]\\
=\exp\bigg[
-T\ln(2\pi)-\frac{T}{2}\ln(\lambda)
-\frac12\sum_{t=1}^T\ln(v_t+\beta\lambda)\bigg]\\
\times\int d\vec{y}\exp\bigg[
-\frac12\Big(
(\vec{y}-\Lambda^{-1}W\vec{x})^\dagger\Lambda(\vec{y}-\Lambda^{-1}W\vec{x})
+\vec{x}^\dagger (V-W^\dagger\Lambda^{-1} W)\vec{x}
\Big)\bigg],
\end{multline}
where $\Lambda,W,V$ are matrices whose \vc{elements} are found by
comparing terms. This is a multivariate Gaussian in $\vec{y}$, with
mean value $\mu=\Lambda^{-1}W\vec{x}$ and covariance matrix
$\Sigma=\Lambda^{-1}$, and the marginalized likelihood is therefore
given by
\begin{equation}\label{eq:pxlam2}
p(x|\lambda)=
\exp\bigg[
-\frac{T-1}{2}\ln(2\pi)-\frac{T}{2}\ln(\lambda)
-\frac12\sum_{t=1}^T\ln(v_t+\beta\lambda)
-\frac12\vec{x}^\dagger (V-W^\dagger\Lambda^{-1}W)\vec{x}
-\frac12\ln|\Lambda|\bigg].
\end{equation}
By comparing terms, we see that $\Lambda$ is symmetric tridiagonal and
positive definite, $V$ is diagonal, and $W$ has only non-zero elements
on the diagonal and first upper diagonal, and so it is possible to
compute the above matrix expression in linear time. We minimize $\ln
p(x|\lambda)$ by standard optimization routines in Matlab.
Repeating the analysis of Fig.~\ref{MT:fig:Dest} in the main text for
this MLE estimator, see Fig.~\ref{fig:DestMLE}, we find almost no
improvement over the CVE that uses average estimated
uncertainties. However, the fact that the MLE estimator is limited to
return positive estimated diffusion constants, and handles missing
data points (by assigning them infinite or very large uncertainties),
may be useful in applications.
\begin{figure}
\centering
\includegraphics{figures/SI_MAP_CVE_MLE_with_uncertainty.eps}
\caption{\label{fig:DestMLE} Mean value and 1\% quantiles of estimated
diffusion constants using the CVE and MLE diffusion constant
estimator, both using estimated positions and uncertainties.}
\end{figure}
\section{Variational EM algorithm for diffusive HMM}\label{sec:HMM}
Here, we extend the above diffusion model to include multiple
diffusion constants. We start by writing down a diffusive HMM, which
includes both the hidden path of the above diffusion model, but also a
set of hidden states with different diffusion constants, that evolve
as a discrete Markov process. Similar models (that however did not
include explicit motion blur effects), have previously been solved by
stochastic EM algorithms \cite{ashley2015,bernstein2016}. Here, we
instead describe a deterministic variational approach inspired by
algorithms for factorial HMMs \cite{ghahramani1997}.
In the rest of this section, we proceed as follows: We start by
specify the diffusive HMM model for a single 1-dimensional
trajectory. We then outline the variational EM approach, derive high
level update equations, and describe the procedure for re-estimating
localized positions. However, do not give a detailed derivation of all
steps in the algorithm, as large parts of it closely resembles
previously published derivations of variational algorithms for HMMs
\cite{persson2013,johnson2014}.
\subsection{Model}
In addition to the measured ($\vec{x}$) and true ($\vec{y}$)
positions, we include a hidden state trajectory
$\vec{s}=[s_1,s_2,\ldots,s_T]$, such that $s_t$ determines the
diffusion constant on the interval $[t,t+1]$. The hidden states are
numbered from 1 to $N$, and evolve according to a Markov process with
transition matrix $A$ and initial state probability $\vpp$. For a
single 1-dimensional trajectory, this leads to a complete data
likelihood of the form
\begin{equation}
p(\x,\y,\s|\lambda,A,\vpp)= p(\x|\y,\s,\lambda)p(\y|\s,\lambda)p(\s|A,\vpp),
\end{equation}
with factors
\begin{eqnarray}
p(\s|A,\vpp)&=&
\prod_{m=1}^N\pp_m^{\delta_{m,s_1}}
\prod_{t=2}^T\prod_{i,j=1}^N A_{ij}^{\delta_{is_t}\delta_{js_{t+1}}},\\
p(\y|\s,\lambda)&=&
\prod_{t=1}^T\prod_{j=1}^N (2\pi \lambda_j)^{-\frac12\delta_{js_t}}
\exp\left[-\delta_{js_t}\frac{(y_{t+1}-y_t)^2}{2\lambda_j}\right],\\
p(\x|\y,\s,\lambda)&=&
\prod_{t=1}^T\prod_{j=1}^N \big(2\pi(v_t+\beta\lambda_j)\big)^{-\frac12\delta_{js_t}}
\exp\left[-\delta_{js_t}
\frac{(x_t-(t-\tau)y_t-\tau y_{t+1})^2}{2(v_t+\beta\lambda_j)}
\right].
\end{eqnarray}
\subsection{Variational EM approach}
We would like to perform maximum-likelihood inference of the model
parameters, which means maximizing the likelihood with latent
variables $\s,\y$ integrated out,
\begin{equation}
L(A,\vpp,\lambda)=\int d\y\sum_{\s}
p(\x|\y,\s,\lambda)p(\y|\s,\lambda)p(\s|A,\vpp).
\end{equation}
Since this problem is intractable, we make a variational
approximation, meaning we approximate $\ln L$ with a lower bound
\begin{equation}\label{eq:lowerbound}
\ln L =\ln \int d\y\sum_{\s}
p(\x|\y,\s,\lambda)p(\y|\s,\lambda)p(\s|A,\vpp)
\ge \int d\y\sum_{\s} q(\s)q(\y)\ln
\frac{p(\x|\y,\s,\lambda)p(\y|\s,\lambda)p(\s|A,\vpp)}{q(\s)q(\y)}\equiv F,
\end{equation}
where the inequality follows from Jensen's inequality. Here, $q(\s)$,
$q(\y)$ are arbitrary variational distributions that need to be
optimized together with the model parameters to achieve the tightest
lower bound, and can be used for approximate inference about the
latent variables. In particular, it turns out that the optimal
variational distributions approximate the posterior distribution of
$\y,\s$ in the sense of minimizing a Kullback-Leibler divergence
\cite{bishop2006}. We use gradient descent for the optimization,
i.e., we iteratively optimize each variational distribution and the
parameters with the others fixed, which leads to an EM-type algorithm.
To optimize $F$ w.r.t.~the variational distributions, we set the
functional derivatives of $F$ to zero and use Lagrange multipliers to
enforce normalization. After some work, one arrives at
\begin{eqnarray}
\ln q(\s)&=&-\ln Z_s
+\mean{\ln p(\x|\y,\s,\lambda}_{q(\y)}
+\mean{\ln p(\y|\s,\lambda}_{q(\y)}
+\ln p(\s|A,\vpp),\label{eq:lnqs}\\
\ln q(\y)&=&-\ln Z_y
+\mean{\ln p(\x|\y,\s,\lambda}_{q(\s)}
+\mean{\ln p(\y|\s,\lambda}_{q(\s)}
+\underbrace{\mean{\ln p(\s|A,\vpp)}_{q(\s)}}_\text{independent of $\y$},
\end{eqnarray}
where $Z_{s,y}$ are normalization constants originating from the
Lagrange multipliers, and $\mean{\cdot}_{f}$ denotes an expectation
value computed with respect to the distribution $f$. As it turns out,
these equations are individually tractable, and results in $q(\y)$
being a multivariate Gaussian, and $q(\s)$ adopting the standard HMM
form amenable to efficient forward-backward iterations. For the
initial state and transition probability parameters, the only
dependence in $F$ comes from $p(\s|A,\lambda)$, and arrives at the
classical Baum-Welch reestimation formulas \cite{rabiner1986}.
However, optimizing $F$ w.r.t.~step length variances does not lead to
a tractable update equation, and we are instead forced to optimize the
$\lambda_j$-dependent parts of $F$ numerically, i.e.,
\begin{equation}
\lambda_j=\mathrm{argmax}_{\lambda_j}
\mean{\ln p(\x|\y\s,\lambda)}_{q(\y)q(\s)}
\mean{\ln p(\y|\s,\lambda)}_{q(\y)q(\s)},
\end{equation}
\subsection{The lower bound}
The lower bound can be computed as well, and gets a
particularly simple form just after the update of
$q(\s)$. Substituting the update equation \Eq{eq:lnqs} into the
expression for $F$, \Eq{eq:lowerbound}, we get
\begin{multline}
F=\int d\y\sum_{\s} q(\s)q(\y)\ln\Big[
\ln p(\x|\y,\s,\lambda)+\ln p(\y|\s,\lambda)+\ln p(\s,A,\vpp)\\
+\ln Z_s
-\mean{\ln p(\x|\y,\s,\lambda)+\ln p(\y|\s,\lambda)+\ln p(\s,A,\vpp)}_{q(\y)}
-\ln q(\y)\Big]
=\ln Z_s-\mean{\ln q(\y)}_{q(\y)}.
\end{multline}
Here, $\ln Z_s$ is the normalization constant of $q(\s)$ that can be
computed as part of the forward-backward iteration, and since (as we
noted above) $q(\y)$ is a multivariate Gaussian with dimension $T+1$,
we get
\begin{equation}
-\mean{\ln q(\y)}_{q(\y)}=
\frac{d}{2}(T+1)\big(1+\ln(2\pi)\big)+\frac12\ln|\Sigma|,
\end{equation}
where $\Sigma$ is the covariance matrix of $q(\y)$. The inverse of
$\Sigma$ is analogous to the matrix $\Lambda$ appearing in the
single-state diffusion estimator,
Eqs.~(\ref{eq:pxlam1},\ref{eq:pxlam2}), and in particular
$\Sigma^{-1}$ is also symmetric, tridiagonal, and positive definite,
and thus the determinant $\ln|\Sigma|=-\ln|\Sigma^{-1}|$ can be
robustly computed in linear time. Since the EM algorithm approximates
the parameter likelihood, the lower bound cannot be used for model
selection as in the case of variational maximum evidence calculations
\cite{bronson2009,persson2013,johnson2014}. However, it is still
useful for numerical convergence control, and possibly for model
selection together with some complexity penalty such as the Bayesian
or Akaike information criterion \cite{burnham}.
\subsection{Refined positions}\label{sec:positions}
An additional use for the HMM analysis is to use it to refine the
localizations. Since the HMM pools information about many spots, it is
in principle possible to beat the Cram\'er-Rao lower bound for single
image localizations in this way. To set up the refinment problem, we
refer back to \Eq{SI:eq:interp2} and the different contributions to
the observed position $x_t=z_t+\sqrt{v_t}\xi_t$, where
\begin{equation}\label{eq:zrefine1}
z_t=y_t(1-\tau)+ y_{t+1}\tau+\sqrt{\lambda_{s_t}}\int_0^1f(t')B_t(t')dt'
\end{equation}
is the motion-averaged position that the localization algorithm tries
to estimate (according to this model). To refine the localization, we
need to compute the posterior density of $z_t$, i.e.,
$p(z_t|\x,\theta)$.
We start by recalling that the Brownian bridge integral in
\Eq{SI:eq:zrefine1} is Gaussian with mean zero and variance
$\beta\lambda_{s_t}$, and using the compact notation
$\overline{y_t}=(1-\tau)y_t+\tau y_{t+1}$ and $\theta$ to denote all
the model parameters, we therefore have
\begin{equation}
p(z_t|\y,\s,\theta)=p(z_t|\overline y_t,s_t,\theta)
=N(\overline y_t,\beta\lambda_{s_t}),
\end{equation}
where $N(a,b)$ denotes a Gaussian density with mean $a$ and variance
$b$. Furthermore, the localization uncertainty is also assumed
Gaussian and independent of the underlying kinetics:
\begin{equation}
p(x_t|\z,\y,\s,\theta)=p(x_t|z_t)=N(z_t,v_t).
\end{equation}
Applying Bayes theorem to these relations, we get
\begin{equation}
p(z_t|x_t,\y,\s,\theta)
=
\frac{p(x_t|z_t)p(z_t|\y,\s,\theta)}{p(x_t|\y,\s,\theta))}
=\frac{N(z_t,v_t)N(\overline y_t,\beta\lambda_{s_t})}{
N(\overline y_t,\beta\lambda_{s_t}+v_t)}
=
N\Big(\frac{v_t\overline y_t+\beta\lambda_{s_t}x_t}{v_t+\beta\lambda_{s_t}},
\frac{\beta\lambda_{s_t}v_t}{v_t+\beta\lambda_{s_t}}
\Big).
\end{equation}
The predictive distribution is finally given by marginalizing
$p(z_t|x_t,\y,\s,\theta)$ over the posterior for $\y,\s$. Using the
variational distribution, this means
\begin{equation}
p(z_t|\x,\theta)
\approx\mean{
N\Big(\frac{v_t\overline y_t+\beta\lambda_{s_t}x_t}{v_t+\beta\lambda_{s_t}},
\frac{\beta\lambda_{s_t}v_t}{v_t+\beta\lambda_{s_t}}\Big)}_{q(\y)q(\s)}.
\end{equation}
In particular, the posterior mean of $z_t$ is then given by
\begin{equation}
\mean{z_t|\x,\theta}\approx\mean{
\frac{v_t\overline y_t+\beta\lambda_{s_t}x_t}{v_t+\beta\lambda_{s_t}}
}_{q(\y)q(\s)}
=\mean{
\frac{(1-\tau)\mu_t+\tau\mu_{t+1}
+\frac{\beta\lambda_{s_t}}{v_t}x_t}{1+\frac{\beta\lambda_{s_t}}{v_t}}
}_{q(\s)},
\end{equation}
which we will use as our estimator for refining the localizations.
Here, $\mu_t=\mean{y_t}_{q(\y)}$ is the variational mean value. The
variational average over hidden states is done numerically.
| f1d034b381b3dc8e3695c5d4d2e1eded7f2fbf5f | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{introduction}
At the outset, we fix the probability space $(\Omega,\Sigma,\mathbf{P})$ and
define the random vector (r.v.) $\mathcal{X}\ni\mathbf{X}=(X_1,\ldots,X_n)'$ as a map from
$(\Omega,\Sigma)$ to the $n(\in\mathbf{N})$-dimensional Borel space
$(\mathbf{R}_+^n,\mathcal{B}(\mathbf{R}_+^n))$. The cumulative distribution
function (c.d.f.) of $\mathbf{X}$ is in the sequel denoted by
$F_{1,\ldots,n}(x_1,\ldots,x_n):=\mathbf{P}[X_1\leq x_1,\ldots,X_n\leq x_n]$, and
the corresponding probability density function (p.d.f.) by
$f_{1,\ldots,n}(x_1,\ldots,x_n):=\partial^n/(\partial x_1\cdots \partial x_n)
F_{1,\ldots,n}(x_1,\ldots,x_n)$, where $(x_1,\ldots,x_n)'\in\mathbf{R}_{+}^n:=(0,\ \infty)^n$.
Finally, $F_i(x)$ and $f_i(x)$ denote, respectively,
the marginal c.d.f. and p.d.f. of $X_i,\ i=1,\ldots,n$.
Clearly, when the coordinates of $\mathbf{X}$ are stochastically independent, then
there is only one way to formulate the c.d.f. $F_{1,\ldots,n}$, whereas the shapes
of the just-mentioned c.d.f. are infinite otherwise.
We further discuss the so-called multivariate reduction approach to creating random
vectors with dependent coordinates. This paves the way to introducing the main object of
our interest in Section \ref{sec2}.
For applications of the multivariate reduction method in insurance, we refer to, e.g.,
Vernic (1997, 2000), Pfeifer and Ne\v slehov\' a (2004), Furman and Landsman (2005, 2010),
Boucher et al. (2008) and Tsanakas (2008), as well as to the references therein.
Let $\mathcal{Y}\ni\mathbf{Y}=(Y_1,\ldots,Y_{n+1})'$ be an $(n+1)$-variate r.v. with
mutually independent univariate margins distributed gamma. Namely, for
$j=1,\ldots,n+1$, the p.d.f. of
$Y_j\backsim Ga(\gamma_j(\in\mathbf{R}_+),\alpha_j(\in\mathbf{R}_+))$ is given by
\begin{equation}
\label{Gampdf}
g(y;\gamma_j,\alpha_j)=e^{-\alpha_j y}\frac{y^{\gamma_j-1}\alpha_j^{\gamma_j}}
{\Gamma(\gamma_j)},\ y\in\mathbf{R}_+
\end{equation}
with the corresponding Laplace transform being well-defined on $\mathbf{R}_{+}:=(0,\ \infty)$ (the interval of
interest herein) and
given by
\begin{equation}
\label{GamLS}
\hat{G}(x;\gamma_j,\alpha_j)=\int_0^\infty e^{-xy}g(y)dy=\left(
1+\frac{x}{\alpha_j}
\right)^{-\gamma_j}.
\end{equation}
\begin{definition}[Furman, 2008; Furman and Landsman, 2010]
Let
$A\in Mat_{n\times (n+1)}(\mathbf{R}_{0,+})$ denote a deterministic $n\times (n+1)$ matrix
with suitable non-negative entries. Then $\mathbf{X}=A\mathbf{Y}$ is distributed
$n$-variate gamma
with shape parameters $\gamma_i^\ast=t_i(\gamma_1,\ldots,\gamma_{n+1})$
and rate parameters
$\alpha_i^\ast=u_i(\alpha_1,\ldots,\alpha_{n+1})$ for appropriate positive Borel functions
$t_i(\cdot),\ u_i(\cdot),\ i=1,\ldots,n$. Succinctly, we write
$\mathbf{X}\backsim Ga_{1,\ldots,n}(\boldsymbol{\gamma}^\ast,\boldsymbol{\alpha}^\ast)$,
where
$\boldsymbol{\gamma}^\ast=(\gamma_1^\ast,\ldots,\gamma_n^\ast)'\in\mathbf{R}_+^n$
and
$\boldsymbol{\alpha}^\ast=(\alpha_1^\ast,\ldots,\alpha_n^\ast)'\in\mathbf{R}_+^n$
are vectors of parameters.
\end{definition}
\begin{example}[Mathai and Moschopoulos, 1991; see also Cherian, 1941 and Ramabhadran, 1951]
\label{ExMMG1991}
Let $Y_j\backsim Ga(\gamma_j,\alpha_j),\ j=1,\ldots,n+1$ be mutually independent
random variables distributed gamma with arbitrary parameters, and
choose
the matrix $A$ such that, for
$i=1,\ldots,n$ and $\sigma_i>0$, it holds that $(A)_{i,n+1}=\alpha_{n+1}/\alpha_i$, $(A)_{i,i}\equiv 1$ and zero otherwise. Then
$\mathbf{X}\backsim Ga_{1,\ldots,n}(\boldsymbol{\gamma}^\ast,\boldsymbol{\alpha})$,
where $\boldsymbol{\gamma}^\ast=(\gamma_{n+1}+\gamma_1,\ldots,\gamma_{n+1}+\gamma_n)'$
and $\boldsymbol{\alpha}=(\alpha_1,\ldots,\alpha_n)'$ are two $n$-variate vectors of
parameters.
\end{example}
\begin{example}[Mathai and Moschopoulos, 1992; see also, Furman, 2008]
\label{ExFG2009}
Let $Y_j\backsim Ga(\gamma_j,\alpha_j)$, $j=1,\ldots,n$ be mutually independent
random variables distributed gamma with arbitrary parameters, and
choose
the matrix $A$ such that, for
$i=1,\ldots,n,\ j=1,\ldots,i$ and $\sigma_i>0$, it holds that
$(A)_{i,j}=\alpha_{j}/\sigma_i$ and zero otherwise.
Then
$\mathbf{X}\backsim Ga_{1,\ldots,n}(\boldsymbol{\gamma}^\ast,\boldsymbol{\sigma})$,
where $\boldsymbol{\gamma}^\ast=(\gamma^\ast_1,\ldots,\gamma^\ast_n)',\ \gamma^\ast_i=
\sum_{j=1}^i \gamma_j$
and $\boldsymbol{\sigma}=(\sigma_1,\ldots,\sigma_n)'$ are two $n$-variate vectors of
parameters.
\end{example}
In the present paper we employ the following modification of Examples \ref{ExMMG1991} and
\ref{ExFG2009}.
\begin{example}[Furman, 2008]
\label{ExFG2009gen}
Let $Y_j\backsim Ga(\gamma_j,\alpha_j),\ j=1,\ldots,n+1$ be again mutually independent
random variables distributed gamma with arbitrary parameters, and
choose
the matrix $A$ such that, for
$i=1,\ldots,n,\ j=1,\ldots,i$ and $\sigma_i>0$, it holds that
$(A)_{i,j}=\alpha_{j}/\sigma_i$, $(A)_{i,n+1}=\alpha_{n+1}/\sigma_i$ and zero otherwise.
Then
$\mathbf{X}\backsim Ga_{1,\ldots,n}(\boldsymbol{\gamma}^\ast,\boldsymbol{\sigma})$,
where $\boldsymbol{\gamma}^\ast=(\gamma^\ast_1,\ldots,\gamma^\ast_n)',\ \gamma^\ast_i=
\gamma_{n+1}+\sum_{j=1}^i \gamma_j$
and $\boldsymbol{\sigma}=(\sigma_1,\ldots,\sigma_n)'$ are two $n$-variate vectors of
parameters.
\end{example}
In the sequel, we embark on the idea in Example \ref{ExFG2009gen} to introduce
an encompassing yet tractable multivariate distribution with univariate
margins distributed Pareto.
We note in passing that a real-valued
r.v.
is said to be distributed Pareto of the $2$nd kind, succinctly
$X\backsim Pa(II)(\mu,\sigma,\alpha),$ where $\mu\in\mathbf{R}$ is a location parameter,
$\sigma\in\mathbf{R}_+$ is a scale parameter and $\alpha\in\mathbf{R}_+$ is a tail index,
if its c.d.f. is given by
\begin{equation}
\label{Par2cdf}
F_X(x;\mu,\sigma,\alpha)=1-\left(
1+\frac{x-\mu}{\sigma}
\right)^{-\alpha},\ x> \mu
\end{equation}
(see, e.g., Pareto, 1897; Arnold, 1983; Kotz et al., 2000).
Similarly to Asimit et al. (2010),
we set $\mu=0$, which conveniently makes the support of the distribution
coincide with the positive half of the real line, i.e.,
$supp_F=\{x\in\mathbf{R}\ :f(x)\neq 0\}=\mathbf{R}_{+}$ and does not lead
to any loss of generality. The resulting distribution (Lomax distribution), notationally $Pa(II)(\sigma,\alpha)$, enjoys
a great variety of applications in all areas of applied mathematics in general and in
actuarial science in particular, as it naturally arises in the extreme value theory
as the limiting distribution of the excess-of-loss r.v. $X_d:=X-d|\ X>d$
where
$d(\in\mathbf{R}_+)$ denotes a threshold
(see, e.g., Balkema and de Haan, 1974; Pickands, 1975).
The rest of the paper is organized as follows. In Section \ref{sec2} a multivariate probability structure
with dependent Pareto-distributed univariate margins is introduced and linked to a number of existing multivariate models.
Then distributional properties of the new structure are derived and some characterization results are proved in
Sections \ref{sec2} and \ref{sec-biv-stuff}.
In Section \ref{SecAppl} the new multivariate Pareto
is reintroduced as a variant of the minima-based multiple risk factor model, and some applications
to notions of actuarial interest are considered.
In Section \ref{sec-numex} an application of the model is elucidated with the help of a
numerical example borrowed from the context of default risk.
Section \ref{sec-con} concludes the paper. All proofs are relegated
to the appendix to facilitate the reading.
\section{New multivariate Pareto distribution }
\label{sec2}
Let $\mathbf{Y}=(Y_1,\ldots,Y_{n+1})'$ be a r.v. with mutually independent coordinates
$Y_j\backsim Ga(\gamma_j,1),$ $\gamma_j\in\mathbf{R}_+$, and choose
the matrix $A_c\in Mat_{n\times (n+1)}$ such
that $(A_c)_{i,j}=c_{i,j}/\sigma_i$, where
$c_{i,j}\in\{0,\ 1\}$ are deterministic constants, $\sigma_i\in\mathbf{R}_+$,
$i=1,\ldots,n$ and $j=1,\ldots,n+1$.
The following definition
unifies Examples \ref{ExMMG1991}, \ref{ExFG2009} and \ref{ExFG2009gen} and serves as an auxiliary tool for constructing the multivariate Pareto distribution of interest.
\begin{definition}
\label{GamDefG}
Let $\mathbf{X}=(X_1,\ldots,X_n)'=A_c\mathbf{Y}$, then it follows an $n$-variate
gamma distribution, notationally $\mathbf{X}\backsim Ga_{1,\ldots,n}(\boldsymbol{\gamma}_c^\ast,\
\boldsymbol{\sigma})$, where $\boldsymbol{\gamma}_c^\ast=(\gamma_{c,1}^\ast,\ldots,\gamma_{c,n}^\ast)'\in\mathbf{R}_+^n$ with $\gamma_{c,i}^\ast=\sum_{j=1}^{n+1} c_{i,j}\gamma_j,\ i=1,\ldots,n$ and
$\boldsymbol{\sigma}=(\sigma_1,\ldots,\sigma_n)'\in\mathbf{R}_+^n$ are two vectors of
parameters.
\end{definition}
We note in passing that Definition \ref{GamDefG} (auxiliary for the present paper)
establishes an encompassing multivariate probability law with gamma-distributed univariate
margins and an additive background risk dependence structure (see, Gollier and Pratt, 1996;
Tsanakas, 2008; Furman and Landsman, 2010 for applications of the additive background risk models in
economics and actuarial science). More specifically,
the following simple special cases of
$Ga_{1,\ldots,n}(\boldsymbol{\gamma}_c^\ast,\
\boldsymbol{\sigma})$ readily recover
the models of, respectively, Mathai and Moschopoulos (1991,\ 1992) and Furman (2008):
\begin{itemize}
\item $c_{i,i}=c_{i,n+1}\equiv 1$ for $i=1,\ldots,n$ and zero otherwise - Example 1.1;
\item $c_{i,j}\equiv1$ for $1\leq j\leq i\leq n$ and zero otherwise - Example 1.2;
\item $c_{i,j}=c_{i,n+1}\equiv1$ for $1\leq j\leq i\leq n$
and zero otherwise - Example 1.3.
\end{itemize}
Some elementary but useful properties of $\mathbf{X}\backsim Ga_{1,\ldots,n}$
follow directly by definition or from the Laplace transform that is established next.
\begin{proposition}
\label{propLTG}
Let $\mathbf{X}\backsim Ga_{1,\ldots,n}(\boldsymbol{\gamma}_c^\ast,\
\boldsymbol{\sigma})$ be the r.v. distributed multivariate gamma as in Definition \ref{GamDefG},
then the corresponding Laplace transform is given by
\[
\hat{G}_{1,\ldots,n}(\mathbf{t})=\prod_{j=1}^{n+1}\left(
1+\sum_{i=1}^n \frac{c_{i,j}}{\sigma_i}t_i
\right)^{-\gamma_j},
\]
and it is well-defined on $\mathbf{R}_{+}^n$.
\end{proposition}
Immediate consequences of Proposition \ref{propLTG} are, for $k,l=1,\ldots,n$, that
\begin{itemize}
\item the distribution of $\mathbf{X}\backsim Ga_{1,\ldots,n}(\boldsymbol{\gamma}_c^\ast,\
\boldsymbol{\sigma})$
is `marginally closed', i.e., $X_k\backsim Ga(\gamma_{c,k}^\ast(\in\mathbf{R}_+),\ \sigma_k(\in\mathbf{R}_+))$;
\item the expectation of the $k$-th coordinate is $\mathbf{E}[X_k]=\gamma_{c,k}^\ast/\sigma_k$;
\item the variance of the $k$-th coordinate is $\mathbf{Var}[X_k]=\gamma_{c,k}^\ast/\sigma_k^2$;
\item for $k\neq l$, the covariance between the coordinates
$X_k$ and $X_l$ is non-negative and
given by
\[
\mathbf{Cov}[X_k,X_l]=\frac{\sum_{j=1}^{n+1} c_{k,j}c_{l,j}\gamma_j}{\sigma_k\sigma_l};
\]
\item for $k\neq l$, the Pearson linear correlation between the coordinates
$X_k$ and $X_l$ is non-negative and given by
\[
\rho[X_k,X_l]=
\frac{\sum_{j=1}^{n+1} c_{k,j}c_{l,j}\gamma_j}{\sqrt{\gamma_{c,k}^\ast\gamma_{c,l}^\ast}}.
\]
\end{itemize}
We are now in a position to introduce the multivariate Pareto distribution of interest.
In fact, simple observation (\ref{GamLS}) along with Proposition \ref{propLTG} result in the following definition.
\begin{definition}
\label{defMPflexGen}
We call the r.v. $\mathbf{X}=(X_1,\ldots,X_n)'$ having the
decumulative distribution function
(d.d.f.)
\begin{equation}
\label{MPflexGenddf}
\overline{F}_{1,\ldots,n}(x_1,\ldots,x_n)=\prod_{j=1}^{n+1}\left(
1+\sum_{i=1}^n \frac{c_{i,j}}{\sigma_i}x_i
\right)^{-\gamma_j}, \textnormal{ where }(x_1,\ldots,x_n)'\in\mathbf{R}_{+}^n,
\end{equation}
a multivariate Pareto of the $2$nd kind. Succinctly, we write
$\mathbf{X}\backsim Pa_{1,\ldots,n}^c$
$(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$, where $\boldsymbol{\sigma}=(\sigma_1,\ldots,\sigma_n)',\ \boldsymbol{\gamma}=(\gamma_1,\ldots,\gamma_n)'$ are two deterministic vectors of positive
parameters, and $\gamma_{n+1}\in\mathbf{R}_+$ and $c\in Mat_{n\times (n+1)}(\{0,\ 1\})$ are scalar-valued
and matrix-valued parameters, respectively.
\end{definition}
Generally, distributions with Paretian tails have been applied in a multitude of areas. Herein we refer to:
Benson et al. (2007) for applications in modelling catastrophic risk; Koedijk et al. (1990), Longin (1996), Gabaix et al. (2003) for applications in general financial phenomena; Cebri\'{a}n et al. (2003) for
applications in insurance pricing; and Soprano et al. (2010), Chavez-Demoulin et al. (2015)
for applications in risk management.
Specifically, the probability law in Definition \ref{defMPflexGen} is a generalization of the classical
multivariate Pareto distribution of Arnold (1983) with the d.d.f. $\overline{F}_{1,\ldots,n}^{Arnold}$. Indeed, set
$c_{i,j}=c_{\bullet,j},\ i=1,\ldots,n,\ j=1,\ldots,n+1$ in (\ref{MPflexGenddf}) and
obtain, for $\gamma^\ast_{c}=\sum_{j=1}^{n+1}c_{\bullet,j}\gamma_j$, that
\begin{equation}
\label{MPArnold}
(2.1)=
\left(
1+\sum_{i=1}^n \frac{x_i}{\sigma_i}
\right)^{-{\gamma}^\ast_c}=\overline{F}^{Arnold}_{1,\ldots,n}(x_1,\ldots,x_n), \textnormal{ where }(x_1,\ldots,x_n)'\in\mathbf{R}_{+}^n.
\end{equation}
That being said, unlike the classical multivariate Pareto distribution of Arnold (1983),
the structure in Definition \ref{defMPflexGen} incorporates stochastic independence -
set $c_{i,i}\equiv 1,\ i=1,\ldots, n$ and zero otherwise and obtain, for $\overline{F}_{1,\ldots,n}^\Pi$ denoting
the d.d.f. of a multivariate Pareto with independent margins, that
\[
(2.1)
= \prod_{i=1}^{n}\left(
1+\frac{x_i}{\sigma_i}
\right)^{-\gamma_i}=\overline{F}^{\Pi}_{1,\ldots,n}(x_1,\ldots,x_n), \textnormal{ where }(x_1,\ldots,x_n)'\in\mathbf{R}_{+}^n.
\]
Consequently, the new multivariate Pareto distribution
meaningfully fills the gap between the multivariate probability distributions with independent
and Arnold-dependent Pareto-distributed margins.
In addition, unlike (\ref{MPArnold}), $Pa_{1,\ldots,n}^c(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$ allows for
distinct marginal tail indices (see, Proposition \ref{Nt1} below). Furthermore, the new multivariate Pareto distribution unifies the
probability models
studied recently in Chiragiev and Landsman
(2009). Namely, in order to obtain their `flexible Pareto type I and II' we choose $c_{i,i}=c_{i,n+1}\equiv 1,\ i=1,\ldots,n$ and zero
otherwise and $c_{i,j}\equiv 1$ for $1\leq j\leq i\leq n$ and zero otherwise, respectively.
Lastly but perhaps most importantly in actuarial applications, d.d.f. (\ref{MPflexGenddf}) admits stochastic representations
that mimic the
multiplicative background risk model (Franke et al., 2006) and the minima-based
common shock model (Bowers et al., 1997) (see, respectively, Theorems \ref{CharLem} and \ref{MinimaCS}
in this paper).
Stochastic representations are a very welcome facet, since they endow probabilistic models with an important feature of interpretability,
and as a result contribute greatly to the process of model selection and implementation.
We further document several simple properties of the multivariate Pareto with d.d.f.
(\ref{MPflexGenddf}).
The proofs are straightforward and thus omitted.
\begin{proposition}
\label{Nt1}
Let $\mathbf{X}\backsim Pa_{1,\ldots,n}^c$
$(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$ as in Definition
\ref{defMPflexGen}, then, for $i=1,\ldots,n$, the marginal d.d.f. of $X_i$ is
\[
\overline{F}_i(x_i)=
\left(1+\frac{x_i}{\sigma_i}\right)^{-\gamma_{c,i}^\ast},\ x_i\in\mathbf{R}_+,
\]
that is $X_i\backsim Pa(II)(\sigma_i,\ \gamma_{c,i}^\ast)$, where
$\gamma_{c,i}^\ast=\sum_{j=1}^{n+1}c_{i,j}\gamma_j$. Also, for $i=1,\ldots,n$ and setting
$\gamma_{c,i}^\ast>1$, we have that
\begin{equation}
\label{Epar}
\mathbf{E}[X_i]=\sigma_i/\left(\gamma_{c,i}^\ast-1\right),
\end{equation}
and furthermore setting $\gamma_{c,i}^\ast>2$, we obtain that
\begin{equation}
\label{Varpar}
\mathbf{Var}[X_i]=\sigma_i^2\gamma_{c,i}^\ast/
\left(\left(
\gamma_{c,i}^\ast-1
\right)^2
\left(
\gamma_{c,i}^\ast-2
\right)
\right).
\end{equation}
\end{proposition}
In Proposition \ref{Nt1},
the substitution $c_{i,i}=c_{i,n+1}\equiv 1,\ i=1,\ldots,n$ and zero otherwise yields Theorem 1 of Chiragiev and Landsman
(2009), whereas the substitution $c_{i,j}\equiv 1$ for $1\leq j\leq i\leq n$ and zero otherwise results in
their Theorem 5.
In what follows, we develop an expression for the joint p.d.f. of the multivariate Pareto distribution of interest. To this end,
let
\begin{equation}\label{pdfaux}
\prod_{i=1}^n\sum_{j=1}^{n+1}c_{i,j} y_j =\sum_{i_j\in I} d_c(i_1,\ldots,i_{n+1})\prod_{j=1}^{n+1} y_j^{i_j},
\end{equation}
where $I$ establishes a set of positive integer indices such that $\sum_{j=1}^{n+1}i_j=n$, and $d_c(i_1,\ldots,i_n)$ are appropriately
chosen constants. Also, let
\[
(\gamma)_n=\frac{\Gamma(\gamma+n)}{\Gamma(\gamma)},\ \textnormal{ where }
\gamma\in\mathbf{R}_+ \textnormal{ and } n\in\mathbf{N}
\]
denote the Pochhammer symbol.
\begin{theorem}
\label{jointpdf}
Let $\mathbf{X}\backsim Pa_{1,\ldots,n}^c(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$ as in Definition \ref{defMPflexGen},
then the corresponding joint p.d.f. is formulated, for $(x_1,\ldots,x_n)'\in \mathbf{R}_+^n$, as
\begin{eqnarray}
\label{joint_pdf}
f_{1,\ldots,n}(x_1,\ldots,x_n)
= \sum_{\forall i_j\in I}d_c(i_1,\ldots,i_{n+1})\prod_{j=1}^{n+1}\frac{(\gamma_j)_{i_j}}{\prod_{l=1}^n\sigma_l }\left(1+\sum_{i=1}^nc_{i,j}\frac{x_i}{\sigma_i}\right)^{-(\gamma_j+i_j)},
\end{eqnarray}
where $d_c(i_1,\ldots,i_{n+1})$ are appropriately chosen constants and $\ i_j\in I$.
\end{theorem}
In general, the constants $d_c(i_1,\ldots,i_n)$ can be rather involved. For an insight, we show
how (\ref{joint_pdf}) reduces to the p.d.f. of the classical multivariate Pareto distribution of Arnold (1983).
To this end,
set $c_{i,j}\equiv 1$ for $i=1,\ldots,n$ and $j=1\ldots,n+1$.
Then from (\ref{pdfaux}), we
have that
\[
d_c(i_1,\ldots,i_{n+1})={n \choose i_1,\ldots, i_{n+1}}
\]
with the right-hand side denoting the multinomial coefficient.
On the other hand, as (\ref{joint_pdf}) must integrate to one and since for the Arnold's
multivariate Pareto distribution, we have, for $\gamma^\ast=\gamma_1+\cdots+\gamma_{n+1}$,
that
\[
\prod_{j=1}^{n+1}\left(1+\sum_{i=1}^nc_{i,j}\frac{x_i}{\sigma_i}\right)^{-(\gamma_j+i_j)}
=\left(1+\sum_{i=1}^n\frac{x_i}{\sigma_i}\right)^{-(\gamma^\ast+n)},
\]
we obtain
\[
(2.6)=\frac{(\gamma^\ast)_n}{\prod_{i=1}^n \sigma_i}\left(1+\sum_{i=1}^n\frac{x_i}{\sigma_i}\right)^{-(\gamma^\ast+n)},\ \textnormal{for}
\ (x_1,\ldots,x_n)'\in\mathbf{R}_+^n,
\]
as required.
The following theorem establishes
a useful characteristic relation in the context of the multivariate Pareto distribution of interest, and
it also plays an important role when deriving the formula for the corresponding Pearson linear
correlation (see, Theorem \ref{cov} in Section \ref{sec-biv-stuff}). In the sequel `$\overset{d}{=}$' denotes equality in
distribution.
\begin{theorem}
\label{CharLem}
Let $\boldsymbol{\Lambda}=(\Lambda_1,\ldots,\Lambda_n)'$ be a r.v. with independent
and exponentially-distributed
univariate margins $\Lambda_i\backsim Exp(1),\ i=1,\ldots,n$, and denote by
$\boldsymbol{\Xi}=(\Xi_1,\ldots,\Xi_n)'$ $\backsim Ga_{1,\ldots,n}(\boldsymbol{\gamma}_c^\ast,\
\boldsymbol{\sigma})$ the $n$-variate gamma distribution introduced in Definition
\ref{GamDefG}; here $\boldsymbol{\gamma}_c^\ast=(\gamma_{c,1}^\ast,\ldots,\gamma_{c,n}^\ast)'\in\mathbf{R}_+^n$ with $\gamma_{c,i}^\ast=\sum_{j=1}^{n+1} c_{i,j}\gamma_j$, and
$\boldsymbol{\sigma}=(\sigma_1,\ldots,\sigma_n)'\in\mathbf{R}_+^n$ are vectors of
parameters. Assume that $\boldsymbol{\Lambda}$ and $\boldsymbol{\Xi}$ are stochastically independent,
then $\mathbf{X}=(X_1,\ldots,X_n)'$ has d.d.f.
(\ref{MPflexGenddf}), and it is thus the $n$-variate Pareto distribution introduced in
Definition \ref{defMPflexGen} if and only if
$(X_1,\ldots,X_n)'\overset{d}{=}(\Lambda_1/ \Xi_1,\ldots,\Lambda_n/ \Xi_n)'$.
\end{theorem}
Theorem \ref{CharLem} establishes the multiplicative background risk representation
of the multivariate probabilistic structure of main interest herein
(see, Franke et al., 2006; Meyers, 2007, Asimit et al., 2013, 2016
for applications of the multiplicative
background risk models in economics and actuarial science).
We conclude this section with yet another characterization of the multivariate Pareto
distribution of interest and its two implications. Namely,
let $\wedge_{i=1}^n X_i=:X_{-}\backsim F_{-}$ and
$\vee_{i=1}^n X_i=:X_+\sim F_+$ denote, respectively the minima and the maxima r.v.'s,
and let $X_i\sim F_i,\ i=1,\ldots,n$ be univariate coordinates of the multivariate Pareto r.v.
of interest in this paper.
\begin{theorem}
\label{minima}
Let $\mathbf{X}=(X_1,\ldots,X_n)'$ be distributed $Pa_{1,\ldots,n}^c(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$ as per Definition \ref{defMPflexGen}, then
$X_{-}$ admits the mixture representation as $X_-|\Lambda=\lambda\backsim Exp(\lambda)$ and
$\Lambda\overset{d}{=}Z_1+\cdots+Z_{n+1}$, where $Z_j,\ j=1,\ldots,n+1$ are univariate
mutually independent r.v.'s distributed gamma.
\end{theorem}
An important corollary of Theorem
\ref{minima} is a random parameter representation
(see, e.g., Feller, 1966)
of the d.d.f.'s of $X_-$ and $X_+$.
The following lemma is crucial in studying the distribution of $X_-$ in Theorem \ref{ggamma}.
\begin{lemma}[Moschopoulos, 1985; Furman and Landsman, 2005]
\label{FL2005} For $i=1,\ldots,n$, let $Z_i\sim Ga(\gamma_i(\in\mathbf{R}_+).\ \alpha_i(\in\mathbf{R}_+))$ denote independent gamma-distributed
r.v.'s. Then the distribution of $Z=Z_1+\cdots+Z_n$ is gamma with a random shape
parameter. More specifically, $Z\sim Ga(\gamma^\ast+K,\ \alpha_{+}),$ where
$\gamma^\ast=\gamma_1+\cdots+\gamma_n$, $\alpha_{+}=\vee_{i=1}^n \alpha_i$ and $K$ is an
integer-valued non-negative r.v. with the probability mass function (p.m.f.) given by
\begin{equation}
\label{pk}
p_k=\mathbf{P}[K=k]=c_{+} \delta_k,\ k=0,\ 1, \ldots,
\end{equation}
where
\[
c_{+}=\prod_{i=1}^n \left(\frac{\alpha_i}{\alpha_{+}}\right)^{\gamma_i},\
\delta_0=1
\]
and
\[
\delta_k=k^{-1}\sum_{l=1}^k\sum_{i=1}^n \gamma_i \left(
1-\frac{\alpha_i}{\alpha_{+}}
\right)^l\delta_{k-l} \textnormal{ for } k>0.
\]
\end{lemma}
\begin{theorem}
\label{ggamma} Let $\mathbf{X}\backsim Pa_{1,\ldots,n}^c(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$ as in Definition \ref{defMPflexGen},
then
$X_{-}\backsim Pa(II)(\alpha_{+}(\boldsymbol{\sigma}),\\ \gamma^\ast +K)$, where
$\alpha_{+}(\boldsymbol{\sigma})=\vee_{j=1}^{n+1} \left(\sum_{i=1}^n\frac{c_{i,j}}{\sigma_i}\right)^{-1}$, $K$ is an integer-valued
r.v. with p.m.f. (\ref{pk})
and $\gamma^\ast=\gamma_1+\cdots+\gamma_{n+1} >1$.
\end{theorem}
While Theorem \ref{ggamma} demonstrates that the minima r.v. $X_-$ is distributed mixed Pareto with random
tail index parameter, the next theorem asserts that the maxima r.v. $X_+$ has a d.d.f.
that is a linear combination of the d.d.f.'s of such mixed Pareto-distributed r.v.'s.
The proof is similar to the one of Proposition 2 in Vernic (2011) and is thus omitted.
\begin{theorem}
\label{maxrv}
Assume that $\mathbf{X}\backsim Pa_{1,\ldots,n}^c(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$ as in Definition \ref{defMPflexGen}, then
the d.d.f. of the maxima r.v.
is given by
\begin{equation}
\label{maxddf}
\overline{F}_+(x)=\sum_{\mathcal{S}\subseteq\{1,\ldots,n\}} (-1)^{|\mathcal{S}|-1}
\overline{F}_{\mathcal{S}-}(x),\ x\in\mathbf{R}_{ +},
\end{equation}
where $X_{\mathcal{S}-}=\wedge_{s\in\mathcal{S}\subseteq\{1,\ldots,n\}}X_s$ and
$X_{\mathcal{S}-}\sim F_{\mathcal{S}-}$.
\end{theorem}
\section{Bivariate quantities of interest}
\label{sec-biv-stuff}
It is worthwhile to make an additional observation before stating the main result of
this section. Namely, we note in passing that for $1\leq k\neq l\leq n$,
a r.v. $(\Xi_k,\ \Xi_l)'$ distributed the bivariate gamma per Definition
\ref{GamDefG}
and an $(n+1)$-variate r.v.
$\mathbf{Y}=(Y_1,\ldots,Y_{n+1})'$ having mutually independent coordinates
$Y_j\backsim Ga(\gamma_j,1),$ $\gamma_j\in\mathbf{R}_+$,
the
following stochastic representation holds
\begin{equation}
\label{StRep}
(\sigma_k \Xi_k,\ \sigma_l \Xi_l)'
\overset{d}{=}
(Y_{c,(k,l)}+Y_{c,k},\ Y_{c,(k,l)}+Y_{c,l})',
\end{equation}
where $Y_{c,(k,l)}=\sum_{j=1}^{n+1}c_{k,j}c_{l,j}Y_j$, $Y_{c,k}=\sum_{j=1}^{n+1}c_{k,j}(1-c_{l,j})Y_j$ and $Y_{c,l}=\sum_{j=1}^{n+1}c_{l,j}(1-c_{k,j})Y_j$ are mutually independent
gamma-distributed r.v.'s with the shape parameters $\gamma_{c,(k,l)}=\sum_{j=1}^{n+1}c_{k,j}c_{l,j}\gamma_j$,
$\gamma_{c,k}=\sum_{j=1}^{n+1}c_{k,j}(1-c_{l,j})\gamma_j$ and
$\gamma_{c,l}=\sum_{j=1}^{n+1}c_{l,j}(1-c_{k,j})\gamma_j$, respectively.
We next show that the covariance of a random couple within the multivariate Pareto of
interest in this paper can be formulated using the $(q+1)\times q$ hypergeometric function
(see, Gradshteyn and Ryzhik, 2007), which is formulated as
\begin{eqnarray}
\label{3F2}
_{q+1}F_q(a_1,\ldots,a_{q+1};b_1,\ldots,b_q;z):=\sum_{k=0}^{\infty}\frac{(a_1)_k,\ldots,(a_{q+1})_k }{(b_1)_k,\ldots,(b_q)_k}\frac{z^k}{k!},\
\end{eqnarray}
where $q\in\mathbf{Z}_+$.
For $a_1,\ldots,a_{q+1}$ all positive, and these are the cases of interest in the present paper,
the radius of convergence of the series is the open disk $|z|<1$. On the boundary $|z|=1$, the series
converges absolutely if $h:=b_1+\cdots + b_q-a_1-\cdots -a_{q+1}>0$, and it
converges except at $z=1$ if $0\geq h>-1$.
\begin{theorem}
\label{cov}
Let $\mathbf{X}\backsim Pa_{1,\ldots,n}^c(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$ as in Definition \ref{defMPflexGen} and assume that
both $\gamma_{c,k}^\ast$ and $\gamma_{c,l}^\ast$ exceed two, then, for
$0\leq k\neq l\leq n$,
\[
\mathbf{Cov}[X_k,\ X_l]=
\sigma_k\sigma_l
\frac{1}{(\gamma_{c,k}^\ast-1)(\gamma_{c,l}^\ast-1)}
\left({}_3F_2\left(\gamma_{c,(k,l)},1,1;\gamma_{c,k}^\ast,\gamma_{c,l}^\ast;1\right)-1\right).
\]
\end{theorem}
An immediate consequence of Theorem \ref{cov} is that the maximal attainable Pearson
correlation in the context of the multivariate Pareto distribution introduced in the
present paper is not one. This consequence is however solely a
result of the fact that the
Pearson index of correlation exists only if the involved second moments are finite, a
pitfall that is well-known to non-life
actuaries, which often deal with heavy-tailed losses (see, Embrechts et al., 2002).
\begin{corollary}
\label{corrbounds}
Let $\mathbf{X}\backsim Pa_{1,\ldots,n}^c(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$ and assume that
both $\gamma_{c,k}^\ast$ and $\gamma_{c,l}^\ast$ exceed two for
$0\leq k\neq l\leq n$, then, for the Pearson correlation, it holds that $\mathbf{Corr}[X_k,\
X_l]\in [0,\ 1/2)$.
\end{corollary}
Another consequence of Theorem \ref{cov} pertains to two special cases of the
multivariate Pareto introduced in this paper, and it is formulated as the following corollary.
\begin{corollary}
\label{CL2009Cov}
Let $\mathbf{X}_1=(X_{1,1},\ldots, X_{1,n})'\sim Pa_{1,\ldots,n}^{(I)}$ and
$\mathbf{X}_2=(X_{2,1},\ldots,X_{2,n})'\sim Pa_{1,\ldots,n}^{(II)}$
be distributed, respectively, the multivariate flexible Pareto of type $I$ and $II$ of Chiragiev and
Landsman (2009). Then the corresponding covariances are readily obtained,
for $0\leq k \neq l\leq n$, as
\begin{eqnarray}
\label{covflex1}
&&\mathbf{Cov}[X_{1,k},\ X_{1,l}] \\
&=&\frac{\sigma_k\sigma_l}{(\gamma_k+\gamma_{n+1}-1)(\gamma_l+\gamma_{n+1}-1)}
\left({}_3F_2\left(\gamma_{n+1},1,1;\gamma_k+\gamma_{n+1},\gamma_l+\gamma_{n+1};1\right)-1\right) \notag,
\end{eqnarray}
for $\gamma_k+\gamma_{n+1}>2$ and $\gamma_l+\gamma_{n+1}>2$,
and as
\begin{eqnarray}
\label{covflex2}
\mathbf{Cov}[X_{2,k},\ X_{2,l}]&=& \frac{\sigma_k\sigma_l}{(\gamma_{c,k}^\ast-1)(\gamma_{c,l}^\ast-1)(\gamma_{c,l}^\ast-2)},
\end{eqnarray}
for $\gamma_{c,k}^\ast=\sum_{j=1}^k\gamma_j>2$ and $\gamma_{c,l}^\ast=\sum_{j=1}^l\gamma_j>2$.
\end{corollary}
We note in passing that expression (\ref{covflex2}) confirms the one derived
in Chiragiev and Landsman (2009), whereas formula (\ref{covflex1}) complements the discussion
therein. Also, the covariance of two r.v.'s coming from the Arnold's multivariate
Pareto distribution (see, Arnold, 1983) is readily obtained from both ($\ref{covflex1}$) and
($\ref{covflex2}$). More specifically, we set $\gamma_k=\gamma_l\equiv 0$
for all $0\leq k \neq l\leq n$ and verify that
(\ref{covflex1}) reduces to
\[
\frac{\sigma_k\sigma_l}{(\gamma_{n+1}-1)(\gamma_{n+1}-1)}
\left({}_2F_1\left(1,1;\gamma_{n+1};1\right)-1\right)=
\frac{\sigma_k\sigma_l}{(\gamma_{n+1}-1)^2(\gamma_{n+1}-2)},
\]
for $\gamma_{n+1}>2$. The verification is straightforward in the case of (\ref{covflex2}).
Generalized hypergeometric function (\ref{3F2}) plays an important role when deriving the
centred regression function $r(y)=\mathbf{E}[X-\mathbf{E}[X]|\ Y=y]$, where $y\in\mathbf{R}_+$
(see, Furman and Zitikis, 2008b, 2010, for applications of the function in insurance and
finance). We next present the conditional d.d.f., followed by the centered regression function for a pair of r.v.'s having the probability law
as in Definition \ref{defMPflexGen}. To this end, let
\[
m(x)=\frac{\sigma_k}{\gamma_{c,(k,l)}}\left(
1+\frac{x}{\sigma_l}
\right),\ x\in\mathbf{R}_+.
\]
\begin{theorem}
\label{cd}
Let $\mathbf{X}\backsim Pa_{1,\ldots,n}^c(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$ as in Definition \ref{defMPflexGen}, the d.d.f. of $X_k$ given $X_l=x_l,\ 0\leq k\neq l\leq n$, is formulated as
\begin{eqnarray}
\label{CondFunction}
&&\mathbf{P}[X_k>x_k|X_l=x_l]\\
=&&\left(\frac{\gamma_{c,(k,l)}}{\gamma_{c,l}^\ast}+\frac{\gamma_{c,l}}{\gamma_{c,l}^\ast} \left(1+\frac{x_k}{\gamma_{c,(k,l)}m(x_l)} \right)\right)\left(1+\frac{x_k}{\sigma_k} \right)^{-\gamma_{c,k}}\left(1+\frac{x_k}{\gamma_{c,(k,l)}m(x_l)} \right)^{-\gamma_{c,(l,k)}-1}, \nonumber
\end{eqnarray}
where $x_k,x_l\in \mathbf{R}_+$.
\end{theorem}
\begin{theorem}
\label{cregTh}
Let $\mathbf{X}\backsim Pa_{1,\ldots,n}^c(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$ as in Definition \ref{defMPflexGen}, the centred regression function of $X_k$ on $X_l,\ 0\leq k\neq l\leq n$, is given, for $\gamma_{c,k}^\ast>1$, by
\begin{equation}
\label{rk}
r_k(x_l)=m(x_l)\sum_{i=1}^2 a_i\ _2F_1\left({\gamma}_{c,k},1;{\gamma}_{c,k}^\ast+2-i;-\frac{x_l}{\sigma_l} \right)-\sigma_k/(\gamma_{c,k}^\ast -1),
\end{equation}
where
\[
a_1=\frac{{{\gamma}^2_{c,(k,l)}}}{{\gamma}^*_{c,k}~ {\gamma}^*_{c,l}},\
a_2=\frac{{\gamma}_{c,l} ~{\gamma}_{c,(k,l)}}{{\gamma}^*_{c,l} ({\gamma}^*_{c,k}-1)}\
\textnormal{ and } x_l\in\mathbf{R}_{+}.
\]
The centred regression function is monotonically-increasing and concave.
\end{theorem}
{\color{blue} We reiterate that our results readily recover the ones derived in Landsman and Chiragiev (2009). More specifically,
by a simple alignment of notation in Theorem \ref{cregTh} above, we obtain Theorem 3 in loc. cit., whereas by choosing
$\gamma_{c,k}=0$ in Theorem \ref{cregTh} and hence for
\[
{}_2F_1\left(0,1;{\gamma}_{c,k}^\ast+2-i;-\frac{x_l}{\sigma_l} \right)=1,\
a_1=\frac{\gamma_{c,k}^\ast}{\gamma_{c,l}^\ast} \textnormal{ and }
a_2=\frac{(\gamma_{c,l}^\ast-\gamma_{c,k}^\ast)\gamma_{c,k}^\ast}{\gamma_{c,l}^\ast(\gamma_{c,k}^\ast-1)},
\]
we end up with Theorem 7 therein.}
Clearly, the centred regression function of the new multivariate Pareto distribution
is not linear, while it is well-known that the classical multivariate Pareto has linear regression
(see, Arnold, 1983). Theorem \ref{cregTh} confirms the latter fact by
setting $\gamma_{c,(k,l)}=\gamma_{c,k}^\ast=\gamma_{c,l}^\ast\equiv \gamma_{n+1}$ and $\gamma_{c,l}=\gamma_{c,k} \equiv 0$ in (\ref{rk}), which then reduces to the following
linear form
\[
r_k(x_l)=\frac{\sigma_k}{\sigma_l}\left(x_l-\frac{\sigma_l}{\gamma_{n+1}(\gamma_{n+1}-1)}\right)
\]
for $\gamma_{n+1}>1,\ 0\leq k\neq l\leq n$ and $x_l\in\mathbf{R}_+$.
\section{Applications to insurance}
\label{SecAppl}
In what follows, we assume that $\mathbf{X}=(X_1,\ldots,X_n)'$ denotes a risk portfolio
(r.p.) with $X_i,\ i=1,\ldots,n$ representing its risk components (r.c.'s). According to Theorem
\ref{CharLem}, if $\mathbf{X}\sim
Pa_{1,\ldots,n}^c(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$, then
it admits the multiplicative background risk representation (see, Franke et al., 2006;
Meyers, 2007; Asimit et al., 2013, 2016).
We next show that the new multivariate Pareto distribution can also be interpreted
as a variant of the classical
minima-based common shock model (see, e.g., Bowers et al,
1997). To this end,
assume that the $i$-th r.c of the r.p. is exposed to the set $\mathcal{R}_i=\{r\in\mathbf{N}:r\leq (n+1)\}$,
$i=1,\ldots,n$ of risk factors (r.f.'s) and
let the r.v.
$\mathbf{Y}=(Y_1,\ldots,Y_{n+1})'$ stipulate the randomness of actuarial interest associated
with the r.f.'s.
The following theorem
establishes the minima-based
multiple risk factor representation of the multivariate Pareto proposed in the present paper.
We note in
passing that `$*$' stands for the mixture operator, i.e., given two appropriately
jointly measurable r.v.'s
${X}_{{\lambda}}\sim C(\cdot;{\lambda})$ and ${\Lambda}\sim H$,
it holds that ${X}_{{\lambda}}*{\Lambda}\overset{d}{=}{X}_{{\Lambda}}$.
\begin{theorem}
\label{MinimaCS}
Let $\mathbf{W}_i=(W_{i,1},\ldots,W_{i.n+1})'$ be r.v.'s with
independent
exponentially-distributed margins
$W_{i,j}\sim Exp(\lambda_{i,j}(\in\mathbf{R}_+)),\ i=1,\ldots,n,\ j=1,\ldots,n+1$, and
let $A_c$ be a deterministic matrix
of zero-one coefficients. Also, let $\mathbf{\Lambda}=(\Lambda_1,\ldots,\Lambda_{n+1})'$
be a r.v. having independent gamma-distributed margins with arbitrary shape parameters
$\gamma_j(\in\mathbf{R}_+)$ and rate parameters equal to $1$,
$j=1,\ldots,n+1$. Set, for $\sigma_i \in \mathbf{R}_+$ and $i=1,\ldots,n$,
\begin{equation}\label{minX-eq}
X_i=\sigma_i
\bigwedge_{j=1,\ c_{i,j}\neq 0}^{n+1}
(W_{i,j}* \Lambda_j),
\end{equation}
then
$\mathbf{X}=(X_1,\ldots,X_n)'\sim Pa_{1,\ldots,n}^c(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$.
\end{theorem}
Theorem \ref{MinimaCS} suggests that the multivariate Pareto distribution
proposed in the present paper might be an appropriate formal framework for modelling
dependent default, survival or failure times when these times are exponentially-distributed
with random parameters. We elaborate on this observation in Section
\ref{sec-numex}.
\subsection{Actuarial risk measurement}
Regulatory accords around the globe require that insurance companies carry out a careful
assessment of their future losses. From now on, the r.v. $X:\Omega \rightarrow \mathbf{R}_+$ is
interpreted as an insurance loss r.v., and $\mathcal{X}$ denotes the collection of such r.v.'s.
\begin{definition}
\label{DefRM}
A risk measure is a functional map
$H:\mathcal{X}\rightarrow[0,\ \infty]$.
\end{definition}
The literature on risk measures is
vast and growing quickly. The following two
indices are arguably the most popular amongst practitioners.
\begin{definition}
\label{VaRDef}
Let $X\in\mathcal{X}$ and fix $q\in[0,\ 1)$, then the Value-at-Risk (VaR) and the Conditional
Tail Expectation (CTE) risk measures are respectively given by
\begin{equation}
\label{VaRdef}
VaR_q[X]=\inf\{x\in\mathbf{R}:\mathbf{P}[X\leq x]\geq q\}
\end{equation}
and
\begin{equation}
\label{CTEdef}
CTE_q[X]=\mathbf{E}[X|\ X>VaR_q[X]].
\end{equation}
\end{definition}
\noindent
We note in passing that both VaR and CTE are distorted as well as weighted risk measures (see, respectively, Wang, 1996; and Furman and
Zitikis, 2008a).
\begin{definition}{Furman and Zitikis (2008a), see also Choo and de Jong (2009, 2010).}
\label{Weightedpcp}
Let $w:\mathbf{R}\rightarrow \mathbf{R}_+$ be a
non-decreasing Borel (weight) function, such
that $0<\mathbf{E}[w(X)]<\infty$,
then the class of weighted risk measures is defined as
\begin{equation}
\label{wpcp}
\pi_{w}[X]=\frac{\mathbf{E}[Xw(X)]}{\mathbf{E}[w(X)]}
\textnormal{ for }X\in\mathcal{X}.
\end{equation}
\end{definition}
Let $v_1,\ w_1:\mathbf{R}^n\rightarrow \mathbf{R}_+$ be two
legitimate weight functions such that all expectations in (\ref{vwpcp}) are finite,
and consider a generalized variant of (\ref{wpcp})
\begin{equation}
\label{vwpcp}
\pi_{v_1,\ w_1}[\boldsymbol{\Lambda}]=
\frac{\mathbf{E}[v_1(\boldsymbol{\Lambda})w_1(\boldsymbol{\Lambda})]}{\mathbf{E}[w_1(\boldsymbol{\Lambda})]}\
\textnormal{ for }\boldsymbol{\Lambda}:\Omega\rightarrow \mathcal{R}\subseteq \mathbf{R}^n.
\end{equation}
\begin{proposition}
\label{CTEmixt}
Let $X|\boldsymbol{\Lambda} =\boldsymbol{\lambda} \backsim C(\cdot;
\boldsymbol{\lambda}(\in\mathcal{R}\subseteq\mathbf{R}^n))$ and assume that $\boldsymbol{\Lambda} \backsim H_{1,\ldots,n}$, then for any legitimate
weight function, the functional $\pi_w[X]$ admits representation (\ref{vwpcp}).
\end{proposition}
\begin{corollary}
\label{VaRProp}
Let $\mathbf{X}\backsim Pa_{1,\ldots,n}^c(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$ as in Definition \ref{defMPflexGen}, then, for $i=1,\ldots,n$ and $q\in[0,\ 1)$,
we have that
\[
VaR_q[X_i]=\sigma_i\left(
(1-q)^{-1/\gamma_{c,i}^\ast}-1
\right).
\]
\end{corollary}
\begin{corollary}
\label{CTEgen}
Let $X|\boldsymbol{\Lambda} =\boldsymbol{\lambda} \backsim C(\cdot;\
\boldsymbol{\lambda}(\in\mathcal{R}\subseteq\mathbf{R}^n))$ and assume that $\boldsymbol{\Lambda} \backsim H_{1,\ldots,n}$, then
the CTE risk measure of $X\backsim F$ is, if exists and for $q\in[0,\ 1)$, given by
\[
CTE_q[X]=\frac{\mathbf{E}[\overline{C}(VaR_q[X];\boldsymbol{\Lambda})CTE_{q^\ast}[X|\boldsymbol{\Lambda}]]}{\overline{F}(VaR_q[X])} \textnormal{ for } X\in\mathcal{X},
\]
where $q^\ast=C(VaR_q[X];\ \boldsymbol{\lambda})$.
\end{corollary}
\begin{corollary}
\label{CTEParXiCor}
Let $\mathbf{X}\backsim Pa_{1,\ldots,n}^c(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$ as in Definition \ref{defMPflexGen}, we have that the CTE risk measure is,
if exists and for $i=1,\ldots,n$ and $q\in[0,\ 1)$, given by
\begin{eqnarray}
\label{CTEpar}
CTE_q[X_i]&=&\mathbf{E}[X_i] \frac{\overline{F}_{X_i^\ast}(VaR_q[X_i])}{1-q}+VaR_q[X_i] \nonumber\\
&=&\mathbf{E}[X_i]+VaR_q[X_i]\frac{\gamma_{c,i}^\ast}{\gamma_{c,i}^\ast-1},
\end{eqnarray}
where $X_i^\ast \sim Pa(II)(\sigma_i,\gamma_{c,i}^\ast-1)$.
\end{corollary}
The minima r.v. $X_{-} = \wedge_{i=1}^n X_i$ plays an important role in insurance mathematics
(recall, e.g., the joint life policies in life insurance), as well as in general finance
(think of, e.g., the first-to-default baskets).
Recall that the r.v. $K$ has been defined as an integer-valued non-negative r.v.
with the following p.m.f.
\begin{equation}
\label{pk2}
p_k=\mathbf{P}[K=k]=c_{+} \delta_k,\ k=0,\ 1, \ldots,
\end{equation}
where
\[
c_{+}=\prod_{i=1}^n \left(\frac{\alpha_i}{\alpha_{+}}\right)^{\gamma_i},\
\delta_0=1
\]
and
\[
\delta_k=k^{-1}\sum_{l=1}^k\sum_{i=1}^n \gamma_i \left(
1-\frac{\alpha_i}{\alpha_{+}}
\right)^l\delta_{k-l} \textnormal{ for } k>0.
\]
\begin{proposition}
\label{CTEparMin}
In the context of the multivariate Pareto of interest, the CTE risk measure of the
minima can be written, if finite and for $q\in[0,\ 1)$, as
\[
CTE_q[X_-]=\mathbf{E}[X_-]\frac{\overline{F}_{X_-^\ast}(VaR_q[X_{-}])}{1-q}+VaR_q[X_{-}],
\]
where $X_-^\ast \sim Pa(\alpha_+(\boldsymbol{\sigma}),\ \gamma^\ast+Q-1)$, where
$\alpha_+(\boldsymbol{\sigma})=\vee_{j=1}^{n+1}\left(
\sum_{i=1}^n \frac{c_{i,j}}{\sigma_i}
\right)^{-1}$, $\gamma^\ast=\gamma_1+\cdots+\gamma_{n+1}$ and $Q$ is an integer-valued r.v. with the p.m.f. obtained from the p.m.f.
of $K$ with the help of the following change of measure
\begin{equation}
\label{cmkq}
q_k=\frac{1}{\mathbf{E}[X_{-}]}\frac{\alpha_+(\boldsymbol{\sigma})}{\gamma^\ast+k-1}p_k,\
k=0,1,\ldots
\end{equation}
\end{proposition}
\begin{proposition}
\label{ctemaxima}
In the context of the multivariate Pareto of interest, the CTE risk measure of the
maxima can be written, if finite and for $q\in[0,\ 1)$, as
the following linear combination
\[
CTE_q[X_+]=\frac{1}{1-q}\sum_{\mathcal{S}\subseteq \{1,\ldots,n\}} (-1)^{|\mathcal{S}|-1} \mathbf{E}[X_{\mathcal{S}-}|X_{\mathcal{S}-}>VaR_q(X_+)] \overline{F}_{{\mathcal{S}-}}(VaR_q(X_+)),
\]
where $X_{\mathcal{S}-}=\wedge_{s\in \mathcal{S}\subseteq\{1,\ldots,n\}}X_s$ and
$X_{\mathcal{S}-}\sim F_{\mathcal{S}-}$.
\end{proposition}
\begin{definition}[Furman and Zitikis, 2008b]
Let $w:\mathbf{R} \rightarrow \mathbf{R}_+$ be a non-decreasing Borel function, such that
$0<\mathbf{E}[w(Y)]<\infty$, then the functional $\Pi:\mathcal{X}\times\mathcal{X}
\rightarrow [0,\ \infty]$ is referred to as the economic risk measure. Moreover,
the special form of $\Pi$, given by
\begin{equation}
\label{ewpcp}
\Pi_w[X,\ Y]=\frac{\mathbf{E}[Xw(Y)]}{\mathbf{E}[w(Y)]}\
\textnormal{for\ }X\in\mathcal{X} \textnormal{ and }Y\in\mathcal{X},
\end{equation}
is called a weighted economic risk measure.
\end{definition}
We further derive an expression for the economic CTE risk measure, which is a particular
case of (\ref{ewpcp}) with $w(y)=\mathbf{1}\{y>VaR_q[Y]\},\ q\in[0,\ 1)$ and $y\in\mathbf{R}_+$. To this
end, we find the next proposition useful. The proof is plain and thus omitted.
\begin{proposition}
\label{conditional_distribution}
Let $\mathbf{X}\backsim Pa_{1,\ldots,n}^c(\boldsymbol{\sigma},\ \boldsymbol{\gamma},\ \gamma_{n+1})$ as in Definition \ref{defMPflexGen}, the d.d.f.
of $X_k$ given $X_l>x_l,\ 0\leq k\neq l\leq n$, is formulated as
\begin{eqnarray}
\label{CondFunction}
\mathbf{P}[X_k>x_k|X_l>x_l]=\left(1+\frac{x_k}{\sigma_k} \right)^{-\gamma_{c,k}}\left(1+\frac{x_k}{\gamma_{c,(k,l)}m(x_l)} \right)^{-\gamma_{c,(l,k)}}, \nonumber
\end{eqnarray}
where
\[
m(x_l)=\frac{\sigma_k}{{\gamma}_{c,(k,l)}}\left(1+\frac{x_l}{\sigma_l}\right)
\]
and
$x_k,\ x_l$ are both in $\mathbf{R}_+$.
\end{proposition}
\begin{proposition}
\label{econCTE}
In the context of the multivariate Pareto of interest,
the economic CTE risk measure is given, for $\gamma^\ast_{c,k}>1$,
$q\in[0,\ 1)$ and $1\leq k \neq l\leq n$, by
\[
\mathbf{E}[X_k|X_l > VaR_q[X_l]]=\frac{\sigma_k}{\gamma_{c,k}^*-1} {}_2F_1\left(\gamma_{c,(k,l)},1;\gamma_{c,k}^*;\frac{VaR_q[X_l]}{\sigma_l+VaR_q[X_l]}\right).
\]
\end{proposition}
To summarize, so far we have introduced and studied a new multivariate probability distribution with the univariate margins
distributed Pareto of the 2nd kind. The dependence structure of the new distribution is driven by a number
of stochastic representations that are variants of the multiplicative background risk and the minima-based
common shock models. We next employ the latter interpretation of the proposed multivariate probability
distribution to exemplify its possible application to modelling and measuring default risk.
\section{Numerical illustration}
\label{sec-numex}
{
For the sake of the discussion in this section,
we adopt the view of the
Financial Stability Board and the International Monetary Fund
that the systemic risk can be caused by impairment of
all or parts of the financial system, and more formally, we call the risk factor
$j\in\{1,\ldots,n+1\}$ `systemic', if $c_{i,j}=1$ for at least two distinct r.c.'s
$i\in\{1,\ldots,n\}$. Similarly, we call the risk factor $j\in\{1,\ldots,n+1\}$
`idiosyncratic', if $c_{i,j}=1$ for only one risk component $i\in\{1,\ldots,n\}$.
Consider obligors in a default risk portfolio,
each of
which is exposed to exactly two distinct categories of fatal risk factors, e.g.,
systemic (category A)
and idiosyncratic (category B).
We assume that the
risk factors from distinct risk categories are independent and that
the hitting times (or occurrences) of defaults of the r.c.'s are exponentially-distributed with random parameters distributed gamma.
In fact,
the future lifetime r.v. of the $i$-th r.c. has exponential
distribution with the random parameter $\sigma_i^{-1}\sum_{j=1}^{n+1}c_{i,j}\Lambda_j$, where $\Lambda_j$ are
distributed gamma with unit rate parameters, and $i$ is $1,\ 2$ or 3.
Then Theorem \ref{MinimaCS} readily implies that the joint default times of the aforementioned
r.c.'s has d.d.f. (\ref{MPflexGenddf}).
To illustrate the effect of the dependence structure on the joint default
probability we further set the dimension to $n=3$ and
specialize the set-up above along the lines in Section 16.8 of Engelmann and Rauhmeier (2011)
as well as employing the 2014's Annual Global Corporate Default Study and Rating Transitions of
Standard \& Poor's. (Standard \& Poor's, 2015).
More specifically, we set
$\mu:=\mathbf{E}[\Lambda_j]\equiv 1.67$, fix the time horizon to $15$
years and choose the corresponding default probability, $p$ say, to be equal to $0.3198$
(on par with the `B' credit rating of speculative entities). This yields the multivariate
probability structure of Definition \ref{defMPflexGen} with identically distributed
margins having the parameters $\sigma_i\equiv \sigma=122.39$ and $\gamma_{c,i}^*\equiv 3.33$, for $i=1,\ 2$ and
$3$.
Then we explore three different exposures
of the obligors to the systemic
and idiosyncratic r.f.'s.
The distinct exposures
are stipulated by appropriate choices of the $c$ parameters gathered by matrices $A_c^{(k)},\
k=1,\ 2,\ 3$. We compare the aforementioned three exposures with the reference case
in which no systemic risk presents, that is the joint d.d.f. of default times
is a trivariate Pareto with independent margins.
We note in passing that the expressions for the d.d.f.'s below readily follow
from Theorem \ref{MinimaCS}, whereas the values of the Pearson correlation coefficient
are in non-trivial cases obtained with the help of Theorem \ref{cov}.
\begin{itemize}
\item [Case (1).]
Only the systemic risk presents, and all risk components are exposed to it. The exposure
is represented schematically with the use of the following matrix, in which the rows
and the columns represent r.c.'s and r.f.'s, respectively
\[
A_c^{(1)}=\left(
\begin{array}{cc|ccc}
1 & 1 & 0 & 0\\
1 & 1 & 0 & 0\\
1 & 1 & 0 & 0
\end{array}
\right).
\]
The joint d.d.f. of the risk components is given by
\[
\overline{F}^{(1)}(x_1,x_2,x_3)=\left(
1+\frac{x_1+x_2+x_3}{\sigma}
\right)^{-2\mu},
\]
where $x_1,x_2,x_3$ are all in $\mathbf{R}_{+}$.
This is obviously the d.d.f. of the classical trivariate Pareto distribution (Arnold, 1983).
In this r.p., the Pearson correlation coefficient between any two of the r.c.'s is $0.3$.
\end{itemize}
\noindent
In the following two cases, both the systemic and idiosyncratic risks present.
\begin{itemize}
\item [Case (2).]
There are overall three uncorrelated idiosyncratic risk factors and one
systemic risk factor. The exposure is gathered by the following block matrix
\[
A_c^{(2)}=\left(
\begin{array}{c|cccc}
1 & 1 & 0 & 0\\
1 & 0 & 1 & 0\\
1 & 0 & 0 & 1
\end{array}
\right).
\]
The joint d.d.f. of the risk components is given by
\begin{eqnarray*}
&&\overline{F}^{(2)}(x_1,x_2,x_3)\\
&&=\left(
1+\frac{x_1+x_2+x_3}{\sigma}
\right)^{-\mu}
\left(
1+\frac{x_1}{\sigma}
\right)^{-\mu}
\left(
1+\frac{x_2}{\sigma}
\right)^{-\mu}
\left(
1+\frac{x_3}{\sigma}
\right)^{-\mu},
\end{eqnarray*}
where $x_1,x_2,x_3$ are all in $\mathbf{R}_{+}$. This case corresponds to the `flexible Pareto type I' of Landsman and
Chiragiev (2009).
In this r.p., the Pearson correlation coefficient between any two of the r.c.'s is $0.09$.
\item [Case (3).] The systemic
risk is represented by two distinct risk factors of which one targets the entire risk
portfolio and the other only hits r.c.'s $\# 1$ and $\# 2$. There is one idiosyncratic
risk factor, and only r.c. $\# 3$ is exposed to it.
The exposure block matrix is given by
\[
A_c^{(3)}=\left(
\begin{array}{cc|ccc}
1 & 1 & 0 & 0\\
1 & 1 & 0 & 0\\
1 & 0 & 1 & 0
\end{array}
\right).
\]
The joint d.d.f. of the risk components is
\[
\overline{F}^{(3)}(x_1,x_2,x_3)=\left(
1+\frac{x_1+x_2+x_3}{\sigma}\right)^{-\mu}
\left(
1+\frac{x_1+x_2}{\sigma}\right)^{-\mu}
\left(
1+\frac{x_3}{\sigma}
\right)^{-\mu},
\]
where $x_1,x_2,x_3$ are all in $\mathbf{R}_{+}$.
In this r.p., the Pearson correlation coefficient between r.c. \#1 and \#2 is $0.3$, and
it is equal to $0.09$ otherwise.
\end{itemize}
\subsection{Expected times of the first default} The left panel of
Figure \ref{fig:CTEmin}
depicts the values of $CTE_q[X_-]$ for $q\in[0,\ 1)$, $X_-\in\mathcal{X}$
and portfolios (1) to (3) as well as the reference portfolio, denoted by $(\perp)$.
{As the risk components are identically distributed, it is not difficult to see that
the following ordering holds
\begin{equation}
\label{stord1}
\overline{F}^{(1)}_-\geq_{st}\overline{F}^{(3)}_-\geq_{st}\overline{F}^{(2)}_-\geq_{st}
\overline{F}^{(\perp)}_-,
\end{equation}
where `$\geq_{st}$' denotes first order stochastic dominance (FSD). Furthermore,
since the CTE risk measure
is known to preserve the FSD ordering, we also have that
\[
CTE^{(1)}_q[X_-]\geq CTE^{(3)}_q[X_-]\geq CTE^{(2)}_q[X_-]\geq CTE^{(\perp)}_q[X_-]
\]
for all $q\in[0,\ 1)$ and $X_-\in\mathcal{X}$. This conforms with
Figure \ref{fig:CTEmin} (left panel), which hints that the r.p.'s with more
significantly correlated r.c.'s
enjoy higher, and thus more favourable, occurrence times of the first default.}
\begin{figure}[h!]
\centering
\includegraphics[width=7cm, height=7cm]{min-plot.eps}
\includegraphics[width=7cm, height=7cm]{max-plot.eps}\\
\caption{Conditional expected times of first (left panel) and last
(right panel) default for portfolios (1) - (3) and the
reference portfolio $(\perp)$ for `B' rating r.p.'s with the probability of
default $p=0.3198$ and
$\mu=1.67$.
Proposition \ref{CTEparMin} is employed to
compute the values of $CTE_q$ for $q\in[0,\ 1)$.}
\label{fig:CTEmin}
\end{figure}
The
downside of high correlations is elucidated in
Figure \ref{fig:minCase}, in which
we leave the probability of default $p$ to be equal to $0.3198$ (`B' rating),
but vary the $\mu$ parameter that stipulates the effect of the risk factors.
In this respect, we observe that the r.p.'s with stronger correlations between r.c.'s are more
sensitive to the changes in the $\mu$ parameter, and therefore such r.p.'s
must be monitored and stress-tested more frequently.
\begin{figure}[h!]
\centering
\includegraphics[width=7cm,height=7cm]{min-c1.eps}
\includegraphics[width=7cm,height=7cm]{min-c2.eps}\\
\includegraphics[width=7cm,height=7cm]{min-c3.eps}
\includegraphics[width=7cm,height=7cm]{min-ind.eps}
\caption{Conditional expected times of first default for portfolios (1) (top left panel),
(2) (top right panel), (3) (bottom left panel) and reference $(\perp)$ (bottom right panel)
with the parameter $\mu$ varying from $1.67$ to $1.15$ and the default probability
$p=0.3198$. Proposition \ref{CTEparMin} is employed to compute the values of
$CTE_q$ for $q\in[0,\ 1)$.}
\label{fig:minCase}
\end{figure}
\subsection{Expected times of the last default}
Figure \ref{fig:CTEmin} (right panel)
depicts the values of $CTE_q[X_+]$ for $q\in[0,\ 1),$ $X_+\in\mathcal{X}$ and portfolios
(1) to (3) as well as the reference portfolio $(\perp)$.
Evoking Theorem \ref{maxrv} along with (\ref{stord1}) results in
\[
\overline{F}^{(\perp)}_+\geq_{st}\overline{F}^{(2)}_+\geq_{st} \overline{F}^{(3)}_+
\geq_{st} \overline{F}^{(1)}_+,
\]
and hence
\[
CTE^{(\perp)}_q[X_+]\geq CTE^{(2)}_q[X_+]\geq CTE^{(3)}_q[X_+] \geq CTE^{(1)}_q[X_+]
\]
for all $q\in[0,\ 1)$ and $X_+\in\mathcal{X}$.
This conforms with the right panel of Figure \ref{fig:CTEmin}.
Unlike in the case of the first default,
we observe that if the time of the last default is
of interest and the distributions of the r.c.'s are fixed,
then assuming stronger correlations between r.c.'s yields a more
conservative assessment of the expected time of the last default.
}
\section{Conclusions}
\label{sec-con}
We have introduced and studied a new form of an absolutely continuous with respect to the
Lebesgue measure multivariate probability law with the univariate margins distributed
Pareto of the 2nd kind.
The genesis of our distribution
is threefold, i.e., it originates as the Laplace transform of a multivariate gamma distribution
with the dependence structure based on the additive form of the multivariate reduction
method, and it also admits variants of the multiplicative background risk model
as well as the minima-based common shock model. We have
meaningfully positioned the proposed multivariate Pareto distribution in the general context of
the current state of the art. We have proved and employed certain characteristic results to derive,
e.g., the (conditional/product) moments of the new multivariate Pareto as well as the
distributions of minima and maxima. Last but not least, we have developed
expressions for some tail-based risk measures of actuarial interest and elucidated our
findings with the help of a numerical example.
\section*{Acknowledgements}
We thank the anonymous referees and the Associate Editor, Prof. Montserrat Guill\' en, for valuable comments
and suggestions that improved the work significantly and resulted in a better
presentation of the material.
We are also grateful to Prof. Paul Embrechts and all participants of the ETHs
Series of Talks in Financial and Insurance Mathematics for feedback and insights.
Our research has been supported by the Natural Sciences and Engineering Research Council (NSERC) of Canada. Jianxi Su also acknowledges the financial support of the Government of Ontario and MITACS Canada via, respectively, the Ontario Graduate Scholarship program
and the Elevate Postdoctoral fellowship.
\section*{References}
\hangindent=\parindent\noindent \textsc{Arnold, B.C. (1983)} \textit{Pareto
Distributions.} International Cooperative Publishing House, Fairland.
\hangindent=\parindent\noindent \textsc{Asimit, V., Furman, E. and Vernic, R. (2010)}
On a multivariate Pareto distribution.
\textit{Insurance: Mathematics and Economics} \textbf{46}(2), 308 -- 316.
\hangindent=\parindent\noindent \textsc{Asimit, V., Vernic, R. and Zitikis, R. (2013)}
Evaluating risk measures and capital allocations based on multi-losses driven by a heavy-tailed background risk: The multivariate Pareto-II model.
\textit{Risk} \textbf{1}(1), 14 -- 33.
\hangindent=\parindent\noindent \textsc{Asimit, V., Vernic, R. and Zitikis, R. (2016)}
Background risk models and stepwise portfolio construction.
\textit{Methodology and Computing in Applied Probability}, in press.
\hangindent=\parindent\noindent \textsc{Balkema, A. and de Haan, L. (1974)}
Residual life time at great age. \textit{Annals of Probability} \textbf{2}(5),
792 -- 804.
\hangindent=\parindent\noindent \textsc{Benson, D.A., Schumer, R. and Meerschaert, M.M. (2007)}
Recurrence of extreme events with power-law interarrival times.
\textit{Geophysical Research Letters} \textbf{34}(16), 1 -- 5.
\hangindent=\parindent\noindent \textsc{Boucher, J.P., Denuit, M. and Guill\'
en, M. (2008)} Models of insurance claim counts with time dependence based
on generalization of Poisson and negative binomial distributions. \textit
Variance} \textbf{2}(1), 135 -- 162.
\hangindent=\parindent\noindent \textsc{Bowers, N.L., Gerber, H.U., Hickman, J.C.,
Jones, D.A. and Nesbitt, C.J. (1997)} \textit{Actuarial Mathematics.} Second edition.
Society of Actuaries, Schaumburg.
\hangindent=\parindent\noindent \textsc{Cebri\'{a}n, A.C., Denuit, M. and Lambert, P. (2003)}
Generalized Pareto fit to the Society of Actuaries large claims database.
\textit{North American Actuarial Journal}
\textbf{7}(3), 18 -- 36.
\hangindent=\parindent\noindent \textsc{Chavez-Demoulin, V., Embrechts, P. and Hofert, M. (2015)}
An extreme value approach for modeling operational risk losses depending on covariates.
\textit{Journal of Risk and Insurance}, in press.
\hangindent=\parindent\noindent \textsc{Cherian, K.C. (1941)}
A bivariate correlated gamma-type distribution function.
\textit{Journal of the Indian Mathematical Society}
\textbf{5}, 133 -- 144.
\hangindent=\parindent\noindent \textsc{Chiragiev, A. and Landsman, Z. (2009)}
Multivariate flexible Pareto model: Dependency structure, properties and
characterizations.
\textit{Statistics and Probability Letters} \textbf{79}(16), 1733 -- 1743.
\hangindent=\parindent\noindent \textsc{Choo, W. and de Jong, P. (2009)}
Loss reserving using loss aversion functions.
\textit{Insurance: Mathematics and Economics} \textbf{45}(2), 271 -- 277.
\hangindent=\parindent\noindent \textsc{Choo, W. and de Jong, P. (2010)}
Determining and allocating diversification benefits for a portfolio of risks.
\textit{ASTIN Bulletin} \textbf{40}(1), 257 -- 269.
\hangindent=\parindent\noindent \textsc{Embrechts, P., McNeil, A. and Straumann, D. (2002)}
Correlation and dependence in risk management: Properties and pitfalls. In: Dempster, M.
et al. (Eds), \textit{Risk Management: Value at Risk and Beyond}. Cambridge University Press,
Cambridge.
\hangindent=\parindent\noindent \textsc{Engelmann, B. and Rauhmeier, R. (2011)}
\textit{The Basel II Risk Parameters: Estimation, Validation, Stress Testing - with Applications to Loan Risk Management}. Springer, Berlin.
\hangindent=\parindent\noindent \textsc{Feller, W. (1966)}
\textit{An Introduction to Probability Theory and its Applications.}
John Wiley and Sons, New York.
\hangindent=\parindent\noindent \textsc{Franke, G., Schlesinger, H.
and Stapleton, R.C. (2006)}
Multiplicative background risk.
\textit{Management Science}
\textbf{52}(1), 146 -- 153.
\hangindent=\parindent\noindent \textsc{Furman, E. (2008)} On a multivariate
gamma distribution. \textit{Statistics and Probability Letters} \textbf{78}(15), 2353 -- 2360.
\hangindent=\parindent\noindent \textsc{Furman, E. and Landsman, Z. (2005)}
Risk capital decomposition for a multivariate dependent gamma portfolio.
\textit{Insurance: Mathematics and Economics} \textbf{37}(3), 635 -- 649.
\hangindent=\parindent\noindent \textsc{Furman, E. and Landsman, Z. (2010)}
Multivariate Tweedie distributions and some related capital-at-risk
analysis. \textit{Insurance: Mathematics and Economics}
\textbf{46}(2), 351 -- 361.
\hangindent=\parindent\noindent \textsc{Furman, E. and Zitikis, R. (2008a)}
Weighted premium calculation principles. \textit{Insurance: Mathematics and
Economics} \textbf{42}(1), 459 -- 465.
\hangindent=\parindent\noindent \textsc{Furman, E. and Zitikis, R. (2008b)}
Weighted risk capital allocations. \textit{Insurance: Mathematics and
Economics} \textbf{43}(2), 263 -- 269.
\hangindent=\parindent\noindent \textsc{Furman, E. and Zitikis, R. (2010)}
General Stein-type covariance decompositions with applications to
insurance and finance. \textit{ASTIN Bulletin} \textbf{40}(1), 369 -- 375.
\hangindent=\parindent\noindent \textsc{Gabaix, X., Gopikrishnan, P., Plerou, V. and Stanley, H.E. (2003)}
A theory of power-law distributions in financial market fluctuations.
\textit{Nature} \textbf{423}, 267 -- 270.
\hangindent=\parindent\noindent \textsc{Gollier, C. and Pratt, J.W. (1996)}
Weak proper risk aversion and the tempering effect of background risk.
\textit{Econometrica} \textbf{64}(5), 1109 -- 1123.
\hangindent=\parindent\noindent \textsc{Gradshteyn, I.S. and Ryzhik, I.M. (2007)}
\textit{Table of Integrals, Series and Products.} Seventh edition. Academic Press, New York.
\hangindent=\parindent\noindent \textsc{Koedijk, K.G., Schafgans, M.M.A. and de Vries, C.G. (1990)}
The tail index of exchange rate returns.
\textit{Journal of International Economics} \textbf{29}(1-2), 93 -- 108.
\hangindent=\parindent\noindent \textsc{Kotz, S., Balakrishnan, N. and Johnson,
N.L. (2000)} \textit{Continuous Multivariate Distributions.} Second edition. Wiley, New York.
\hangindent=\parindent\noindent \textsc{Longin, F.M. (1996)}
The asymptotic distribution of extreme stock market returns.
\textit{Journal of Business} \textbf{69}(3), 383 -- 408.
\hangindent=\parindent\noindent \textsc{Mathai, A.M. and Moschopoulos, P.G. (1991)}
On a multivariate gamma.
\textit{Journal of Multivariate Analysis} \textbf{39}(1), 135 -- 153.
\hangindent=\parindent\noindent \textsc{Mathai, A.M. and Moschopoulos, P.G. (1992)}
A form of multivariate gamma distribution.
\textit{Annals of the Institute of Statistical Mathematics}
\textbf{44}(1), 97 -- 106.
\hangindent=\parindent\noindent \textsc{Meyers, G.G. (2007)}
The common-shock model for correlated insurance losses.
\textit{Variance} \textbf{1}(1), 40 --52.
\hangindent=\parindent\noindent \textsc{Moschopoulos, P.G. (1985)}
The distribution of the sum of independent gamma random variables.
\textit{Annals of the Institute of Statistical Mathematics}
\textbf{37}, 541 -- 544.
\hangindent=\parindent\noindent \textsc{Pareto, V. (1897)} The new theories of economics. \textit{Journal of Political Economy} \textbf{5}(4), 485 -- 502.
\hangindent=\parindent\noindent \textsc{Pfeifer, D. and Ne\v slehov\' a, J. (2004)}
Modeling and generating dependent risk processes for IRM and DFA. \textit
ASTIN Bulletin} \textbf{34}(2), 333 -- 360.
\hangindent=\parindent\noindent \textsc{Pickands, J. (1975)} Statistical
inference using extreme order statistics. \textit{The Annals of Statistics}
\textbf{3}(1), 119 -- 131.
\hangindent=\parindent\noindent \textsc{Ramabhadran, V. (1951)}
A multivariate gamma-type distribution.
\textit{Journal of Multivariate Analysis}
\textbf{38}, 213 -- 232.
\hangindent=\parindent\noindent \textsc{Soprano, A., Crielaard, B., Piacenza, F. and Ruspantini, D. (2009)} \textit{Measuring Operational and Reputational Risk: A Practitioner's Approach.} Wiley, Chichester.
\hangindent=\parindent\noindent \textsc{Standard \& Poor's (2015)}
Default, transition and recovery:
2014 annual global corporate default study and rating transitions. Technical report, Standard and Poor's, New York.
\hangindent=\parindent\noindent \textsc{Tsanakas, A. (2008)} Risk
measurement in the presence of background risk. \textit{Insurance:
Mathematics and Economics} \textbf{42}(2), 520 -- 528.
\hangindent=\parindent\noindent \textsc{Vernic, R. (1997)} On the bivariate
generalized Poisson distribution. \textit{ASTIN Bulletin} \textbf{27}(1), 23
-- 32.
\hangindent=\parindent\noindent \textsc{Vernic, R. (2000)} A multivariate
generalization of the generalized Poisson distribution. \textit{ASTIN
Bulletin} \textbf{30}(1), 57 -- 67.
\hangindent=\parindent\noindent \textsc{Vernic, R. (2011)} Tail conditional
expectation for the multivariate
Pareto distribution of the second
kind: Another approach. \textit{Methodology and Computing in
Applied Probability} \textbf{13}(1), 121 -- 137.
\hangindent=\parindent\noindent \textsc{Wang, S. (1996)}
Premium calculation by transforming the layer premium density.
\textit{ASTIN Bulletin} \textbf{26}(1), 71 -- 92.
| ce4540ec93ca866b704e0d9207a87d8d512f2406 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Humanoid robots have been widely studied in the research of robotics.
With the recent development of motion capture devices such as RGB-D camera
(e.g., Kinect) and wearable sensor system (e.g., Xsens MVN), efforts have
been made to generate human-like motions for humanoid robots with high degree-of-freedoms.
However, directly applying captured poses of human to humanoids
is difficult because of the difference in human's and humanoid's kinematics. Therefore, a variety
of kinematics based approaches for humanoid imitation have been investigated, which can be classified into two
categories. Many of them perform an offline optimization step to compute the corresponding
configurations that conform to the mechanical structures and kinematics of humanoids from input
human data\cite{suleiman2008human, chalodhorn2007learning, nakaoka2007learning, ude2004programming,
kim2009stable, safonova2003optimizing}. It is obvious that the significant computational overhead
in those techniques prevents us from applying them to real-time imitation.
Methods in the other thread of research compute online imitation following captured human
motion\cite{ott2008motion, dariush2009online, do2008imitation,
yamane2010controlling, koenemann2012whole, koenemann2014real}.
In this paper, we consider about the problem of realizing real-time human-to-humanoid motion imitation.
Unfortunately, it is not an easy task due to:
\begin{itemize
\item full sampling of human-to-humanoid correspondence often leads to large data size;
\item high non-linearity of underlying mechanical rules results in significant computational cost;
\item how to find the configuration of a humanoid according to the input poses of human in real-time is not intuitive.
\end{itemize
Artificial neural networks have been adopted to ease the difficulties, with which a lot of
efforts have been made in simulation and for robots with small degree-of-freedoms\cite{morris1997finding,
aleotti2004position, neto2009accelerometer, neto2010high,
stanton2012teleoperation, van1993control, jung1996neural, larsen2004case, wang2005improving}.
A recent work\cite{stanton2012teleoperation} by
Stanton et al directly introduced neural networks with particle swarm optimization to find the
mapping between human movements and joint angle positions of humanoid. However, there is no measurement presented in their work to evaluate the quality of humanoid poses generated by the trained neural system. On the other aspect, our method is also different from this work in terms of the training data set. We use the sparse correspondence
instead of the densely recorded raw data, which can help eliminate the redundancy in data set and
improve the training speed. Moreover, only requiring a sparse set of correspondence samples leads to a lower barrier of system implementation.
We propose a framework that allows efficient projection of a pose from
human's space to the configuration space of humanoid based on sparsely sampled correspondence
extracted from recorded raw data, which can be used to realize motion imitation in real time (see Fig. \ref{figWorkflow}).
Experimental results show that our framework can be successfully used in the motion imitation of humanoid (see Fig.\ref{figTeleoperation} for an example of tele-operation using a NAO humanoid).
\begin{figure}\centering
\includegraphics[width=\linewidth]{Teleoperation}
\caption{An example of humanoid imitation realized by our framework.}
\label{figTeleoperation}
\end{figure}
\begin{figure*} \centering
\captionsetup{justification=centering}
\includegraphics[width=\linewidth]{Workflow}
\caption{An illustration of our framework for motion imitation using configuration projection.}
\label{figWorkflow}
\end{figure*}
\section{Framework of Configuration Projection} \label{secFramework}
\subsection{Problem Definition} \label{subsecProblem}
A human pose can be uniquely represented as a point~(abbreviated as $C$-point) $\mathbf{h} \in \mathbb{R}^m$
in the configuration space (abbreviated as $C$-space -- $\mathcal{H}$) of human's motion and its corresponding pose of humanoid can be denoted as a point $\mathbf{r} \in \mathbb{R}^n$ in the $C$-space of humanoid -- $\mathcal{R}$. We assume one-to-one correspondence between the poses of human body and humanoid,
i.e. the mapping between human and humanoid's $C$-spaces is bijective. A pair of
human's and humanoid's configurations is denoted as $(\mathbf{h}, \mathbf{r}) \in \mathbb{R}^{m+n}$.
Given stored correspondence pairs $\{(\mathbf{h}, \mathbf{r})\}$ as the known knowledge and a new input pose $\mathbf{h}^*\in \mathbb{R}^m$, the configuration projection $\Omega(\cdot)$ can be
defined as finding a corresponding $\mathbf{r}^* \in \mathbb{R}^n$ that satisfies two basic properties:
\begin{itemize}
\item \textbf{Identity} -- for any sample pair $(\mathbf{h}_i, \mathbf{r}_i)$ in the data-set, it should have
\begin{center}
$\Omega(\mathbf{h}_i) = \mathbf{r}_i$.
\end{center}
\item \textbf{Similarity} -- for an input $C$-point of human $\mathbf{h}^*$, if
\begin{displaymath}
\max \{ \min_i \|\mathbf{h}_i - \mathbf{h}^*\| \} < \delta
\end{displaymath}
then it should have
\begin{displaymath}
\|\Omega(\mathbf{h}^*) - \tilde{\mathbf{r}}(\mathbf{h}^*)\| < \epsilon,
\end{displaymath}
where $\delta$ and $\epsilon$ are two constant values, and $\tilde{\mathbf{r}}(\mathbf{h}^*)$ is a $C$-point of humanoid that
can be obtained by more accurate but computational intensive methods (e.g., inverse kinematics) as the ground truth.
\end{itemize}
All sample pairs should be repeated with the projection $\Omega(\cdot)$ according to the property of \textit{identity}.
The demand on \textit{similarity} indicates that
if a new input is close to the known samples,
its projected result should not deviate too much from its corresponding ground truth.
The main difficulty of finding the projection $\mathbf{r}^*$ lies in the lack of explicit functions
to determine the mapping between two $C$-spaces with different dimensions (i.e., degree-of-freedoms).
Given sparsely aligned pairs of poses as samples,
we try to solve this problem by proposing a strategy of kernel-based projection to
find a good approximation for $\mathbf{r}^*$.
\subsection{Data Pre-processing}\label{subsecPreprocessing}
The knowledge of correspondence $\{(\mathbf{h}, \mathbf{r})\}$ can be established through experiments. Although aligning a pose of human body with a corresponding pose of humanoid can be taken manually, it is a task almost impossible if thousands of such correspondence samples need to be specified. Therefore, in our experiments, we first capture continuous motions of human bodies by using a motion capture system. The data-set obtained in this way often results in
large size and redundancy. To resolve this problem, we perform a pre-processing step to extract marker poses from the raw data-set recorded from human's motion. Specifically, mean shift clustering~\cite{comaniciu2002meanshift} is employed to
generate the marker set denoted as $\mathcal{H}$. For each sample $\hat{\mathbf{h}}\in \mathcal{H}$,
its corresponding pose $\hat{\mathbf{r}}$ in the configuration space of humanoid can be either specified manually (when the number of samples in $\mathcal{H}$ is small) or generated automatically by a sophisticated method (e.g., the inverse kinematics methods). The pairs of correspondence, $\{(\hat{\mathbf{h}}_i,\hat{\mathbf{r}}_i) \}_{i=1,\cdots,N}$, extracted in this way is treated as landmarks to be used in our framework.
\subsection{ELM Based Kernels}
As the configuration pairs of marker data-set are discrete in space, we define a kernel $\kappa(\cdot)$
on each marker configuration $\hat{\mathbf{h}}_i$ and $\hat{\mathbf{r}}_i$
as a local spatial descriptor using the technique of \textit{Extreme Learning Machine} (ELM) \cite{huang2011extreme}.
ELM method has been widely used in regression and classification
problems as a \textit{single hidden layer feed-forward network} (SLFN) with its advantageous properties of fast training
speed, tuning-free neurons and easiness in implementation (ref.~\cite{huang2012extreme}).
Basically, the training formula of ELM can be expressed as $\mathbf{H} \mathbf{b} = \mathbf{T}$,
where $\mathbf{H}$ is the hidden layer output matrix of SLFN, $\mathbf{b}$ is the output weight
vector to be computed, and $\mathbf{T}$ is the target feature vector.
Given a new input $\mathbf{x}$,
the prediction function of ELM is $\mathbf{f}(\mathbf{x}) = \mathbf{Q}(\mathbf{x})\mathbf{b}$,
where the $\mathbf{Q}(\mathbf{x})$ is the hidden layer feature mapping of $\mathbf{x}$. It has been pointed
out in~\cite{huang2011extreme} that the training errors will be eliminated if the number of hidden nodes is
not less than the number of training samples, indicating the trained ELM can be used as a fitting function that
interpolates all training samples
\begin{displaymath}
\mathbf{Q}(\hat{\mathbf{h}}_i)\mathbf{b} = \hat{\mathbf{r}}_i, ~~ (i = 1, \cdots, N).
\end{displaymath}
In this case, the output weight vector is computed as
\begin{displaymath}
\mathbf{b} = \mathbf{H}^{T}(\mathbf{H}\mathbf{H}^{T})^{-1}\mathbf{T},
\end{displaymath}
where $\mathbf{H}^{T}(\mathbf{H}\mathbf{H}^{T})^{-1}$ is the Moore-Penrose generalized inverse of
$\mathbf{H}$. Regularized ELM is proposed in\cite{deng2009regularized} to improve its numerical
stability, leading to the following training formula with $\lambda$ (a very small value in practice)
as the regularization factor
\begin{displaymath}
\mathbf{b} = \mathbf{H}^{T}(\lambda + \mathbf{H}\mathbf{H}^{T})^{-1}\mathbf{T}.
\end{displaymath}
With the help of ELM, a kernel $\kappa^h_{i}(\cdot) \in \mathbb{R}^n$ for a human's landmark point $\hat{\mathbf{h}}_i$ can be built with its nearest neighbors.
Specifically, we find $k$ spatial nearest neighbors of $\hat{\mathbf{h}}_i$ in the set of human's landmarks as
$\{ \hat{\mathbf{h}}_j \}_{j \in \mathcal{N}(\hat{\mathbf{h}}_i)}$, where $\mathcal{N}(\cdot)$ denotes the set of nearest neighbors. Then, the ELM kernel of $\kappa^h_{i}(\cdot)$ is trained using $\{ ( \hat{\mathbf{h}}_j, \hat{\mathbf{r}}_j ) \}_{j \in \mathcal{N}(\hat{\mathbf{h}}_i)}$, which is regarded as an approximate local descriptor of the nearby mapping of $\hat{\mathbf{h}}_i$: $\mathcal{H} \mapsto \mathcal{R}$. When inputting a new human pose $\mathbf{h}^* \in \mathbb{R}^m$, a local estimation of mapping with reference to this kernel can be represented as
\begin{displaymath}
\kappa^h_{i}(\mathbf{h}^*)=\mathbf{Q}(\mathbf{h}^*) \mathbf{b}.
\end{displaymath}
This function is called a \textit{forward} kernel.
Similarly, for each $C$-point ${r^m}_i$ of a humanoid, an ELM based kernel $\kappa^r_{i}(\cdot) \in \mathbb{R}^m$ can be constructed in the same way for the inverse mapping: $\mathcal{R} \mapsto \mathcal{H}$. $\kappa^r_{i}(\cdot)$ is called a \textit{backward} kernel. These two types of kernel functions will be used in our framework for realizing the projection.
\subsection{Projection}\label{subsecFramework}
For an input pose $\mathbf{h}^* \in \mathbb{R}^m$, the point determined by the ELM kernel function, $\kappa^h_{i}(\mathbf{h}^*)$, is not guaranteed to satisfy the requirement of bijective mapping (i.e., $\kappa^r_{i}(\kappa^h_{i}(\mathbf{h}^*)) \neq \mathbf{h}^*$). To improve the bijection of mapping, the projection of a human's $C$-point is formulated as determining an optimal point from all candidates generated from different forward kernels.
\begin{figure}[t] \centering
\includegraphics[width=\linewidth]{Framework}
\caption{An illustration of finding an optimal point that minimizes a back-projected deviation (with $L=M=4$).
}
\label{figFramework}
\end{figure}
First of all, $L$ nearest neighbors of $\mathbf{h}^*$ are retrieved in $\mathcal{H}$ as $\{\hat{\mathbf{h}}_j\}$ ($j=1,\cdots,L$). From the forward kernel associated with each of these $L$ points in $\mathcal{H}$, a candidate point in $\mathcal{R}$ can be determined by $\mathbf{r}^c_j=\kappa^h_j(\mathbf{h}^*)$. For each $\mathbf{r}^c_j$, we search for its $M$ nearest neighbors in $\mathcal{R}$ as $\mathcal{N}(\mathbf{r}^c_j)=\{\hat{\mathbf{r}}_{j,k}\}$ ($k=1,\cdots,M$). In other words, there are $M$ backward kernels associated with $\mathbf{r}^c_j$, which are $\{ \kappa^r_{j,k}\}$. In each cluster of backward kernels, we determine a set of weights $w_{j,k}$ that leads to a point formed as the convex combination of $\{\hat{\mathbf{r}}_{j,k}\}$
\begin{displaymath}
\tilde{\mathbf{r}}^c_j = \sum_{k} w_{j,k} \mathbf{r}^c_{j,k}.
\end{displaymath}
An optimal point $\tilde{\mathbf{r}}^c_j$ minimizes the deviation of back-projection with regard to the cluster of kernels $\{\kappa^r_{j,k}(\cdot)\}_k$ is defined as
\begin{equation} \label{eqSubFunction}
\begin{aligned}
&\min_{w_{j,k}} \{ \| \kappa^r_{j,k} (\sum_{k} w_{j,k} \mathbf{r}^c_{j,k}) - \mathbf{h}^* \| \}_k, \\
s.t. \quad & \sum_{k=1}^M w_{j,k} = 1, ~~ w_{j,k} \geq 0.
\end{aligned}
\end{equation}
The final projected point $\mathbf{r}^*$ is then defined as
\begin{equation} \label{eqCoreFunction}
\mathbf{r}^* = \sum_{k} w_{l,k} \mathbf{r}^c_{l,k}
\end{equation}
according to the cluster of $\mathcal{N}(\mathbf{r}^c_l)$ that gives the minimal back-projected deviation, which is a solution of
\begin{equation}\begin{aligned}\label{eqCoreFunction2}
& \min_j \left\{ \min_{w_{j,k}} \{ \| \kappa^r_{j,k} (\sum_{k} w_{j,k}\mathbf{r}^c_{j,k}) - \mathbf{h}^* \| \}_k \right\}, \\
s.t. \quad & \sum_{k=1}^M w_{j,k} = 1, ~~ w_{j,k} \geq 0.
\end{aligned}\end{equation}
The computation for solving above optimization problem can be slow in many cases. Therefore, we propose a sub-optimal objective function as a relaxation of Eq.(\ref{eqCoreFunction2}) to be used in real-time applications (e.g., the tele-operation shown in Fig.\ref{figTeleoperation}). The problem is relaxed to
\begin{equation}
\min_j \left\{ \min_{k} \{ \| \kappa^r_{j,k} (\mathbf{r}^c_{j}) - \mathbf{h}^* \| \}_k \right\},
\end{equation}
the solution of which can be acquired very efficiently by checking each candidate $\mathbf{r}^c_j$ with
regard to all its $M$ reference backward kernels. Figure \ref{figFramework} gives an illustration for the evaluation of back-projected deviation.
\vspace{5pt} \noindent \textbf{Motion Smoothing:} A dynamic motion is processed as a sequence of continuous poses
in our system, where the projected poses in the configuration space of humanoid are generated separately. To avoid the generation of jerky motion, we use a method modified from the double exponential smoothing~\cite{laviola2003double} to post-process the projected poses. Given a projected pose $\mathbf{r}_t$ at time frame $t$, the update rules of a smoothed pose $\mathbf{s}_t$ are defined as
\begin{equation}
\begin{aligned}
& \mathbf{s}_t = \alpha y_t + (1-\alpha)(\mathbf{s}_{t-1} + \mathbf{b}_{t-1}), ~~ 0 \leq \alpha \leq 1 \\
& \mathbf{b}_t = \gamma (\mathbf{s}_t - \mathbf{s}_{t-1}) + (1 - \gamma) \mathbf{b}_{t-1}, ~~ 0 \leq \gamma \leq 1 \\
& \mathbf{s}_t = \mathbf{s}_{t-1}, ~\text{if} ~\|\mathbf{s}_t - \mathbf{s}_{t-1}\| < \eta
\end{aligned}
\end{equation}
$\alpha$, $\gamma$ and $\eta$ are parameters to control the effectiveness of smoothing, where $\alpha=0.75$, $\gamma=0.3$ and $\eta=0.15$ are used to give satisfactory results in our practice.
\vspace{5pt} \noindent \textbf{Remark:} It must be clarified the \textit{Identity} property introduced in Section
\ref{subsecProblem} is relaxed to $\Omega(h_i) \approx r_i$ in practice due to the following reasons:
\begin{itemize}
\item Regularized ELM method is employed to construct the kernels, which changes the corresponding
energy function where a regularization term is added to improve its numerical stability.
\item Double exponential smoothing is applied for smoothing a motion, which introduces minor adjustments
on the output values.
\end{itemize}
\begin{figure} \centering
\includegraphics[width=0.9\linewidth]{Skeleton}
\caption{Feature vectors of human and humanoid: (a) the human skeleton from a Kinect sensor,
(b) the corresponding pose descriptor of a human body consists of $19$ unit vectors, and
(c) the pose descriptor for a NAO humanoid formed by all DOFs on its joints (source: http//:www.ez-robot.com).}
\label{figSkeleton}
\end{figure}
\section{Real-time Projection on NAO} \label{secReconstruction}
Our framework is testified on real-time motion imitation of a NAO humanoid robot with a Kinect RGB-D camera as the device to capture the motion of human.
\subsection{Human-to-humanoid Motion Imitation}
The human skeleton provided by a Kinect sensor is a set of line segments based on predefined key joints
as shown in Fig.\ref{figSkeleton}(a).
We define an abstraction consisting of $19$ unit vectors for a pose
as illustrated in Fig.\ref{figSkeleton}(b), which is independent different body dimensions.
It should be pointed out that it is unnecessary to always use the full set of unit vectors unless full
body motion must be sensed. The NAO humanoid robot has 26 degree-of-freedoms, including the roll,
pitch, and yaw of all its joints (see Fig.\ref{figSkeleton}(c)).
Posing a NAO humanoid can be executed by specifying the values of all its degree-of-freedoms.
To collect the data-set of human's motion, a user is asked to do arbitrary motion in front of a Kinect camera.
Meanwhile, we have implemented a straightforward \textit{inverse kinematics} (IK) based scheme for upper-body motion.
The roll, pitch, and yaw of every joint can be computed directly by the unit vectors of a human's skeleton model.
After using mean shift to extract the landmarks of motion from the raw set, their corresponding landmark poses in the $C$-space of humanoid can be generated by this IK. Besides, we also define \textit{eight} basic poses (see Fig.\ref{figBasicPoses}) which play a critical role when evaluating the similarity between the projected poses of humanoid and the poses of human.
\begin{figure}[t]\centering
\includegraphics[width=\linewidth]{BasicPoses}
\caption{Basic poses serve as benchmarks for similarity evaluation.}\label{figBasicPoses}
\end{figure}
Using the landmark poses defined in this way, human-to-humanoid motion imitation has been implemented by a single-core C++ program. All the tests below are taken on a personal computer with Intel Core i7-3770 3.4 GHz and 8 GB RAM memory.
\subsection{Experimental Results}
We evaluate our method mainly from three perspectives, including the computational
efficiency of projection, the quality of reconstructed motion, and the influence by the size
of landmark set.
\vspace{5pt}\noindent \textbf{Efficiency of Projection:} From Section \ref{subsecFramework}, we know that the complexity for computing projection depends on the size of neighbors (i.e., $L$ and $M$).
The cost of computation increases with larger $L$ and $M$ as more candidates and more reference
kernels will be involved. In all our experiments, we use $L = M = 10$ and the average time for
making a configuration projection is $0.00273ms$. When increasing to $L = M = 50$, the average time cost
is still only $0.0201ms$. In summary, the overhead of our method for motion imitation is very light -- i.e., it fits well for different real-time applications.
\begin{figure*} \centering
\includegraphics[width=\linewidth]{PoseReconstruct}
\caption{Eight basic poses are reconstructed by our method (left of each pair) and compared with the ground truth (right of each pair). The similarity metrics, $M_{max}$ and $M_{avg}$, of each pair are also reported. The evaluation is taken on a projection defined by using $1,644$ landmark pairs.}\label{figPoseReconstruct}
\end{figure*}
\begin{figure*} \centering
\includegraphics[width=\linewidth]{Statistics}
\caption{Statistics in eight motions for the change of two metrics: $M_{max}$ (blue) and $M_{avg}$ (red). The evaluation is also taken on a projection with $1,644$ landmark pairs.}
\label{figStatistics}
\end{figure*}
\vspace{5pt}\noindent \textbf{Quality of Reconstruction:} Two metrics are used in our experiments to estimate the quality of a projected configuration $\mathbf{r}^* \in \mathbb{R}^{n}$ referring to its corresponding ground truth value $\mathbf{r}_{gt}$ -- the maximum absolute deviation in degree as
\begin{displaymath}
M_{max} = \frac{180^{\circ}}{\pi} \|r^* - r_{gt}\|_{\infty},
\end{displaymath}
and the average absolute deviation in degree as
\begin{displaymath}
M_{avg} = \frac{1}{n} \left( \frac{180^{\circ}}{\pi}\|r^* - r_{gt}\|_1 \right).
\end{displaymath}
The evaluation is taken with a set holding $1,644$ configuration pairs as landmarks. All those eight poses shown in Fig.\ref{figBasicPoses} are tested, and the results are shown in Fig.\ref{figPoseReconstruct}. The results of comparison (in terms of $M_{max}$ and $M_{avg}$) indicates that the poses generated by our method share good similarity with the ground truths. Besides of static poses, we also evaluate the quality of reconstructed motion in the $C$-space of humanoid as a sequence of poses. We define eight basic motion sequences, each of which starts from the rest pose and ends at one of the basic poses. The complete human motions are recorded for the reconstruction using our projection in the $C$-space of humanoid. The projected poses are compared with the poses generated by IK, serving as the ground truths. The values of $M_{max}$ and $M_{avg}$ in these eight motions are shown in Fig.\ref{figStatistics}. It is easy to find that the errors are bounded to less than $10^{\circ}$ in all motions.
\begin{figure}[t] \centering
\includegraphics[width=\linewidth]{Size}
\caption{To reconstruct motion using landmark sets having different number of corresponding samples, statistics indicate that more landmark pairs lead to better results.}\label{figSize}
\end{figure}
\vspace{5pt}\noindent \textbf{Size of Landmarks:} As presented in Section \ref{subsecPreprocessing}, the correspondence samples used to formulate projection in our framework is extracted from the captured motions. In our implementation, it is generated by a user moving in front of a Kinect sensor for 5 minutes. Then, three sets with different number of landmarks ($1,644$, $961$, and $86$ respectively) are extracted. The corresponding pairs of poses are then constructed with the help of IK. The 8-th pose in Fig.\ref{figBasicPoses} -- POSE\_8 and the motion from the rest pose to POSE\_8 are constructed from the projections defined on the sets with different number of landmarks. From the statistics and comparisons shown in Fig.\ref{figSize}, it is easy to conclude that our projection based formulation converges when the number of landmarks increases. In other words, more landmarks result in a more accurate projection. However, it should also be noted that the projection from the smaller set may still be useful in some applications with low requirement on quality but having more restrictions on speed and memory usage.
\subsection{Application of tele-operation}
We have tested the motion imitation realized by our method in an application of tele-operation using a NAO humanoid.
As illustrated in Fig.\ref{figTeleoperation} and the supplementary video of this paper, a user can remotely control the motion of a NAO robot to grasp an object and put it into a box. The scene that can be seen from the camera of NAO is displayed on a screen placed in front of the user as the visual feedback. The imitation realized by our system has good accuracy. As a result, the tele-operation can be performed very smoothly.
\section{Conclusion \& Future Work} \label{secConclusion}
In this paper, we have proposed a framework to realize motion imitation. Different from conventional methods, our method is based on a novel formulation of projection between two configuration spaces with different dimensions. Given a new input pose of human, its projection in the space of humanoid is defined as finding the optimal $C$-point that minimizes a back-projection deviation referring to the built kernels. We have validated our idea by reconstructing humanoid motion on a NAO robot. The experimental results are encouraging and motions of good quality can be reconstructed efficiently.
There are several potential improvements can be made to our method.
First, an intuitive improvement is to extend the current setup to full-body motion reconstruction by
incorporating the constraint of whole-body balance.
Second, the ELM based kernels currently used in our framework do not have a explicit bound for prediction
with a new input. Finding kernel functions that can provide a bound on prediction could be another future work.
Lastly, we are interested in exploring more applications beyond tele-operation.
\bibliographystyle{IEEEtran}
| d0a6c9d3b3d99f1753e95a7ad2f148b53981801a | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Structure-from-Motion targets at recovering 3D structure and camera motion from monocular 2D feature tracks. Conventional SFM primarily concerns with the 3D reconstruction of a single rigidly moving object seen by a static camera, or a static and rigid scene observed by a moving camera --in both cases there are only one relative rigid motion involved. SFM has been extended to the areas of {\em multi-body} SFM, as well as {\em single body} {non-rigid SFM} (see Table.\ref{tab:summary} for a classification of different SFM problems).
Since, NRSFM is central to many computer vision applications such as motion capture, activity recognition, human-computer interaction, and dynamic scene reconstruction, etc. Therefore, it has been extensively researched (e.g. in \cite{Bregler:CVPR-2000}\cite{Xiao-Chai-Kanade:ECCV-2004}\cite{Akhter-Sheikh-Khan-Kanade:Trajectory-Space-2010}\cite{Dai-Li-He:CVPR-2012}\cite{Dense-NRSFM:CVPR-2013}\cite{PND-Mixture-Model:IJCV-2016}) in the past.
So far, all existing methods for NRSFM have implicitly assumed that there is only one deformable shape or object in the view. However, the real world scenarios can be much more complicated involving multiple, independently deforming objects in the scene. Multiple nonrigid objects are commonly encountered in our daily lives, for example, in motion capture, multiple persons perform different activities with possible interactions (see Fig.~\ref{fig:multiple_nonrigid} for an example); in human-computer interaction, different users may conduct different gesture commands; and in traffic scene, multiple vehicles and walking pedestrians create {\bf multi-body non-rigid deformations}.
To handle such multiple non-rigid deformations in 3D reconstruction, a natural idea would be to simply treat the multiple non-rigid deformations as a single (though more complex) non-rigid deformation (with higher order or higher rank), and then apply any state-of-the-art non-rigid structure-from-motion methods such as \cite{Dai-Li-He:CVPR-2012}\cite{Procrustean-Normal-Distribution:CVPR-2013}. However, by this idea, the inherent structure in the problem has not been exploited, which may hinder the success of 3D reconstruction. Even if the method succeeds in obtaining 3D reconstruction, it cannot tell meaningful segmentation of the multiple non-rigid objects. Another choice would be conducting the tasks of non-rigid motion segmentation \cite{SSC:PAMI-2013} and non-rigid 3D reconstruction \cite{Dai-Li-He:CVPR-2012} successively. However, in this way, the solution of each sub-task does not benefit from the solution of the other sub-task. For example, non-rigid motion segmentation provides critical information to correct non-rigid 3D reconstruction while non-rigid 3D reconstruction constrains the corresponding multi-body motion segmentation. Therefore, we would like to emphasize that since non-rigid deformation originally occurs in 3D space, it's more intuitive to perform non-rigid motion segmentation and reconstruction simultaneously in 3D space rather than solving this problem using two step process.
{\small \begin{table*}[h!]
\caption{A classification of different SFM problems defined by the number of objects and the rigidity of each object. This paper aims to fill in the currently missing work of {\bf Multi-body Non-rigid SFM} shown in blue.\label{tab:summary}}\vspace{-0.15in}
\center
\resizebox{1.0\linewidth}{!} {\small
\begin{tabular}{|l||c|c|}
\hline
& Single body & Multi-body\\\hline\hline Rigid & \begin{tabular}{@{}c@{}} Single-body Rigid SFM \cite{Tomasi:IJCV-1992} \\ $\m W_{2\m F\times \m P} = \m R_{2\m F\times 3} \m S_{3\times \m P}, \mbox{\rm rank}(\m S) = 3$ \end{tabular} & \begin{tabular}{@{}c@{}} Multi-body Rigid SFM \cite{Multi-body-Rigid:IJCV-1998}\cite{Yan-Pollefeys:PAMI-2008} \\ $\m W_{2\m F\times \m P} = \m R_{2\m F\times 3\m F} \m S_{3\m F\times \m P}, \mbox{\rm rank}(\m S) = 3\m K $ \end{tabular} \\ \hline
Non-rigid & \begin{tabular}{@{}c@{}} Single-body Non-rigid SFM \cite{Bregler:CVPR-2000} \\ $\m W_{2\m F\times \m P} = \m R_{2\m F\times 3\m F} \m S_{3\m F\times \m P}, \mbox{\rm rank}(\m S) = 3\m K$ \end{tabular} & \begin{tabular}{@{}c@{}} \textcolor{blue}{Multi-body Non-Rigid SFM}\\ \textcolor{blue}{$\m W_{2\m F\times \m P} = \m R_{2\m F\times 3\m F} \m S_{3\m F\times \m P}, \m S = \m S \m C$. i.e, } \\ \textcolor{blue}{3D trajectory should lie in union of linear/affine subspace.} \end{tabular} \\ \hline
\end{tabular}
}
\label{tab:multi-scal}
\end{table*}}
\begin{figure}[!htp]
\begin{center}
\subfigure[\label{fig:dance_yoga_2D_1}]{\includegraphics[width=0.24\linewidth]{Figures/Image1.png}}
\subfigure[\label{fig:dance_yoga_2D_2}]{\includegraphics[width=0.24\linewidth]{Figures/Image2.png}}
\subfigure[\label{fig:dance_yoga_3D_1}]{\includegraphics[width=0.24\linewidth]{Figures/3DImage1.png}}
\subfigure[\label{fig:dance_yoga_3D_2}]{\includegraphics[width=0.24\linewidth]{Figures/3DImage2.png}}
\end{center}
\caption{\small Illustration of complex multi-body non-rigid 3D reconstruction, where two persons perform different activities (Dancing and Yoga \cite{Akhter-Sheikh-Khan-Kanade:Trajectory-Space-2010}). Given the multi-frame 2D feature tracks as input, our method simultaneously outputs the 3D non-rigid reconstruction and the segmentation of the track trajectory in 3D. (a) First frame of the 2D track; (b) Second frame of the 2D track; (c) 3D reconstruction and segmentation of (a) using our method, where different colors index shows the corresponding segmentation; (d) Similarly, 3D reconstruction and segmentation of (b). }
\noindent\makebox[\linewidth]{\rule{0.83\paperwidth}{0.4pt}}
\label{fig:multiple_nonrigid}
\end{figure}
\begin{figure}
\centering
\subfigure[\label{fig:multi_object3}] {\includegraphics[width=0.40\textwidth]{Figures/recons_seg_multi_class.png}}
\subfigure[\label{fig:diag_multi3}] {\includegraphics[width=0.40\textwidth]{Figures/diagonal_multi_class.png}}
\caption{(a) Three subjects are performing actions such as stretch(red), dance(cyan) and yoga(green). Our approach is able to reconstruct and segment each action faithfully with reconstruction error of 0.0413 and 0 segmentation error. Here, different color corresponds to segmentation with dark and light color circles for each subject shows ground-truth and reconstructed 3D coordinates respectively. (b) Obtained block diagonal matrix.}
\noindent\makebox[\linewidth]{\rule{0.83\paperwidth}{0.4pt}}
\label{fig:result_3object}
\end{figure}
To demonstrate the advantage of the concurrent procedure as shown in Fig. \ref{fig:flow_diagram} and Table \ref{tab:summary}, this paper introduces an approach to perform non-rigid 3D reconstruction and its motion segmentation simultaneously. Specifically, we represent the multiple non-rigid motion as union of 3D trajectory subspaces \footnote{Zhu \etal \cite{Union_Subspaces:CVPR-2014} used the union of subspaces representation (different non-rigid deformations lie in different subspaces), where the subspace is defined in shape space contrast to our trajectory space. As we will show later, this difference provides uniqueness of our formulation in dealing with multiple non-rigid deformation and in dealing with dense case.}. By using the self-expressiveness model in representing multiple linear/affine subspace, where each 3D trajectory can be expressed with other trajectories in the same subspace only, enables us in compact representation of trajectories. In this way, we are able to exploit the inherent grouping structure in \emph{3D trajectory space}. For dense non-rigid reconstruction, we could further enforce the spatial coherence constraint. By contrast to existing methods, this endows us the following benefits:
\begin{enumerate}[(i)]
\item A compact representation for component non-rigid deformation in 3D trajectory space;
\item Joint multiple deformable objects motion segmentation and 3D non-rigid structure reconstruction;
\item Improved spatial regularity in 3D non-rigid dense reconstruction (in contrast, a hard segmentation of the 2D tracks may result in discontinuity at the segmentation boundary). \end{enumerate}
The readers are invited to have a preview of results obtained by our method, as shown in Fig.~\ref{fig:multiple_nonrigid}, where we show a daily life motion capture of two person performing dancing and yoga individually as in Fig.~\ref{fig:dance_yoga_2D_1} and Fig.~\ref{fig:dance_yoga_2D_2}. Similarly, in Fig.~\ref{fig:result_3object} three different person are performing stretch, dance and yoga. By applying our proposed method, we are able to achieve both non-rigid 3D reconstruction and non-rigid motion segmentation as shown in Fig.~\ref{fig:dance_yoga_3D_1}, Fig.~\ref{fig:dance_yoga_3D_2} and Fig.~\ref{fig:multi_object3}.
\paragraph{Contributions:} (1) This paper is first to model multiple non-rigid deformations as points in the union of multiple affine 3D trajectory subspace. This enables us to jointly solve the non-rigid reconstruction and non-rigid motion segmentation problems in 3D trajectory space; (2) Our formulation can handle both sparse and dense multiple non-rigid reconstruction problems uniformly; (3) We propose an efficient optimization procedure based on ADMM.
\section{Related Work}
Ever since the seminal work by Bregler \etal \cite{Bregler:CVPR-2000} modeling a non-rigid shape as lying in a ``\emph{shape space}'' (a linear combination of basis shapes), considerable progress has been made in the area of non-rigid 3D reconstruction. In 2004, Xiao \etal \cite{Xiao-Chai-Kanade:ECCV-2004} showed the inherent ambiguity in modeling non-rigid shape and proposed a remedy of ``basis constraints'' to derive a closed-form solution. In 2008, Akhter \etal \cite{Akhter-Sheikh-Khan-Kanade:Trajectory-Space-2010} presented a dual approach by modeling 3D trajectories, \ie ``\emph{trajectory space}''. In 2009, Akhter \etal \cite{Defence-orthonormality:CVPR-2009} proved that even there is an ambiguity in shape bases or trajectory bases, non-rigid shapes can still be solved uniquely without any ambiguity. Based on this, in 2012, Dai \etal \cite{Dai-Li-He:CVPR-2012} proposed a ``prior-free'' method to recover camera motion and 3D non-rigid deformation by exploiting the low rank constraint only. Besides shape basis model and trajectory basis model, the shape-trajectory approach \cite{Complementary-rank-3:CVPR-2011} combines two models and formulates the problems as revealing trajectory of the shape basis coefficients. Besides linear combination model, Lee \etal \cite{Procrustean-Normal-Distribution:CVPR-2013} proposed a Procrustean Normal Distribution (PND) model, where 3D shapes are aligned and fit into a normal distribution. Simon \etal \cite{Spatiotemporal-Priors:ECCV-2014} proposed to exploit the Kronecker pattern in the shape-trajectory (spatial-temporal) priors. Zhu and Lucey \cite{Convolutional-Sparse-Coding-Trajectory:PAMI-2015} applied the convolutional sparse coding technique to NRSFM using point trajectories. However, the method requires to learn an over-complete basis of 3D trajectories, prior to performing 3D reconstruction.
Despite of the above success, NRSFM is still far behind its rigid counterpart. This is mainly due to difficulty in modeling real world non-rigid deformation. Real world non-rigid reconstruction generally requires the ability to handle long-term, complex and dense non-rigid shape variations. Such complex and dense non-rigid motion not only increases the computational complexity but also adds difficulty in modeling various kinds of different motions. Zhu \etal \cite{Union_Subspaces:CVPR-2014} proposed to represent long-term complex non-rigid motion as lying in a union of shape sub-spaces rather than sum of sub-spaces. Cho \etal \cite{PND-Mixture-Model:IJCV-2016} represented complex shape variations probabilistically by a mixture of primitive shape variations.
By contrast to the above methods dealing with sparse NRSFM, dense NRSFM methods such as \cite{Dense-NRSFM:3DIMPVT-2012}\cite{Dense-NRSFM:CVPR-2013}\cite{Video-Registration:IJCV-2013}\cite{Video-Pop-Up:ECCV-2014} aim at achieving 3D reconstruction for each pixel in a video sequence, where spatial constraint has been widely exploited to regularize the problem. Garg \etal \cite{Dense-NRSFM:CVPR-2013} presented a variational formulation to dense non-rigid reconstruction by exploiting the spatial smoothness in 3D shapes, which in principle deals with single non-rigid deformation in contrast to our multiple non-rigid deformations. Fragkiadaki \etal \cite{Grouping-Trajectory-Reconstuction:NIPS-2014} solved the problem in sequel, namely, video segmentation by multi-scale trajectory clustering, 2D trajectory completion, rotation estimation and 3D reconstruction. Recently, Yu \etal \cite{Direct-Dense-Deformable:ICCV-2015} bridges template based method and feature track based method by proposing a dense template based direct approach to deformable shape reconstruction from monocular sequences.
Russell \etal \cite{Video-Pop-Up:ECCV-2014} proposed to simultaneously segment a complex dynamic scene containing mixture of multiple objects into constituent objects and reconstruct a 3D model of the scene by formulating the problem as hierarchical graph-cut based segmentation, where the entire scene is decomposed into background and foreground objects and the complex motion of non-rigid or articulated objects are modeled as a set of overlapping rigid parts. Our proposed method differs from this method in the following aspects: 1) We provide a compact representation to multiple non-rigid deformation problem; 2) We propose an efficient and elegant optimization based on ADMM; 3) Our method could deal with both sparse and dense scenarios elegantly.
\section{Formulation}
We seek to reconstruct the 3D trajectories such that they satisfy the union of affine subspace constraint (\ie the 3D trajectories lie in a union of affine subspaces) and non-rigid shape constraints (low rank and spatial coherent).
Let us consider a monocular camera observing multiple non-rigid objects. In this paper, we use the $orthographic$ $camera$ model and eliminate the translation component in camera motion as suggested in \cite{Bregler:CVPR-2000}. The image measurement $\m w_{ij} = [u_{ij}, v_{ij}]^T$ and 3D point $\m S_{ij}$ on the non-rigid shape are related by the camera motion $\m R_i$ as: $\mathbf{w}_{ij} = \m R_i \m{S}_{ij}$, where $\m R_i \in \mathbb{R}^{2\times 3}$ denotes the first two rows of the $i$-th camera rotation. Under this representation, stacking all the $\m F$ frames of measurements and all the $\m P$ points in a matrix form will give us:
\begin{equation}
\label{eq:motion_shape}
\m W = \m R \m S,
\end{equation}
where $\m R = \mathrm{blkdiag}(\m R_1,\cdots,\m R_F) \in \mathbb{R}^{2\m F\times 3\m F}$ denotes the camera motion matrix. NRSFM aims at recovering the \emph{camera motion} $\m R$ and 3D non-rigid reconstruction $\m S \in \mathbb{R}^{3\m F\times \m P}$ from the 2D \emph{measurement matrix} $\m W \in \mathbb{R}^{2\m F \times \m P}$ such that $\m W=\m R\m S$.
\subsection{Representing multi-body non-rigid structure as a union of affine subspace}
We propose that multiple non-rigid structures that corresponds to distinct motion lies in a union of affine subspace. Here, the underlying assumption is that the trajectories belong to different non-rigid objects spans a distinct affine subspace. Figure \ref{fig:A_algo} clearly validates such assumption as there are only connection within clusters and no connection between clusters or block diagonal structure.
Now, consider each trajectory $\m S_i$ that corresponds to a $3 \m F$ dimensional vector formed by stacking the 3D feature tracks of point $i$ across all frames.
\begin{equation}
\m S_i = [\m S_{1i}^T, \m S_{2i}^T, ..., \m S_{\m Fi}^T]^T \in \mathbb{R}^{3\m F},
\end{equation}
where $\m S_{fi}$ $\in $ $\mathbb{R}^3$. Under the orthographic camera model, feature trajectory associated with non-rigid motion lie in an affine subspace of dimension $\mathbb{R}^{3 \m F}$. After taking $\m P$ such $3\m F$ dimensional trajectory and stacking into the column of a matrix we form $\m S$ matrix $\in $ $\mathbb{R}^{3\m F \times \m P}$. Mathematically, it implies that $\m S$ is a real valued matrix whose columns are drawn from a union of $n$ subspace of $\mathbb{R}^{3\m F}$ of dimension less than min($3\m F$, $\m P$). Therefore, each trajectory in a union of affine subspace could be faithfully reconstructed by a combination of other trajectories in the same subspace. This leads to $self$-$expressiveness$ of 3D trajectories. Concretely,
\begin{equation}
\m S_i = \m S \m C_i, \m C_{ii} = 0.
\end{equation}
\begin{figure}
\centering
\includegraphics[width=0.6\textwidth] {Figures/Trajectory_Space_Intuition.pdf}~~~
\caption{Visual illustration of the affine subspace constraint $\m S_i$ = $\m S$ $\m C_i$ for multi-body NRSFM. Each column of $\m S$ is a trajectory of a 3D point(shown in green). This visualization tries to state that a trajectory $\m S_i$ can be reconstructed using affine combination of few other trajectories(sparse). $Note :$ This pictorial representation is provided for better understanding and is only for illustration purpose, signals shown here are not generated mathematically. }
\label{fig:S_Sci}
\end{figure}
Here, $\m C_i$ is a $\m P \times 1$ coefficient vector and $\m C_{ii}$ = $0$ takes care of the trivial solution $\m S_i = \m S_i$. Stacking all such coefficient vectors, we form a matrix $\m C \in \mathbb{R}^{\m P \times \m P}$ that captures the similarity between different trajectories. Using the fact that any trajectory of $\m S$ in an affine subspace of dimension $3\m F$ can be written as affine combination of $3 \m F + 1$ other points from $\m S$ and to cluster trajectories that lies near to union of affine subspaces, we arrive at the following equation.
\begin{equation}
\label{eq:SSC_constraints}
\m S = \m S \m C, \m 1^T \m C = 1^T, \mathrm{diag}(\m C) = 0.
\end{equation}
Figure \ref{fig:Union of Subspace representation} shows the affinity matrix $\m A = |\m C| + |\m C^T|$ obtained for two objects that undergo non-rigid deformation. The obtained sparse solution clearly shows that multi-body non-rigid structure can be represented as union of affine subspace.
\begin{figure}[!htp]
\begin{center}
\subfigure[\label{fig:affine_s}]{\includegraphics[width=0.32\linewidth]{Figures/affine_subspace1.png}}
\subfigure[\label{fig:A_algo}]{\includegraphics[width=0.32\linewidth]{Figures/CMatrix_Dance_Yoga1.png}}
\subfigure[\label{fig:A_clean}]{\includegraphics[width=0.32\linewidth]{Figures/CMatrix_Dance_Yoga_Clean.png}}
\end{center}
\caption{\small (a) Two subjects performing different non-rigid motion that are dance and yoga. Red and green color shows the entire trajectory of each objects over $F$ frames. (b) Visualisation of the affinity matrix obtained using our formulation. (c) Clean affinity matrix obtained after incorporating spectral clustering \cite{Spectral-Clustering:NIPS-2001}. }
\label{fig:Union of Subspace representation}
\noindent\makebox[\linewidth]{\rule{0.83\paperwidth}{0.4pt}}
\end{figure}
\subsection{Representing multiple non-rigid deformations in case of sparse feature tracks}
To solve the problem of multi-body NRSFM in case of sparse feature tracks, we propose the following optimization framework for simultaneous reconstruction and segmentation of objects that are undergoing non-rigid deformation,
\begin{equation}
\begin{aligned}
& \displaystyle \underset{\m C, \m S, \m S^{\sharp}} {\text{minimize}} \frac{1}{2}\|\m W - \m R \m S \|_F^2 + \lambda_1\|\m C\|_1 + \lambda_2\|\m S^{\sharp}\|_{*} \\
& \text{{subject to: }} \\
& \displaystyle \m S^{\sharp} = g(\m S),
\displaystyle \m S = \m S \m C,
\displaystyle \m 1^{T} \m C = \m 1^{T},
\displaystyle \mathrm{diag}(\m C) = \m 0.
\end{aligned}
\label{eq:sparse_version}
\end{equation}
The first term in the above optimization is meant for penalizing reprojection error under $orthographic$ projection. Under single-body NRSFM configuration, 3D shape $\m S$ can be well characterized as lying in a single low dimensional linear subspace. However, when there are multiple non-rigid objects, each non-rigid object could be characterized as lying in an affine subspace. One can argue that affine subspace of dimension $n$ can be considered as a subset of $(n+1)$ dimension that includes origin. Nonetheless, such representation may result in ambiguous solution while clustering different subspace.
The fact that the 3D trajectories lie in a union of affine subspaces as argued previously we put the Eq.~\ref{eq:SSC_constraints} as our optimization constraints.
In addition to this, to reveal the intrinsic structure of multi-body non-rigid structure-from-motion (NRSFM), we seek for sparsest solution of $\m C$ (\cite{SSC:PAMI-2013}). So, the second term in {\bf{Eq.}} \eqref{eq:sparse_version} enforces $l_1$ norm minimization of $\m C$ matrix. Lastly, we enforce a global shape constraint for compact representation of multi-body non-rigid objects by penalizing the rank of entire non-rigid shape. Similar to \cite{Dai-Li-He:CVPR-2012} and \cite{Dense-NRSFM:CVPR-2013}, we penalize the nuclear norm of reshuffled shape matrix $\m S^{\sharp} \in \mathbb{R}^{\m F\times 3 \m P}$, this is because nuclear norm is known as convex envelope of the rank function. Here, $\m g(\m S)$ denotes mapping from $\m S$ $\in$ $\mathbb{R}^{3\m F\times \m P}$ to $\m S^\sharp$ $\in$ $\mathbb{R}^{\m F\times 3 \m P}$.
\subsection{Representing multi-body non-rigid deformations in case of dense feature tracks}
When per pixel feature tracks are available, we can enforce spatial regularization ($Markovian$ $assumption$) i.e there will be a high correlation between neighbouring features. To exploit this property, the spatial smoothness can be used as a regularization term to further constrain the non-rigid reconstruction. Garg \etal \cite{Dense-NRSFM:CVPR-2013} proposed to use the total variation of the 3D shape $\|\m S\|_{TV}$. By contrast, we propose to enforce the spatial smoothness constraint on the coefficient matrix $\m C$ directly by using the $l_1$ norm,
\begin{equation}
\label{eq:C_TV}
\sum_{(i,j) \in \mathcal{N}} \|\m C_i - \m C_j\|_1,
\end{equation}
\ie, the total variation of $\m C$. This definition gives us the benefit in solving the problem as proved later. In essence, the total variations of $\m C$ and $\m S$ are correlated. However, it is desirable that $\m C$ matrix must cater the self-expressiveness of the non-rigid shape deformation as compact as possible. So, we incorporate spatial smoothness constraint on coefficient matrix than on shape matrix $\m S$.
By introducing an appropriately defined matrix $\m D$ encoding the neighbouring relation, Eq.~\eqref{eq:C_TV} can be equivalently expressed as:
\begin{equation}
\|\m C \m D \|_1 = \sum_{(i,j) \in \mathcal{N}} \|\m C_i - \m C_j\|_1.
\label{eq:C_D}
\end{equation}
In {{Fig.}} \ref{fig:DMatrix_exp}, we illustrate the process of how to obtain the matrix $\m D$.
\begin{figure}
\centering
\includegraphics[width=0.6\textwidth] {Figures/Dmatrix_explanation.pdf}~~~
\caption{D matrix caters the neighboring trajectory relation. In the above illustration $\m P_2$, $\m P_4$, $\m P_6$, $\m P_8$, are the 4 immediate neighboring trajectories of $\m P_5$. Therefore, corresponding elements of $\m D$ matrix has $\m -1$ entries, $\m P_5$ has value $\m 1$ and the rest entries are $\m 0$. Therefore, the corresponding coefficient column of $\m C$ matrix now rely on relations defined in the $\m D$ matrix. Here, $\m C$ and $\m D$ are $\m P \times \m P$ and $\m P \times \m 4\m P$ matrix respectively. Where, $\m P$ is the number of total feature tracks. For handling the border pixels one can think of zero padding techniques in 2D image and add zero column vector to the corresponding pixel index in D matrix.}
\label{fig:DMatrix_exp}
\end{figure}
Adding the spatial constraint to the optimization equation \eqref{eq:sparse_version}, that facilitates this neighboring constraint, we procure the following optimization for dense tracks
\begin{equation}
\begin{aligned}
& \displaystyle \underset{\m C, \m S, \m S^{\sharp} }{\text{minimize}}~ \frac{1}{2}\|\m W - \m R \m S \|_F^2 + \lambda_1\|\m C\|_1 + \lambda_2 \| \m C\m D \|_1 + \lambda_3\|\m S^{\sharp}\|_{*} \\
& \text{{subject to:}} \\
& \displaystyle \m S^{\sharp} = g(\m S),
\displaystyle \m S = \m S \m C,
\displaystyle \m 1^{T} \m C = \m 1^{T},
\displaystyle \mathrm{diag}(\m C) = \m 0.
\end{aligned}
\label{eq:Optimization_function}
\end{equation}
\section{Solution}
Due to the bilinear term $\m S = \m S \m C$, the overall optimization of Eq.-\eqref{eq:Optimization_function} is non-convex. We solve it via the ADMM, which has a proven effectiveness for many non-convex problems and is widely used in computer vision. The ADMM works by decomposing the original optimization problem into several sub-problems, where each sub-problem can be solved efficiently. To this end, we seek to decompose Eq.-\eqref{eq:Optimization_function} into several sub-problems.
First note that the two $l_1$ terms $\|\m C\|_1$ and $\|\m C \m D\|_1$ can be put together as $\|\m C [\m I ~\m D]\|_1$. Without loss of generality, we still denote the new term as $\m C \m D$, the only difference is, the new dimension of $\m D$ will be $\m P \times 5\m P$, thus the cost function becomes: $\frac{1}{2} \|\m W - \m R \m S \|_F^2 + \lambda_1\|\m C \m D \|_1 + \lambda_2\|\m S^{\sharp}\|_{*} $. To further decouple the constraint, we introduce an auxiliary variable $\m E = \m C \m D$. With these operations, the optimization problem Eq.-\eqref{eq:Optimization_function} can be reformulated as:
\begin{equation}
\begin{aligned}
& \displaystyle \underset{\m E, \m S, \m S^{\sharp}, \m C} {\text{minimize}} ~\frac{1}{2}\|\m W - \m R \m S \|_F^2 + \lambda_1\|\m E \|_1 + \lambda_2\|\m S^{\sharp}\|_{*} \\
& \text{{subject to:}} \\
& \displaystyle \m S^{\sharp} = g(\m S),
\displaystyle \m S = \m S \m C,
\displaystyle \m C \m D = \m E,
\displaystyle \m 1^{T} \m C = \m 1^{T},
\displaystyle \mathrm{diag}(\m C) = \m 0.
\end{aligned}
\label{eq:main_optimization}
\end{equation}
The Augmented Lagrangian formulation for Eq.-\eqref{eq:main_optimization} is:
\begin{eqnarray}
\mathcal{L}(\m S, \m S^{\sharp}, \m C, \m E, \{\m Y_i\}_{i=1}^4) = & \frac{1}{2}\|\m W - \m R \m S \|_F^2 + \lambda_1\|\m E \|_1 + \lambda_2\|\m S^{\sharp}\|_{*} + <\m Y_1, \m S^{\sharp} - g(\m S)> \nonumber \\
& + \frac{\beta}{2}\|\m S^{\sharp} - g(\m S) \|_F^2 + <\m Y_2, \m S - \m S \m C> + \frac{\beta}{2}\|\m S - \m S \m C \|_F^2 \\ \nonumber & + <\m Y_3, \m C \m D - \m E> + \frac{\beta}{2}\|\m C \m D - \m E \|_F^2 \nonumber \\ \nonumber & + <\m Y_4, 1^T \m C - \m 1^T > + \frac{\beta}{2}\|\m 1^T \m C - \m 1^T \|_F^2.
\label{eq:Lagrangian}
\end{eqnarray}
where $\{\m Y_i\}_{i=1}^4$ are the matrices of Lagrange multipliers corresponding to the four equality constraints, and $\beta$ is the penalty parameter. We do not need to introduce a Lagrange multiplier for the diagonal constraint of $\mathrm{diag}(\m C) = \m 0$ as we will enforce this constraint exactly in the solution of $\m C$.
The ADMM iteratively updates the individual variable so as to minimize $\mathcal{L}$ while the other variables are fixed. In the following, we derive the update for each of the variables.
\subsubsection{The solution of $\m S$:}
\begin{equation}
\begin{array}{l}
\m {S} = \arg\min_{\m S} \frac{1}{2}\|\m W - \m R \m S \|_F^2 + <\m Y_1, \m S^{\sharp} - g(\m S)> + \frac{\beta}{2}\|\m S^{\sharp} - g(\m S) \|_F^2 \\ \qquad \qquad \qquad + <\m Y_2, \m S - \m S \m C> + \frac{\beta}{2}\|\m S - \m S \m C \|_F^2.
\end{array}
\end{equation}
The sub-problem for $\m S$ reaches a least squares problem. The closed-form solution of $\m S$ can be derived as:
\begin{equation}
\label{eq:Update_S}
\frac{1}{\beta}(\m R^T \m R + \beta \m I) \m S + \m S(\m I - \m C)(\m I - \m C^T) = \frac{1}{\beta}\m R^T \m W + (g^{-1}(\m S^{\sharp}) + \frac{g^{-1}(\m Y_1)}{\beta } - \frac{\m Y_2}{\beta}(\m I - \m C^T)),
\end{equation}
which is a Sylvester equation.
\subsubsection{The solution of $\m S^{\sharp}$:}
\begin{equation}
\label{eq:update_S_Sharp}
\m {S}^{\sharp} = \arg\min_{\m {S}^{\sharp}} ~ \lambda_2\|\m S^{\sharp}\|_{*} + <\m Y_1, \m S^{\sharp} - g(\m S)> + \frac{\beta}{2}\|\m S^{\sharp} - g(\m S) \|_F^2
\end{equation}
A close-form solution exists for this sub-problem. Let's define the soft-thresholding operation as $\mathcal{S}_{\tau}[x] = \mathrm{sign}(x)\max(|x|-\tau, 0)$. The optimal solution to Eq.-\eqref{eq:update_S_Sharp} can be obtained as:
\begin{equation}
\label{eq:Update_S_Sharp}
\m S^{\sharp} = \m U \mathcal{S}_{\lambda_2/\beta}(\Sigma) \m V,
\end{equation}
where $[\m U, \Sigma, \m V] = \mathrm{svd}(g(\m S) - \m Y_1/\beta)$.
\subsubsection{The solution of $\m E$:}
\begin{equation}
\m E = \arg\min_{\m E} \lambda_1\|\m E \|_1 + <\m Y_3, \m C \m D -\m E> + \frac{\beta}{2}\| \m C \m D - \m E \|_F^2,
\end{equation}
A close-form solution exists for this sub-problem by using element-wise shrinkage.
\begin{equation}
\label{eq:Update_E}
\m E = \mathcal{S}_{\lambda_1/\beta}(\m C \m D + \frac{\m Y_3}{\beta}).
\end{equation}
\subsubsection{The solution of $\m C$:}
\begin{equation}
\begin{array}{l}
\m C = \arg\min_{\m C} ~ <\m Y_2, \m S - \m S \m C> + \frac{\beta}{2}\|\m S - \m S \m C\|_F^2 + <\m Y_3, \m C \m D - \m E> + \frac{\beta}{2}\|\m C \m D - \m E\|_F^2 + \\ \qquad \qquad \qquad <\m Y_4, \m 1^T \m C - \m 1^T> + \frac{\beta}{2}\|\m 1^T \m C - \m 1^T \|_F^2
\end{array}
\end{equation}
The closed-form solution of $\m C$ is derived as:
\begin{equation}\label{eq:update_C}
(\m S^T \m S + \m 1 \m 1^T)\m C + \m C(\m D \m D^T) = \m S^T \m S + \m S^T \frac{\m Y_2}{\beta} + \m E \m D^T - \m Y_3 \frac{\m D^T}{\beta} + \m 1 \m 1^T - \m 1\frac{\m Y_4}{\beta}
\end{equation}
\begin{equation}
\label{eq:Update_C}
\m C = \m C - \mathrm{diag}(\m C),
\end{equation}
Finally, the Lagrange multipliers $\{\m Y_i\}_{i=1}^4$ and $\beta$ are updated as:
\begin{equation}
\label{eq:update_Y1}
\m Y_1 = \m Y_1 + \beta(\m S^{\sharp} - g(\m S)), \m Y_2 = \m Y_2 + \beta(\m S - \m S \m C),
\end{equation}
\begin{equation}
\label{eq:update_Y3}
\m Y_3 = \m Y_3 + \beta(\m C \m D - \m E), \m Y_4 = \m Y_4 + \beta(\m 1^T \m C - \m 1^T),
\end{equation}
\begin{equation}
\label{eq:update_beta}
\beta = \min(\beta_m, \rho\beta),
\end{equation}
\subsubsection{Initialization:} As our method tries to solve a non-convex optimization problem \eqref{eq:main_optimization}, a proper initialization is needed. In this paper, we initialize our implementation by using the single-body non-rigid structure-from-motion method \cite{Dai-Li-He:CVPR-2012}. Other methods can also be used. Note that these single-body non-rigid structure-from-motion was only used for a proper camera motion. In our current implementation, we have fixed the camera motion while updating the 3D non-rigid reconstruction and segmentation. In future, we will put the update of camera rotation in the loop.
\subsubsection{Sparse feature tracks:} In the above section, we provide the derivation for solving the dense feature tracks case. Sparse feature tracks case could be viewed as a simplified form of the dense case by removing the spatial constraint term $\|\m C \m D \|$ directly.
\begin{algorithm}[t!]
\caption{Multi-body non-rigid structure-from-motion and segmentation via the ADMM}
\label{Algorithm 1}
\begin{algorithmic}
\REQUIRE ~~\\
2D feature track matrix $\m W$, camera motion $\m R$, $\lambda_1$, $\lambda_2$, $\rho>1$, $\beta_m$, $\epsilon$; \\ \vspace{0.2cm}
\hspace{-0.3cm}{\bf Initialize:} ${\m S}^{(0)}$, ${\m S^{\sharp}}^{(0)}$, $\m C^{(0)}$, $\m E^{(0)}$, $\{{\bf Y}_i^{(0)}\}_{i=1}^4 = {\bf 0}$, $\beta^{(0)}$;\\ \vspace{0.2cm}
\WHILE {not converged}
\STATE 1. Update $(\m S, \m S^{\sharp}, \m E, \m C)$ by Eq.~\eqref{eq:Update_S}, Eq.~\eqref{eq:Update_S_Sharp}, Eq.~\eqref{eq:Update_E}, Eq.~\eqref{eq:update_C} and Eq.~\eqref{eq:Update_C};\\
\STATE 2. Update $\{{\bf Y}_i\}_{i=1}^4$ and $\beta$ by Eq.~\eqref{eq:update_Y1}-Eq.~\eqref{eq:update_beta};\\
\STATE 3. Check the convergence conditions $\| \m S^{\sharp} - g(\m S)\|_{\infty} \leq \epsilon$, $\| \m S - \m S\m C\|_{\infty} \leq \epsilon$, $\| \m 1^T \m C - \m 1^T \|_{\infty} \leq \epsilon$, and $\| \m C \m D - \m E \|_{\infty} \leq \epsilon$; \\
\ENDWHILE
\vspace{0.2cm}
\ENSURE ~${\m C}$, $\m S, \m S^{\sharp}$.
\STATE Form an affinity matrix $\m A = |\m C| + |\m C^T|$, then apply spectral clustering \cite{Spectral-Clustering:NIPS-2001} to $\m A$.
Note: Spectral clustering demands number of clusters as input. One way to obtain optimal number of cluster is to use eigen gap information of Laplacian matrix, but in our experiments, we assumed that number of deformable objects are known a priori.
\end{algorithmic}
\end{algorithm}
\section{Experiments}
To evaluate the effectiveness of our approach for multi-body non-rigid structure-from-motion, we conducted extensive experiments on both synthetic data and real images, under both sparse and dense scenarios. Our implementation is not trying to alternate between shape recovery and SSC \cite{SSC:PAMI-2013}, rather its a unified optimization framework.
As our method jointly perform non-rigid 3D reconstruction and segmentation, we use the following criteria to measure the performances of the algorithm: \\
(i) Relative error in multi-body non-rigid 3D reconstruction
\begin{equation}
e_{3D} = \|\m S_f^{est} - \m S_f^{GT}\|_F/\| \m S_f^{GT} \|_F,
\end{equation}
(ii) Error in multi-body non-rigid motion segmentation,
\begin{equation}
e_{MS} = \frac{\mathrm{total ~number ~of ~incorrectly ~segmented ~trajectories}}{\mathrm{total ~number ~of trajectories}}.
\end{equation}
\subsection{Multi-body non-rigid data}
To prove the correctness of our formulation, we first test our method on synthetic dataset. We synthesize two non-rigid objects by using the CMU Mocap dataset \cite{Akhter-Sheikh-Khan-Kanade:Trajectory-Space-2010}, for example one person performs Yoga while the other person is dancing. In Fig.~\ref{fig:synthetic_sequence}, we illustrate four sample multi-body non-rigid structure-from-motion sequences.
\begin{figure}
\centering
\subfigure [\label{fig:1}] {\includegraphics[width=0.24\textwidth]{Figures/Face_Pickup_Affine_Subspace.png}}
\subfigure [\label{fig:2}] {\includegraphics[width=0.24\textwidth]{Figures/Shark_Yoga_Affine_Subspace.png}}
\subfigure [\label{fig:3}]{\includegraphics[width=0.24\textwidth]{Figures/Stretch_Yoga_Affine_Subspace.png}}
\subfigure [\label{fig:4}]{\includegraphics[width=0.24\textwidth]{Figures/Walking_Yoga_Affine_Subspace.png}}
\caption{3D reconstruction and segmentation of different multi-body non-rigid motion sequences a) Face-Pickup Sequence; b) Shark-Yoga Sequence; c) Stretch-Yoga Sequence; d) Walking-Yoga Sequence. The dark small circles in the respective segments shows the Ground-Truth 3D points.}
\label{fig:synthetic_sequence}
\noindent\makebox[\linewidth]{\rule{0.83\paperwidth}{0.4pt}}
\end{figure}
\subsection{Experiment 1: Convergence} Given noise free input, we want to check whether or not our proposed algorithm converge; and if it does converge, whether it converges to the correct solution. Note that we use the sparse sequences from the CMU MoCap dataset \cite{Akhter-Sheikh-Khan-Kanade:Trajectory-Space-2010} directly without any dimension reduction or projection. Typical convergence curves of the objective function and the primal residuals are illustrated in Fig.~\ref{fig:Convergence_Curve}.
\begin{figure}
\centering
\includegraphics[width=1.0\textwidth,height=0.5\textwidth] {Figures/convergence_curve.png}~~~
\caption{Typical convergence curves of the objective function and the primal residuals $\|\m S^{\sharp} - g(\m S)\|_\infty, \| \m S - \m S \m C \|_\infty, \| \m C \m D - \m E\|_\infty$ and $\| \m 1^T \m C - \m 1^T\|_\infty$. The above plot shows the convergence statistics for Dance+Yoga Sequence. *ADMM Convergence Curve = maximum of ($\|\m S^{\sharp} - g(\m S)\|_\infty, \| \m S - \m S \m C \|_\infty, \| \m C \m D - \m E\|_\infty$ and $\| \m 1^T \m C - \m 1^T\|_\infty$). }
\label{fig:Convergence_Curve}
\end{figure}
\subsection{Experiment 2: Performance on noisy feature tracks}
We also conducted experiments to analyze the performance of our method under different level of noise on input track features. In the same manner as above, we generated multi-body non-rigid sequences (``Dance + Yoga'', ``Face + Pickup'', ``Face + Yoga'', ``Shark + Stretch'', ``Shark + Yoga'', ``Stretch + Yoga'' and ``Walking + Yoga''), then zero-mean Gaussian noise with standard deviation $\sigma$ were added to the feature tracks. For each noisy input, we ran our code for 5 times and recorded the mean 3D reconstruction error and non-rigid motion segmentation error.
\begin{figure}
\centering
\includegraphics[width=0.47\textwidth,height=0.166\textheight] {Figures/recons_noise.png}~
\includegraphics[width=0.47\textwidth,height=0.166\textheight]{Figures/seg_noise.png}
\caption{Left: 3D Reconstruction error VS noise levels; Right: non-rigid motion segmentation error VS noise levels. }
\label{fig:noise_performance}
\end{figure}
Fig.~\ref{fig:noise_performance}, illustrate the statistical results of 3D non-rigid reconstruction and non-rigid motion segmentation. From the statistics as plotted in Fig.~\ref{fig:noise_performance}, we conclude that both the 3D reconstruction error and the motion segmentation error increases with the increase of noise level. Our 3D reconstruction based non-rigid motion segmentation achieves smaller motion segmentation error compared with 2D trajectory based motion segmentation methods such as sparse subspace clustering (SSC) \cite{SSC:PAMI-2013} and efficient dense subspace clustering (EDSC) \cite{EDSC:WACV-2014}.
\subsection{Experiment 3: Performance Comparison on Sparse Dataset} In Table~\ref{tab:segmentation_comparison} we compared the segmentation results of our approach with SSC \cite{SSC:PAMI-2013} and EDSC \cite{EDSC:WACV-2014} on multi-body non-rigid sequences over 2D feature tracks. It clearly demonstrate that performing non-rigid motion segmentation in 3D space using our approach leads to remarkable results. Since, our method jointly solves 3D reconstruction and multi-body non-rigid motion segmentation, we compare our method with the two stage method, namely
\begin{enumerate}[1)]
\item Baseline method 1: Single body non-rigid structure-from-motion (State-of-the-art ``block-matrix method'' \cite{Dai-Li-He:CVPR-2012} was used) followed by subspace clustering of the 3D trajectories (SSC \cite{SSC:PAMI-2013} was used), denoted as ``BMM+SSC(3D)'';
\item Baseline method 2: Subspace clustering of the 2D feature tracks (2D trajectories) followed by single body non-rigid structure-from-motion for each cluster of 2D feature tracks, denoted as ``SSC(2D)+BMM''.
\end{enumerate}
Table~\ref{tab:comparisons_baseline}, provides experimental comparisons between our method and two baseline methods in dealing with multi-body non-rigid structure-from-motion. In all the sequences, our method achieves zero multi-body non-rigid motion segmentation error and comparable 3D non-rigid reconstruction.
Experiments clearly shows that our 3D reconstruction is very close to the accuracy of BMM \cite{Dai-Li-He:CVPR-2012} method in almost all dataset. However, the advantage with our framework we can have robust segmentation along with better reconstruction at the same time. Fig. \ref{fig:comparison_efficacy} shows the robustness of our approach. Our method faithfully reconstruct and segment two different complex non-rigid motion than the two baseline methods. Extensive experiments were performed on synthetic sparse data-sets with different combination of non-rigid motion Fig. \ref{fig:multi_nonrigid} shows some of the results on these different combinations of non-rigid motion on CMU Mocap dataset \cite{Akhter-Sheikh-Khan-Kanade:Trajectory-Space-2010}, where as Fig. \ref{fig:umpm_result} and Table \ref{tab:result_umpm} presents results obtained on UMPM dataset \cite{UMPM}.
We also compared the affinity matrices of our method with SSC \cite{SSC:PAMI-2013}. It's visually apparent from Fig.~\ref{fig:affinity_matrix}, that our method outputs an affinity matrix with better structure, which results in better non-rigid motion segmentation performance.
\begin{table}
\begin{center}
\caption{Motion segmentation performance comparison with SSC \cite{SSC:PAMI-2013} and EDSC \cite{EDSC:WACV-2014} over 2D feature tracks.}
\label{tab:segmentation_comparison}
\begin{tabular}{|c|c|c|c|}
\hline
Dataset & SSC ($e_{MS}$) & EDSC ($e_{MS}$) & Ours\\
\hline
Dance+Yoga & 0.025 & 0.0345 & 0.0\\
\hline
Drink+Walking & 0.0 & 0.01 & 0.0\\
\hline
Shark+Stretch & 0.3939 & 0.0 & 0.0 \\
\hline
Walking+Yoga & 0.0 & 0.0 & 0.0 \\
\hline
Face+Pickup & 0.098 & 0.0 & 0.0\\
\hline
Face+Yoga & 0.012 & 0.0 & 0.0\\
\hline
Shark+Yoga & 0.41 & 0.0 & 0.0 \\
\hline
Stretch+Yoga & 0.0 & 0.0 & 0.0\\
\hline
\end{tabular}
\end{center}
\end{table}
\begin{table}[!htp]
\centering
\caption{Performance comparison between our method and the baseline methods, where 3D reconstruction error ($e_{3D}$) and non-rigid motion segmentation error ($e_{MS}$) are used as error metrics.}
\begin{tabular}{|c|c|c|c|c|c|c|}
\hline
\multirow{2}{*}{Dataset} &
\multicolumn{2}{c|}{BMM + SSC (3D)} &
\multicolumn{2}{c|}{SSC(2D) + BMM} &
\multicolumn{2}{c|}{Our Method} \\ \cline{2-7}
& $e_{3D}$ & $e_{MS}$ & $e_{3D}$ & $e_{MS}$ & $e_{3D}$ & $e_{MS}$ \\
\hline
Dance + Yoga & 0.0456 & 0.0345 & 0.0588 & 0.0259 & 0.046 & {\bf{0.0}} \\
\hline
Drink + Walking & 0.0745 & 0.0 & 0.0858 & 0.0 & {\bf{0.073}} & {\bf{0.0}} \\
\hline
Shark + Stretch & 0.0246 & 0.4015 & 0.0979 & 0.3939 & 0.025 & {\bf{0.0}} \\
\hline
Walking + Yoga & 0.0702 & 0.0 & 0.0900 & 0.0 & {0.0702} & {\bf{0.0}} \\
\hline
Face + Pickup & 0.0324 & 0.0988 & 0.0239 & 0.0988 & 0.025 & {\bf{0.0}} \\
\hline
Face + Yoga & 0.0172 & 0.012 & 0.0332 & 0.012 & 0.019 & {\bf{0.0}} \\
\hline
Shark + Yoga & 0.0356 & 0.4167 & 0.1049 & 0.4091 & 0.0371 & {\bf{0.0}}\\
\hline
Stretch + Yoga & 0.0392 & 0.0 & 0.0557 & 0.0 & 0.0393 & {\bf{0.0}} \\
\hline
\end{tabular}
\label{tab:comparisons_baseline}
\end{table}
\begin{figure}[!htp]
\begin{center}
\subfigure[\label{fig:NRSFM_3D}]{\includegraphics[width=0.32\linewidth]{Figures/NrSFM_SSC_3D.png}}
\subfigure[\label{fig:SSC_NRSFM}]{\includegraphics[width=0.32\linewidth]{Figures/SSC_2D_NRSFM.png}}
\subfigure[\label{fig:our_method}]{\includegraphics[width=0.32\linewidth]{Figures/Our_Method.png}}
\end{center}
\caption{\small Demonstrating the efficacy of our approach. The above plot shows the results on Dance + Yoga sequence. (a) Result obtained by applying BMM method \cite{Dai-Li-He:CVPR-2012} to get 3D points and then use SSC \cite{SSC:PAMI-2013} to segment 3D points. (b) Result obtained by applying SSC \cite{SSC:PAMI-2013} to 2D feature tracks and then use BMM \cite{Dai-Li-He:CVPR-2012} separately to each segment to get 3D reconstruction. (c) Result by applying simultaneous reconstruction and segmentation framework(Our approach). }
\label{fig:comparison_efficacy}
\noindent\makebox[\linewidth]{\rule{0.83\paperwidth}{0.4pt}}
\end{figure}
\begin{figure}[!htp]
\begin{center}
\subfigure[\label{fig:shark_yoga}]{\includegraphics[width=0.24\linewidth]{Figures/Shark_Yoga_Affine_Subspace.png}}
\subfigure[\label{fig:stretch_yoga}]{\includegraphics[width=0.24\linewidth]{Figures/Dance_Yoga_Affine_Subspace.png}}
\subfigure[\label{fig:face_yoga}]{\includegraphics[width=0.24\linewidth]{Figures/Face_Yoga_Affine_Subspace.png}}
\subfigure[\label{fig:shark_stretch}]{\includegraphics[width=0.24\linewidth]{Figures/Shark_Stretch_Affine_Subspace.png}}
\end{center}
\caption{\small Illustration of multi-body non-rigid reconstruction with segmentation on different data-sets. (a) Shark + Yoga Sequence (b) Dance + Yoga Sequence (c) Face + Yoga Sequence (d) Shark + Stretch Sequence. The dark small circles in the respective segments shows the Ground-Truth 3D points.}
\label{fig:multi_nonrigid}
\noindent\makebox[\linewidth]{\rule{0.83\paperwidth}{0.4pt}}
\end{figure}
\begin{table}
\centering
\caption{Quantitative results on UMPM data-set \cite{UMPM}. The rotation were obtained using BMM method \cite{Dai-Li-He:CVPR-2012} (K = 11). Different K value may results in different reconstruction error ($e _{MS}$) value. Statistics clearly reveals that we can get high fidelity motion segmentation using our approach and reasonable 3D reconstruction simultaneously on very complex data-set as well.}
\begin{tabular}{|c|c|c|c|c|c|c|}
\hline
\multirow{2}{*}{Dataset} &
\multicolumn{2}{c|}{BMM + SSC (3D)} &
\multicolumn{2}{c|}{SSC(2D) + BMM} &
\multicolumn{2}{c|}{Our Method} \\ \cline{2-7}
& $e_{3D}$ & $e_{MS}$ & $e_{3D}$ & $e_{MS}$ & $e_{3D}$ & $e_{MS}$ \\
\hline
p2\_free\_2 & 0.1973 & 0.0 & 0.2177 & 0.0 & 0.1992 & {0.0} \\
\hline
p3\_ball\_1 & 0.1356 & 0.0 & 0.1422 & 0.0 & 0.1348 & 0.0 \\
\hline
p4\_grab\_2 & 0.2018 & 0.0 & 0.2182 & 0.0 & 0.2080 & 0.0 \\
\hline
p4\_table\_12 & 0.2313 & 0.0 & 0.2418 & 0.0 & 0.2313 & 0.0 \\
\hline
p4\_meet\_12 & 0.0800 & 0.0135 & 0.0941 & 0.0 & 0.0821 & 0.0135 \\
\hline
\end{tabular}
\label{tab:result_umpm}
\end{table}
\begin{figure}
\centering
\includegraphics[width=0.6\textwidth, height=0.15\textheight] {Figures/result_umpm.pdf}~~~
\caption{Reconstruction and segmentation result on (a) p3\_ball\_1. (b) p4\_meet\_12 data-set from UMPM database \cite{UMPM}. Light green and light red circles shows the 3D reconstruction, where as dark circles of the respective color shows their 3D ground-truth. Here, different color shows the corresponding segmentation.}
\label{fig:umpm_result}
\noindent\makebox[\linewidth]{\rule{0.83\paperwidth}{0.4pt}}
\end{figure}
\begin{figure}
\centering
\includegraphics[width=0.25\textwidth]{Figures/Affinity_Matrix_SSC.jpg}~~~
\includegraphics[width=0.25\textwidth]{Figures/Affinity_Matrix_Our_Method.jpg}
\caption{Obtained Affinity Matrix $\m A = |\m C| + |\m C^T|$. a) Affinity matrix from SSC; b) Affinity matrix from our Method. Best Viewed on Screen.}
\label{fig:affinity_matrix}
\end{figure}
\subsection{Experiment 4: Analysis on Dense Dataset}
To expeditiously test the effectiveness of our implementation over available dense sequences, we tested our method on the uniformly sampled version of the original sequences. We performed experimentation on benchmark NRSFM synthetic and real data-set sequence \cite{Dense-NRSFM:CVPR-2013} introduced by Grag {\it{et al.}} Table \ref{tab:dense_3d_synthetic} lists the obtained 3D reconstruction error on these freely available standard synthetic data-sequence \cite{Dense-NRSFM:CVPR-2013}. Figure \ref{fig:Dense_result_synthetic} and \ref{fig:real_face_back_} provides visual insight of the procured 3D shapes over synthetic and real sequence respectively. To test the segmentation of different structure on real dataset, we performed experiments by combining two real dataset sequence. Figure \ref{fig:real_face_heart_} shows the segmentation of non-rigid feature tracks to their corresponding classes.
\begin{table}
\begin{center}
\caption{Quantitative results on synthetic face sequence, without and with neighboring constraints.}
\label{tab:dense_3d_synthetic}
\begin{tabular}{|c|c|c|c|}
\hline
Dataset & No.of Feature points & e3d(C constraint) & e3d(CD constraint)\\
\hline
Face Sequence 1 & 3275 & 0.0749 & 0.0745\\
\hline
Face Sequence 2 & 3275 & 0.0506 & 0.050\\
\hline
Face Sequence 3 & 3275 & 0.0384 & 0.0380\\
\hline
Face Sequence 4 & 3275 & 0.0446 & 0.0443 \\
\hline
\end{tabular}
\end{center}
\end{table}
\begin{figure}[t!]
\centering
\includegraphics[width=0.8\textwidth, height=0.20\textheight] {Figures/Diagram_dense_recons.pdf}~~~
\caption{Reconstruction results on the synthetic Face Sequence benchmark data-set \cite{Dense-NRSFM:CVPR-2013}. The above shown results were obtained on uniformly sampled trajectories of Face Sequence. }
\label{fig:Dense_result_synthetic}
\end{figure}
\begin{figure}[t!]
\centering
\includegraphics[width=0.8\textwidth, height=0.22\textheight] {Figures/real_sequence_face_back.pdf}~~~\\
\includegraphics[width=0.8\textwidth, height=0.22\textheight] {Figures/Diagram_real_sequence_heart.pdf}
\caption{Reconstruction results on the real sequence data-sets: Face, Back and Heart Sequence \cite{Dense-NRSFM:CVPR-2013}. The above shown results were obtained on uniformly sampled trajectories.}
\label{fig:real_face_back_}
\end{figure}
\begin{figure}[t!]
\centering
\includegraphics[width=0.8\textwidth, height=0.20\textheight] {Figures/Face+Heart.pdf}~~~
\caption{Segmentation result on the real sequence data-sets: (Face + Heart) Sequence \cite{Dense-NRSFM:CVPR-2013}. (a) Top-View, (b) \& (c) Side-View of the scene. Different color signature (Green for heart and Red for Face) symbolizes the corresponding class labels.}
\label{fig:real_face_heart_}
\end{figure}
\section{Conclusions}
This paper has filled in a missing gap in the Structure-from-Motion family by proposing a new framework for ``Multi-body Non-rigid-Structure-from-Motion''. It achieves a joint non-rigid reconstruction and non-rigid shape segmentation of multiple deformable structures observed in a single image sequence. Under our new multi-body NRSFM framework, the solutions for motion segmentation can better constrain the solutions for 3D reconstruction. This way, we achieved superior performance in both 3D non-rigid reconstruction and non-rigid motion segmentation, compared with the alternative, two stage method (first segment, then reconstruct). In future, we plan to investigate the scalability issue observed in our current implementation; we aim to apply the new method to denser feature tracks in longer video sequences. In addition to this, we plan to improve over state-of-the-art 3D reconstruction method \cite{Dai-Li-He:CVPR-2012} to get much robust reconstruction for the objects undergoing non-rigid deformation.
\section*{Acknowledgments}
Y. Dai is funded in part by ARC Grants (DE140100180, LP100100588) and National Natural Science Foundation of China (61420106007). H. Li's research is funded in part by ARC grants (DP120103896, LP100100588, CE140100016) and NICTA (Data61).
\bibliographystyle{splncs}
| da3c1e83ef2052435a927df2141e54e0a9e70ee1 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Numerical simulations of the quantum dynamics and the associated transient-absorption spectra}
\label{Equations of motion and transient-absorption spectra}
\setcounter{equation}{0}
We model Rb atoms as three- or five-level systems, as shown in Fig.~2(b) in the article. We consider a quantum system in the state $|\psi(t)\rangle = \hat{\mathcal{U}}(t,t_0)|\psi(t_0)\rangle$, where the evolution operator $\hat{\mathcal{U}}(t,t_0)$, describing the dynamics from $t_0$ to $t$, is solution of the Schr\"odinger equation
\begin{equation}
\frac{\mathop{}\!\mathrm{d} \hat{\mathcal{U}}(t,t_0)}{ \mathop{}\!\mathrm{d} t} = -\mathrm{i}[\hat{H}_0 + \hat{H}_{\mathrm{int}}]\,\hat{\mathcal{U}}(t,t_0)\ \ \ \hat{\mathcal{U}}(t_0,t_0) = \hat{I}.
\label{eq:evolution-operator}
\end{equation}
The total Hamiltonian consists of the atomic-structure Hamiltonian $\hat{H}_0 = \sum_i{(\omega_{i} -\mathrm{i}\gamma_i/2 ) |i \rangle\langle i|}$, complex-valued and including energies $\omega_{i}$ and linewidths $\gamma_{i}$ of each atomic state $|i\rangle$, and of the electric-dipole ($E1$) light-matter interaction $\hat{H}_{\mathrm{int}} = - [\hat{\boldsymbol{D}}^{-}\cdot\,\boldsymbol{\mathcal{E}}^{+}(t) + \mathrm{H.\,c.}]$, in the rotating-wave approximation \cite{Scully:QuantumOptics}. Here, $\hat{I} = \sum_i{|i \rangle\langle i|}$ is the identity operator, while $\hat{\boldsymbol{D}}^{-} = \sum_{i>j} \boldsymbol{D}_{ij} |i \rangle\langle j|$ is the negative-frequency part of the dipole-moment operator, of matrix elements $\boldsymbol{D}_{ij}$, and $\boldsymbol{\mathcal{E}}^{+}(t)$ is the complex electric field \cite{diels2006ultrashort}. Pump and probe pulses are modeled by functions of the form $\boldsymbol{\mathcal{E}}(t-t_{\mathrm{c}},\phi) = \boldsymbol{\mathcal{E}}^{+}(t -t_{\mathrm{c}},\phi ) + \mathrm{c.c.} = \mathcal{E}_0\,f\left(t-t_{\mathrm{c}}\right)\,\cos\left[\omega_{\mathrm{L}} \left(t-t_{\mathrm{c}}\right)+\phi\right]\hat{\boldsymbol{e}}_z$, centered around $t_{\mathrm{c}}$, with carrier-envelope phase (CEP) $\phi$ and envelope function of the form $f(t) = \cos^2{(\pi t/T)}\,R(t/T)$. Here, $R(x) = \theta(x+1/2)-\theta(x-1/2)$ is defined in terms of the Heaviside step function $\theta(x)$, such that $T = \pi T_{\mathrm{FWHM}}/[2\arccos{(\sqrt[4]{1/2})}]$, with $T_{\mathrm{FWHM}}$ being the full width at half maximum of $f^2(t)$. The evolution operator $\hat{\mathcal{U}}_{0}(t, t_0)$ introduced in the article is associated with a pulse centered on $t_{\mathrm{c}} = 0$ and with vanishing CEP $\phi = 0$. The generic evolution operator reads $\hat{\mathcal{U}}(t, t_0) = \hat{\varPhi}^{\dagger}(\phi)\,\hat{\mathcal{U}}_0\left(t - t_{\mathrm{c}},t_0-t_{\mathrm{c}}\right)\,\hat{\varPhi}(\phi)$, with the diagonal operator $\hat{\varPhi}(\phi)$ accounting for the pulse CEP.
{In order to better present our key idea, we employ a Schr\"odinger-equation-based formalism throughout the manuscript, with decay rates $\gamma_2 = \gamma_3 = 1/(1500\,\mathrm{fs})$ modeling experimental linewidth due to, e.g., Doppler or collision-induced broadening. Radiative decay times of $27\,\mathrm{ns}$ and $25.7\,\mathrm{ns}$, from the excited states 2 and 3, respectively, are more than 4 orders of magnitude larger than the duration of the femtosecond pulses used for our control scheme. Therefore, spontaneous decay can be neglected for our system. In this time scale, also the employed decay rates $\gamma_2 = \gamma_3 = 1/(1500\,\mathrm{fs})$ are small and the difference with a density-matrix treatment is therefore negligible. In Sec.~\ref{Density-matrix formulation} we display how our results can be generalized to a more complex case requiring a density-matrix formulation.}
We solve the Schr\"odinger equation~(\ref{eq:evolution-operator}) to calculate the time evolution of the system interacting with delayed pump and probe pulses, and use the obtained solution to simulate transient-absorption spectra. The same procedure is used to calculate the dynamics of the controlled quantum system, displayed in Figs.~3(c) and 4 in the article, for an optimal sequence of two intense pulses.
To avoid repetitions, in the following we immediately refer to the more general case of a five-level scheme.
The state of the system $|\psi^{(5)}(t,\tau)\rangle = \sum_{i=1}^5c^{(5)}_i(t,\tau)|i\rangle$ has a vector representation given by $\boldsymbol{c}^{(5)} = (c^{(5)}_1,\,c^{(5)}_2,\,c^{(5)}_3,\,c^{(5)}_4,\,c^{(5)}_5)^{\mathrm{T}}$. The free-evolution operator is given by
\begin{widetext}
\begin{equation}
\hat{V}^{(5)}(t) = \mathrm{diag} (1,\,\mathrm{e}^{-(\frac{\gamma_2}{2} + \mathrm{i} \omega_{21})t},\,\mathrm{e}^{-(\frac{\gamma_3}{2} + \mathrm{i} \omega_{31})t},\,\mathrm{e}^{-(\frac{\gamma_4}{2} + \mathrm{i} \omega_{41})t},\,\mathrm{e}^{-(\frac{\gamma_5}{2} + \mathrm{i} \omega_{51})t}).
\end{equation}
Here, the two highest-lying excited states are almost degenerate, with transition energies $\omega_{41} = \omega_{51} = 3.19\,\mathrm{eV}$. For the $E1$-allowed transitions depicted in Fig.~2(b) in the article, we introduce the time-dependent Rabi frequencies $\varOmega_{ij}(t) = D_{ij}\,\mathcal{E}_0\,f(t)$, with dipole-moment matrix elements $\hat{\boldsymbol{D}}_{ij} = D_{ij}\hat{\boldsymbol{e}}_z$ aligned along the pulse polarization vector \cite{PhysRevA.61.050502, PhysRevA.69.022509}. We calculate the time evolution of the atomic system interacting with a pump pulse centered on $t_{\mathrm{c}} = 0$ and a probe pulse centered on $t_{\mathrm{c}} =\tau$. We assume a weak probe pulse, with duration $T_{\mathrm{FWHM, pr}} = 15\,\mathrm{fs}$ and intensity $I_{\mathrm{pr}} = 1\times10^8\,\mathrm{W/cm^2}$, and pump pulses of duration $T_{\mathrm{FWHM, pu}} = 30\,\mathrm{fs}$ and intensities $I_{\mathrm{pu}}$ varying between $0.1\times10^{10}\,\mathrm{W/cm^2}$ and $5\times10^{10}\,\mathrm{W/cm^2}$. For $(t-t_{\mathrm{c}})\in[-T/2,\,T/2]$, the time evolution of $\boldsymbol{c}^{(5)}$ in the rotating-wave approximation is given by
\begin{equation}
\frac{\mathop{}\!\mathrm{d} \boldsymbol{c}^{(5)}}{\mathop{}\!\mathrm{d} t} = [\hat{\varPhi}^{(5)}(\phi)]^{\dagger}\,[\hat{\varLambda}^{(5)}(t-t_{\mathrm{c}})]^{\dagger}\,\hat{M}^{(5)}(t-t_{\mathrm{c}})\,\hat{\varLambda}^{(5)}(t-t_{\mathrm{c}})\,\hat{\varPhi}^{(5)}(\phi )\,\boldsymbol{c}^{(5)}(t),
\label{eq:timeevolution5}
\end{equation}
with the phase of the pulse accounted for by $\hat{\varPhi}^{(5)}(\phi) = \mathrm{diag}(1,\,\mathrm{e}^{\mathrm{i}\phi},\,\mathrm{e}^{\mathrm{i}\phi},\,\mathrm{e}^{2\mathrm{i}\phi},\,\mathrm{e}^{2\mathrm{i}\phi})$ and $\hat{\varLambda}^{(5)}(t) = \mathrm{diag}(1,\,\mathrm{e}^{\mathrm{i}\omega_{\mathrm{L}}t},\,\mathrm{e}^{\mathrm{i}\omega_{\mathrm{L}}t},\,\mathrm{e}^{2\mathrm{i}\omega_{\mathrm{L}}t},\,\mathrm{e}^{2\mathrm{i}\omega_{\mathrm{L}}t})$, while
\begin{equation}
\hat{M}^{(5)}(t) =
\begin{pmatrix}
0 & \mathrm{i} \frac{\varOmega_{12}(t)}{2} & \mathrm{i} \frac{\varOmega_{13}(t)}{2} & 0 & 0\\
\mathrm{i} \frac{\varOmega_{12}^*(t)}{2} &-\frac{\gamma_2}{2}-\mathrm{i} \omega_{21} & 0& \mathrm{i} \frac{\varOmega_{24}(t)}{2} & 0\\
\mathrm{i} \frac{\varOmega_{13}^*(t)}{2} & 0 &-\frac{\gamma_3}{2}-\mathrm{i} \omega_{31} & \mathrm{i} \frac{\varOmega_{34}(t)}{2} & \mathrm{i} \frac{\varOmega_{35}(t)}{2}\\
0 & \mathrm{i} \frac{\varOmega_{24}^*(t)}{2} &\mathrm{i} \frac{\varOmega_{34}^*(t)}{2} & -\frac{\gamma_4 + \gamma_{\mathrm{P},4}(t)}{2}-\mathrm{i} \omega_{41} & 0\\
0 & 0 &\mathrm{i} \frac{\varOmega_{35}^*(t)}{2} & 0 & -\frac{\gamma_5 + \gamma_{\mathrm{P},5}(t)}{2}-\mathrm{i} \omega_{51}
\end{pmatrix}.
\end{equation}
\end{widetext}
Photoionization of states $|4\rangle$ and $|5\rangle$ in the presence of an optical pulse has also been included as an effective loss of amplitude at the rate $\gamma_{\mathrm{P},l}(t) = \sigma_l(\omega_{L}) \mathcal{I}_{\mathrm{L}}(t)$, with photoionization cross sections $\sigma_l(\omega_{L})$ calculated with \cite{LANL}, and optical flux $\mathcal{I}_{\mathrm{L}}(t) = I_{\mathrm{L}}(t)/\omega_{L}$ defined in terms of the time-dependent pulse intensity $I_{L}(t) = [\mathcal{E}_0\,f(t)]^2/(8\pi\alpha)$.
The dynamics of the five-level system are associated with the $5\times 5$ evolution operator $\hat{\mathcal{U}}^{(5)}(t, t_0)$ and the related interaction operator
\begin{equation}
\begin{split}
\hat{U}^{(5)} &= \hat{V}^{(5)}(-T/2)\,\hat{\mathcal{U}}^{(5)}_{0}(T/2, -T/2)\,[\hat{V}^{(5)}(T/2)]^{-1} \\
&=
\begin{pmatrix}
\hat{U}^{(5)}_{3\times3} & \hat{U}^{(5)}_{3\times2}\\
\hat{U}^{(5)}_{2\times3} & \hat{U}^{(5)}_{2\times2}
\end{pmatrix}.
\end{split}
\end{equation}
The time evolution of the five-level system, calculated from Eq.~(\ref{eq:timeevolution5}), is used to numerically simulate transient-absorption spectra $\mathcal{S}^{(5)}(\omega,\tau)$.
Fast oscillations as a function of the time delay $\tau$ are averaged out in a noncollinear geometry \cite{PhysRevLett.115.033003} and cannot be discerned. We numerically calculate the averaged time-delay-dependent spectrum by performing a convolution of the fast oscillating spectrum with a normalized Gaussian function $G(\tau, \Delta\tau)$ of width $\Delta\tau = 5\times2\pi/\omega_{\mathrm{L}}$ \cite{PhysRevLett.115.033003}.
For a three-level model associated with the state $\left|\psi^{(3)}(t,\tau)\right\rangle =\sum_{i=1}^3 c^{(3)}_i(t,\tau)|i\rangle$ and the density-matrix elements $\rho^{(3)}_{ij}(t,\tau)= c^{(3)}_i(t,\tau)\,[c^{(3)}_j(t,\tau)]^*$, $i,\,j\in\{1,\,2,\,3\}$, then the absorption spectrum results from
\begin{equation}
\mathcal{S}^{(3)}(\omega,\tau)\propto -\omega\mathop{}\!\mathbf{Im}\left[\sum_{k=2}^3 D_{1k}^* \int_{-\infty}^{\infty}\rho^{(3)}_{1k}(t,\tau)\,\mathrm{e}^{-\mathrm{i} \omega (t- \tau)}\mathop{}\!\mathrm{d} t\right].
\label{eq:spectrum3}
\end{equation}
Similarly, the dynamics of the three-level system is described in terms of the operators $\hat{V}^{(3)}(t) = \mathrm{diag} (1,\,\mathrm{e}^{-(\frac{\gamma_2}{2} + \mathrm{i} \omega_{21})t},\,\mathrm{e}^{-(\frac{\gamma_3}{2} + \mathrm{i} \omega_{31})t})$, $\hat{\varPhi}^{(3)}(\phi) = \mathrm{diag}(1,\,\mathrm{e}^{\mathrm{i}\phi},\,\mathrm{e}^{\mathrm{i}\phi})$, $\hat{\varLambda}^{(3)}(t) = \mathrm{diag}(1,\,\mathrm{e}^{\mathrm{i}\omega_{\mathrm{L}}t},\,\mathrm{e}^{\mathrm{i}\omega_{\mathrm{L}}t})$, while
\begin{equation}
\hat{M}^{(3)}(t) =
\begin{pmatrix}
0 & \mathrm{i} \frac{\varOmega_{12}(t)}{2} & \mathrm{i} \frac{\varOmega_{13}(t)}{2} \\
\mathrm{i} \frac{\varOmega_{12}^*(t)}{2} &-\frac{\gamma_2}{2}-\mathrm{i} \omega_{21} & 0\\
\mathrm{i} \frac{\varOmega_{13}^*(t)}{2} & 0 &-\frac{\gamma_3}{2}-\mathrm{i} \omega_{31}
\end{pmatrix},
\label{eq:M3}
\end{equation}
resulting in a $3\times 3$ evolution operator $\hat{\mathcal{U}}^{(3)}(t, t_0)$ and a related interaction operator
$\hat{U}^{(3)} = \hat{V}^{(3)}(-T/2)\,\hat{\mathcal{U}}^{(3)}_{0}(T/2, -T/2)\,[\hat{V}^{(3)}(T/2)]^{-1}$.
It is important to stress that the $3\times 3$ matrix $\hat{U}^{(5)}_{3\times3}$ is different from $\hat{U}^{(3)}$, since $\hat{U}^{(5)}_{3\times3}$ is influenced by the presence of states $|4\rangle$ and $|5\rangle$, which could not be taken into account if one solved the Schr\"odinger equation for a three-level system exclusively. By using TAS to extract strong-field interaction (SFI) operators by means of the analytical fitting model presented in Sec.~\ref{Fitting analytical model of transient-absorption spectra in terms of intensity-dependent control operators}, one has thus access to $\hat{U}^{(5)}_{3\times3}$, in contrast to methods based exclusively on a theory which could only provide $\hat{U}^{(3)}$.
\section{Analytical fitting model of transient-absorption spectra in terms of interaction operators}
\label{Fitting analytical model of transient-absorption spectra in terms of intensity-dependent control operators}
To interpret transient-absorption spectra from Rb atoms, modeled by the few-level scheme depicted in Fig.~2(b) in the article, we focus on a $V$-type three-level model, accounting for the two transitions $|1\rangle\rightarrow|k\rangle$, $k\in\{2,\,3\}$.
Starting from these assumptions and the associated definition of the spectrum in Eq.~(\ref{eq:spectrum3}), we derive an analytical fitting model to relate experimental transient-absorption spectra to the operators $\hat{U}_{\mathrm{pu}}(I)$ and $\hat{U}_{\mathrm{pr}}$
\subsection{Interaction operator for weak short probe pulses}
\label{Evolution operator for weak short probe pulses}
The operator $\hat{U}_{\mathrm{pr}}$ models a weak probe pulse, with FWHM of $15\,\mathrm{fs}$ and intensity of $1\times 10^{8}\,\mathrm{W/cm^2}$. In order to interpret results from TAS, we approximate the envelope function $f_{\mathrm{pr}}(t)$ of such a short (broadband) weak pulse with a Dirac-$\delta$-like peak
\begin{equation}
\tilde{f}_{\mathrm{pr}}(t) = \delta(t)\,\int_{-T_{\mathrm{pr}}/2}^{T_{\mathrm{pr}}/2} f_{\mathrm{pr}}(t')\,\mathop{}\!\mathrm{d} t',
\end{equation}
where $T_{\mathrm{pr}}$ is the duration of the pulse.
An explicit solution of Eq.~(\ref{eq:evolution-operator}) provides
\begin{equation}
\hat{\mathcal{U}}_{\mathrm{pr}}(t,t_0) = \hat{V}(t)\,\mathrm{e}^{\hat{B}\,[\theta(t) - \theta(t_0 )]}\,\hat{V}(-t_0),
\end{equation}
where $\theta(x)$ is the Heaviside step function, $\hat{V}(t)$ is the free-evolution operator, and
\begin{equation}
\hat{B} =
\begin{pmatrix}
0 & \mathrm{i} \frac{\vartheta_2}{2} & \mathrm{i} \frac{\vartheta_3}{2}\\
\mathrm{i} \frac{\vartheta_2^*}{2} &0 & 0\\
\mathrm{i} \frac{\vartheta_3^*}{2} & 0 &0
\end{pmatrix}
\end{equation}
is defined in terms of the pulse areas $\vartheta_k =\int_{-T_{\mathrm{pr}}/2}^{T_{\mathrm{pr}}/2} \varOmega_{\mathrm{pr},1k}(t)\,\mathop{}\!\mathrm{d} t$ of the probe-pulse Rabi frequencies $\varOmega_{\mathrm{pr},1k}(t)$, for $k\in\{2,\,3\}$. Taking advantage of the weak probe-pulse intensity, the probe operator $\hat{U}_{\mathrm{pr}} = \hat{V}(-T_{\mathrm{pr}}/2)\,\hat{\mathcal{U}}_{\mathrm{pr}}(T_{\mathrm{pr}}/2, -T_{\mathrm{pr}}/2)\,\hat{V}^{-1}(T_{\mathrm{pr}}/2)$ reads
\begin{equation}
\hat{U}_{\mathrm{pr}} = \mathrm{e}^{\hat{B}} \approx \hat{I}_{3} + \hat{B} =
\begin{pmatrix}
1 & \mathrm{i} \frac{\vartheta_2}{2} & \mathrm{i} \frac{\vartheta_3}{2}\\
\mathrm{i} \frac{\vartheta_2^*}{2} &1 & 0\\
\mathrm{i} \frac{\vartheta_3^*}{2} & 0 &1
\end{pmatrix}.
\label{eq:Upr}
\end{equation}
\subsection{Interpretation of transient-absorption spectra and reconstruction of strong-field interaction operators}
\label{Transient-absorption spectra}
While first-order perturbation theory is used to write $\hat{U}_{\mathrm{pr}}$ explicitly [Eq.~(\ref{eq:Upr})], no assumption is required for the operator $\hat{U}_{\mathrm{pu}}$, describing the interaction with strong pump pulses. We show how its 9 matrix elements are linked to measurable transient-absorption spectra, deriving an analytical fitting model which can be used to extract $\hat{U}_{\mathrm{pu}}$ directly from observable data.
\subsubsection{Probe-pump scheme}
For a probe-pump scheme ($\tau<0$), the weak probe pulse generates the initial excited state, which is nonperturbatively modified by the action of the second-arriving intense pump pulse. In terms of the operators $\hat{U}_{\mathrm{pr}}$ and $\hat{U}_{\mathrm{pu}}(I)$, modeling light-matter interactions as effectively instantaneous transformations, the effective evolution of the time-delay-dependent state $\left|\psi_{\mathrm{fit}}(t,\tau)\right\rangle = \sum_{i=1}^3 c_i(t,\tau)|i\rangle$ from the effective initial state $|\psi_0\rangle = |1\rangle$ is given by
\begin{equation}
|\psi_{\mathrm{fit}}(t,\tau)\rangle = \left\{
\begin{aligned}
&|\psi_0\rangle, &\text{if $t<\tau$},\\
&\hat{V}(t-\tau)\hat{U}_{\mathrm{pr}}|\psi_0\rangle, &\text{if $\tau<t<0$},\\
&\hat{V}(t)\hat{U}_{\mathrm{pu}}(I)\,\hat{V}(-\tau)\,\hat{U}_{\mathrm{pr}}|\psi_0\rangle, &\text{if $t>0$}.
\end{aligned}
\right.
\label{eq:evolutionprobepump}
\end{equation}
We can analogously define the effective density-matrix elements $\rho_{\mathrm{fit},ij}(t,\tau)= c_i(t,\tau)\,[c_j(t,\tau)]^*$, in terms of the components $c_{i}(t,\tau)$ of $\left|\psi_{\mathrm{fit}}(t,\tau)\right\rangle$. By inserting $\rho_{\mathrm{fit},ij}(t,\tau)$ into Eq.~(\ref{eq:spectrum3}), an analytical interpretation model for the probe-pump spectrum is obtained as
\begin{equation}
\begin{split}
&\mathcal{S}_{\mathrm{fit}}(\omega,\tau,U_{\mathrm{pu},ij}) \\
\propto\,& -\omega\mathop{}\!\mathbf{Im}\biggl\{\sum_{k=2}^3 D_{1k}^* \biggr[\int_{\tau}^{0}\rho_{\mathrm{fit},1k}(t,\tau)\,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)}\,\mathop{}\!\mathrm{d} t \\
&\ \ \ \ \ \ \ \ \ \ \ \ + \int_{0}^{\infty}\rho_{\mathrm{fit},1k}(t,\tau)\,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)}\,\mathop{}\!\mathrm{d} t\biggr]\biggr\}\,.
\end{split}
\label{eq:spprpu}
\end{equation}
The first integral in Eq.~(\ref{eq:spprpu}) is equal to
\begin{equation}
\begin{split}
&\,\int_{\tau}^{0}\rho_{\mathrm{fit},1k}(t,\tau)\,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)}\,\mathop{}\!\mathrm{d} t \\
=&\,U_{\mathrm{pr,}11}\, U^*_{\mathrm{pr,}k1} \int_{\tau}^{0} V^*_{kk}(t-\tau) \,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)}\,\mathop{}\!\mathrm{d} t \\
= &\, -\mathrm{i}\frac{\vartheta_k}{2}\,\frac{1-\mathrm{e}^{\mathrm{i}(\omega - \omega_{k1}) \tau}\,\mathrm{e}^{\frac{\gamma_k}{2}\tau}}{\mathrm{i}(\omega - \omega_{k1}) + \frac{\gamma_k}{2}},
\end{split}
\label{eq:spprpu1stcomponent}
\end{equation}
which, for $\omega \approx \omega_{k1}$, does not feature fast oscillations as a function of the time delay $\tau$. The second integral in Eq.~(\ref{eq:spprpu}) is given by
\begin{equation}
\begin{split}
&\,\int_{0}^{\infty}\rho_{\mathrm{fit},1k}(t,\tau)\,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)}\,\mathop{}\!\mathrm{d} t \\
=&\,\sum_{j,j' =1}^3U_{\mathrm{pu},1j}\,U_{\mathrm{pu},kj'}^*\,V_{jj}(-\tau)\,V_{j'j'}^*(-\tau)\,U_{\mathrm{pr},j1}\,U_{\mathrm{pr},j'1}^*\\
&\times\,\int_{0}^{\infty} V_{kk}^*(t)\,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)} \,\mathop{}\!\mathrm{d} t \\
= &\ \frac{1}{\,\mathrm{i}(\omega - \omega_{k1}) + \frac{\gamma_k}{2}}\sum_{j,j' =1}^3U_{\mathrm{pu},1j}\,U_{\mathrm{pu},kj'}^*\,\\
&\ \ \times\,\left[V_{jj}(-\tau)\,V_{j'j'}^*(-\tau)\,\mathrm{e}^{\mathrm{i}\omega\tau}\right]\,U_{\mathrm{pr},j1}\,U_{\mathrm{pr},j'1}^*.
\end{split}
\label{eq:spprpu2ndcomponent}
\end{equation}
Firstly, all terms in the above sum depend upon nonvanishing elements of the probe-pulse operator $\hat{U}_{\mathrm{pr}}$ [Eq.~(\ref{eq:Upr})]. In a collinear geometry all terms in the above sum contribute to the resulting transient-absorption spectrum, which can be used to extract SFI matrix-element products such as $U_{\mathrm{pu}, 1j}\,U_{\mathrm{pu}, kj'}$. In a noncollinear geometry, however, such as the one utilized in Ref.~\cite{PhysRevLett.115.033003}, fast oscillating terms due to $[V_{jj}(-\tau)\,V_{j'j'}^*(-\tau)\,\mathrm{e}^{\mathrm{i}\omega\tau}]$ are averaged out and do not contribute to the resulting spectra. Equation~(\ref{eq:spprpu2ndcomponent}) can be employed to recognize and eliminate all terms which, at $\omega\approx \omega_{k1}$, feature fast oscillations in $\tau$. As a result,
in terms of the two functions
\begin{equation}
\begin{split}
A_2(\tau) &= U_{\mathrm{pu,}11}\,U^*_{\mathrm{pu,}22} + U_{\mathrm{pu,}11}\,U^*_{\mathrm{pu,}23}\, \frac{\vartheta_{3}}{\vartheta_{2}}\, \mathrm{e}^{-\mathrm{i}\omega_{32}\tau}\,\mathrm{e}^{\frac{\gamma_3-\gamma_2}{2}\tau} ,\\
A_3(\tau) &= U_{\mathrm{pu,}11}\,U^*_{\mathrm{pu,}33}+ U_{\mathrm{pu,}11}\,U^*_{\mathrm{pu,}32}\,\frac{\vartheta_{2}}{\vartheta_{3}}\, \mathrm{e}^{\mathrm{i}\omega_{32}\tau}\,\mathrm{e}^{\frac{\gamma_2-\gamma_3}{2}\tau} ,
\end{split}
\end{equation}
average experimental spectra can be modeled by
\begin{equation}
\begin{split}
\langle \mathcal{S}_{\mathrm{fit}}(\omega,\tau,U_{\mathrm{pu},ij})\rangle_{\tau} & \propto-\omega\mathop{}\!\mathbf{Im}\biggl\{\sum_{k=2}^3\frac{-\mathrm{i}\,D_{1k}^*\frac{\vartheta_k}{2}}{\mathrm{i}(\omega - \omega_{k1}) + \frac{\gamma_k}{2}}\\
&\ \times \bigl[1+\mathrm{e}^{\mathrm{i}(\omega - \omega_{k1})\tau}\,\mathrm{e}^{\frac{\gamma_k}{2}\tau}\, (A_k(\tau)-1)\bigr]\biggr\}.
\end{split}
\label{eq:Sexpprpu}
\end{equation}
The proportionality symbol $\propto$ stresses that these spectra depend on a density-dependent multiplication factor $K$, which we treat here as a fitting parameter, along with the SFI matrix-element products $U_{\mathrm{pu,}11}\,U^*_{\mathrm{pu,}22}$, $U_{\mathrm{pu,}11}\,U^*_{\mathrm{pu,}33}$, $U_{\mathrm{pu,}11}\,U^*_{\mathrm{pu,}32}$, and $U_{\mathrm{pu,}11}\,U^*_{\mathrm{pu,}23}$.
\subsubsection{Pump-probe scheme}
For a pump-probe scheme ($\tau>0$), in terms of the operators $\hat{U}_{\mathrm{pr}}$ and $\hat{U}_{\mathrm{pu}}(I)$, the effective evolution of the time-delay-dependent state $\left|\psi_{\mathrm{fit}}(t,\tau)\right\rangle $ from the effective initial state $|\psi_0\rangle = |1\rangle$ can be modeled as
\begin{equation}
|\psi_{\mathrm{fit}}(t,\tau)\rangle = \left\{
\begin{aligned}
&|\psi_0\rangle, &\text{if $t<0$},\\
&\hat{V}(t)\hat{U}_{\mathrm{pu}}(I)|\psi_0\rangle, &\text{if $0<t<\tau$},\\
&\hat{V}(t-\tau)\hat{U}_{\mathrm{pr}}\,\hat{V}(\tau)\,\hat{U}_{\mathrm{pu}}(I)|\psi_0\rangle, &\text{if $t>\tau$}.
\end{aligned}
\right.
\label{eq:evolutionpumpprobe}
\end{equation}
By including in Eq.~(\ref{eq:spectrum3}) the effective evolution of the matrix elements $\rho_{\mathrm{fit},ij}(t,\tau)$ from Eq.~(\ref{eq:evolutionpumpprobe}), an analytical interpretation model for the pump-probe spectrum is derived, which can be split into the following sum:
\begin{equation}
\begin{split}
&\mathcal{S}_{\mathrm{fit}}(\omega,\tau, U_{\mathrm{pu},ij}) \\
\propto\,& -\omega\mathop{}\!\mathbf{Im}\biggl\{\sum_{k=2}^3 D_{1k}^* \biggr[\int_{0}^{\tau}\rho_{\mathrm{fit},1k}(t,\tau)\,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)}\,\mathop{}\!\mathrm{d} t \\
&\ \ \ \ \ \ \ \ \ + \int_{\tau}^{\infty}\rho_{\mathrm{fit},1k}(t,\tau)\,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)}\,\mathop{}\!\mathrm{d} t\biggr]\biggr\}\,.
\end{split}
\label{eq:sppupr}
\end{equation}
The first integral in Eq.~(\ref{eq:sppupr}) is equal to
\begin{equation}
\begin{split}
&\,\int_{0}^{\tau}\rho_{\mathrm{fit},1k}(t,\tau)\,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)}\,\mathop{}\!\mathrm{d} t \\
=&\,U_{\mathrm{pu,}11}\,U^*_{\mathrm{pu,}k1}\,\int_{0}^{\tau}V_{kk}^*(t)\,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)}\,\mathop{}\!\mathrm{d} t \\
= &\, U_{\mathrm{pu,}11}\,U^*_{\mathrm{pu,}k1}\,\mathrm{e}^{\mathrm{i}\omega \tau}\,\frac{1-\mathrm{e}^{-\mathrm{i}(\omega - \omega_{k1}) \tau}\,\mathrm{e}^{-\frac{\gamma_k}{2}\tau}}{\mathrm{i}(\omega - \omega_{k1}) + \frac{\gamma_k}{2}}.
\end{split}
\label{eq:sppupr1stcomponent}
\end{equation}
As described in the case of the probe-pump scheme, the fast oscillations at frequencies $\omega \approx \omega_{k1}$ are averaged out in a noncollinear geometry, and this first integral does not contribute to the associated average absorption spectrum. The second integral in Eq.~(\ref{eq:sppupr}) is given by
\begin{equation}
\begin{split}
&\,\int_{\tau}^{\infty}\rho_{\mathrm{fit},1k}(t,\tau)\,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)}\,\mathop{}\!\mathrm{d} t \\
= &\,\sum_{j,j' =1}^3U_{\mathrm{pr},1j}\,U_{\mathrm{pr},kj'}^*\,V_{jj}(\tau)\,V_{j'j'}^*(\tau)\,U_{\mathrm{pu},j1}\,U_{\mathrm{pu},j'1}^*\\
&\times\,\int_{\tau}^{\infty} V_{kk}^*(t-\tau)\,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)} \,\mathop{}\!\mathrm{d} t \\
= &\,\frac{1}{\,\mathrm{i}(\omega - \omega_{k1}) + \frac{\gamma_k}{2}} \sum_{j,j' =1}^3U_{\mathrm{pr},1j}\,U_{\mathrm{pr},kj'}^*\,\\
&\times\,\left[V_{jj}(\tau)\,V_{j'j'}^*(\tau)\right]\,U_{\mathrm{pu},j1}\,U_{\mathrm{pu},j'1}^*.
\end{split}
\label{eq:sppupr2ndcomponent}
\end{equation}
By using Eq.~(\ref{eq:Upr}) and removing fast oscillating terms due to $V_{jj}(\tau)\,V_{j'j'}^*(\tau)$, which would not appear in a noncollinear geometry, we conclude that
\begin{equation}
\begin{split}
&\langle \mathcal{S}_{\mathrm{fit}}(\omega,\tau, U_{\mathrm{pu},ij})\rangle_{\tau} \\
\propto &-\omega\mathop{}\!\mathbf{Im}\biggl\{\frac{D_{12}^*}{\mathrm{i}(\omega - \omega_{21}) + \frac{\gamma_2}{2}}\biggl[ \Bigl( -\mathrm{i}\frac{\vartheta_{2}}{2}\, |U_{\mathrm{pu,}11}|^2\\
&\ \ \ \ \ \ \ \ \ +\mathrm{i}\frac{\vartheta_{2}}{2}\, |U_{\mathrm{pu,}21}|^2 \,\mathrm{e}^{-\gamma_2\tau} \Bigr)\\
&\ \ \ \ \ \ \ \ \ +\mathrm{i}\frac{\vartheta_{3}}{2}\,U_{\mathrm{pu,}31} \,U^*_{\mathrm{pu,}21} \,\mathrm{e}^{-\mathrm{i}\omega_{32}\tau}\,\mathrm{e}^{-\frac{\gamma_2+\gamma_3}{2}\tau} \biggr]\\
&\ \ \ \ \ + \frac{D_{13}^*}{\mathrm{i}(\omega - \omega_{31}) + \frac{\gamma_3}{2}}\biggl[ \Bigl( -\mathrm{i}\frac{\vartheta_{3}}{2}\, |U_{\mathrm{pu,}11}|^2\\
&\ \ \ \ \ \ \ \ \ +\mathrm{i}\frac{\vartheta_{3}}{2}\, |U_{\mathrm{pu,}31}|^2 \,\mathrm{e}^{-\gamma_3\tau}\Bigr)\\
&\ \ \ \ \ \ \ \ \ +\mathrm{i}\frac{\vartheta_{2}}{2}\, U_{\mathrm{pu,}21}\,U^*_{\mathrm{pu,}31}\,\mathrm{e}^{\mathrm{i}\omega_{32}\tau}\,\mathrm{e}^{-\frac{\gamma_2+\gamma_3}{2}\tau}\biggr]\biggr\},
\end{split}
\label{eq:Sexppupr}
\end{equation}
which explicitly depends upon $|U_{\mathrm{pu,}11}|^2$, $|U_{\mathrm{pu,}22}|^2$, $|U_{\mathrm{pu,}33}|^2$, and $(U_{\mathrm{pu,}31} \,U^*_{\mathrm{pu,}21})$. The same multiplication factor $K$ should be used which was extracted from the probe-pump spectrum [see Eq.~(\ref{eq:Sexpprpu})].
\subsubsection{Additional remarks}
The matrix elements of the SFI operator $\hat{U}_{\mathrm{pu}}$ are reconstructed by using the above formulas to fit experimentally measurable transient-absorption spectra. In particular, one can fit probe-pump spectra to quantify $U_{\mathrm{pu},11}\,U_{\mathrm{pu},22}^*$, $U_{\mathrm{pu},11}\,U_{\mathrm{pu},33}^*$, $U_{\mathrm{pu},11}\,U_{\mathrm{pu},32}^*$ and $U_{\mathrm{pu},11}\,U_{\mathrm{pu},23}^*$, and the common multiplication factor $K$. Once $K$ is known, it can be employed to fit pump-probe spectra, and thereby extract $|U_{\mathrm{pu},11}|^2$, $|U_{\mathrm{pu},21}|^2$, $|U_{\mathrm{pu},31}|^2$, and $(U_{\mathrm{pu,}31} \,U^*_{\mathrm{pu,}21})$.
All elements of the interaction operator $\hat{U}_{\mathrm{pu}}$ which can be retrieved from probe-pump spectra are inferred from product terms $U_{\mathrm{pu},11} U_{\mathrm{pu}, ij}^*$, with $|U_{\mathrm{pu},11}|$ coming from pump-probe spectra. This has two consequences. Firstly, the phase $\beta$ of $U_{\mathrm{pu},11} = |U_{\mathrm{pu},11}|\mathrm{e}^{\mathrm{i}\beta}$ cannot be accessed, resulting in SFI matrix elements
\begin{equation}
U_{\mathrm{pu}, ij} = \frac{(U_{\mathrm{pu},11} U_{\mathrm{pu}, ij}^*)^*}{|U_{\mathrm{pu},11}|}\mathrm{e}^{\mathrm{i}\beta},
\label{eq:ratio}
\end{equation}
known up to this common phase $\beta$. However, this is not a limitation, since it only implies that the final state can be measured and controlled up to a nonrelevant phase term, with access to the information about relevant relative phases. Secondly, when $|U_{\mathrm{pu}, 11}|$ is very close to $0$, small uncertainties in its reconstructed value are amplified when the division in Eq.~(\ref{eq:ratio}) is performed to retrieve the remaining matrix elements. For a certain range of pump-pulse intensities, we verified that this is indeed the main source of uncertainty in the extraction of $\hat{U}_{\mathrm{pu}}$, yet always below the relative level of $8\%$.
In a noncollinear geometry, the remaining pump-probe fitting parameter $(U_{\mathrm{pu,}31} \,U^*_{\mathrm{pu,}21})$ can be used to quantify the relative phase $\arg{(U_{\mathrm{pu,}31})} - \arg{(U_{\mathrm{pu,}21})}$, but not the absolute phases of $U_{\mathrm{pu,}21}$ and $U_{\mathrm{pu,}31}$---at least up to the same common phase $\beta$ we introduced before. However, this can be easily circumvented by directly observing the absorption spectrum of a single intense pump pulse. The associated spectral lines are given by
\begin{equation}
\mathcal{S}_{\mathrm{fit}}(\omega)
\propto\,-\omega\mathop{}\!\mathbf{Im}\biggl\{\mathcal{M} + \sum_{k=2}^3 D_{1k}^* \int_{0}^{\infty}\rho_{\mathrm{fit},1k}(t)\,\mathrm{e}^{-\mathrm{i}\omega t}\,\mathop{}\!\mathrm{d} t \biggr\}\,,
\label{eq:Sexpnopr}
\end{equation}
where $\mathcal{M}$ is a fitting parameter modeling the broadband (and hence constant for small frequency intervals) Fourier transform of $\rho^{(3)}_{1k}(t)$ in the interval $[-T/2,\,T/2]$, i.e., in the presence of the pump pulse. The remaining integral
\begin{equation}
\begin{split}
&\,\int_{0}^{\infty}\rho_{\mathrm{fit},1k}(t)\,\mathrm{e}^{-\mathrm{i}\omega t }\,\mathop{}\!\mathrm{d} t \\
= &\,U_{\mathrm{pu},11}\,U_{\mathrm{pu},k1}^*\int_{0}^{\infty} \,\mathrm{e}^{-\mathrm{i}(\omega - \omega_{k1})t} \,\mathrm{e}^{-\frac{\gamma_k}{2}t}\,\mathop{}\!\mathrm{d} t \\
= &\,U_{\mathrm{pu},11}\,U_{\mathrm{pu},k1}^*\,\frac{1}{\mathrm{i}(\omega - \omega_{k1}) + \frac{\gamma_k}{2}}
\end{split}
\label{eq:integral1pulse}
\end{equation}
exploits the effective evolution of the system, given by
\begin{equation}
|\psi_{\mathrm{fit}}(t)\rangle = \left\{
\begin{aligned}
&|\psi_0\rangle, &\text{if $t<0$},\\
&\hat{V}(t)\hat{U}_{\mathrm{pu}}(I)|\psi_0\rangle, &\text{if $t>0$}.
\end{aligned}
\right.
\label{eq:evolutionnoprobe}
\end{equation}
Up to the same common phase $\beta$ already mentioned above, this allows one to retrieve the phases of $U_{\mathrm{pu},21}$ and $U_{\mathrm{pu},31}$, complementing the amplitude information accessible by fitting pump-probe spectra.
In order to independently extract $U_{\mathrm{pu},12}$ and $U_{\mathrm{pu},13}$ in a noncollinear geometry, we show that, when the envelope function $f(t) = f(-t)$ is a symmetric function of time, then the associated interaction operator $\hat{U} = \hat{V}(-T/2)\,\hat{\mathcal{U}}_0(T/2, -T/2)\,\hat{V}^{-1}(T/2)$ is a symmetric matrix. We recall that the time evolution of $\hat{\mathcal{U}}_0(t,t_0)$ is calculated for a pulse with $\phi = 0$ and $t_{\mathrm{c}} = 0$, such that
\begin{equation}
\begin{split}
\frac{\mathop{}\!\mathrm{d} \hat{\mathcal{U}}_0(t,t_0)}{\mathop{}\!\mathrm{d} t} &= [\hat{\varLambda}^{(3)}(t)]^{\dagger}\,\hat{M}^{(3)}(t)\,\hat{\varLambda}^{(3)}(t)\,\hat{\mathcal{U}}(t,t_0),\\
\hat{\mathcal{U}}(t_0,t_0) &= \hat{I},
\end{split}
\label{eq:timeevolution3U}
\end{equation}
compare this with Eq.~(\ref{eq:timeevolution5}).
The assumption of a symmetric envelope, $f(t) = f(-t)$, implies that also $ \hat{M}^{(3)}(t)$ is symmetric in time.
We thus introduce the operator $\hat{\mathcal{Z}}(t,t_0) = \hat{\mathcal{U}}_0(-t, -t_0)$, which is solution of the differential equation
\begin{equation}
\begin{split}
\frac{\mathop{}\!\mathrm{d} \hat{\mathcal{Z}}(t,t_0)}{\mathop{}\!\mathrm{d} t} &= -\hat{\varLambda}^{(3)}(t)\,\hat{M}^{(3)}(t)\,[\hat{\varLambda}^{(3)}(t)]^{-1}\,\hat{\mathcal{Z}}(t,t_0),\\
\hat{\mathcal{Z}}(t_0,t_0) &= \hat{I},
\end{split}
\label{eq:evolutionZ}
\end{equation}
where $\hat{\varLambda}^{(3)}(-t) = [\hat{\varLambda}^{(3)}(t)]^{-1}$ and $\hat{M}^{(3)}(-t) = \hat{M}^{(3)}(t)$. Both $\hat{\varLambda}^{(3)}(t) = [\hat{\varLambda}^{(3)}(t)]^{\mathrm{T}}$ and $\hat{M}^{(3)}(t) = [\hat{M}^{(3)}(t)]^{\mathrm{T}}$ are symmetric matrices. As a result,
\begin{equation}
\begin{split}
\frac{\mathop{}\!\mathrm{d} \hat{\mathcal{Z}}^{-1}(t,t_0)}{\mathop{}\!\mathrm{d} t}
&= \hat{\mathcal{Z}}^{-1}(t,t_0)\hat{\varLambda}^{(3)}(t)\,\hat{M}^{(3)}(t)\,[\hat{\varLambda}^{(3)}(t)]^{-1},\\
\hat{\mathcal{Z}}^{-1}(t_0,t_0) & = \hat{I},
\end{split}
\label{eq:evolutionZ-1}
\end{equation}
and
\begin{equation}
\begin{split}
\frac{\mathop{}\!\mathrm{d} (\hat{\mathcal{Z}}^{-1})^{\mathrm{T}}(t,t_0)}{\mathop{}\!\mathrm{d} t}
&\,=[\hat{\varLambda}^{(3)}(t)]^{-1}\,\hat{M}^{(3)}(t)\,\hat{\varLambda}^{(3)}(t)\,(\hat{\mathcal{Z}}^{-1})^{\mathrm{T}}(t,t_0),\\
(\hat{\mathcal{Z}}^{-1})^{\mathrm{T}}(t_0,t_0) &\,= \hat{I}.
\end{split}
\label{eq:evolutionZ-1T}
\end{equation}
Since this equation corresponds to Eq.~(\ref{eq:timeevolution3U}), the evolution operators $(\hat{\mathcal{Z}}^{-1})^{\mathrm{T}}(t,t_0)$ and $\hat{\mathcal{U}}_0(t, t_0)$ are solution of the same differential equations and are therefore identical, and hence $\hat{\mathcal{U}}_0(T/2,0) = \hat{\mathcal{U}}_0^{-1}(0, T/2) = \hat{\mathcal{Z}}^{-1}(0,-T/2) = \hat{\mathcal{U}}_0^{\mathrm{T}}(0,-T/2)$.
As a result,
\begin{equation}
\begin{split}
\hat{\mathcal{U}}_0(T/2,-T/2) &= \hat{\mathcal{U}}_0(T/2,0)\,\hat{\mathcal{U}}_0(0,-T/2)\\
&=\hat{\mathcal{U}}_0^{\mathrm{T}}(0,-T/2) \,\hat{\mathcal{U}}_0(0,-T/2),
\end{split}
\end{equation}
and
\begin{equation}
\begin{split}
\hat{\mathcal{U}}_0^{\mathrm{T}}(T/2,-T/2) &= \hat{\mathcal{U}}_0^{\mathrm{T}}(0,-T/2) \,\hat{\mathcal{U}}_0(0,-T/2) \\
&= \hat{\mathcal{U}}_0(T/2,-T/2),
\end{split}
\end{equation}
i.e., $\hat{\mathcal{U}}_0(T/2,-T/2) $ is a symmetric matrix. Since $\hat{V}(t)$ is a diagonal (and hence symmetric) matrix, it follows that $\hat{U} = \hat{V}(-T/2)\,\hat{\mathcal{U}}_0(T/2, -T/2)\,\hat{V}^{-1}(T/2)$ is symmetric.
We hence use the above result for the operator $\hat{U}_{\mathrm{pu}}$ reconstructed from strong-field TAS, and fix $U_{\mathrm{pu},1k} = U_{\mathrm{pu},k1}$. We stress that this is not a disadvantage of the operator-reconstruction scheme proposed here. As shown by Eq.~(\ref{eq:spprpu2ndcomponent}), an independent extraction of $U_{\mathrm{pu},12}$ and $U_{\mathrm{pu},13}$, without additional assumptions on the envelope $f(t)$, is possible in a collinear geometry.
{
\section{Density-matrix formulation}
\label{Density-matrix formulation}
The equations of motion have been investigated thus far in the framework of the Schr\"odinger equation, assuming a decay-rate model with a complex-valued atomic structure Hamiltonian. For our implementation with Rb atoms, this was a proper treatment, in the light of the very different time scales in which the strong-field dynamics and the relaxation of the system take place.
For more general cases, however, a density-matrix formulation of our operator-reconstruction and quantum-control scheme is possible, as we show in the following. In this section, we will focus on the case of a three-level system, although the equations can be easily generalized to higher numbers of levels if necessary.
We describe the system in terms of a $3\times 3$ density matrix $\hat{\rho}(t)$, with a diagonal element $\rho_{ii}(t)$ associated with the population of the $i$th level, and off-diagonal matrix elements for the atomic coherences. A density-matrix formulation may be necessary, e.g., to explicitly account for the spontaneous decay of the two excited levels $2$ and $3$ into the ground state $1$, at rates respectively given by $\gamma_{22}$ and $\gamma_{33}$. Similarly, it is necessary to account for additional decoherence processes affecting the atomic coherences $\rho_{12}$, $\rho_{13}$, and $\rho_{23}$. In this Section, this is modeled by the decoherence rates $\gamma_{21}$, $\gamma_{31}$, and $\gamma_{23}$, respectively. The evolution of the density matrix $\hat{\rho}(t)$ is then given by the master equation
\begin{equation}
\frac{\mathop{}\!\mathrm{d} \hat{\rho}(t)}{\mathop{}\!\mathrm{d} t} = \mathrm{i} [\hat{\rho}(t),\,\hat{H}_0 + \hat{H}_{\mathrm{int}}] + \mathcal{L}[\hat{\rho}(t)],
\end{equation}
with the atomic structure Hamiltonian now given by $\hat{H}_ 0 = \sum \omega_i |i\rangle\langle i|$ and the superoperator $\mathcal{L}[\hat{\rho}(t)]$ describing dissipation. In the absence of external fields, the equations of motion satisfied by the elements of the density matrix can hence be written as:
\begin{equation}
\begin{split}
\frac{\mathop{}\!\mathrm{d} \rho_{11}}{\mathop{}\!\mathrm{d} t} &= \gamma_{22}\rho_{22} + \gamma_{33}\rho_{33},\\
\frac{\mathop{}\!\mathrm{d} \rho_{kk}}{\mathop{}\!\mathrm{d} t} &= -\gamma_{kk}\rho_{kk},\\
\frac{\mathop{}\!\mathrm{d} \rho_{1k}}{\mathop{}\!\mathrm{d} t} &= - \gamma_{k1}\rho_{1k} + \mathrm{i} \omega_{k1} \rho_{1k},\\
\frac{\mathop{}\!\mathrm{d} \rho_{23}}{\mathop{}\!\mathrm{d} t} &= - \gamma_{23}\rho_{23} + \mathrm{i} \omega_{32} \rho_{23},
\end{split}
\end{equation}
with $k\in\{2,3\}$ and $\rho_{ij} = \rho_{ji}^*$. The explicit solution reads
\begin{equation}
\begin{split}
\rho_{11}(t) &= \rho_{11}(0) + \rho_{22}(0)\left(1 - \mathrm{e}^{-\gamma_{22}t} \right) \\
&\ \ \ + \rho_{33}(0)\left(1 - \mathrm{e}^{-\gamma_{33}t} \right),\\
\rho_{kk}(t) &= \rho_{kk}(0) \mathrm{e}^{-\gamma_{kk}t},\\
\rho_{1k}(t) &= \rho_{1k}(0) \mathrm{e}^{ - \gamma_{k1}t} \mathrm{e}^{\mathrm{i} \omega_{k1} t},\\
\rho_{23}(t) &= \rho_{23}(0) \mathrm{e}^{ - \gamma_{23}t} \mathrm{e}^{\mathrm{i} \omega_{32} t},
\end{split}
\end{equation}
summarized by the elements of the operator $\boldsymbol{X}(t)$ via
\begin{equation}
\rho_{ij}(t) = \sum_{i'j'}\boldsymbol{X}_{ij,i'j'}(t)\rho_{i'j'}(0).
\end{equation}
In analogy to Eq.~(1) in the article, we can introduce an operator $\boldsymbol{Y}$ describing the effect of ultrashort pump and probe pulses on the density matrix of the system as an instantaneous interaction, i.e.,
\begin{equation}
\rho^+ = \boldsymbol{Y}(I)\rho^-.
\end{equation}
For ultrashort pulses, acting on a time scale much shorter than the time scale of the decay processes, we can assume that $\boldsymbol{Y}(I) = \hat{U}(I)\otimes \hat{U}^*(I)$, i.e., $\boldsymbol{Y}_{ij,i'j'} = U_{ii'}\,U_{jj'}^*$. In contrast to the previous Sections, however, here we fully account for the relaxation of the system during its free evolution in terms of the operator $\boldsymbol{X}(t)$. A similar approach can be used to model quantum systems also in the presence of a more structured reservoir, provided that the free evolution of the system is known and can be used to define the operator $\boldsymbol{X}(t)$.
For a TAS experiment in a probe-pump setup, we can describe the effective evolution of the time-delay-dependent density matrix $\hat{\rho}_{\mathrm{fit}}(t,\tau)$ from the effective initial state $\rho_0 = |1\rangle\langle 1|$ as
\begin{equation}
\hat{\rho}_{\mathrm{fit}}(t,\tau) = \left\{
\begin{aligned}
&\hat{\rho}_0, &\text{if $t<\tau$},\\
&\boldsymbol{X}(t-\tau)\boldsymbol{Y}_{\mathrm{pr}}\hat{\rho}_0, &\text{if $\tau<t<0$},\\
&\boldsymbol{X}(t)\boldsymbol{Y}_{\mathrm{pu}}(I)\,\boldsymbol{X}(-\tau)\,\boldsymbol{Y}_{\mathrm{pr}}\hat{\rho}_0, &\text{if $t>0$}.
\end{aligned}
\right.
\end{equation}
By inserting $\rho_{\mathrm{fit},ij}(t,\tau)$ into Eq.~(\ref{eq:spectrum3}), an analytical interpretation model for the probe-pump spectrum is obtained consisting also in this case of two integrals. The first integral reads
\begin{equation}
\begin{split}
&\,\int_{\tau}^{0}\rho_{\mathrm{fit},1k}(t,\tau)\,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)}\,\mathop{}\!\mathrm{d} t \\
=&\int_{\tau}^{0} \sum_{i,j=1}^3\boldsymbol{X}_{1k,ij}(t-\tau)\,\boldsymbol{Y}_{\mathrm{pr};ij,11} \,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)}\,\mathop{}\!\mathrm{d} t \\
= &\, -\mathrm{i}\frac{\vartheta_k}{2}\,\frac{1-\mathrm{e}^{\mathrm{i}(\omega - \omega_{k1}) \tau}\,\mathrm{e}^{\gamma_{k1}\tau}}{\mathrm{i}(\omega - \omega_{k1}) + \gamma_{k1}},
\end{split}
\end{equation}
where we have used the fact that $\boldsymbol{X}_{1k,ij}(t) = \mathrm{e}^{-\gamma_{k1}t}\,\mathrm{e}^{\mathrm{i}\omega_{k1}t}\,\delta_{1i}\,\delta_{kj}$. The second integral is similarly given by
\begin{equation}
\begin{split}
&\,\int_{0}^{\infty}\rho_{\mathrm{fit},1k}(t,\tau)\,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)}\,\mathop{}\!\mathrm{d} t \\
=&\,\sum_{i,j =1}^3\sum_{i',j' =1}^3\sum_{i'',j'' =1}^3 \boldsymbol{Y}_{\mathrm{pu};ij,i'j'}\,\boldsymbol{X}_{i'j',i''j''}(-\tau)\,\boldsymbol{Y}_{\mathrm{pr},i''j'',11}\\
&\times\,\int_{0}^{\infty} \boldsymbol{X}_{1k,ij}(t)\,\mathrm{e}^{-\mathrm{i}\omega (t-\tau)} \,\mathop{}\!\mathrm{d} t \\
= &\,\frac{1}{\,\mathrm{i}(\omega - \omega_{k1}) + \gamma_{k1}}\sum_{i',j' =1}^3\sum_{i'',j'' =1}^3 U_{\mathrm{pu},1i'}\,U_{\mathrm{pu},kj'}^*\,\\
&\ \ \times\,\left[\boldsymbol{X}_{i'j',i''j''}(-\tau)\,\mathrm{e}^{\mathrm{i}\omega\tau}\right]\, U_{\mathrm{pr},i''1}\,U_{\mathrm{pr},j''1}^*.
\end{split}
\end{equation}
By including explicitly the value of $\boldsymbol{X}_{i'j',i''j''}(-\tau)$, one obtains an equation in which, in contrast to Eq.~(\ref{eq:spprpu2ndcomponent}), different decay rates for coherences and populations are present and additional terms appear due to the spontaneous decay from the excited states to the ground state.
In order to model spectra from a noncollinear geometry, fast oscillating terms due to $[\boldsymbol{X}_{i'j',i''j''}(-\tau)\,\mathrm{e}^{\mathrm{i}\omega\tau}]$ can be eliminated, resulting in the average spectrum
\begin{equation}
\begin{split}
&\langle \mathcal{S}_{\mathrm{fit}}(\omega,\tau,U_{\mathrm{pu},ij})\rangle_{\tau} \\
\propto &-\omega\mathop{}\!\mathbf{Im}\biggl\{\sum_{k=2}^3 D_{1k}^* \,\frac{1}{\mathrm{i}(\omega - \omega_{k1}) + \gamma_{k1}}\\
&\ \ \ \ \ \ \ \biggl[(-\mathrm{i}{\vartheta_k}/{2})\,\bigl(1-\mathrm{e}^{\mathrm{i}(\omega - \omega_{k1}) \tau}\,\mathrm{e}^{\gamma_{k1}\tau}\bigr)\\
&\ \ \ \ \ \ \ \ + U_{\mathrm{pu},11}\,U_{\mathrm{pu},k2}^*\, (-\mathrm{i}{\vartheta_2}/{2})\,\mathrm{e}^{\mathrm{i}(\omega-\omega_{21})\tau}\,\mathrm{e}^{\gamma_{21}\tau}\\
&\ \ \ \ \ \ \ \ \ +U_{\mathrm{pu},11}\,U_{\mathrm{pu},k3}^*\, (-\mathrm{i}{\vartheta_3}/{2})\, \mathrm{e}^{\mathrm{i}(\omega-\omega_{31})\tau}\,\mathrm{e}^{\gamma_{31}\tau} \biggr]\biggr\}.
\end{split}
\end{equation}
In a pump-probe setup, the same formalism can be used, resulting in an effective evolution given by
\begin{equation}
\hat{\rho}_{\mathrm{fit}}(t,\tau) = \left\{
\begin{aligned}
&\hat{\rho}_0, &\text{if $t<0$},\\
&\boldsymbol{X}(t)\,\boldsymbol{Y}_{\mathrm{pu}}(I)\,\hat{\rho}_0, &\text{if $0<t<\tau$},\\
&\boldsymbol{X}(t-\tau)\,\boldsymbol{Y}_{\mathrm{pr}}\,\boldsymbol{X}(\tau)\,\boldsymbol{Y}_{\mathrm{pu}}(I)\,\hat{\rho}_0, &\text{if $t>\tau$}.
\end{aligned}
\right.
\end{equation}
By following the same procedure outlined for the probe-pump case, and after removing the fast oscillating terms, one obtains the average spectrum given by
\begin{equation}
\begin{split}
&\langle \mathcal{S}_{\mathrm{fit}}(\omega,\tau, U_{\mathrm{pu},ij})\rangle_{\tau} \\
\propto &-\omega\mathop{}\!\mathbf{Im}\biggl\{\frac{D_{12}^*}{\mathrm{i}(\omega - \omega_{21}) + \gamma_{21}}\biggl[ \Bigl( -\mathrm{i}\frac{\vartheta_{2}}{2}\, |U_{\mathrm{pu,}11}|^2\\
&\ \ \ \ \ \ \ \ \ +\mathrm{i}\frac{\vartheta_{2}}{2}\, |U_{\mathrm{pu,}21}|^2 \,(2\mathrm{e}^{-\gamma_{22}\tau}-1) \Bigr)\\
&\ \ \ \ \ \ \ \ \ +\mathrm{i}\frac{\vartheta_{3}}{2}\,U_{\mathrm{pu,}31} \,U^*_{\mathrm{pu,}21} \,\mathrm{e}^{-\mathrm{i}\omega_{32}\tau}\,\mathrm{e}^{-\gamma_{23}\tau} \biggr]\\
&\ \ \ \ \ + \frac{D_{13}^*}{\mathrm{i}(\omega - \omega_{31}) + \gamma_{31}}\biggl[ \Bigl( -\mathrm{i}\frac{\vartheta_{3}}{2}\, |U_{\mathrm{pu,}11}|^2\\
&\ \ \ \ \ \ \ \ \ +\mathrm{i}\frac{\vartheta_{3}}{2}\, |U_{\mathrm{pu,}31}|^2 \,(2\mathrm{e}^{-\gamma_{33}\tau}-1)\Bigr)\\
&\ \ \ \ \ \ \ \ \ +\mathrm{i}\frac{\vartheta_{2}}{2}\, U_{\mathrm{pu,}21}\,U^*_{\mathrm{pu,}31}\,\mathrm{e}^{\mathrm{i}\omega_{32}\tau}\,\mathrm{e}^{-\gamma_{23}\tau}\biggr]\biggr\}.
\end{split}
\end{equation}
Also in this case, in contrast to Eq.~(\ref{eq:Sexppupr}), we notice different decay rates and the presence of additional terms to account for the decay into the ground state 1.
This model can be employed to fit experimental results and extract interaction operators $\boldsymbol{Y}$ in those cases in which a density-matrix description of the free evolution of the system is required. Once SFI operators are reconstructed, then the sequence of pulses constituting our control method can be modeled as
\begin{equation}
\hat{\rho}_{N_{\mathrm{p}}} = \tilde{\boldsymbol{Y}}_{N_{\mathrm{p}}} \ldots \boldsymbol{X}(\bar{\tau}_m)\tilde{\boldsymbol{Y}}_{m}\ldots \boldsymbol{X}(\bar{\tau}_1)\tilde{\boldsymbol{Y}}_1\hat{\rho}_0,
\end{equation}
generalizing Eq.~(2) in the article to the case when the system's dynamics have to be described by a density matrix.
}
| 9865cb8237469622bed5f056f8f76ac7d3ff2863 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Proof-of-Concept Applications}
This paper demonstrates backscatter communication on commodity devices, and in this section we evaluate prototypes of three novel applications enabled by these techniques. We evaluate only the communication aspect of our prototypes and consider fully exploring their security and usability issues outside the of scope of this work.
\subsection{Smart Contact Lens}
Smart contact lens systems~\cite{liao20123,yao2012contact} measure biomarkers like glucose, cholesterol and sodium in tears that can help with unobtrusive tracking for diabetic patients. The lens consists of a miniature glucose sensor IC, and an antenna. Although the power required for glucose sensing is minimal, real-time communication is power consuming and rules out conventional radios. Today these systems are limited to periodically sampling and storing the glucose data that is then transmitted sporadically using backscatter whenever the lens is next to a dedicated, powered, RFID-like reader.
We leverage our design to show that a smart contact lens can communicate with commodity Wi-Fi and Bluetooth radios, without the need for dedicated reader hardware. We develop the form factor antenna prototype lens shown in Fig.~\ref{fig:rssi_contact_lens}. The prototype consists of a 1~cm diameter loop antenna similar to prior contact lens systems~\cite{liao20123, yao2012contact}. The antenna was built using 30~AWG wire and then encapsulated in a 200 $\mu$m thick layer of poly-dimethylsiloxane (PDMS) for biocompatibility and structural integrity. We then connect the antenna to our FPGA prototype. Unlike traditional antennas that have a 50 $\Omega$ impedance, small loop antennas have non-standard impedances and so we optimize the impedance of the backscatter switch network for the impedance of our loop antenna. We note that for ease of prototyping, we use FPGA hardware but an IC should conform to the contact lens size form factor, similar to prior backscattering lens designs~\cite{lens1}.
To evaluate the system, we immerse our antenna prototype in contact lens solution. We configure the TI Bluetooth chip to transmit every 40~ms and place it 12 inches from the lens. We use the Intel Wi-Fi card as a receiver and vary its distances to the lens antenna. The RSSIs follow a similar trend when using a Samsung Galaxy S4 as the receiver. We configure our FPGA to detect the Bluetooth beacons and backscatter 2~Mbps Wi-Fi packets. Fig.~\ref{fig:rssi_contact_lens} plots the Wi-Fi RSSI for different Bluetooth transmit power values. The plot shows that we can achieve ranges of more than 24~inches, demonstrating the feasibility of a smart contact lens that communicate directly with commodity radios. We note that the range is smaller than prior plots because the antenna is much smaller and is immersed in liquid leading to high signal attenuation. The range however is similar to prior smart contact lens with a dedicated RFID-like reader~\cite{lens1}.
\begin{figure}[!t]
\subfigure
{
\includegraphics[width=.25\columnwidth]{./figures/contact_ant.eps}
}
\subfigure
{
\includegraphics[width=0.7\columnwidth]{./figures/contact_lens.eps}
}
\vskip -0.1in
\caption{{\bf RSSI with smart contact lens antenna prototype.}}
\vskip -0.15in
\label{fig:rssi_contact_lens}
\end{figure}
\vfill
\subsection{Implanted Neural Recording Devices}
Implantable neural recording devices have recently demonstrated promising results towards use in brain-computer interfaces (BCIs)~\cite{leuthardt2004brain} that help paralyzed individuals operate motor prosthetic devices, command electronic devices, or even regain control of their limbs~\cite{hochberg2006neuronal}. These systems use either penetrating neural probes~\cite{bmi2} or a surface electrode grid~\cite{wyler1984subdural} that is implanted to collect local field potentials and electrocorticography (ECoG) signals~\cite{bmi2}. The recording sensors today consume around 2~$\mu$W/channel~\cite{wyler1984subdural}, with 8--64 parallel recording channels.
\begin{figure}[!t]
\subfigure
{
\includegraphics[width=.25\columnwidth]{./figures/bmi_ant.eps}
}
\subfigure
{
\includegraphics[width=0.7\columnwidth]{./figures/bmi.eps}
}
\vskip -0.1in
\caption{{\bf RSSI with implantable neural recording antenna.}}
\vskip -0.15in
\label{fig:rssi_bmi}
\end{figure}
Instead of using a custom backscatter reader as prior prototypes~\cite{bmi1,bmi2,bmi3}, we leverage our design to transmit the data directly to commodity radios. To demonstrate this, we built a form factor antenna for the implantable neural recording device shown in Fig.~\ref{fig:rssi_bmi}. We construct a 4~cm diameter full wavelength loop antenna using 16 AWG magnet wire and encapsulate it in a 2~mm thick layer of PDMS to isolate it from biological tissue.
These devices are typically implanted either on top of the dura beneath the skull or just under the skin inside the skull cavity. For our in-vitro evaluation, we choose muscle tissue whose electromagnetic properties are similar to grey matter at 2.4~GHz~\cite{gabriel1996dielectric}. We took a 0.75~inch thick pork chop and created a slit 0.0625~inch from the surface and inserted the antenna. We use the TI Bluetooth device as our RF source and place it at a distance of 3~inches from the meat surface. We use Intel Link 5300 Wi-Fi card on channel 11 as our Wi-Fi receiver; the RSSI results were similar with a Samsung S4. We set our FPGA prototype to synthesize 2~Mbps Wi-Fi packets. Fig.~\ref{fig:rssi_bmi} shows the RSSI for different Bluetooth powers. The plots show the feasibility of communicating with implanted devices, despite significant attenuation due to the muscle tissue. The range we achieve is better than the 1-2~cm target range for existing prototypes~\cite{bmi1, bmi2}. Finally, since Samsung Note 5 and iPhone 6 support 10~dBm Bluetooth transmissions, this demonstrates an implantable device that can communicate with commodity mobile devices.
\subsection{Card to Card Communication}
Finally, we can use the single-tone transmission from Bluetooth devices to enable communication between two passive cards similar to ambient backscatter~\cite{abc-sigcomm2013}, but without the need for strong TV transmissions. Since most users have Bluetooth-enabled devices (e.g., smartphone), this could be used for money transfer between credit cards, bus passes, splitting of checks, or transferring content to digital paper devices.
We prototype proof-of-concept credit card form factor devices shown in Fig.~\ref{fig:card_2_card} that backscatter the single-tone from the Bluetooth device to communicate between each other. We replicate prior work on ambient backscatter~\cite{abc-sigcomm2013} and use its receiver architecture. We however, tune the hardware to operate at 2.4~GHz instead of the UHF TV band. The prototype also has an IGLOO Nano FPGA to implement digital baseband processing for \name\, an MSP430 micro-controller for running application software, LEDs for visual feedback, an accelerometer and capacitive touch sensors for user input.
\begin{figure}[!t]
\subfigur
{
\includegraphics[width=.25\columnwidth]{./figures/card_to_card.eps}
}
\subfigure
{
\includegraphics[width=0.7\columnwidth]{./figures/tag2tag.eps}
}
\vskip -0.15in
\caption{{\bf BER for communication between two cards with integrated antennas.} Amazon card shown for comparison.}
\vskip -0.15in
\label{fig:card_2_card}
\end{figure}
We evaluate card to card communication by configuring one card to periodically transmit an 18 bit payload at 100~kbps to another, at time instances corresponding to the single-tone Bluetooth transmissions. The cards use their energy detectors to synchronize with the Bluetooth transmissions. We place the transmit card 3~inches from a 10~dBm TI Bluetooth device and vary the location of the receiver. Fig.~\ref{fig:card_2_card} shows the bit error rate achieved. The plot demonstrates that up to a distance of 30~inches, 10~dBm Bluetooth transmissions achievable on smartphones such as the Note 5 and iPhone 6, can enable card to card communication.
\subsection{Bluetooth as an RF source}
\label{sec:bluetooth}
The high level idea is to transform a Bluetooth chip into a single tone transmitter, i.e., with constant amplitude and frequency. We leverage two insights about GFSK modulation used in Bluetooth. First, Bluetooth uses two frequencies to encode the zero and one data bits. Thus, if we could transmit a stream of constant ones or zeros, we create a single frequency tone. Second, passing a single tone through the Gaussian filter used by Bluetooth does not change its spectral properties since the filter only smooths out abrupt changes to the frequency. Thus, if we could get the Bluetooth chipsets to transmit a continuous stream of zeros or ones, then we effectively create a single tone. Achieving this in practice requires us to address two key challenges: data whitening and the BLE packet structure.
\begin{figure}[!t]
\includegraphics[width=0.9\columnwidth]{./figures/ble_scrambler.eps}
\vskip -0.15in
\caption{Shift registers used in Bluetooth to perform data whitening.}
\vskip -0.1in
\label{fig:ble_scrambler}
\end{figure}
\vskip 0.05in\noindent{\bf Data whitening.} While our goal is to create long sequences of either zeros or ones, Bluetooth uses data whitening to avoid such sequences so as to enable accurate timing recovery on a Bluetooth receiver. Specifically, Bluetooth uses the 7-bit linear feedback shift register circuit in Fig.~\ref{fig:ble_scrambler} with the polynomial $x^7 + x^4 +1$. Given an initial state, the circuit outputs a sequence of bits that are used to whiten the incoming data by XORing the data bits with the bits output by the circuit. We reverse this whitening process to create the desired sequence of ones or zeros. To do this, we observe that given an initial state for each of the registers, we can deterministically generate the whitening sequence. We also note from the Bluetooth specification that the shift registers are initialized using the Bluetooth channel number. In particular, Bluetooth initializes the zeroth register to a one and the rest of the six registers to the binary representation of the Bluetooth channel number. For instance, while transmitting on the Bluetooth advertising channel 37, the zeroth register in Fig.~\ref{fig:ble_scrambler} is set to 1 and the rest are set to the binary representation of 37. Thus, given an advertising channel, we can initialize the Bluetooth whitening algorithm and compute the whitening sequence. The data bits are then set to the same bits in the whitening sequence or their bit complement to generate long sequences of zeros or ones respectively. In~\xref{sec:blue-results}, we show this process works with unmodified Bluetooth chipsets.
\vskip 0.05in\noindent{\bf Link-layer packet structure.} The above discussion assumes that we can control all the bits in a Bluetooth packet; however an advertising packet has fields including the preamble and access address, shown in Fig.~\ref{fig:ble_packet}, that cannot be arbitrarily modified. The preamble is fixed to an alternating sequence of zeros and ones, and the access address is set to {\it 0x8E89BED6} for advertising packets. This is followed by a length field and an advertiser address field. Finally, the packet has the data payload and a 3-byte CRC. Of the above fields, only the data payload can be set to arbitrary values.\footnote{The Android API only allows 24 of the 32 bytes in the payload to be arbitrarily set.} Therefore, our system takes the following steps: 1) we use the Bluetooth preamble, access address and the header (56~$\mu$s in total) to enable Bluetooth packet detection at the backscattering device using an ultra-low power envelope detection circuit, 2) we estimate the beginning of the payload and start backscattering using the techniques introduced in~\xref{sec:genwifi} to generate Wi-Fi packets, and 3) we complete the Wi-Fi transmission before the start of the Bluetooth CRC field. Since the Bluetooth Bluetooth CRC is being transmitted on a different channel than the generated Wi-Fi packet, it does not affect the Wi-Fi receiver.
\begin{figure}[!t]
\includegraphics[width=0.9\columnwidth]{./figures/ble_packet.eps}
\vskip -0.05in
\caption{Structure of a Bluetooth advertising packet.}
\vskip -0.15in
\label{fig:ble_packet}
\end{figure}
The above design requires estimating the beginning of the Bluetooth packet. Our receiver uses an envelope detector, similar to prior ultra-low power designs~\cite{allsee-nsdi14,abc-sigcomm2013}, for energy detectors. Our energy detection circuit can be customized to only trigger for Bluetooth transmitter up to 8--10 feet, to prevent false positives. Energy detection however does not allow us to accurately detect the beginning of the Bluetooth packet since we do not decode the Bluetooth preamble and hence cannot use typical synchronization techniques. This results in inaccurate estimates of the beginning of the payload. Our implementation uses a guard interval of 4~$\mu$s in our estimate of the start of the payload to address this.
\section{Discussion and Conclusion}
We introduce a novel approach that transforms wireless transmissions from one technology to the other, on the air. Specifically, we show for the first time that Bluetooth transmissions can be used to generate Wi-Fi and ZigBee-compatible signals using backscatter communication. Using this approach, we build proof-of-concepts for previously infeasible applications including the first contact lens form-factor antenna prototype and an implantable neural recording interface that communicate directly with commodity devices such as smartphones and watches, thus enabling the vision of Internet connected implanted devices.
\textred{Finally we outline research directions for improving upon this work. First, interscatter enables high bit rates which allow data to be transmitted in shorter periods of time. This reduces the time for which transmissions occupy the channel and can reduce power consumption by duty cycling the device. Future work could improve overall throughput by using BLE data packets which can be transmitted at a faster rate than BLE advertisements. Additionally the latest Bluetooth standard increases the maximum length for these data packets which would further improve throughput. Lastly, while our current design focuses on generating 802.11b transmissions using backscatter, future work could explore OFDM based protocols such as 802.11g/n/ac to improve the bit rates we achieve by an additional order of magnitude.}
\subsection{Communication to Backscatter Device}
\label{sec:downlink}
Achieving communication to the backscattering device is a challenge because it cannot decode Wi-Fi and Bluetooth transmissions: Bluetooth uses frequency modulation while 802.11b uses phase modulation with DBPSK/DQPSK; so both have relatively constant amplitudes. Traditional receivers for such phase/frequency modulated signals require synthesis of a high frequency carrier that is orders of magnitude more power consuming than backscatter transmitter. In fact, existing ultra-low power backscatter receiver designs~\cite{abc-sigcomm2013,allsee-nsdi14,nsdi16} use amplitude modulation (AM) which does not require phase and frequency selectivity; unfortunately, Wi-Fi and Bluetooth radios do not support AM.
\begin{figure}[!t]
\includegraphics[width=\columnwidth]{./figures/ofdm_zeros.eps}
\vskip -0.15in
\caption{{\bf Creating AM signals using OFDM.} We show four OFDM symbols (in black). Random OFDM symbols are constructed using IFFT over random modulated bits while constant OFDM symbols are constructed using the same modulated bits. The red line show the output of our peak detector receiver.}
\label{fig:ofdm_timedomain}
\vskip -0.15in
\end{figure}
\vskip 0.05in\noindent{\bf Our Approach.} We transform the payload of 802.11g packets into AM modulated signals. In 802.11g, each OFDM symbol is generated by taking an IFFT over QAM modulated bits to generate 64~time domain samples. Fig.~\ref{fig:ofdm_timedomain} shows a time-domain OFDM symbol created from random modulated bits, which we call {\it random OFDM}. We also show the symbol created when the IFFT is performed over constant modulated bits, which we call {\it constant OFDM}. The figure shows that while random OFDM symbols have the energy spread across the time samples, in constant OFDM the energy is in the first time sample and is zero elsewhere. This can be used to create an amplitude modulated signal. Constructing a constant OFDM symbol using Wi-Fi radios is however not straightforward due to scrambling, coding and interleaving. Next, we describe how to to achieve this goal.
\vskip 0.05in\noindent{\it Scrambler.} 802.11g data scrambling is performed by XORing the incoming data bits with the output of a 7-bit linear feedback shift register using the same polynomial shown in Fig.~\ref{fig:ble_scrambler}. Given the scrambling seed, the output sequence of the circuit is deterministic. According to the Wi-Fi standard, the scrambling seed is set to a pseudorandom non-zero value. In principle, this information should be available on the Wi-Fi hardware. Our experiments in~\xref{sec:res_downlink} reveal that a number of commercial Wi-Fi chipsets use a predictable sequence of scrambling seeds. Further, certain Atheros chipsets also allow us to set the scrambling seed to a fixed value in the driver. Using this, we reverse the effect of the scrambler.
\vskip 0.05in\noindent{\it Convolutional Encoder.} 802.11g uses convolutional encoding on the scrambled bits to be resilient to noise and interference. It uses a 1/2 rate convolutional encoder where two coded bits are output for each incoming scrambled bit. The higher 2/3 and 3/4 coding rates are obtained by dropping some of the 1/2 rate encoded bits. Specifically, given the scrambled bits, $b[k]$, the two encoded bits are,
\begin{eqnarray*}
C_1[k] = b[k] \oplus b[k-2] \oplus b[k-3] \oplus b[k-5] \oplus b[k-6]\\
C_2[k] = b[k] \oplus b[k-1] \oplus b[k-2] \oplus b[k-3] \oplus b[k-6]
\end{eqnarray*}
This is a 1-to-2 mapping which cannot generate every desired sequence of encoded bits. We observe however that if all the incoming scrambled bits are ones (zeros), then all the encoded bits are ones (zeros). Thus, we can generate a sequence of all zeros or all ones as encoded bits.
\vskip 0.05in\noindent{\it Interleaver.} The encoded bits are interleaved across different OFDM frequency bins to make adjacent encoded bits more robust to frequency selective channel fading. We note however that when using a sequence of all ones or zeros as our encoded bits, interleaving again results in a sequence of all ones or zeros and does not require us to perform any special operation.
\vskip 0.05in\noindent{\it Modulator.} The interleaved bits are modulated using BPSK, 4QAM, 16QAM or 64QAM. Since our interleaved bits are all ones or zeros within an OFDM symbol, the modulation operation results in using the same constellation point across all the OFDM bins, achieving our goal of creating an OFDM symbol constructed using a constant modulated symbol.
We note that OFDM symbols have pilot bits inserted in specific frequency bins, which cannot be controlled. This however does not significantly change the desired constant OFDM pattern since the fraction of pilot to data symbols is low. Also, 802.11g convolutional encoders have a delay length of $7$, i.e., the last six data bits from the previous OFDM symbol impact the first few encoded bits in the current OFDM symbol. This could be a problem when the constant OFDM symbol follows a random OFDM symbol. We address this by setting the last six data bits in the random OFDM symbol to ones and use 16/64 QAM to ensure that the random OFDM symbol will still result in a high amplitude signal.
\begin{figure}[!t]
\includegraphics[width=\columnwidth]{./figures/ofdm_ones.eps}
\vskip -0.15in
\caption{{\bf Encoding bits with OFDM symbols.} The figure shows the encoding used to convey bits to the backscatter device.}
\label{fig:ofdm_timedomain1}
\vskip -0.15in
\end{figure}
\vskip 0.05in\noindent{\bf Encoding process.} Ideally, we can encode a 1 and 0 bit by random and constant OFDM symbols respectively. Fig.~\ref{fig:ofdm_timedomain} shows why this is a problem: constant OFDM symbols have a peak at the beginning of the time-domain symbol. This is a problem since we use a passive peak detection receiver that tracks the peaks in this signal (shown in red). This creates false peaks at the beginning of each constant OFDM symbol, which can confuse the receiver when there are consecutive constant OFDM symbols. To avoid this, we encode each bit with two OFDM symbols as shown in Fig.~\ref{fig:ofdm_timedomain1}. A one bit is represented by a random OFDM symbol followed by a constant OFDM symbol, while a zero bit is represented as two random OFDM symbols. Since each 802.11g OFDM symbol is 4~$\mu$s, this achieves a bit rate of $125$~kbps.
Finally, OFDM symbols have a cyclic prefix where the last few time samples are repeated at the beginning. Since the cyclic prefix in the case of a constant OFDM symbol is all zero, this could create a glitch. To avoid this, we pick the preceding random OFDM symbol such that its last time sample has a high amplitude. This ensures that the peak detector circuit sees a high peak at the end of the first OFDM symbol and does not create a glitch during the cyclic prefix.
\subsection{Putting it all together}
We use a query-reply protocol. The Wi-Fi device queries the backscatter device using the reverse channel in~\xref{sec:downlink}. The backscatter device responds using the backscatter channel in~\xref{sec:genwifi}. This design works with multiple backscatter devices since the Wi-Fi device can query them one after the other.
\subsection{Generating Wi-Fi using backscatter}
\label{sec:genwifi}
We describe how to generate 802.11b signals by backscatter the single-tone from the Bluetooth device. We first present our design that can create a frequency shift to the incoming single tone signal. We then show how to synthesize 802.11b signals from this shifted tone.
\subsubsection{Single Sideband Backscatter Design}
\label{sec:singleside}
While existing sideband modulation techniques have been recently used to create frequency shifts~\cite{bluetoothbackscatter, nsdi16}, they also create a mirror copy. These works demonstrate that modulating the radar cross-section of an antenna effectively multiples a single tone signal by the modulating signal. Since, $2\sin(f_c t)\sin(\Delta ft) = \cos((f_c-\Delta f)t) - \cos((f_c+\Delta f)t)$, modulating the antenna at $\Delta f$ in the presence of the incoming single-tone carrier signal, $\sin(f_c t)$, creates the desired shift $\cos((f_c+\Delta f)t)$. However, it also creates the mirror copy $\cos((f_c-\Delta f)t)$. This is problematic when used with Bluetooth: both the advertising channels 37 and 39 are at either end of the ISM band, as shown in Fig.~\ref{fig:versus}. Thus, creating any frequency shifts to the corresponding Bluetooth signal will create a mirror copy outside the ISM band. The second advertising channel, 38, overlaps with Wi-Fi channel 6 and is close to Wi-Fi channel 1, and hence can create strong interference to the weak backscattered Wi-Fi signals. Further, generating packets on Wi-Fi channel 11 using advertising channel 39 shown in Fig.~\ref{fig:versus}, would again create a mirror copy that lies outside the ISM band. Thus, existing sideband modulation techniques cannot be used by \name\ on any of the BLE/Wi-Fi channels.
\vskip 0.05in\noindent{\bf Our Solution.} We present the first single sideband backscatter architecture that produces a frequency shift on only one side of the single tone Bluetooth transmission. Our intuition is to emulate radios. Specifically, radios use 2.4~GHz oscillators to generate the orthogonal signals, $\cos(2\pi f_c t)$ and $\sin(2\pi f_c t)$. These are multiplied with digital in-phase, $I(t)$, and quadrature phase components, $Q(t)$, to create the signal: $$[I(t)+j Q(t)]\times[\cos(2\pi f_c t) + j\sin(2\pi f_c t)]$$ By setting $I(t)$ and $Q(t)$ to $\cos(2\pi \Delta ft)$ and $\sin(2\pi \Delta ft)$, radios can easily create the desired shifted signal, $e^{j2\pi (f_c +\Delta f)t}$, without any mirror copies. The challenge is that we cannot use oscillators running at 2.4~GHz since they consume significant amounts of power~\cite{razavi1998rf}.
Our insight is that mathematically we can imitate the above operations using complex impedances on the backscatter device without 2.4~GHz oscillators. Say, we could create the complex signal $e^{j2\pi\Delta ft}$, then backscattering such a complex signal with the incoming single-tone Bluetooth transmission, $\cos(2\pi f_c t)$, results in:
\begin{eqnarray*}
e^{j2\pi\Delta ft} \cos(2\pi f_c t) = \frac{1}{2} (e^{j2\pi(f_c+\Delta f)t} + e^{j2\pi(-f_c+\Delta f)t})
\end{eqnarray*}
The first term is the desired shifted signal while the second term has a negative frequency and does not occur in practice. Thus, the above operation creates the desired shift without a mirror copy. So if we can create the complex signal $e^{j2\pi\Delta ft}$ using backscatter, we can achieve single-sideband backscatter modulation. The desired signal can be written as,
\begin{eqnarray}
\label{eqn:1}
e^{j2\pi\Delta ft} = \cos(2\pi\Delta ft) + j\sin(2\pi\Delta ft)
\end{eqnarray}
To create this on a backscatter device, we generate the sin/cos terms using square waves. We then use complex impedances at the switch to generate the desired complex values.
\vskip 0.05in\noindent{\it Step 1.} We approximate the sin/cos terms in Eq.~\ref{eqn:1} using square waves alternating between +1 to -1,\footnote{ Since digital operations are on 0 and 1 bits, in practice we perform step 1 using a square wave between 1 and 0 instead of +1 and -1. This is however a straightforward mapping with a DC offset. For simplicity however, we explain our design using +1 and -1.} at a frequency of $\Delta f$. From Fourier analysis a square wave at $\Delta f$ can be written as $\frac{4}{\pi}\sum_{n=1,3,5,\cdots}\frac{1}{n} \sin(2\pi n\Delta ft)$. The first harmonic is the desired sine term while the third and fifth harmonic have a power of $\frac{1}{n^2}$ which are 9.5~dB and 14~dB lower than the first. Since all 802.11b bit rates can operate at SNRs lower than 14~dB, such an approximation is sufficient for our purposes. To generate the cosine term, we time shift this square wave by a quarter of the time period. This square wave can easily be generated by clocking the switch and the digital operations at multiples of the desired offset, $\Delta f$.
\vskip 0.05in\noindent{\it Step 2.} Now that we have approximated the sin/cos terms to be either +1 or -1, Eq.~\ref{eqn:1} can take one of four values: {\it 1+j, 1-j, -1+j,} and {\it -1-j}. We create these complex values by changing the impedance of the backscatter hardware. RF signals are reflected when they cross two materials that have different impedances. Since the impedance of an antenna is different from the medium around it, a fraction of the incident RF signals get reflected off the antenna. Backscatter works by creating an additional impedance boundary between the antenna and the backscatter circuit. Specifically, given an incoming signal $S_{in}$, the reflected signal from the backscatter device is given by,
\begin{equation*}
S_{out} = \frac{Z_a - Z_c}{Z_a +Z_c} S_{in}
\end{equation*}
Here $Z_a$ and $Z_c$ are the impedance of the antenna and the backscatter circuit respectively. In traditional backscatter, the impedance of the backscatter circuit is set to either $Z_a$ or $0$ corresponding to no reflection or maximum reflection of the incoming signal. We note however that the impedance of the backscatter circuit can be set to complex values by changing the inductance of the circuit~\cite{fullduplex-mobicom2014,matt-qambacksactter}. Specifically, at the frequency $f$, the impedance of the backscatter circuit can be written as $j2\pi fL$ where the inductance is $L$. Thus, by changing this inductance, we can create complex values for the fraction, $\frac{Z_a-Z_c}{Z_a+Z_c}$. Specifically, to get the four desired complex values, {\it 1+j, 1-j, -1+j, -1-j}, we set the impedances of the backscatter circuit to $\frac{-j}{2+j}Z_a$, $\frac{j}{2-j}Z_a$, $\frac{2-j}{j}Z_a$ and $\frac{2+j}{-j}Z_a$ respectively. The antenna impedance, $Z_a$, is typically tuned to 50 ohms. By switching between these impedance states, we can generate the desired complex signal $e^{j2\pi \Delta ft}$ and hence achieve single-sideband backscatter modulation. \textred{Optimizing for these constraints, for our 2.4~GHz backscatter devices we used a 3~pF capacitor, open impedance, 1~pF and 2~nH to achieve the four complex values.}
\subsubsection{Synthesizing 802.11b signals}
\label{sec:ssb_iq}
A Wi-Fi signal can be written as, $(I_{wifi}(t) + jQ_{wifi}(t))e^{j2\pi f_{c}t}$, where $I_{wifi}(t)$ and $Q_{wifi}(t)$ correspond to the in-phase and quadrature-phase components for the baseband Wi-Fi signal. This can be rewritten as,
\begin{eqnarray*}
(I_{wifi}(t) + jQ_{wifi}(t)) e^{j2\pi\Delta ft} e^{j2\pi f_{bluetooth}t}
\end{eqnarray*}
Thus, to generate Wi-Fi signals, we need to create $(I_{wifi}(t) + jQ_{wifi}(t))e^{j2\pi\Delta ft}$ using backscatter. We have already demonstrated earlier how to generate $e^{j2\pi\Delta ft}$, so we can multiply it with the in-phase and quadrature-phase components of 802.11b to generate Wi-Fi signals. As described in~\xref{sec:versus}, 802.11b signals use DSSS/CCK coding that creates coded bits that are then modulated using either DBPSK or DQPSK. Thus, if we can show that we can transmit DBPSK and DQOPK, we can create 802.11b signals.
\vskip 0.05in\noindent{\bf DBPSK.} In this case, the one and zero bits are represented as +1 and -1, which translates to always setting $Q_{wifi}(t)$ to zero and $I_{wifi}(t)$ to either +1 or -1. Since $e^{j2\pi\Delta ft}$ takes the values in the set {\it \{1+j, 1-j, -1+j, -1-j\}}, multiplying it with +1 or -1 results in values within the same set, which we know how to generate using our impedance values. Thus, we can generate \label{eqn:2} DBPSK modulation and hence achieve 1 and 5.5~Mbps 802.11b transmissions.
\vskip 0.05in\noindent{\bf DQPSK.} In this case, both $I_{wifi}(t)$ and $Q_{wifi}(t)$ are set to either +1 or -1. Thus, the baseband Wi-Fi signal can take one of the following values: {\it \{1+j, 1-j, -1+j, -1-j\}}. Multiplying this with $e^{j2\pi\Delta ft}$ which takes one of the following values {\it \{1+j, 1-j, -1+j, -1-j\}}, results in one of these four normalized values: {\it \{1,-1,j,-j\}}. We observe that {\it\{1,-1,j,-j\}} and {\it \{1+j, 1-j, -1+j, -1-j\}} (which are the four impedance values generated in our backscatter hardware) are constellation points that are shifted by $\pi/4$. Since 802.11b uses differential QPSK, we ignore this constant phase shift of $\pi/4$ and instead map them to the four complex impedance values generated by our hardware. Wi-Fi receivers ignore this constant phase shift since the bits are encoded using differential phase modulation. This allows us to generate DQPSK modulation and the corresponding 2 and 11~Mbps 802.11b transmissions.
\begin{figure}[!t]
\includegraphics[width=\columnwidth]{./figures/ssb_spectra.eps}
\vskip -0.1in
\caption{{ Comparing single sideband backscatter (red signal) to prior double sideband backscatter (blue signal).}}
\vskip -0.1in
\label{fig:ssb_spectra}
\end{figure}
Fig.~\ref{fig:ssb_spectra} shows the frequency spectrum for the backscatter generated WiFi signals at 2~Mbps using our single-sideband backscatter with an arbitrary frequency shift of 22~MHz. For comparison, we also plot the spectrum using prior sideband backscatter approaches~\cite{nsdi16}. The plots show that prior sideband backscatter designs create a strong mirror copy on the other side of the single tone. However, single-sideband backscatter, introduced in this paper, eliminates this mirror copy and hence improves spectral efficiency.
\subsubsection{Practical Design Considerations}
The Bluetooth advertising packet can have a payload up to 31 bytes or 248~$\mu$s. Since Wi-Fi packets at different bit rates occupy the channel for different times, this translates to different packet sizes. At 2, 5.5 and 11~Mbps the Wi-Fi payload can be 38, 104, and 209 bytes within a single Bluetooth advertising packet. Given its size, we however cannot fit a 1~Mbps Wi-Fi packet in a single Bluetooth advertising packets. \textred{We note that Bluetooth data transmissions, which are sent at faster rates and can last up to 2~$ms$, would enable 1~Mbps packets and greater overall throughput.} We focus on Bluetooth advertising packets in this paper since they are easier to control on commodity devices.
\vskip 0.05in\noindent{\it Additional Optimizations.}
Bluetooth does not perform carrier sense before transmitting. Further, the backscatter Wi-Fi packet is at a different frequency that could be occupied, resulting in a collision. Since Bluetooth advertisements are small and sent once every 20~ms, such collisions have a negligible impact on Wi-Fi which operates at a much finer time granularity. Collisions however are not desirable at the backscattering device; they would require the backscattering device to retransmit its data, consuming more energy. We describe three optimizations that can reduce these collisions.
\vskip 0.05in\noindent 1) We could ensure that the Wi-Fi channel is unoccupied for the backscatter duration. Since most devices have both Wi-Fi and Bluetooth, they could coordinate and hence the Wi-Fi radio could schedule a $CTS\_to\_Self$ packet to be transmitted before the Bluetooth packet. $CTS\_to\_Self$ can reserve the channel for the duration of the Bluetooth packet, preventing other Wi-Fi devices from making concurrent transmissions. The ability to schedule $CTS\_to\_Self$ packets has been demonstrated in the past~\cite{tep,wifibackscatter-sigcomm2014} and requires driver and firmware access to the commodity device.
\vskip 0.05in\noindent 2) We leverage that the advertisement packets are sent on all Bluetooth advertising channels one after the other, separated by a fixed duration $\Delta T$ (around 400~$\mu$s for TI Bluetooth chipsets). Using this we imitate an RTS-CTS exchange: when Bluetooth transmits on channel 37, we backscatter an RTS packet on the desired Wi-Fi channel. If the channel is free, the Wi-Fi device responds with a CTS packet, effectively reserving Wi-Fi channel 11 for the next $2\Delta T + T_{Bluetooth}$, where $T_{Bluetooth}$ is the duration of the Bluetooth packet. The backscattering device detects the presence of this CTS using our peak detection hardware. It then transmits data packets on the desired Wi-Fi channel using the remaining advertising packets sent on channel 38 and 39 over the next $2\Delta T + T_{Bluetooth}$ seconds.
\vskip 0.05in\noindent 3) To reduce the energy overhead of transmitting the RTS packet, we could instead transmit a data packet. If the Wi-Fi receiver can decode this packet, we would have exchanged useful data to the receiver. The Wi-Fi device can then send a $CTS\_to\_Self$ packet reserving the channel for the next $2\Delta T + T_{Bluetooth}$, which can then be used to backscatter additional Wi-Fi packets using the two remaining advertising packets, without collisions. This eliminates the energy overhead of sending a data-free RTS packet. Evaluating the effectiveness of this is not in the scope of this paper.
\vskip 0.05in\noindent\textred{Finally we note that having a large number of backscatter devices will not be affected by the RTS transmission since the multiple devices will be scheduled using the downlink transmissions before any of the backscatter devices even transmit.}
\section{FPGA and IC Design}
We began by developing the hardware on an FPGA platform to characterize the system and build proof of concept applications. We then translated the design into an IC and used it to quantify the power consumption.
\noindent\textbf{FPGA design.} Our system has two components: the RF front end and the baseband digital circuit. The front end consists of a backscatter modulator and a passive receiver. We isolate the receiver from the antenna using a single pole double throw (SPDT) switch that switches between transmit and receive modes. The backscatter modulator switches between four impedance states and is implemented using a cascaded two-stage SPDT switch network. We use HMC190BMS8 SPDT switches both for isolating the transmitter and receiver and in the backscatter modulator. The front end is implemented on a low loss Rodgers 4350 substrate~\cite{hmc190bms8}. As explained in~\xref{sec:genwifi}, the impedances connected to the four terminals of the switch network are a 3~pF capacitor, open impedance, a 1~pF capacitor and a 2~nH inductor which achieve the four desired complex values. The receiver is an energy detector consisting of passive analog components and a comparator to distinguish between the presence and absence of energy. We replicate the receiver described in ~\cite{allsee-nsdi14, abc-sigcomm2013, nsdi16}.
The 802.11b scrambling, DSSS/CCK encoding, CRC encoding, DQSPK encoding and single-side band backscatter were implemented in Verilog \textred{and translated onto the DE1 Cyclone II FPGA development board by Altera~\cite{altera_de1}}. We implement a 35.75~MHz shift which we found to be optimal for rejecting the interference from the Bluetooth RF source. The digital output of the FPGA was connected to the backscatter modulator and the energy detector was fed to its digital input. A 2~dBi antenna was used on the \name\ device.
\begin{figure*}[!t]
\subfigure[TI CC2650]
{
\includegraphics[width=.3\textwidth]{./figures/cc2650.eps}
}
\subfigure[Galaxy S5 Smartphone]
{
\includegraphics[width=.3\textwidth]{./figures/s5.eps}
}
\subfigure[Moto360 (2nd gen) Smartwatch]
{
\includegraphics[width=0.3\textwidth]{./figures/moto360.eps}
}
\vskip -0.1in
\caption{{\bf Generating Bluetooth single tone.} The red lines show Bluetooth transmissions with random application data and the green lines show the single-tone transmission created by \name.}
\label{fig:spectra}
\vskip -0.15in
\end{figure*}
\noindent \textbf{IC design.} As CMOS technology has scaled, the power consumption of digital computation has decreased significantly~\cite{koomey}. Unfortunately, active radios require power hungry \textit{analog} components which operate at RF frequencies and do not scale in either power or area. Interscatter relies exclusively on baseband digital computation with no components operating at RF frequencies, so it can leverage CMOS scaling for ultra-low power operations. We implement \name\ on the TSMC 65~nm low power CMOS technology node. For context, Atheros AR6003~\cite{atheros_ar6003} chipsets released in 2009 used 65~nm CMOS and so, it is a fair comparison with industry standard. The \name\ IC implementation can be broken down into the following three main components described in detail below: the frequency synthesizer, baseband processor, and backscatter modulator.
\noindent \textit{Frequency synthesizer.} This block takes a frequency reference as an input and generates the 802.11b baseband 11~MHz clock as well as the four phases of the 35.75~MHz frequency clock required for backscatter. \textred{We use an integer N charge pump and ring oscillator based PLL to generate a 143~MHz clock which is fed to a Johnson counter to generate the four clocks for the 35.75~MHz frequency shift each 90\degree~out of phase.} The same 143~MHz clock is divided by 13 to generate the 11~MHz baseband clock. Thus, 11~MHz and 35.75~MHz are phase synchronized and we avoid glitches. This block consumes 9.69 $\mu W$ of power.
\noindent\textit{Baseband processor.} This block takes the payload as the input and generates the baseband 802.11b packet. We use the same Verilog code which was verified on the FPGA and synthesize the transistor level implementation using the Design Compiler tool by Synopsis~\cite{synopsis}. This block has a power consumption of 8.51 $\mu W$ for 2~Mbps Wi-Fi transmissions.
\noindent\textit{Backscatter modulator.} We implement our single sideband backscatter in the digital domain by independently generating the in-phase and quadrature phase components. We take the two bit output of the baseband processor and feed it to two multiplexers that map the four phases of the 35.75~MHz carrier to corresponding in-phase and quadrature-phase components. Then, at each time instant, the in-phase and quadrature phase components are mapped to the four required impedance states in~\xref{sec:ssb_iq}. We use CMOS switches to choose between open, short, capacitive and inductive states. The single side band backscatter modulator consumes 9.79 $\mu W$. {\it In total, generating 2~Mbps 802.11b packets consumes 28 $\mu W$.}
\section{Introduction}
There has been recent interest in medical applications including smart contact lens platforms~\cite{lens1,lens2,lens3,lens4} that measure biomarkers like glucose, cholesterol and sodium for diabetes management, as well as implantable neural devices that help in the treatment of epilepsy, Parkinson's disease~\cite{ecog-parkinson}, reanimation of limbs~\cite{ecog-limb} and development of brain-computer interfaces~\cite{ecog-bmi}. As part of an ecosystem of connected devices, these implants have the potential to transform the management of chronic medical conditions and enable novel interaction capabilities.
In this paper we ask the following question: can these implanted devices communicate directly with mobile devices such as smartphones, watches and tablets? The key challenge is that owing to their severe power constraints, these devices cannot use conventional radios to generate Wi-Fi, Bluetooth or ZigBee transmissions and hence cannot communicate directly with commodity devices. While recent work on passive Wi-Fi~\cite{nsdi16} significantly reduces the power consumption of Wi-Fi transmissions using backscatter communication, it requires infrastructure support in the form of specialized hardware that can generate the continuous wave RF signal needed for backscattering Wi-Fi signals. Thus, it cannot fully plug and play with existing mobile devices.
\begin{figure}[!t]
\includegraphics[width=\columnwidth]{./figures/watch_to_phone.eps}
\caption{{\bf \Name\ Communication.} The backscatter device (e.g., contact lens) reflects transmissions from a Bluetooth device (e.g., smart watch) to generate standards-compatible Wi-Fi signals that are decodable on a Wi-Fi device (e.g., smartphone).}
\label{fig:page1}
\vskip -0.15in
\end{figure}
\begin{figure*}[!t]
\centering
\includegraphics[width=0.8\textwidth]{./figures/applications.eps}
\vskip -0.1in
\caption{{\bf Potential applications of \Name\ communication.}
(a) Active contact lens systems can backscatter Bluetooth transmissions from a watch to generate Wi-Fi signals to a phone, (b) implantable brain interfaces can communicate by using Bluetooth headsets and smartphones, and (c) credit cards can communicate with each other by backscattering Bluetooth transmissions from a smartphone.}
\label{fig:applications}
\vskip -0.15in
\end{figure*}
We introduce {\it\name}\footnote{Short for inter-technology backscatter communication.}, a novel backscatter communication system that works using only commodity devices by transforming transmissions from one technology to the other, on the air. We observe that most mobile devices have a Bluetooth, Wi-Fi or ZigBee radio. Further, users increasingly carry smart watches, fitness trackers and Bluetooth headsets in addition to their smartphones. \Name\ reuses the radios on these devices as both RF sources for backscatter as well as receivers the backscattered signals. This allows us to benefit from the Wi-Fi, ZigBee and Bluetooth economies of scale (a few dollars per chip~\cite{ti_ble_chip_buy}) and eliminate the cost of a dedicated reader. Further, it significantly lowers the barrier to adoption, as the wireless hardware is widely available on commodity devices and does not require the user to carry a specialized backscatter reader.
Our key idea is a novel approach to backscatter communication: create Wi-Fi signals by backscattering Bluetooth transmissions. To understand this, consider an \name\ device in the presence of a Bluetooth transmitter (e.g., smart watch) and a Wi-Fi receiver (e.g., smartphone), as shown in Fig.~\ref{fig:page1}. We backscatter advertising packets from the Bluetooth device to synthesize 802.11b signals at 2--11~Mbps. In the figure, the \name\ device backscatters Bluetooth packets on channel 38 to generate standards-compliant 802.11b packets on Wi-Fi channel 11.
Realizing this idea is challenging for at least three reasons:
\begin{Itemize}
\item Wi-Fi and Bluetooth have different physical layer specifications --- Wi-Fi requires a 22~MHz bandwidth and uses spread spectrum coding. Bluetooth occupies a 1-2~MHz bandwidth and uses Gaussian Frequency Shift Keying (GFSK).
\item Bluetooth operates at carrier frequencies that are different from Wi-Fi. While sideband-backscatter modulation~\cite{nsdi16,bluetoothbackscatter} could shift the carrier by tens of Megahertz, it also creates a redundant copy on the opposite side of the carrier.\footnote{Say an RF source transmits a signal $\sin(f_ct)$, sideband modulation backscatters at a frequency of $\Delta f$, resulting in $sin(\Delta f t)$, which would in turn create the multiplicative signal $2\sin(f_ct)\sin(\Delta ft) = \cos((f_c-\Delta f)t) - \cos((f_c+\Delta f)t)$. The two signals correspond to a signal with a positive frequency shift and its mirror copy with a negative shift.} This not only wastes bandwidth but the redundant copy would also lie outside the unlicensed ISM band (see~\xref{sec:singleside}).
\item For bi-directional communication, we need a receiver at the \name\ device. Wi-Fi and Bluetooth receivers consume orders of magnitude higher power than backscatter and would offset its power saving. In fact, existing ultra-low power receiver designs rely on amplitude modulation (AM)~\cite{haykin2008communication}, which is not supported by Wi-Fi or Bluetooth.
\end{Itemize}
At a high level, we first transform a Bluetooth transmission into a single tone signal and use backscatter to create standards-compliant Wi-Fi packets on a single side of the resulting single tone Bluetooth signal. Specifically, we make three key technical contributions to achieve this design.
\begin{Itemize}
\item We show for the first time that {\it Bluetooth radios can be used to create single-tone transmissions}. We leverage that Bluetooth uses GFSK that encodes bits using two frequency tones. Thus, if we could transmit a stream of constant ones or zeros, we can create a single tone transmission. In~ \xref{sec:bluetooth}, we describe how to achieve this on commodity Bluetooth devices in the presence of the data whitening, CRC and headers.
\item We present the {\it first single-sideband backscatter design} that creates frequency shifts on a single side of the carrier. This lets us create 2--11~Mbps Wi-Fi signals shifted by tens of Megahertz on only {\it one side} of our single tone Bluetooth transmissions. We achieve this using complex impedances at the backscatter switch, without the need for a power-consuming 2.4~GHz oscillators (see~\xref{sec:genwifi}).
\item We {\it transform OFDM Wi-Fi devices into AM modulators}. At a high level, the Wi-Fi device in our design modulates the amplitude profile of OFDM symbols to create an AM signal. We show that this can be achieved by just setting the appropriate data bits in a Wi-Fi packet, without the need for any hardware power control.~\xref{sec:downlink} describes how we perform this in the presence of Wi-Fi scrambling, convolutional encoding and interleaving.
\end{Itemize}
We build prototype backscatter hardware on an FPGA platform and experiment with various Bluetooth and Wi-Fi devices. Our evaluation shows that we can generate 2--11~Mbps Wi-Fi signals from Bluetooth transmissions. To estimate the power consumption, we also design an integrated circuit using Cadence and Synopsis~\cite{cadence_RFspectre, synopsis}, which backscatters Bluetooth to create Wi-Fi signals. Our results show that backscattering 2~Mbps Wi-Fi signals using Bluetooth consumes only 28~$\mu$W. Finally, to demonstrate the generality of our approach, we also show the feasibility of generating ZigBee signals by backscattering Bluetooth transmissions.
To show the potential of our design, we implement proof-of-concepts for the three applications shown in Fig.~\ref{fig:applications}. We build a contact lens form-factor antenna and evaluate it in-vitro to demonstrate that it can communicate with commodity devices. We also build an implantable neural recording interface antenna and evaluate it in-vitro using muscle tissue. \textred{Finally, to demonstrate the applicability of our techniques beyond medical implants, we create credit card form-factor devices that can communicate with each other using Bluetooth transmissions as the RF signal source for backscatter.}
\section{System Design}
We use backscatter to transform transmissions from Bluetooth devices into Wi-Fi signals. In this section, we first provide an overview of Bluetooth and Wi-Fi physical layers and then describe how to create single-tone transmissions using Bluetooth devices. We then show how to create an 802.11b signal from this single tone Bluetooth transmission. Finally, we outline our design for bi-directional communication using OFDM Wi-Fi as an AM modulator.
\section{\@startsection {section}{1}{\z@}%
{2ex \@plus 1ex \@minus .1ex}%
{1ex \@plus.2ex}%
{\normalfont\bfseries\scshape\fontsize{11}{13}\selectfont}}
\global\def\subsection{\@startsection{subsection}{2}{\z@}%
{2ex\@plus 1ex \@minus .1ex}%
{1ex \@plus .2ex}%
{\normalfont\bfseries\fontsize{10}{12}\selectfont}}
\def\@maketitle{\newpage
\null
\setbox\@acmtitlebox\vbox{%
\baselineskip 20pt
\vskip 2em
\begin{center}
{\ttlfnt \@title\par}
\vskip 0.25em
{\subttlfnt \the\subtitletext\par}\vskip 1.25e
{\baselineskip 16pt\aufnt
\lineskip .5em
\begin{tabular}[t]{c}\hspace{-.15cm}\@author
\end{tabular}\par}
\vskip 0.75em
\end{center}}
\dimen0=\ht\@acmtitlebox
\advance\dimen0 by -10pc\relax
\unvbox\@acmtitlebox
\ifdim\dimen0<0.0pt\relax\vskip-\dimen0\fi}
\fi
\def\sharedaffiliation{%
\end{tabular}
\begin{tabular}{c}}
\begin{document}
\font\ttlfnt=phvb8t at 16pt
\newcommand{\supsym}[1]{\raisebox{6pt}{{\footnotesize #1}}}
\widowpenalty = 10000
\title{\textred{Inter-Technology Backscatter: Towards \\Internet Connectivity for Implanted Devices}}
\numberofauthors{1}
\author{%
Vikram Iyer\supsym{$\dagger$}, Vamsi Talla\supsym{$\dagger$}, Bryce Kellogg\supsym{$\dagger$}, Shyamnath Gollakota and Joshua R. Smith\\
\affaddr{University of Washington}\\
\coprimary{\supsym{$\dagger$}Co-primary Student Authors}\\
\affaddr{\{vsiyer, vamsit, kellogg, gshyam, jrsjrs\}@uw.edu}\\
}
\maketitle
\begin{sloppypar}
\input{abstract-9}
\input{intro-8}
\input{overview-2}
\input{versus-4}
\input{bluesource-3}
\input{genwifi-2}
\input{downlink-5}
\input{imp-1}
\input{results-1}
\input{applications-2}
\input{related-3}
\input{conclusion-3}
\bibliographystyle{abbrv}
\section{Related Work}
Interscatter is related to our prior work on ambient and Wi-Fi backscatter communication~\cite{abc-sigcomm2013,abc-sigcomm2014, wifibackscatter-sigcomm2014} where devices communicate by backscattering ambient signals such as TV and Wi-Fi transmissions. Specifically, Wi-Fi backscatter~\cite{wifibackscatter-sigcomm2014} conveys information by either reflecting or not reflecting existing Wi-Fi packets on the wireless channel. This creates CSI changes that encode information. Since the minimum resolution at which the bits are conveyed is greater than a Wi-Fi packet, this limits the bit rate to 100--1000~bps. \textred{Recent work on FS-Backscatter~\cite{fs-backscatter} combines this packet-level backscatter approach with double sideband subcarrier modulation~\cite{nsdi16,rfid-textbook1,fm-backscatter-paper}. Specifically, it uses a ring oscillator to shift each Wi-Fi packet to a different frequency and achieves higher reliability than Wi-Fi backscatter. Interscatter differs from these designs in two ways. First, interscatter achieves standards-compliant 2--11~Mbps Wi-Fi packets where the content is generated by the backscattering device. This allows devices to transmit data at the high bit rates supported by Wi-Fi. In contrast, Wi-Fi backscatter and FS backscatter (in the context of Wi-Fi), encode information by either reflecting or not reflecting multiple packets from a Wi-Fi transmitter. As a result, the backscattered bits are encoded at the granularity of the Wi-Fi transmitter's packets and the backscatter device cannot change the packet content. Thus, the achieved bit rates are significantly lower. Second, these approaches require a large number of packets to be flooded by the Wi-Fi transmitter to send a small number of backscatter bits. This is not an efficient use of wireless resources and is likely to be unacceptable in crowded Wi-Fi networks. Finally in large scale production open loop ring oscillators are known to have frequency variation between devices~\cite{ring-otis, process-otis} and high phase noise which limit data rates and range performance. Given these constraints as well as the significant attenuation and detuning of antennas in human tissue, it is unclear whether this approach would be feasible for medical implants. }
\cite{nasa,wifibackscatter-sigcomm2015} use full-duplex radios to cancel the high-power Wi-Fi transmissions from the reader and decode weak backscattered signals to enable high data rates of 5--330 Mbps. These approaches require a custom full-duplex radio to be incorporated at the receiver and hence do not work with existing devices.
Passive Wi-Fi~\cite{nsdi16} generates 802.11b transmissions using double sideband subcarrier modulation~\cite{rfid-textbook1, fm-backscatter-paper} from a continuous wave transmitter. Specifically, it shows that backscatter can be used to generate 802.11b transmissions when backscattering transmissions from a custom transmitter that sends out constant wave signals.~\cite{bluetoothbackscatter} generates FSK Bluetooth transmissions by backscattering a continuous wave transmitter. We build on this work, but differ in three critical ways. 1) Prior work requires custom specialized continuous wave transmitter hardware. Thus, they cannot plug and play with commodity devices. We introduce a novel approach to backscatter communication where we demonstrate for the first time that one can transform transmissions from a Bluetooth device into Wi-Fi signals, and achieve a system that can work using only commodity radios. 2) Prior work could not perform single-sideband backscatter and hence used 44~MHz of bandwidth to generate 22~MHz Wi-Fi transmissions. This is undesirable given that the 2.4~GHz ISM band is becoming increasingly crowded. In contrast, we introduce the first single-sideband backscatter design that generates Wi-Fi signals on only one side of the carrier and has double the spectral efficiency over prior designs. 3) Prior work uses custom plugged-in hardware to communicate to the backscattering device. In contrast, we reuse Wi-Fi transmitters to convey information to the backscattering devices. Specifically, we introduce a novel mechanism that encodes information in OFDM symbols of a Wi-Fi transmission.
Finally, Anynfc~\cite{cheap_rfid_reader} sells RFID readers that connect to smartphones via the headphone jack for \$185. While promising, RFID readers do not have the same economy of scale as Wi-Fi or Bluetooth which cost a few dollars~\cite{ti_ble_chip_buy}. Further, existing devices already have Wi-Fi and Bluetooth radios. This takes us closer to the vision of backscatter as a general-purpose communication mechanism.
\section{Evaluation}
We first check if we can create single tone transmissions on various Bluetooth platforms. We then measure the communication range and packet loss rate for the Wi-Fi packets generated by backscattering Bluetooth. Next, we evaluate the efficacy of our single sideband backscatter architecture. We then evaluate the communication link from an 802.11g transmitter to our low-power receiver. Finally, we demonstrate the feasibility of generating ZigBee transmissions.
\subsection{Generating Bluetooth single tone}
\label{sec:blue-results}
We run experiments with three different Bluetooth devices: the Texas Instruments CC2650, a Moto 360 2nd gen smart watch, and a Samsung Galaxy S5. We connected the exposed antenna connector on the TI chipsets to a spectrum analyzer and recorded data during the payload section of a Bluetooth packet. The Android platforms do not expose antenna connectors, so instead we performed the same experiment by recording the signal through a a 2~dBi monopole antenna connected to the spectral analyzer. We set the application layer data as described in~\xref{sec:bluetooth} to create a single tone. Fig.~\ref{fig:spectra} shows the spectrum of random Bluetooth transmissions in green. It also shows the same spectrum (in red) after performing the operations described in~\xref{sec:bluetooth} for comparison. The plots show that we can create single tone transmissions from commodity Bluetooth devices.
\subsection{Measuring the Wi-Fi RSSI}
We measure Wi-Fi RSSI for different distances between the backscatter device, Bluetooth transmitter and Wi-Fi receiver. We first fix the distance between the backscatter device and the Bluetooth transmitter. We then move the Wi-Fi receiver perpendicular from the mid-point of the line connecting the other two devices and measure the RSSI of the backscattered Wi-Fi packets reported by the Wi-Fi receiver. We set the backscatter device to generate 2~Mbps Wi-Fi packets on channel 11. The Bluetooth transmitter sends advertising packets with a 31 byte payload on BLE channel 38, once every 40~ms. We use a TI Bluetooth device and an Intel Link 5300 Wi-Fi card as our Bluetooth transmitter and Wi-Fi receiver respectively. We use four power values at the Bluetooth transmitter: 1) 0~dBm which is the typical configuration for Bluetooth devices, 2) 4~dBm, which is supported on Samsung S6~\cite{fcc-samsungs6} and One Plus 2 ~\cite{fcc-oneplus2}, 3) 10~dBm, which is supported by Samsung Note 5~\cite{fcc-samsungnote5} and iPhone 6~\cite{fcc-iphone6}, and 4) 20~dBm which is supported by class 1 Bluetooth devices.
\begin{figure}[!t]
\subfigure[1~feet between the backscatter device and Bluetooth transmitter]
{
\includegraphics[width=\columnwidth]{./figures/rssi_vs_dist_1ft.eps}
}
\subfigure[3 feet between the backscatter device and Bluetooth transmitter]
{
\includegraphics[width=\columnwidth]{./figures/rssi_vs_dist_3ft.eps}
}
\caption{{\bf Measuring the Wi-Fi RSSI.} RSSI versus distances between the backscattering device, Bluetooth transmitter and Wi-Fi receiver at different transmit power values.}
\label{fig:rssi_dist}
\vskip -0.15in
\end{figure}
Fig.~\ref{fig:rssi_dist} shows the results when the Bluetooth transmitter and the backscatter device are 1 and 3 feet apart. We pick these distances since we expect the user to use their mobile devices close to the backscatter device they intend to interrogate. The x-axis plots the distance between the Wi-Fi receiver and the backscatter device while the y-axis plots the reported Wi-Fi RSSI values. The figure shows that:
\begin{Itemize}
\item At higher transmit powers, the range at which Wi-Fi packets are reported is high --- with 20~dBm, we achieve a range of around 90~feet. This is because higher transmit powers translate to high-powered Bluetooth signals at the backscatter device. This in turns increases the RSSI of the backscatter generated Wi-Fi packets.
\item For a similar reason, at larger distances between the Bluetooth transmitter and the backscatter device, the Wi-Fi RSSI and range are lower. We however note that the achieved ranges are more than sufficient for the target personal area networking applications shown in Fig.~\ref{fig:applications}.
\item While the RSSI values are lower than typical 802.11n/g deployment, the reported RSSI values are mostly sufficient for decoding 2~Mbps 802.11b transmissions given that theoretically 2~Mbps Wi-Fi transmissions require an SNR of only 6~dB for reliable communication.
\end{Itemize}
Finally, we compute the packet error rate (PER) observed on the backscatter generated Wi-Fi packets across the whole spectrum of RSSI values observed in our experiments. To do this, we configure the backscatter device to consecutively transmit 200 unique sequence numbers in a loop that we use to compute the error rate the Wi-Fi receiver. We compute the packet error rate for both 2 and 11~Mbps Wi-Fi transmissions. For 2~Mbps and 11~Mbps, we generate packets with a payload of 31 and 77~bytes respectively so as to fit within a Bluetooth advertising packet. Fig.~\ref{fig:per} shows a CDF of the observed packet loss rate. The figure shows that,
\begin{Itemize}
\item Both 2~Mbps and 11~Mbps transmissions have similar packet loss rates. This is because the Wi-Fi packet payload sizes that can fit within a Bluetooth advertising packet are small. Further, both our 2~Mbps and 11~Mbps Wi-Fi packets use preamble and header encoded at the same bit rate. Thus, we see similar packet error rates between the two transmissions. This provides an interesting design choice in our system, where given the small packet sizes, if we were to optimize for the number of retransmissions, it is better to use the higher Wi-Fi bit rates and transmit more bits. Exploring this and how it interacts with power consumption would be an interesting future direction.
\item The packet error rate was greater than 30\% for low RSSI values. In principle, we can leverage prior work in our community~\cite{marnello,mrd} that use diversity and coding to account for such packet losses in Wi-Fi networks.
\end{Itemize}
\begin{figure}[!t]
\includegraphics[width=\columnwidth]{./figures/per.eps}
\vskip -0.1in
\caption{{\bf Wi-Fi packet error rate.} We measured packet error rate for backscatter-generated Wi-Fi packets at both 2 and 11 Mbps.}
\vskip -0.1in
\label{fig:per}
\end{figure}
\subsection{Efficacy of single sideband backscatter}
We compare the single sideband backscatter design introduced in this paper with prior double sideband backscatter hardware designs~\cite{nsdi16}. We use a USRP to transmit a single tone carrier such that the unintended mirror copy generated by double sideband backscatter appears in Wi-Fi channel 6. We then configure a standard Wi-Fi transmitter-receiver pair on channel 6 and evaluate the effect of the backscatter interference by running iperf using TCP with the default Wi-Fi rate adaptation algorithm. We use a Linksys WRT54G Wi-Fi AP as the Wi-Fi transmitter and a Nexus 4 smartphone as the Wi-Fi receiver separated by 10~feet. We set the backscatter device to generate 2~Mbps Wi-Fi packets with 32 byte payloads, at a distance of 2~feet from the Wi-Fi receiver.
Fig.~\ref{fig:interference} shows the iperf throughput in the presence of single and double-sideband backscatter hardware. The x-axis shows the rate at which Wi-Fi packets are generated using backscatter. For comparison, we show the baseline throughput in the absence of both backscatter devices. The plot shows that,
\begin{Itemize}
\item When the backscattering device generates a small number of packets (50 pkts/s), it has negligible impact on the concurrent iperf flow. This is true with both single-sideband and prior double sideband backscatter designs. This is expected because the backscattered packets are small and are transmitted at a very low rate that they do not affect the throughput of concurrent Wi-Fi connections.
\item When we backscatter Wi-Fi packets at a higher rate, the iperf throughput is reduced with prior double sideband backscatter hardware. This is because it creates a mirror copy on Wi-Fi channel 6 that reduces the throughput of any other Wi-Fi connection. However, we see negligible impact on the iperf throughput with our single-sideband hardware. This demonstrates that our single-sideband backscatter design can double spectral efficiency.
\end{Itemize}
\begin{figure}[!t]
\includegraphics[width=\columnwidth]{./figures/iperf_interfere.eps}
\caption{{\bf Efficacy of single sideband backscatter.} We compare our design with double sideband backscatter designs on the throughput of an iperf flow on a concurrent Wi-Fi transmitter-receiver pair. Baseline is the throughput in the absence of any backscatter device.}
\label{fig:interference}
\end{figure}
\subsection{Communication in reverse direction}
\label{sec:res_downlink}
\begin{figure}[!t]
\includegraphics[width=\columnwidth]{./figures/downlink.eps}
\vskip -0.1in
\caption{BER from 802.11g device to our low-power receiver.}
\vskip -0.15in
\label{fig:downlink}
\end{figure}
As described in~\xref{sec:downlink}, we create an AM modulated signal using OFDM by setting the appropriate modulated bits on each OFDM symbol. This however requires us to know the scrambling seed that is used by the Wi-Fi transmitter. We run experiments to track how different chipsets change the scrambling seed between packets by transmitting 802.11g packets at a bit rate of 36~Mbps from each Wi-Fi device. Since existing Wi-Fi receivers do not expose the scrambling seed, we instead use the gr-ieee802-11 package~\cite{usrp-802.11g} for GNURadio which implements the complete 802.11g receive chain to view the scrambling seed for each packet. Our experiments reveal that the AR5001G, AR5007G and AR9580 Atheros chipsets simply increment the scrambling seed by one between frames which is consistent with other studies~\cite{scrambler-seed}. We were also able keep the seed value fixed on ath5k cards by setting the GEN\_SCRAMBLER field in the AR5K\_PHY\_CTL register of the driver.
Next, we evaluate how our reverse link works with our low-power peak detector receiver. We transmit 36~Mbps 802.11g packets scrambled with a known seed. The Wi-Fi transmitter is configured to send a pre-defined repeating sequence of bits using the encoding in~\xref{sec:downlink}. We move our low-power receiver away from the Wi-Fi transmitter and compute the observed bit error rate at each location. Fig.~\ref{fig:downlink} shows the BER results as a function of the distance between the Wi-Fi transmitter and our peak detector receiver. The plot shows that despite the significant variability of OFDM, our peak detector achieves bit error rates less than 0.01 at distance of up to 18. Our current receiver uses off-the-shelf components and has a sensitivity of -32~dBm for 160~kbps. This however can be improved in a custom IC implementation that could give us higher ranges.
\subsection{Generating ZigBee Using Backscatter}
Finally, we show the feasibility of generating ZigBee signals by backscattering Bluetooth transmissions. ZigBee operates in the 2.4~GHz band over 16 channels where each channel is 5~MHz wide. At the physical layer, ZigBee achieves bit rates of 250~kbps using DSSS coding and offset-QPSK and has a better noise sensitivity than Wi-Fi. Since we have demonstrated that we can generate 802.11b that uses DSSS and QPSK, we adapt the techniques in~\xref{sec:genwifi} to also generate ZigBee-compliant packets.
\begin{figure}[!t]
\includegraphics[width=\columnwidth]{./figures/zigbee.eps}
\vskip -0.1in
\caption{{\bf CDF of ZigBee RSSI.} We measure the RSSI of backscatter-generated ZigBee packets at five different locations.}
\vskip -0.15in
\label{fig:zigbee}
\end{figure}
To evaluate this, we use the TI CC2650 Bluetooth device as our Bluetooth transmitter on advertising channel 38 and set our backscatter device to generate packets on channel 14, i.e., 2.420~GHz. We use the TI CC2531 as our commodity ZigBee receiver to receive the packets generated by the backscatter device. We place the backscattering device two feet away from the Bluetooth transmitter and the ZigBee receiver at five different locations up to 15~feet from the backscatter device. Fig.~\ref{fig:zigbee} shows the CDF of the ZigBee RSSI values for various distance locations, showing the feasibility of generating ZigBee packets with backscatter. We note that existing ZigBee transmitters consume tens of milliwatts of power when actively transmitting. In contrast, our backscatter based approach would consume tens of microwatts while transmitting a packet and could be beneficial for short range communication with nearby ZigBee devices.
\subsection{Bluetooth Versus Wi-Fi}
\label{sec:versus}
\vskip 0.05in\noindent{\bf Bluetooth.} Bluetooth low energy (BLE) devices use the advertisement channels to broadcast information about their presence and to initiate connections. Once the connection is established with a nearby Bluetooth device, they communicate by hopping across the 36 data channels spread across the 2.4~GHz ISM band. The three advertisement channels labeled as channels 37, 38 and 39 are shown in Fig.~\ref{fig:versus}. Since transmissions on data channels require establishing a connection with another device, we focus on Bluetooth advertisement channels where we can broadcast packets. Bluetooth LE uses Gaussian Frequency Shift Keying (GFSK) modulation with a bandwidth of 2~MHz. Specifically, a `1' (`0') bit is represented by a positive (negative) frequency shift of approximately 250~kHz from the center frequency. The resulting FSK signal is then passed through a Gaussian filter to achieve the desired spectral shape.
\vskip 0.05in\noindent{\bf Wi-Fi.} While Wi-Fi supports a suite of standards, we focus on 802.11b for the purpose of backscatter. Wi-Fi operates on three non-overlapping channels, each 22~MHz wide. To create 1 and 2~Mbps transmissions, 802.11b first XORs each data bit with a Barker sequence to create a sequence of eleven coded bits for each incoming data bit, which it then modulates using DBPSK and DQPSK. To create 5.5 and 11~Mbps transmissions, 802.11b uses CCK where each block of four incoming bits is mapped to 8-bit code words, which are then transmitted using DBPSK and DQPSK.
| 8f3f566c641fadbb5bb050f7af22374b6e3bd60e | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
The rapid growth of mobile wireless applications and consumers
continues to strain the limited cellular network capacity and has motivated the exploration of next generation 5G wireless networks.
The mobile industry predicts a $1000$x data
traffic growth by 2020. To meet this anticipated data growth demand, both industry and academia are exploring various
advanced solutions to boosting network capacity while continually providing high-level user experience to wireless customers.
One natural direction is developing new technologies to improve the efficiency of limited licensed spectrum.
Such popular advances include small cell base station (SBS) coverage, massive multiple-input multiple-output
links, and device-to-device communications.
Despite these exciting solutions, one key obstacle to network capacity expansion still lies in
the scarce licensed spectral resources.
To overcome spectrum shortage, LTE in unlicensed spectrum (LTE-U) technology has been proposed and is currently under
consideration by the 3GPP into its future standards.
LTE-U technology allows users to access both licensed and unlicensed spectra under a unified LTE network infrastructure.
It can provide better link performance, medium access control (MAC), mobility management, and larger coverage than
simple Wi-Fi offloading \cite{traffic,offload}.
Despite the many advantages of LTE-U, it still faces certain technical challenges. One primary issue is the coexistence with the incumbent Wi-Fi systems.
On the one hand, LTE systems are specifically designed to operate in the licensed spectrum under the centralized control of
network units based on non-contention MAC protocols
to prevent packet collision among subscribers.
On the other hand, Wi-Fi users rely on carrier sensing multiple access with collision avoidance (CSMA/CA) to reduce packet collision and
use a contention based MAC protocol, namely distributed coordination function (DCF),
to resolve package collision through a random backoff mechanism.
Therefore, how to ensure a fair and harmonious coexistence environment for both networks becomes a
major challenge for LTE-U.
First, for the two conflicting MAC technologies to coexist in the same unlicensed band, their mutual
impact and interference must be carefully controlled and managed.
It is important that a) Wi-Fi transmissions
will not frequently collide with LTE-U transmissions; b) LTE-U transmissions
will not lead to a substantial drop of Wi-Fi throughput.
Secondly, we should require a coexistence proposal of LTE-U and Wi-Fi so as not to demand
fundamental changes to the existing protocols defined in LTE and Wi-Fi standards.
This requirement is essential to avoid any impractical retrofitting of IEEE 802.11-compliant
Wi-Fi radios, or the re-design of existing 3GPP standards.
To overcome such challenges, we will first propose a new wireless access point (AP) architecture, namely hyper-AP (HAP), which is deployed by cellular operators and functions simultaneously as an LTE SBS and
a Wi-Fi AP. By integrating both LTE and Wi-Fi functions within the same HAP node, it can then jointly
coordinate the spectrum allocation and the interference management for Wi-Fi and LTE-U.
Based on the use of HAP, we develop a novel LTE-U and Wi-Fi coexistence mechanism by
embedding LTE-U signalling within the existing Wi-Fi protocol for seamless integration. Specifically,
the HAP will provide LTE user access in the contention free period (CFP) under a centralized coordinator,
whereas the HAP continues to provide normal Wi-Fi user access through its contention period
(CP) using the traditional CSMA/CA mechanism.
The main merit of the newly proposed HAP is that it can well support the standalone mode for LTE-U transmission,
which is found to be a formidable hurdle in many other coexistence mechanisms.
In what follows, we shall first provide an LTE-U technology overview, including operation modes,
coexistence technologies for LTE and Wi-Fi, and current considerations of combining LTE-U technology within
Wi-Fi MAC protocols. Next, we introduce the new wireless HAP that provides both LTE-U access
and Wi-Fi access in shared unlicensed band.
To improve the system performance, our key contribution is the novel LTE and Wi-Fi coexistence mechanism that embeds LTE-U channel access
within a standardized Wi-Fi protocol.
We shall also provide simulation results to
illustrate the achievable network performance gain in terms of Wi-Fi per-user throughput and overall system throughput.
Compared with networks under traditional coexistence technology without LTE-U deployments,
the proposed network access demonstrates better performance for both LTE-U and Wi-Fi user groups.
\section{Overview of LTE-U Technology}
\subsection{LTE-U Operation Modes}
Presently, there are mainly three operation modes proposed for LTE-U networking
based on different control channel configurations and supplementary links:
supplementary downlink (SDL), carrier aggregation (CA) time division LTE (TD-LTE), and standalone LTE-U \cite{coe1},\cite{coexistence}.
In SDL mode, the licensed spectrum serves as the anchor carrier.
When needed, the unlicensed spectrum is used for additional downlink data transmission since the downlink
demand is often higher than uplink.
Since the SDL mode merely utilizes the unlicensed spectrum as a downlink supplement, carrier aggregation technique can be applied
with few changes on the existing LTE standards.
In carrier aggregation TD-LTE mode, the unlicensed spectrum is used as an auxiliary TDD channel to carry both downlink and uplink traffic.
Its control channels remain on the licensed spectrum.
Once again, CA techniques can be adopted to execute the carrier aggregation TD-LTE mode.
In standalone LTE-U mode, both data and control signals from LTE-U nodes
are transmitted on unlicensed spectrum without relying on the additional licensed spectrum.
This mode can be used in the areas without guaranteed coverage of licensed cellular infrastructure,
and hence is a more flexible form of LTE-U.
Moreover, this mode also lacks an intelligent and centralized control signalling on licensed spectrum to ensure
reliable transmission, resource utilization, channel measurement, among others.
\subsection{LTE and Wi-Fi Coexistence Technologies} \label{section_coexistence}
Wi-Fi and LTE have substantially different PHY/MAC protocols, making their joint coordination
exceptionally challenging. In a nutshell, the LTE system uses a centralized MAC protocol
which allocates a non-overlapping set of physical resource blocks in time-frequency domain to mobile users within a cell.
There is no contention among users within such centralized multiple access in both downlink and uplink.
On the other hand, the Wi-Fi MAC protocol uses a decentralized, contention-based, random access mechanism
based on CSMA/CA and DCF.
If properly designed, LTE-U networks should be a good spectrum-sharing partner with the
incumbent Wi-Fi networks within the unlicensed band \cite{neighbor}.
Next, we will introduce several known LTE and Wi-Fi coexistence mechanisms, which are also summarized in Table \ref{tab_coe}.
\subsubsection{Channel Splitting Coexistence}
In some cases, the unlicensed spectrum is sufficiently wide.
It becomes possible for the LTE-U SBS to choose the cleanest channel, i.e., the channel exhibiting the lowest sensed interference power level,
based on power and activity measurements.
In this case, the SBS is almost free of interference from its
neighboring Wi-Fi devices if the Wi-Fi APs are given a ``channel-clearing'' request by the SBS.
Qualcomm has developed a simple and efficient dynamic channel selection method \cite{neighbor}.
Other works have also discussed ways to leverage existing underlay techniques in both LTE-U and Wi-Fi systems to ensure the least congested channel selection.
We shall note that, although the unlicensed spectrum is not unlimited, there is still a chance that it has not been fully utilized by Wi-Fi users.
There may exist little Wi-Fi activity in some locations such as certain outdoor environments.
Therefore, dynamic channel selection may suffice to meet the coexistence requirement in various low or medium-density Wi-Fi
deployment scenarios.
\subsubsection{Channel Sharing Coexistence}
When the Wi-Fi APs are densely deployed, it is very likely that the SBS has to share the unlicensed channel with Wi-Fi systems.
According to the regional regulatory requirements, co-channel coexistence mechanisms can be separated into two kinds:
listen-before-talk (LBT) and non-LBT markets.
In LBT markets, such as Europe and Japan, the mechanism of CSMA/CA will be used.
Before data transmission, the cognitive SBS will attempt to access the unlicensed spectrum to assess whether
there exists sufficient access opportunities. When the unlicensed spectrum is deemed idle, the SBS will activate; otherwise,
the SBS will back off one period.
Given the simplicity of this CSMA/CA protocol, it is not effective in highly dense Wi-Fi networks that rarely have
idle channels.
In non-LBT markets, such as the United States, South Korea, China, and India,
several mechanisms have also been proposed to facilitate the coexistence.
Qualcomm has proposed a duty-cycle based carrier sensing adaptive transmission (CSAT) mechanism that
measures channel utilization by neighboring nodes and adopts the on/off duty cycle of secondary cells in the unlicensed band \cite{neighbor}.
The duty cycle period is determined by LTE MAC control elements, and is adaptively chosen based on the number of measured
active Wi-Fi APs and their channel occupancy to allow a fair sharing between LTE-U and Wi-Fi.
In \cite{coe6}, a similar duty cycle mechanism called LTE muting was proposed to silence some LTE-U subframes for
Wi-Fi usage. The almost blank subframes mechanism in existing LTE standard can also be exploited to
enable the duty cycle mechanism \cite{coe8}.
Furthermore, the authors in \cite{bib_Yun} proposed to utilize multiple
antenna configurations and to exploit different frame structures between the two systems for efficient coexistence.
The cognitive radio schemes have also been taken into account for LTE and Wi-Fi coexistence in \cite{bib_Guan}.
Generally, the proposed mechanisms in non-LBT markets
are more compatible with the existing LTE standards and are simpler to implement than the LBT mechanism.
Moreover, in the high density Wi-Fi environment, the LBT mechanism is likely to fail since open channel opportunities
for LTE-U are slim. In such cases, non-LBT mechanisms can achieve a good performance by carefully selecting
the duty cycle (or muting) period.
\begin{table}[!hbp]
\renewcommand{\multirowsetup}{\centering}
\caption{LTE and Wi-Fi coexistence technologies} \label{tab_coe}
\small
\center
\begin{tabular}{|c|c|c|c|}
\hline
\rowcolor{mygray}
\bf{Types} & \bf{Features} & \bf{Ideas} & \bf{Examples} \\
\hline
\tabincell{c}{Channel-\\ Splitting} & \tabincell{c}{Channel\\ selection} & \tabincell{l}{SBS selects the channel having the \\ lowest interference power to avoid \\interference with nearby nodes.} & \tabincell{c}{Dynamic frequency\\ selection \cite{neighbor}}\\
\hline
\multirow{5}{2.2cm}{Channel-Sharing}
& LBT & \tabincell{l}{\hspace{1em}SBS adopts the CSMA/CA with\\ \hspace{1em}collision avoidance mechanism\\ \hspace{1em}to access unlicensed spectrum.} & \tabincell{c}{Dynamic duty cycle\\ with LBT \cite{DLBT}} \\
\cline{2-4}
& non-LBT & \tabincell{l}{SBS senses the channel\\ utilization then adopts\\ duty cycle of the unlicensed\\ spectrum.} & \tabincell{c}{Duty cycle based\\ CSAT \cite{neighbor}; \\ LTE muting \cite{coe6, coe8, bib_Yun, bib_Guan}} \\
\hline
\end{tabular}
\end{table}
\subsection{Coexistence from Wi-Fi Perspective} \label{section_structure_A}
Thus far, our discussions have centered on the necessary aspects of an LTE-U system to
coexist with Wi-Fi networks in the unlicensed band. Here, we will briefly summarize how
Wi-Fi MAC protocols may be affected by the presence of coexisting LTE-U
and discuss their coexistence challenges.
We start with the default and the most commonly utilized Wi-Fi channel access mechanism known as DCF.
Under the DCF mechanism, pure Wi-Fi users will have less opportunities to access the channel if they sense that LTE users are continuously occupying certain channels. Even with LBT or non-LBT mechanisms at the LTE side,
Wi-Fi transmissions could still be disrupted because of the well-known exposed-node problem, which occurs when a node is prevented from sending packets to other nodes owing to a neighboring transmitter.
Conversely, LTE transmissions will also be interfered by the Wi-Fi systems under DCF for the same reason.
Despite the popularity of the DCF mechanism, point coordination function (PCF) is a well-known optional protocol capability for Wi-Fi infrastructure mode.
Under the PCF mechanism, there exists a CFP followed by a CP in each periodic access interval and Wi-Fi nodes can be centrally coordinated by an AP for controlled channel
access with a point coordinator (PC) in the CFP. Although the PCF option has seldom been used until now,
its centralized, contention-free nature provides a perfect mechanism for LTE-U channel sharing without
interference from contention based Wi-Fi transmission.
By activating PCF, the CFP can be used by LTE-U channel access whereas the CP can be continually occupied by Wi-Fi nodes
in sharing the unlicensed band without mutual interference.
We further note that, in addition to the classic DCF and PCF, a more versatile hybrid coordination function (HCF) is another option for infrastructure Wi-Fi
access. Unlike in PCF, the CFP in the HCF includes several contention
free intervals known as HCF Controlled Channel Access (HCCA) transmission opportunity (TXOP).
Moreover, HCF based Wi-Fi nodes access the Wi-Fi channels by utilizing a QoS-aware hybrid coordinator (HC).
To ensure QoS of LTE-U users, medium access proportion of each node may be allocated according to the priority of services.
Because of the flexibility and the self-control ability of HCF, it is more suitable to LTE-U applications given
only unlicensed band.
\section{LTE-U Embedded in Wi-Fi Protocols}
In this section, we first propose a novel HAP for LTE and Wi-Fi coexistence and then introduce an innovative LTE and Wi-Fi coexistence mechanism by embedding LTE-U within existing
Wi-Fi protocol.
\subsection{A New LTE-U Framework}
One major obstacle to LTE-U is the lack of centralized coordinator in the unlicensed band, making it difficult
to manage mutual interference between the two channel sharing networks.
To tackle this problem, we propose a novel wireless AP, called HAP, which integrates both the Wi-Fi AP and SBS functionalities
into a single networking unit.
The Wi-Fi side of the HAP exclusively accesses the unlicensed spectrum,
while the LTE side of the HAP can access both licensed and unlicensed bands.
Because the HAP is deployed by the same network service provider,
it can more efficiently mitigate the mutual interference between the two networks
through careful management of both licensed and unlicensed bands.
\begin{figure}
\begin{centering}
\includegraphics[width=0.7\textwidth]{unlicensed_model.eps}
\par\end{centering}
\caption{LTE-U embedded in Wi-Fi protocol.}\label{fig_unlicensd_model}
\end{figure}
We propose an approach different from the existing LTE-U mechanisms, which is depicted in Fig. \ref{fig_unlicensd_model}.
In the CFP, a centralized coordinator controls the resource allocation scheme. When one LTE-U radio transmits during CFP,
other LTE-U and Wi-Fi radios are muted. In this way, interference between different radios and from the Wi-Fi transmission can be avoided.
On the other hand, since the transmission periods for both LTE and Wi-Fi users are completely separated,
Wi-Fi users will no longer need to compete for unlicensed resources with LTE users.
Hence, the performance of the Wi-Fi network can be separately controlled and managed.
The proposed LTE and Wi-Fi coexistence technique functions as follows. When an HAP is on, it shall first scan the unlicensed band and select the channel with the least background interference power
and the lowest user occupancy.
Thereafter, each HAP determines its CFP and CP allocations for the chosen channel at a CFP repetition interval
according to its traffic needs and other QoS considerations.
Once the setup of channel-sharing between contention-free LTE-U and contention-based
Wi-Fi completes, LTE-U and Wi-Fi users can access the unlicensed spectrum via the CFP and the CP, respectively.
In the following, we will present two LTE-U embedding proposals: unlicensed carrier aggregation (UCA) TD-LTE and standalone LTE-U operation modes.
We note that the SDL mode is a special case of the carrier aggregation TD-LTE
without the uplink transmission in LTE-U. Therefore, we do not devote special
attention to the SDL mode henceforth. We shall also note that, along with the other LTE-U networks, the following two modes
are better suited to support delay-tolerant services due to the volatility
of the Wi-Fi traffic and the uncertainty of background interference on the unlicensed spectrum.
\subsection{Unlicensed Carrier Aggregation of TD-LTE}
We now demonstrate how to implement the UCA TD-LTE mode in the proposed HAP.
As shown in the carrier aggregation phase of Fig. \ref{fig_LTEU-CA}, the HAP will first read its Wi-Fi
beacons, which are located at the start of each CFP. The beacons contain the information of the time stamp and the CFP length.
The HAP then will decide whether and when to aggregate the unlicensed and licensed bands according to the beacon, i.e., based on the available transmission interval calculated by the time stamp and CFP length embedded in the beacon.
The decision is then signalled to the LTE-U users in the licensed downlink control channel (DCCH).
Since the unlicensed spectrum is only used for
data payload transmission (i.e., to expand the downlink shared channel (DSCH) bandwidth),
both PCF and HCF mechanisms can be adopted.
The signalling exchange procedure of the proposed scheme is also shown in Fig. \ref{fig_LTEU-CA}.
First, an association request will be sent by a user equipment (UE). After receiving the association request, the
HAP will search for channels with the lowest interference power and the highest CFP vacancy, then
decide whether and when to aggregate these unlicensed bands as LTE-U channels based on its time-stamp and CFP length,
and later respond to the user with its control signals and an uplink (UL) grant.
Thereafter, a signal carrying the user identity will be sent to the HAP,
and the HAP will send Radio Resource Control (RRC) signal to decide the channel allocation given its licensed bands and the newly acquired carrier aggregation resources.
\begin{figure}
\begin{centering}
\includegraphics[width=1\textwidth]{CA_TD-LTE.eps}
\par\end{centering}
\caption{Procedure for UCA
TD-LTE mode.}\label{fig_LTEU-CA}
\end{figure}
\subsection{Standalone LTE-U}
We now propose a standalone (SA) LTE-U mode in our HAP framework. In the SA LTE-U mode,
both the control and data channels of the LTE-U system must reside in the unlicensed band without the help of
an auxiliary licensed spectrum.
In this mode, it is likely that the allowable duration of each TXOP
is too short to accommodate a LTE frame (i.e., $10$ ms). Thus,
to properly embed LTE-U into the Wi-Fi networks,
the active length of each LTE frame should be adapted.
In light of standardized LTE transmissions, we shall exploit the
discontinuous reception (DRX) and discontinuous transmission (DTX) mechanisms in the LTE protocol for LTE-U embedding \cite{DX2}.
According to the existing DTX and DRX standard formats,
the LTE frame can be shortened through silencing to fit into a TXOP duration.
By shortening an LTE frame to fit within a TXOP interval, we are able to embed SA LTE-U transmission into an existing Wi-Fi timing structure in
the unlicensed band. In fact, the Wi-Fi only users can be fully oblivious to the presence of SA LTE-U without the need for any protocol
or standard readjustment.
Fig. \ref{fig_DTX_DRX_LTEU} shows an example of a shortened LTE frame with $6$ subframes that are embedded into the TXOP duration. Each LTE subframe begins with a control region that covers $1$ to $3$ OFDM symbols. The primary and secondary synchronization signals (PSS and SSS, respectively) of the SA LTE-U
are transmitted in subframes $0$ and $5$ and the physical broadcast channel (PBCH) is transmitted in subframe $0$.
Because subframes $0$ and $5$ carry crucial synchronization information, we shall define
each TXOP duration for LTE-U to span at least $6$ subframes\footnote{According to the IEEE 802.11n standard,
TXOP duration is in the range from $32~\mu s$ to $8160~\mu s$ in increments of $32~\mu s$.} or 6 ms.
The beacons in Wi-Fi HCF carry the system information such as the time stamp, the length of CFP,
and the starting time and the maximum length of each TXOP.
Before each (shortened) SA LTE frame, there should also be a header to help with the frame synchronization. Moreover, an ACK signal will be sent at the end of each TXOP.
In what follows, we will show the operation process for both uplink and downlink.
\begin{figure}
\begin{centering}
\includegraphics[width=1\textwidth]{standalone_LTE.eps}
\par\end{centering}
\caption{Procedure for SA LTE-U mode.}\label{fig_DTX_DRX_LTEU}
\end{figure}
The part of DTX LTE-U mode in Fig. \ref{fig_DTX_DRX_LTEU} shows the process of uplink signal transmission with the DTX mechanism. In the discovery and association phase, a UE will first send an association request to the serving HAP once it has data to transmit. The serving
HAP will respond to the UE with the
beacon information and a UL grant. Thereafter, the UE will check the Wi-Fi Beacon and determine
whether to work with that HAP. After the UE joins LTE-U,
the signal transmission will go into the phase of normal data transfer,
during which both the user identity and DTX length $n$ will be sent to the HAP.
On the unlicensed band, HAP also uses SA LTE-U to exchange RRC information with the UE.
Finally, the uplink and downlink transmissions will stay active for $n$ subframes.
The UE will return to DTX mode in the ensuing $10-n$ subframes, and then wake up
again to send more data.
Similarly, the part of the DRX LTE-U mode in the Fig. \ref{fig_DTX_DRX_LTEU} shows the signalling
exchange procedure of the downlink transmission with the DRX mechanism.
A UE will wake up periodically to check PDCCH. If the UE cannot monitor the PDCCH, it would
return to sleep again; otherwise, the UE would send the pending data transmission request to the serving
HAP. Upon receiving the pending data transmission request, the serving HAP will allocate physical resources on the unlicensed spectrum to the UE. From the beacon information and the LTE-U broadcast channel,
the UE can locate the LTE-U control channels. Information such as UE identity, buffer status,
and the DRX length $n$ will be sent to
the serving HAP. Accordingly, the HAP determines the resource allocation.
Finally, the shortened $n$ subframes will carry downlink and uplink transmissions, and the UE
will sleep for the next $10-n$ subframes in DRX before waking up for more downlink reception.
\section{Performance Evaluation}
\begin{table}[]
\caption{System Parameters}\label{parameter}
\centering{
\small
\begin{tabular}{|c|c|c|c|}
\hline
Parameters & Settings & Parameters & Settings \tabularnewline
\hline
\hline
Noise power & $-174$~dBm/Hz & PHY header & $192$ bits \tabularnewline
\hline
Path loss exp. & $5$ & MAC header & $224$ bits \tabularnewline
\hline
Transmit power & $30$~dBm & Propagation delay & $20$ $\mu s$ \tabularnewline
\hline
Payload length & $1500$ byte & SIFS & $16$ $\mu s$ \tabularnewline
\hline
Bandwidth & $20$ MHz & DIFS & $50$ $\mu s$ \tabularnewline
\hline
Min. backoff window & 16 & Slot time & $9$ $\mu s$ \tabularnewline
\hline
Max. backoff window & $1024$ & ACK & $112$ bits + PHY header \tabularnewline
\hline
Max. backoff stage & $6$ & RTS & $160$ bits + PHY header \tabularnewline
\hline
WiFi channel bit rate & $130$ Mbps & CTS & $112$ bits + PHY header \tabularnewline
\hline
\end{tabular}
\end{table}
In this section, we will consider a two-tier network in which HAPs operate under the coverage of a
macrocell. We deploy $N$ Wi-Fi users and $M$ LTE-U users according to Poisson distribution
within the coverage of an HAP. Each user will associate with the proper serving
HAP according to standard LTE association strategies, and the licensed bands are orthogonally allocated to the macrocell as well as each HAP.
We let the Wi-Fi be the most popular IEEE 802.11n protocol in
the $5$ GHz unlicensed band with $20$ MHz bandwidth and certain useful parameters can be found in \cite{ave_LTE}.
We let the HAP coverage have a radius of $100$ m.
The length of the CFP repetitional interval is $100$ ms, which is the default value in the IEEE 802.11n.
The channel fading between LTE-U users and the HAP follows standard 3GPP fading model \cite{fading}. The parameters are listed in Table \ref{parameter}.
In our performance evaluation tests, we compare the HAP based network performance against two
benchmark access schemes. Specifically, our proposed LTE-U network based on the HAP framework is
compared against the pure Wi-Fi access without LTE-U deployment (labelled as ``original'' henceforth in all figures).
The LTE-U access based on the LBT protocol is also included.
For simplicity, we do not differentiate specific traffic types for each LTE user and let available channel resources be equally allocated among LTE-U and Wi-Fi users. In other words, we let the fractions of CFP and CP in the HAP network are $\frac{M}{M+N}$ and $\frac{N}{M+N}$, respectively. Note that this assignment is simple but not necessarily optimum.
Fig. 4(a) presents the comparative results of average Wi-Fi per-user throughput under
the three network service protocols.
Recall that the ``original'' network denotes the pure Wi-Fi case without serving LTE-U users,
which naturally delivers the highest Wi-Fi throughput among all three scenarios.
Because of the channel contention, the per-user Wi-Fi throughput decreases as the number of Wi-Fi users grows
under each of the three protocols.
What is more interesting is that, for smaller number of LTE-U users ($M$ = 5), the proposed HAP can improve the per-user Wi-Fi throughput by 20\%- 30\%
above that of the LBT protocol. For larger number of LTE-U users ($M = 10$), the improvement becomes more significant (30\% - 70\%).
The reason is that LBT is still a contention-based protocol whose performance would deteriorate as the number of competing users grows. On the
other hand, our newly proposed HAP-based access protocol divides the LTE-U spectrum access orthogonally from the Wi-Fi user access such that
the cross-network interference between LTE-U and Wi-Fi is avoided. The results in Fig. 4(a) demonstrate that the proposed HAP coverage
can better preserve Wi-Fi network throughput than the LBT scheme.
\begin{figure}
\center
\subfigure[Per-user Wi-Fi throughput under different networks.]{
\begin{minipage}[b]{0.5\textwidth}
\includegraphics[width=10cm]{wifi_throughput.eps}
\end{minipage}
}
\subfigure[LTE-U, Wi-Fi, and total throughput comparison in unlicensed spectrum under different LTE-U modes.]{
\begin{minipage}[b]{0.5\textwidth}
\includegraphics[width=10cm]{throughput2.eps}
\end{minipage}
}
\caption{Simulation results.}
\end{figure}
In Fig. 4(b), we further compare the
LTE-U throughput, the Wi-Fi throughput, and the overall system throughput under the three different network scenarios.
We randomly place $30$ Wi-Fi users and $10$ LTE-U users within the HAP coverage area.
In addition, the results in Fig. 4(b) illustrate that the proposed HAP architecture
achieves higher LTE-U throughput, higher Wi-Fi throughput, and higher overall system throughput when compared
with the LBT protocol.
Under HAP, the overall system throughput can be $50\%$ plus with only a mild loss of Wi-Fi throughput when compared against the original pure Wi-Fi network.
On the other hand, the LBT scheme not only exhibits a noticeable throughput loss in Wi-Fi, but also a significant drop in
the overall throughput than the proposed HAP network.
Clearly, our HAP serviced network can better utilize the unlicensed spectrum while better protecting the Wi-Fi user experience.
Furthermore, the improvement of unlicensed LTE-U throughput is also significant compared with the standalone licensed LTE throughput. Suppose the licensed bandwidth in each HAP is the same as the unlicensed bandwidth, i.e., $20$ MHz. Then we assume that one LTE user's rate within a typical 4G-LTE connection is at $8$ Mbps, which is within the average per-user throughput in practical LTE systems from $3$ to $10$ Mbps.
By allowing this user to be one of the $10$ users to access LTE-U, its throughput can be improved by as much as $71.6\%$ in the proposed HAP-based LTE-U network, rather than by $36.8\%$ in the LBT-based LTE-U network. We also test the case with $40$ Wi-Fi users, in which the proposed HAP-based LTE-U network still achieves as much as $57.3\%$ throughput gain, better than $27.8\%$ in the LBT-based LTE-U network.
Overall, we demonstrate that the shared spectrum access by LTE-U and Wi-Fi is more efficient than the conventional LBT protocol because of the more efficient LTE-U channel access. Furthermore, our proposed LTE-U embedding protocols are easy to implement.
They can be directly integrated into existing IEEE 802.11 networks without any Wi-Fi adjustment. In fact, non-LTE users can
function in normal mode and appear fully oblivious to the presence of any LTE-U users. Our test results
of the proposed LTE-U access protocol establish its network performance advantages in terms of sum throughput and per-user Wi-Fi throughput.
\section{Further Works and Open Research Issues}
Although the proposed HAP frameworks can substantially improve the spectrum efficiency and facilitate the deployment of LTE-U,
there still exist several open issues worth investigating:
\noindent
\textbf{Interference management}:
The first issue is the interference management in the unlicensed spectrum. Although we require HAPs to scan for clear
unlicensed band for LTE-U embedding, the criteria for such decision deserve further detailed study, particularly
for dense Wi-Fi deployment areas. Without a truly clean and open unlicensed channel, LTE-U has to reuse certain channels and
manage interference even during CFP. In terms of interoperability, HAPs belonging to
different operators may compete for the same unlicensed bands and, if unregulated, could lead to major
failure of LTE-U. Therefore, problems involving channel selection and interoperability must be investigated.
\noindent
\textbf{CFP and CP Allocation}:
Another question lies in the resource division between CFP and CP. For example,
when many hybrid users are admitted into the LTE-U system from Wi-Fi, it is natural to allocate longer CFP.
However, it may substantially degrade the performance of remaining Wi-Fi users relying on the shorter CP.
Conversely, if insufficient CFP is allocated, per-user LTE-U throughput would suffer whereas precious resources
on the Wi-Fi side may be under-utilized. It is therefore important to optimize the CFP and CP parameters (e.g. period, duration)
according to the network environment. The optimization depends not only on the number of users but also
on co-channel interference from adjacent cells or networks as well as the user traffic load.
\noindent
\textbf{Multiple HAP cooperation and clustering}:
Multiple HAPs
may be deployed within the coverage of a macro-cell. In this case, how to cluster and jointly
control HAPs for better system performance and how to allocate wireless resources among
different HAPs are important open problems. Distributed resource optimizations based on
game theoretic models and machine learning approaches are expected to provide important insight.
\section{Conclusion}
This article aims to propose novel frameworks for the deployment of LTE-U in the
unlicensed spectrum. We first reviewed various operation modes and coexistence techniques for LTE and Wi-Fi
networks, and then presented a novel concept of HAP by combining SBS and Wi-Fi AP functionality to
overcome many thorny issues in LTE-U and Wi-Fi coexistence.
Our focus for embedding LTE-U fully within the present Wi-Fi protocol is motivated by the natural need
to avoid retrofitting Wi-Fi networks to accommodate LTE-U channel sharing.
We as well proposed two basic LTE-U embedding approaches within the existing Wi-Fi MAC protocol based on the novel HAP framework.
Our simulation results demonstrated the substantial throughput advantage of the proposed LTE-U embedding mechanisms
over existing methods.
We further identified several open research issues for fine-tuning and optimizing our proposed LTE-U embedding
mechanisms in practical implementation.
| 3bd5dc1ac903b089b0d0b7b785ea551da9a641dd | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}\label{sec-intro}
In this paper we introduce a new higher categorical structure, called a weakly globular Tamsamani $n$-category and we prove its equivalence with a simpler higher categorical structure introduced by the author in \cite{Pa2}, called a weakly globular $n$-fold category.
Higher category theory witnessed an explosion of interest in recent years and found applications to disparate fields, such as homotopy theory \cite{Be2} \cite{L2}, algebraic geometry \cite{Simp}, mathematical physics \cite{Lu3}, logic and computer science \cite{HTT},\cite{Voe}.
In a category there are objects, arrows, identity arrows, and composition of arrows that are associative and unital. Higher categories generalize categories by admitting higher arrows (also called higher cells) and compositions between them. The behaviour of these compositions determine two main classes: strict and weak higher categories.
For each of these classes, there are higher categories which admits cells in every dimensions, and those that have cells in dimensions only up $0$ to $n$. The former have been studied extensively in relation to applications to homotopy theory, giving rise to several notions of $(\infty,n)$-category \cite{BeRe}, \cite{Be3}, \cite{Be2}, \cite{Bk1}, \cite{Lu3}, \cite{L2} as well as weak $\omega$-category \cite{Ve}.
In this work we concentrate on the second 'truncated' case, with cells in dimensions $0$ up to $n$. This is close to one of the original motivations for the development of higher categories, namely the modelling of the building blocks of spaces, the $n$-types, and it is of fundamental importance for higher category theory. It also leads to applications to homotopy theory in the search for a combinatorial description of the $k$-invariants of spaces and of simplicial categories.
In a strict higher category, compositions are associative and unital, and there is a simple way to describe strict $n$-categories via iterated enrichment. Although simple to define, strict $n$-categories are insufficient for many applications and the wider class of weak $n$-categories is needed.
In a weak $n$-category, higher cells compose in a way that is associative and unital only up to an invertible cell in the next dimension, and these associativity and unit isomorphisms are suitably compatible or coherent.
In low dimensions, the prototype of a weak $n$-category is the classical the notion of bicategory \cite{Ben} (when $n$=2) and tricategory \cite{GPS}(when $n$=3).
In dimension $n>3$, writing explicitly the coherence data for a weak $n$-category is too complex. Instead, different combinatorial models of weak $n$-category were developed, using a variety of techniques such as operads \cite{B}, \cite{Cheng2}, \cite{Lu2}, simplicial sets \cite{Simp}, \cite{Ta}, opetopes \cite{Cheng1} and many others.
A good notion of weak $n$-category needs to satisfy the homotopy hypothesis of giving an algebraic model of the building blocks of spaces (the $n$-types) when the cells in the structure are weakly invertible, that is in the weak $n$-groupoid case.
A very important model of higher categories was developed by Tamsamani and Simpson \cite{Simp}, \cite{Ta}. This model is based on the combinatorics of multi-simplicial sets. The latter are combinatorial models of spaces and feature prominently in homotopy theory \cite{Jard}.
The idea of the Tamsamani-Simpson model stems from a well-known connection between categories and simplicial sets, via the nerve functor
\begin{equation*}
N:\Cat \rw \funcat{}{\Set}\;.
\end{equation*}
This functor is fully faithful and a characterization of its image can be given in terms of Segal maps. Given a simplicial set $X$ the Segal maps for each $k\geq 2$ are maps of sets
\begin{equation*}
X_k \rw \pro{X_1}{X_0}{k}.
\end{equation*}
Then $X$ is the nerve of a category if and only if all the Segal maps are isomorphisms. Thus the Segal maps are natural candidates for the composition in a category, and the condition that they are isomorphisms determines the structure of a category.
Segal maps can be defined for any $n$-fold simplicial set $X\in \funcat{n}{\Set}$. Namely, viewing $X$ as a simplicial object in $(n-1)$-fold simplicial sets in $n$ different directions affords corresponding Segal maps. These are candidates for the compositions of the higher cells in the structures.
In the Tamsamani-Simpson model $\ta{n}$, the higher cells are encoded as discrete structures (that is, sets) in the respective dimensions. This is called the \emph{globularity condition}, since it gives rise to the globular shape of the higher cells in the structure.
If in addition to the globularity condition we require the Segal maps in all simplicial directions to be isomorphisms we obtain a strict $n$-category, described in multi-simplicial language.
The Tamsamani-Simpson model is a suitable weakening of this construction. Namely, the globularity condition is maintained, but the Segal maps are no longer isomorphisms but suitably defined higher categorical equivalences.
In this paper we introduce a new Segal-type model of higher categories based on multi-simplicial structures, called a weakly globular Tamsamani $n$-category and denoted by $\tawg{n}$. This model uses a new paradigm to weaken higher categorical structures, which is the notion of \emph{weak globularity}. This notion was introduced by the author in \cite{Pa2} with the category $\catwg{n}$ of weakly globular \nfol categories.
In the weakly globular approach to higher categories the globularity condition is replaced by weak globularity: the cells in each dimension ($0$ up to $n$) no longer form a set but have a categorical structure on their own which is equivalent in a suitable sense to a discrete one. More precisely, it is a homotopically discrete $n$-fold category in the sense defined by the author in
\cite{Pa1}.
The behaviour of the compositions is controlled by the induced Segal maps, which for $X\in\tawg{n}$ have the form for each $k\geq 2$\;
\begin{equation*}
X_k\rw\pro{X_1}{X_0^d}{k}.
\end{equation*}
Here $X_0^d$ is the discretization of the homotopically discrete object $X_0$. In a weakly globular Tamsamani $n$-category the induced Segal maps are required to be suitable higher categorical equivalences. The formal definition of $X$ is given by induction on dimension. The Segal maps are in general not isomorphisms.
Weakly globular $n$-fold categories satisfy the weak globularity condition and their induced Segal maps are higher categorical equivalences but, being $n$-fold categories, all their Segal maps are isomorphisms.
There are embeddings
\begin{equation*}
\catwg{n}\subset\tawg{n} \qquad \text{and}\qquad \ta{n}\subset\tawg{n}\;.
\end{equation*}
Thus weakly globular Tamsamani $n$-categories are a generalization both of the Tamsamani-Simpson model and of weakly globular $n$-fold categories.
The main result of this paper, Theorem \ref{the-funct-Qn}, asserts the existence of a functor, called \emph{rigidification}
\begin{equation*}
Q_n:\tawg{n}\rw\catwg{n}
\end{equation*}
such that for each $X\in\tawg{n}$ there is a suitable higher categorical equivalence $Q_n X\rw X$. This result means that $X$ can be approximated with the more rigid and therefore simpler structure $Q_n X$. In particular, this implies (see Corollary \ref{cor-the-funct-Qn}) that the two categories $\catwg{n}$ and $\tawg{n}$ are equivalent after localization with respect to the $n$-equivalences.
Rigidification of structures defined up to homotopy is an important topic in homotopy theory, where it is often tackled with model theoretic techniques \cite{Badz}. In the higher categorical setting, rigidification results have been proved for $(\infty,1)$-categories \cite{Be2} but not for the more complex case of weak $n$-categories for general $n>0$. Our result is an important step in the study of rigidification in the general higher categorical setting, and paves the way for further developments and applications.
In a subsequent paper \cite{Pa4} we will show that the rigidification functor $Q_n$ leads to a suitable equivalence between $\catwg{n}$ and the Tamsamani $n$-categories $\ta{n}$, exhibiting $\catwg{n}$ as a new model of weak $n$-categories satisfying, in particular, the homotopy hypothesis. This model will lead to new connections between different combinatorics modelling higher structures, as well as to applications to homotopy theory.
The rigidification functor factors through the category
\begin{equation*}
\segpsc{n-1}{\Cat}
\end{equation*}
of Segalic pseudo-functors introduced by the author in \cite{Pa2} whose strictification gives weakly globular \nfol categories (see \cite{Pa2}). That is, $Q_n$ is the composite
\begin{equation*}
Q_n:\tawg{n}\xrw{\ \;\;} \segpsc{n-1}{\Cat}\xrw{\St} \catwg{n}\;.
\end{equation*}
where $\St$ is the classical strictification of pseudo-functors, see \cite{Lack}, \cite{PW}.
The functor from $\tawg{n}$ to Segalic pseudo-functors further factorizes as
\begin{equation*}
\tawg{n}\xrw{P_n} \lta{n} \xrw{\Tr_{n}} \segpsc{n-1}{\Cat}
\end{equation*}
where $\lta{n}$ is a full subcategory of $\tawg{n}$ from which it is easy to construct a pseudo-functor using a general categorical technique known as of 'transport of structure along an adjunction'. The functor $P_n$ produces a functorial approximation (up to $n$-equivalence) of an object of $\tawg{n}$ with an object of $\lta{n}$.
This paper is organized as follows. In Section \ref{sec-prelim} we recall some basic multi-simplicial techniques used throughout the paper, the classical theory of strictification of pseudo-functors as well as the technique of transport of structure to produce pseudo-functors.
In Section \ref{sec-WG-nfol-cat} we recall the definition of the category $\catwg{n}$ of weakly globular \nfol categories and of Segalic pseudo-functors introduced in \cite{Pa2}.
In Section \ref{sec-WG-Tam-cat} we introduce the category $\tawg{n}$ of weakly globular Tamsamani $n$-categories and of $n$-equivalences and we prove its main properties. In particular we show in Proposition \ref{pro-crit-lev-nequiv} a criterion for an $n$-equivalence in $\tawg{n}$ to be a levelwise equivalence of categories. This is used crucially in Section \ref{sec-cat-lta} in the proofs of Lemma \ref{lem-jn-alpha} and Theorem \ref{the-repl-obj-1}.
In Section \ref{sec-func-qn} we continue the study of the category $\tawg{n}$ and we focus on the important functor
\begin{equation*}
\q{n}:\tawg{n}\rw \tawg{n-1}\;.
\end{equation*}
This functor generalizes to higher dimensions the connected components functor $q:\Cat\rw\Set$. For each $X\in\tawg{n}$ there is a map $\zg\up{n}: X\rw \di{n}\q{n}X$ where $\di{n}$ is the inclusion of $\tawg{n-1}$ in $\tawg{n}$ as a discrete structure in the top dimension. We study the properties of pullbacks along this map playing a crucial role in Section \ref{sec-cat-lta}.
In Section \ref{sec-cat-lta} we introduce the subcategory $\lta{n}\subset \tawg{n}$ and we show in Theorem \ref{the-repl-obj-1} and Corollary \ref{cor-repl-obj-2} that every object of $\tawg{n}$ is $n$-equivalent to an object of $\lta{n}$.
In Section \ref{sec-wg-tam-to-psefun}, Theorem \ref{the-XXXX} we prove the existence of a functor
\begin{equation*}
Tr_n: \lta{n}\rw \segpsc{n-1}{\Cat}
\end{equation*}
We then use the functor $Tr_{n}$ and the results of Section \ref{sec-cat-lta} to build the rigidification functor $Q_n$.
\bk
\textbf{Acknowledgements}: This work is supported by a Marie Curie International Reintegration Grant No 256341. I thank the Centre for Australian Category Theory for their hospitality and financial support during August-December 2015, as well as the University of Leicester for its support during my study leave. I also thank the University of Chicago for their hospitality and financial support during April 2016.
\section{Preliminaries}\label{sec-prelim}
In this section we review some basic simplicial techniques that we will use throughout the paper as well as some categorical background on pseudo-functors and their strictification, and on a technique to produce pseudo-functors. The material in this section is well-known, see for instance \cite{Borc}, \cite{Jard}, \cite{lk}, \cite{PW}, \cite{Lack}.
\subsection{Simplicial objects}\label{sbs-simp-tech}
Let $\zD$ be the simplicial category and let $\Dnop$ denote the product of $n$ copies of $\Dop$. Given a category $\clC$, $\funcat{n}{\clC}$ is called the category of $n$-simplicial objects in $\clC$ (simplicial objects in $\clC$ when $n=1$).
\begin{notation}\label{not-simp}
If $X\in \funcat{n}{\clC}$ and $\uk=([k_1],\ldots,[k_n])\in \Dnop$, we shall denote $X ([k_1],\ldots,[k_n])$ by $X(k_1,\ldots,k_n)$, as well as $X_{k_1,\ldots,k_n}$ and $X_{\uk}$. We shall also denote $\uk(1,i)=([k_1],\ldots,[k_{i-1}],1,[k_{i+1}],\ldots,[k_n]) \in \Dnop$ for $1\leq i\leq n$.
Every $n$-simplicial object in $\clC$ can be regarded as a simplicial object in $\funcat{n-1}{\clC}$ in $n$ possible ways. For each $1\leq i\leq n$ there is an isomorphism
\begin{equation*}
\xi_i:\funcat{n}{\clC}\rw\funcat{}{\funcat{n-1}{\clC}}
\end{equation*}
given by
\begin{equation*}
(\xi_i X)_{r}(k_1,\ldots,k_{n-1})=X(k_1,\ldots,k_{i-1},r,k_{i+1},\ldots,k_{n-1})
\end{equation*}
for $X\in\funcat{n}\clC$ and $r\in\Dop$.
\end{notation}
\begin{definition}\label{def-fun-smacat}
Let $F:\clC \rw \clD$ be a functor, $\clI$ a small category. Denote
\begin{equation*}
\ovl{F}:[\clI,\clC]\rw [\clI,\clD]
\end{equation*}
the functor given by
\begin{equation*}
(\ovl{F} X)_i = F(X(i))
\end{equation*}
for all $i\in\clI$.
\end{definition}
\begin{definition}\label{def-seg-map}
Let ${X\in\funcat{}{\clC}}$ be a simplicial object in any category $\clC$ with pullbacks. For each ${1\leq j\leq k}$ and $k\geq 2$, let ${\nu_j:X_k\rw X_1}$ be induced by the map $[1]\rw[k]$ in $\Delta$ sending $0$ to ${j-1}$ and $1$ to $j$. Then the following diagram commutes:
\begin{equation}\label{eq-seg-map}
\xymatrix@C=20pt{
&&&& X\sb{k} \ar[llld]_{\nu\sb{1}} \ar[ld]_{\nu\sb{2}} \ar[rrd]^{\nu\sb{k}} &&&& \\
& X\sb{1} \ar[ld]_{d\sb{1}} \ar[rd]^{d\sb{0}} &&
X\sb{1} \ar[ld]_{d\sb{1}} \ar[rd]^{d\sb{0}} && \dotsc &
X\sb{1} \ar[ld]_{d\sb{1}} \ar[rd]^{d\sb{0}} & \\
X\sb{0} && X\sb{0} && X\sb{0} &\dotsc X\sb{0} && X\sb{0}
}
\end{equation}
If ${\pro{X_1}{X_0}{k}}$ denotes the limit of the lower part of the
diagram \eqref{eq-seg-map}, the \emph{$k$-th Segal map for $X$} is the unique map
$$
\muk:X\sb{k}~\rw~\pro{X\sb{1}}{X\sb{0}}{k}
$$
\noindent such that ${\pr_j\,\muk=\nu\sb{j}}$ where
${\pr\sb{j}}$ is the $j^{th}$ projection.
\end{definition}
\begin{definition}\label{def-ind-seg-map}
Let ${X\in\funcat{}{\clC}}$ and suppose that there is a map $\zg: X_0 \rw X^d_0$ in $\clC$ $\zg: X_0 \rw X^d_0$ such that the limit of the diagram
\begin{equation*}
\xymatrix@R25pt@C16pt{
& X\sb{1} \ar[ld]_{\zg d\sb{1}} \ar[rd]^{\zg d\sb{0}} &&
X\sb{1} \ar[ld]_{\zg d\sb{1}} \ar[rd]^{\zg d\sb{0}} &\cdots& k &\cdots&
X\sb{1} \ar[ld]_{\zg d\sb{1}} \ar[rd]^{\zg d\sb{0}} & \\
X^d\sb{0} && X^d\sb{0} && X^d\sb{0}\cdots &&\cdots X^d\sb{0} && X^d\sb{0}
}
\end{equation*}
exists; denote the latter by $\pro{X_1}{X_0^d}{k}$. Then the following diagram commutes, where $\nu_j$ is as in Definition \ref{def-seg-map}, and $k\geq 2$
\begin{equation*}
\xymatrix@C=20pt{
&&&& X\sb{k} \ar[llld]_{\nu\sb{1}} \ar[ld]_{\nu\sb{2}} \ar[rrd]^{\nu\sb{k}} &&&& \\
& X\sb{1} \ar[ld]_{\zg d\sb{1}} \ar[rd]^{\zg d\sb{0}} &&
X\sb{1} \ar[ld]_{\zg d\sb{1}} \ar[rd]^{\zg d\sb{0}} && \dotsc &
X\sb{1} \ar[ld]_{\zg d\sb{1}} \ar[rd]^{\zg d\sb{0}} & \\
X^d\sb{0} && X^d\sb{0} && X^d\sb{0} &\dotsc X^d\sb{0} && X^d\sb{0}
}
\end{equation*}
The \emph{$k$-th induced Segal map for $X$} is the unique map
\begin{equation*}
\hmuk:X\sb{k}~\rw~\pro{X\sb{1}}{X^d\sb{0}}{k}
\end{equation*}
such that ${\pr_j\,\hmuk=\nu\sb{j}}$ where ${\pr\sb{j}}$ is the $j^{th}$ projection.
\end{definition}
\subsection{$\mathbf{n}$-Fold internal categories}\label{sbs-nint-cat}
Let $\clC$ be a category with finite limits. An internal category $X$ in $\clC$ is a diagram in $\clC$
\begin{equation}\label{eq-nint-cat}
\xymatrix{
\tens{X_1}{X_0} \ar^(0.65){m}[r] & X_1 \ar^{d_0}[r]<2.5ex> \ar^{d_1}[r] & X_0
\ar^{s}[l]<2ex>
}
\end{equation}
where $m,d_0,d_1,s$ satisfy the usual axioms of a category (see for instance \cite{Borc} for details). An internal functor is a morphism of diagrams like \eqref{eq-nint-cat} commuting in the obvious way. We denote by $\Cat \clC$ the category of internal categories and internal functors.
The category $\cat{n}(\clC)$ of \nfol categories in $\clC$ is defined inductively by iterating $n$ times the internal category construction. That is, $\cat{1}(\clC)=\Cat$ and, for $n>1$,
\begin{equation*}
\cat{n}(\clC)= \Cat(\cat{n-1}(\clC)).
\end{equation*}
When $\clC=\Set$, $\cat{n}(\Set)$ is simply denoted by $\cat{n}$ and called the category of \nfol categories (double categories when $n=2$).
\subsection{Nerve functors}\label{sus-ner-funct}
There is a nerve functor
\begin{equation*}
N:\Cat\clC \rw \funcat{}{\clC}
\end{equation*}
such that, for $X\in\Cat\clC$
\begin{equation*}
(N X)_k=
\left\{
\begin{array}{ll}
X_0, & \hbox{$k=0$;} \\
X_1, & \hbox{$k=1$;} \\
\pro{X_1}{X_0}{k}, & \hbox{$k>1$.}
\end{array}
\right.
\end{equation*}
When no ambiguity arises, we shall sometimes denote $(NX)_k$ by $X_k$ for all $k\geq 0$.
The following fact is well known:
\begin{proposition}\label{pro-ner-int-cat}
A simplicial object in $\clC$ is the nerve of an internal category in $\clC$ if and only if all the Segal maps are isomorphisms.
\end{proposition}
By iterating the nerve construction, we obtain the multinerve functor
\begin{equation*}
\Nn :\cat{n}(\clC)\rw \funcat{n}{\clC}\;.
\end{equation*}
\begin{definition}\label{def-discrete-nfold}
An internal $n$-fold category $X\in \cat{n}(\clC)$ is said to be discrete if $\Nn X$ is a constant functor.
\end{definition}
Each object of $\cat{n}(\clC)$ can be considered as an internal category in $\cat{n-1}(\clC)$ in $n$ possible ways, corresponding to the $n$ simplicial directions of its multinerve. To prove this, we use the following lemma, which is a straightforward consequence of the definitions.
\begin{lemma}\label{lem-multin-iff}\
\begin{itemize}
\item [a)] $X\in\funcat{n}{\clC}$ is the multinerve of an \nfol category in $\clC$ if and only if, for all $1\leq r\leq n$ and $[p_1],\ldots,[p_r]\in\Dop$, $p_r\geq 2$
\begin{equation}\label{eq-multin-iff}
\begin{split}
& X(p_1,...,p_r,\mi)\cong\\
&\resizebox{1.0\hsize}{!}{$\cong\pro{X(p_1,...,p_{r-1},1,\mi)}{X(p_1,...,p_{r-1},0,\mi)}{p_r}$}
\end{split}
\end{equation}
\item [b)] Let $X\in\cat{n}(\clC)$. For each $1\leq k\leq n$, $[i]\in\Dop$, there is $X_i\up{k}\in\cat{n-1}(\clC)$ with
\begin{equation*}
\Nb{n-1}X_i\up{k} (p_1,\ldots,p_{n-1})=\Nn X(p_1,\ldots,p_{k-1},i,p_k,\ldots,p_{n-1})
\end{equation*}
\end{itemize}
\end{lemma}
\begin{proof}\
a) By induction on $n$. By Proposition \ref{pro-ner-int-cat}, it is true for $n=1$. Suppose it holds for $n-1$ and let $X\in\Cat(\cat{n-1}(\clC))$ with objects of objects (resp. arrows) $X_0$ (resp. $X_1$); denote $(NX)_p=X_p$. By definition of the multinerve
\begin{equation*}
(\Nb{n} X)(p_1,\ldots,p_r,\mi)=\Nb{n-1}X_{p_1}(p_2,\ldots,p_r,\mi)\;.
\end{equation*}
Hence using the induction hypothesis
\begin{align*}
&\Nb{n}X(p_1...p_r\,\mi)=\Nb{n-1}X_{p_1}(p_2... p_r\,\mi)\cong\\
&\resizebox{1.0\hsize}{!}{$
\cong \pro{\Nb{n-1} X_{p_1}(p_2... p_{r-1}\,1\,\mi)}
{\Nb{n-1} X_{p_1}(p_2... p_{r-1}\,0\,\mi)}{p_r}=$}\\
&\resizebox{1.0\hsize}{!}{
$=\pro{\Nb{n} X(p_1... p_{r-1}\,1\,\mi)}
{\Nb{n} X(p_1... p_{r-1}\,0\,\mi)}{p_r}.$}
\end{align*}
Conversely, suppose $X\in\funcat{n}{\clC}$ satisfies \eqref{eq-multin-iff}. Then for each $[p]\in\Dop$, $X(p,\mi)$ satisfies \eqref{eq-multin-iff}, hence
\begin{equation*}
X(p,\mi)=\Nb{n-1}X_p
\end{equation*}
for $X_p\in\cat{n-1}(\clC)$. Also, by induction hypothesis
\begin{equation*}
X(p,\mi)=\pro{X(1,\mi)}{X(0,\mi)}{p}\;.
\end{equation*}
Thus we have the object $X\in\cat{n}(\clC)$ with objects $X_0$, arrows $X_1$ and $X_p=X(p,\mi)$ as above.
\bigskip
b) By part a), there is an isomorphism for $p_r\geq 2$
\newpage
\begin{flalign*}
&\Nb{n}X(p_1...p_n)=&
\end{flalign*}
\begin{equation*}
\resizebox{1.0\hsize}{!}{$\pro{\Nb{n}X(p_1...p_{r-1}\, 1 ...p_n)}{\Nb{n}X(p_1...p_{r-1}\, 0 ...p_n)}{p_r}$}\;.
\end{equation*}
In particular, evaluating this at $p_k=i$, this is saying the $(n-1)$-simplicial group taking $(p_1...p_n)$ to $\Nb{n}X(p_1...p_{k-1}\, i ...p_{n-1})$ satisfies condition \eqref{eq-multin-iff} in part a). Hence by part a) there exists $X_i\up{k}$ with
\begin{equation*}
\Nb{n-1}X_i\up{k}(p_1...p_{n-1})=\Nb{n}X(p_1...p_{k-1}\, i ...p_{n-1})
\end{equation*}
as required.
\end{proof}
\begin{proposition}\label{pro-assoc-iso}
For each $1\leq k\leq n$ there is an isomorphism $\xi_k:\cat{n}(\clC)\rw \cat{n}(\clC)$ which associates to $X=\cat{n}(\clC)$ an object $\xi_k X$ of $\Cat(\cat{n-1}(\clC))$ with
\begin{equation*}
(\xi_k X)_i=X_i\up{k}\qquad i=0,1
\end{equation*}
with $X_i\up{k}$ as in Lemma \ref{lem-multin-iff}.
\end{proposition}
\begin{proof}
Consider the object of $\funcat{}{\funcat{n-1}{\clC}}$ taking $i$ to the $\nm$-simplicial object associating to $(\seqc{p}{1}{n-1})$ the object
\begin{equation*}
\Nb{n}X(p_1...p_{k-1}\, i \, p_{k+1}...p_{n-1})\;.
\end{equation*}
By Lemma \ref{lem-multin-iff} b), the latter is the multinerve of $X_i\up{k}\in \cat{n-1}(\clC)$. Further, by Lemma \ref{lem-multin-iff} a), we have
\begin{equation*}
\Nb{n-1}X_i\up{k}\cong \pro{\Nb{n-1}X_1\up{k}}{\Nb{n-1}X_0\up{k}}{i}\;.
\end{equation*}
Hence $\Nb{n}X$ as a simplicial object in $\funcat{n-1}{\clC}$ along the $k^{th}$ direction, has
\begin{equation*}
(\Nb{n}X)_i=
\left\{
\begin{array}{ll}
\Nb{n-1}X_i\up{k}, & \hbox{$i=0,1$;} \\
\Nb{n-1}(\pro{X_1\up{k}}{X_0\up{k}}{i}), & \hbox{for $i\geq 2$.}
\end{array}
\right.
\end{equation*}
This defines $\xi_k X\in\Cat(\cat{n-1}(\clC))$ with
\begin{equation*}
(\xi_k X)_i=\Nb{n-1}X_i\up{k}\qquad i=0,1\;.
\end{equation*}
We now define the inverse for $\xi_k$. Let $X\in\Cat(\cat{n-1}(\clC))$, and let $X_i=\pro{X_1}{X_0}{i}$ for $i\geq 2$. The $n$-simplicial object $X_{k}$ taking $(p_1,\ldots,p_n)$ to
\begin{equation*}
\Nb{n}X_{p_k}(p_1...p_{k-1}p_{k+1}...p_n)
\end{equation*}
satisfies condition \eqref{eq-multin-iff}, as easily seen. Hence by Lemma \ref{lem-multin-iff} there is $\xi'_k X\in\cat{n}\clC$ such that $\Nb{n }\xi'_k X=X_k$. It is immediate to check that $\xi_k$ and $\xi'_k$ are inverse bijections.
\end{proof}
\begin{definition}\label{def-ner-func-dirk}
The nerve functor in the $k^{th}$ direction is defined as the composite
\begin{equation*}
\Nu{k}:\cat{n}(\clC)\xrw{\xi_k}\Cat(\cat{n-1}(\clC))\xrw{N}\funcat{}{\cat{n-1}(\clC)}
\end{equation*}
so that, in the above notation,
\begin{equation*}
(\Nu{k}X)_i=X\up{k}_i\qquad i=0,1\;.
\end{equation*}
Note that $\Nb{n}=\Nu{n}...\Nu{2}\Nu{1}$.
\end{definition}
\begin{notation}\label{not-ner-func-dirk}
When $\clC=\Set$ we shall denote
\begin{equation*}
J_n=\Nu{n-1}\ldots \Nu{1}:\cat{n}\rw\funcat{n-1}{\Cat}\;.
\end{equation*}
\end{notation}
Thus $J_n$ amounts to taking the nerve construction in all but the last simplicial direction. The functor $J_n$ is fully faithful, thus we can identify $\cat{n}$ with the image $J_n(\cat{n})$ of the functor $J_n$.
Given $X\in\cat{n}$, when no ambiguity arises we shall denote, for each $(s_1,\ldots ,s_{n-1})\in\Dmenop$
\begin{equation*}
X_{s_1,\ldots ,s_{n-1}}=(J_n X)_{s_1,\ldots ,s_{n-1}}\in\Cat
\end{equation*}
and more generally, if $1\leq j \leq n-1$,
\begin{equation*}
X_{s_1,\ldots ,s_{j}}=(\Nu{j}\ldots \Nu{1} X)_{s_1,\ldots ,s_{j}}\in\cat{n-j}\;.
\end{equation*}
Let $ob : \Cat \clC \rw \clC$ be the object of object functor. The left adjoint to $ob$ is the discrete internal category functor $d$. By Proposition \ref{pro-assoc-iso} we then have
\begin{equation*}
\xymatrix{
\cat{n}\clC \oset{\xi_n}{\cong}\Cat(\cat{n-1}\clC) \ar@<1ex>[r]^(0.65){ob} & \cat{n-1}\clC \ar@<1ex>[l]^(0.35){d}\;.
}
\end{equation*}
We denote
\begin{equation*}
\di{n}=\xi^{-1}_{n}\circ d \text{\;\;for\;\;} n>1,\; \di{1}=d \;.
\end{equation*}
Thus $\di{n}$ is the discrete inclusion of $\cat{n-1}\clC$ into $\cat{n}\clC$ in the $n^{th}$ direction.
\bk
The following is a characterization of objects of $\funcat{n-1}{\Cat}$ in the image of the functor $J_n$ in \ref{not-ner-func-dirk}.
\begin{lemma}\label{lem-char-obj}
Let $L\in \funcat{n-1}{\Cat}$ be such that, for all $\uk\in\dop{n-1}$, $1\leq i\leq n-1$ and $k_i\geq 2$, the Segal maps are isomorphisms:
\begin{equation}\label{eq-lem-char-obj}
L_{\uk} \cong\pro{L_{\uk(1,i)}}{L_{\uk(0,i)}}{k_i}\;.
\end{equation}
Then $L\in\cat{n}$.
\end{lemma}
\begin{proof}
By induction on $n$. When $n=2$, $L\in\funcat{}{\Cat}$, $k\in \Dop$, $i=1$, $\uk(1,i)=1$, $\uk(0,i)=0$, $k_i=k=2$ and
\begin{equation*}
L_k\cong\pro{L_1}{L_0}{k}\;.
\end{equation*}
Thus by Proposition \ref{pro-ner-int-cat}, $L\in\cat{2}$.
Suppose the lemma holds for $(n-1)$ and let $L \in \funcat{n-1}{\Cat}$ be as in the hypothesis. Consider $L_j\in\funcat{n-2}{\Cat}$ for $j\geq 0$. Let $\ur\in\dop{n-2}$ and denote $\uk=(j,\ur)\in\dop{n-1}$. Then, for any $2\leq i\leq n-1$, $k_i=r_{i-1}$ and
\begin{equation*}
L_{\uk}=(L_j)_{\ur},\quad L_{\uk(1,i)}=(L_j)_{\ur(1,i-1)},\quad L_{\uk(0,i)}=(L_j)_{\ur(0,i-1)}\;.
\end{equation*}
Therefore \eqref{eq-lem-char-obj} implies
\begin{equation*}
(L_j)_{\ur}=\pro{(L_j)_{\ur(1,i-1)}}{(L_j)_{\ur(0,i-1)}}{r_{i-1}}\;.
\end{equation*}
This means that $L_j$ satisfies the inductive hypothesis and therefore $L_j\in\cat{n-1}$.
Taking $i=1$ in \eqref{eq-lem-char-obj} we see that, for each $k_1\geq 2$ and $\ur=(k_2,\ldots,k_{n-1})\in\dop{n-2}$,
\begin{equation*}
(L_{k_1})_{\ur}=\pro{(L_1)_{\ur}}{(L_0)_{\ur}}{k_1}\;.
\end{equation*}
That is, we have isomorphisms in $\cat{n-1}$
\begin{equation*}
L_{k_1}\cong \pro{L_1}{L_0}{k_1}\;.
\end{equation*}
We conclude from Proposition \ref{pro-ner-int-cat} that $L\in\cat{n}$.
\end{proof}
\begin{lemma}\label{lem-char-obj-II}
Let $P$ be the pullback in $\funcat{n-1}{\Cat}$ of the diagram in $A\rw C \lw B$. Suppose that for each $k\geq 2$ there are isomorphisms of Segal maps in $\funcat{n-2}{\Cat}$
\begin{equation*}
A_k\cong \pro{A_1}{A_0}{k},\;\; C_k\cong \pro{C_1}{C_0}{k},\;\; B_k\cong \pro{B_1}{B_0}{k}\;.
\end{equation*}
Then $P_k\cong \pro{P_1}{P_0}{k}$.
\end{lemma}
\begin{proof}
We show this for $k=2$, the case $k>2$ being similar. Since the nerve functor $N:\Cat\rw\funcat{}{\Set}$ commutes with pullbacks (as it is right adjoint) and pullbacks in $\funcat{n-1}{\Cat}$ are computed pointwise, for each $\us\in\dop{n-1}$ we have a pullback in $\Set$
\begin{equation*}
\xymatrix{
(NP)_{2\us} \ar[r] \ar[d] & (NA)_{2\us} \ar[d]\\
(NC)_{2\us} \ar[r] & (NC)_{2\us}
}
\end{equation*}
where
\begin{equation*}
(NA)_{2\us}=\tens{(NA)_{1\us}}{(NA)_{0\us}}
\end{equation*}
and similarly for $NC$ and $NB$. We then calculate
\begin{align*}
&(NP)_{2\us}=(NA)_{2\us} \tiund{(NC)_{2\us}} (NB)_{2\us} =\\
& \resizebox{1.0\hsize}{!}{$
=\{ \tens{(NA)_{1\us}}{(NA)_{0\us}} \}\tiund{\tens{(NC)_{1\us}}{(NC)_{0\us}}} \{ \tens{(NB)_{1\us}}{(NB)_{0\us}}\}\cong $}\\
& \resizebox{1.0\hsize}{!}{$
\cong\{ \tens{(NB)_{1\us}}{(NC)_{0\us}} \}\tiund{\tens{(NA)_{1\us}}{(NC)_{0\us}}} \{ \tens{(NB)_{1\us}}{(NC)_{0\us}}\}= $}\\
&= \tens{(NP)_{1\us}}{(NP)_{0\us}}\;.
\end{align*}
In the above, the isomorphism before the last takes $(x_1,x_2,x_3,x_4)$ to $(x_1,x_3,x_2,x_4)$. Since this holds for all $\us$, it follows that
\begin{equation*}
P_2\cong \tens{P_1}{P_0}\;.
\end{equation*}
The case $k>2$ is similar.
\end{proof}
\subsection{Some functors on $ \Cat$}\label{sbs-funct-cat}
The connected component functor
\begin{equation*}
q: \Cat\rw \Set
\end{equation*}
associates to a category its set of paths components. This is left adjoint to the discrete category functor
\begin{equation*}
\di{1}:\Set \rw \Cat
\end{equation*}
associating to a set $X$ the discrete category on that set. We denote by
\begin{equation*}
\zgu{1}:\Id\Rw \di{1}q
\end{equation*}
the unit of the adjunction $q\dashv \di{1}$.
\begin{lemma}\label{lem-q-pres-fib-pro}
$q$ preserves fiber products over discrete objects and sends
equivalences of categories to isomorphisms.
\end{lemma}
\begin{proof}
We claim that $q$ preserves products; that is, given categories
$\clC$ and $\clD$, there is a bijection
\begin{equation*}
q(\clC\times \clD)=q(\clC)\times q(\clD)\;.
\end{equation*}
In fact, given $(c,d)\in q(\clC\times \clD)$ the map
$q(\clC\times\clD)\rw q(\clC)\times q(\clD)$ given by
$[(c,d)]=([c],[d])$ is well defined and is clearly surjective. On
the other hand, this map is also injective: given $[(c,d)]$ and
$[(c',d')]$ with $[c]=[c']$ and $[d]=[d']$, we have paths in $\clC$
\newcommand{\lin}{-\!\!\!-\!\!\!-}
\begin{equation*}
\xymatrix @R5pt{c \hspace{2mm} \lin \hspace{2mm}\cdots \hspace{2mm}
\lin
\hspace{2mm}c'\\
d \hspace{2mm} \lin \hspace{2mm}\cdots \hspace{2mm} \lin
\hspace{2mm}d' }
\end{equation*}
and hence a path in $\clC\times \clD$
\begin{equation*}
(c,d)\hspace{2mm}\lin\hspace{2mm}\cdots\hspace{2mm}\lin\hspace{2mm}(c',d)
\hspace{2mm}\lin\hspace{2mm}\cdots\hspace{2mm}\lin\hspace{2mm}(c',d')\;.
\end{equation*}
Thus $[(c,d)]=[(c',d')]$ and so the map is also injective, hence it
is a bijection, as claimed.
Given a diagram in $\Cat$ $\xymatrix{\clC\ar_{f}[r] & \clE & \clD
\ar^{g}[l]}$ with $\clE$ discrete, we have
\begin{equation}\label{eq-q-pres-fib-pro}
\clC\tiund{\clE}\clD=\underset{x\in\clE}{\coprod}\clC_x\times
\clD_x
\end{equation}
where $\clC_x,\;\clD_x$ are the full subcategories of $\clC$ and
$\clD$ with objects $c,\;d$ such that \;$f(c)=x=g(d)$. Since $q$ preserves
products and (being left adjoint) coproducts, we conclude by
\eqref{eq-q-pres-fib-pro} that
\begin{equation*}
q(\clC\tiund{\clE}\clD)\cong q(\clC)\tiund{\clE}\,q(\clD)\;.
\end{equation*}
Finally, if $F:\clC\simeq \clD:G$ is an equivalence of categories,
$FG\,\clC\cong\clC$ and $FG\,\clD\cong \clD$ which implies that
$qF\,qG\,\clC\cong q\clC$ and $qF\,qG\,\clD\cong q\clD$, so $q\clC$
and $q\clD$ are isomorphic.
\end{proof}
The isomorphism classes of objects functot
\begin{equation*}
p:\Cat\rw\Set
\end{equation*}
associates to a category the set of isomorphism classes of its objects. Notice that if $\clC$ is a groupoid, $p\clC=q\clC$.
\begin{lemma}\label{lem-iso-cla-obj-fib-pro}\
\begin{itemize}
\item [a)] Let $X\xrw{f} Z \xlw{g}Y$ be a diagram in $\Cat$. Then
\begin{equation*}
p(X\tiund{Z}Y)\subseteq pX\tiund{pZ}pY\;.
\end{equation*}
\item [b)] Suppose, further, that $g_0=\Id$. Then
\begin{equation*}
p(X\tiund{Z}Y)\cong pX\tiund{pZ}pY\;.
\end{equation*}
\end{itemize}
\end{lemma}
\begin{proof}\
\nid a) The map
\begin{equation*}
j:p(X\tiund{Z}Y)\rw pX\tiund{pZ}pY
\end{equation*}
is determined by the maps
\begin{equation*}
p(X\tiund{Z}Y)\rw pX\quad \text{and}\quad p(X\tiund{Z}Y)\rw pY
\end{equation*}
induced by the projections
\begin{equation*}
X\tiund{Z}Y \rw X \quad \text{and}\quad X\tiund{Z}Y \rw Y\;.
\end{equation*}
Thus, for each $(a,b)\in X\tiund{Z}Y$,
\begin{equation}\label{eq1-lem-iso-cla}
j\,p(a,b)=(p(a),p(b))\;.
\end{equation}
Let $(a,b), (a',b')\in X\tiund{Z}Y$ be such that $p(a,b)=p(a',b')$. It follows by \eqref{eq1-lem-iso-cla} that $p(a)=p(a')$ and $p(b)=p(b')$. Thus there are isomorphisms $\za:a\cong a'$ in $X$ and $\zb: b\cong b'$ in $Y$ and in $Z$ we have
\begin{equation*}
f\za=g\zb: fa=gb\cong fa'=gb'\;.
\end{equation*}
Thus $(\za,\zb):(a,b)\cong (a',b')$ is an isomorphism in $X\tiund{Z} Y$ and so $p(a,b) = p(a',b')$. This shows that $j$ is injective, proving a).
\mk
\nid b) By a) the map $j$ is injective. We now show that, if $g_0=\Id$, then $j$ is also surjective. Let $(px,py)\in pX\tiund{pZ} pY$ be such that $g_0=\Id$. Then
\begin{equation*}
f p(x)= p f(x)=gp(y)=pg(y)=p(y)\;.
\end{equation*}
Also, $f(x)=g(f(x))$, so that $(x,f(x))\in X\tiund{Z}Y$. It follows that
\begin{equation*}
j(x,f(x))=(p(x),pf(x))=(p(x),p(y))
\end{equation*}
so that $j$ is surjective. Hence $j$ is a bijection.
\end{proof}
\subsection{Pseudo-functors and their strictification}\label{sbs-pseudo-functors}
We recall the classical theory of strictification of pseudo-functors, see \cite{PW}, \cite{Lack}.
The functor 2-category $\funcat{n}{\Cat}$ is 2-monadic over $[ob(\Dnop),\Cat]$ where $ob(\Dnop)$ is the set of objects of $\Dnop$. Let
\begin{equation*}
U:\funcat{n}{\Cat}\rw [ob(\Dnop),\Cat]
\end{equation*}
be the forgetful functor $(UX)_{\uk}=X_{\uk}$. Its left adjoint $F$ is given on objects by
\begin{equation*}
(FH)_{\uk}=\underset{\ur\in ob(\Dnop)}{\coprod}\Dnop(\ur,\uk)\times H_{\ur}
\end{equation*}
for $H\in [ob(\Dmenop),\Cat]$, $\uk\in ob(\Dmenop)$. If $T$ is the monad corresponding to the adjunction $F\dashv U$, then
\begin{equation*}
(TH)_{\uk}=\underset{\ur\in ob(\Dnop)}{\coprod}\Dnop(\ur,\uk)\times H_{\ur}
\end{equation*}
A pseudo $T$-algebra is given by $H\in [ob(\Dnop),\Cat]$,
\begin{equation*}
h_{n}: \underset{\ur\in ob(\Dnop)}{\coprod}\Dnop(\ur,\uk)\times H_{\ur} \rw H_{\uk}
\end{equation*}
and additional data, as described in \cite{PW}. This amounts precisely to functors from $\Dnop$ to $\Cat$ and the 2-category $\sf{Ps\mi T\mi alg}$ of pseudo $T$-algebras corresponds to the 2-category $\Ps\funcat{n}{\Cat}$ of pseudo-functors, pseudo-natural transformations and modifications.
The strictification result proved in \cite{PW} yields that every pseudo-functor from $\Dnop$ to $\Cat$ is equivalent, in $\Ps\funcat{n}{\Cat}$, to a 2-functor.
Given a pseudo $T$-algebra as above, \cite{PW} consider the factorization of $h:TH\rw H$ as
\begin{equation*}
TH\xrw{v}L\xrw{g}H
\end{equation*}
with $v_{\uk}$ bijective on objects and $g_{\uk}$ fully faithful, for each $\uk\in\Dnop$. It is shown in \cite{PW} that it is possible to give a strict $T$-algebra structure $TL\rw L$ such that $(g,Tg)$ is an equivalence of pseudo $T$-algebras. It is immediate to see that, for each $\uk\in\Dnop$, $g_{\uk}$ is an equivalence of categories.
Further, it is shown in \cite{Lack} that $\St:\psc{n}{\Cat}\rw\funcat{n}{\Cat}$ as described above is left adjoint to the inclusion
\begin{equation*}
J:\funcat{n}{\Cat}\rw\psc{n}{\Cat}
\end{equation*}
and that the components of the units are equivalences in $\psc{n}{\Cat}$.
\subsection{Transport of structure}\label{transport-structure} We now recall a general categorical technique, known as transport of structure along an adjunction, with one of its applications. This will be used crucially in the proof of Theorem \ref{the-XXXX}.
\begin{theorem}\rm{\cite[Theorem 6.1]{lk}}\em\label{s2.the1}
Given an equivalence $\;\eta,\;\zve : f \dashv f^* : A\rw B$ in the complete and locally small
2-category $\clA$, and an algebra $(A,a)$ for the monad $T=(T,i,m)$ on $\clA$, the
equivalence enriches to an equivalence
\begin{equation*}
\eta,\zve:(f,\ovll{f})\vdash (f^*,\ovll{f^*}):(A,a)\rw(B,b,\hat{b},\ovl{b})
\end{equation*}
in $\PsTalg$, where $\hat{b}=\eta$, $\;\ovl{b}=f^* a\cdot T\zve \cdot
Ta\cdot T^2 f$, $\;\ovll{f}=\zve^{-1} a\cdot Tf$, $\;\ovll{f^*}=f^* a\cdot
T\zve$.
\end{theorem}
Let $\eta',\zve':f'\vdash f'^{*}:A'\rw B'$ be another equivalence in $\clA$ and
let $(B',b',\hat{b'},\ovl{b'})$ be the corresponding pseudo-$T$-algebra as in
Theorem \ref{s2.the1}. Suppose $g:(A,a)\rw(A',a')$ is a morphism in $\clA$ and
$\gamma$ is an invertible 2-cell in $\clA$
\begin{equation*}
\xy
0;/r.10pc/:
(-20,20)*+{B}="1";
(-20,-20)*+{B'}="2";
(20,20)*+{A}="3";
(20,-20)*+{A'}="4";
{\ar_{f^*}"3";"1"};
{\ar_{h}"1";"2"};
{\ar^{f'^*}"4";"2"};
{\ar^{g}"3";"4"};
{\ar@{=>}^{\gamma}(0,3);(0,-3)};
\endxy
\end{equation*}
Let $\ovl{\gamma}$ be the invertible 2-cell given by the following pasting:
\begin{equation*}
\xy
0;/r.15pc/:
(-40,40)*+{TB}="1";
(40,40)*+{TB'}="2";
(-40,-40)*+{B}="3";
(40,-40)*+{B'}="4";
(-20,20)*+{TA}="5";
(20,20)*+{TA'}="6";
(-20,-20)*+{A}="7";
(20,-20)*+{A'}="8";
{\ar^{Th}"1";"2"};
{\ar_{b}"1";"3"};
{\ar^{b'}"2";"4"};
{\ar_{h}"3";"4"};
{\ar_{Tg}"5";"6"};
{\ar^{}"5";"7"};
{\ar_{}"6";"8"};
{\ar^{g}"7";"8"};
{\ar_{Tf^*}"5";"1"};
{\ar^{f^*}"7";"3"};
{\ar^{Tf'^*}"6";"2"};
{\ar^{f'^*}"8";"4"};
{\ar@{=>}^{(T\gamma)^{-1}}(0,33);(0,27)};
{\ar@{=>}^{\gamma}(0,-27);(0,-33)};
{\ar@{=>}^{\ovll{f'^*}}(30,3);(30,-3)};
{\ar@{=>}^{\ovll{f^*}}(-30,3);(-30,-3)};
\endxy
\end{equation*}
Then it is not difficult to show that
$(h,\ovl{\gamma}):(B,b,\hat{b},\ovl{b})\rw(B',b',\hat{b'},\ovl{b'})$ is a
pseudo-$T$-algebra morphism.
The following fact is essentially known and, as sketched in the proof below, it is an instance of Theorem \ref{s2.the1}
\begin{lemma}\cite{PP}\label{lem-PP}
Let $\clC$ be a small 2-category, $F,F':\clC\rw\Cat$ be 2-functors, $\alpha:F\rw F'$
a 2-natural transformation. Suppose that, for all objects $C$ of $\clC$, the
following conditions hold:
\begin{itemize}
\item [i)] $G(C),\;G'(C)$ are objects of $\Cat$ and there are adjoint equivalences of
categories $\mu_C\vdash\eta_C$, $\mu'_C\vdash\eta'_C$,
\begin{equation*}
\mu_C:G(C)\;\rightleftarrows\;F(C):\eta_C\qquad\qquad
\mu'_C:G'(C)\;\rightleftarrows\;F'(C):\eta'_C,
\end{equation*}
\item [ii)] there are functors $\beta_C:G(C)\rw G'(C),$
\item [iii)] there is an invertible 2-cell
\begin{equation*}
\gamma_C:\beta_C\,\eta_C\Rightarrow\eta'_C\,\alpha_C.
\end{equation*}
\end{itemize}
Then
\begin{itemize}
\item [a)] There exists a pseudo-functor $G:\clC\rw\Cat$ given on objects by $G(C)$,
and pseudo-natural transformations $\eta:F\rw G$, $\mu:G\rw F$ with
$\eta(C)=\eta_C$, $\mu(C)=\mu_C$; these are part of an adjoint equivalence
$\mu\vdash\eta$ in the 2-category $\Ps[\clC,\Cat]$.
\item [b)] There is a pseudo-natural transformation $\beta:G\rw G'$ with
$\beta(C)=\beta_C$ and an invertible 2-cell in $\Ps[\clC,\Cat]$,
$\gamma:\beta\eta\Rightarrow\eta\alpha$ with $\gamma(C)=\gamma_C$.
\end{itemize}
\end{lemma}
\begin{proof}
Recall \cite{PW} that the functor 2-category $[\clC,\Cat]$ is 2-monadic over
$[ob(\clC),\Cat]$, where $ob(\clC)$ is the set of objects in $\clC$. Let
\begin{equation*}
\clU:[\clC,\Cat]\rw[ob(\clC),\Cat]
\end{equation*}
be the forgetful functor. Let $T$ be the 2-monad; then the pseudo-$T$-algebras are precisely the pseudo-functors from
$\clC$ to $\Cat$.
Then the adjoint equivalences $\mu_C\vdash\eta_C$ amount precisely to an
adjoint equivalence in $[ob(\clC),\Cat]$, $\;\mu_0\vdash\eta_0$,
$\;\mu_0:G_0\;\;\rightleftarrows\;\;\clU F:\eta_0$ where $\;G_0(C)=G(C)$ for
all $C\in ob(\clC)$. This equivalence enriches to an
adjoint equivalence $\mu\vdash\eta$ in $\Ps[\clC,\Cat]$
\begin{equation*}
\mu:G\;\rightleftarrows\; F:\eta
\end{equation*}
between $F$ and a pseudo-functor $G$; it is $\clU G=G_0$, $\;\clU\eta=\eta_0$,
$\;\clU\mu=\mu_0$; hence on objects $G$ is given by $G(C)$, and
$\eta(C)=\clU\eta(C)=\eta_C$, $\;\mu(C)=\clU\mu(C)=\mu_C$.
Let $\nu_C:\Id_{G(C)}\Rw\eta_C\mu_C$ and $\zve_C:\mu_C\eta_C\Rw\Id_{F(C)}$ be
the unit and counit of the adjunction $\mu_C\vdash\eta_C$. Given a morphism $f:C\rw D$ in $\clC$, it is
\begin{equation*}
G(f)=\eta_D F(f)\mu_C
\end{equation*}
and we have natural isomorphisms:
\begin{align*}
& \eta_f : G(f)\eta_C=\eta_D F(f)\mu_C\eta_C\overset{\eta_D F(f)\zve_C}{=\!=\!=\!=\!\Rw} \eta_D F(f)\\
& \mu_f : F(f)\mu_C\overset{\nu_{F(f)}\mu_C}{=\!=\!=\!\Rw}\mu_D\eta_D F(f)\mu_C=\mu_D
G(f).
\end{align*}
Also, the natural isomorphism
\begin{equation*}
\beta_f: G'(f)\beta_C\Rw\beta_D G(f)
\end{equation*}
is the result of the following pasting
\begin{equation*}
\xy
0;/r.15pc/:
(-40,40)*+{G(C)}="1";
(40,40)*+{G'(C)}="2";
(-40,-40)*+{G(D)}="3";
(40,-40)*+{G'(D)}="4";
(-20,20)*+{F(C)}="5";
(20,20)*+{F'(C)}="6";
(-20,-20)*+{F(D)}="7";
(20,-20)*+{F'(D)}="8";
{\ar^{\beta_C}"1";"2"};
{\ar_{G(f)}"1";"3"};
{\ar^{G'(f)}"2";"4"};
{\ar_{\beta_D}"3";"4"};
{\ar^{\alpha_C}"5";"6"};
{\ar^{F(f)}"5";"7"};
{\ar_{F'(f)}"6";"8"};
{\ar_{\alpha'_D}"7";"8"};
{\ar^{}"5";"1"};
{\ar^{}"7";"3"};
{\ar^{}"6";"2"};
{\ar^{}"8";"4"};
{\ar@{=>}^{\gamma_C}(0,33);(0,27)};
{\ar@{=>}^{\gamma_D^{-1}}(0,-27);(0,-33)};
{\ar@{=>}^{\eta_f'}(30,3);(30,-3)};
{\ar@{=>}^{\eta_f}(-30,3);(-30,-3)};
\endxy
\end{equation*}
\end{proof}
\section{Weakly globular $n$-fold categories}\label{sec-WG-nfol-cat}
In this section we review the notions of weakly globular \nfol category and of Segalic pseudo-functors introduced by the author in \cite{Pa2}. These will be used throughout the paper.
\subsection{Homotopically discrete $n$-fold categories}\label{homotopically-discrete} We first recall the category of homotopically discrete \nfol categories, introduced by the author in \cite{Pa1}. This is needed to define weakly globular $n$-fold categories.
\begin{definition}\label{def-hom-dis-ncat}
Define inductively the full subcategory $\cathd{n}\subset\cat{n}$ of homotopically discrete \nfol categories.
For $n=1$, $\cathd{1}=\cathd{}$ is the category of equivalence relations. Denote by $\p{1}=p:\Cat\rw\Set$ the isomorphism classes of objects functor.
Suppose, inductively, that for each $1\leq k\leq n-1$ we defined $\cathd{k}\subset\cat{k}$ and $k$-equivalences such that the following holds:
\begin{itemize}
\item [a)] The $k^{th}$ direction in $\cathd{k}$ is groupoidal; that is, if $X\in\cathd{k}$, $\xi_{k}X\in\Gpd(\cat{k-1})$ (where $\xi_{k}X$ is as in Proposition \ref{pro-assoc-iso}).
\item [b)] There is a functor $\p{k}:\cathd{k}\rw\cathd{k-1}$ making the following diagram commute:
\begin{equation}\label{eq1-p-fun-def}
\xymatrix{
\cathd{k} \ar^{\Nu{k-1}...\Nu{1}}[rrr] \ar_{p^{(k)}}[d] &&& \funcat{k-1}{\Cat} \ar^{\bar p}[d]\\
\cathd{k-1} \ar_{\N{k-1}}[rrr] &&& \funcat{k-1}{\Set}
}
\end{equation}
Note that this implies that $(\p{k}X)_{s_1 ... s_{k-1}}=p X_{s_1 ... s_{k-1}}$ for all $(s_1 ... s_{k-1})\in\dop{k-1}$.
\end{itemize}
$\cathd{n}$ is the full subcategory of $\funcat{}{\cathd{n-1}}$ whose objects $X$ are such that
\bigskip
\begin{itemize}
\item [(i)] $\hspace{30mm} X_s\cong\pro{X_1}{X_0}{s} \quad \mbox{for all} \; s \geq 2.$
\medskip
In particular this implies that $X\in \Cat(\Gpd(\cat{n-2})) =\Gpd(\cat{n-1})$ and the $n^{th}$ direction in $X$ is groupoidal.
\medskip
\item [(ii)] The functor
\begin{equation*}
\op{n-1}:\cathd{n}\subset \funcat{}{\cathd{n-1}}\rw\funcat{}{\cathd{n-2}}
\end{equation*}
restricts to a functor
\begin{equation*}
\p{n}:\cathd{n}\rw\cathd{n-1}
\end{equation*}
Note that this implies that $(\p{n}X)_{s_1 ... s_{n-1}}=p X_{s_1 ... s_{n-1}}$ and that the following diagram commutes
\begin{equation}\label{eq2-p-fun-def}
\xymatrix{
\cathd{n} \ar^{\Nu{n-1}...\Nu{1}}[rrr] \ar_{p^{(n)}}[d] &&& \funcat{n-1}{\Cat} \ar^{\bar p}[d]\\
\cathd{n-1} \ar_{\N{n-1}}[rrr] &&& \funcat{n-1}{\Set}
}
\end{equation}
\end{itemize}
\end{definition}
\mk
\begin{definition}\label{def-hom-dis-ncat-0}
Denote by $\zgu{n}_X:X\rw \di{n}\p{n}X$ the morphism given by
\begin{equation*}
(\zgu{n}_X)_{s_1...s_{n-1}} :X_{s_1...s_{n-1}} \rw d p X_{s_1...s_{n-1}}
\end{equation*}
for all $(s_1,...,s_{n-1})\in \dop{n-1}$. Denote by
\begin{equation*}
X^d =\di{n}\di{n-1}...\di{1}\p{1}\p{2}...\p{n}X
\end{equation*}
and by $\zg\lo{n}$ the composite
\begin{equation*}
X\xrw{\zgu{n}}\di{n}\p{n}X \xrw{\di{n}\zgu{n-1}} \di{n}\di{n-1}\p{n-1}\p{n}X \rw \cdots \rw X^d\;.
\end{equation*}
For each $a,b\in X_0^d$ denote by $X(a,b)$ the fiber at $(a,b)$ of the map
\begin{equation*}
X_1 \xrw{(d_0,d_1)} X_0\times X_0 \xrw{\zg\lo{n}\times\zg\lo{n}} X_0^d\times X_0^d\;.
\end{equation*}
\end{definition}
\begin{definition}\label{def-hom-dis-ncat-1}
Define inductively $n$-equivalences in $\cathd{n}$. For $n=1$, a 1-equivalence is an equivalence of categories. Suppose we defined $\nm$-equivalences in $\cathd{n-1}$. Then a map $f:X\rw Y$ in $\cathd{n}$ is an $n$-equivalence if, for all $a,b \in X_0^d$, $f(a,b):X(a,b) \rw Y(fa,fb)$ and $\p{n}f$ are $\nm$-equivalences.
\end{definition}
\subsection{Weakly globular $n$-fold categories}\label{weakly-globular} We recall the category of weakly globular \nfol categories and some of its main properties needed in this paper.
\begin{definition}\label{def-n-equiv}
For $n=1$, $\catwg{1}=\Cat$ and $1$-equivalences are equivalences of categories.
Suppose, inductively, that we defined $\catwg{n-1}$ and $(n-1)$-equivalences. Then $\catwg{n}$ is the full subcategory of $\funcat{}{\catwg{n-1}}$ whose objects $X$ are such that
\begin{itemize}
\item [a)] \textsl{Weak globularity condition} $X_0\in\cathd{n-1}$.\mk
\item [b)] \textsl{Segal condition} For all $k\geq 2$ the Segal maps are isomorphisms:
\begin{equation*}
X_k\cong\pro{X_1}{X_0}{k}\;.
\end{equation*}
\item [c)] \textsl{Induced Segal condition} For all $k\geq 2$ the induced Segal maps
\begin{equation*}
X_k\rw\pro{X_1}{X^d_0}{k}
\end{equation*}
(induced by the map $\zg:X_0\rw X_0^d$) are $(n-1)$-equivalences.\mk
\item [d)] \textsl{Truncation functor} There is a functor $\p{n}:\catwg{n}\rw\catwg{n-1}$ making the following diagram commute
\begin{equation*}
\xymatrix{
\catwg{n} \ar^{J_n}[rr] \ar_{\p{n}}[d] && \funcat{n-1}{\Cat} \ar^{\ovl p}[d]\\
\catwg{n-1} \ar^{\Nb{n-1}}[rr] && \funcat{n-1}{\Set}
}
\end{equation*}
\end{itemize}
Given $a,b\in X_0^d$, denote by $X(a,b)$ the fiber at $(a,b)$ of the map
\begin{equation*}
X_1\xrw{(\pt_0,\pt_1)} X_0\times X_0 \xrw{\zg\times \zg} X^d_0\times X^d_0\;.
\end{equation*}
We say that a map $f:X\rw Y$ in $\catwg{n}$ is an $n$-equivalence if
\begin{itemize}
\item [i)] For all $a,b\in X_0^d$
\begin{equation*}
f(a,b): X(a,b) \rw Y(fa,fb)
\end{equation*}
is an $(n-1)$-equivalence.\mk
\item [ii)] $\p{n}f$ is an $(n-1)$-equivalence.
\end{itemize}
This completes the inductive step in the definition of $\catwg{n}$.
\end{definition}
\begin{remark}\label{rem-n-equiv}
It follows by Definition \ref{def-n-equiv}, Definition \ref{def-hom-dis-ncat} and \cite[Proposition 3.4]{Pa2} that $\cathd{n}\subset \catwg{n}$.
\end{remark}
\begin{definition}\label{def-kdir-wg-ncat}
Given $X\in\cat{n}$ and $k\geq 0$, let $\Nu{2}X\in\funcat{}{\cat{n-1}}$ as in Definition \ref{def-ner-func-dirk} so that for each $k\geq 0$, $(\Nu{2}X)_k=X_k\up{2}\in \funcat{}{\cat{n-2}}$ is given by
\begin{equation*}
(X_k\up{2})_s =
\left\{
\begin{array}{ll}
X_{0k}, & \hbox{$s=0$;} \\
X_{1k}, & \hbox{$s=1$;} \\
X_{sk}=\pro{X_{1k}}{X_{0k}}{s}, & \hbox{$s\geq 2$.}
\end{array}
\right.
\end{equation*}
\end{definition}
The following property of weakly globular \nfol categories will be used in Section \ref{sec-cat-lta}.
\begin{proposition}\label{pro-property-WG-nfol}
The functor $\Nu{2}:\cat{n}\rw \funcat{}{\cat{n-1}}$ restricts to a functor
\begin{equation*}
\Nu{2}:\catwg{n} \rw \funcat{}{\catwg{n-1}}\;.
\end{equation*}
\end{proposition}
\begin{proof}
See \cite[Proposition 3.16 a)]{Pa2}
\end{proof}
\begin{definition}\label{def-pn}
For each $1\leq j \leq n$ denote
\begin{align*}
\p{j,n} & = \p{j}\p{j-1}\cdots \p{n}:\catwg{n}\rw \catwg{j-1}\\
\p{n,n}& = \p{n}\;.
\end{align*}
\end{definition}
\begin{lemma}\label{lem-prop-pn}\cite [Lemma 3.8]{Pa2}
For each $X\in\catwg{n}$, $1\leq j < n$ and $s\geq 2$ it is
\begin{equation}\label{eq-lem-prop-pn}
\begin{split}
&\p{j,n-1}X_s \cong \p{j,n-1}(\pro{X_1}{X_0}{s})=\\
& =\pro{\p{j,n-1} X_1}{\p{j,n-1} X_0}{s}\;.
\end{split}
\end{equation}
\end{lemma}
\begin{remark}\label{rem-eq-def-wg-ncat}
It follows immediately from Lemma \ref{lem-prop-pn} that if $X\in\catwg{n}$, for all $s\geq 2$
\begin{equation}\label{eq1-rem-eq-def-wg-ncat}
X_{s0}^d=(\pro{X_{10}}{X_{00}}{s})^d\cong\pro{X^d_{10}}{X^d_{00}}{s}\;.
\end{equation}
In fact, by \eqref{eq-lem-prop-pn} in the case $j=2$, taking the 0-component, we obtain
\begin{equation*}
\begin{split}
& \p{1,n-2}(\pro{X_{10}}{X_{00}}{s})\cong \\
=\ & \pro{\p{1,n-2}X_{10}}{\p{1,n-2}X_{00}}{s}
\end{split}
\end{equation*}
which is the same as \eqref{eq1-rem-eq-def-wg-ncat}.
\end{remark}
\subsection{Segalic pseudo-functors}\label{segalic-pseudo} We now recall the notion of Segalic pseudo-functor from \cite{Pa2}.
Let $H\in\Ps\funcat{n}{\Cat}$ be such that $H_{\uk(0,i)}$ is discrete for all $\uk\in\Dmenop$ and all $i\geq 0$. Then the following diagram commutes, for each $k_i\geq 2$.
\begin{equation*}
\xy
0;/r.8pc/:
(0,0)*+{H_{\uk}}="1";
(-7,-5)*+{H_{\uk(1,i)}}="2";
(-2,-5)*+{H_{\uk(1,i)}}="3";
(9,-5)*+{H_{\uk(1,i)}}="4";
(-10,-9)*+{H_{\uk(0,i)}}="5";
(-5,-9)*+{H_{\uk(0,i)}}="6";
(0,-9)*+{H_{\uk(0,i)}}="7";
(6,-9)*+{H_{\uk(0,i)}}="8";
(12,-9)*+{H_{\uk(0,i)}}="9";
(3,-5)*+{\cdots}="10";
(3,-9)*+{\cdots}="11";
{\ar_{\nu_1}"1";"2"};
{\ar^{\nu_2}"1";"3"};
{\ar^{\nu_k}"1";"4"};
{\ar_{d_1}"2";"5"};
{\ar^{d_0}"2";"6"};
{\ar^{d_1}"3";"6"};
{\ar^{d_0}"3";"7"};
{\ar_{d_1}"4";"8"};
{\ar^{d_0}"4";"9"};
\endxy
\end{equation*}
There is therefore a unique Segal map
\begin{equation*}
H_{\uk}\rw \pro{H_{\uk(1,i)}}{H_{\uk(0,i)}}{k_i}\;.
\end{equation*}
\begin{definition}\label{def-seg-ps-fun}
We\; define\; the \;subcategory \; $\segpsc{n}{\Cat}$\; of $\psc{n}{\Cat}$ as follows:
For $n=1$, $H\in \segpsc{}{\Cat}$ if $H_0$ is discrete and the Segal maps are isomorphisms: that is, for all $k\geq 2$
\begin{equation*}
H_k\cong\pro{H_1}{H_0}{k}
\end{equation*}
Note that, since $p$ commutes with pullbacks over discrete objects, there is a functor
\begin{equation*}
\begin{split}
& \p{2}:\segpsc{}{\Cat} \rw \Cat\;, \\
& (\p{2}X)_{k}=p X_k\;.
\end{split}
\end{equation*}
That is the following diagram commutes:
\begin{equation*}
\xymatrix{
\segpsc{}{\Cat} \ar@{^{(}->}[rr]\ar_{\p{2}}[d] && \psc{}{\Cat} \ar^{\ovl{p}}[d]\\
\Cat \ar[rr] && \funcat{}{\Set}
}
\end{equation*}
When $n>1$, $\segpsc{n}{\Cat}$ is the full subcategory of $\psc{n}{\Cat}$ whose objects $H$ satisfy the following: \mk
\begin{itemize}
\item [a)] \emph{Discreteness condition}: $H_{\uk(0,i)}$ is discrete for all $\uk\in\Dmenop$ and $1 \leq i \leq n$.\mk
\item [b)] \emph{Segal condition}: All Segal maps are isomorphisms
\begin{equation*}
H_{\uk}\cong\pro{H_{\uk(1,i)}}{H_{\uk(0,i)}}{k_i}
\end{equation*}
for all $\uk\in\Dmenop$, $1 \leq i \leq n$ and $k_i\geq2$.\mk
\item [c)] \emph{Truncation functor}: There is a functor
\begin{equation*}
\p{n+1}:\segpsc{n}{\Cat}\rw \catwg{n}
\end{equation*}
\end{itemize}
making the following diagram commute:
\begin{equation*}
\xymatrix{
\segpsc{n}{\Cat} \ar@{^(->}[r] \ar_{\p{n+1}}[d] & \psc{n}{\Cat} \ar^{\ovl{p}}[d]\\
\catwg{n} \ar_{\Nb{n}}[r] & \funcat{n}{\Set}
}
\end{equation*}
\end{definition}
\bigskip
The main property of Segalic pseudo-functors is the theorem below stating that that the classical strictification of pseudo-functors, when applied to a Segalic pseudo-functors, yields weakly globular \nfol categories. This result is used crucially in Section \ref{sec-cat-lta} to build the rigidification functor $Q_n$.
\begin{theorem}\cite[Theorem 4.5]{Pa2}\label{the-strict-funct}
The strictification functor
\begin{equation*}
\St : \psc{n-1}{\Cat}\rw \funcat{n-1}{\Cat}
\end{equation*}
restricts to a functor
\begin{equation*}
L_n: \segpsc{n-1}{\Cat}\rw J_n\catwg{n}
\end{equation*}
where $J_n\catwg{n}$ denotes the image of the fully faithful functor $J_n:\catwg{n}\rw\funcat{n-1}{\Cat}$\;.
Further, for each $H\in \segpsc{n-1}{\Cat}$ and $\uk\in\dop{n-1}$, the map $(L_n H)_{\uk}\rw H_{\uk}$ is an equivalence of categories.
\end{theorem}
\section{Weakly globular Tamsamani n-categories}\label{sec-WG-Tam-cat}
In this section we introduce the category $\tawg{n}$ of weakly globular Tamsamani $n$-categories and discuss their properties. The definition is inductive on dimension starting with $\Cat$ when $n=1$.
In dimension $n>1$, a weakly globular Tamsamani $n$-category $X$ is a simplicial object in the category $\tawg{n-1}$ satisfying additional conditions. These conditions encode the weakness in the structure in two ways: one is the weak globularity condition, requiring $X_{0}$ to be a homotopically discrete $\nm$-fold category. The second is the induced Segal maps condition, requiring that the induced Segal maps for all $k\geq 2$
\begin{equation}\label{eq1-sec-prelim}
\hmu{k}:X_k \rw \pro{X_1}{X_0^d}{k}
\end{equation}
are $\nm$-equivalences.
By unravelling the inductive Definition \ref{def-wg-ps-cat} we obtain the embedding
\begin{equation*}
J_n : \tawg{n}\rw \funcat{n-1}{\Cat}\;.
\end{equation*}
The inductive Definition \ref{def-wg-ps-cat} also requires the existence of a truncation functor
\begin{equation*}
\p{n}:\tawg{n}\rw \tawg{n-1}
\end{equation*}
obtained by applying levelwise to $J_n X$ the isomorphism classes of object functor $p:\Cat\rw\Set$.
This truncation functor is used to define $\nequ$s in $\tawg{n}$. This notion is a higher dimensional generalization of a functor which is fully faithful and essentially surjective on objects. A useful characterization of $n$-equivalences is given in Proposition \ref{pro-n-equiv}.
Under certain conditions, an $\nequ$ $f$ is such that $J_n f$ is a levelwise equivalence of categories (see Proposition \ref{pro-crit-lev-nequiv}) and this will be used in Section \ref{sec-func-qn}.
\begin{definition}\rm\label{def-wg-ps-cat}
We define the category $\tawg{n}$ by induction on $n$. For $n=1$, $\tawg{1}=\Cat$ and 1-equivalences are equivalences of categories. We denote by $\p{1}=p:\Cat\rw \Set$ the isomorphism classes of object functor.
Suppose, inductively, that we defined for each $1 < k\leq n-1$
\begin{equation*}
\xymatrix{\tawg{k}\;\ar@{^{(}->}^(0.4){}[r] & \;\funcat{k-1}{\Cat}}
\end{equation*}
and $k$-equivalences in $\tawg{k}$ as well as a functor
\begin{equation*}
\p{k}:\tawg{k}\rw \tawg{k-1}
\end{equation*}
sending $k$-equivalences to $(k-1)$-equivalences and making the following diagram commute:
\begin{equation}\label{eq-wg-ps-cat}
\xymatrix@C=30pt{
\tawg{k} \ar^{J_{k}}[rr]\ar_{\p{k}}[d] && \funcat{k-1}{\Cat} \ar^{\ovl{p}}[d]\\
\tawg{k-1} \ar^{\Nb{k-1}}[rr] && \funcat{k-1}{\Set}
}
\end{equation}
Define $\tawg{n}$ to be the full subcategory of $\funcat{}{\tawg{n-1}}$ whose objects $X$ are such that
\begin{itemize}
\item [a)] \textsl{Weak globularity condition} \;$X_0\in\cathd{n-1}$.\bk
\item [b)] \textsl{Induced Segal maps condition}. For all $s\geq 2$ the induced Segal maps
\begin{equation*}
X_s \rw \pro{X_1}{X^d_0}{s}
\end{equation*}
(induced by the map $\zg:X_0\rw X_0^d$) are $(n-1)$-equivalences.
\end{itemize}
\medskip
To complete the inductive step, we need to define $\p{n}$ and $n$-equivalences.
Note that the functor
\begin{equation*}
\ovl{\p{n-1}}:\funcat{}{\tawg{n-1}}\rw \funcat{}{\tawg{n-2}}
\end{equation*}
restricts to a functor
\begin{equation*}
\p{n}:\tawg{n}\rw \tawg{n-1}\;.
\end{equation*}
In fact, by \eqref{eq-wg-ps-cat} $\p{n-1}$ preserves pullbacks over discrete objects so that
\begin{equation*}
\p{n-1}(\pro{X_1}{X_0^d}{s}) \cong\pro{\p{n-1}X_1}{(\p{n-1}X_0^d)}{s}\; .
\end{equation*}
Further, $\p{n-1}X_0^d = (\p{n-1}X_0)^d$ and $\p{n-1}$ sends $(n-1)$-equivalences to $(n-2)$-equivalences.
Therefore, the induced Segal maps for $s\geq 2$
\begin{equation*}
X_s \rw\pro{X_1}{X_0^d}{s}
\end{equation*}
being $(n-1)$-equivalences, give rise to $(n-2)$-equivalences
\begin{equation*}
\p{n-1}X_s \rw \pro{\p{n-1}X_1}{(\p{n-1}X_0)^d}{s}\; .
\end{equation*}
This shows that $\p{n}X \in \tawg{n-1}$. It is immediate that \eqref{eq-wg-ps-cat} holds at step $n$.
Given $a,b\in X_0^d$, denote by $X(a,b)$ the fiber at $(a,b)$ of the map
\begin{equation*}
X_1\xrw{(\pt_0,\pt_1)} X_0\times X_0 \xrw{\zg\times \zg} X^d_0\times X^d_0\;.
\end{equation*}
We say that a map $f:X\rw Y$ in $\tawg{n}$ is an $n$-equivalence if
\begin{itemize}
\item [i)] For all $a,b\in X_0^d$
\begin{equation*}
f(a,b): X(a,b) \rw Y(fa,fb)
\end{equation*}
is an $(n-1)$-equivalence.\mk
\item [ii)] $\p{n}f$ is an $(n-1)$-equivalence.
\end{itemize}
This completes the inductive step in the definition of $\tawg{n}$.
\end{definition}
\begin{remark}\label{rem-wg-ps-cat}
It follows by Definition \ref{def-n-equiv} that $\catwg{n}\subset\tawg{n}$.
\end{remark}
\begin{definition}\label{def-x-tawg-disc}
An object $X\in\tawg{n}$ is called discrete if $\Nb{n-1}X$ is a constant functor.
\end{definition}
\begin{example}\label{ex-tam}
Tamsamani $n$-categories.
\mk
A special case of weakly globular Tamsamani $n$-category occurs when $X\in\tawg{n}$ is such that $X_0$ and $X_{\oset{r}{1...1}0}$ are discrete for all $1 \leq r \leq n-2$. The resulting category is the category $\Tan$ of Tamsamani's $n$-categories. Note that, if $X\in\Tan$ then $X_s \in \ta{n-1}$ for all $n$, the induced Segal maps $\hmu{s}$ coincide with the Segal maps
\begin{equation*}
\nu_s:X_s\rw \pro{X_1}{X_0}{s}
\end{equation*}
and $\p{n}X\in\ta{n-1}$. Hence this recovers the original definition of Tamsamani's weak $n$-category \cite{Ta}.
\end{example}
\begin{example}\label{ex-wg-2-3}
Weakly globular Tamsamani 2-categories.
\mk
\nid From the definition, $X\in \tawg{2}$ consists of a simplicial object $X\in\funcat{}{\Cat}$ such that $X_0\in\cathd{}$ and the induced Segal maps
\begin{equation*}
\hmuk:X_k \xrw{\muk}\pro{X_1}{X_0}{k}\xrw{\nu_k}\pro{X_1}{X^d_0}{k}
\end{equation*}
are equivalences of categories.
The existence of the functor $\p{2}:\tawg{2}\rw \Cat$ making diagram \eqref{eq-wg-ps-cat} commute is in this case automatic so this condition does need to be imposed as part of the definition. In fact, since $p:\Cat\rw \Set$ sends equivalences to isomorphisms and preserves pullbacks over discrete objects, we obtain isomorphisms for each $k\geq 2$
\begin{flalign*}
p\hmuk: p X_k & \cong p(\pro{X_1}{X_0^d}{k})\cong &\\
& \cong \pro{p X_1}{p X_{0}^{d}}{k}\cong \pro{p X_1}{(p X_{0})^{d}}{k}\;.&
\end{flalign*}
Since $p X_k =(\p{2}X)_k$ for all $k$, the Segal maps for the simplicial set $\bar p X$ are isomorphisms. Therefore $\bar p X$ is the nerve of a category. We therefore have a functor $\p{2}$ given as composite
\begin{equation*}
\p{2}:\tawg{2}\xrw{\bar p} \mbox{\sf{Ner}} \Cat \subset \funcat{}{\Set}\xrw{P}\Cat
\end{equation*}
where $\mbox{\sf{Ner}} \Cat$ is the full subcategory of $\funcat{}{\Set}$ consisting of nerves of categories and $P$ is the left adjoint to the nerve functor $N:\Cat\rw\funcat{}{\Set}$.
\end{example}
\begin{notation}\label{not-ex-tam}
For each $1 \leq j\leq n-1$ denote\mk
$\begin{array}{ll}
\p{j,n}=\p{j}\p{j+1}...\p{n}:\tawg{n}\rw \tawg{j-1},\\
\p{n,n}=\p{n}.
\end{array}$
\end{notation}
In the following lemma we give a criterion for a weakly globular Tamsamani $n$-category to be in $\catwg{n}$.
\begin{lemma}\label{lem-x-in-tawg-x-in-catwg}
Let $X\in\tawg{n}$ be such that
\begin{itemize}
\item [a)] $X_s\in \catwg{n-1}$ for all $s$.\mk
\item [b)] $X_s \cong \pro{X_1}{X_0}{s}$ for all $s\geq 2$.\mk
\item [c)] For all $s\geq 2$ and $1\leq j\leq n-1$
\begin{equation}\label{eq-lem-eq-def-wg}
\begin{split}
& \p{j,n-1}X_s \cong \p{j,n-1}(\pro{X_1}{X_0}{s})= \\
= \ &\pro{\p{j,n-1} X_1}{\p{j,n-1} X_0}{s}
\end{split}
\end{equation}
\end{itemize}
then $X\in\catwg{n}$.
\end{lemma}
\begin{proof}
By induction on $n$. When $n=2$, let $X\in\tawg{2}$ satisfy a), b), c). Then $X_0\in\cathd{}$ and by b), $X\in\cat{2}$. Since $X\in\tawg{2}$ the induced Segal maps
\begin{equation*}
X_s \rw \pro{X_1}{X_0^d}{s}
\end{equation*}
are equivalences of categories for all $s\geq 2$. Thus, by definition, $X\in\catwg{2}$.
Suppose, inductively, that the lemma holds for $(n-1)$ and let $X\in\tawg{n}$ satisfy a), b), c). Then $X_0\in\cathd{n-1}$, $X_k\in\catwg{n-1}$ for all $k\geq 0$ and, since $X\in\tawg{n}$, the induced Segal maps
\begin{equation*}
X_s \rw \pro{X_1}{X_0^d}{s}
\end{equation*}
are $(n-1)$-equivalences for all $s\geq 2$.
By hypothesis b) and the definition of $\catwg{n}$, to show that $X\in\catwg{n}$ it is enough to prove that $\p{n}X\in\catwg{n-1}$. We do so by proving that $\p{n-1}X$ satisfies the inductive hypothesis.
For all $s\geq 0$, $(\p{n}X)_s=\p{n-1}X_s\in\catwg{n-2}$ since, by hypothesis a), $X_s\in\catwg{n-1}$. Thus $\p{n}X$ satisfies the inductive hypothesis a). Also, from hypothesis c) for $X$ in the case $j=n-1$ for all $s\geq 2$,
\begin{equation*}
\begin{split}
& \p{n-1}X_s \cong\p{n-1}(\pro{X_1}{X_0}{s})\cong \\
\cong \ & \pro{\p{n-1}X_1}{\p{n-1}X_0}{s}
\end{split}
\end{equation*}
that is, $\p{n}X$ satisfies inductive hypothesis b). From this and from hypothesis c) for $X$ we deduce, for all $s\geq 2$ and $1\leq j\leq n-2$
\begin{align*}
& \p{j,n-2}(\pro{(\p{n}X)_1}{(\p{n}X)_0}{s})=\\
= \ & \p{j,n-2}(\pro{\p{n-1}X_1}{\p{n-1}X_0}{s})=\\
= \ & \p{j,n-2}\p{n-1}(\pro{X_1}{X_0}{s})=\\
= \ & \pro{\p{j,n-1}X_1}{\p{j,n-1}X_0}{s}=\\
= \ & \pro{\p{j,n-2}(\p{n}X)_1}{\p{j,n-2}(\p{n}X)_0}{s}\;.
\end{align*}
This shows that $\p{n}X$ satisfies inductive hypothesis c). Hence we conclude that $\p{n}X\in\catwg{n-1}$, as required.
\end{proof}
\begin{lemma}\label{lem-flevel-fneq}
Let $f:X\rw Y$ in $\tawg{n}$ be a levelwise $(n-1)$-equivalence in $\tawg{n-1}$. Then $f$ is an $n$-equivalence.
\end{lemma}
\begin{proof}
By induction on $n$. Let $n=2$. If $f_0$ is an equivalence of categories, $X_0^d\cong Y_0^d$. Hence
\begin{align}\label{eq1-lem-flevel-fneq}
& Y_1=\uset{a',b'\in Y_0^d}{\coprod} Y(a',b') \cong \uset{fa,fb\in Y_0^d}{\coprod} Y(fa,fb)\;.
\end{align}
\begin{align}\label{eq2-lem-flevel-fneq}
& X_1=\uset{a,b\in X_0^d}{\coprod}X(a,b)\;.
\end{align}
Since $f_1$ is an equivalence of categories it follows from \eqref{eq1-lem-flevel-fneq} and \eqref{eq2-lem-flevel-fneq} that $f(a,b)$ is an equivalence of categories. Further, $f_k$ is an equivalence of categories for all $k\geq 0$ so that $p f_k=(\p{2}f)_k$ is an isomorphism, hence $\p{2}f$ is an isomorphism; we conclude that $f$ is a 2-equivalence.
Suppose the lemma holds for $(n-1)$ and let $f$ be as in the hypothesis. Since $f_0$ is a $(n-1)$-equivalence in $\cathd{n-1}$, $X_0^d \cong Y_0^d$, so that \eqref{eq1-lem-flevel-fneq} holds. Since $f_1$ is a $(n-1)$-equivalence it follows from \eqref{eq1-lem-flevel-fneq} and \eqref{eq2-lem-flevel-fneq} that $f(a,b)$ is a $(n-1)$-equivalence for all $a,b\in X_0^d$.
Since $f_k$ is an $(n-1)$-equivalence for all $k\geq 0$, $\p{n-1}f_k=(\p{n}f)_k$ is a $(n-2)$-equivalence. So $\p{n}f$ satisfies the induction hypothesis and is therefore a $(n-1)$-equivalence. In conclusion, $f$ is an $n$-equivalence.
\end{proof}
\begin{remark}\label{rem-local-equiv}
Applying inductively Lemma \ref{lem-flevel-fneq} it follows immediately that if a morphism $f$ in $\tawg{n}$ is such that $J_n f$ is a levelwise equivalence of categories, then $f$ is an $n$-equivalence.
\end{remark}
\begin{definition}\label{def-local-equiv}
A morphism $f:X\rw Y$ in $\tawg{n}$ is said to be a local $\nm$-equivalence if for all $a,b \in X_0^d$, $f(a,b):X(a,b)\rw Y(fa,fb)$ is a $\nm$-equivalence in $\tawg{n-1}$.
\end{definition}
\nid In the following proposition, we give a useful description of $n$-equiva-lences in $\tawg{n}$.
\begin{proposition}\label{pro-n-equiv}\
\sk
\begin{itemize}
\item [a)] Let $f$ be a morphism in $\tawg{n}$ which is an $\nequ$. Then $f$ is a local $(n-1)$-equivalence and $\p{1,n} f$ is an isomorphism.
\item [b)] Let $f$ be a morphism in $\tawg{n}$ which is a local $(n-1)$-equivalence and is such that $\p{1,n} f$ is surjective. Then $f$ is an $\nequ$.
\item [c)] Let $X\xrw{g} Z \xrw{h} Y$ be morphisms in $\tawg{n}$, $f= hg$ and suppose that $f$ and $h$ are $n$-equivalences. Then $g$ is an $n$-equivalence.
\item [d)] Let $X\xrw{g} Z \xrw{h} Y$ be morphisms in $\tawg{n}$, $f= hg$ and suppose that $g$ and $h$ are $n$-equivalences. Then $f$ is an $n$-equivalence.
\item [e)] Let $X\xrw{g} Z \xrw{h} Y$ be morphisms in $\tawg{n}$, $f= hg$ and let $g_0^d: X_0^d \rw Z_0^d$ be surjective; suppose that $f$ and $g$ are $n$-equivalences. Then $h$ is an $n$-equivalence.
\end{itemize}
\end{proposition}
\begin{proof}
By induction on $n$. It is clear for $n=1$. Suppose it is true for $n-1$.
a) Let $f:X\rw Y$ be an $\nequ$ in $\tawg{n}$. Then, by definition, $f$ is a local $(n-1)$-equivalence and $\p{n}f$ is a $\equ{n-1}$. Therefore, by induction hypothesis applied to $\p{n}f$, $\p{1,n} f$ is an isomorphism.
\mk
b) Suppose that $f:X\rw Y$ is a local $(n-1)$-equivalence in $\tawg{n}$ and $\p{1,n}f$ is surjective. To show that $f$ is a $\nequ$ we need to show that $\p{n}f$ is a $\equ{n-1}$. For each $a,b\in X_0^d$
\begin{equation*}
(\p{n}f)(a,b)=\p{n-1}f(a,b)
\end{equation*}
Since $f(a,b)$ is a $(n-1)$-equivalence, $\p{n-1}f(a,b)$ is a $(n-2)$-equivalence; that is, $\p{n}f$ is a local $(n-2)$-equivalence.
Since $\p{1,n} f = \p{1,n-1}\p{n} f$ is surjective, by inductive hypothesis applied to $\p{n}f$ we conclude that $\p{n}f$ is a $\equ{n-1}$ as required.
\mk
c) For all $a,b\in X_0^d$,
\begin{equation}\label{eq1-pro-n-equiv}
f(a,b)=h(ga,gb)g(a,b)
\end{equation}
with $f(a,b)$ and $h(ga,gb)$ \equ{n-1}s. By inductive hypothesis, $g(a,b)$ is therefore a $\equ{n-1}$.
By hypothesis and by part a), $\p{1,n} f$ and $\p{1,n} h$ are isomorphisms. Since
\begin{equation}\label{eq2-pro-n-equiv}
\p{1,n} f=(\p{1,n} h)(\p{1,n} g)
\end{equation}
it follows that $\p{1,n} g$ is an isomorphism, hence in particular it is surjective. By part b), this implies that $g$ is a $\equ{n-1}$.
\mk
d) Suppose that $h$ and $g$ are $\nequ$s. By \eqref{eq1-pro-n-equiv}, $f$ is a local $(n-1)$-equivalence and by \eqref{eq2-pro-n-equiv} $\p{1,n} f$ is an isomorphism. By b), $f$ is thus a $\nequ$.
\mk
e) By hypothesis, for each $a',b'\in Z_0^d$, $a'=ga$, $b'=gb$ for $a,b\in X_0^d$. It follows that $h(a',b')=h(ga,gb)$. Since, by induction hypothesis and by \eqref{eq1-pro-n-equiv}, $h(ga,gb)$ is a $\nm$-equivalence, it follows that such is $h(a',b')$. That is, $h$ is a local equivalence.
By hypothesis and by part a), $\p{1,n} f$ and $\p{1,n} g$ are isomorphisms, so by \eqref{eq2-pro-n-equiv}, such is $\p{1,n} h$. We conclude by part b) that $h$ is a $\nequ$.
\end{proof}
\begin{lemma}\label{lem-crit-lev-nequiv}
Consider the diagram in $\tawg{n}$
\begin{equation*}
\xymatrix{
X \ar^{f}[r]\ar_{\za}[d] & Z \ar@{=}[d] & Y \ar_{g}[l]\ar^{\zb}[d]\\
X' \ar_{f'}[r] & Z & Y' \ar^{g'}[l]
}
\end{equation*}
with $Z$ discrete. Then:
\begin{itemize}
\item [a)] $X\tiund{Z} Y\,,\,X'\tiund{Z'} Y'\in\tawg{n}$, where the pullback are taken in $\funcat{n-1}{\Cat}$.\mk
\item [b)] $\p{n}(X\tiund{Z}Y)\cong \p{n}X\tiund{\p{n}Z}\p{n}Y$.\mk
\item [c)] If $\za,\,\zb$ are $n$-equivalences such is
\begin{equation*}
(\za,\zb):X\tiund{Z} Y\rw X'\tiund{Z'} Y'\;.
\end{equation*}
\end{itemize}
\end{lemma}
\begin{proof}
By induction on $n$. It is clear for $n=1$ since the maps $f,g,f',g'$ are isofibrations as their target is discrete. Suppose, inductively, that the lemma holds for $(n-1)$.
\mk
a) Since pullbacks in $\funcat{n-1}{\Cat}$ are computed pointwise, for each $k\geq 0$
\begin{equation*}
(X\tiund{Z}Y)_k = X_k \tiund{Z} Y_k \in \funcat{n-2}{\Cat}
\end{equation*}
with $X_k,Y_k\in\tawg{n-1}$. It follows from inductive hypothesis a) that $(X\tiund{Z}Y)_k\in\tawg{n-1}$. Also, $(X\tiund{Z}Y)_0 = X_0 \tiund{Z} Y_0 \in\cathd{n-1}$ since $X_0,Y_0\in\cathd{n-1}$ and $Z$ is discrete, (see \cite[Lemma 3.10 c)]{Pa1}).
To show that $X\tiund{Z}Y\in\tawg{n}$ it remains to prove that the induced Segal maps $\hmu{k}$ for $X\tiund{Z}Y$ are $\nm$-equivalences. We prove this for $k=2$, the case $k>2$ being similar. Note that, by \cite[Lemma 3.10 c)]{Pa1}.
\begin{equation}\label{eq1-crit-lev-nequiv}
\tens{(X\tiund{Z}Y)_1}{(X\tiund{Z}Y)_0^d} \cong (\tens{X_1}{X_0^d})\tiund{Z}(\tens{Y_1}{Y_0^d})\;.
\end{equation}
Consider the commutative diagram in $\tawg{n-1}$
\begin{equation}\label{eq2-crit-lev-nequiv}
\xymatrix{
X_2 \ar^{}[r]\ar_{\hmu{2}(X)}[d] & Z \ar@{=}[d] & Y_2 \ar^{}[l] \ar^{\hmu{2}(Y)}[d]\\
\tens{X_1}{X_0^d} \ar^{}[r] & Z & \tens{Y_1}{Y_0^d} \ar^{}[l]
}
\end{equation}
The vertical maps are the induced Segal maps for $X$ and $Y$, hence they are $\nm$-equivalences. By inductive hypothesis b) applied to \eqref{eq2-crit-lev-nequiv} we conclude that the induced map of pullbacks is a $\nm$-equivalence. By \eqref{eq1-crit-lev-nequiv} the latter is the induced Segal map $\hmu{2}$ for $X\tiund{Y}Z$. The proof for $k>2$ is similar and we conclude that $X\tiund{Y}Z\in\tawg{n}$.
\mk
b) By \eqref{eq-wg-ps-cat}, for all $\uk\in\dop{n-2}$
\begin{equation}\label{eq1-lem-crit-lev-nequiv}
(\p{n}(X\tiund{Z}Y))_{\uk} = p (X\tiund{Z}Y)_{\uk}\;.
\end{equation}
Since pullbacks in $\funcat{n-1}{\Cat}$ are computed pointwise and $p$ commutes with pullbacks over discrete objects, we have
\begin{equation}\label{eq2-lem-crit-lev-nequiv}
\begin{split}
& p (X\tiund{Z}Y)_{\uk} = p (X_{\uk}\tiund{Z_{\uk}}Y_{\uk})_{\uk} = pX_{\uk}\tiund{pZ_{\uk}}pY_{\uk}=\\
= & (\p{n} X)_{\uk}\tiund{(\p{n}Z)_{\uk}} (\p{n} Y)_{\uk}\;.
\end{split}
\end{equation}
Since this holds for all $\uk$, \eqref{eq1-lem-crit-lev-nequiv} and \eqref{eq2-lem-crit-lev-nequiv} imply b).
\mk
c) For each $(a,b),\, (c,d)\in (X\tiund{Z}Y)^d_0 = X_0^d\tiund{Z}Y_0^d$ we have
\begin{equation*}
\begin{split}
& (X\tiund{Z}Y)((a,b),(c,d))=X(a,c)\times Y(b,d) \\
& (X'\tiund{Z}Y')((fa,fb),(gc,gd))=X'(fa,gc)\times Y'(fb,gd)\;.
\end{split}
\end{equation*}
Since $\za,\zb$ are $n$-equivalences, $\za(a,c)$ and $\zb(b,d)$ are $\nm$-equivalences, hence such is
\begin{equation*}
(\za,\zb)((a,b),(c,d))=\za(a,c)\times \zb(b,d)\;.
\end{equation*}
Since $\p{n}$ commutes with pullbacks over discrete objects for each $n$, so does $\p{1,n}$, hence
\begin{equation*}
\p{1,n}(X\tiund{Z}Y) = \p{1,n}X\tiund{Z}\p{1,n}Y\;.
\end{equation*}
From the hypothesis and from Proposition \ref{pro-n-equiv} a), $\p{1,n}\za$ and $\p{1,n}\zb$ are isomorphisms, thus so is $\p{1,n}(\za,\zb)$. Since, from above, $(\za,\zb)$ is a local $\nm$-equivalence, we conclude by Proposition \ref{pro-n-equiv} b) that $(\za,\zb)$ is a $n$-equivalence.
\end{proof}
\begin{proposition}\label{pro-crit-lev-nequiv}
Let $f:X\rw Y$ be a morphism in $\tawg{n}$ with $n\geq 2$, such that
\begin{itemize}
\item [a)] $f$ is a $\nequ$.
\item [b)] $\p{n-1}X_0 \cong \p{n-1}Y_0$,
\item [c)] For each $1\leq r < n-1$ and all $k_1, \ldots, k_r \geq 0$,
\begin{equation*}
\p{n-r-1}X_{k_1, \ldots, k_r,\, 0}\cong \p{n-r-1}Y_{k_1, \ldots, k_r,\, 0}\;.
\end{equation*}
Then $J_nf$ is a levelwise equivalence of categories.
\end{itemize}
\end{proposition}
\begin{proof}
By induction on $n$. Let $f:X\rw Y$ be a 2-equivalence in $\tawg{2}$ such that $p X_0=X_0^d\cong p Y_0=Y_0^d$. Since $f_0$ is a morphism in $\cathd{}$ it follows that $f_0$ is an equivalence of categories. Since $f(a,b)$ is an equivalence of categories for all $a,b\in X_0^d$ there is an equivalence of categories
\begin{equation*}
f_1:X_1=\underset{a,b\in X_0^d}{\cop}X(a,b)\rw Y_1= \underset{a',b'\in Y_0^d}{\cop}Y(a',b')=\underset{fa,fb\in Y_0^d}{\cop}Y(fa,fb).
\end{equation*}
Hence there are equivalences of categories for $k\geq 2$
\begin{equation*}
X_k \sim \pro{X_1}{X_0^d}{k}\sim \pro{Y_1}{Y_0^d}{k}\sim Y_k\;.
\end{equation*}
In conclusion $X_k\sim Y_k$ for all $k\geq 0$.
Suppose, inductively, that the statement holds for $(n-1)$ and let $f:X\rw Y$ be as in the hypothesis. We show that $J_{n-1}f_k$ is a levelwise equivalence of categories for each $k\geq 0$ by showing that $f_k$ satisfies the inductive hypothesis. It then follows that $J_n f$ is a levelwise equivalence of categories since
\begin{equation*}
(J_n f)_{k_1...k_{n-1}}=(J_{n-1}f_{k_1})_{k_2...k_{n-1}}\;.
\end{equation*}
Since $X_0\in\cathd{n-1}$, from b) and \cite[Lemma 3.8]{Pa1} we obtain
\begin{equation}\label{eq1-pro-crit-lev-nequiv}
X_0^d=\p{1,n}X_0 \cong \p{1,n}Y_0 \cong Y_0^d\;.
\end{equation}
Thus, by \cite[Lemma 3.8]{Pa1} again, $f_0:X_0\rw Y_0$ is a $\equ{n-1}$. Further, by hypothesis c),
\begin{align*}
& \p{n-2}X_{00} \cong \p{n-2}Y_{00}\\
& \p{n-r-2}X_{0\,k_1...k_r\,0} \cong \p{n-r-2}Y_{0\,k_1...k_r\,0}
\end{align*}
for each $1\leq r < n-2$ and all $k_1...k_r$. Thus $f_0:X_0\rw Y_0$ satisfies the inductive hypothesis and we conclude that $f_0$ is a levelwise equivalence of categories. By \eqref{eq1-pro-crit-lev-nequiv} we also have
\begin{equation*}
f_1:\underset{a,b\in X_0^d}{\cop}X(a,b)\rw Y_1 = \underset{a',b'\in Y_0^d}{\cop}Y(a',b')=\underset{fa,fb\in Y_0^d}{\cop}Y(fa,fb)\;.
\end{equation*}
Since $f$ is a local $\equ{n-1}$, it follows that $f_1:X_1 \rw Y_1$ is a $\equ{n-1}$. Further, by hypothesis c)
\begin{align*}
& \p{n-2}X_{10} \cong \p{n-2}Y_{10}\\
& \p{n-r-2}X_{1\,k_1...k_r\,0} \cong \p{n-r-2}Y_{1\,k_1...k_r\,0}
\end{align*}
for all $1 \leq r < n-2$. Thus $f_1$ satisfies the inductive hypothesis, and is therefore a levelwise equivalence of categories.
For each $k\geq 2$ consider the map
\begin{equation*}
(f_1,...,f_1):\pro{X_1}{X_0^d}{k}\rw \pro{Y_1}{Y_0^d}{k}\;.
\end{equation*}
Since $X_0^d \cong Y_0^d$ and, from above, $f_1$ is a $\equ{n-1}$, then $(f_1,...,f_1)$ is also a $\equ{n-1}$.
There is a commutative diagram in $\tawg{n-1}$
\begin{equation*}
\xymatrix{
X_k \ar^(0.3){\hmuk}[rr] \ar_{f_k}[d] && \pro{X_1}{X_0^d}{k} \ar^{(f_1,...,f_1)}[d]\\
Y_k \ar_(0.3){\hmuk}[rr] && \pro{Y_1}{Y_0^d}{k}
}
\end{equation*}
where the horizontal induced Segal maps are $(n-1)$-equivalences since $X,Y\in\tawg{n-1}$ and the right vertical map is a $(n-1)$-equivalence from above. It follows from Proposition \ref{pro-n-equiv} c) and d) that $f_k$ is a $(n-1)$-equivalence. Further, from hypothesis c),
\begin{align*}
& \p{n-1}X_{k0} \cong \p{n-2}Y_{k0}\\
& \p{n-r-2}X_{k\,k_1...k_r\,0} \cong \p{n-r-2}Y_{k\,k_1...k_r\,0}\;.
\end{align*}
Thus $f_k$ satisfies the induction hypothesis and we conclude that $f_k$ is a levelwise equivalence of categories.
In conclusion, $f_k$ is a levelwise equivalence of categories for all $k\geq 0$. Since this holds for each $k\geq0$ this implies that $f$ is a levelwise equivalence of categories.
\end{proof}
\section{The functor $\q{n}$.}\label{sec-func-qn}
This section introduces the functor
\begin{equation*}
\q{n}:\tawg{n}\rw \tawg{n-1}
\end{equation*}
This functor is a higher dimensional generalization of the connected component functor $q:\Cat\rw\Set$ and comes equipped with a morphism
\begin{equation*}
\zg\up{n}:X\rw \di{n}\q{n}X
\end{equation*}
for each $X\in \tawg{n}$, where $\di{n}$ is as in Notation \ref{not-ner-func-dirk} . It will be used crucially in Section \ref{sec-cat-lta} to replace a weakly globular \nfol category $X$ with a simpler one (Theorem \ref{the-repl-obj-1}). This will involve taking pullbacks along the map $\zg\up{n}$. The last part of this Section establishes several properties of these pullbacks needed in Section \ref{sec-cat-lta}.
\begin{proposition}\label{pro-post-trunc-fun}
There is a functor $\qn:\tawg{n}\rw \tawg{n-1}$ making the following diagram commute.
\begin{equation}\label{eq-lem-post-trunc}
\xymatrix@C=30pt{
\tawg{n} \ar^{J_{n}}[rr]\ar_{\qn}[d] && \funcat{n}{\Cat} \ar^{\ovl{q}}[d]\\
\tawg{n-1} \ar_{\Nb{n-1}}[rr] & & \funcat{n-1}{\Set}
}
\end{equation}
where $q:\Cat\rw \Set$ is the connected component functor. The functor $\qn$ sends $n$-equivalences to $\nm$-equivalences and preserves pullbacks over discrete objects. If $X\in\cathd{n}$, then $\q{n}X=\p{n}X$; further, for each $X\in \tawg{n}$, there is a map $\zgu{n}:X\rw \dn\qn X$ natural in $X$.
\end{proposition}
\begin{proof}
By induction on $n$; for $n=1$, $\q{1}=q:\Cat\rw\Set$ is the connected components functor which, by Lemma \ref{lem-q-pres-fib-pro}, has the desired properties. If $X\in\cathd{}$, in particular $X$ is a groupoid, so $pX=qX$.
Suppose there is a $\q{n-1}$ with the desired properties, let $X\in\tawg{n}$. We claim that $ \ovl{\q{n-1}}X\in\tawg{n-1}$. In fact, for each $s\geq 0$, by induction hypothesis
\begin{equation*}
( \ovl{\q{n-1}}X)_s=\q{n-1}X_s\in\tawg{n-2}\;.
\end{equation*}
Also, by induction hypothesis and by Definition \ref{def-hom-dis-ncat},
\begin{equation*}
( \ovl{\q{n-1}}X)_0 =\q{n-1}X_0 = \p{n-1}X_0 \in \cathd{n-1}
\end{equation*}
as $X_0 \in \cathd{n-1}$. Further, since
\begin{equation*}
\hmuk : X_k\rw \pro{X_1}{X_0^d}{k}
\end{equation*}
is a $\equ{n-1}$, by induction hypothesis the map
\begin{equation*}
\q{n-1}X_k\rw\q{n-1}(\pro{X_1}{X_0^d}{k})\cong \pro{\q{n-1}X_1}{X_0^d}{k}
\end{equation*}
is a $\equ{n-2}$; in this we used the fact that $\q{n-1}X_0^d\cong X_0^d$, which follows from diagram \eqref{eq-lem-post-trunc} at step $(n-1)$.
This shows that $ \ovl{\q{n-1}}X\in\tawg{n-1}$. We therefore define
\begin{equation*}
\q{n}X = \ovl{\q{n-1}}X\;.
\end{equation*}
The fact that $\q{n}$ satisfies diagram \eqref{eq-lem-post-trunc} is immediate from the definitions and the induction hypothesis.
If $X\in\cathd{n}$, by definition $X_k\in\cathd{n-1}$ for each $k$, so by induction hypothesis $\p{n-1}X_k=\q{n-1}X_k$. It follows that $(\p{n}X)_k=(\q{n}X)_k$ for all $k$. That is $\p{n}X=\q{n}X$.
Let $f:X\rw Y$ be a $\nequ$ in $\tawg{n}$ and let $a,b\in X_0^d$. Then from the definitions
\begin{equation*}
(\q{n}f)(a,b)=\q{n-1}f(a,b)\;.
\end{equation*}
Since $f(a,b)$ is a $\equ{n\mi1}$, by induction hypothesis $\q{n-1} f(a,b)$ is a $\equ{n-2}$. By Proposition \ref{pro-n-equiv}, to prove that $\q{n}f$ is a $\equ{n-1}$, it is enough to show that $\p{1,n-1}\q{n}f$ is an surjective.
Recall that for any category $\clC$ there is a surjective map $p\,\clC\rw q\,\clC$ natural in $\clC$. Applying this levelwise to $J_nX$ we obtain a map
\begin{equation*}
\za_X\up{n}:\p{n}X \rw \q{n}X
\end{equation*}
natural in $X$. The map $\za\up{n}$ induces a functor
\begin{equation}\label{eq-post-trunc-fun}
\p{2,n-1}\za_X\up{n}: \p{2,n-1}\p{n}X \rw \p{2,n-1}\q{n}X\;,
\end{equation}
which is identity on objects. In fact, on object this map is given by
\begin{equation*}
\p{1,n-2}\p{n-1}X_0 \rw \p{1,n-2}\q{n-1}X_0
\end{equation*}
and since $X_0\in\cathd{n-1}$, $\p{n-1}X_0=\q{n-1}X_0$ so this map is the identity. It follows that the map in $\Set$
\begin{equation*}
\p{1,n-1}\za_X\up{n-1}: \p{1,n-1}\p{n}X \rw \p{1,n-1}\q{n}X
\end{equation*}
is surjective. We thus have a commuting diagram
\begin{equation*}
\xymatrix{
\p{1,n-1}\p{n}X \ar^{\p{1,n-1}\p{n}f}[rr] \ar_{\p{1,n-1}\za_X\up{n}}[d] && \p{1,n-1}\p{n}Y \ar^{\p{1,n-1}\za_Y\up{n}}[d]\\
\p{1,n-1}\q{n}X \ar^{\p{1,n-1}\q{n}f}[rr] && \p{1,n-1}\q{n}Y
}
\end{equation*}
in which the top arrow is an isomorphism (by Proposition \ref{pro-n-equiv}) and from above the vertical arrows are surjective. It follows that the bottom map is also surjective. By Proposition \ref{pro-n-equiv} b) we conclude that $\q{n}f$ is a $\equ{n-1}$.
Finally, the map $\zgu{n}:X\rw \q{n}X$ is given levelwise by the maps $X_s\rw \q{n-1}X_s$, which exist by induction hypothesis.
\end{proof}
\begin{remark}\label{rem-spec-isofib}
For each $X\in \tawg{n}$, from the proof of Proposition \ref{pro-post-trunc-fun} the functor $\p{2,n-1}\za_X\up{n}$ is identity on objects. It is also surjective on morphisms since, by the proof of Proposition \ref{pro-post-trunc-fun}, the map
\begin{equation*}
(\p{2,n-1}\za_X\up{n})_1:\p{1,n-1}\p{n-1} X_1 \rw \p{1,n-2}\q{n-1} X_1 \;,
\end{equation*}
is surjective. It follows that $\p{2,n-1}\za_X\up{n}$ is an isofibration.
\end{remark}
\begin{lemma}\label{lem-spec-pulbk-eqr}
Let $X\in\cathd{n}$, $Z\in\cathd{n-1}$, $r: Z\rw \qn X$. Consider the pullback in $\funcat{n-1}{\Cat}$
\begin{equation*}
\xymatrix@R=35pt @C=40pt{
P \ar^{}[r] \ar^{}[d] & X \ar^{\zgu{n}}[d] \\
\dn Z \ar_{\dn r}[r] & \dn \qn X
}
\end{equation*}
then $P\in\cathd{n}$ and $\p{n}P=Z$.
\end{lemma}
\begin{proof}
By induction on $n$. For $n=1$, since $\di{1}\q{1}X$ is discrete, the map $\zgu{1}:X\rw \di{1}\q{1}X=X^d$ is an isofibration. Therefore, since $\zgu{1}$ is an equivalence of categories (as $X\in\cathd{}$) we have an equivalence of categories
\begin{equation*}
P=\di{1}Z\tiund{\di{1}\q{1}X} X\simeq \di{1}Z\tiund{\di{1}\q{1}X} \di{1}\q{1}X = \di{1}Z
\end{equation*}
Thus $P\in\cathd{}$ and $pP=Z$.
Suppose, inductively, that the lemma holds for $n-1$ and let $P$ be as in the hypothesis. Since pullbacks in $\funcat{n-1}{\Cat}$ are computed pointwise, for each $k\geq 0$ we have a pullback in $\funcat{n-2}{\Cat}$
\begin{equation*}
\xymatrix{
P_k \ar^{}[rr] \ar^{}[d] && X_k \ar^{}[d]\\
\di{n-1}Z_k \ar^{}[rr] && \di{n-1}\q{n-1}X_k
}
\end{equation*}
where $X_k\in\cathd{n-1}$ (since $X\in \cathd{n}$) and $Z_k \in \cathd{n-2}$ (since $Z\in\cathd{n-1}$). By induction hypothesis, we conclude that $P_k\in\cathd{n-1}$.
We now show that, for each $k \geq 2$
\begin{equation}\label{eq1-lem-spec-pulbk-eqr}
P_k\cong \pro{P_1}{P_0}{k}\;.
\end{equation}
We illustrate this for $k=2$, the case $k>2$ being similar. Since $X\in\cathd{n}$, $\q{n}X = \p{n}X\in\cathd{n-1}$, so
\begin{equation*}
\begin{split}
& \q{n-1}X_2=\p{n-1}X_2 =\p{n-1}(\tens{X_1}{X_0})= \\
=\ & \tens{\p{n-1} X_1}{\p{n-1} X_0}=\tens{\q{n-1} X_1}{\q{n-1} X_0}\;.
\end{split}
\end{equation*}
Since $X_2 \cong \tens{X_1}{X_0}$ and $Z_2 \cong \tens{Z_1}{Z_0}$, it follows from Lemma \ref{lem-char-obj-II} that $P_2 \cong \tens{P_1}{P_0}$.
To prove that $P\in\cathd{n}$ it remains to show that $\p{n}P\in\cathd{n-1}$. Since $p$ commutes with fiber products over discrete objects, for each $\us\in\dop{n-1}$ we have
\begin{align*}
&(\p{n}P)_{\us}=p\,P_{\us}=
p(d\,Z_{\us}\tiund{d q X_{\us}} X_{\us})
= Z_{\us}\tiund{qX_{\us}}p X_{\us}=Z_{\us}
\end{align*}
where we used the fact that, since $X_{\us}$ is a groupoid, $pX_{\us}=qX_{\us}$. Since this holds for each ${\us}$ we conclude that $\p{n}P=Z\in\cathd{n-1}$ as required.
\end{proof}
\begin{lemma}\label{lem-p2-n-1}
Let $Y\in\tawg{n}$ and let
\begin{equation}\label{eq1-lem-p2-n-1}
X\rw \q{n}Y \lw \p{n}Y
\end{equation}
be a diagram in $\tawg{n-1}$ such that $X\tiund{\q{n}Y}\p{n}Y\in \tawg{n-1}$. Then for all $1\leq j \leq n-1$
\begin{equation*}
\p{j,n-1}(X\tiund{\q{n}Y}\p{n}Y)=\p{j,n-1}X\tiund{\p{j,n-1}\q{n}Y}\p{j,n-1}\p{n}Y\;.
\end{equation*}
\end{lemma}
\begin{proof}\
By induction on $n$. For $n=2$, the functor $\p{2}Y\rw \q{2}Y$ is the identity on objects, therefore by Lemma \ref{lem-iso-cla-obj-fib-pro}
\begin{equation*}
p(X\tiund{\q{2}Y}\p{2}Y)=pX\tiund{p\q{2}Y}p\p{2}Y\;.
\end{equation*}
Suppose, inductively, that it holds for $n-1$. Then for each $k\geq 0$
\begin{equation}\label{eq2-lem-p2-n-1}
\begin{split}
& (\p{j,n-1}(X\tiund{\q{n}Y}\p{n}Y))_k= \\
=\; &\p{j-1,n-2}(X_k\tiund{\q{n-1}Y_k}\p{n-1}Y_k)=\\
=\; &\p{j-1,n-2}X_k \tiund{\p{j-1,n-2}\q{n-1}Y_k} \p{j-1,n-2}\p{n-1}Y_k=\\
=\; &(\p{j,n-2}X)_k \tiund{(\p{j,n-2}\q{n-1}Y)_k} (\p{j,n-2}\p{n-1}Y)_k\;.
\end{split}
\end{equation}
Since this holds for each $k\geq 0$, the lemma follows.
\end{proof}
\begin{proposition}\label{pro-spec-plbk-pscatwg}
Let $\di{n}A\xrw{\ \di{n}f\ }\di{n}\q{n}C \xlw{\ g\ } C$ be a diagram in $\tawg{n}$ where $f: A\rw\q{n}C$ is a morphism in $\tawg{n-1}$ and consider the pullback in $\funcat{n-1}{\Cat}$
\begin{equation*}
P=\di{n}A \tiund{\di{n}\q{n}C} C\;.
\end{equation*}
\begin{itemize}
\item [a)] Then $P\in \tawg{n}$ and
\begin{equation}\label{eq-pro-spec-plbk-pscatwg0}
\p{1,n}P = \p{1,n-1} A \tiund{\p{1,n-1}\q{n}C} \p{1,n} C\;.
\end{equation}
\item [b)]
Consider the commutative diagram in $\tawg{n}$
\begin{equation}\label{eq-pro-spec-plbk-pscatwg}
\xymatrix@R=35pt @C=60pt{
\di{n}A \ar^{\di{n}f}[r] \ar_{a}[d] & \di{n}\q{n} C \ar^{b}[d] & C \ar^{c}[d] \ar_{g}[l] \\
\di{n}D \ar_{\di{n}h}[r] & \di{n}\q{n} F & F \ar^{l}[l]
}
\end{equation}
where $a,b,c$ are $\nequ$s. Then the induced maps of pullbacks
\begin{equation*}
(a,c): \di{n}A\tiund{\di{n}\q{n} C} C \rw \di{n}D \tiund{\di{n}\q{n} F} F
\end{equation*}
is a $n$-equivalence in $\tawg{n}$.
\medskip
\item [c)] If $f$ is an $(n-1)$-equivalence, $P\xrw{w}C$ is an $n$-equivalence.
\end{itemize}
\end{proposition}
\begin{proof}
By induction on $n$. When $n=1$, the maps $f,g,h,l$ are isofibrations since their target is a discrete category. Therefore, since $a,b,c$ are equivalences of categories, the induced map of pullbacks
\begin{equation*}
(a,c):d A\tiund{d q C} C \rw d D \tiund{d q F} F
\end{equation*}
is an equivalence of categories and $p (d A\tiund{dqC}C) = A \tiund{qC} pC$.
Suppose, inductively, that the proposition holds for $n-1$.
\bk
a) We have
\begin{equation*}
(\di{n}A\tiund{\di{n}\q{n} C} C)_0=\di{n-1}A_0\tiund{\di{n-1}\q{n-1} C_0} C_0\;.
\end{equation*}
Since, by hypothesis, $A \in\tawg{n-1}$ and $C$ is in $\tawg{n}$, by definition $A_0 \in\cathd{n-2}$ and $C_0 \in \cathd{n-1}$. Therefore by Lemma \ref{lem-spec-pulbk-eqr}
\begin{equation*}
\di{n-1}A_0\tiund{\di{n-1}\q{n-1} C_0} C_0\in \cathd{n-1}\;.
\end{equation*}
Further, for each $k\geq 1$,
\begin{equation*}
(\di{n}A\tiund{\di{n}\q{n} C} C)_k = \di{n-1}A_k\tiund{\di{n-1}\q{n-1} C_k} C_k
\end{equation*}
where, by hypothesis, $A_k \in\tawg{n-2}$ and $C_k \in \tawg{n-1}$. It follows by inductive hypothesis that
\begin{equation*}
\di{n-1}A_k\tiund{\di{n-1}\q{n-1} C_k} C_k \in \tawg{n-1}\;.
\end{equation*}
To prove that $\di{n}A\tiund{\di{n}\q{n} C} C \in \tawg{n}$, it remains to show that its induced Segal maps $\hmu{s}$ are $\equ{n-1}$s for all $s\geq 2$. We show this for $s=2$, the case $s>2$ being similar. We have
\begin{equation*}
\begin{split}
& \hmu{2}:(\di{n}A\tiund{\di{n}\q{n} C} C)_2 \rw \\
& \rw \tens{(\di{n}A\tiund{\di{n}\q{n} C} C)_1}{(\di{n}A\tiund{\di{n}\q{n} C} C)_0^d}\;.
\end{split}
\end{equation*}
By Lemma \ref{lem-spec-pulbk-eqr},
\begin{equation*}
\begin{split}
& (\di{n}A\tiund{\di{n}\q{n} C} C)_0^d= (\p{n-1}(\di{n-1}A_0\tiund{\di{n-1}\q{n-1} C_0} C_0))^d=A_0^d = \\
& = A_0^d\tiund{\di{n-1}\q{n-1} C_0^d} C_0^d
\end{split}
\end{equation*}
where we used the fact that, since $\q{n-1}\di{n-1}=\Id$,
\begin{equation*}
\begin{split}
& \di{n-1}\q{n-1}C_0^d= \di{n-1}\q{n-1}\di{n-1}...\di{1}p...\p{n-1}C_0= \\
=\; &\di{n-1}\di{n-2}...\di{1}p...\p{n-1}C_0=C_0^d\;.
\end{split}
\end{equation*}
Recalling that $\q{n-1}$ commutes with pullbacks over discrete objects, we obtain
\begin{equation*}
\begin{split}
& \tens{(\di{n}A\tiund{\di{n}\q{n}C}C)_1}{(\di{n}A\tiund{\di{n}\q{n}C}C)_0^d}= \\
& \resizebox{1.0\hsize}{!}{$ =\tens{(\di{n-1}A_1\tiund{\di{n-1}\q{n-1}C_1}C_1)}{(A_0^d\tiund{\di{n-1}\q{n-1}C_0^d}C_0^d)}=$} \\
& = \di{n-1}(\tens{A_1}{A_0^d})\tiund{\di{n-1}\q{n-1}(\tens{C_1}{C_0^d})}(\tens{C_1}{C_0^d})\;.
\end{split}
\end{equation*}
On the other hand
\begin{equation*}
(\di{n}A\tiund{\di{n}\q{n}C} C)_2 = \di{n-1}A_2 \tiund{\di{n-1}\q{n-1}C_2} C_2 \;.
\end{equation*}
Hence we see that the map $\hmu{2}$ is the induced map on pullbacks from the diagram in $\tawg{n-1}$
\begin{equation}\label{eq1-pro-spec-plbk-pscatwg}
\xymatrix@C=15pt{
\di{n-1}A_2 \ar^{}[rr] \ar^{}[d] && \di{n-1}\q{n-1}C_2 \ar^{}[d] && C_2 \ar^{}[ll] \ar^{}[d]\\
\di{n-1}(\tens{A_1}{A_0^d}) \ar^{}[rr] && \di{n-1}\q{n-1}(\tens{C_1}{C_0^d}) && \tens{C_1}{C_0^d} \ar^{}[ll]
}
\end{equation}
In this diagram, the left and right vertical maps are $\nm$-equivalences since they are induced Segal maps of $\di{n}A$ and $C$ respectively. The map
\begin{equation*}
\q{n-1}C_2 \rw \q{n-1}(\tens{C_1}{C_0^d}) = \tens{\q{n-1}C_1}{(\q{n-1}C_0)^d}
\end{equation*}
is the induced Segal map for $\q{n}C\in\tawg{n-1}$, and is therefore a $(n-2)$-equivalence. The central vertical map in \eqref{eq1-pro-spec-plbk-pscatwg} is therefore in particular a $(n-1)$-equivalence. We can therefore apply the induction hypothesis b) to the diagram \eqref{eq1-pro-spec-plbk-pscatwg} and conclude that the induced map on pullbacks is a $\equ{n-1}$. That is $\hmu{2}$ is a $\nm$-equivalence.
Similarly one shows that $\hmu{s}$ is a $\nm$-equivalence for all $s \geq 2$. This concludes the proof that $\di{n}A \tiund{\di{n}\q{n}C} C \in \tawg{n}$ and \eqref{eq-pro-spec-plbk-pscatwg0} follows from Lemma \ref{lem-p2-n-1}.
\bk
b) We now show that $(a,c)$ is a $\nequ$. By Proposition \ref{pro-n-equiv} b), it is enough to show that it is a local $\equ{n-1}$ and that $\p{1,n}(a,c)$ is an isomorphism.
Let
\begin{equation*}
(x_1,y_1),(x_2,y_2) \in (\di{n}A \tiund{\di{n}\q{n}C} C)_0^d = A_0^d \tiund{\di{n-1}\q{n-1}C_0^d} C_0^d.
\end{equation*}
Then the map
\begin{equation*}
(a,c)((x_1,y_1),(x_2,y_2))
\end{equation*}
is the induced map on pullbacks of the diagrams
\begin{equation}\label{eq2-pro-spec-plbk-pscatwg}
\xymatrix@C=14pt{
\di{n}A(x_1,x_2) \ar^(0.38){f(x_1,x_2)}[rr] \ar_{a(x_1,x_2)}[d] && \di{n-1}\q{n-1}C(fx_1,fx_2) \ar^{b(fx_1,fx_2)}[d] && C(y_1,y_2) \ar_(0.38){g(y_1,y_2)}[ll] \ar_{c(y_1,y_2)}[d]\\
\di{n-1}D(a x_1,a x_2) \ar_(0.45){h(a x_1,a x_2)}[rr] && \di{n-1}\q{n-1}F(h a x_1,h a x_2) && C(cy_1,cy_2) \ar^(0.35){C(cy_1,cy_2)}[ll]
}
\end{equation}
The vertical maps in \eqref{eq2-pro-spec-plbk-pscatwg} are $\equ{n-1}$s. By the inductive hypothesis b), since $a,b,c$ are $n$-equivalences we conclude that the induced map of pullbacks is a $\equ{n-1}$. This shows that $(a,c)$ is a local $\equ{n-1}$. By Lemma \ref{lem-p2-n-1}, $\p{n-1}(a,c)=(\p{1,n}a,\p{1,n}c)$.
Applying the functor $\p{1,n}$ to the diagram \eqref{eq-pro-spec-plbk-pscatwg} we obtain a commutative diagram in $\Set$
\begin{equation*}
\xymatrix{
\p{1,n-1}A \ar^{}[rr] \ar_{\p{1,n-1}a}[d] && \p{1,n-1}\q{n}C \ar^{\p{1,n}b}[d] && \p{1,n-1}\p{n}C \ar^{}[ll] \ar^{\p{1,n}c}[d]\\
\p{1,n-1}D \ar^{}[rr] && \p{1,n-1}\q{n}F && \p{1,n-1}\p{n}C \ar^{}[ll]
}
\end{equation*}
Since, by hypothesis, $a,b$ and $c$ are $\nequ$s, by Proposition \ref{pro-n-equiv} a) the vertical maps are isomorphisms, hence such is $\p{1,n}(a,c)$ as required. We conclude from Proposition \ref{pro-n-equiv} b) that $(a,c)$ is an $n$-equivalence.
\bk
c) By a), we have a commutative diagram in $\tawg{n}$
\begin{equation*}
\xymatrix{
\di{n}A \ar^{f}[rr] \ar_{f}[d] && \di{n}\q{n}C \ar@{=}[d] && C \ar_{g}[ll] \ar@{=}[d] \\
\di{n}\q{n}C \ar@{=}[rr] && \di{n}\q{n}C && C \ar^{g}[ll]
}
\end{equation*}
in which the vertical maps are $n$-equivalences. It follows by b) that the induced map of pullbacks
\begin{equation*}
P= \di{n}A\tiund{\di{n}\q{n}}C \xrw{w} \di{n}\q{n} C \tiund{\di{n}\q{n}} C = C
\end{equation*}
is an $n$-equivalence.
\end{proof}
\section{The category $\mathbf{\lta{n}}$.}\label{sec-cat-lta}
In this section we introduce the subcategory $\lta{n}$ of $\tawg{n}$. We will show in Section \ref{sec-wg-tam-to-psefun} how to associate to this category a Segalic pseudo-functor, which in turn will lead to the contruction of the rigidification functor $Q_n$.
The main result of this section is Theorem \ref{the-repl-obj-1} establishing that that every object of $\tawg{n}$ can be approximated up to an $n$-equivalence with an object of $\tawg{n}$.
\begin{definition}\label{def-ind-sub-ltawg}
Define inductively the subcategory $\lta{n}\subset\tawg{n}$. For $n=2$, $\lta{2}=\tawg{2}$. Suppose we defined $\lta{n-1}\subset \tawg{n}$. Let $\lta{n}$ be the full subcategory of $\tawg{n}$ whose objects $X$ are such that
\begin{itemize}
\item [i)] $X_k\in\lta{n-1}$ for all $k\geq 0$.\bk
\item [ii)] The maps in $\funcat{n-2}{\Cat}$
\begin{equation*}
v_k: J_{n-1} X_k \rw J_{n-1}(\pro{X_1}{\di{n-1}\p{n-1}X_0}{k})
\end{equation*}
are levelwise equivalences of categories for all $k\geq 2$\bk
\item [iii)] $\p{n}X\in\catwg{n-1}$.
\end{itemize}
\end{definition}
\begin{definition}\label{def-ltawg-equiv}
Let $\lnta{n}{n}$ be the full subcategory of $\tawg{n}$ whose objects $X$ are such that
\begin{itemize}
\item [i)] The maps in $\funcat{n-2}{\Cat}$
\begin{equation*}
v_k: J_{n-1}X_k \rw J_{n-1}(\pro{X_1}{\di{n-1}\p{n-1}X_0}{k})
\end{equation*}
are levelwise equivalences of categories for all $k\geq 2$\bk
\item [ii)] $\p{n}X\in\catwg{n-1}$.
\end{itemize}
\end{definition}
\begin{lemma}\label{lem-lev-wg-pscat}
Let $X\in\tawg{n}$. Then $X\in \lta{n}$ if and only if
\begin{itemize}
\item [a)] $X\in \lnta{n}{n}$.\bk
\item [b)] For each $1 < r \leq n-1$ and each $k_1,...,k_{n-r}\in\Dop$, $X_{k_1,...,k_{n-r}}\in \lnta{r}{r}$
\end{itemize}
\end{lemma}
\begin{proof}
By induction on $n$. It holds for $n=2$ since $\lta{2}=\Lb{2}\tawg{2}=\tawg{2}$. Suppose it holds for $(n-1)$ and let $X\in\lta{n}$. Then by definition $X\in\Lb{n}\tawg{n}$. Also by definition $X_{k_1\cdots k_{n-r}}\in\lta{r}$ and therefore $X_{k_1\cdots k_{n-r}}\in\Lb{r}\tawg{r}$.
Conversely, suppose that $X\in\tawg{n}$ satisfies a) and b). By a), $v_k$ is a levelwise equivalence of categories and $\p{n}X\in\catwg{n-1}$; by b), $X_k\in\Lb{n-1}\tawg{n-1}$ and, further, $X_k$ itself satisfies b). Thus by induction hypothesis applied to $X_k$ we conclude that $X_k\in\lta{n-1}$. By definition, this shows that $X\in\lta{n}$.
\end{proof}
\begin{corollary}\label{cor-lev-wg-ncat}
Let $X\in\catwg{n}$. Then $X\in \lta{n}$.
\end{corollary}
\begin{proof}
We claim that $X\in\lnta{n}{n}$. We prove the claim by induction on $n$. For $n=2$, if $X\in\catwg{2}$, for each $s\geq 2$ there is an equivalence of categories
\begin{equation*}
X_s\simeq \pro{X_1}{X_0^d}{s}=\pro{X_1}{\di{1}\p{1}X_0}{s}\;.
\end{equation*}
Thus, by definition, $X\in\lnta{2}{2}$. Suppose, inductively, that the statement holds for $n-1$ and let $X\in\catwg{n}$. Then, by Proposition \ref{pro-property-WG-nfol}, $X_k\up{2}\in \catwg{n-1}$ for all $k\geq 0$. Hence by inductive hypothesis $X_k\up{2}\in\lnta{n-1}{n-1}$. From the definition of $X_k\up{2}$ this means that, for all $s\geq 2$ and $k\geq 0$, the map
\begin{equation*}
J_{n-2} X_{sk}\rw J_{n-2}(\pro{X_{1k}}{\di{n-2}\p{n-2}X_{0k}}{s})
\end{equation*}
is a levelwise equivalence of categories. Since this holds for every $k\geq 0$, this implies that the map
\begin{equation*}
J_{n-1} X_s \rw J_{n-1}(\pro{X_1}{\di{n-1}\p{n-1}X_0}{s})
\end{equation*}
is a levelwise equivalence of categories. Since $\p{n}X\in\catwg{n-1}$ this implies that $X\in\lnta{n}{n}$, as claimed.
Since $X\in\catwg{n}$, by definition $X_{\seq{k}{1}{n-r}}\in \catwg{r}$, thus, from above $X_{\seq{k}{1}{n-r}}\in \lnta{r}{r}$. Hence, by Lemma \ref{lem-lev-wg-pscat}, $X\in\lta{n}$.
\end{proof}
\begin{remark}\label{rem-lev-wg-ncat}
If $X\in\catwg{n}$, not only the map
\begin{equation*}
J_{n-1}X_k\rw J_{n-1}(\pro{X_1}{\di{n-1}\p{n-1}X_0}{k})
\end{equation*}
is a levelwise equivalence of categories, but so is the map
\begin{equation}\label{eq-rem-lev-wg-ncat}
J_{n-1} X_k\rw J_{n-1}(\pro{X_1}{\di{j,n-1}\p{j,n-1}X_0}{k})
\end{equation}
where $\di{j,n-1}=\di{n-1}...\di{j}$ for $j=1,...,n-1$. We can see this by induction on $n$. The case $n=2$ is as in proof of Corollary \ref{cor-lev-wg-ncat}. Suppose, inductively, that the statement holds for $n-1$ and let $X\in\catwg{n}$. Then, by Proposition \ref{pro-property-WG-nfol}, $X_k\up{2}\in\catwg{n-1}$ for all $k\geq 0$. Hence by inductive hypothesis, for all $k\geq 2$ the map
\begin{equation*}
J_{n-2} X_{sk}\rw J_{n-2}(\pro{X_{1k}}{\di{j-1,n-2}\p{j-1,n-2}X_{0k}}{k})
\end{equation*}
is a levelwise equivalence of categories. Since this holds for all $k\geq 0$, this means that \eqref{eq-rem-lev-wg-ncat} is a levelwise equivalence of categories.
\end{remark}
The following lemma is used in the proof of the main result of this section, Theorem \ref{the-repl-obj-1}.
\begin{lemma}\label{lem-jn-alpha}
Let $B \xrw{\pt_0} X \xlw{\pt_1}B$ be a diagram in $\tawg{n}$ with $X\in\cathd{n}$ and let
\begin{equation*}
A\xrw{\za}\tens{B}{X}\xrw{\zb}\tens{B}{X^d}
\end{equation*}
be maps $\tawg{n}$ (where $\zb$ is induced by the map $\zg: X\rw X^d$) such that\mk
\begin{itemize}
\item [i)] $\p{n-1}\za_0$, and $\p{n-r-1}\za_{k_1...k_r\,0}$ are isomorphisms for all $1\leq r< n-1$.\mk
\item [ii)] $(\tens{B}{X})^d_0\cong\tens{B_0^d}{X_0^d}$\mk
\nid $(\tens{B}{X})^d_{k_1...k_r\,0}\cong\tens{B^d_{k_1...k_r\,0}}{X^d_{k_1...k_r\,0}}$\mk
\item [iii)] $\zb\za$ is a $n$-equivalence.\mk
\end{itemize}
Then $J_{n}\za$ is a levelwise equivalence of categories.
\end{lemma}
\begin{proof}
Let $x,x'\in A_0^d$. By hypothesis i) and ii),
\begin{equation*}
A_0^d\cong(\tens{B}{X})^d_0=\tens{B_0^d}{X_0^d}\subset \tens{B_0^d}{X^d}
\end{equation*}
where the last inclusion holds because the map $\zg_0:X_0\rw (X^d)_0$ factors through $X_0^d$. Let $\za x=(a,b)\in \tens{B_0^d}{X_0^d}$ and $\za x'=(a',b')$. We have
\begin{flalign*}
&(\tens{B}{X})(x,x')=B(a,a')\tiund{X(\pt_0 a,\pt_0 a')}B(b,b')&\\
&(\tens{B}{X^d})(\zb x,\zb x')=B(a,a')\tiund{X^d(\zg a,\zg a')}B(b,b')=B(a,a')\times B(b,b')&
\end{flalign*}
where in the last equality we used the fact that $da=da'$ so that $X^d(\zg a,\zg a')=\{\cdot\}$ since $X^d$ is discrete.
The map $\zg:X\rw X^d$ factors as
\begin{equation*}
X\rw \di{n}...\di{2}\p{2}...\p{n}X\rw \di{n}...\di{2}d p...\p{n}X=X^d
\end{equation*}
and we have
\begin{equation*}
(\di{n}...\di{2}\p{2}...\p{n}X)(p \pt_0 a,p \pt_0 a')=X(\pt_0 a,\pt_0 a')^d\;.
\end{equation*}
Thus the map $\zb(x,x')$ factors as
\begin{equation*}
\resizebox{1.0\hsize}{!}{$
B(a,a')\tiund{X(\pt_0 a,\pt_0 a')}B(b,b')\xrw{s} B(a,a')\tiund{X(\pt_0 a,\pt_0 a')^d}B(b,b')\xrw{t}B(a,a')\times B(b,b')\;.$}
\end{equation*}
On the other hand, since $\p{2,n}X\in\cathd{}$, the set $\p{2,n}X(p\pt_0 a,p\pt_0 a')$ contains only one element. Thus $X(\pt_0 a,\pt_0 a')^d$ is the terminal object, and $t=\Id$.
Since, by hypothesis, $\zb\za$ is a $n$-equivalence $\zb\za(x,y)$ is a $\nm$-equivalence, so by the above the composite
\begin{equation}\label{eq1-lem-jn-alpha}
A(x,y)\xrw{\za(x,y)} B(a,a')\tiund{X(\pt a,\pt a')}B(b,b')\xrw{s} B(a,a')\tiund{X(\pt a,\pt a')^d}B(b,b')
\end{equation}
is a $\equ{n-1}$.
\smallskip
We now proceed to the rest of the proof by induction on $n$. When $n=2$, since $X(\pt_0 a,\pt_0 a')\in\cathd{}$, the map $s$ is fully faithful. Since $s\za(x,y)$ is an equivalence of categories, it is essentially surjective on objects, and therefore $s$ is essentially surjective on objects. It follows that $s$ is an equivalence of categories, and therefore such is $\za(x,y)$. Since by hypothesis $p\za_0$ is a bijection, the map $\p{2}\za$ is bijective on objects, thus $p\p{2}\za$ is surjective. From Proposition \ref{pro-n-equiv} we deduce that $\za$ is a 2-equivalence.
Thus $\za$ satisfies the hypotheses of Proposition \ref{pro-crit-lev-nequiv} and we conclude that it is a levelwise equivalence of categories.
Suppose, inductively, that the lemma holds for $n-1$. We show that the maps \eqref{eq1-lem-jn-alpha} satisfy the inductive hypothesis.
Since $s\za(x,y)$ is a $\equ{n-1}$, inductive hypothesis iii) holds. Since by hypothesis $\p{n-1}\za_0$ is an isomorphism, so is
\begin{equation*}
\p{n-2}\za_0(x,y)=(\p{n-1}\za_0)(x,y)
\end{equation*}
as well as
\begin{equation*}
\p{n-r-2}\{\za_{k_1...k_r\,0}(x,y)\}=(\p{n-r-1}\za_{k_1...k_r\,0})(x,y)\;.
\end{equation*}
Thus inductive hypothesis i) holds for the maps \eqref{eq1-lem-jn-alpha}. Further, using hypothesis ii) we compute
\begin{equation*}
\begin{split}
& ((\tens{B}{X})(x,y))^d_0=(\tens{B_1}{X_1})^d_0(x,y)= \\
=\ & (\tens{B^d_{10}}{X^d_{10}})(x,y)=\tens{B(x,y)^d_0}{X(x,y)^d_0}
\end{split}
\end{equation*}
and similarly
\begin{equation*}
\begin{split}
& ((\tens{B}{X})(x,y))^d_{k_1...k_r\,0}=(\tens{B}{X})^d_{1 k_1...k_r\,0}(x,y)= \\
\cong\ & \tens{B^d_{1 {k_1...k_r\,0}}(x,y)}{X^d_{1 {k_1...k_r\,0}}(x,y)}=\\
=\ &\tens{B_{k_1...k_r\,0}(x,y)^d}{X_{k_1...k_r\,0}(x,y)^d}\;.
\end{split}
\end{equation*}
Thus the inductive hypothesis ii) holds for the maps \eqref{eq1-lem-jn-alpha}.
We conclude by induction that $J_{n-1}\za(x,y)$ is a levelwise equivalence of categories, hence in particular it is a $\equ{n-1}$. Since by hypothesis $\p{n-1}\za_0$ is an isomorphism, so is
\begin{equation*}
(\p{2,n}\za)_0= \p{1,n-1}\za_0
\end{equation*}
so that $\p{1,n}\za$ is surjective. Since, from above, $\za$ is a local $\nm$-equivalence, from Proposition \ref{pro-n-equiv} it follows that $\za$ is a $\nequ$. Together with hypothesis i) this shows that $\za$ satisfies the hypotheses of Proposition \ref{pro-crit-lev-nequiv} and we conclude that $J_{n}\za$ is a levelwise equivalence of categories.
\end{proof}
\begin{theorem}\label{the-repl-obj-1}
Let $X\in \tawg{n}$, and let $r:Z\rw \di{n} \qn X$ be a map in $\tawg{n-1}$ with $Z\in\catwg{n-1}$ and consider the pullback in $\funcat{n-1}{\Cat}$
\begin{equation*}
\xymatrix@R=35pt @C=40pt{
P \ar^{w}[r] \ar^{}[d] & X \ar^{\zgu{n}}[d] \\
\dn Z \ar_{\dn r}[r] & \dn \qn X
}
\end{equation*}
Then
\begin{itemize}
\item [a)] For all $ k\geq 2$
\begin{equation*}
\za: J_{n-1}P_k\rw J_{n-1}(\pro{P_1}{\di{n-1}\p{n-1}P_0}{k})
\end{equation*}
is a levelwise equivalence of categories.\mk
\item [b)] $\q{n}P$ and $\p{n}P$ are in $\catwg{n-1}$.\mk
\item [c)] $P\in\lnta{n}{n}$.\mk
\item [d)] If $r$ is a $\nm$-equivalence then $w$ is a $\nequ$.\mk
\item [e)] $P\in\lta{n}$.
\end{itemize}
\end{theorem}
\begin{proof}
By induction in $n$. When $n=2$, we know by Proposition \ref{pro-spec-plbk-pscatwg} that $P\in\tawg{2}=\lnta{2}{2}$, proving a) and c). Part b) is trivial since $\p{2}P$ and $\q{2}P$ are in $\Cat$. Part a) follows from Lemma \ref{pro-spec-plbk-pscatwg} c). Part e) holds since $tawg{2}=\lta{2}$
\medskip
Suppose, inductively that the theorem holds for $n-1$.
\bk
a) We first show that for each $k\geq 2$
\begin{equation*}
\pro{P_1}{\di{n-1}\p{n-1}P_0}{k}\in \tawg{n-1}\;.
\end{equation*}
We illustrate this for $k=2$, the case $k>2$ being similar. Since $p$ commutes with pullbacks over discrete objects we have
\begin{equation*}
\p{n}P=Z\tiund{\q{n}X}\p{n}X
\end{equation*}
and, (since $\q{n-1}X_0=\p{n-1}X_0$ as $X_0\in\cathd{n-1}$)
\begin{equation*}
\p{n-1}P_0=Z_0\tiund{\q{n-1}X_0}\p{n-1}X_0=Z_0\;.
\end{equation*}
Also, $P_1=\di{n-1}Z_1\tiund{\di{n-1}\q{n-1}X_1}X_1$. Therefore
\begin{equation*}
\begin{split}
& \tens{P_1}{\di{n-1}\p{n-1}P_0}= \\
=\ & \tens{(\di{n-1}Z_1\tiund{\di{n-1}\q{n-1}X_1}X_1)}{\di{n-1}Z_0}=\\
=\ & \di{n-1}(\tens{Z_1}{Z_0})\tiund{\di{n-1}\q{n-1}({X_1}\times{X_1})} ({X_1}\times{X_1})\;.
\end{split}
\end{equation*}
By Proposition \ref{pro-spec-plbk-pscatwg}, this is an object of $\tawg{n-1}$.
The induced Segal map $\hmu{2}$ for $P$ can therefore be written as composite of maps in $\tawg{n-1}$
\begin{equation}\label{eq-the-repl-obj-1}
P_2 \xrw{\za} \tens{P_1}{\di{n-1}\p{n-1}P_0} \xrw{\zb} \tens{P_1}{P_0^d}\;.
\end{equation}
We show that the maps \eqref{eq-the-repl-obj-1} satisfies the hypotheses of Lemma \ref{lem-jn-alpha}.
Since $P\in\tawg{n}$, $\hmu{2}=\zb\za$ is a $\equ{n-1}$, so hypothesis iii) of Lemma \ref{lem-jn-alpha} holds for the maps \eqref{eq-the-repl-obj-1} .
To check hypothesis i), note that
\begin{equation*}
\begin{split}
& \p{n-2}P_{20}=\p{n-2}(\di{n-2}Z_{20}\tiund{\di{n-2}\q{n-2}X_{10}} \di{n-2}\p{n-2}X_{10})=Z_{20} \\
& \p{n-2}(\tens{P_1}{\di{n-1}\p{n-1}P_0})_0=\\
=\ &\tens{\p{n-2}P_{10}}{\p{n-2}P_{00}}=\tens{Z_{10}}{Z_{00}}\cong Z_{20}
\end{split}
\end{equation*}
where the last isomorphism holds since $Z\in\catwg{n-1}$.
Hence $\p{n-2}\za_0$ is an isomorphism. Similarly
\begin{equation*}
P_{s_1...s_r}=Z_{s_1...s_r}\tiund{\q{n-r-1}X_{s_1...s_r}}X_{s_1...s_r}
\end{equation*}
\begin{equation*}
\p{n-r-1}P_{s_1...s_r}=Z_{s_1...s_r}\tiund{\q{n-r-1}X_{s_1...s_r}}X_{s_1...s_r}\;.
\end{equation*}
Thus
\begin{equation*}
\begin{split}
& \p{n-r-2}P_{2 k_1...k_r 0}=Z_{2 k_1...k_r 0} \\
& \p{n-r-2}(\tens{P_1}{\di{n-1}\p{n-1}P_0})_{k_1...k_r 0}=\\
=\ & \p{n-r-2}(\tens{P_{1 k_1...k_r 0}}{\di{n-r-1}\p{n-r-1}P_{0 k_1...k_r 0}})=\\
=\ & \tens{\p{n-r-2}P_{1 k_1...k_r 0}}{\p{n-r-1}P_{0 k_1...k_r 0}}=\\
=\ & \tens{Z_{1 k_1...k_r 0}}{Z_{0 k_1...k_r 0}}\cong Z_{2 k_1...k_r 0}\;.
\end{split}
\end{equation*}
where the last isomorphism holds since $Z\in\catwg{n-1}$. This shows that $\p{n-r-2}\za_{k_1...k_r 0}$ is an isomorphism, proving hypothesis i) of Lemma \ref{lem-jn-alpha} for the maps \eqref{eq-the-repl-obj-1}.
To check hypothesis ii) of Lemma \ref{lem-jn-alpha} note that $P^d_{20}=Z^d_{20}$ while
\begin{equation*}
\begin{split}
& (\tens{P_1}{\di{n-1}\p{n-1}P_0})^d_0=(\tens{P_{10}}{\di{n-2}\p{n-2}P_{00}})^d_0= \\
=\ & (\tens{Z_{10}}{\di{n-2}\p{n-2}Z_{00}})^d=\tens{Z_{10}^d}{Z_{00}^d}\cong Z_{20}^d
\end{split}
\end{equation*}
where the last isomorphism holds because $Z\in\catwg{n}$ (apply Remark \ref{rem-eq-def-wg-ncat} to $Z_0\up{2}$ which is an object of $\catwg{n-2}$ by Proposition \ref{pro-property-WG-nfol}). Similarly
\begin{equation*}
\begin{split}
& (P_2)^d_{k_1...k_r 0}=Z^d_{2 k_1...k_r 0} \\
& (\tens{P_1}{\di{n-1}\p{n-1}P_0})^d_{k_1...k_r 0}=\\
=\ & (\tens{Z_{1 k_1...k_r 0}}{\di{n-1}\p{n-1}Z_{0 k_1...k_r 0}})^d=\\
=\ & \tens{Z^d_{1 k_1...k_r 0}}{Z^d_{0 k_1...k_r 0}}=Z^d_{2 k_1...k_r 0}\;.
\end{split}
\end{equation*}
where the last equality holds because $Z\in\catwg{n-1}$ and therefore $Z_{k_0...k_r}\in\catwg{n-r-1}$ by applying Remark \ref{rem-eq-def-wg-ncat} to $(Z_{k_0...k_r})_0\up{2}$ which is an object of $\catwg{n-r-1}$ by Proposition \ref{pro-property-WG-nfol}. This proves that hypothesis ii) of Lemma \ref{lem-jn-alpha} holds for \eqref{eq-the-repl-obj-1}.
So all hypotheses of Lemma \ref{lem-jn-alpha} holds for the maps \eqref{eq-the-repl-obj-1} and we conclude that $J_{n-1}\za$ is a levelwise equivalence of categories.
The proof that, for each $s > 2$ the map
\begin{equation*}
P_s\rw \pro{P_1}{\di{n-1}\p{n-1}P_0}{s}
\end{equation*}
is a levelwise equivalence of categories is similar.
\bk
b) We have $\q{n}P=Z\in\catwg{n}$, $\p{n}P\in\tawg{n-1}$ since $P\in \tawg{n}$ by Proposition \ref{pro-spec-plbk-pscatwg}. We now show that $\p{n}P$ satisfies the hypotheses of Lemma \ref{lem-x-in-tawg-x-in-catwg}, which then shows that $\p{n}P\in\catwg{n-1}$.
We have the pullback in $\funcat{n-2}{\Cat}$ for each $k\geq 0$,
\begin{equation*}
\xymatrix{
P_k \ar[rr] \ar[d] && X_k \ar[d]\\
Z_k \ar[rr] && \di{n-1}\p{n-1}X_k
}
\end{equation*}
which\, satisfies\, the\, induction\, hypothesis. \,Therefore, \,by \,induction,\, $\p{n-1}P_k\in\catwg{n-2}$ which is hypothesis a) of Lemma \ref{lem-x-in-tawg-x-in-catwg}. Since, by part a), the map
\begin{equation*}
t_2:P_2\rw \tens{P_1}{\di{n-1}\p{n-1}P_0}
\end{equation*}
is a levelwise equivalence of categories, it induces an isomorphism
\begin{equation*}
\p{n-1}P_2=\p{n-1}(\tens{P_1}{\di{n-1}\p{n-1}P_0})=\tens{\p{n-1}P_1}{\p{n-1}P_0}
\end{equation*}
and similarly all the other Segal maps of $\p{n}P$ are isomorphisms. This proves hypothesis b) of Lemma \ref{lem-x-in-tawg-x-in-catwg} for $\p{n}X$.
To prove hypothesis c) of Lemma \ref{lem-x-in-tawg-x-in-catwg}, we first note that
\begin{equation}\label{eq0-the-repl-obj-1}
\tens{P_1}{\di{j,n-1}\p{j,n-1}P_0}\in \tawg{n-1}
\end{equation}
where we denote
\begin{equation*}
\p{j,n-1}=\p{j}...\p{n-1}, \qquad \di{j,n-1}=\di{n-1}...\di{j}\;.
\end{equation*}
In fact, since $\p{n-1}P_0=Z_0$ we have
\begin{equation}\label{eq1-the-repl-obj-1}
\begin{split}
& \tens{P_1}{\di{j,n-1}\p{j,n-1}P_0}= \\
& \resizebox{1.0\hsize}{!}{$ \;= \tens{(\di{n-1}Z_1 \tiund{\di{n-1}\q{n-1}X_1} X_1)}{\di{j,n-1}\p{j,n-1}Z_0}=$}\\
& = \{\tens{\di{n-1}Z_1}{\di{j,n-1}\p{j,n-1}Z_0}\}\tiund{\di{n-1}\q{n-1}(X_1\times X_1)} (X_1\times X_1)=\\
& = \di{n-1}(\tens{Z_1}{\di{j,n-2}\p{j,n-1}Z_0})\tiund{\di{n-1}\q{n-1}(X_1\times X_1)} (X_1\times X_1)\;.
\end{split}
\end{equation}
By Remark \ref{rem-lev-wg-ncat}, $\tens{Z_1}{\di{j,n-2}\p{j,n-2}Z_0}\in\catwg{n-1}$; by Proposition \ref{pro-spec-plbk-pscatwg} and \eqref{eq1-the-repl-obj-1} we conclude that \eqref{eq0-the-repl-obj-1} holds. It follows that
\begin{equation*}
\begin{split}
& \p{j+1,n-1}(\tens{P_1}{\di{j,n-1}\p{j,n-1}P_0})= \\
=\ & \tens{\p{j+1,n-1}P_1}{\di{j}\p{j}\p{j+1,n-1}P_0}\in\tawg{j}\;.
\end{split}
\end{equation*}
Since $P\in\tawg{n}$, $\p{j+1,n}P\in\tawg{j}$ so that the induced Segal map
\begin{equation*}
\p{j+1,n-1}P_2\rw \tens{\p{j+1,n-1}P_1}{(\p{j+1,n-1}P_0)^d}
\end{equation*}
are $\equ{j-1}$s. From above, this map factorizes as composite of maps in $\tawg{j}$
\begin{equation*}
\begin{split}
& \p{j+1,n-1}P_2 \xrw{\;\za\;} \tens{\p{j+1,n-1}P_1}{\di{j}\p{j}\p{j+1,n-1}P_0} \xrw{\;\zb\;}\\
\xrw{\;\zb\;}\ &\tens{\p{j+1,n-1}P_1}{(\p{j+1,n-1}P_0)^d}\;.
\end{split}
\end{equation*}
It is not difficult to check that this satisfies the hypotheses of Lemma \ref{lem-jn-alpha}; thus we conclude that $J_j\za$ is a levelwise equivalence of categories. Therefore $\p{j}\za$ is an isomorphism, that is
\begin{equation*}
\begin{split}
& \p{j,n-1}P_2 \cong \p{j}(\tens{\p{j+1,n-1}P_1}{\di{j}\p{j}\p{j+1,n-1}P_0})= \\
=\ & \tens{\p{j,n-1}P_1}{\p{j,n-1}P_0}\;.
\end{split}
\end{equation*}
Similarly one shows that all the other Segal maps for $\p{j+1,n}P$ are isomorphisms, which proves condition c) in Lemma \ref{lem-x-in-tawg-x-in-catwg} for $\p{n}P$. Thus by Lemma \ref{lem-x-in-tawg-x-in-catwg} we conclude that $\p{n}P\in\catwg{n}$, proving b).
\bk
c) By definition of $\lnta{n}{n}$, this follows from a) and b).
\bk
d) Consider the commuting diagram in $\tawg{n}$
\begin{equation*}
\xymatrix@R=35pt @C=40pt{
\di{n}Z \ar^{\di{n}r}[r] \ar_{\di{n}r}[d] & \di{n}\q{n}X \ar@{=}^{}[d] & X \ar^{}[l] \ar@{=}^{}[d] \\
\di{n}\q{n}X \ar@{=}^{}[r] & \di{n}\q{n}X & X \ar^{}[l]
}
\end{equation*}
By hypothesis, $\di{n}r$ is a $\nequ$. Thus by Proposition \ref{pro-spec-plbk-pscatwg} the induced map of pullbacks
\begin{equation*}
P=\di{n}Z \tiund{\di{n}\q{n}X} X \xrw{w} \di{n}\q{n}X \tiund{\di{n}\q{n}X} X =X
\end{equation*}
is a $\nequ$, as required.
\bk
e) By c) and by Lemma \ref{lem-lev-wg-pscat}, to show that $P\in\lta{n}$ it is enough to show that, for each $\seqc{k}{1}{n-s}\in\Dop$, $P_{\seqc{k}{1}{n-s}}\in\lnta{s}{s}$, $1< s \leq n-1$. We have a pullback in $\funcat{n-1} {\Cat}$
\begin{equation}\label{eq2-the-repl-obj-1}
\xymatrix@R=35pt @C=45pt{
P_{\seq{k}{1}{n-s}} \ar^{w_{\seq{k}{1}{n-s}}}[r] \ar^{}[d] & X_{\seq{k}{1}{n-s}} \ar^{}[d] \\
\di{s} Z_{\seq{k}{1}{n-s}} \ar_{\di{s}r_{\seq{k}{1}{n-s}}}[r] & \di{s}\q{s} X_{\seq{k}{1}{n-s}}
}
\end{equation}
where $X_{\seq{k}{1}{n-s}}\in \tawg{s}$ (since $X\in\tawg{n}$) and $Z_{\seq{k}{1}{n-s}}\in \catwg{s-1}$ (since $Z\in\catwg{n-1}$).
Thus \eqref{eq2-the-repl-obj-1} satisfies the hypotheses of the theorem and we conclude from c) that $P_{\seq{k}{1}{n-s}}\in\lnta{s}{s}$, as required.
\end{proof}
\medskip
\section{Ridigifying weakly globular Tamsamani ${n}$-categories}\label{sec-wg-tam-to-psefun}
In this section we show the main result of the paper, Theorem \ref{the-funct-Qn}, establishing the existence of a rigidification functor
\begin{equation*}
Q_n:\tawg{n}\rw \catwg{n}
\end{equation*}
replacing $X\in\tawg{n}$ with an $n$-equivalent object $Q_n X$.
The construction of the functor $Q_n$ uses three main ingredients: the first is the approximation up to $n$-equivalence of any object of $\tawg{n}$ with an object of $\lta{n}$ established in the previous section. The second ingredient is a functor $Tr_n$ from the category $\lta{n}$ to the category of Segalic pseudo-functors, which we establish in this section in Theorem \ref{the-XXXX}. The third ingredient is the functor from Segalic pseudo-functors to weakly globular $n$-fold categories from \cite{Pa2} recalled in Theorem \ref{the-strict-funct}.
The construction of the functor $Tr_n$ is based on a general categorical technique which we recalled in Lemma \ref{lem-PP}. We can apply this technique to objects of the category $\lta{n}$ thanks to a property of the latter, which we establish in Lemma \ref{lem-maps-nu-eqcat}.
\bigskip
We first fix a notation for the induced Segal maps of categories for an object $X$ of $\tawg{n}$ when all but one simplicial directions in $J_nX$ are fixed.
\begin{notation}\label{not-ind-seg-map}
Let $X\in\tawg{n}$, $\uk=(k_1,\ldots,k_{n-1})\in\dop{n-1}$, $1\leq i\leq n-1$. Then there is $X^i_{\uk}\in\funcat{}{\Cat}$ with
\begin{equation*}
(X^i_{\uk})_r=X_{\uk(r,i)}=X_{k_1...k_{i-1}r k_{i+1}...k_{n-1}}
\end{equation*}
so that $(X^i_{\uk})_{k_i}=X_{\uk}$. Since $X_{k_1,\ldots,k_{i-1}}\in\tawg{n-i+1}$, $X_{k_1,\ldots,k_{i-1}0}\in\cathd{n-i}$ and thus by \cite[Lemma 3.1]{Pa1}
\begin{equation*}
X_{k_1...k_{i-1} 0 k_{i+1}...k_{n-1}}= X_{\uk(0,i)}\in\cathd\;.
\end{equation*}
We therefore obtain induced Segal maps in $\Cat$ for all $k_i\geq 2$.
\begin{equation}\label{eq1-not-ind-seg-map}
\nu(\uk,i):X_{\uk}\rw \pro{X_{\uk(1,i)}}{X^d_{\uk(0,i)}}{k_i}\;.
\end{equation}
\end{notation}
\begin{lemma}\label{lem-maps-nu-eqcat}
Let $X\in\lta{n}$; then for each $\uk\in\dop{n-1}$, $1\leq i\leq n-1$ and $k_i\geq 2$ the maps $ \nu(\uk,i)$ as in \eqref{eq1-not-ind-seg-map} are equivalences of categories.
\end{lemma}
\begin{proof}
By induction on $n$. It holds when $n=2$ since $\lta{2}=\tawg{2}$ and, for $k\geq 2$, $\nu(k,1)$ is the induced Segal map
\begin{equation*}
X_k\rw \pro{X_1}{X_0^d}{k}
\end{equation*}
which is an equivalence of categories by definition of $\tawg{2}$.
Suppose the lemma holds for $(n-1)$ and let $X$ be as in the hypothesis. Consider first the case $1<i\leq n-1$. Since $X\in\lta{n}$, by definition $X_{k_1}\in\lta{n-1}$. Denoting $\ur=(k_2...k_{n-1})$ we have
\begin{equation}\label{eq2-not-ind-seg-map}
\begin{split}
& X_{\uk}=(X_{k_1})_{\ur} \\
& X_{\uk(1,i)}=(X_{k_1})_{\ur(1,i-1)}\qquad X_{\uk(0,i)}=(X_{k_1})_{\ur(0,i-1)}\;.
\end{split}
\end{equation}
The induction hypothesis applied to $Y=X_{k_1}$ implies the equivalence of categories for each $r_{i-1}\geq 2$
\begin{equation*}
Y_{\ur}\simeq \pro{Y_{\ur(1,i-1)}}{Y^d_{\ur(0,i-1)}}{r_{i-1}}\;.
\end{equation*}
By \eqref{eq2-not-ind-seg-map} this means that $\nu(\uk,i)$ is an equivalence of categories for all $1<i\leq n-1$.
\nid Consider the case $i=1$. By definition of $\lta{n}$ the map in $\funcat{n-2}{\Cat}$
\begin{equation*}
X_{k_1}\rw\pro{X_1}{\di{n-1}\p{n-1}X_0}{k_1}
\end{equation*}
is a levelwise equivalence of categories for each $k_i\geq 2$. Therefore, for each $\ur=(k_2...k_{n-1})$ there is an equivalence of categories.
\begin{equation*}
(X_{k_1})_{\ur}\simeq \pro{(X_1)_{\ur}}{d p (X_0)_{\ur}}{k_1}\;.
\end{equation*}
Since $(X_1)_{\ur}=X_{\uk(1,1)}$ and $dp(X_0)_{\ur}=X^d_{\uk(0,1)}$ we obtain the equivalence of categories
\begin{equation*}
X_{\uk}\simeq \pro{X_{\uk(1,1)}}{X^d_{\uk(0,1)}}{k_i}\;.
\end{equation*}
In conclusion $\nu(\uk,i)$ is an equivalence of categories for all $1\leq i\leq n-1$.
\end{proof}
\begin{theorem}\label{the-XXXX}
There is a functor
\begin{equation*}
Tr_{n}: \lta{n} \rw \segpsc{n-1}{\Cat}
\end{equation*}
together with a pseudo-natural transformation $t_n (X):Tr_{n}X\rw X$ for each $X\in\lta{n}$ which is a levelwise equivalence of categories.
\end{theorem}
\begin{proof}
Given $X\in\lta{n}$ define the diagram $Tr_{n} X$ in $[ob(\dop{n-1}),\Cat]$ as follows: for each $\uk\in\dop{n-1}$ and $1\leq i\leq n-1$
\begin{itemize}
\item [i)] If $k_j=0$ for some $1\leq j\leq n-1$
\begin{equation*}
(Tr_{n} X)_{\uk}=X^d_{\uk}\;.
\end{equation*}
\item [ii)] If $k_j\neq 0$ for all $1\leq j\leq n-1$ and $k_i=1$,
\begin{equation*}
(Tr_{n} X)_{\uk}=X_{\uk(1,i)}=X_{\uk}\;.
\end{equation*}
\item [iii)] If $k_j\neq 0$ for all $1\leq j\leq n-1$ and $k_i>1$,
\begin{equation*}
(Tr_{n} X)_{\uk}=\pro{X_{\uk(1,i)}}{X^d_{\uk(0,i)}}{k_i} \;.
\end{equation*}
\end{itemize}
We claim that, for all $\uk\in \dop{n-1}$ there is an equivalence of categories
\begin{equation*}
(Tr_{n} X)_{\uk}=X_{\uk}\;.
\end{equation*}
In fact, in case i), $X_{\uk}\in\cathd{}$ thus there is an equivalence of categories
\begin{equation*}
(Tr_{n} X)_{\uk}= X^d_{\uk}\simeq X_{\uk}\;.
\end{equation*}
In case ii) we have
\begin{equation*}
(Tr_{n} X)_{\uk}=X_{\uk(1,i)}=X_{\uk}\;.
\end{equation*}
In case iii) by Lemma \ref{lem-maps-nu-eqcat} we have an equivalence of categories
\begin{equation*}
(Tr_{n} X)_{\uk}=\pro{X_{\uk(1,i)}}{X^d_{\uk(0,i)}}{k_i}\simeq X_{\uk} \;.
\end{equation*}
This proves the claim.
We can therefore apply Lemma \ref{lem-PP} with $\clC=\dop{n-1}$ and conclude that $Tr_{n}X$ lifts to a pseudo-functor
\begin{equation*}
Tr_{n}X \in \psc{n-1}{\Cat}
\end{equation*}
with $(Tr_{n} X)_{\uk}$ as in i), ii) and iii). We check that $Tr_{n}X$ is a Segalic pseudo-functor.
Conditions a) in Definition \ref{def-seg-ps-fun} is satisfied by construction. As for condition b), let $\uk\in\dop{n-1}$ be such that $k_j=0$ for some $1\leq j\leq n-1$ and let $k_i\geq 2$ for $1\leq i\leq n$. Since by Lemma \ref{lem-maps-nu-eqcat} there is an equivalence of categories
\begin{equation*}
X_{\uk}\simeq \pro{X_{\uk(1,i)}}{X^d_{\uk(0,i)}}{k_i}
\end{equation*}
and since $X_{\uk}\in \cathd{}$, $X_{\uk(1,i)}\in \cathd{}$, there is an isomorphism
\begin{equation*}
X^d_{\uk}\cong dp( \pro{X_{\uk(1,i)}}{X^d_{\uk(0,i)}}{k_i})\cong \pro{X^d_{\uk(1,i)}}{X^d_{\uk(0,i)}}{k_i}
\end{equation*}
which is the same as
\begin{equation*}
(Tr_{n}X )_{\uk}\cong\pro{(Tr_{n}X )_{\uk(1,i)}}{(Tr_{n}X )_{\uk(0,i)}}{k_i}\;.
\end{equation*}
Thus condition b) in Definition \ref{def-seg-ps-fun} holds in this case.
Suppose $k_j\neq 0$ for all $1\leq j\leq n-1$, $k_i\geq 2$. Then
\begin{equation*}
\begin{split}
& (Tr_{n}X )_{\uk}=\pro{X _{\uk(1,i)}}{X^d_{\uk(0,i)}}{k_i}\cong \\
\cong \, & \pro{(Tr_{n}X )_{\uk(1,i)}}{(Tr_{n}X )_{\uk(0,i)}}{k_i} \;.
\end{split}
\end{equation*}
So condition b) in Definition \ref{def-seg-ps-fun} also holds in this case. As for condition c), we note that the equivalence of categories $(Tr_{n} X)_{\uk}\simeq X_{\uk}$ for each $\uk\in\dop{n-1}$ implies the isomorphism
\begin{equation*}
p (Tr_{n} X)_{\uk} \cong p X_{\uk} =(\p{n}X)_{\uk}\;.
\end{equation*}
Since $X\in\lta{n}$, $\p{n}X\in\catwg{n-1}$ hence $\p{n}Tr_{n} X\in\catwg{n-1}$. By definition this means that $Tr_{n} X\in\segpsc{n-1}{\Cat}$. By Lemma \ref{lem-PP} there is a morphism in $\psc{n-1}{\Cat}$, \, $t_n(X):Tr_{n} X\rw X$ which is a levelwise equivalence of categories.
\end{proof}
We now prove the main result of this paper, which is the existence of a rigidification functor from $\tawg{n}$ to $\catwg{n}$ replacing any object of $\tawg{n}$ with a suitably equivalent object in $\catwg{n}$.
\begin{theorem}\label{the-funct-Qn}
There is a functor
\begin{equation*}
Q_n:\tawg{n} \rw \catwg{n}
\end{equation*}
and for each $X\in\tawg{n}$ a morphism in $\tawg{n}$ $s_n(X):Q_n X\rw X$, natural in $X$, such that $(s_n(X))_{k}$ is a $(n-1)$-equivalence for all $k\geq0$. In particular, $s_{n}(X)$ is an $n$-equivalence.
\end{theorem}
\begin{proof}
By induction on $n$. When $n=2$, let $Q_2$ be the composite
\begin{equation*}
Q_2: \tawg{2}\xrw{Tr_{2}} \segpsc{}{\Cat} \xrw{\St} \catwg{2}
\end{equation*}
where $Tr_{2}$ is as in Theorem \ref{the-XXXX}. By Theorem \ref{the-strict-funct}, $Q_2X \in \catwg{2}$. Recall \cite{Lack} that the strictification functor
\begin{equation*}
\St:\psc{}{\Cat} \rw \funcat{}{\Cat}
\end{equation*}
is left adjoint to the inclusion
\begin{equation*}
J: \funcat{}{\Cat} \rw \psc{}{\Cat}
\end{equation*}
and that the components of the unit are equivalences in $\psc{}{\Cat}$. By Theorem \ref{the-XXXX} there is a morphism in $\psc{}{\Cat}$
\begin{equation*}
t_2(X):Tr_{2}X \rw JX \;.
\end{equation*}
By adjunction this corresponds to a morphism in $\funcat{}{\Cat}$
\begin{equation*}
Q_2 X=\St Tr_{2} X\xrw{s_2(X)} X
\end{equation*}
making the following diagram commute
\begin{equation*}
\xymatrix@C=50pt@R=50pt{
Tr_{2}X \ar^(0.4){\eta}[r] \ar_{t_2(X)}[dr] & J\St Tr_{2}X
\ar^{J s_2(X)}[d] \\
& JX
}
\end{equation*}
Since $\eta$ and $t_2(X)$ are levelwise equivalences of categories, such is $s_2(X)$.
Suppose, inductively, that we defined $Q_{n-1}$. Define the functor
\begin{equation*}
P_n:\tawg{n}\rw\lta{n}
\end{equation*}
as follows. Given $X\in\tawg{n}$, consider the pullback in $\funcat{n-1}{\Cat}$
\begin{equation*}
\xymatrix@C=70pt@R=30pt{
P_n X \ar^{w(X)}[r] \ar[d] & X \ar^{\zg_{n}}[d] \\
\di{n}Q_{n-1}\q{n}X \ar_{\di{n}s_{n-1}(\q{n}X)}[r] & \di{n}\q{n}X
}
\end{equation*}
By Theorem \ref{the-repl-obj-1}, $P_n X\in\lta{n}$. Define
\begin{equation*}
Q_n X=\St Tr_{n} P_n X.
\end{equation*}
By Theorem \ref{the-strict-funct}, $Q_nX \in \catwg{n}$. Let $s_n(X):Q_n X\rw X$ be the composite
\begin{equation*}
s_n(X): Q_n X \xrw{h_n(P_n X)} P_n X \xrw{w(X)} X\;.
\end{equation*}
where the morphism in $\funcat{n-1}{\Cat}$
\begin{equation*}
Q_n X= \St Tr_{n}P_n X \xrw{h_n(P_n X)} P_n X
\end{equation*}
corrersponds by adjunction to the morphism in $\psc{n-1}{\Cat}$
\begin{equation*}
Tr_{n}P_n X \xrw{t_n(P_n X)} J P_n X
\end{equation*}
(where $t_n(P_n X)$ is as in Theorem \ref{the-XXXX}) such that the following diagram commutes
\begin{equation*}
\xymatrix@C=50pt@R=50pt{
Tr_{n}P_n X \ar^(0.35){\eta}[r] \ar_{t_n(P_n X)}[dr] & J \St Tr_{n}P_n
X= J Q_n X \ar^{J h_n(P_n X)}[d]\\
& J P_n X
}
\end{equation*}
We need to show that $(s_n(X))_{k}$ is an $(n-1)$-equivalence. Since $\eta$ and $t_n(P_n X)$ are levelwise equivalence of categories, such is $h_n(P_n X)$ so in particular $(h_n(P_n X))_{k}$ is a levelwise equivalence of categories, and thus is a $(n-1)$-equivalence (see Remark \ref{rem-local-equiv}).
Since pullbacks in $\funcat{n-1}{\Cat}$ are computed pointwise, there is a pullback in $\funcat{n-2}{\Cat}$
\begin{equation*}
\xymatrix@C=90pt@R=40pt{
(P_n X)_k \ar^{(w(X))_k}[r] \ar[d] & X_k \ar[d]\\
\di{n-1}(Q_{n-1}\q{n}X)_k \ar_(0.55){\di{n-1}(s_{n-1}(\q{n}X))_k}[r] & \di{n-1}\q{n-1}X_k
}
\end{equation*}
where $X_k\in\tawg{n-1}$ (since $X\in\tawg{n}$) and $(Q_{n-1}\q{n}X)_k\in\catwg{n-2}$ (since $Q_{n-1}\q{n}X\in\catwg{n-1}$) and, by induction hypothesis, $(s_{n-1}(\q{n}X))_k$ is a $(n-2)$-equivalence. It follows by Theorem \ref{the-repl-obj-1} that $(w(X))_k$ is a $(n-1)$-equivalence.
In conclusion both $(h_n(P_n X))_k$ and $(w(X))_k$ are $(n-1)$-equivalences so by Proposition \ref{pro-n-equiv} such is their composite
\begin{equation*}
(s_n(X))_k : (Q_n X)_k \xrw{(h_n(P_n X))_k} (P_n X)_k \xrw{(w(X))_k} X_k
\end{equation*}
as required. By Lemma \ref{lem-flevel-fneq}, it follows that $s_n(X)$ is an $n$-equivalence.
\end{proof}
\begin{corollary}\label{cor-the-funct-Qn}
The functor $Q_n:\tawg{n}\rw \catwg{n}$ induces an equivalence of categories
\begin{equation}\label{eq-cor-the-funct-Qn}
\tawg{n}\bsim^n\; \simeq\; \catwg{n}\bsim^n
\end{equation}
after localization with respect to the $n$-equivalences.
\end{corollary}
\begin{proof}
Let $i:\catwg{n}\hookrightarrow \tawg{n}$ denote the embedding of $\catwg{n}$ into $\tawg{n}$. Given $X\in\tawg{n}$, by Theorem \ref{the-funct-Qn} there is an $n$-equivalence in $\tawg{n}$ $i Q_n X\rw X$, therefore $i Q_n X\cong X$ in $\tawg{n}\bsim^n$.
Let $Y\in\catwg{n}$; then $i Y\in\tawg{n}$ so by Theorem \ref{the-funct-Qn} there is an $n$-equivalence in $\tawg{n}$ $i Q_n i Y \rw i Y$. Since $i$ is fully faithful, $ Q_n i Y \rw Y$ is an $n$-equivalence in $\catwg{n}$. It follows that $Q_n i Y \cong Y$ in $\catwg{n}\bsim^n$.
In conclusion $Q_n$ and $i$ induce the equivalence of categories \eqref{eq-cor-the-funct-Qn}.
\end{proof}
| ba9101e1304eee64cc897962b80726d2651909ea | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{INTRODUCTION}
\label{sec:intro}
A period of accelerated exponential expansion in the early Universe,
while a radical extrapolation from understood physics, naturally
solves the flatness, horizon, and monopole problems of standard
cosmology\cite{planck_p15xx_15}. This inflationary period also
explains the origin of structure by stretching quantum fluctuations to
macroscopic scales. Inflation predicts perturbations to the metric in
both scalars (density waves) and tensors (gravitational waves). Both
types of perturbations affect the polarization of the cosmic microwave
background (CMB) at last scattering\cite{polnarev85}: scalars can only
generate a gradient-type ($E$-mode) pattern, while tensors can also
generate a curl-type ($B$-mode) pattern. Measurement of $B$-mode\ power
in the CMB at degree angular scales in excess of the expectation from
gravitational lensing would be direct evidence for an inflationary
period, and the amplitude of the signal -- parametrized by $r$, the
tensor-to-scalar ratio -- would indicate the energy scale of
inflation.
Since 2006, the {\sc Bicep}/{\it Keck Array}\ CMB polarization experiments have
been mapping $\sim 1\%$ of low-foreground sky from the Amundsen-Scott
South Pole Station. The telescopes are optimized to detect
fluctuations at degree angular scales, and therefore use a compact,
on-axis refracting optical system with the minimum aperture necessary
to resolve the $\ell \sim 100$ peak in the inflationary gravitational
wave $B$-mode\ spectrum. In 2014 we reported a detection of
$B$-mode\ power at degree angular scales at 150 GHz by
{\sc Bicep2}\cite{biceptwo14_I}, which was quickly confirmed by data from
the {\it Keck Array}\cite{biceptwo15_V}. Multifrequency maps are required
to separate Galactic foreground emission (thermal dust at high
frequencies and synchrotron at low frequencies) from the CMB signal.
Using all available {\it Planck}\ and {\sc Wmap}\ maps and the latest
{\it Keck Array}\ maps at both 95 and 150 GHz, we find that the excess at
150 GHz is consistent with dust emission, and have constrained the
tensor-to-scalar ratio to $r < 0.07$ at $95\%$
confidence\cite{biceptwo15_BKP_all,biceptwo16_VI}. To push this limit
further -- or achieve a detection of primordial $B$-mode s -- we have
deployed 220 GHz detectors in the {\it Keck Array}\ to measure dust at high
precision, and have deployed {\sc Bicep3}\cite{wu15} at 95 GHz to probe
deeply at a low-foreground frequency. For the 2016 observing season,
{\sc Bicep3}\ was outfitted with a full complement of 20 detector tiles
(2400 optically coupled detectors).
The {\sc Bicep}/{\it Keck Array}\ telescopes measure polarization by differencing
co-located, orthogonally polarized detector pairs\cite{biceptwo14_II}.
Any mismatch in beam shape between the two polarizations will leak
some of the bright CMB temperature signal into polarization.
Temperature-to-polarization leakage is the most prominent instrumental
systematic effect which may cause a false $B$-mode\ signal and must be
precisely characterized\cite{biceptwo15_IV}. We have developed an
analysis technique called deprojection\cite{biceptwo15_III} to
marginalize over any false polarization signal which would arise from
the coupling of the CMB temperature sky to a second-order expansion of
the measured differential beam; however, the contribution from
higher-order beam mismatch remains. In the {\sc Bicep2}\ final results,
using high-fidelity far field beam maps we constrained the
undeprojected residual contribution to the final $B$-mode\ spectrum to
an equivalent $r < 3.0 \times 10^{-3}$.
In these proceedings, we present an analysis of the far field beam
response of the {\sc Bicep3}\ detectors observing in the 2016 season.
In February and March of 2016, we obtained far field beam maps of all
{\sc Bicep3}\ detectors \emph{in situ} at the South Pole using a
chopped thermal source in the far field of the telescope. These
measurements serve several purposes: the array-averaged beam profile
is used to smooth {\tt Healpix}\ maps for simulations, the measured beam
shapes offer a cross-check that the modes removed by deprojection
correspond to actual beam mismatch and also allow for direct
subtraction of the deprojection templates, and the beam maps are used
in dedicated simulations to explicitly predict the residual
$B$-mode\ power after deprojection.
In Section~\ref{sect:optics} we review the optical design of
{\sc Bicep3}. In Section~\ref{sect:ffbm} we describe the far field
beam measurement setup and the new chopped source constructed in 2016.
In Section~\ref{sect:beamparam} we present two-dimensional Gaussian
fits to each detector and characterize beamwidth, ellipticity,
differential pointing, differential beamwidth, and differential
ellipticity across the entire array. In Section~\ref{sect:composites}
we coadd all beam measurements to obtain high signal-to-noise
composite maps of individual beams, and use these beams to calculate
the array-averaged beam response. In Section~\ref{sect:deprojection}
we discuss the measured beam mismatch in the context of deprojection
and the residual beam systematics levels achieved by {\sc Bicep2}\ and
{\it Keck Array}. Three companion papers -- a {\sc Bicep3}\ status
update\cite{grayson16}, {\sc Bicep3}\ detector performance\cite{hui16},
and measurements of {\it Keck Array}\ detector polarization angles with a
dielectric sheet calibrator\cite{bullock16} -- are also presented at
this conference.
\section{OPTICAL DESIGN}
\label{sect:optics}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=0.9\columnwidth]{b3keck_optics.pdf}
\caption{\label{fig:optics} Optical schematic of
{\sc Bicep2}/{\it Keck Array}\ (left) and {\sc Bicep3}\ (right) with optical
elements labeled, to scale. {\sc Bicep3}'s larger aperture (520 mm
vs 264 mm) and faster optics result in optical throughput similar to
that of five {\it Keck Array}\ receivers combined.}
\end{center}
\end{figure}
Like {\sc Bicep1}, {\sc Bicep2}\, and {\it Keck Array}, {\sc Bicep3}\ is a compact,
two-lens on-axis telecentric refracting telescope\cite{ahmed14}.
However, the aperture is twice as large as that of its predecessors
($520$ mm vs $264$ mm), the field of view is larger ($27^{\circ}$ vs
$15^{\circ}$), and the optics are faster ($f/1.7$ vs $f/2.4$). For
observations at a wavelength of $\lambda = 3 \mbox{ mm}$, the far
field is at roughly $2D^2 / \lambda = 180 \mbox{ m}$.
Figure~\ref{fig:optics} shows a schematic of the optical system
compared to that of {\sc Bicep2}/{\it Keck Array}. Mounted directly on the
detector modules are metal-mesh low pass filters with a cutoff at $4
\mbox{ cm}^{-1}$ (120 GHz) to block out-of-band power. Two lenses,
made of 99.6\% pure alumina ceramic, perform the imaging; a cold Lyot
stop skyward of the objective lens defines the aperture. The
telescope housing, lenses, and aperture stop are all cooled to 4 K.
The alumina lenses are anti-reflection coated with a mix of Stycast
1090 and 2850 epoxy tuned to the correct index of refraction ($n=1.75$
for alumina at $n=3.1$), machined to the correct thickness and laser
diced into patches to relieve mechanical stress from differential
thermal contraction.
The $670 \mbox{ mm}$ aperture high-density polyethylene vacuum window
is $31.75 \mbox{ mm}$ thick and anti-reflection coated with Teadit
24RGD. The large aperture allows over $100 \mbox{ W}$ of infrared
power to enter the cryostat, and the purely absorptive filtering used
in {\sc Bicep2}\ and {\it Keck Array}\ would overload the Cryomech PT415 pulse
tube; we therefore reject a majority of the loading with a stack of 10
metal-mesh IR-reflective filters mounted just below the receiver
window at ambient temperature. Below these are flat absorptive
filters: a $10 \mbox{ mm}$ thick alumina filter at 50 K above the
objective lens, a $5 \mbox{ mm}$ thick nylon filter at 4 K above the
field lens, and $9.5 \mbox{ mm}$ thick nylon filter at 4 K between the
field lens and focal plane. There is an additional reflective metal
mesh filter below the alumina filter at 50 K. The resulting IR load
allows continuous operation of the pulse tube at $< 4$ K and $> 48$
hour hold time of the sub-Kelvin refrigerator.
To terminate far sidelobes in CMB observations, we install a co-moving
absorptive forebaffle coated with a combination of Eccosorb
HR-25/AN-75 and weatherproofed with Volara foam. The forebaffle is
$128.9 \mbox{ cm}$ tall, $141 \mbox{ cm}$ in diameter, and intercepts
radiation $> 14^{\circ}$ from the telescope boresight. The telescope
is surrounded by a stationary reflective ground shield which redirects
off-axis rays to cold sky; the geometry requires that a ray must
diffract twice -- around the forebaffle and ground shield -- before it
hits the ground.
\section{FAR FIELD BEAM MEASUREMENT SETUP}
\label{sect:ffbm}
The far field beam mapping setup for {\sc Bicep3}\ is very similar to
that used for {\sc Bicep2}\ and the {\it Keck Array}; see the Beams
paper\cite{biceptwo15_IV} for additional details.
\subsection{Flat Mirror and Thermal Chopped Microwave Source}
\label{ss:mirrorchopper}
Since the telescope is contained within its ground shield and can only
tip down to $\sim 40^{\circ}$ from zenith, we use a flat mirror to
redirect the beams over the ground shield and towards the thermal
chopper (Figure~\ref{fig:mirrorchopper}, left panel). The 1.7 x 2.5 m
aluminum honeycomb mirror is flat to 0.2 mm across its surface and is
mounted at a $45^{\circ}$ angle, such that when the telescope is
pointed at zenith, the beams are redirected towards the horizon (the
source is $\sim 2^{\circ}$ above the horizon). The forebaffle is
removed to provide room for the mirror.
\begin{figure}[h!]
\begin{center}
\includegraphics[width=1.0\columnwidth]{mirrorchopper.pdf}
\caption{\label{fig:mirrorchopper} Left: The aluminum honeycomb mirror
mounted on {\sc Bicep3}\ at the Dark Sector Laboratory in the far
field beam mapping configuration. The source is visible in the
distance. Right: The chopped thermal source mounted on a 40 ft tall
mast at the Martin A. Pomerantz Observatory. The {\it Keck Array}\ flat
mirror is also visible. Right inset: the chopped source in lab.
The 24'' aperture is sealed with white HD-30 Zotefoam. The
reflective housing contains a blackened rotating blade, behind which
is a flat mirror redirecting up to sky.}
\end{center}
\end{figure}
For the 2016 deployment season, we built two identical chopped thermal
microwave sources so that {\sc Bicep3}\ and {\it Keck Array}\ could take far
field beam map data simultaneously. These choppers consist of a
rotating blade, an enclosure surrounding the blade, and a motor drive
assembly to spin the blade. The blade is a composite material made
from carbon fiber, low density foam, and honeycomb Nomex, coated with
Eccosorb HR-10. The enclosure is made of a similar carbon fiber
composite, coated with aluminum foil so that rays which do not
terminate on the source aperture are reflected to cold sky (see the
right inset of Figure~\ref{fig:mirrorchopper}). The source aperture
is 24'' in diameter, behind which is a flat mirror directed to zenith,
so that the source chops between ambient ($\sim 260$ K) and sky ($\sim
10$ K). The entire assembly weighs 55 pounds. The right panel of
Figure~\ref{fig:mirrorchopper} shows the chopper mounted on a 40 ft
high mast on the Martin A. Pomerantz Observatory, 211 meters away from
{\sc Bicep3}. The chopper can spin at a maximum of 9 Hz (18 Hz chop
rate due to the two blades); for beam mapping in 2016 we operated at a
14 Hz chop rate. The aperture is sealed with HD-30
Zotefoam\footnote{HD-30 is also used for the {\it Keck Array}\ vacuum window
and has $< 2\%$ transmission loss in our band.} to prevent air
turbulence from slowing down the blade. Compared to previous chopped
thermal microwave sources deployed at South Pole, the 24'' diameter
chopper offers higher signal due to the larger aperture and lower
noise due to the faster chop rate.
\subsection{2016 Data Set}
\label{ss:dataset}
The far field beam maps presented in these proceedings were taken in
February and March of 2016. Due to the comparatively high loading of
the $\sim 250$ K source, beam maps are taken with the detectors on the
aluminum superconducting transition\footnote{Our transition edge
sensor (TES) detectors contain superconductors with different
transition temperatures suitable for various loading conditions:
titanium for CMB observations and aluminum for the higher loading in
calibration measurements.}. In this higher-loading configuration
the sub-Kelvin refrigerator lasts for 12 hours. We scan across the
source in $30^{\circ}$ azimuth rasters at $0.8^{\circ}$ per second,
increasing the elevation by $0.05^{\circ}$ in between each scan. The
scan speed was chosen so that at the fast detector sample rate (150
Hz) and source chop rate (14 Hz), multiple chop cycles would be
recorded across the $0.167^{\circ}$ beam. The full field of view of
the instrument is therefore covered in $\sim 8$ hours. To fill in the
gap before the next fridge cycle, we additionally run ``half''
schedules which only cover the top, middle, or bottom of the focal
plane. Schedules are evenly divided among five boresight angles:
$0^{\circ}$, $45^{\circ}$, $90^{\circ}$, $135^{\circ}$, and
$315^{\circ}$. Consistency of beam maps across boresight angles
offers a robust check on the repeatability of our beam measurements.
Including both full and half focal plane configurations, we took a
total of 77 beam mapping schedules. Once the data are taken, we
demodulate the timestream data using a reference signal from the
chopper and bin into maps with $0.1^{\circ}$ pixelization. The
reflection off the mirror is handled with a pointing model that
describes both the mount and the mirror.
\section{GAUSSIAN BEAM PARAMETERS}
\label{sect:beamparam}
\subsection{Coordinate System and 2D Gaussian Fits}
\label{ss:coord}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=0.75\columnwidth]{fpu_bicep3_2016.pdf}
\caption{\label{fig:fpu} Instantaneous projection of {\sc Bicep3}\ beam
centers on the sky; non-functioning detectors are not plotted. Tile
numbers are indicated in black, and the $A$ detector polarization
direction is indicated with arrows on each tile (the $B$ orientation
is perpendicular to $A$). The tiles corresponding to each MCE are
plotted by color. The $(x',y')$ coordinate system is indicated, in
addition to the directions corresponding to plus ($p$) and cross
($c$) ellipticity. Many MCE3 detectors are non-functioning due to
an electrical open in 7 multiplexing rows. Tile 21 does not exist
because more detectors would require another MCE system.}
\end{center}
\end{figure}
To facilitate comparison of beam parameters across measurements taken
at multiple telescope orientations, we define a coordinate system
similar to that used for {\sc Bicep2}\ and {\it Keck Array}\ beams analysis --
see the Beams paper\cite{biceptwo15_IV} for further details. Each
pixel $P$, consisting of two orthogonally polarized detectors, is at a
location $(r,\theta)$ from the boresight. $r$ is the radial distance
away from the boresight, and $\theta$ is the counterclockwise angle
looking out from the telescope toward the sky from the
$\theta=0^{\circ}$ ray, defined to be the great circle that runs along
tiles 11/10/9 on the focal plane. For each pixel we then define a
local $(x',y')$ coordinate system. The positive $x'$ axis is along
the great circle passing through $P$ which is an angle $-\theta$ from
the $\hat{r}$ direction of the pixel, while the $y'$ axis is along the
great circle $+90^{\circ}$ counterclockwise away from the $x'$ axis.
The $(x',y')$ coordinate system is then projected onto a plane at $P$.
This coordinate system is fixed to the instrument and rotates on the
sky with the boresight rotation angle $K$ and the angle with which the
receiver is clocked with respect to the $K=0$ line, or the drum angle
$K'$; for {\sc Bicep3}, $K'=1.5^{\circ}$. In a simplified sense, to
transform from mount coordinates to detector-centered coordinates, we
flip the elevation axis to account for the mirror and rotate by $K +
K'$ to arrive at $(x',y')$. All focal plane and beam map plots in
these proceedings are rendered as the instantaneous projection of each
beam onto the sky, with the same parity (viewing the projection from
inside the celestial sphere) and orientation (with the $+x'$ axis
pointing up and the $+y'$ axis pointing to the left); this is
consistent with the conventions established in our other publications.
Figure~\ref{fig:fpu} shows the coordinate axes, the location of each
nominally working beam, tile numbers and polarization axes, and the
division of tiles into separate Multi-Channel Electronics (MCE)
systems used to control the multiplexing and readout of the detectors.
Each beam is fit to a two-dimensional elliptical Gaussian model with
six free parameters:
\begin{equation}
B(\bm{x}) = \frac{1}{\Omega} e^{-\frac{1}{2}(\bm{x}-\bm{\mu})^{T} \Sigma^{-1} (\bm{x}-\bm{\mu})},
\end{equation}
where $\bm{x} = (x,y)$ is the location of the beam center, $\bm{\mu} =
(0,0)$ is the origin, $\Omega$ is the normalization, and $\Sigma$ is
the covariance matrix parametrized as
\begin{equation}
\Sigma =
\begin{pmatrix}
\sigma^2 (1 + p) & c \sigma^2 \\
c \sigma^2 & \sigma^2 (1 - p)
\end{pmatrix}.
\end{equation}
$\sigma$ is the beamwidth and $p$ and $c$ are the ellipticities in the
``plus'' and ``cross'' directions respectively, such that an
elliptical Gaussian with a major axis oriented along $x'$ or $y'$
would have a $+p$ or $-p$ ellipticity, and one with a major axis
oriented diagonally would have a $\pm c$ ellipticity (see
Figure~\ref{fig:fpu}). Parametrization with ($\sigma, p, c$), while
equivalent to the more common ($\sigma_{maj}, \sigma_{min}, \phi$), is
more convenient because differential parameters are simple
differences: $d\sigma = \sigma_A - \sigma_B$, $dx = x_A - x_B$, $dy =
y_A - y_B$, $dp = p_A - p_B$, and $dc = c_A - c_B$, where $A$ and $B$
refer to the orthogonally polarized detectors within a pair. These
parameters directly scale the deprojection templates used to remove
temperature-to-polarization leakage in analysis. Note that we do not
use the absolute $(x,y)$ beam locations because of the complicated
mirror model, instead obtaining beam centers from cross-correlation
with the {\it Planck}\ CMB temperature maps.
In each of the 77 beam map schedules, we find the best fit $\Omega$,
$\bm{x}$, $\sigma$, $p$, and $c$ for every beam. Not all beams are
covered in each schedule, and occasionally some detectors are not
working normally even when they are covered. We then remove maps in
which the fit failed, or in which only one of the $A$ or $B$
polarizations were working, so that the same number of measurements
and detectors are used in per-detector and per-pair parameters. After
cuts, the average pair has $\sim 30$ good measurements.
Table~\ref{tab:stats} shows summary statatistics derived from these
measurements. For each detector/pair, we take the median across the
$\sim 30$ measurements as the best estimate of each parameter, and
take half the width of the central 68\% of the distribution of those
measurements as the measurement uncertainty.\footnote{This statistic
is relatively insensitive to outliers, and would equal $1 \sigma$
for a Gaussian distribution of measurements.} The characteristic
uncertainty for an individual measurement -- taken to be the median
measurement uncertainty for all detectors/pairs across the array -- is
shown in the ``Individual Measurement Uncertainty'' column of
Table~\ref{tab:stats}.
To characterize the distribution of parameters, we also find the
median across the focal plane (``FPU Median'' column) and quantify the
variation across the focal plane as half the width of the central 68\%
of the distribution of \emph{best estimates for each detector/pair}
(``FPU Scatter'' column). Note that the ``FPU Scatter'' measures
the spread of best-estimate parameters across the array and is not a
measurement uncertainty.
\begin{table}[t]
\caption{{\sc Bicep3}\ 2016 Beam Parameter Summary Statistics. See
Section~\ref{ss:coord} for a discussion of measurement
uncertainties.}
\label{tab:stats}
\begin{center}
\begin{tabular}{lccc}
\hline
\hline
\rule[-1ex]{0pt}{3.5ex} Parameter & FPU Median & FPU Scatter & Individual Measurement Uncertainty\\
\hline
Beamwidth $\sigma$ (degrees) & 0.167 & 0.002 & 0.002 \\
Ellipticity plus $p$ (+) & 0.010 & 0.021 & 0.026 \\
Ellipticity cross $c$ ($\times$) & -0.004 & 0.016 & 0.026 \\
Diff. X pointing $dx$ (arcmin) & 0.03 & 0.13 & 0.05 \\
Diff. Y pointing $dy$ (arcmin) & -0.12 & 0.17 & 0.05 \\
Diff. Beamwidth $d\sigma$ (degrees) & -0.001 & 0.001 & 0.001 \\
Diff. Ellipticity plus $dp$ (+) & -0.006 & 0.017 & 0.004 \\
Diff. Ellipticity cross $dc$ ($\times$) & 0.000 & 0.005 & 0.004 \\
\hline
\end{tabular}
\end{center}
\end{table}
The median Gaussian beamwidth $\sigma$ across the focal plane is
$0.167^{\circ}$, corresponding to a FWHM of 23.6 arcminutes, and has
been corrected for the non-negligible size of the source aperture (see
Section~\ref{ss:arrayaverage}). The per-detector ellipticity has
large variation across the focal plane, with a central value close to
zero for both $p$ and $c$. The ellipticity individual measurement
uncertainty is somewhat large (but similar to those obtained for
{\sc Bicep2}/{\it Keck Array}) because random artifacts in the beam maps
occasionally cause the beam fitting routine to choose an ellipticity
which is a poor fit to the beam. These measurements can be improved
by careful deglitching or visually inspecting beam residuals and
removing failed fits. We also note that the typical measurement
uncertainties of differential parameters are much smaller; this is
because artifacts tend to affect both detectors in a pair equally and
thus bias the parameter estimate in the same way.
\subsection{Per-Pair Parameters}
\label{ss:perpair}
In pair differencing experiments, differential beam parameters
correspond to components of the intensity field which leak into
polarization. Here we characterize the distribution of beam
parameters corresponding to the difference of elliptical Gaussians,
which approximately couple to the CMB temperature map and its
derivatives, and which are marginalized over in our deprojection
procedure. In Section~\ref{sect:deprojection} we compare the
magnitude and distribution of these measurements to those found for
{\sc Bicep2}\ and {\it Keck Array}.
Figure~\ref{fig:diffsig} shows the differential beamwidth across the
focal plane. The distribution is centered around zero, incoherent
across the array, and small compared to the beamwidth -- the median
absolute differential beamwidth is $0.0007^{\circ}$ compared to
$0.167^{\circ}$. Figure~\ref{fig:diffpoint} shows the differential
pointing, which for {\sc Bicep2}\ and {\it Keck Array}\ has traditionally been
the largest beam shape mismatch mode. The distribution appears random
across the focal plane, although several tiles exhibit coherence
across the tile (e.g. Tiles 1 and 17); the median offset magnitude
$\sqrt{dx^2 + dy^2}$ is 0.103 arcminutes. Figure~\ref{fig:diffellip}
shows the differential ellipticity across the focal plane. We notice
that differential ellipticity is largely coherent within tiles, and is
mostly $\pm dp$ corresponding to ellipse orientations along the $x'$
or $y'$ axis. Tiles 8, 12, 13, 17, and 18 (those corresponding to
MCE0; see Figure~\ref{fig:fpu}) are rotated $90^{\circ}$ with respect
to the rest of the tiles. When their parameters are rotated so that
all tiles have the same polarization orientation (i.e. with $A$ along
the $x'$ axis), we find that differential ellipticity is primarily
$-dp$, corresponding to the $y'$ direction on the focal plane and the
$B$ polarization axis on the tiles. The median differential
ellipticity for the array in the same orientation is $dp = -0.014$ and
$dc = 0.000$.
\begin{figure}[h!]
\begin{center}
\includegraphics[width=0.75\columnwidth]{dsig_bicep3_2016.pdf}
\caption{\label{fig:diffsig} {\sc Bicep3}\ differential beamwidth
$d\sigma$. The size of the circle is proportional to the beamwidth
difference between $A$ and $B$; blue circles indicate $A$ is larger
than $B$. The distribution is incoherent across the focal plane and
small compared to the median beamwidth $\sigma = 0.167^{\circ}$.}
\end{center}
\end{figure}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=0.75\columnwidth]{diffpoint_ffbm_bicep3_2016.pdf}
\caption{\label{fig:diffpoint} {\sc Bicep3}\ differential pointing
$dx$, $dy$. The arrows point from the $A$ detector location to the
$B$ detector location, but the tails are centered on the measured
$A/B$ centroid. The length of the arrow corresponds to the degree
of mismatch, scaled up by a factor of 40 so that the small offsets
are visible. The distribution does not show a pattern across the
focal plane, but several tiles have a coherent pointing direction.}
\end{center}
\end{figure}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=0.75\columnwidth]{dellip_bicep3_2016.pdf}
\caption{\label{fig:diffellip} {\sc Bicep3}\ differential ellipticity
$\sqrt{dp^2 + dc^2}$. The direction of the ellipse indicates the
direction of the differential ellipticity -- the legend indicates
the axes corresponding to the $p$ and $c$ axes. The ellipticity is
highly exaggerated to highlight small deviations from the nominal
circular beams. Tiles 8, 12, 13, 17, and 18 -- the three in the
bottom row and the left two in the row immediately above it -- are
rotated $90^{\circ}$ with respect to the others. Tiles mostly
exhibit $\pm dp$ differential ellipticity, which is along the tile
$B$ polarization axis.}
\end{center}
\end{figure}
\section{COMPOSITE BEAM MAPS AND ARRAY-AVERAGED MAPS}
\label{sect:composites}
\subsection{Composite Maps}
\label{ss:composites}
Since each detector has many beam maps at multiple boresight rotation
angles, we can coadd them to obtain a high signal-to-noise composite
map which covers all parts of the beam. We only use beam maps which
have acceptable 2D Gaussian fits and in which both the $A$ and $B$
detectors were measured. Each component map is rotated to account for
the boresight angle and centered on the common centroid for each
detector pair. The ground is masked out since it causes visible
response even in the demodulated maps. The composite map is then made
from the median amplitude of each pixel across component maps.
Figure~\ref{fig:composite} shows an example composite and the
components which were used to make it. Spurious pixels in the
component maps do not contribute to the final composite, which has
visibly lower noise. In the 2016 dataset, only 38 optically
responsive detectors have fewer than 10 good component maps (the
typical detector has $\sim 30$). Although the composite beams shown
in these proceedings only extend to $5^{\circ}$ away from the main
beam, we can generate maps out to much larger angles. Future work
with extended composite maps will allow for explicit measurement of
extended beam response, including detailed constraints on reflections
and total integrated sidelobe power out to large angles.
\begin{figure}[h!]
\begin{center}
\includegraphics[trim=0 0 0 0,clip,width=1.0\columnwidth]{composite_fig.pdf}
\caption{\label{fig:composite} Composite beam map for a single
detector (center), surrounded by the component maps used in the
composite. Each component map is labeled with the schedule number
(out of 77 total) and the boresight rotation angle. All maps are
rotated to standard $(x',y')$ coordinates and are in dB relative to
the peak amplitude -- the composite clearly has a lower noise level
than the components. The ground is masked out in each of the
component maps, $\sim 2^{\circ}$ below the source.}
\end{center}
\end{figure}
\subsection{Array-Averaged Beam}
\label{ss:arrayaverage}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=0.6\columnwidth]{avgbeam_b3.pdf}
\caption{\label{fig:avgbeam} {\sc Bicep3}\ average beam, made by
coadding composite beam maps from all optically active detectors.
The peak is at 52.6 dBi.}
\end{center}
\end{figure}
\begin{figure}[h!]
\begin{center}
\includegraphics[width=0.7\columnwidth]{bl_keckb3.pdf}
\caption{\label{fig:bl} {\sc Bicep3}\ $B_{\ell}$ profile (black),
compared to {\it Keck Array}\ $B_{\ell}$ at 3 frequencies. All
{\it Keck Array}\ apertures are 264 mm in diameter, while {\sc Bicep3}'s is
520 mm.}
\end{center}
\end{figure}
From the composite beam maps, we integral normalize and coadd all
operational beams to form an array-averaged beam map.
Figure~\ref{fig:avgbeam} shows the averaged map in dBi (i.e. relative
to an isotropic radiator); Airy ring structure is clearly visible.
This map is then averaged in radial bins and Fourier transformed to
obtain the circularly symmetric beam window function $B(\ell)$. The
measured $B(\ell)$ is corrected for the effect of the finite size of
the source aperture:
\begin{equation}
B(\ell)_{measured} = B(\ell)_{true} \times \frac{2J_1 \left( \ell D/2 \right)}{\ell D/2},
\end{equation}
where $J_1$ is the Bessel function of the first kind and $D =
0.179^{\circ}$ is the diameter of the thermal source as seen from
{\sc Bicep3}. Figure~\ref{fig:bl} shows $B(\ell)_{true}$ for
{\sc Bicep3}\ and {\it Keck Array}, which we use to smooth the {\it Planck}\ input
maps sampled in our timestream-level simulations.
\section{DEPROJECTION AND EXPECTED SYSTEMATICS CONTROL}
\label{sect:deprojection}
\begin{table}[ht]
\caption{Differential Beam Parameters for {\sc Bicep2}, {\it Keck Array}, and
{\sc Bicep3}. Experiments are ordered by decreasing median beam size
(see Section~\ref{sect:deprojection} for a discussion about the
scaling of leakage modes with beam size). Detector count only
includes optically active detectors -- the total number of TESs is
higher. {\it Keck Array}\ parameters are combined over all receivers
operating at the indicated frequency -- five for 150 GHz, and two
each for 95 and 220 GHz -- so coherent per-receiver effects may
average out (see e.g. $dx$ in {\it Keck Array}\ 2012). For more detailed
{\it Keck Array}\ beam parameters, see Table 3 of the Beams
paper\cite{biceptwo15_IV}. Values are Array Median $\pm$ Array
Scatter $\pm$ Individual Measurement Uncertainty.}
\label{tab:statsvb2}
\begin{center}
\begin{tabular}{lccc}
\hline
\hline
\rule[-1ex]{0pt}{3.5ex} Parameter & {\it Keck Array}\ 2015 (95 GHz) & {\sc Bicep2}\ (150 GHz) & {\it Keck Array}\ 2012 (150 GHz) \\
\hline
$\sigma$ (degrees) & 0.306 & 0.220 & 0.216 \\
Detector count & 544 & 500 & 2480 \\
$dx$ (arcmin) & 0.19 $\pm$ 0.51 $\pm$ 0.05 & 0.81 $\pm$ 0.29 $\pm$ 0.14 & 0.00 $\pm$ 0.97 $\pm$ 0.07 \\
$dy$ (arcmin) & -0.24 $\pm$ 0.48 $\pm$ 0.04 & 0.78 $\pm$ 0.35 $\pm$ 0.14 & -0.21 $\pm$ 0.54 $\pm$ 0.06 \\
$d\sigma$ (degrees) & 0.000 $\pm$ 0.001 $\pm$ 0.001 & 0.000 $\pm$ 0.001 $\pm$ 0.001 & 0.000 $\pm$ 0.001 $\pm$ 0.001 \\
$dp$ (+) & -0.010 $\pm$ 0.008 $\pm$ 0.002 & -0.002 $\pm$ 0.013 $\pm$ 0.011 & -0.007 $\pm$ 0.015 $\pm$ 0.003 \\
$dc$ ($\times$) & 0.002 $\pm$ 0.003 $\pm$ 0.002 & -0.003 $\pm$ 0.012 $\pm$ 0.005 & -0.007 $\pm$ 0.009 $\pm$ 0.003 \\
\hline
\end{tabular}
\vspace{10pt}
\begin{tabular}{lccc}
\hline
\hline
\rule[-1ex]{0pt}{3.5ex} Parameter & {\it Keck Array}\ 2013 (150 GHz) & {\sc Bicep3}\ (95 GHz) & {\it Keck Array}\ 2015 (220 GHz)\\
\hline
$\sigma$ (degrees) & 0.216 & 0.167 & 0.142 \\
Detector count & 2480 & 2400 & 992 \\
$dx$ (arcmin) & 0.11 $\pm$ 0.58 $\pm$ 0.07 & 0.03 $\pm$ 0.13 $\pm$ 0.05 & -0.05 $\pm$ 0.19 $\pm$ 0.04 \\
$dy$ (arcmin) & -0.22 $\pm$ 0.56 $\pm$ 0.06 & -0.12 $\pm$ 0.17 $\pm$ 0.05 & -0.01 $\pm$ 0.28 $\pm$ 0.04 \\
$d\sigma$ (degrees) & 0.000 $\pm$ 0.001 $\pm$ 0.001 & -0.001 $\pm$ 0.001 $\pm$ 0.001 & 0.000 $\pm$ 0.001 $\pm$ 0.001 \\
$dp$ (+) & -0.011 $\pm$ 0.012 $\pm$ 0.002 & -0.006 $\pm$ 0.017 $\pm$ 0.004 & -0.017 $\pm$ 0.008 $\pm$ 0.006\\
$dc$ ($\times$) & -0.004 $\pm$ 0.007 $\pm$ 0.002 & 0.000 $\pm$ 0.005 $\pm$ 0.004 & 0.003 $\pm$ 0.007 $\pm$ 0.006 \\
\hline
\end{tabular}
\end{center}
\end{table}
In this section, we compare {\sc Bicep3}'s differential beam parameters
with those from {\sc Bicep2}\ and {\it Keck Array}\ (Table~\ref{tab:statsvb2})
and comment on the expected level of temperature-to-polarization
leakage. In our CMB analysis pipeline, for each detector pair we
perform a deprojection procedure in which the timestream data are
regressed against templates of the real CMB temperature sky and its
derivatives corresponding to the difference modes of two mismatched
beams. The template multiplied by the best-fit regression coefficient
is removed from the timestream, eliminating all leakage caused by a
particular mode without external knowledge of the beams. As an
alternative to deprojection, we may also choose to fix the coefficient
which scales the deprojection template to the beam-map-derived value
-- we refer to this procedure as ``subtraction.'' Since differential
gain is likely time-variable, we deproject it. Gain is not measured
in beam maps because of the different TES used and loading environment
between calibration measurements and CMB observations. In our
previous results, we have chosen to deproject differential pointing;
the measured correlation between beam-map-derived values and
deprojection coefficients offers a robust check that our beam maps
correspond to reality. Differential beamwidth has been found to be
negligible, and so has neither been deprojected nor subtracted. We
subtract differential ellipticity, because the real-sky $TE$
correlation would be filtered out of our $TE$ and $EE$ spectra.
We first consider differential pointing, which for {\sc Bicep2}\ and
{\it Keck Array}\ has been the beam mismatch mode contributing the most
leakage and is independent of beam size. While {\sc Bicep2}'s $dx$ and
$dy$ were quite large, significant development effort\cite{obrient12}
resulted in subsequent {\it Keck Array}\ focal planes at 95, 150, and 220
GHz with differential pointing that was several times smaller.
{\sc Bicep3}\ is most similar to {\it Keck Array}\ 220 GHz in both the
magnitude and the measurement precision. We can expect {\sc Bicep3}'s
pre-deprojection leakage from differential pointing to be much less
than that of {\sc Bicep2}; the distribution is also incoherent across the
focal plane and will likely average down significantly.
{\sc Bicep3}'s differential beamwidth is incoherent across the focal
plane, centered close to zero, and is measured with precision similar
to that of previous receivers. The level of leakage scales with one
power of the beamwidth. Though in previous results we have found
differential beamwidth to contribute negligible leakage, if beam map
simulations show that this mode is significant for {\sc Bicep3},
deprojection is a possibility.
Differential ellipticity is subtracted in our standard results, and
therefore the precision requirements become stronger. We first
observe that the measurement uncertainties in Table 2 are lower than
those achieved by {\sc Bicep2}, so we may expect that {\sc Bicep3}\ will
be able to subtract leakage from differential ellipticity with higher
precision than in previous results. We also note that the
pre-subtraction leakage scales as the square of the beam size --
{\sc Bicep3}'s beams are $\sim 75\%$ the size of {\sc Bicep2}'s, so a
similar level of inherent differential ellipticity would produce
significantly less leakage in {\sc Bicep3}.
Ultimately, the effectiveness of deprojection is simulated explicitly
using beam map simulations in which we convolve the {\it Planck}\ CMB
temperature sky with the individual beam maps and process simulated
timestreams just like the real data. We can then deproject these maps
which contain only temperature-to-polarization leakage from measured
beam mismatch. Based on comparison of the magnitude of the dominant
differential pointing in Table~\ref{tab:statsvb2}, we expect the
inherent leakage from beam mismatch modes to be equal to or lower than
that from {\sc Bicep2}.
Though deprojection removes all power corresponding to the coupling of
the CMB temperature sky to a second-order expansion of the measured
differential beam shape, real beams contain higher-order power which
deprojection does not remove. Since the undeprojected beam mismatch
residuals are often complicated, vary significantly across the focal
plane, and may nontrivially cancel or amplify in the final map, the
degree to which they will affect the final $B$-mode\ power spectrum is
not immediately obvious. The undeprojected residual level is the
final result of the beam map simulations after deprojection. In the
{\sc Bicep2}\ final results, we found that after deprojection of relative
gain and differential pointing, and subtraction of differential
ellipticity, we were able to constrain the residual
temperature-to-polarization leakage to an equivalent $r < 3.0 \times
10^{-3}$. Similar simulations for {\sc Bicep3}\ will be run using the
composite beam maps presented above, and we expect the measurement
uncertainty on the beam maps to be much lower than that of
{\sc Bicep2}\ because of higher signal strength afforded by the new
thermal chopper.
\section{CONCLUSIONS}
\label{sect:conclusions}
In these proceedings we have presented a preliminary analysis of
{\sc Bicep3}'s far field beam performance for the 2016 observing
season. Measurements were made during the 2015-2016 deployment season
using a new chopped thermal microwave source with a 24'' aperture. We
find that {\sc Bicep3}\ has a median beamwidth of $0.167^{\circ}$ (23.6
arcminutes) and present the array-averaged $B(\ell)$ profile. When
detector pairs are fit to 2D elliptical Gaussians, we find that
differential pointing, beamwidth, and ellipticity are similar to or
lower than those measured for {\sc Bicep2}\ and {\it Keck Array}. The
high-fidelity per-detector composite beam maps we have constructed
from 77 beam mapping schedules will allow for explicit simulation of
the undeprojected residual temperature-to-polarization leakage
expected in the final polarization maps produced by {\sc Bicep3}.
\acknowledgments
The {\sc Bicep3}\ project has been made possible through support from
the National Science Foundation (grant Nos. 1313158, 1313010, 1313062,
1313287, 1056465, and 0960243), the SLAC Laboratory Directed Research
and Development Fund, the Canada Foundation for Innovation, and the
British Columbia Development Fund. The development of antenna-coupled
detector technology was supported by the JPL Research and Technology
Development Fund and grants 06-ARPA206-0040 and 10-SAT10-0017 from the
NASA ARPA and SAT programs. The development and testing of focal
planes were supported by the Gordon and Betty Moore Foundation at
Caltech. The computations in these proceedings were run on the
Odyssey cluster supported by the FAS Science Division Research
Computing Group at Harvard University. Tireless administrative
support was provided by Irene Coyle, Kathy Deniston, Donna Hernandez,
and Dana Volponi.
We are grateful to Samuel Harrison and Hans Boenish as our 2015 and
2016 winterovers, respectively. We thank the staff of the US Antarctic
Program and in particular the South Pole Station without whose help
this research would not have been possible. We thank our {\sc Bicep1},
{\sc Bicep2}, {\it Keck Array}\, and {\sc Spider}\ colleagues for useful discussions
and shared expertise.
| 7eda8787c1702bf9969e9504f71ebe7eb240464b | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}\label{secIntr}
Understanding the nature and structure of the excited states of the
nucleon is a key contemporary problem in QCD. The Roper resonance,
$N^*(1440)$, was first deduced from the analysis of $\pi N$ phase shifts
in 1963 \cite{Roper1964}. However, its structure and nature have
aroused interest ever since \cite{Aznauryan2008,Joo2005}; it is lighter than the first odd-parity
nucleon excitation, $N^*(1535)$, and has a significant branching ratio
into $N\pi\pi$. Although it is recognized as a well established
resonance (four-star ranking in the Review of Particle Physics)
\cite{PDG2014}, the properties of the Roper, such as the mass, width,
and decay branching ratios, still suffer large experimental
uncertainties \cite{Mokeev2012,Aznauryan2012,Thiel2015}.
On the theoretical side, there are widely varying models describing
the Roper resonance, such as early classical quark models
\cite{Weber1990,Julia-Diaz2006,Barquilla-Cano2007,Golli:2007sa,Golli:2009uk}, bag
\cite{Meissner1984} and Skyrme models \cite{Hajduk1984}, dynamically
generated by meson-nucleon interactions \cite{Krehl2000,
Schuetz1998,Matsuyama2007,Kamano2010,Kamano2013,Hernandez2002a}, or
a monopole gluonic excitation
\cite{Barnes1983,Golowich1983,Kisslinger1995}. However, these
descriptions do encounter challenges. For example, predictions of the
mass are often too large or predictions for its width are too small.
Difficulties are also encountered in explaining its electromagnetic
coupling \cite{Sarantsev2008}.
One expects that lattice QCD simulations will provide unique
information concerning the Roper in a finite volume
\cite{Mahbub:2013ala,Mahbub:2010rm,Roberts:2013ipa,Roberts:2013oea,Alexandrou2015,Edwards2011,Kiratidis2015,Liu2014a}.
Current simulation results near the physical quark masses on lattices
with spatial length $L \simeq 3$ fm
\cite{Mahbub:2013ala,Mahbub:2010rm,Alexandrou2015,Kiratidis2015} reveal a $2S$-like
radial excitation \cite{Roberts:2013ipa,Roberts:2013oea} of the
nucleon near 1800 MeV, much higher than the infinite volume mass of
1440 MeV. The main task of this paper is to examine the physical
phase shifts, inelasticities and pole position associated with the
$N^*(1440)$ Roper resonance and connect these infinite-volume
observables to the finite-volume spectrum of lattice QCD. We use the
formalism of Hamiltonian effective field theory (HEFT) to achieve this
goal and seek an understanding of the observed finite-volume spectrum
in the context of empirical scattering observables.
The investigations of recent papers \cite{Hall2013, Wu2014} have shown
how to relate the eigenvalues of a finite-volume Hamiltonian matrix to
the spectrum of states observed in lattice QCD. These two papers
explored the spectrum of states with the quantum numbers of the
$\Delta(1232)$ resonance \cite{Hall2013} and the $\pi\pi$-$K\bar{K}$
system \cite{Wu2014} via solutions of the eigen-equation of a
finite-volume Hamiltonian matrix. Both papers showed that this
Hamiltonian matrix approach is equivalent to the well known L\"uscher
formulation \cite{Luscher1990,Luescher1991}. Furthermore,
Ref.~\cite{Wu2014} showed that this method is sufficient for the
multi-channel scattering case, where the L\"uscher method is more
difficult to apply because it needs the phase shifts and inelastic
factors in every channel. In contrast, the parameters of the
Hamiltonian can be constrained by the empirical phase shifts and
inelasticities. As a result, the spectrum is easily obtained in HEFT.
This work is a direct application of HEFT in the $N\frac{1}{2}^+$
multi-channel case including three channels, $\pi N$, $\pi\Delta$, and
$\sigma N$. The parameters of the Hamiltonian are fitted to describe the phase shifts and inelasticities of $\pi N$
scattering up to 1800 MeV. Then, in the finite-volume relevant to
lattice QCD, the energy eigenvalues and their associated eigenvectors (describing the wave functions of the eigenstates) are obtained from the
Hamiltonian matrix. Both the energy eigenvalues and the eigenstate
wave functions are important in understanding the spectrum of the
Roper channel obtained in today's lattice QCD simulations.
The framework of HEFT is described in Sec.~\ref{secFrm}. We illustrate how
the phase shifts and inelasticities in the infinite volume of
Nature are obtained and the manner in which the finite-volume energy
eigenstates are calculated. The numerical results and associated
discussion are presented in Sec.~\ref{secNum}. Here we present
results for
three different hypotheses for the internal structure of the Roper.
In the first case, the Roper is postulated to have a triquark-like
bare or core component with a mass exceeding the resonance mass. This
component mixes with attractive virtual meson-baryon contributions,
including the $\pi N$, $\pi \Delta$, and $\sigma N$ channels, to
reproduce the observed pole position.
In the second hypothesis, the Roper resonance is dynamically generated
purely from the $\pi N$, $\pi \Delta$, and $\sigma N$ channels.
In the third hypothesis, the Roper resonance is coupled to the
low-lying bare component associated with the ground state nucleon.
Through coupling with the virtual meson-baryon contributions the
scattering data and pole position are reproduced. The merit of this
third approach lies in its ability to not only describe the observed
nucleon energy levels in large-volume lattice QCD simulations but also
explain why other low-lying states have been missed in today's lattice
QCD results. Finally, a short summary is given in Sec.~\ref{secSum}.
\section{Framework}\label{secFrm}
In this section we provide a short introduction to HEFT and illustrate
how it is used in both infinite and finite volumes. The Hamiltonian
interactions associated with the $N\frac{1}{2}^+$ resonance channel
are described in Sec.~\ref{secInt}. In Sec.~\ref{secPha}, the
phase shifts and inelasticities are derived from the Hamiltonian model
and the pole positions of states are easily obtained via the
$T$-matrix. The Hamiltonian is then momentum discretised for the
finite-volume of the lattice in Sec.~\ref{secHam} and the spectrum of
energy eigenstates is obtained by solving the Hamiltonian
eigen equation.
\subsection{Hamiltonian in Channels with $\mathbf{I(J^P)=\frac12(\frac12^+)}$} \label{secInt}
The main channels strongly coupled to the Roper are the $\pi N$, $\pi
\Delta$ and $\sigma N$ channels. In the rest frame, the Hamiltonian of
the system with $I(J^P)=\frac12(\frac12^+)$ has the following
energy-independent form
\cite{Matsuyama2007,Kamano2010,Kamano2013,Thomas1984},
\begin{eqnarray}
H = H_0 + H_I.
\label{eq:h}
\end{eqnarray}
The non-interacting part is
\begin{eqnarray}
H_0 &=&\sum_{B_0} |B_0\rangle \, m^{0}_{B} \, \langle B_0|+ \sum_{\alpha}\int d^3\vec{k}\nonumber\\
&& |\alpha(\vec{k})\rangle \, \left [ \sqrt{m_{\alpha_1}^2 + \vec{k}^2} +
\sqrt{m_{\alpha_2}^2 + \vec{k}^2}\, \right ] \langle\alpha(\vec{k})|,
\label{eq:h0}
\end{eqnarray}
where $B_0$ is the bare baryon (including a bare nucleon $N_0$ or a
bare Roper $R_0$) with mass $ m^{0}_{B}$ and $\alpha$ denotes the
included channels $\pi N$, $\pi \Delta$, and $\sigma N$. The masses
$m_{\alpha_1}$ and $m_{\alpha_2}$ are the masses of the meson and
baryon in the channel $\alpha$, respectively.
The interaction Hamiltonian of this system includes two parts
\begin{eqnarray}
H_I = g + v,\label{eq:hi}
\end{eqnarray}
where $g$ describes the vertex interaction between the bare particle
and the two-particle channels $\alpha$
\begin{eqnarray}
g &=& \sum_{\alpha\, B_0} \int d^3\vec{k} \left \{ \,
|\alpha(\vec{k})\rangle \, G^\dagger_{\alpha, B_0}(k)\, \langle B_0|
\right . \nonumber \\
&&\qquad\qquad\quad \left . + |B_0\rangle\, G_{\alpha, B_0}(k)\, \langle \alpha(\vec{k})|
\, \right \} \, ,
\label{eq:int-g}
\end{eqnarray}
while the direct two-to-two particle interaction is defined by
\begin{eqnarray}
v = \sum_{\alpha,\beta} \int d^3\vec{k}\; d^3\vec{k}'\,
|\alpha(\vec{k})\rangle\, V^{S}_{\alpha,\beta}(k,k')\, \langle
\beta(\vec{k}')| \, .
\label{eq:int-v}
\end{eqnarray}
For the vertex interaction between the bare baryon and two-particle
channels, the following form is used:
\begin{eqnarray}
G_{\pi N, B_0}^2(k)&=&\frac{3 g_{B_0\pi N}^2 }{4\pi^2 f^2} \frac{k^2 u_{\pi N}^2(k)}{\omega_\pi(k)},\\
G_{\pi \Delta, B_0}^2(k)&=&\frac{g_{B_0\pi\Delta}^2 }{3\pi^2 f^2} \frac{k^2 u_{\pi \Delta}^2(k)}{\omega_\pi(k)},\\
G_{\sigma N, B_0}^2(k)&=&\frac{g_{B_0\sigma N}^2 }{4\pi^2} \frac{u_{\sigma N}^2(k)}{\omega_\sigma(k)},
\end{eqnarray}
where $f=92.4$ MeV is the pion decay constant, $\omega_X(k)=\sqrt{k^2+m_X^2}$ is the corresponding
energy, and $u_\alpha(k)$ is the regulator \cite{Leinweber:2003dg,Wang:2007iw}. We consider the exponential form
\begin{equation}
u_\alpha(k)=\exp\left ( -\frac{k^2}{\Lambda_\alpha^2} \right ) \, ,
\end{equation}
where $\Lambda_\alpha$ is the regularization scale.
Although we adopt the exponential form, our main conclusions are not affected if other form factors
are used. We have explicitly checked the selection of a dipole form factor
$u_\alpha(k)=\left(1+{k^2}/{\Lambda_\alpha^2}\right)^{-2}$. The phase shifts and inelasticities
are fit well and we obtain similar finite-volume results for the three scenarios
considered in Sec.~\ref{secNum}.
For the two-to-two particle interaction, we introduce the separable potentials for the following five channels
\begin{eqnarray}
V_{\pi N, \pi N}^S(k,k')&=&g^S_{\pi N}\frac{\bar G_{\pi N}(k)}{\sqrt{\omega_\pi(k)}}
\frac{\bar G_{\pi N}(k')}{\sqrt{\omega_\pi(k')}}\, ,\label{eqVspiN}\\
V_{\pi \Delta, \pi \Delta}^S(k,k')&=&g^S_{\pi \Delta}\frac{\bar G_{\pi \Delta}(k)}{\sqrt{\omega_\pi(k)}}
\frac{\bar G_{\pi \Delta}(k')}{\sqrt{\omega_\pi(k')}}\, , \label{eqVsPiDelta} \\
V_{\pi N, \pi \Delta}^S(k,k')&=&g^S_{\pi N,\pi \Delta}\frac{\bar G_{\pi N}(k)}{\sqrt{\omega_\pi(k)}}
\frac{\bar G_{\pi \Delta}(k')}{\sqrt{\omega_\pi(k')}}\, , \label{eqVPiNPiDel} \\
V_{\sigma N, \sigma N}^S(k,k')&=&g^S_{\sigma N,\sigma N}\frac{\bar G_{\sigma N}(k)}{\sqrt{\omega_\sigma(k)}}
\frac{\bar G_{\sigma N}(k')}{\sqrt{\omega_\sigma(k')}}\, , \label{eqVSigmaN}\\
V_{\pi N, \sigma N}^S(k,k')&=&g^S_{\pi N,\sigma N}\frac{\bar G_{\pi N}(k)}{\sqrt{\omega_\pi(k)}}
\frac{\bar G_{\sigma N}(k')}{\sqrt{\omega_\sigma(k')}}\, , \label{eqVPiNSigmaN}
\end{eqnarray}
where $\bar G_{\alpha}(k)=G_{\alpha, B_0}(k)/g_{B_0 \alpha}$.
\subsection{Phase Shift and Inelasticity}\label{secPha}
The T-matrices for two particle scattering can be obtained by solving
a three-dimensional reduction of the coupled-channel Bethe-Salpeter
equations for each partial wave
\begin{eqnarray}
T_{\alpha, \beta}(k,k';E)&=&V_{\alpha, \beta}(k,k';E)+\sum_\gamma \int
q^2 dq \times \\
&&\hspace{-1.0cm} V_{\alpha, \gamma}(k,q;E)\,
\frac{1}{E-\omega_\gamma(q)+i \epsilon}\, T_{\gamma, \beta}(q,k';E)
\, , \nonumber
\end{eqnarray}
where $\omega_\alpha(k)$ is the center-of-mass energy of channel $\alpha$
\begin{equation}
\omega_{\alpha}(k)=\sqrt{m_{\alpha_1}^2+k^2}+\sqrt{m_{\alpha_2}^2+k^2},
\end{equation}
and the coupled-channel potential can be calculated from the interaction Hamiltonian
\begin{eqnarray}
V_{\alpha, \beta}(k,k') &=& \sum_{B_0}\, G^\dag_{\alpha, B_0}(k)\,
\frac{1}{E-m_B^0}\, G_{\beta, B_0}(k') \nonumber \\
&&+V^S_{\alpha,\beta}(k,k') \, .
\label{eq:lseq-2}
\end{eqnarray}
With the normalization $\langle
\alpha(\vec{k})\, |\, \beta(\vec{k}^{\,\,'})\rangle =
\delta_{\alpha,\beta}\, \delta (\vec{k}-\vec{k}^{\,\,'})$, the S-matrix
for $\pi N \to \pi N$ is related to the T-matrix by
\begin{eqnarray}
S_{\pi N}(E_{\rm cm})&=&1 -2i\pi \frac{\omega_\pi(k_{\rm cm})\, \omega_N(k_{\rm cm})}{E_{\rm cm}}\, k_{\rm cm}\nonumber\\
&&\quad\quad \times T_{\pi N,\pi N}(k_{\rm cm}, k_{\rm cm}; E_{\rm
cm}) \, ,
\end{eqnarray}
where $k_{\rm cm}$ satisfies the on-shell condition $\sqrt{m_N^2+k_{\rm
cm}^2}+\sqrt{m_\pi^2+k_{\rm cm}^2}=E_{\rm cm}$. One can obtain
phase shifts $\delta$ and inelasticities $\eta$ with $S_{\pi N}(E_{\rm
cm})=\eta \exp(2i\delta)$.
In addition to the phase shifts and inelasticities, the pole positions
of bound states or resonances can also be obtained by searching for
the poles of the T-matrix.\\
\subsection{Finite-Volume Matrix Hamiltonian Model} \label{secHam}
We present the formalism of the finite-volume matrix Hamiltonian model by
following Refs.~\cite{Hall2013,Hall2015}. The Hamiltonian $\mathcal H$
at finite volume is the momentum discretisation of the Hamiltonian $H$ at
infinite volume. It can also be written as a sum of free and
interacting Hamiltonians $\mathcal H=\mathcal H_0+\mathcal H_I$.
In the center-of-mass frame, the meson and the baryon in the
two-particle states carry the same magnitude of momentum with
back-to-back orientation, while the bare baryon is at rest. In the
finite periodic volume of the lattice with length $L$, the momentum of
a particle is restricted to $k_n=2\pi\sqrt{n}/L$, with $n = n_x^2 +
n_y^2 + n_z^2$ such that $n=0,1,2,\ldots$
In the finite volume, it is convenient to express the Hamiltonian $\mathcal H$ as a matrix. $\mathcal H_0$ is a diagonal matrix
\begin{equation}
\mathcal H_0={\rm diag}\{m_B^0,\, \omega_{\sigma N}(k_0),~ \omega_{\pi
N}(k_1),\, \omega_{\pi\Delta}(k_1),\, \omega_{\sigma N}(k_1),~...\}.
\end{equation}
The corresponding symmetric matrix $\mathcal H_I$ is
\begin{widetext}
\begin{equation}
\mathcal H_I=\left( \begin{array}{ccccccc}
0 & \mathcal G_{\sigma N, B_0}(k_0) & \mathcal G_{\pi N, B_0}(k_1) & \mathcal G_{\pi \Delta, B_0}(k_1) & \mathcal G_{\sigma N, B_0}(k_1) & \mathcal G_{\pi N, B_0}(k_2) & \ldots \\
\mathcal G_{\sigma N, B_0}(k_0) &\mathcal V^S_{\sigma N,\sigma N}(k_0,k_0) & \mathcal V^S_{\sigma N,\pi N}(k_0,k_1)& \mathcal V^S_{\sigma N,\pi \Delta}(k_0,k_1) & \mathcal V^S_{\sigma N,\sigma N}(k_0,k_1)& \mathcal V^S_{\sigma N,\pi N}(k_0,k_2) & \ldots \\
\mathcal G_{\pi N, B_0}(k_1) &\mathcal V^S_{\pi N,\sigma N}(k_1,k_0) & \mathcal V^S_{\pi N,\pi N}(k_1,k_1) & \mathcal V^S_{\pi N,\pi \Delta}(k_1,k_1) & \mathcal V^S_{\pi N,\sigma N}(k_1,k_1) & \mathcal V^S_{\pi N,\pi N}(k_1,k_2) & \ldots \\
\mathcal G_{\pi \Delta, B_0}(k_1) & \mathcal V^S_{\pi \Delta,\sigma N}(k_1,k_0) & \mathcal V^S_{\pi \Delta,\pi N}(k_1,k_1) & \mathcal V^S_{\pi \Delta,\pi \Delta}(k_1,k_1) & \mathcal V^S_{\pi \Delta,\sigma N}(k_1,k_1) & \mathcal V^S_{\pi N, \pi \Delta}(k_1,k_2) & \ldots \\
\mathcal G_{\sigma N, B_0}(k_1) &\mathcal V^S_{\sigma N,\sigma N}(k_1,k_0) & \mathcal V^S_{\sigma N,\pi N}(k_1,k_1)& \mathcal V^S_{\sigma N,\pi \Delta}(k_1,k_1) & \mathcal V^S_{\sigma N,\sigma N}(k_1,k_1)& \mathcal V^S_{\sigma N,\pi N}(k_1,k_2) & \ldots \\
\mathcal G_{\pi N, B_0}(k_2) &\mathcal V^S_{\pi N,\sigma N}(k_2,k_0) & \mathcal V^S_{\pi N,\pi N}(k_2,k_1) & \mathcal V^S_{\pi N, \pi \Delta}(k_2,k_1) & \mathcal V^S_{\pi N,\sigma N}(k_2,k_1) & \mathcal V^S_{\pi N,\pi N}(k_2,k_2) & \ldots \\
\vdots & \vdots & \vdots & \vdots & \vdots & \vdots & \ddots \\
\end{array}
\right),
\end{equation}
\end{widetext}
where
\begin{equation}
\mathcal G_{\alpha,B_0}(k_n)=\sqrt{\frac{C_3(n)}{4\pi}}\, \left
(\frac{2\pi}{L}\right )^{3/2}\, G_{\alpha,B_0}(k_n)\, ,
\end{equation}
\begin{equation}
\mathcal V^S_{\alpha,\beta}(k_n,
k_m)=\frac{\sqrt{C_3(n)\,C_3(m)}}{4\pi}\, \left
(\frac{2\pi}{L}\right )^{3}\,
V^S_{\alpha,\beta}(k_n, k_m) \, .
\end{equation}
$C_3(n)$ represents the degeneracy factor for summing the squares of
three integers to equal $n$.
One can obtain the eigenstate energy levels on the lattice and analyse the
corresponding eigenvector wave functions describing the constituents
of the eigenstates with the above Hamiltonian $\mathcal H$.
In addition to the results at physical pion mass, we can also extend
the formalism to unphysical pion masses. Using $m_\pi^2$ as a measure
of the light quark masses, we consider the variation of the bare mass
and $\sigma$-meson mass as
\begin{eqnarray}
m_B^0(m_\pi^2)&=&m_B^0|_{\rm phy}+\alpha_B^0\, (m_\pi^2-m_\pi^2|_{\rm phy})\, , \\
m_\sigma^2(m_\pi^2)&=&m_\sigma^2|_{\rm phy}+\alpha_\sigma^0\, (m_\pi^2-m_\pi^2|_{\rm phy}),
\end{eqnarray}
where the slope parameter $\alpha_B^0$ is constrained by lattice QCD data from the CSSM. In the
large quark mass regime where constituent quark degrees of freedom become relevant one expects
$m_\sigma \sim m_\sigma|_{\rm phy} + (2/3)\, \alpha_N^0 \, (m_\pi^2-m_\pi^2|_{\rm phy})$
\cite{Cloet:2002eg} providing $\alpha_\sigma^0 \simeq \frac43 \, m_\sigma|_{\rm phy} \,
\alpha_N^0$. The nucleon and Delta masses away from the physical point, $m_N(m_\pi^2)$ and
$m_\Delta(m_\pi^2)$, are obtained via linear interpolation between the corresponding data of
lattice QCD. With $\alpha_N^0 = 1.00~{\rm GeV}^{-1}$, $\alpha_\sigma^0 \simeq 0.67$. In the
following results, we find the $\sigma N$ channel couples weakly and therefore our conclusions are
not sensitive to this value.
For the other parameters constrained by experimental data, it is difficult to predict their
quark-mass dependence. However, Refs.~\cite{Liu:2015ktc,Hall2013} show examples where lattice data
can be described well without a quark-mass dependence for the couplings. A similar approach has
been employed successfully in chiral effective field theory, where one expands in small momenta and
masses about the chiral limit. Using fixed couplings, lattice data is described over a wide range
of pion masses. In any event, the lightest pion mass considered herein is very close to the
physical pion mass. The couplings should not change significantly over this small change in pion
mass.
\section{Numerical Results and Discussion} \label{secNum}
\subsection{Fitting the Phase Shift and Inelasticity} \label{secFitData}
Here we examine the phase shifts and inelasticities associated with the $N^*(1440)$ Roper resonance
and connect these infinite-volume observables to the finite-volume spectrum of lattice QCD using
HEFT. It is natural to think that the Roper resonance might be dominated by a bare state dressed
by meson-baryon states, like the nucleon, but some authors also propose that the Roper may be a
dynamically-generated molecular state arising purely from multi-particle meson-baryon interactions.
These considerations lead us to explore three hypotheses for the structure of the Roper resonance.
\begin{figure}[tbp]
\begin{center}
\subfigure{\includegraphics[width=1.0\columnwidth]{PS}}
\subfigure{\includegraphics[width=1.0\columnwidth]{Eta}}
\caption{{\bf Colour online:} Phase shifts (upper) and inelasticities (lower) for $\pi N$ scattering
with $I(J^P)=\frac12(\frac12^+)$. The dot-dashed, dotted and dashed lines represent our best
fits for scenario I with the bare Roper, scenario II without a bare baryon, and scenario III with
the bare nucleon, respectively. }\label{figPSEta}
\end{center}
\end{figure}
In the first case, the Roper is postulated to have a triquark-like bare or core component with a
mass exceeding the resonance mass. This component mixes with virtual meson-baryon
contributions, including the $\pi N$, $\pi \Delta$, and $\sigma N$ channels, to reproduce the
observed pole position. We will refer to this first scenario (Scenario I) as the ``bare Roper''
scenario.
In the second hypothesis, the Roper resonance is dynamically generated purely from the $\pi N$,
$\pi \Delta$, and $\sigma N$ channels. We will refer to this second scenario (Scenario II) as the
``without bare baryon'' scenario.
In the third hypothesis, the Roper resonance is composed of the low-lying bare component associated
with the ground state nucleon. Through coupling with the virtual meson-baryon contributions the
scattering data and pole position are reproduced. We will refer to this third scenario (Scenario
III) as the ``bare Nucleon'' scenario.
We fit the experimental data up to centre of mass energies of 1800 MeV. Since the $\sigma$ meson
has a large width and the $N(\pi\pi)_\texttt{S-wave}$ plays an important role for the
inelasticities of the $\pi N$ channel up to 1450 MeV, we fix the $\sigma$ mass to a small value of
350 MeV to describe the threshold behavior well.
In this way we have included the threshold effects of the $N(\pi\pi)_\texttt{S-wave}$ channel in an
effective manner. The contributions of both $N(\pi\pi)_\texttt{S-wave}$ and $N \sigma$ are
included in a single effective channel.
Similar difficulties are encountered in lattice QCD. Here one needs to include both five-quark
momentum-projected $N \sigma$ and seven-quark momentum-projected $N \pi \pi$ interpolating fields
to separate the $N(\pi\pi)_\texttt{S-wave}$ and $N \sigma$ contributions. In the absence of the
seven-quark interpolating fields, the $N \pi \pi$ and $N \sigma$ contributions will be treated in a
similar effective manner, where the combined contributions are treated as a single state.
The fitted results for the phase shifts and inelasticities are plotted in Fig. \ref{figPSEta}, for
the three aforementioned scenarios. The best-fit parameters and the pole positions for each
scenario are presented in Table \ref{tabPara}. It is interesting to observe that a pole
corresponding to the Roper resonance is generated in all three scenarios, whether a bare state is
introduced or not. While the imaginary part in Scenarios II and III deviates from the Review of
Particle Physics \cite{PDG2014}, we note the model is in agreement with the phase shift and
inelasticity data.
\begin{table}[tbp]
\caption{Best-fit parameters and resultant pole positions in the three scenarios: I, the system
with the bare Roper; II, the system without a bare state; and III, the system with bare nucleon.
Underlined parameters were fixed in the fitting of that scenario. The experimental pole position
for the Roper resonance is $( 1365 \pm 15 ) - ( 95 \pm 15)\,i$ MeV \cite{PDG2014}.}
\label{tabPara}
\begin{ruledtabular}
\begin{tabular}{cccc}
\noalign{\smallskip}
Parameter & I & II & III \\
\noalign{\smallskip}
\hline
\noalign{\smallskip}
$ g^S_{\pi N} $ & $0.161$ & $0.489$ & $0.213$ \\
$ g^S_{\pi \Delta} $ & $-0.046$ & $-1.183$ & $-1.633$ \\
$ g^S_{\pi N,\pi \Delta} $ & $0.006$ & $-1.008$ & $-0.640$ \\
$ g^S_{\pi N,\sigma N} $ & $\underline{0}$ & $2.176$ & $2.401$ \\
$ g^S_{\sigma N} $ & $\underline{0}$ & $9.898$ & $9.343$ \\
$ g_{B_0\pi N} $ & $0.640$ & $\underline{0}$ & $-0.586$ \\
$ g_{B_0\pi\Delta} $ & $1.044$ & $\underline{0}$ & $1.012$ \\
$ g_{B_0\sigma N} $ & $2.172$ & $\underline{0}$ & $2.739$ \\
$ m_B^0/{\rm GeV} $ & $2.033$ & $\underline{\infty}$ & $1.170$ \\
$ \Lambda_{\pi N}/{\rm GeV} $ & $\underline{0.700}$ & $0.562$ & $\underline{0.562}$ \\
$ \Lambda_{\pi \Delta}/{\rm GeV} $ & $\underline{0.700}$ & $0.654$ & $\underline{0.654}$ \\
$ \Lambda_{\sigma N}/{\rm GeV} $ & $\underline{0.700}$ & $1.353$ & $\underline{1.353}$ \\
\noalign{\medskip}
Pole (MeV) & $1380 - 87\,i$ & $1361 - 39\,i$ & $1357 - 36\,i$ \\
\end{tabular}
\end{ruledtabular}
\end{table}
With the parameters of the interactions constrained by the experimental phase shifts and
inelasticities, one can proceed to compare the predictions of the matrix Hamiltonian model in a
finite volume with results from lattice QCD. Fig.~\ref{figLDNonInt3fm} summarises world lattice
QCD results \cite{Leinweber:2015kyz} for the positive parity nucleon spectrum
\cite{Mahbub:2013ala,Mahbub:2010rm,Kiratidis2015,Alexandrou2015} for volumes with $L \simeq 2.9$
fm. Here the statistics of the CSSM results \cite{Mahbub:2013ala} have been increased through the
consideration of approximately 29,472 propagators on the PACS-CS configurations \cite{Aoki:2008sm}.
The $2S$ orbital structure of the CSSM's first excited states of the nucleon reported in
Fig.~\ref{figLDNonInt3fm} was established in Refs.~\cite{Roberts:2013ipa} and
\cite{Roberts:2013oea}.
It may be interesting to note that the CSSM collaboration performed a rather exhaustive search for
a low-lying Roper-like state in Ref.~\cite{Mahbub:2010rm}. There a broad range of smeared-source
interpolators were considered in the hope that one could form a correlation matrix that would
reveal a low-lying Roper-like state. Instead, one found that the first excitation energy was
insensitive to the basis of interpolating fields explored and no state approaching 1440 MeV could
be found.
It is important to note that these lattice QCD results have been obtained through the use of local
three-quark interpolating fields. This approach will make it difficult to access multi-particle
scattering states. If there is little attraction to localize the multi-particle state in a
finite volume, $V$, then the overlap of the interpolator is volume-suppressed by a factor of $1/V$;
{\it i.e} the probability of finding the second hadron at the position of the first is $1/V$.
As a consequence, it may be that only the states composed of a significant bare-state component in
the Hamiltonian model will be excited by the three-quark interpolating fields. We will examine
this possibility in detail in the following.
The non-interacting energies of the two-particle meson-baryon channels considered herein for
$L = 2.90$ fm are also illustrated in Fig.~\ref{figLDNonInt3fm}. The predictions of the three
scenarios for the finite-volume spectra of lattice QCD are presented in Secs.\ \ref{secSBR},
\ref{secSNB} and \ref{secSBN}.
\begin{figure}[t]
\begin{center}
\includegraphics[width=1.0\columnwidth]{Exp3fm1Bare_2LDNonInt}
\caption{{\bf Colour online:} The lowest lying $I(J^P)=\frac12(\frac12^+)$ baryon states observed
in lattice QCD simulations with length $L \simeq 2.90$ fm, as a function of the input quark mass
($\propto m_\pi^2$). The data with filled symbols are from the CSSM group
\cite{Mahbub:2013ala,Mahbub:2010rm,Kiratidis2015} updated to high statistics herein. Those with
hollow symbols are from the Cyprus group \cite{Alexandrou2015}. The non-interacting energies of
the low-lying two-particle meson-baryon channels for this lattice are also illustrated. }
\label{figLDNonInt3fm}
\end{center}
\end{figure}
\subsection{System with the Bare Roper Basis State} \label{secSBR}
In the traditional quark model, the nucleon is thought to be made up of three constituent quarks
with small five-quark components \cite{Julia-Diaz2006}. In the view of effective field theory, the
nucleon is a mixed state of a bare nucleon component, dressed by attractive $\pi N$, $\pi \Delta$,
etc.\ components, with the bare component dominating. As a simple analogy, the Roper is popularly
treated as a state dominated by a bare Roper component with a mass above the Roper resonance
position, dressed by attractive $\pi N$, $\pi \Delta$, etc.\ components. The bare Roper component
is thought to coincide with a three-quark core while other components like $\pi N$ contain states
with at least five quarks.
Our first scenario follows this picture. In this scenario the Roper is composed of a large-mass
bare Roper state dressed by $\pi N$, $\pi \Delta$, and $\sigma N$ channels. Fits to the
phase shifts and inelasticities in this model are plotted as the red dot-dashed lines in
Fig.~\ref{figPSEta}, and the associated parameters are listed in the column labeled I in Table
\ref{tabPara}. In this fit, two of the separable potentials are removed via $g^S_{\sigma N,\sigma
N}=0$ and $g^S_{\pi N,\sigma N}=0$, because their effect on the fits is accommodated by other
interactions.
\begin{figure}[t]
\begin{center}
\includegraphics[width=1.0\columnwidth]{ExpBRoper3fm_4NoDashedMany}
\caption{{\bf Colour online}: The pion mass dependence of the $L = 2.90$ fm finite-volume
energy eigenstates for the Hamiltonian model scenario with a bare Roper basis state. The different
line types and colours used in illustrating the energy levels indicate the strength of the bare
basis state in the Hamiltonian-model eigenvector describing the composition of the state. The
thick-solid (red), dashed (blue) and dotted (green) lines correspond to the states having the
first, second, and third largest bare-state contributions. Since three-quark operators are used
to excite the states observed in lattice QCD, we label these states as the first, second, and
third most probable states to be seen in the lattice QCD simulations.}
\label{figSpecDpl3fm}
\end{center}
\end{figure}
These fit parameters enable the determination of the eigen-energy spectra in a finite volume at the
physical pion mass. To obtain the spectrum at higher quark masses, we proceed to determine the
quark mass dependence of the bare mass, $m_R^0(m_\pi^2)$, governed by the slope parameter
$\alpha_R^0$. To do this, we consider the lowest-lying excitation energies observed by the CSSM at
the largest two quark masses considered, illustrated by the filled blue-square markers in
Fig.~\ref{figLDNonInt3fm}. We assume that these lattice results, obtained with three-quark
operators, correspond to Hamiltonian-model states in which the bare state plays an important role.
The parameter $\alpha_R^0$ is then constrained via a standard $\chi^2$ measure between the first
Hamiltonian model excitation with a significant bare state eigenvector component the aforementioned
lattice QCD results. The remainder of the spectra are then a prediction. Our best fit gives
$\alpha_R^0=2.14~{\rm GeV^{-1}}$.
The energy levels of the Hamiltonian model for the lattice with $L = 2.90$ fm are illustrated
in Fig.~\ref{figSpecDpl3fm}. In this case, the lowest lying excitation in the Hamiltonian model is
almost pure $\sigma N$. The second excitation has a bare state component exceeding 10\% and
therefore this state is constrained to the lowest-lying excitation energies at the largest two
quark masses considered, illustrated by the filled blue-square markers in
Fig.~\ref{figLDNonInt3fm}, in determining $\alpha_R^0$.
\begin{figure}[t]
\begin{center}
\includegraphics[width=1.0\columnwidth]{ExpBRoper3fm_5BareRDistribt}
\caption{{\bf Colour online:} The fraction of the bare-Roper basis state, $| m_0 \rangle$, in the
Hamiltonian energy eigenstates $| E_i \rangle$ for the three states having the largest bare-state
contribution. States are labeled by the energy-eigenstate integers $i$ and these state labels
are indicated next to the curves. For example, at light quark masses, the second energy
eigenstate has the largest bare-Roper component and therefore the second excitation energy in
Fig.~\ref{figSpecDpl3fm} is highlighted with a thick red line. The dark-green dots plotted at
$y = 0.25$ indicate the positions of the five quark masses considered in the CSSM results. While
the line type and colour scheme match that of Fig.~\ref{figSpecDpl3fm}, the thick and thin
lines alternate to indicate a change in the energy eigenstate associated with that value.}
\label{figBareRDpl3fm}
\end{center}
\end{figure}
The (coloured) thick-solid, dashed and dotted line-type decorations of the Hamiltonian model
eigenstates in Fig.~\ref{figSpecDpl3fm} reflect the magnitude of the bare-state contribution to
the eigenstates.
Denoting $|E_i\rangle$ as the $i$-th energy eigenstate from the matrix Hamiltonian model the
structure of $|E_i\rangle$ is obtained through the overlap of the eigenvector with each of the
basis states. For example, the proportion of the bare state $|m_0\rangle$ in $|E_i\rangle$ is
$|\langle m_0|E_i\rangle|^2$. For the meson-baryon basis states, we sum over all the back-to-back
momenta considered when reporting their contributions to the energy eigenstates.
Since three-quark operators are used to excite the states observed in lattice QCD, Hamiltonian
model states with a large proportion of the bare-Roper basis state are more likely to be observed
in the lattice QCD calculations. To identify these states, we seek the first three eigenstates,
$|E_i\rangle$, among the first twenty lowest-lying excitations, which contain the largest
bare-Roper basis state contributions. This is done at each pion mass considered in generating the
curves. We label these states in Fig.~\ref{figSpecDpl3fm} as the first, second, and third most
probable states to be seen in the lattice QCD simulations.
Fig.~\ref{figBareRDpl3fm} reports the bare-Roper fraction, $|\langle m_0|E_i\rangle|^2$, for these
three states. The integer next to each section of the curves indicates the $i$-th energy
eigenstate associated with the fraction plotted.
We see that the bare-Roper basis state strength is spread across many energy eigenstates. None of
the first twenty eigenstates contain more than 30\% of the state in the bare-Roper basis
state. This situation contrasts that for the nucleon ground state where more than 80\% of the
energy eigenstate is composed with the bare-nucleon basis state.
To further illustrate the composition of the energy eigenstates created in this scenario,
Fig.~\ref{figCompnDpl3fm} reports the fractions of the bare-state and meson-baryon channels
composing the energy eigenvectors as a function of the squared pion mass for the first four
low-lying states in the finite volume with $L = 2.90$ fm.
The first panel, Fig.~\ref{figComDpl3fm1}, shows the first state is nearly a pure $|\sigma N\rangle$
scattering state associated with $N(\pi\pi)_\texttt{S-wave}$ contributions.
\begin{figure}[t]
\begin{center}
\subfigure[\ 1st eigenstate]{\scalebox{0.23}{\includegraphics{ExpBRoper3fm_zDisVSMpi1}}\label{figComDpl3fm1}}
\subfigure[\ 2nd eigenstate]{\scalebox{0.23}{\includegraphics{ExpBRoper3fm_zDisVSMpi2}}\label{figComDpl3fm2}}
\subfigure[\ 3rd eigenstate]{\scalebox{0.23}{\includegraphics{ExpBRoper3fm_zDisVSMpi3}}}
\subfigure[\ 4th eigenstate]{\scalebox{0.23}{\includegraphics{ExpBRoper3fm_zDisVSMpi4}}}
\caption{{\bf Colour online:} The pion-mass evolution of the Hamiltonian eigenvector components for
the scenario with the bare Roper on the lattice volume with $L = 2.90$ fm. The fractions of
the bare-state, $|\langle m_0|E_i\rangle|^2$, and meson-baryon channels,
$\sum_{\vec k} |\langle \alpha(\vec k) | E_i\rangle|^2$, composing the energy
eigenvectors are illustrated for the first four states observed in the model. Here all momenta
for a particular meson-baryon channel have been summed to report the relative importance of the
$\alpha = m_0$, $\pi N$, $\pi \Delta$ and $\sigma N$ channels. The (green) dots plotted horizontally at $y
= 0.45$ indicate the positions of the five pion masses considered by the CSSM.
\vspace{-12pt}}
\label{figCompnDpl3fm}
\end{center}
\end{figure}
Fig.~\ref{figComDpl3fm2} reveals the second lowest state is dominated by the $|\pi N\rangle$ basis
state, as argued in the original publications reporting this state
\cite{Mahbub:2013ala,Mahbub:2010rm}. Away from the avoided level crossing at the largest quark
mass considered, the state is typically 60\% $|\pi N\rangle$ 20\% bare Roper, 10\% $|\pi
\Delta\rangle$ and 10\% $|\sigma N\rangle$. The presence of a significant bare-state contribution
explains the ability of the three-quark interpolating fields used in the lattice QCD calculations
to excite this state. Similarly the absence of a significant bare-state contribution to the first
excitation explains the omission of this state in present-day lattice QCD simulations.
Of particular note is the prediction of a very significant bare-Roper contribution to the second
energy eigenstate at light quark masses. The bare-state contribution exceeds 15\% for the three
lightest quark masses considered by the CSSM, {\it i.e.} $0 \le m_\pi^2 \le 0.15\ {\rm GeV^2}$.
Therefore this scenario predicts that these low-lying states should be readily observed in the
lattice QCD calculations. However, neither the CSSM nor the Cyprus groups have observed these
states. Therefore, this scenario in inconsistent with lattice QCD. Thus the popular notion of the
Roper resonance being described by a large bare-Roper mass dressed by attractive meson-baryon
scattering channels is not supported by lattice QCD.
\subsection{System Without a Bare Baryon Basis State}\label{secSNB}
In light of the discrepancy between the first scenario and the results of lattice QCD, we proceed
to explore the possibility that Roper resonance is a pure molecular state. In this scenario, the
Roper is assumed to be void of any triquark core and therefore we do not introduce a bare-baryon
basis state. If this model can describe the experimental data, then it can also explain the void
of low-lying states in lattice QCD, as the overlap of three-quark operators with multi-particle
states is volume suppressed.
The fitted phase shifts and inelasticities are plotted as dotted lines in Fig.~\ref{figPSEta}
indicating the scattering data can be fit in the absence of a bare baryon contribution. The
corresponding fit parameters are reported in the middle column of Table \ref{tabPara}.
Fig.~\ref{figSpecExp3fm} displays the energy levels in the finite-volume lattice. The high density
of eigenstate levels from the Hamiltonian model provides easy overlap with the lattice QCD results.
The fractional meson-baryon components for the eight lowest-lying eigenstates of this scenario are
plotted in Fig.~\ref{figCompnExp3fm}. Again, the lowest-lying state is predominantly $|\sigma
N\rangle$. Noting that the second and third eigenstates are associated with the low-lying lattice
QCD results at large pion masses, Figs.~\ref{figComExp3fm2} and \ref{figComExp3fm3} indicate these
states are dominated by $|\pi N\rangle$ and $|\sigma N\rangle$ basis states. After an avoided
level crossing at large pion masses, the composition of these two states is exchanged.
The fifth and sixth states of Figs.~\ref{figComExp3fm5} and \ref{figComExp3fm6} are more
interesting. At light pion masses these states are a non-trivial superposition of all three basis
states, $|\pi N\rangle$, $|\pi \Delta\rangle$, and $|\sigma N\rangle$. These states appear to be
more than weakly mixed scattering states and it is interesting that these are the levels consistent
with the lattice QCD results at light pion masses.
\begin{figure}[t]
\begin{center}
\includegraphics[width=1.0\columnwidth]{Exp3fm_4NoDashedMany}
\caption{{\bf Colour online:} The pion mass dependence of the $L = 2.90$ fm finite-volume
energy eigenstates for the Hamiltonian model scenario without a bare baryon basis state.}
\label{figSpecExp3fm}
\end{center}
\end{figure}
\begin{figure}[tb]
\begin{center}
\subfigure[\ 1st eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm_zDisVSMpi1}}\label{figComExp3fm1}}
\subfigure[\ 2nd eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm_zDisVSMpi2}}\label{figComExp3fm2}}
\subfigure[\ 3rd eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm_zDisVSMpi3}}\label{figComExp3fm3}}
\subfigure[\ 4th eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm_zDisVSMpi4}}\label{figComExp3fm4}}
\subfigure[\ 5th eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm_zDisVSMpi5}}\label{figComExp3fm5}}
\subfigure[\ 6th eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm_zDisVSMpi6}}\label{figComExp3fm6}}
\subfigure[\ 7th eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm_zDisVSMpi7}}\label{figComExp3fm7}}
\subfigure[\ 8th eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm_zDisVSMpi8}}\label{figComExp3fm8}}
\caption{{\bf Colour online:} The pion-mass evolution of the Hamiltonian eigenvector components for
the first eight states observed in the scenario without a bare baryon state on the lattice volume
with $L = 2.90$ fm. The fifth and sixth states display a non-trivial superposition of all
three basis states, $|\pi N\rangle$, $|\pi \Delta\rangle$, and $|\sigma N\rangle$ at light pion
masses.}
\label{figCompnExp3fm}
\end{center}
\end{figure}
While only a few of the eigenstates illustrated in Fig.~\ref{figSpecExp3fm} have been seen on the
lattice, one should in principle be able to observe all of these states in future lattice QCD
calculations. The key is to move beyond local three quark operators. Five quark operators
\cite{Kiratidis2015} have successfully revealed low-lying scattering states in the odd-parity
nucleon channel that were missed with three-quark operators. Moreover, five-quark multi-particle
operators where the momentum of both the meson and the baryon are projected at the source are
particularly efficient at exciting the lowest-lying scattering states \cite{Lang:2012db}. Future
lattice QCD simulations will draw on these techniques to fill in the missing states predicted by
our Hamiltonian model.
In summary, scenario II describes the experimental scattering data and also describes the lattice
QCD results as non-trivial mixings of the basis states. More trivial mixings of the basis states
are not seen on the large-volume lattice simulations because the overlap of weakly mixed two-particle
scattering states with local three-quark operators is suppressed by the spatial volume of the lattice.
\subsection{System with a Bare Nucleon Basis State} \label{secSBN}
In light of the success of our second scenario describing the Roper resonance as a pure molecular
meson-baryon state, we proceed to a third scenario in which these channels have an opportunity to
mix with the bare baryon state associated with the ground state nucleon. There is {\it a priori}
no reason to omit such couplings.
We find fits to the the phase shifts and inelasticities to be rather insensitive to the couplings and mass of
the bare nucleon state, $|N_0\rangle$. Thus, to constrain the couplings and the bare mass, we fit
the CSSM lattice QCD results simultaneously with the experimental phase shifts and inelasticities.
We restrict the cutoffs $\Lambda_\alpha$ of the exponential regulators to be the same as those in
the second scenario. In addition, we restrict the nucleon pole to be $939$ MeV at infinite volume.
We plot the best fit results for the scattering data as dashed lines in Fig.~\ref{figPSEta}
and summarise the parameters in the right-hand column of Table \ref{tabPara} labeled III.
To obtain the eigen-energy spectrum at finite volume, we need the nucleon mass as a function of the
squared pion mass, $m_N(m_\pi^2)$. In the previous two scenarios, we used a linear interpolation between
the nucleon lattice results from the CSSM. Here, we obtain $m_N(m_\pi^2)$ via iteration where the
lowest eigen-energy of the Hamiltonian model output is used as the input for $m_N(m_\pi^2)$ in the
next iteration. Convergence is obtained without difficulty.
The slope of the bare nucleon mass as a function of $m_\pi^2$ is found to be
\begin{eqnarray}
\alpha_N^0=0.995~{\rm GeV}^{-1} \, .
\label{eqParaN0}
\end{eqnarray}
The energy levels predicted by the Hamiltonian Model and the proportion of the bare nucleon basis
state in the excited states are illustrated in Figs.~\ref{figSpecExpBN3fm} and \ref{figBareRExp3fm}
respectively.
The CSSM ground-state nucleon data are fit well by the Hamiltonian model and those from the Cyprus
group are clustered near the ground-state curve in Fig.~\ref{figSpecExpBN3fm}. We obtain the
nucleon mass on the $L \simeq 2.90$ fm volume at the physical pion mass of $0.140$ GeV to be
\begin{equation}
m_N(m_\pi^2|_{\rm phy})_{L\simeq 2.90~{\rm fm}}=0.957~{\rm GeV} \, , \quad
\end{equation}
revealing that the finite volume of the lattice increases the nucleon mass by nearly 20 MeV.
The nucleon ground state on the 3 fm lattice contains 80\% $\sim$ 90\% of the bare nucleon basis
state.
As in the first scenario, we anticipate that excited states having a large bare-state component will have a
more significant coupling with the three-quark operators used to excite the states in contemporary
lattice QCD calculations. Figure \ref{figBareRExp3fm} identifies excited states having the largest
bare-state components and thus the most probable states to be seen on the lattice.
For example, at the lightest quark mass, the sixth energy eigenstate has the largest
bare-nucleon component and is the most likely state to be observed in current lattice QCD
calculations. Correspondingly, the sixth excitation energy in Fig.~\ref{figSpecExpBN3fm} is
highlighted with a thick red line.
Both the CSSM and Cyprus lattice calculations produce an excited state consistent with the sixth
energy level.
Remarkably the second most probable state to be seen in lattice QCD simulations lies even higher in
energy at approximately 2 GeV.
For $m_\pi^2$ lying in the range $0.07 \sim 0.27\ {\rm GeV}^2$, the seventh eigenstate is predicted
to be the most easily seen with the proportion of bare nucleon basis state at 2.5\% $\sim$ 5\%.
The second most probable state to be seen in this regime is state 10 at 2.1 to 2.3 GeV. For
$m_\pi^2 > 0.27\ {\rm GeV}^2$ states 7 and 10 contain roughly equal amounts of bare state
contributions. The lattice QCD results are consistent with the lower-lying state of these two most
probable states to be seen in lattice QCD.
We do not analyse the lattice QCD results near the tenth eigen-energy level as this
energy regime surpasses the realm of our model constraints.
While the lattice QCD simulations reveal a low-lying state at the largest two quark masses
considered (blue filled squares in Fig.~\ref{figSpecExpBN3fm}) the trend does not continue into the
lighter quark mass regime. Figure \ref{figBareRExp3fm} provides an explanation for this
observation. For the second and third lightest quark masses at $m_\pi^2 \simeq 0.08$ and 0.15
GeV${}^2$ the bare-state contribution to the low-lying (green curve) state is three to five times
smaller than that for the states having the largest contribution. This reduced bare-state
component is expected to coincide with a reduced coupling of the state to three-quark operators.
Thus a possible explanation for the omission of the lower-lying state at light quark masses in
current lattice QCD simulations is that its relatively small coupling to three-quark interpolators
is insufficient for it to be seen in the correlation matrix analysis with current levels of
statistical accuracy.
The components of the eigenstate vectors are illustrated for this scenario in
Fig.~\ref{figCompnExpBN3fm}.
At the physical pion mass, Fig.~\ref{figCnExpBN1} indicates the ground state nucleon is 80\% bare
nucleon dressed with 20\% meson-baryon states spread evenly over the three meson-baryon channels
considered.
As this analysis now includes the ground-state nucleon as the first state, the state labels have
changed in scenario III being one larger than in scenarios I and II.
Once again the first excitation is predominantly $|\sigma N\rangle$.
Noting that the second and third excitations (states 3 and 4) are in the realm of the low-lying
lattice QCD results at large pion masses, Figs.~\ref{figCnExpBN3} and \ref{figCnExpBN4} indicate
these states are dominated by $|\pi N\rangle$ and $|\sigma N\rangle$ basis states. As in scenario
II an avoided level crossing at large pion masses causes the composition of these two states to be
exchanged. However, state three has the largest bare state component at the two largest quark
masses considered and it is this state that has most likely been produced in the lattice
simulations.
The next excitation, state 5, resembles state 4 of scenario II being predominantly $|\pi N\rangle$.
The sixth and seventh eigenstates (corresponding to the fifth and sixth eigenstates in scenario II)
continue to show a non-trivial mixing of the meson-baryon basis states near the lightest two quark
masses considered by the CSSM. It is precisely in this region of non-trivial mixing that the bare
state component becomes manifest.
At the lightest quark mass, Fig.~\ref{figCnExpBN6} illustrates that the sixth eigenstate is 50\%
$|\pi N\rangle$, 45\% $|\pi \Delta\rangle$ and 5\% bare nucleon.
At the second lightest quark mass considered by the CSSM, the seventh state in Fig.~\ref{figCnExpBN7} has the largest
bare-state component with
40\% $|\pi \Delta\rangle$,
40\% $|\sigma N \rangle$,
15\% $|\pi N\rangle$
and
5\% bare nucleon.
It is precisely these excited states having the largest bare state component that correspond to the
states seen in lattice QCD simulations at light quark masses.
\begin{figure}[t]
\begin{center}
\includegraphics[width=1.0\columnwidth]{Exp3fm1Bare_4NoDashedMany}
\caption{{\bf Colour online}: The pion mass dependence of the $L = 2.90$ fm finite-volume
energy eigenstates for the Hamiltonian model scenario with a bare nucleon basis state. As in
Fig.~\ref{figSpecDpl3fm}, the different line types and colours used in illustrating the energy
levels indicate the strength of the bare basis state in the Hamiltonian-model eigenvector. The
thick-solid (red), dashed (blue) and dotted (green) lines correspond to the states having the
first, second, and third largest bare-state contributions and therefore represent the most
probable states to be observed in the lattice QCD simulations.}
\label{figSpecExpBN3fm}
\end{center}
\end{figure}
\begin{figure}[t]
\begin{center}
\includegraphics[width=1.0\columnwidth]{Exp3fm1Bare_5BareRDistribt}
\caption{{\bf Colour online:} The fraction of the bare nucleon basis state, $| m_0 \rangle$, in the
Hamiltonian energy eigenstates $| E_i \rangle$ for the three states having the largest bare-state
contribution. As in Fig.~\ref{figBareRDpl3fm}, states are labeled by the energy-eigenstate
integers, $i$. The dark-green dots plotted at $y = 0.015$ indicate the
positions of the five quark masses considered in the CSSM lattice results. While the line type
and colour scheme matches that of Fig.~\ref{figSpecExpBN3fm}, the thick and thin lines alternate
to indicate a change in the energy eigenstate associated with that value.}
\label{figBareRExp3fm}
\end{center}
\end{figure}
\begin{figure}[t]
\begin{center}
\subfigure[\ 1st eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm1Bare_zDisVSMpi1}}\label{figCnExpBN1}}
\subfigure[\ 2nd eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm1Bare_zDisVSMpi2}}\label{figCnExpBN2}}
\subfigure[\ 3rd eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm1Bare_zDisVSMpi3}}\label{figCnExpBN3}}
\subfigure[\ 4th eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm1Bare_zDisVSMpi4}}\label{figCnExpBN4}}
\subfigure[\ 5th eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm1Bare_zDisVSMpi5}}\label{figCnExpBN5}}
\subfigure[\ 6th eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm1Bare_zDisVSMpi6}}\label{figCnExpBN6}}
\subfigure[\ 7th eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm1Bare_zDisVSMpi7}}\label{figCnExpBN7}}
\subfigure[\ 8th eigenstate]{\scalebox{0.23}{\includegraphics{Exp3fm1Bare_zDisVSMpi8}}\label{figCnExpBN8}}
\caption{{\bf Colour online:} The pion-mass evolution of the Hamiltonian eigenvector components for
the first eight states (including the ground state) observed in the scenario with a bare nucleon
basis state on the lattice volume with $L = 2.90$ fm. Of all the excited states, the sixth
and seventh states have the largest bare-state component at light quark masses. This component
is accompanied by a non-trivial superposition of meson-baryon basis states.}
\label{figCompnExpBN3fm}
\end{center}
\end{figure}
\subsection{Comparison of the Three Scenarios}
We have studied the phase shifts and inelasticities at infinite volume and the finite-volume
eigenstates on the lattice in three scenarios, a bare Roper basis state in the first, no bare
baryon basis state in the second and a bare nucleon basis state in the third scenario. As
illustrated in Fig.~\ref{figPSEta}, there are differences in the phase shifts and inelasticities
among the three scenarios. Within the constraints of these models, it is not possible to find
sets of fit parameters which can make the phase shifts and inelasticities of the three scenarios
overlap everywhere. Moreover, the fits cannot give the same phase shifts and inelasticities in each
of the $\pi N$, $\pi\Delta$, and $\sigma N$ channels. This will directly lead to different
eigen-energy spectra on the lattice, based on L\"uscher's Theorem.
In performing a direct comparison of the energy levels predicted in our three scenarios, the pion
mass dependence of the ground state nucleon mass $m_N(m_\pi^2)$ obtained in scenario III is used
in all three scenarios. The energy levels for the $L = 2.90$ fm lattice are compared in
Fig.~\ref{figCompare3fm}.
\begin{figure}[t]
\begin{center}
\includegraphics[width=1.0\columnwidth]{eigCom3fm}
\caption{{\bf Colour online:} Comparison of the excited-state energy levels for the three scenarios
at finite volume with $L = 2.90$ fm. All scenarios use $m_N(m_\pi^2)$ obtained in the third
scenario.}
\label{figCompare3fm}
\end{center}
\end{figure}
We can see a significant difference between cases with and without the bare Roper basis state in
Fig. \ref{figCompare3fm}. However, differences between scenarios III and II, with and without a
bare nucleon basis state respectively, are subtle. The main feature provided through the inclusion
of a bare nucleon basis state is a clear understanding of the states to be observed in contemporary
lattice QCD calculations.
\subsection{Results for a Volume with $\mathbf{L\simeq 1.98}$ fm}
In this section, we consider the smaller spatial lattice volume with $L \simeq 1.98$ fm considered
by the Hadron Spectrum Collaboration (HSC) \cite{Edwards2011}. Drawing on the fit results to the
experimental data summarized in Table \ref{tabPara}, one can proceed to explore the predictions of
the Hamiltonian model on this very small volume lattice.
First we consider Scenario I with the bare Roper basis state. The finite-volume spectrum of states
are compared with the HSC results in Fig.~\ref{figSpec2fmR} while the proportion of the bare Roper
basis state is illustrated in Fig.~\ref{figBareR2fmR}. The HSC results sit near the energy levels
of the matrix Hamiltonian model. However, the same problem encountered in the $L = 2.90$ fm volume
case appears here. The Hamiltonian model predicts states approaching 1.6 GeV in the light
quark-mass regime having a bare state component exceeding 35\% of the eigenvector. Such a state
should be easy to excite with the three-quark operators considered by the HSC. The absence of such
a state at the two lightest quark masses considered by the HSC in the mass range 1.85 to 2.00 GeV
provides further evidence that the Roper resonance is not composed with a bare basis state with mass
$\simeq 2.0$ GeV.
\begin{figure}[t]
\begin{center}
\subfigure[\ Spectrum]
{\scalebox{0.23}{\includegraphics{ExpBRoper2fm_4NoDashedMany}}\label{figSpec2fmR}}
\subfigure[\ Bare Roper Component]
{\scalebox{0.23}{\includegraphics{ExpBRoper2fm_5BareRDistribt}}\label{figBareR2fmR}}
\caption{{\bf Colour online:} The pion mass dependence of the $L = 1.98$ fm finite-volume energy
eigenstates (left) and the fraction of the bare-Roper basis state, $| m_0 \rangle$, in the
Hamiltonian energy eigenstates $| E_i \rangle$ for the three states having the largest bare-state
contribution (right) for the Hamiltonian model scenario with a bare Roper basis state.
The spectrum results with filled symbols (left) are from the Hadron Spectrum Collaboration
\cite{Edwards2011}.
The dark-green dots plotted at $y = 0.30$ (right) indicate the positions of the three quark
masses considered in the HSC results.
The line types and colour schemes match those of Figs.~\ref{figSpecDpl3fm} and
\ref{figBareRDpl3fm}.
}
\end{center}
\end{figure}
As for the $L = 2.90$ fm results, there is little difference in the finite volume spectra of
Scenarios II and III, and therefore we proceed directly to an illustration of the results for
Scenario III. The results with the bare nucleon basis state in the finite volume of $L\simeq 1.98$
fm are presented in Fig.~\ref{fig2fmN}. This time the HSC results do not coincide with the
finite volume energy levels of the Hamiltonian model illustrated in Fig.~\ref{figSpec2fmN}. There
are two finite-volume based concerns that can contribute to the origin of this discrepancy.
\begin{figure}[htbp]
\begin{center}
\subfigure[\ Spectrum]
{\scalebox{0.23}{\includegraphics{Exp2fm1Bare_4NoDashedMany}}\label{figSpec2fmN}}
\subfigure[\ Bare Nucleon Component]
{\scalebox{0.23}{\includegraphics{Exp2fm1Bare_5BareRDistribt}}\label{figBareR2fmN}}
\caption{{\bf Colour online:}
The pion mass dependence of the $L = 1.98$ fm finite-volume energy
eigenstates (left) and the fraction of the bare-Roper basis state, $| m_0 \rangle$, in the
Hamiltonian energy eigenstates $| E_i \rangle$ for the three states having the largest bare-state
contribution (right) for the Hamiltonian model scenario with a bare Nucleon basis state.
The line types, symbols and colour schemes are as in Figs.~\ref{figSpec2fmR} and
\ref{figBareR2fmR}.
}
\label{fig2fmN}
\end{center}
\end{figure}
One concern is the interference of the coarse infrared momentum discretisation induced by the small
periodic volume and the ultraviolet regulators of the loop integrals constrained by the
experimental phase shifts and inelasticities of Fig.~\ref{figPSEta}. For $P$-wave meson-baryon
dressings, the zero-momentum contribution is absent and the finite volume acts as an infrared
regulator. The form factors in the third scenario are constrained by experiment to have small
volume such that even the first momenta available on the small value are already suppressed, almost
rendering the meson-baryon dressings negligible.
For example, for a finite volume with $L=2$ fm and $k_i=2\pi/L$
\begin{equation}
u_{\pi N}^{\rm III}(k_1)=0.296,~
u_{\pi N}^{\rm III}(k_2)=0.088,~
u_{\pi N}^{\rm III}(k_3)=0.026,
\end{equation}
in scenario III. These form factors are small compared to those of the first scenario on the
same volume
\begin{equation}
u_{\pi N}^{\rm I}(k_1)=0.456,~
u_{\pi N}^{\rm I}(k_2)=0.208,~
u_{\pi N}^{\rm I}(k_3)=0.095.
\end{equation}
They are also significantly smaller than those of the third scenario with $L=3$ fm where
\begin{equation}
u_{\pi N}^{\rm III}(k_1)=0.582,~
u_{\pi N}^{\rm III}(k_2)=0.339,~
u_{\pi N}^{\rm III}(k_3)=0.197.
\end{equation}
\begin{figure}[t]
\begin{center}
\includegraphics[clip=true,trim=2.9cm 0.0cm 2.9cm 0.0cm,width=0.48\columnwidth]{MassDepState2mass5.png}
\includegraphics[clip=true,trim=2.9cm 0.0cm 2.9cm 0.0cm,width=0.48\columnwidth]{MassDepState2mass2.png}
\caption{{\bf Colour online:} Lattice QCD calculation of the $d$-quark probability distribution in
the first excited state of the proton, reproduced from Ref.~\cite{Roberts:2013oea}. The CSSM's
lattice calculations at the (left) lightest quark mass (state 6 in scenario III) and (right)
second heaviest quark mass (state 3 in scenario III) reveal the effect of the finite periodic
volume of the lattice with $L \simeq 2.90$ fm. Here, two $u$ quarks are fixed at the origin at the centre
of the plot and the lattice spacing is approximately 0.09 fm. }
\label{figWavefunction}
\end{center}
\end{figure}
Another concern is that the effects of the small finite volume may induce distortions that cannot
be accounted for by meson-baryon dressings alone. Figure \ref{figWavefunction} reproduced from
Ref.~\cite{Roberts:2013oea} illustrates the influence of the periodic volume of the $d$-quark
probability distribution in the first excited state of the nucleon. This lattice QCD calculation
was carried out on lattice volumes with $L \simeq 2.90$ fm. Even at large quark masses
corresponding to $m_\pi^2 = 0.27$ GeV${}^2$ distortions in the spherically symmetric $2S$ radial
wave function are readily observed. They become even more severe as the value of the isovolume cut
is lowered and the tails of the wave function are considered \cite{Roberts:2013oea}. Therefore we
caution that an even smaller volume with $L \simeq 1.98$ fm is likely to have finite volume
distortions that cannot be described by effective field theory alone.
\section{Conclusion}\label{secSum}
We have studied the infinite volume phase shifts and inelasticities for scattering states with
$I(J^P)=\frac12(\frac12^+)$ in effective field theory. Through the consideration of a
finite-volume Hamiltonian Matrix we have explored the corresponding finite-volume spectra of states
relevant to contemporary lattice QCD calculations of the spectrum.
In doing so we have explored three scenarios for the underlying theory describing the available
data. All three scenarios are able to describe the scattering data well and all three create a
pole position for Roper similar to that reported by the PDG. However the finite volume spectrum
predicted by the scenarios has important differences.
In the first case, the Roper is postulated to have a triquark-like bare or core component with a
mass exceeding the resonance mass. This component mixes with attractive virtual meson-baryon
contributions, including the $\pi N$, $\pi \Delta$, and $\sigma N$ channels, to reproduce the
observed pole position.
With the advent of new insight from the Hamiltonian model and lattice QCD results we have been able to discard
this popular description of including a bare Roper basis state. This model predicts a low-lying state
in the finite volume having a very large bare-state component that makes it accessible to current
lattice QCD techniques. The absence of this state in today's lattice QCD calculations exposes an
inconsistency in the model predictions.
In the second hypothesis, the Roper resonance is dynamically generated purely from the $\pi N$,
$\pi \Delta$, and $\sigma N$ channels in the absence of a bare-baryon basis state. This scenario
identifies the lattice QCD results as non-trivial superpositions of the basis states that have a
qualitative difference from the weak mixing of basis states in the scattering channels. However,
given the presence of a bare state associated with the ground state nucleon, we proceed to consider
a third scenario incorporating the presence of this basis state.
In the third scenario the Roper resonance is composed of the low-lying bare basis component
associated with the ground state nucleon. The merit of this scenario lies in its ability to not
only identify and describe the finite-volume energy levels to be observed in contemporary
large-volume lattice QCD simulations but also explain why other low-lying states have been missed
in today's lattice QCD results for the nucleon spectrum.
We conclude that the Roper resonance of Nature is predominantly a dynamically-generated molecular
meson-baryon state with a weak coupling to a low-lying bare basis state associated with the ground
state nucleon.
This conclusion is in sharp contrast to a conventional state with a large three-quark core
component like the ground state nucleon or even the $N(1535)$ resonance where a significant
bare-state contribution was manifest \cite{Liu:2015ktc}. It also suggests that relativistic
three-quark bound state approaches \cite{Eichmann:2016yit} will fail as these models do not have
the full influence of the meson-baryon sector required to generate the full coupled-channel
physics.
Future work should investigate the role of three-body coupled channel effects in the structure of
the Roper resonance. Of particular interest in the role of the $N(\pi \pi)_\texttt{S-wave}$
channel. While our consideration of the $\sigma N$ channel does model the effects of the $N(\pi
\pi)_\texttt{S-wave}$ channel, the $\sigma$ meson is a broad state with a large width and it is
desirable to accommodate this important physics in a more direct manner. For example the imaginary
part of the Roper pole position is likely to be sensitive to this physics.
Similarly, it may be interesting to explore other models of the Roper resonance and their finite
volume implementation. For example one could further explore the nature of the bare basis state
and its impact on resonance structure.
It is also desirable to advance lattice QCD simulations to include five-quark interpolating
fields where the momentum of each of the meson-baryon pairs can be defined at the source. Not only
does this approach address the volume suppression of multi-particle states through a double sum in
the Fourier projection, it also enables the creation of a state very similar to the scattering
state in the finite volume of the lattice. With this approach it should be possible to observe all
the states predicted by the Hamiltonian model and eventually reverse the process such that the
experimental phase shifts and inelasticities are determined from the finite volume spectra of
lattice QCD.
Such developments will be key in obtaining a full understanding of the Roper resonance.
After this paper was submitted, a very important lattice QCD simulation was released by Lang {\it
et al.} \cite{Lang:2016hnn}. In addition to standard three-quark operators, these authors included
explicit momentum-projected $\pi N$ and $\sigma N$ interpolating fields in a lattice QCD
analysis of the Roper channel. The $\sigma N$ operator was included to simulate the effect of the
$N \pi \pi$ channel. By comparing their energy levels with those calculated here they reached
similar conclusions. Their results provide strong support for the third scenario and disfavor the
first scenario considered herein. The success of the Hamiltonian effective field theory in predicting
the position of these energy levels confirms the consideration of resonant two-body channels (such as the $\sigma N$ and $\pi \Delta$ channels) is effective in linking lattice QCD results to the Roper resonance of nature. We note the inclusion of the $\pi \Delta$ contributions is essential for describing the inelasticity of the $\pi N$ to $\pi N$ amplitude. We anticipate that when the $\pi \Delta$ channel is explored in future lattice QCD simulations, a new low-lying energy level will be observed consistent with our 5'th state at 1.7 GeV for the lightest quark mass.
\begin{acknowledgments}
We would like to thank T.-S.H. Lee for helpful discussions.
We thank the PACS-CS Collaboration for making the $2+1$ flavor configurations underpinning the
current investigation available and the ILDG for their ongoing support in the sharing of these
configurations.
This research was undertaken with the assistance of resources at the NCI National Facility in
Canberra, the iVEC facilities at the Pawsey Centre and the Phoenix GPU cluster at the University of
Adelaide, Australia. These resources were provided through the National Computational Merit
Allocation Scheme, supported by the Australian Government, and the University of Adelaide through
their support of the NCI Partner Share and the Phoenix GPU cluster.
This research is supported by the Australian Research Council through the ARC Centre of Excellence
for Particle Physics at the Terascale (CE110001104), and through Grants No.\ LE160100051, DP151103101 (A.W.T.), DP150103164,
DP120104627 and LE120100181 (D.B.L.).
\end{acknowledgments}
| 82bb706264abda72c2c0a4c1d66d4e60af05f2a8 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction and results} \label{sec:Intro}
Consider an $n$-variate polynomial of degree at most $d$:
\[ p=\sum_{|\alpha| \le d} c_{\alpha} x^{\alpha} \]
where $x:=(x_1,\ldots,x_n)$, $\alpha \in \ZZ_{\ge 0}^n$,
$|\alpha|:=\sum_i \alpha_i$, $x^{\alpha}:=\prod_i x_i^{\alpha_i}$,
and where each coefficient $c_\alpha$ is taken from a ground field $K$.
A {\em determinantal representation} of $p$ is an $N \times N$-matrix $M$
of the form
\[ M=A_0 + \sum_{i=1}^n x_i A_i, \]
where each $A_i \in K^{N \times N},$ with $\det(M)=p$. We call $N$
the {\em size} of the determinantal representation. Clearly, since the
entries of $M$ are affine-linear forms in $x_1,\ldots,x_n$, $N$ must be
at least the degree of $p$.
Determinantal representations of polynomials play a fundamental
role in several mathematical areas: from {\em algebraic geometry}
it is known that each plane curve ($n=2$) of degree $d$ over an
algebraically closed field $K$ admits a determinantal representation of
size $d$ \cite{Dickson,Dixon}. Over non-algebraically closed fields, and
especially when restricting to symmetric determinantal representations,
the situation is much more subtle \cite{IshitshukaIto}. For larger $n$,
only certain hypersurfaces have a determinantal representation of size
equal to their degree \cite{Beauville,Dickson}. In {\em optimisation},
and notably in the theory of {\em hyperbolic polynomials} \cite{Wagner},
one is particularly interested in the case where $K=\RR$, $A_0$ is
symmetric positive definite, and the $A_i$ are symmetric. In this case,
the restriction of $p$ to any line through 0 has only real roots. For
$n=2$ the converse also holds \cite{HeltonVinnikov,LewisParriloRamana};
for counterexamples to this converse holding for higher $n$, see \cite{Branden}.
In {\em complexity theory} a central role is played by Valiant's
conjecture that the permanent of an $m \times m$-matrix does not admit a
determinantal representation of size polynomial in $m$ \cite{Valiant}. Via
the {\em geometric complexity theory} programme \cite{MulmuleySohoniI}
this leads to the study of polynomials in the boundary of the orbit of
the $N \times N$-determinant under the action of the group $\GL_{N^2}(K)$
permuting matrix entries. Recent developments in this field include the study of
this boundary for $N=3$ \cite{HuttenhainLairez} and the exciting negative
result in \cite{BurgisserIkenmeyerPanova} that Valiant's conjecture can
{\em not} be proved using occurrence obstructions proposed earlier in
\cite{MulmuleySohoniII}.
Our motivation comes from {\em scientific computing}, where determinantal
representations of polynomials have recently been proposed for efficiently
solving systems of equations \cite{BorMichiel}. For this application, it is
crucial to have determinantal representations not of a {\em single}
polynomial $p$, but rather of all $n$-variate polynomials of degree
at most $d$. Moreover, the representation should be easily computable
from the coefficients of $p$. Specifically, in \cite{BorMichiel} determinantal
representations are constructed for the bivariate case ($n=2$) in which
the entries of the matrices $A_0,\ldots,A_n$ themselves {\em depend
affine-linearly} on the coefficients $c_\alpha$. This is what we call a
{\em uniform determinantal representation} of the generic polynomial $p$
of degree $d$ in $n$ variables; see Section~\ref{sec:Problem} for a
precise definition.
\begin{ex}[The binary quadric] \label{ex:BinaryQuadric}
{\rm
The identity
\[ c_{00} + c_{10}x + c_{01}y + c_{20} x^2 + c_{11} xy +
c_{02} y^2= \det
\begin{bmatrix}
-x & 1 & 0 \\
-y & 0 & 1 \\
c_{00} & c_{10}+c_{20}x+c_{11}y & c_{01}+c_{02}y\\
\end{bmatrix}
\]
exhibits the matrix on the right as a uniform determinantal representation
of the generic bivariate quadric.
\hfill $\clubsuit$
}
\end{ex}
In applications, the matrix $M$ is used as input to algorithms
in numerical linear algebra that scale unfavourably with $N$,
such as a complexity of $O(N^6)$.
Consequently, we are led to consider the following fundamental question.
\begin{que}
What is the minimal size $N^*(n,d)$ of any uniform determinantal
representation of the generic polynomial of degree $d$ in $n$ variables?
\end{que}
A construction from \cite{BorMichiel} shows that
for fixed $n=2$ and $d \to \infty$ we have
$N^*(2,d) \le \frac{1}{4} \, d^2 + O(d)$;
this construction is reviewed in Section~\ref{sec:First}.
We improve the construction from \cite{BorMichiel} by giving a particularly elegant uniform
determinantal representations of bivariate polynomials of size $2d+1$ in
Example~\ref{ex:2d}, and of size $2d-1$ in Example~\ref{ex:2d1}.
In view of the obvious lower bound of $d$ this is
clearly sharp up to a constant factor for $d \to \infty$, although we do
not know where in the interval $[d,2d-1]$ the true answer lies.
We show in Section~\ref{sec:Numerics} how to use these small determinantal
representations of bivariate polynomials for solving systems of equations.
Before that, we focus on the asymptotic behaviour of $N^*(n,d)$
for fixed $n$ and $d \to \infty$. In this setting, we derive the
following result.
\begin{thm} \label{thm:Main}
For fixed $n \in \ZZ_{\ge 2}$ there exist positive constants $C_1,C_2$
(depending on $n$) such that for each $d \in \ZZ_{\ge 0}$ the smallest
size $N^*(n,d)$ of a uniform determinantal representation of the generic
polynomial of degree $d$ in $n$ variables satisfies
$C_1 d^{n/2} \le N^*(n,d) \le C_2 d^{n/2}$.
\end{thm}
We will also compare our results with previous constructions,
most notably with those by Quarez \cite[Thm.~4.4]{Quarez}, who proves the
existence of a symmetric
representation of size ${n+\lfloor \frac{d}{2} \rfloor} \choose n$.
For fixed $n$ and $d \to \infty$, \cite{Quarez} therefore has the asymptotic
rate $\sim d^n$, meaning that the results of this paper represent a clear improvement.
For fixed $d$ and $n \to \infty$, \cite{Quarez} leads to the asymptotic behavior
$\sim n^{\lfloor d/2 \rfloor}$, which is similar to our bounds;
we will discuss more details in Section~\ref{sec:Outlook}.
In Section~\ref{sec:Problem} we formalise the notion of uniform
determinantal representations, study their symmetries, and
derive some simple properties. In particular, we relate uniform
determinantal representations to spaces of singular $N \times N$-matrices. In
Section~\ref{sec:Singular} we briefly review some of the existing literature on
these singular spaces, and we prove that for $N>4$ there are
infinitely many equivalence classes of such objects; this poses an
obstruction to a ``brute-force'' approach towards finding lower bounds
on $N^*(n,d)$. In Section~\ref{sec:First} we present a first
construction, of which however the size is of the order of $d^n$,
rather than $d^{n/2}$, for $d \to \infty$.
In Section~\ref{sec:Second} we give a more efficient construction
and prove Theorem~\ref{thm:Main}.
In Section~\ref{sec:Small} we give upper bounds on $N^*(n,d)$
for small $n$ and $d$ and
determine $N^*(2,2)$ and $N^*(3,2)$ exactly. We extend representations
from scalar to matrix polynomials in Section~\ref{sec:MatPol}.
In Section~\ref{sec:Numerics} we give some numerical results that
show that for $n=2$ and small $d$ we get a competitive method
for computing zeros of polynomials systems.
Finally, in Section~\ref{sec:Outlook} we summarise our main conclusions
and collect some questions that arise naturally from our work.
\section{Problem formulation and symmetries} \label{sec:Problem}
In this section we give a formal definition of uniform determinantal
representations, and introduce a group that acts on such representations.
We also show that a uniform determinantal representation gives rise to
a vector space consisting entirely of singular matrices; such spaces
are the topic of next section.
Let $K$ be a field and fix $d,n \in \ZZ_{\ge 0}$. Let $F_d$ denote
the polynomials of degree at most $d$ in the polynomial ring $K[x_1,\ldots,x_n]$.
Furthermore, let $p_{n,d}$ be the generic polynomial of that degree, i.e.,
\begin{equation}\label{eq:polnd}
p_{n,d}=\sum_{|\alpha| \le d} c_\alpha x^\alpha,
\end{equation}
where $x:=(x_1,\ldots,x_n)$, $\alpha \in \ZZ_{\ge 0}^n$,
$|\alpha|:=\sum_i \alpha_i$, $x^{\alpha}:=\prod_i x_i^{\alpha_i}$,
and where we consider $c_\alpha$ as a variable for each $\alpha$.
\begin{de}
For $n,d \in \ZZ_{\ge 0}$, a {\em uniform determinantal representation}
of $p_{n,d}$ is an $N \times N$-matrix $M$ with entries from
$K[(x_1,\ldots,x_n),(c_\alpha)_{|\alpha| \le d}]$, of degree at most 1
in each of these two sets of variables, such that $\det(M)=p_{n,d}$. The
number $N$ is called the {\em size} of the determinantal representation.
\end{de}
To be explicit, we require each entry of $M$ to be a $K$-linear
combination of the monomials $1,x_i,c_\alpha,c_\alpha x_i,\
(i=1,\ldots,n, |\alpha| \le d)$. This means that we can decompose $M$
as $M_0+M_1$, where $M_0$ contains all terms in $M$ that do not
contain any $c_\alpha$, and where $M_1$ contains all terms in $M$ that
do. We will use the notation $M=M_0+M_1$ throughout the paper. When $n$
and $d$ are fixed in the context, we will also speak of a
uniform determinantal representation without reference to $p_{n,d}$. Our
ultimate aim is to determine the following quantity.
\begin{de}
For $n,d \in \ZZ_{\ge 0}$, $N^*(n,d) \in \ZZ_{>0}$ is the minimum among
all sizes of uniform determinantal representations of $p_{n,d}$.
\end{de}
This minimal size could potentially depend on the ground field $K$,
but the bounds that we will prove do not. Note that in the definition
of $N^*(n,d)$ we do not allow terms in $M$ of degree strictly larger
than one in the $c_\alpha$. Relaxing this condition to polynomial
dependence on the $c_\alpha$ might affect the exact value of $N^*$,
but it will not affect our bounds---see Remark~\ref{re:Poly}.
Given a uniform determinantal representation $M$ of size $N$, and given
matrices $g,h$ in $\SL_N(K)$, the group of determinant-one matrices
with entries in $K$,
the matrix $gMh^{-1}$ is another uniform determinantal representation
of $p_{n,d}$. In this manner, the group $\lieg{SL}_N(K) \times
\lieg{SL}_N(K)$ acts on the set of uniform determinantal representations
of $p_{n,d}$. Moreover, there exist further symmetries, arising from affine
transformations of the $n$-space. Recall that these transformations form the
group $\AGL_n(K)=\GL_n(K) \ltimes K^n$ generated by invertible linear
transformations and translations.
\begin{lm} \label{lm:Affine}
The group $\AGL_n(K)$ acts on uniform determinantal representations of $p_{n,d}$.
\end{lm}
The statement of this lemma is empty without making the action
explicit, as we do in the proof.
\begin{proof}
Let $g \in \AGL_n(K)$ be an affine transformation of $K^n$, and expand
\[ p_{n,d}(g^{-1}x,c)=\sum_{|\alpha| \le d} c'_\alpha x^\alpha, \]
where the $c'_\alpha$ are linear combinations of the $c_\alpha$. More
precisely, the vector $c'$ can be written as $\rho(g)c$, where $\rho$
is the representation of $\AGL_n(K)$ on polynomials of degree at most $d$
regarded as a matrix representation relative to the monomial basis.
Now let $M=M(x,c)$ be a uniform determinantal representation of
$p_{n,d}$. Then
\[
\det(M(g^{-1}x,\rho(g)^{-1}c))=p_{n,d}(g^{-1}x,\rho(g)^{-1}c)=p_{n,d}(x,c),
\]
i.e., $M(g^{-1}x,\rho(g)^{-1}c)$ is another uniform determinantal
representation of $p_{n,d}$. The action of $g$ is given by $M \mapsto
M(g^{-1}x,\rho(g)^{-1}c)$.
\end{proof}
\begin{ex}[The binary quadric revisited]
{\rm
Take $n=d=2$ and the affine transformation
$g(x,y):=(y,x+1)$ with inverse $g^{-1}(x,y)=(y-1,x)$. We have
\[ p_{2,2}(g^{-1}(x,y))=
(c_{00} - c_{10} + c_{20})
+ (c_{01} - c_{11})x
+ (c_{10} - 2 c_{20})y
+ c_{02} x^2
+ c_{11} xy
+ c_{20} y^2.
\]
We find
{\footnotesize
\[
c'=
\begin{bmatrix}
c'_{00}\\c'_{10}\\c'_{01}\\c'_{20}\\c'_{11}\\c'_{02}
\end{bmatrix}
=
\begin{bmatrix}
1 & -1 & 0 & 1 & 0 & 0\\
0 & 0 & 1 & 0 &-1 & 0\\
0 & 1 & 0 &-2 & 0 & 0\\
0 & 0 & 0 & 0 & 0 & 1\\
0 & 0 & 0 & 0 & 1 & 0\\
0 & 0 & 0 & 1 & 0 & 0
\end{bmatrix}
\begin{bmatrix} c_{00}\\c_{10}\\c_{01}\\c_{20}\\c_{11}\\c_{02}
\end{bmatrix}
=\rho(g)c; \text{ }
\rho(g)^{-1}=
\begin{bmatrix}
1& 0& 1& 0& 0& 1\\
0& 0& 1& 0& 0& 2\\
0& 1& 0& 0& 1& 0\\
0& 0& 0& 0& 0& 1\\
0& 0& 0& 0& 1& 0\\
0& 0& 0& 1& 0& 0
\end{bmatrix}.
\]
If we make the substitutions
\begin{align*}
c_{00} &\mapsto c_{00}+c_{01}+c_{02} &
c_{10} &\mapsto c_{01}+2c_{02} &
c_{01} &\mapsto c_{10}+c_{11} &
x &\mapsto y-1\\
c_{20} &\mapsto c_{02} &
c_{11} &\mapsto c_{11} &
c_{02} &\mapsto c_{20} &
y &\mapsto x
\end{align*}
in the uniform determinantal representation of
Example~\ref{ex:BinaryQuadric}, then we arrive at the matrix
\[
\begin{bmatrix}
1 - y& 1& 0\\
-x& 0& 1\\
c_{00} + c_{01} + c_{02}& c_{01} + c_{02} + c_{02}y + c_{11}x& c_{10}
+ c_{11} + c_{20}x
\end{bmatrix}
\]
whose determinant also equals $p_{2,2}$.
\hfill $\clubsuit$
}
\end{ex}
The action of the affine group will be used in Section~\ref{sec:Small}
to determine the exact value of $N^*(n,2)$ for $n=2$ and $3$. We now turn our
attention to the component $M_0$ of a uniform determinantal representation $M$.
\begin{lm} \label{lm:SingSpace}
For any uniform determinantal representation $M=M_0+M_1$ of
size $N$, the determinant of $M_0$ is the zero polynomial in
$K[x_1,\ldots,x_n]$. Moreover, at every point $\bar{x} \in K^n$, the rank
of the specialisation $M_0(\bar{x}) \in K^{N \times N}$ is exactly
$N-1$.
\end{lm}
\begin{proof}
The first statement follows from the fact that $\det(M_0)$ is the part
of the polynomial $\det(M)$ which is homogeneous of degree zero in the
$c_\alpha$; hence zero.
By specialising the vector $x$ of variables to a point $\bar{x} \in K^n$,
the rank of $M_0$ can only drop, so the rank of $M_0(\bar{x})$ is at most
$N-1$. However, if it were at most $N-2$, then after column operations on
$M$ by means of determinant-one matrices with entries in $K$ we may assume
that $M_0(\bar{x})$ has its last two columns equal to 0. This means that
all entries of $M(\bar{x})=M_0(\bar{x})+M_1(\bar{x})$ in these columns are linear
in the $c_\alpha$. This in turn implies that any term in the polynomial $\det
M(\bar{x})$ is at least quadratic in the $c_\alpha$. But on the other hand
$\det M(\bar{x})$ equals $p_{n,d}(\bar{x})$, which is a non-zero {\em linear}
polynomial in the $c_\alpha$ (nonzero since not every polynomial of
degree at most $d$ vanishes at $\bar{x}$). This contradiction implies that
the rank of $M_0(\bar{x})$ is $N-1$.
\end{proof}
\begin{lm} \label{lm:Dets}
If $M=M_0+M_1$ is a uniform determinantal representation of size
$N$, then $V \subseteq F_{N-1}$ spanned by the $(N-1) \times
(N-1)$-subdeterminants of $M_0$ satisfies $F_1 \cdot V \supseteq F_d$.
\end{lm}
Here, as in the rest of this paper, by the product of two spaces of
polynomials we mean the $K$-linear span of all the products.
\begin{proof}
Let $D_{ij}$ be the determinant of the submatrix of $M_0$ obtained
by deleting the $i$th row and the $j$th column. On the one hand,
$\det(M)=p_{n,d}$ is linear in the $c_\alpha$ by assumption, and on the
other hand, by expanding $\det(M)$ we see that the part that is homogeneous of
degree one in the $c_\alpha$ is
\[ \sum_{i,j}(-1)^{i+j} (M_1)_{ij} D_{ij}; \]
this therefore equals $p_{n,d}$. Hence any element $q$ of $F_d$
is obtained from the expression above by specialising the variables
$c_\alpha$ to the coefficients of $q$. Since each $(M_1)_{ij}$ is then
specialised to an element of $F_1$, we find $q \in F_1 \cdot V$.
\end{proof}
\begin{re} \label{re:Poly}
{\rm
Note that the proof above still applies if we allow determinantal
representations of the form $M_0+M_1+\cdots+M_e$ where $M_r$ is
homogeneous of degree $r$ in the $c_\alpha$ and affine-linear in the
$x_i$. Since our upper bound in
Section~\ref{sec:Second} builds directly on this lemma, the upper bound
holds in this more general setting, as well.
}
\end{re}
\section{Spaces of singular matrices} \label{sec:Singular}
Let $M=M_0+M_1$ be a uniform determinantal representation of $p_{n,d}$.
Writing $M_0=B_0+\sum_{i=1}^n x_i B_i$, Lemma~\ref{lm:SingSpace} implies
that the linear span $\langle B_0,\ldots,B_n \rangle_K \subseteq K^{N
\times N}$ consists entirely of singular matrices (and indeed that this
remains true when extending scalars from $K$ to an extension field). There
is an extensive literature on such {\em singular matrix spaces}; see,
e.g., \cite{DraismaBLMS,FillmoreLaurieRadjavi} and the references therein.
The easiest examples are the following.
\begin{de}
A subspace $\mathcal{A} \subseteq K^{N \times N}$ is called a {\em
compression space} if there exists a subspace $U \subseteq K^N$ with
$\dim(\langle u^T A \mid A \in \mathcal{A},u \in U\rangle_K)<\dim U$. We call
the space $U$ a {\em witness} for the singularity of $\mathcal{A}$.
\end{de}
Given any two subspaces $U,V \subseteq K^N$ with $\dim V=-1+\dim U$,
the space of all matrices which map $U$ into $V$ (acting on row vectors)
is a compression space with witness $U$. It is easy to see that these
spaces are inclusion-wise maximal among all singular spaces.
If $\mathcal{A}$ is a singular matrix space, then so is $g \mathcal{A}
h^{-1}$ for any pair $(g,h) \in \GL_N(K) \times \GL_N(K)$. We call the latter space {\em
conjugate} to the former.
\begin{ex} \label{ex:SmallSingSpace}
{\rm
For $N=2$, every singular matrix space is a compression space, hence
conjugate to a subspace of one of the two spaces
\[
\left\{\begin{bmatrix} * & * \\ 0 & 0 \end{bmatrix} \right\},
\left\{\begin{bmatrix} 0 & * \\ 0 & *\end{bmatrix} \right\},
\]
where the $*$s indicate entries that can be filled arbitrarily.
A witness for the first space is the span $\langle e_2 \rangle$ of the
second standard basis vector, and a witness for the second space is $K^2$.
For $N=3$, there are four conjugacy classes of
inclusion-maximal singular matrix spaces, represented by the three
maximal compression spaces
\[
\left\{\begin{bmatrix} * & * & * \\ * & * & * \\ 0 & 0 &
0\end{bmatrix} \right\},
\left\{\begin{bmatrix} * & * & * \\ 0 & 0 & * \\ 0 & 0 &
*\end{bmatrix} \right\},
\left\{\begin{bmatrix} 0 & * & * \\ 0 & * & * \\ 0 & * &
*\end{bmatrix} \right\},
\]
and the space of skew-symmetric $3 \times 3$-matrices \cite{EisenbudHarris}; the latter
is not a compression space.
For $N=4$, there are still finitely many (namely, $10$) conjugacy classes of
inclusion-maximal singular matrix spaces
\cite{EisenbudHarris,FillmoreLaurieRadjavi}, but this is not true
for $N \ge 5$, as Theorem~\ref{thm:Sing5} below shows.
This theorem is presumably folklore; we include a proof since
we have not been able to find a literature reference for it.
\hfill $\clubsuit$
}
\end{ex}
\begin{prop} \label{prop:Max}
Assume that $K$ is algebraically closed. For any $m$ and $N\in \ZZ_{\ge
0}$ the locus $X_m$ in the Grassmannian $\Gr(m, K^{N \times N})$ of
$m$-dimensional subspaces of $K^{N \times N}$ consisting of all {\em
singular}
subspaces is closed in the Zariski topology. Moreover, the locus $U_m$ in
$X_m$ consisting of all {\em inclusion-wise maximal} singular
subspaces is open inside $X_m$.
\end{prop}
\begin{proof}
The first statement is standard. For the second statement, consider the
incidence variety
\[
Z:=\{(\mathcal{A},\mathcal{A'}) \in X_m \times X_{m+1} \mid
\mathcal{A} \subseteq \mathcal{A'} \} \subseteq X_m \times X_{m+1},
\]
which is a closed subvariety of $X_m \times X_{m+1}$.
The projection of $Z$ into $X_m$ is the complement of $U_m$, and it is
closed because $X_{m+1}$ is a projective variety.
\end{proof}
\begin{thm} \label{thm:Sing5}
Assume that $K$ is infinite and of characteristic unequal to two. For
$N \ge 5$ there are infinitely many conjugacy classes of inclusion-wise
maximal singular $N \times N$-matrix spaces.
\end{thm}
\begin{proof}
Take $N \ge 5$. For sufficiently general skew-symmetric matrices
$A_1,\ldots,A_N \in K^{N \times N}$ set $A:=(A_1,\ldots,A_N)$ and define
the space
\[ \cB_A:=\{(A_1x|\cdots|A_Nx) \mid x \in K^N\} \subseteq K^{N \times N}. \]
Each matrix in this space is singular, since for $x \neq 0$ we have
\[ x^T(A_1x|\cdots|A_Nx)=(x^TA_1x,\ldots,x^TA_Nx)=0. \]
In \cite{FillmoreLaurieRadjavi} it is proved that, for a specific
choice of the tuple $A$, the
space $\cB_A$ is maximal among the singular subspaces of $K^{N
\times N}$. By
Proposition~\ref{prop:Max}, $\cB_A$ is maximal for sufficiently general
$A$, as well (note that we may first extend $K$ to its algebraic closure
to apply the proposition). In the notation of that proposition,
we have a rational map
\[
\phi:S^N \dashrightarrow U_N,\quad A \mapsto \cB_A,
\]
where $S \subseteq K^{N \times N}$ denotes the subspace of
skew-symmetric matrices; the dashed arrow
indicates that the map is defined only in an open dense
subset of $S^N$. For any nonzero scalar $t$, $\phi(tA)=\phi(A)$.
We claim that, in fact, the general fibre of $\phi$ is indeed
one-dimensional. As the fibre dimension is semicontinuous, it suffices
to verify this at a particular point where $\phi$ is defined. We take
$A_i=E_{i,i+1}-E_{i+1,i}$ for $i=1,\ldots,N-1$ and $A_N$ general;
here $E_{ij}$ is the matrix with zeros everywhere except for a 1
at position $(i,j)$. Let $B \in S^N$; if $\phi(A)=\phi(B)$, then there
exists an invertible matrix $g \in \GL_N(K)$ such that
\[ (A_1 g x|\cdots|A_N g x)=(B_1 x |\cdots|B_N x) \]
for all $x$, so that $A_i g =B_i$. Using skew-symmetry of $A_i$ and
$B_i$, we find that $A_ig=g^T \! A_i$. Substituting our choice of $A_i$
for $i \in \{1,\ldots,N-1\}$ yields $g_{i,j}=g_{j,i}=0$ for all $j$
with $|i-j|>1$, $g_{i,i+1}=-g_{i,i+1}$, so $g_{i,i+1}=0$ since $\cha
K \neq 2$, and $g_{i,i}=g_{i+1,i+1}$. Hence $g$ is a scalar multiple
of the identity. It follows that the fibre of $\phi$ through $A$ is
one-dimensional as claimed.
Since $\dim S=\binom{N}{2}$, we have thus constructed an
$(N\binom{N}{2}-1)$-dimensional family inside $U_N$. Given any point
$\mathcal{A}$ in $U_N$, its orbit under $\GL_N(K) \times \GL_N(K)$ has dimension
at most $2 \, (N^2-1)$ (scalars act trivially). Now for $N=5$ we have
\[
N \, \binom{N}{2}-1=5\cdot 10 -1=49 \quad
\text{ and } \quad 2 \, (N^2-1)=48,
\]
so that we have found (at least) a one-parameter family of conjugacy
classes of singular spaces. For $N>5$ the difference between
$N\binom{N}{2}-1$ and $2 \, (N^2-1)$ is even larger.
\end{proof}
For large $N$ it seems impossible to classify maximal singular matrix
spaces. The construction above already gives an infinite number of
conjugacy classes, but there are many other sources of examples.
For instance, for infinitely many $N$ there exists a maximal singular
matrix space in $K^{N \times N}$ of constant dimension $8$, at least if
we assume that $K$ has characteristic 0 \cite{DraismaBLMS}. On the other hand, if
the singular matrix space $\mathcal{A}$ has dimension at least $N^2-N$,
then it is a compression space with either a one-dimensional witness
or all of $K^N$ as witness \cite{Dieudonne} (and hence of dimension exactly
$N^2-N$). A sharpening of this result is proved in \cite{FillmoreLaurieRadjavi} (see also
\cite{deSeguinsPazzis}).
It should be noted that in many cases not even the
dimension of such singular matrix spaces is known, for fixed values of the size and rank of the matrices.
There is a considerable body of work devoted to giving
lower and upper bounds for such dimensions, both in the
case of bounded and constant rank, but these bounds are rarely sharp, see,
among many other references, \cite{Flanders, IlicLandsberg, Sylvester, Westwick}
and the more recent works on skew-symmetric matrices of constant rank
\cite{BoraleviFaenziMezzetti,ManivelMezzetti}.
Hence the fact that $M_0$ represents a singular matrix space of
dimension (at most) $n+1$ does not much narrow down our search for good
uniform determinantal representations, except in small cases discussed
in Section~\ref{sec:Small}. However, for our constructions in
Sections~\ref{sec:First} and~\ref{sec:Second} we will only use compression
spaces where the witness has dimension 1 or about $\frac{1}{2} N$, respectively;
and our lower bounds on $N^*(n,d)$ are independent of the literature on
singular matrix spaces.
\section{A first construction} \label{sec:First}
In this section we restrict our attention to determinantal
representations $M=M_0+M_1$ where $M_0$ represents a compression space
with a one-dimensional witness (or, dually by transposition, with a
full-dimensional witness). Under this assumption we will prove quite
tight bounds on the minimal size of a uniform matrix representation. The
following fundamental notion will be used throughout below.
\begin{de}
We say that a subspace $V \subseteq K[x_1,\ldots,x_n]$ is
{\em connected to $1$}
if it is nonzero and its intersections $V_e:=V \cap F_e$ satisfy
$F_1 \cdot V_e \supseteq V_{e+1}$ for each $e \ge 0$.
\end{de}
Note that this implies that $V_0=\langle 1 \rangle$. We borrow the
terminology from the theory of border bases \cite{LLMRT}, where a set
$S$ of monomials is called connected to $1$ if $1 \in S$ and each
nonconstant monomial in $S$ can be divided by some variable to obtain
another monomial in $S$. The linear span of $S$ is then
connected to $1$
in our sense. Translating monomials to their exponent vectors, we will
call a subset $S$ of $\ZZ_{\ge 0}^n$ {\em connected to
$0$} if it contains
0 and for each $\alpha \in S \setminus \{0\}$ there exists an $i$
such that $\alpha-e_i \in S$, where $e_i$ is the $i$-th
standard basis vector.
Let $V$ be a finite-dimensional subspace of $K[x_1,\ldots,x_n]$ connected
to 1. Choose a $K$-basis $f_1,\ldots,f_m$ of $V$ whose total degrees
increase weakly. For each $i=2,\ldots,m$ write
\[ f_i=\sum_{j<i} \ell_{ij} f_j \]
for suitable elements $\ell_{ij} \in F_1$. Let $M_V$ be the $(m-1)
\times m$-matrix whose $i$th row equals
\[ (-\ell_{i1},-\ell_{i2},\ldots,-\ell_{i,i-1},1,0,\ldots,0). \]
Note that $M_V$ depends on the choice of basis, but we suppress this
dependence in the notation, since the property of $M_V$ in the next
lemma does not depend on the choice of basis.
\begin{lm} \label{lm:Block}
The $K$-linear subspace of $K[x_1,\ldots,x_n]$ spanned by
the $(m-1) \times (m-1)$-subdeterminants of $M_V$ equals $V$.
\end{lm}
\begin{proof}
By construction, $M_V$ has rank $m-1$ over the field $K(x_1,\ldots,x_n)$
and satisfies $M_V \cdot (f_1,\ldots,f_m)^T=0$. By (a version
of) Cramer's rule, the kernel of $M_V$ is also spanned by
$(D_1,-D_2,\ldots,(-1)^{m-1}D_m)$ where $D_j$ is the determinant of the
submatrix of $M_V$ obtained by removing the $j$th column. So these two
vectors differ by a factor in $K(x_1,\ldots,x_n)$. Since $D_1=1=f_1$
we find that they are, in fact, equal. Hence $\langle D_1,\ldots,D_m
\rangle=V$ as claimed.
\end{proof}
We can now formulate our first general construction. This
generalises a construction from \cite{BorMichiel} to the multivariate case.
\begin{prop} \label{prop:Cons1}
Let $V$ be an $m$-dimensional space connected to $1$ and suppose that
$F_1 \cdot V \supseteq F_d$. Then there exists a uniform determinantal
representation of size $m$ for the generic polynomial of degree $d$
in $n$ variables.
\end{prop}
\begin{proof}
Let $D_1,\ldots,D_m \in K[x_1,\ldots,x_n]$ be the
$(m-1)\times(m-1)$-subdeterminants of $M_V$. By $F_1 \cdot V=F_d$
and Lemma~\ref{lm:Block} we can find, for each monomial $x^\alpha$
of degree at most $d$, affine-linear forms
$l_{\alpha 1},\ldots,l_{\alpha m} \in F_1$ such that $x^{\alpha}=\sum_j
(-1)^{j-1} l_{\alpha j} D_j$. Define
\[ s:=\sum_{|\alpha| \le d} c_\alpha (l_{\alpha
1},\ldots,l_{\alpha m}), \]
a row vector of bi-affine linear forms in the $x_i$ and the
$c_{\alpha}$. Then, by Laplace expansion along the first row, we find
that the determinant of
\[ \begin{bmatrix} s\\M_V \end{bmatrix} \]
is the generic polynomial of degree $d$ in $x_1,\ldots,x_n$.
\end{proof}
\begin{ex}\label{ex:lin1}
{\rm
For $n=2$ the following picture gives a space $V$, connected to $1$
and spanned by the monomials marked with black vertices, such that $F_1
\cdot V=F_6$:
\begin{center}
\includegraphics{tree}
\end{center}
This is the construction of \cite{BorMichiel}, which shows that there exists a
uniform determinantal representation of the generic bivariate polynomial of
degree $d$ of size $\frac{1}{4} \, d^2 + O(d)$ as $d \to \infty$.
\hfill $\clubsuit$
}
\end{ex}
The bivariate case generalises as follows.
\begin{thm} \label{thm:Cons1}
For fixed $n$, there exists a determinantal representation $M=M_0+M_1$ of
the generic $n$-variate polynomial of degree $d$ of size
$\frac{1}{n \cdot n!} \, d^n + O(d^{n-1})$
such that, moreover, the singular matrix space
represented by $M_0$ is a compression space with a one-dimensional
witness. Moreover, under this latter additional condition on $M_0$,
the bound is sharp.
\end{thm}
\begin{proof}
Note that $\dim F_d=\binom{n+d}{n}= \frac{1}{n!} \, d^n + O(d^{n-1})$. Hence by
Proposition~\ref{prop:Cons1} it suffices to show the existence of a
subspace $V \subseteq F_d$ connected to $1$ and such that $F_1 \cdot
V=F_d$, where $\dim V=\frac{1}{n} \, \dim F_d + O(d^{n-1})$. We will, in fact, show
that $V$ can be chosen to be spanned by monomials.
First, recall that there exists a lattice $\Lambda$ in $\ZZ^{n-1}$
such that $\ZZ^{n-1}$ is the disjoint union of $\Lambda$ and its cosets
$e_i+\Lambda$ for $i=1,\ldots,n-1$, namely, the root lattice of type $A_n$
generated by the rows of the $(n-1) \times (n-1)$-Cartan matrix
\[
\left[
\begin{array}{rrrrr}
2&-1& & &\\
-1&2&-1& &\\
&\ddots&\ddots&\ddots&\\
& &-1&2&-1\\
& & &-1&2
\end{array}
\right],
\]
where the empty positions represent zeros
\cite[Planche 1]{Bourbaki}. In particular,
the index of $\Lambda$ in $\ZZ^{n-1}$ equals $n$. For example, if
$n=3$, here is the root lattice $\Lambda$ (in black) and its two
cosets (in gray and white):
\begin{center}
\includegraphics[scale=.5]{A2}
\end{center}
Now let $\Delta_d$ be the simplex in $\RR^n$ with vertices
$0,de_1,\ldots,de_n$, for
$i=1,\ldots,n$ let $S_i$ be the set of lattice points in $\Delta_d$
that have $i$th coordinate zero, and set $S_0:=(\ZZ \times \Lambda)
\cap \Delta_d$. Define
\[ S:=S_1 \cup S_2 \cup \ldots \cup S_n \cup S_0,
\]
a subset of the lattice points in $\Delta_d$. We claim that $S$ is connected
to 0. Indeed, for each $i=1,\ldots,n$ the set $S_i$ is
connected to $0$,
and from each point $\alpha$ in $S_0$ one can walk within $S_0$
to $S_1$ by subtracting $\alpha_1$ times an $e_1$.
Next, we claim that for each $\alpha \in \Delta_d \cap \ZZ^n$ there
exists a $\beta \in S$ with $\alpha-\beta \in \{0,e_2,\ldots,e_n\}$.
Indeed, there is a (unique) $\beta'$ with this property in $\ZZ
\times \Lambda$. If this $\beta'$ has nonnegative entries, then set
$\beta:=\beta' \in S_0$. Otherwise, $\alpha$ itself has a
zero entry, say on the $i$th position, and we set $\beta:=\alpha \in S_i$.
Furthermore, for $i=1,\ldots,n$ the set $S_i$ contains $O(d^{n-1})$
vertices, and $S_0$ contains
$\frac{1}{n} \cdot \frac{1}{n!} \, d^n + O(d^{n-1})$
vertices. This concludes the construction--note that in the construction of
Proposition~\ref{prop:Cons1} the matrix $M_0$ has a zero
row, so that it represents a compression space with a
one-dimensional witness.
For sharpness, assume that $M=M_0+M_1$ is a uniform determinantal
representation of size $N$ such that the singular matrix space represented
by $M_0$ is a compression space with a one-dimensional witness. After a
choice of basis of $K^n$, we may assume that the first row of $M_0$
is identically zero; write $M_0=[0|M_0']^T$ accordingly. Let $u$ be the
first row of $M_1$ and write $M_1=[u|M_1']^T$. Then we have
\[
p=\sum_{|\alpha| \le d} c_\alpha x^\alpha = \det[u|M_0'+M_1'].
\]
Let $D_1,\ldots,D_N$ denote the $(N-1) \times (N-1)$ subdeterminants
of $M_0'$. By Lemma~\ref{lm:Dets}, the space $V$ spanned by these satisfies $F_1
\cdot V \supseteq F_d$. This already gives a lower bound of $V$ equal to
$d^n/((n+1)n!) + O(d^{n-1})$. To improve the $n+1$ in the denominator
into an $n$, we observe that by Cramer's rule, the map
\[
F_1^N \to K[x_1,\ldots,x_n],\ (\ell_1,\ldots,\ell_N) \mapsto \sum_i
(-1)^i \ell_i D_i
\]
has every column of $M_0'$ in its kernel. These
columns are linearly independent over $K$ (indeed over
$K(x_1,\ldots,x_n)$; see Lemma~\ref{lm:SingSpace}). We conclude that
\begin{equation} \label{eq:Ineq}
N \cdot \dim F_1 - (N-1) \ge \dim F_d,
\end{equation}
so that
\[ N \ge ((\dim F_d)-1)/n = d^n/(n \cdot n!) + O(d^{n-1}), \]
as desired.
\end{proof}
In the next section we derive a second
general construction of uniform determinantal representations, which we
use to prove Theorem~\ref{thm:Main}.
\section{A second construction} \label{sec:Second}
For a while, we believed that the uniform determinantal
representations of Theorem~\ref{thm:Cons1} were optimal.
But then we realised that
if one relaxes the condition that $M_0$ represent a compression space
with one-dimensional witness to the condition that $M_0$ represent
{\em some} compression space, smaller-size representations are possible.
The basic example is the following.
\begin{ex} \label{ex:2d}
{\rm
Let $p=\sum_{i+j \le 4} c_{ij} x^i y^j$ be the generic
polynomial of degree $d=4$ in $n=2$ variables. It has the
following uniform determinantal representation:
\begin{equation}\label{eq:repjan}
p=\det
\begin{bmatrix}
-x & \ph-1 & & & & & & & \\
& -x & \ph-1 & & & & & & \\
& & -x & \ph-1 & & & & & \\
& & & -x & 1 & & & & \\
c_{00}&c_{10}&c_{20}&c_{30}&c_{40}& -y & & & \\
c_{01}&c_{11}&c_{21}&c_{31}& & \ph-1 & -y & & \\
c_{02}&c_{12}&c_{22}& & & & \ph-1 & -y & \\
c_{03}&c_{13}& & & & & & \ph-1 & -y \\
c_{04}& & & & & & & & \ph-1
\end{bmatrix},
\end{equation}
where the empty positions denote zeros. Let $M=M_0+M_1$ be the matrix
on the right-hand side. In this case, $M_0$ represents a
compression space with witness $U=\langle e_5,\ldots,e_9 \rangle_K$,
which is mapped into $\langle e_6,\ldots,e_9 \rangle_K$.
To verify the identity above without too many calculations, note that the
5 maximal subdeterminants of the $4 \times 5$-block with $x$'s are,
consecutively, $1,-x,x^2,-x^3,x^4$, and similarly for $y$. The matrix
obtained from $M$ by deleting the column corresponding to $x^i$ and the
row corresponding to $y^j$ has determinant $x^iy^j$.
This example extends to a uniform determinantal representation of size
$2d+1$ for the generic bivariate polynomial $p$ of degree $d$. We get
$p=\det(M)$, where
\[
M=(-1)^d\begin{bmatrix}M_x & 0\cr L & M_y^T\end{bmatrix},
\]
$M_x$ and $M_y$ are $d\times (d+1)$ matrices with 1 on the first
upper diagonal and $-x$ and $-y$ respectively on the main diagonal,
while $L$ is a $(d+1)\times (d+1)$ triangular matrix such that
$\ell_{ij}=p_{j-1,i-1}$ for $i+j\le d+2$ and 0 otherwise.
Note that we will slightly improve on the size $2d+1$ in Example~\ref{ex:2d1}.
\hfill $\clubsuit$
}
\end{ex}
Example~\ref{ex:2d} generalises as follows.
\begin{prop} \label{prop:Cons2}
Let $V,W \subseteq K[x_1,\ldots,x_n]$ be subspaces connected to $1$ such
that $F_1 \cdot V \cdot W \supseteq F_d$. Then there exists a uniform
determinantal representation of the generic $n$-variate polynomial of
degree $d$ of size $-1 + \dim V + \dim W.$
\end{prop}
\begin{proof}
Set $m_1:=\dim V$ and $m_2:=\dim W$. Consider the matrix
\[
M:=
\begin{bmatrix}
M_V & 0 \\ L & M_W^T
\end{bmatrix},
\]
with $M_V$ and $M_W$ the matrices of sizes $(m_1-1) \times m_1$
and $(m_2-1)\times m_2$ from Lemma~\ref{lm:Block}, and where
$L=(\ell_{ij})_{ij}$ is an $m_2 \times m_1$-matrix to be determined. Note
that the determinant of $M$ is linear in the entries of $L$. Indeed,
setting $L=0$ yields the singular matrix $M_0$, so $\det(M)$ contains
no terms of degree 0 in the entries of $L$.
Furthermore, deleting from $M$ two or more
of the first $m_1$ columns from $M_V$, we end up with a matrix that
is singular since, when acting on rows, it maps the span of $\langle
e_1,\ldots,e_{m_1-1} \rangle$ into a space of dimension at most $m_1-2$,
so $\det(M)$ does not contain terms that are of degree $> 1$
in the entries of $L$.
Hence the determinant equals $\sum_{ij} \pm \ell_{ij} D_j E_i$ where
the $D_j$ are the maximal subdeterminants of $M_V$ and the $E_i$ are
the maximal subdeterminants of $M_W$. By Lemma~\ref{lm:Block} we have
$V=\langle D_1,\ldots,D_{m_1} \rangle_K$ and $W=\langle E_1,\ldots,E_{m_2}
\rangle_K$. Hence the assumption that $F_1 \cdot V \cdot W \supseteq F_d$
ensures that we can choose the $\ell_{ij} \in F_1$ in such a manner that
the determinant of $M$ equals the generic polynomial $p$.
\end{proof}
\begin{ex} \label{ex:2d1}
{\rm
Example~\ref{ex:2d} can be slightly improved to a representation of size
$2d-1$ by taking $V=\langle 1,x,\ldots,x^{d-1} \rangle$ and $W=\langle
1,y,\ldots,y^{d-1} \rangle$; note that, indeed, $F_1 \cdot V \cdot W
\supseteq F_d$.
A representation of size $2d-1$ for the polynomial $p$ from \eqref{eq:repjan} is
\begin{equation}\label{eq:repjan2}p=\det
\begin{bmatrix}
-x & \ph-1 & & & & & \\
& -x & \ph-1 & & & & \\
& & -x & -1 & & & \\
c_{00}&c_{10}&c_{20}&c_{30}+c_{40} x& -y & \\
c_{01}&c_{11}&c_{21}+c_{31}x & & \ph-1 & -y & \\
c_{02}+ c_{03} y&c_{12}+c_{22}x& & & & \ph-1 & -y \\
c_{13} x+c_{04} y& & & & & & \ph-1 \\
\end{bmatrix}.
\end{equation}
We do not know whether the factor 2 can be improved.
\hfill $\clubsuit$
}
\end{ex}
\begin{re}
{\rm
A representation of the form \eqref{eq:repjan2} can also be obtained
from the linearisations based on dual basis from \cite{Robol}. There,
linearisations of a univariate polynomial are presented that use the basis
of the form $\phi_i(x)\psi_j(x)$, where $\phi_i$ and $\psi_j$ are
polynomials. If we use the same approach for a bivariate polynomial
with the standard basis $\phi_i=x^i$ and $\psi_j=y^j$, we get a representation of the form
\eqref{eq:repjan2} up to permutations of rows and columns.
}
\end{re}
We will now prove our main theorem.
\begin{proof}[Proof of Theorem~\ref{thm:Main}, lower bound.]
For the lower bound on $N^*(n,d)$, let $M=M_0+M_1$ be a size-$N$ uniform
determinantal representation of the generic $n$-variate polynomial
of degree $d$. Let $D_{ij}$ be the $(N-1) \times (N-1)$-determinant
of the submatrix of $M_0$ obtained by deleting the entry at position
$(i,j)$. Then the image of the linear map
\[ \phi: F_1^{N \times N} \to K[x_1,\ldots,x_n],\
(\ell_{ij})_{i,j} \mapsto \sum_{i,j} (-1)^{i+j} \ell_{ij}
D_{ij} \]
contains $F_d$ (see Lemma~\ref{lm:Dets}).
We claim that $\phi$ has a kernel of dimension at
least $N(N-1)$. Indeed, fix any row index $i_0$. If the $D_{i_0,j}$ are
all zero, then we obtain an $(n+1)N$-dimensional subspace of $\ker \phi$
by setting all $\ell_{i,j}$ with $i \neq i_0$ equal to zero and choosing
the $\ell_{i_0,j} \in F_1$ arbitrarily. If they are not all zero, then
the $N-1$ rows with indices $i_1 \neq i_0$ are linearly independent over
$K(x_1,\ldots,x_n)$ and hence {\em a fortiori} over $K$. For
each such $i_1$ define $\ell^{(i_1)} \in F_1^{N \times N}$
by
\[ \ell^{(i_1)}_{ij}:=
\begin{cases}
(M_0)_{i_1,j} & \text{ if } i=i_0, \text{ and }\\
0 &\text{otherwise.}
\end{cases}
\]
Then, by Cramer's rule, we have $\phi(\ell^{(i_1)})=0$, and these $N-1$
vectors are linearly independent. Hence, for each $i_0$ we find a subspace
of $\ker \phi$ of dimension at least $N-1$, and these subspaces are
linearly independent. Thus we find that
\[ N^2(n+1)-N(N-1) = N^2 n + N \ge \dim F_d =
\frac{d^n}{n!}+O(d^{n-1}), \]
from which the existence of $C_1$ follows.
\end{proof}
\begin{re}
{\rm
In the proof of the lower bound we have been a bit more careful than
strictly needed: without the discussion of the kernel it follows
that $N^2 \ge (\dim F_d)/(n+1)$. But one derives a better
constant (for $d\to \infty$) by using the kernel.
}
\end{re}
For the upper bound, we first give a simple construction for even $n$.
For odd $n$, a trickier analysis is needed (which also applies in
the even case); see below.
\begin{proof}[Proof of Theorem~\ref{thm:Main}, upper bound for even $n$]
Assume that $n=2m$ with $m \in \ZZ_{\ge 0}$. Let $V$ be the space of
polynomials in $x_1,\ldots,x_m$ of degree at most $d$, and let $W$ be
the space of polynomials in $x_{m+1},\ldots,x_n$ of degree at most $d$.
Then $V$ and $W$ are connected to $1$ and we have $F_1 \cdot V \cdot W \supseteq
F_d$, so that by Proposition~\ref{prop:Cons2} we have $N^*(n,d) \le -1 +
\dim V+\dim W$. Now compute
\[ \dim V=\dim W=\binom{m+d}{m}=\frac{d^{n/2}}{(n/2)!}+O(d^{m-1}). \]
This implies the existence of $C_2$ for even $n$.
\end{proof}
We now give a construction that works for all $n>2$, for which we thank
Aart Blokhuis.
\begin{proof}[Proof of Theorem~\ref{thm:Main}, upper bound for $n>2$]
For $i=0,1$ let $B_i \subseteq \ZZ_{\ge 0}$ denote the set of nonnegative
integers that can be expressed as $\sum_{j=0}^e b_j 2^{2j+i}$ with
$b_j \in \{0,1\}$, i.e., whose binary expansions have ones only at
even positions (for $i=0$, counting the least significant bit as
zeroeth position) or only at odd positions (for $i=1$). Observe that
$B_0+B_1=\ZZ_{\ge 0}$ and that both $B_0$ and $B_1$ contain roughly
$\sqrt{d}$ of the first $d$ nonnegative integers for every $d$--they
have ``dimension $1/2$''. Now set $A_i:=B_i^n \subseteq \ZZ_{\ge 0}^n$
for $i=0,1$, so that $A_0+A_1=\ZZ_{\ge 0}^n$ and the number of elements
of $A_i$ intersected with a large box $[0,d]^n$ is roughly $d^{n/2}$.
Unfortunately, $A_0$ and $A_1$ are not connected to $0$.
However, we can connect them to 0 as follows. For a lattice point $\alpha \in A_i
\setminus \{0\}$ let $l$ be the minimum among the 2-adic valuations of
its entries, attained, say, by $\alpha_j$. Then
set $\tilde{\alpha}_j:=\alpha_j-2^l \in B_i$. Setting the remaining
coordinates of $\tilde{\alpha}$ equal to those of $\alpha$ we have
$\tilde{\alpha} \in A_i$ and
\[ \|\alpha-\tilde{\alpha}\|_1=2^l. \]
We propose to add to $A_i$ the sequence
\[ \alpha-e_j,\alpha-2e_j,\ldots,\alpha-(2^l-1)e_j \]
to connect $\alpha$ to $\tilde{\alpha}$. We need to verify, however, that
in doing this, $A_i$ retains dimension $n/2$. The fraction of $\alpha$
in (a large box intersected with) $A_i$ for which the minimal valuation
is at least $l$ equals roughly $(2^{(-l+i)/2})^n$--after all, the
condition is that for each $j=1,\ldots,n$, $\alpha_j$ has zeros on
the first $(l-i)/2$ positions where it is allowed to have ones. Write
$l=i+2m$. Thus by adding the sequences above, the total increase of $A_i$
is by a factor of at most
\[ \sum_{m=0}^\infty 2^{2m+i} (2^{-m})^n = \sum_{m=0}^\infty 2^{(2-n)m+i}. \]
This is a convergent series as $n>2$, and hence $A_i$ retains
dimension $n/2$.
We have thus constructed subsets $A_0,A_1 \subseteq \ZZ_{\ge 0}^n$ that
satisfy $A_0+A_1=\ZZ_{\ge 0}^n$, are connected to $1$, and that contain
roughly a constant times $d^{n/2}$ points in each box $[0,d]^n$. For
$i=0,1$ let $V_i$ be the space spanned by the monomials whose exponent
vectors lie in $A_i \cap [0,d]^n$. Then $V_1 \cdot V_2 \supseteq F_d$
and $V_1,V_2$ are connected to $1$, so by Proposition~\ref{prop:Cons2}
there exists a uniform determinantal representation for $p_{n,d}$ of
size $\dim V_1+\dim V_2=O(d^{n/2})$, as desired.
\end{proof}
\begin{re}
{\rm
The construction in the proof is by no means tight. For example, one could
also replace $\alpha_j$ by $\alpha_j-2^l+2^{l-2}+2^{l-4}+\ldots+2^i$,
which yields a shorter sequence to be added; and for large $l$ we have
also counted additional, shorter sequences since they also have valuation
larger than numbers smaller than $l$. We think that for even $n$ the
previous construction, subdividing the variables into two sets of equal
size, may lead to a better constant, but we have not verified this.
}
\end{re}
\begin{ex}
{\rm
Carrying out the construction in the proof for $A_0$ with $n=3$, always choosing
for $j$ the smallest index of a coordinate of $\alpha$ with minimal
valuation, we arrive at the following fractal-like structure (the circles indicate
the points of $A_0$, the black edges show that $A_0$ is
connected to $0$):
\begin{center}
\includegraphics[width=.4\textwidth]{fractal}
\end{center}
}
\end{ex}
\section{Small $n$ and $d$} \label{sec:Small}
In this section we give several uniform representations of--to our knowledge--the smallest possible size
for cases where $n$ and $d$ are small. We start with the two cases
where we can compute $N^*(n,d)$ exactly.
\begin{prop}
$N^*(2,2)=3$.
\end{prop}
\begin{proof}
Taking $V=\langle 1,x,y \rangle$ in Proposition~\ref{prop:Cons1}
we see that $N^*(2,2) \le 3$; this is the representation
of Example~\ref{ex:BinaryQuadric}. Suppose that a uniform
determinantal representation $M=M_0+M_1$ of size $N=2$ exists. Then,
by Example~\ref{ex:SmallSingSpace}, after acting with $\SL_2(K) \times
\SL_2(K)$ and transposing if necessary, we may assume that the singular
space represented by $M_0$ is a compression space with a one-dimensional
witness. But then \eqref{eq:Ineq} reads
\[ 2 \cdot 3 - 1 = N \cdot \dim F_1-(N-1) \ge \dim F_2 = 6, \]
a contradiction. Hence $N^*(2,2)=3$.
\end{proof}
\begin{prop}
$N^*(3,2)=4$.
\end{prop}
\begin{proof}
Taking $V=\langle 1,x,y,z \rangle$ in Proposition~\ref{prop:Cons1}
we see that $N^*(3,2)\le 4$. Suppose that a uniform representation of
size $N=3$ exists. Up to transposition, there are three possibilities
for the singular space $\mathcal{A}$ represented by $M_0$; see
Example~\ref{ex:SmallSingSpace} (where the third
compression space is conjugate to the transpose of the
first):
\begin{enumerate}
\item
Assume that $\mathcal{A}$ is a compression space with a
one-dimensional witness, so that after acting with $\SL_3(K) \times
\SL_3(K)$ we have
\[ M_0=\begin{bmatrix}
0 & 0 & 0 \\
* & * & *\\
* & * & *
\end{bmatrix}.
\]
Let $D_j$ denote the determinant of the minor of $M_0$ obtained by
deleting the first row and $j$th column. Then the linear map
\[ \Omega:F_1^3 \mapsto K[x,y,z],\ (l_1,l_2,l_3) \mapsto
l_1D_1-l_2D_2+l_3D_3 \]
has $F_2 \subseteq \im \Omega$. Now inequality~\eqref{eq:Ineq} reads
\[ 3 \cdot 4 - 2 = N \cdot \dim F_1-(N-1) \ge \dim F_2 = 10, \]
which holds with equality. This means that, in fact, $\im \Omega$
must {\em equal} $F_2$. In particular, $D_1,D_2,D_3$ must all be of
degree one (or else $\im \Omega$ would contain cubic polynomials). The
image of $\Omega$ depends only on the span $V:=\langle D_1,D_2,D_3
\rangle \subseteq F_1$. If $1 \not \in V$, then there exists an
affine transformation in $\AGL_3(K)$ that maps $V$ into a subspace of
$\langle x,y,z \rangle$. Then $1 \not \in F_1 \cdot V=\im \Omega$,
a contradiction. If $1 \in V$, then after an affine transformation we
find $\langle 1 \rangle \subseteq V \subseteq \langle 1,x,y \rangle$. In
that case, $z^2 \not \in F_1 \cdot V$, another contradiction.
\item Assume that $\mathcal{A}$ is a compression space with a
two-dimensional witness, so that after row and column operations we
have
\[
M_0=\begin{bmatrix}
0&0&q \\
0&0&r \\
s&t&*
\end{bmatrix},
\]
where $q,r,s,t \in F_1$. Write $M_1=(m_{ij})_{ij}$. Using that $\det(M)$
is assumed to be linear in the $c_\alpha$s, we find that
\[ \det(M)=-m_{11} rt + m_{12} rs + m_{21} qt - m_{22} qs. \]
Consequently, setting $V_1:=\langle q,r \rangle$ and $V_2:=\langle s,t
\rangle$, we have $F_1 \cdot V_1 \cdot V_2 \supseteq F_2$. If $1 \not \in
V_1$, then by acting with a suitable element of $\AGL_3(K)$ we achieve
that $V_1 \subseteq \langle x,y,z \rangle$. But then $F_1 \cdot V_1
\cdot V_2 \not \ni 1$. The same applies when $1 \not \in V_2$. On the
other hand, if $1 \in V_1 \cap V_2$, then by an element in $\AGL_3(K)$
we achieve that $\langle 1 \rangle \subseteq V_1,V_2 \subseteq \langle
1,x,y\rangle$. In that case, $z^2 \not \in F_1 \cdot V_1 \cdot V_2$.
\item Finally, assume that $\mathcal{A}$ is conjugate to a space of
skew-symmetric matrices, so that after conjugation
\[ M_0=
\left[
\begin{array}{rrr}
0 & q & r \\
-q& 0 & s \\
-r&-s & 0
\end{array}
\right]
\]
where $q,r,s \in F_1$. Set $V:=\langle q,r,s \rangle \subseteq F_1$. Then
the space spanned by the $2 \times 2$-determinants of $M_0$ is $V \cdot
V$, of dimension at most $6$. Moreover, we have $F_1 \cdot V \cdot V
\supseteq F_2$. If $1 \not \in V$, then by acting with $\AGL_3(K)$ we
achieve that $V \subseteq \langle x,y,z \rangle$, and hence $1 \not \in
F_1 \cdot V \cdot V$. If, on the other hand, $1 \in V$, then we achieve
that $\langle 1 \rangle \subseteq V \subseteq \langle 1,x,y \rangle$,
and $z^2 \not \in F_1 \cdot V \cdot V$.
\end{enumerate}
In each of these cases we arrive at a contradiction. Consequently,
$N^*(3,2)=4$ as claimed.
\end{proof}
The proofs above use the classification of spaces of small singular
matrices in an essential manner, as well as the action of $\AGL_n(K)$
on uniform determinantal representations. We conjecture that
$N^*(4,2)=5$, and that this can still be proved in the same manner,
using the classification of $4 \times 4$-singular matrix spaces from
\cite{FillmoreLaurieRadjavi}. But as Theorem~\ref{thm:Sing5} shows,
fundamentally new ideas will be needed to prove lower bounds in larger
situations.
For some pairs of small $n$ and $d$ we now give the smallest
uniform representations that we have been able to find.
For the constructions we use Proposition~\ref{prop:Cons2} with
subspaces $V,W\subseteq K[x_1,\ldots,x_n]$ spanned by the monomials and
connected to $1$. First, we give
in Table~\ref{tab:minnd} the minimal sizes known to us of
uniform determinantal representations for some small values
of $n$ and $d$.
\begin{table}[!htbp]
\caption{Minimal known sizes of uniform determinantal representations
we have been able to construct for $n$-variate polynomials of degree $d$;
cf.~Table~\ref{tab:wv}.\label{tab:minnd}}
\begin{center}
{\footnotesize
\begin{tabular}{c|ccccccccc} \hline \rule{0pt}{2.3ex
$n$ & $d=2$ & $d=3$ & $d=4$ & $d=5$ & $d=6$ & $d=7$ & $d=8$ & $d=9$ \\ \hline \rule{0pt}{2.3ex
2 & 3 & \ph{1}5 & \ph{1}7 & \ph{1}9 & 11& 13& 15& 17 \\
3 & 4 & \ph{1}7 & 10 & 14& 18& 22& 27& 34 \\
4 & 5 & \ph{1}9 &14 & 19& 26& 34& 44 \\
5 & 6 &11 &18 & 26& \\
6 & 7 &13 &22 & 33& \\
7 & 8 &15 &27 & 39& \\
8 & 9 &17 &32 & \\ \hline
\end{tabular}
}
\end{center}
\end{table}
The corresponding representations for the entries
in Table~\ref{tab:minnd} for $n=2$, which are of size $2d-1$, are given in
Example~\ref{ex:2d1}.
For $d=2$ we take $V= \langle 1,x_1,\ldots,x_n \rangle$ and
$W= \langle 1 \rangle$,
therefore $N^*(n,2)\le n+1$, while for $d=3$ we can
take $V=W= \langle 1,x_1,\ldots,x_n \rangle$, and hence $N^*(n,3)\le
2n-1$.
In Table~\ref{tab:wv}
we give sets $V$ and $W$ for the remaining nonzero entries in Table~\ref{tab:minnd}.
The subspaces $V$ and $W$ have the form $V=V_0\cup V_1$ and
$W=W_0\cup W_1$, where
\begin{equation}\label{eq:vw0}
\begin{matrix}
V_0=\langle1,x_1,\ldots,x_n,\ldots,x_1^e,\ldots,x_n^e\rangle,\\[0.4em]
W_0=\langle1,x_1,\ldots,x_n,\ldots,x_1^f,\ldots,x_n^f\rangle
\end{matrix}
\end{equation}
for $e=\lceil (d-1)/2\rceil$ and $f=\lfloor (d-1)/2\rfloor$, which yields $d-1=e+f$.
For clarity and brevity, the variables $x,y,z,w,u,v,q,s$ in Table~\ref{tab:wv}
stand for $x_1,\ldots,x_8$, respectively.
\begin{table}[!htbp]
\begin{center}
\caption{List of monomials in $V_1$ and $W_1$ that,
together with $V_0$ and $W_0$ from \eqref{eq:vw0},
lead to uniform representations of sizes as in Table~\ref{tab:minnd}.\label{tab:wv}}
{\footnotesize
\begin{tabular}{ll|ll} \hline \rule{0pt}{2.3ex
$n$ & $d$ & $V_1$ & $W_1$\\
\hline \rule{0pt}{2.3ex
3 & 4 & $-$ & $-$\\
3 & 5 & $-$ & $xy$\\
3 & 6 & $-$ & $xy,\, x^2y$\\
3 & 7 & $-$ & $x^2y,\, y^2z,\, z^2 x$\\
3 & 8 & $-$ & $x^2y,\, y^2z,\, z^2x,\, x^2y^2,\, z^2w^2$ \\
3 & 9 & $x^3y,\,y^3z,\,z^3x$ & $x^2y,\,x^2z,\,y^2z,\,x^2y^2,\,x^2z^2,\,y^2z^2$\\[2mm]
4 & 4 & $-$ & $xy$\\
4 & 5 & $-$ & $xy,\, zw$\\
4 & 6 & $x^2y,\, y^2z,\, z^2w$ &
$xy,\, x^2y,\, z w$\\
4 & 7 & $x^2y,\, y^2x,\, z^2w,\, w^2x,\, xy$ &
$x^2z,\, xz^2,\,y^2w,\, yw^2$\\
4 & 8 & $x^2y,\, x^2y^2,\, z^2x,\, x^3 y,\, y^3z,\,
z^3w,\,w^3x$ &
$ xy,\,xyz,\, xyw,\, y^2z,\, z^2w,\,
w^2x,\, w^2y,\, x^2z$\\[2mm]
5 & 4 & $-$ & $ xy,\, zw$\\
5 & 5 & $xy,\, yz,\, zw$ & $wu,\, xu$\\[2mm]
6 & 4 & $-$ & $xy,\, zw,\, uv$\\
6 & 5 & $xy,\, zw,\,uv,\,wy$ & $yz,\, wu,\, xv,\, xz$\\[2mm]
7 & 4 & $-$ & $xy,\, zw,\, uv,\, xq,\, yq$\\
7 & 5 & $xy,\, z w,\, u v,\, w y,\, q u$ &
$y z,\, w u,\, v q,\, x z,\, w x$\\[2mm]
8 & 4 & $-$ & $x y,\, y z,\, x z,\, w u,\, w v,\, u v,\, q s$ \\ \hline
\end{tabular}}
\end{center}
\end{table}
\begin{ex} \label{ex:turan}
{\rm
To show how things get complicated, let us
consider the construction for $d=4$. We take $V=
\langle 1,x_1,\ldots,x_n,x_1^2,\ldots,x_n^2 \rangle$ and
\[W= \langle
1,x_1,\ldots,x_n,x_{\alpha_1}x_{\beta_1},\ldots,x_{\alpha_m}x_{\beta_m}
\rangle,\]
where $1\le \alpha_i<\beta_i\le n$ and $m$ is as small as possible. If we take all possible pairs
$x_{\alpha}x_\beta$, then clearly $F_1 \cdot V \cdot W \supseteq F_4$,
while on the other hand, when $m=0$, $F_1 \cdot V \cdot W$ does not contain
any monomials of the form
\begin{equation}
\label{eq:ijkl}
x_i \, x_j \, x_k \, x_\ell
\end{equation}
for $1\le i<j<k<\ell\le n$.
We need a minimal set of $x_\alpha x_\beta$ to cover all possible
monomials \eqref{eq:ijkl}, which is related to the following covering problem.
Given positive integers $r\le k\le n$, we say that a system $S$ of $r$-subsets
(called blocks) of $\{1,\dots, n\}$ is called
a Tur\'an $(n,k,r)$-system if every $k$-subset of $\{1,\dots, n\}$ contains at
least one block from $S$ \cite{Sidorenko}. The minimum size of $S$
is called the Tur\'an number $T(n,k,r)$.
In our case, additional terms $x_{\alpha_1}x_{\beta_1},\ldots,x_{\alpha_m}x_{\beta_m}$ form
a Tur\'an $(n,4,2)$-system. While for most cases
only upper and lower bounds for $T(n,k,r)$ are known, Tur\'an proved
that
\begin{equation}\label{eq:Turan}
T(n,4,2)=mn-3 \, {m(m+1)\over 2},
\end{equation}
where $m=\lfloor n/3\rfloor$. To obtain the minimal set one has to divide $\{1,\ldots,n\}$
into three nearly equal groups (their sizes do not differ for more than one)
and then take all pairs $x_\alpha x_\beta$
such that $\alpha$ and $\beta$ belong to the same group.
As a result, such construction gives a uniform representation
of size $N$, where $N=\frac{1}{6} \, n^2+O(n)$,
which therefore implies $N^*(n,4) \le \frac{1}{6} \, n^2+O(n)$.
\hfill $\clubsuit$
}
\end{ex}
\section{Matrix polynomials} \label{sec:MatPol}
Suppose that we have a uniform representation $M$ of
$p_{n,d}$ as in \eqref{eq:polnd}, and write
\begin{equation}\label{eq:linM}
M=M_0+M_1=M_0 + \sum_{|\alpha| \le d} c_\alpha M_\alpha,
\end{equation}
where each $M_\alpha \in F_1^{N \times N}$.
Now consider the matrix polynomial (cf.~\eqref{eq:polnd})
\[ P_{n,d}=\sum_{|\alpha| \le d} x^\alpha C_\alpha , \]
where $C_\alpha$ is a $k\times k$ matrix. We will show that under
certain assumptions we can construct from $M$ a matrix
$\widetilde{M}$ that represents $P_{n,d}$ in the sense
that
$\det(\widetilde M)=\det(P_{n,d})$.
We obtain $\widetilde M$ from $M$ in the following way. Each element of the form
$\alpha +\beta x +\gamma y$ is replaced by the $k\times k$ matrix
$(\alpha+\beta x + \gamma y)I_k$, where $I_k$ is the $k \times k$ identity,
and each $c_\alpha$ is replaced by the matrix $C_\alpha$.
\begin{thm}\label{thm:matpollin} Let \eqref{eq:linM} be a uniform representation
of the generic polynomial \eqref{eq:polnd} of degree $d$ in $n$ variables
and assume that there exist matrices $Q$ and $Z$, whose elements are polynomials in
$x_1,\ldots,x_n$, such that $\det(Q) = \det(Z) = 1$, and
$QMZ$ is a triangular matrix with one diagonal element equal to $p_{n,d}$
and all other diagonal elements equal to 1.
Then
\[\widetilde M = M_0\otimes I_k+
\sum_{|\alpha|\le d}M_\alpha\otimes C_\alpha\]
is a representation for the matrix polynomial $P_{n,d}$, i.e.,
$\det(\widetilde M)=\det(P_{n,d})$.
\end{thm}
\begin{proof}
It is easy to see that $(Q\otimes I_k)\widetilde M(Z\otimes I_k)$
is a block triangular matrix with one diagonal block $P_{n,d}$ while
all other diagonal blocks are equal to $I_k$. Since $\det(Q\otimes I_k)=\det(Z\otimes I_k)=1$,
it follows that $\det(\widetilde M)=\det(P_{n,d})$.
\end{proof}
\begin{ex}
{\rm
Theorem~\ref{thm:matpollin} applies to the uniform representation~\eqref{eq:repjan}.
Indeed, take
\[
Q=\begin{bmatrix}
1 & & & & & & &\cr
& \ddots & & & & & & \cr
& & 1 & & & & & \cr
& & & 1 & y& y^2 &y^3 &y^4 \cr
& & & & 1 & y & y^2& y^3\cr
& & & & & 1 & y& y^2\cr
& & & & & & 1 & y\cr
& & & & & & & 1\end{bmatrix},
\qquad
Z=\begin{bmatrix} 1 & & & & & & & \cr
x & 1 & & & & & & \cr
x^2 & x & 1 & & & & & \cr
x^3 & x^2 & x& 1 & & & & \cr
x^4 & x^3 & x^2 & x & 1 & & & \cr
& & & & & 1 & & \cr
& & & & & & \ddots & \cr
& & & & & & & 1\end{bmatrix},
\]
then
\[
QMZ=
\begin{bmatrix}
& 1 & & & & & & & \\
& & 1 & & & & & & \\
& & & 1 & & & & & \\
& & & & 1 & & & & \\
p & \times & \times & \times &c_{40}& & & & \\
\times &\times &\times &c_{31}& & 1 & & & \\
\times &\times &c_{22}& & & & 1 & & \\
\times &c_{13}& & & & & & 1 & \\
c_{04}& & & & & & & & 1
\end{bmatrix}.
\]
It is easy to see that there exist permutation matrices $P_L$ and $P_R$ such that
\[
P_L(QMZ)P_R=
\begin{bmatrix}
1 & & & & & & & & \\
& 1 & & & & & & & \\
& & 1 & & & & & & \\
& & & 1 & & & & & \\
c_{40} & \times & \times & \times &p& & & & \\
&\times &\times &c_{31}& \times & 1 & & & \\
&\times &c_{22}& & \times & & 1 & & \\
&c_{13}& & & \times & & & 1 & \\
& & & & c_{04} & & & & 1
\end{bmatrix}
\]
is triangular and has the diagonal which satisfies
Theorem \ref{thm:matpollin}. Therefore, we can apply
\eqref{eq:repjan} for matrix polynomials by using block
matrices. This can be generalised to a uniform
representation of size $2d+1$ of the form
\eqref{eq:repjan}. In a similar way we can show that
this also holds for representations of the form
\eqref{eq:repjan2} of size $2d-1$.
\hfill $\clubsuit$
}
\end{ex}
Unfortunately, not all uniform determinantal representations induce a determinantal
representation of a general matrix polynomial in this manner. As a counterexample,
let $M$ be such a uniform determinantal representation of the polynomial
$p_{n,d}$, $|\alpha|,|\beta| \le d$, and construct a representation of larger size
\[
M'=\begin{bmatrix}
M & 0\\
0 & N\\
\end{bmatrix},
\quad\quad \hbox{with}\quad N=
\left[
\begin{array}{rrrr}
0 & c_{\alpha} & c_{\beta} & 1\\
-c_{\alpha} & 0 & 1 & 0\\
-c_{\beta} & -1 & 0 & 0\\
-1 & 0 & 0 & 1
\end{array}
\right].
\]
Then $\det(M')=\det(M)\det(N)=
p_{n,d}(1+c_{\alpha}c_{\beta}-c_{\beta}c_{\alpha})=p_{n,d}$, but
$\widetilde{M'}$ is not a representation for the matrix polynomial
$P_{n,d}$ as the coefficient matrices $C_{\alpha}$ and $C_{\beta}$
do not commute in general.
This motivates the following definition.
\begin{de}
A uniform determinantal representation $M$ is \emph{minimal} if there do not exist
constant matrices $P$ and $Z$ such that $\det(P)=\det(Z)=1$ and
\[
QMZ=\begin{bmatrix}
* & *\\
0 & M_2\\
\end{bmatrix},
\quad \quad \hbox{where $M_2$ is square with} \quad \det(M_2) =1.
\]
\end{de}
We speculate that each minimal uniform representation gives rise to a
representation for a matrix polynomial.
\section{Numerical experiments} \label{sec:Numerics}
Recently, a new numerical approach for computing roots of systems of bivariate
polynomials was proposed in \cite{BorMichiel}. The main idea is
to treat the system as a two-parameter
eigenvalue problem using determinantal representations.
Suppose that we are looking for roots of a system of bivariate polynomials
\begin{equation}\label{eq:syspol}
\begin{matrix}p=\sum_{i+j \le d_1} \alpha_{ij} x^i y^j=0,\\[0.2em]
q=\sum_{i+j \le d_2} \beta_{ij} x^i y^j=0,\end{matrix}
\end{equation} where
$p$ and $q$
are polynomials
of degree $d_1$ and $d_2$
over $\CC$. Let
$P=A_0 + x A_1 + y A_2$ and $Q=B_0 + x B_1 + y B_2$,
where $A_0,A_1,A_2 \in \CC^{N_1 \times N_1}$ and $B_0,B_1,B_2 \in \CC^{N_2 \times N_2}$, with $\det(P)=p$ and
$\det(Q)=q$, be determinantal representations of $p$ and $q$, respectively.
Then a root $(x,y)$ of \eqref{eq:syspol} is an
eigenvalue of the two-parameter eigenvalue problem
\begin{equation}\label{eq:twopar}
\begin{matrix}
(A_0+x A_1+y A_2) \, u=0,\\[0.2em]
(B_0+x B_1+y B_2) \, v=0,
\end{matrix}
\end{equation}
where $u\in\CC^{N_1}$ and $v\in\CC^{N_2}$ are nonzero vectors.
The standard way to solve \eqref{eq:twopar} is to
consider a joint pair
of generalized eigenvalue problems \cite{Atkinson}
\begin{equation}\label{eq:twopardelta}
\begin{matrix}
(\Delta_1-x \Delta_0) \, w=0,\\[0.2em]
(\Delta_2-y \Delta_0) \, w=0,
\end{matrix}
\end{equation}
where
\[
\Delta_0=A_1\otimes B_2-A_2\otimes B_1,\quad
\Delta_1=A_2\otimes B_0-A_0\otimes B_2,\quad
\Delta_2=A_0\otimes B_1-A_1\otimes B_0,
\]
and $w=u\otimes v$.
In this particular application we can expect that the
pencils in \eqref{eq:twopardelta} are singular, i.e., $\det(\Delta_1-x\Delta_0)\equiv 0$ and
$\det(\Delta_1-y\Delta_0)\equiv 0$. Namely, by
B\'ezout's theorem a generic system \eqref{eq:syspol}
has $d_1d_2$ solutions, while a generic problem \eqref{eq:twopar} has $N_1N_2$ eigenvalues.
Unless $(d_1,d_2)=(N_1,N_2)$, both pencils in \eqref{eq:twopardelta} are singular. In this
case we first have to apply the staircase algorithm from \cite{MuhicPlestenjakLAA}
to extract the finite regular eigenvalues. The method returns smaller matrices
$\widetilde \Delta_0$,
$\widetilde \Delta_1$, and $\widetilde \Delta_2$ (of size $d_1d_2\times d_1d_2$ for a generic \eqref{eq:syspol})
such that $\widetilde \Delta_0$ is nonsingular and ${\widetilde \Delta_0}^{-1}\widetilde \Delta_1$ and
${\widetilde \Delta_0}^{-1}\widetilde \Delta_2$ commute. From
\[
\begin{matrix}
(\widetilde\Delta_1-x \widetilde\Delta_0) \, \widetilde w=0,\\[0.2em]
(\widetilde\Delta_2-y \widetilde\Delta_0) \, \widetilde w=0,
\end{matrix}
\]
we compute the eigenvalues $(x,y)$ using a variant of the QZ algorithm \cite{HKP}
and thus obtain the roots of \eqref{eq:syspol}.
The above approach is implemented in the Matlab package BiRoots \cite{BorBR} together with
the two determinantal representations
from \cite{BorMichiel}. The first one, to which we refer as {\tt Lin1}, is a
uniform one from Example \ref{ex:lin1} of size $\frac{1}{4} \, d^2+ O(d)$ for a polynomial of degree $d$.
The second one, which we refer to as {\tt Lin2}, is not uniform and involves
some computation to obtain a smaller size $\frac{1}{6} \, d^2+ O(d)$.
Although the construction of {\tt Lin2} is more time consuming,
this pays off later, when the staircase algorithm is applied to \eqref{eq:twopardelta}.
Table~\ref{tbl:compare1} shows the sizes of determinantal representations for polynomials of
small degree.
As expected, the new uniform determinantal representation of size
$2d-1$, to which we refer as {\tt MinUnif}, returns smaller matrices, which reflects later
in faster computational times. It is also
important that {\tt Lin1} and {\tt MinUnif} return real matrices for polynomials with real
coefficients, which is not true for {\tt Lin2}.
\begin{table}[!htbp]
\begin{footnotesize}
\begin{center}
\caption{Size of the matrices for {\tt Lin1} and {\tt Lin2}
for bivariate polynomials ($n=2$) and various degrees $d$.}\label{tbl:compare1}
\begin{tabular}{l|cccccccccc} \hline \rule{0pt}{2.3ex
Method & $d=3$ & $d=4$ & $d=5$ & $d=6$ & $d=7$ & $d=8$ & $d=9$ & $d=10$ & $d=11$ & $d=12$\\
\hline \rule{0pt}{2.3ex
{\tt Lin1} & 5 & 8 & 11 & 15 & 19 & 24 & 29 & 35 & 41 & 48 \\
{\tt Lin2} & 3 & 5 & \ph{1}8 & 10 & 13 & 17 & 20 & 24 & 29 & 34\\
{\tt MinUnif} & 5 & 7 & \ph{1}9 & 11 & 13 & 15 & 17 & 19 & 21 & 23\\
\hline
\end{tabular}
\end{center}
\end{footnotesize}
\end{table}
It was reported in
\cite{BorMichiel} that the determinantal representation approach for solving
systems of bivariate polynomials is competitive for polynomials of degree 9 or less.
As we show below, the new uniform representation {\tt MinUnif} extends
this to degree 15 and, in addition, performs
better than the existing representations for polynomials of degree 6 or more.
In \cite{BorMichiel} the approach was compared numerically to the following
state-of-the art numerical methods for polynomial systems:
{\tt NSolve} in Mathematica~9~\cite{Wolfram},
BertiniLab~1.4~\cite{Bertini} running Bertini~1.5~\cite{BertiniExe},
NAClab~3.0~\cite{NAClab},
and {\tt PHCLab}~1.04~\cite{PHClab} running PHCpack~2.3.84,
which turned out as the fastest of these methods.
To show the improved performance of the new determinantal representation, we compare
{\tt MinUnif} to {\tt Lin1}, {\tt Lin2}, and {\tt PHCLab}
in Table~\ref{tbl:two}. For each $d$ we run the methods on the same set of
50 real and 50 complex random polynomial systems of degree $d$
and measure the average time.
For {\tt Lin1} and {\tt MinUnif}, where determinantal representations have real
matrices for real polynomials, we report separate results
for polynomials with real and complex coefficients. The timings
for {\tt Lin1} and {\tt Lin2} are given only for $n\le 10$ as
for larger $n$ these two linearisations are no longer competitive.
\begin{table}[!htbp]
\begin{footnotesize}
\begin{center}
\caption{Average computational times in milliseconds
for {\tt Lin1}, {\tt Lin2}, {\tt MinUnif}, and {\tt PHCLab} for
random full bivariate polynomial systems of degree $3$ to $15$. For {\tt Lin1} and {\tt MinUnif}
separate results are included for real $(\RR)$ and complex polynomials $(\CC)$.
}\label{tbl:two}
\begin{tabular}{r|rrrrrr} \hline \rule{0pt}{2.3ex
$d$ & {\tt Lin1} ($\RR$) & {\tt Lin1} ($\CC$) & {\tt Lin2} & {\tt PHCLab} & {\tt MinUnif} ($\RR$) & {\tt MinUnif} ($\CC$) \\
\hline \rule{0pt}{2.3ex
3 & \ph{111}6 & \ph{111}8 & \ph{11}4 & \ph{1}116 & \ph{11}6 & \ph{11}7 \\
4 & \ph{111}9 & \ph{11}11 & \ph{11}6 & \ph{1}130 & \ph{1}12 & \ph{11}13 \\
5 & \ph{11}20 & \ph{11}26 & \ph{1}13 & \ph{1}151 & \ph{1}18 & \ph{11}20 \\
6 & \ph{11}39 & \ph{11}71 & \ph{1}28 & \ph{1}174 & \ph{1}27 & \ph{11}27 \\
7 & \ph{11}96 & \ph{1}160 & \ph{1}51 & \ph{1}217 & \ph{1}36 & \ph{11}44 \\
8 & \ph{1}205 & \ph{1}395 & 118 & \ph{1}264 & \ph{1}59 & \ph{11}74 \\
9 & \ph{1}467 & 1124 & {279} & \ph{1}329 & \ph{1}{95} & \ph{1}125 \\
10 & 1424 & 3412 & 600 & \ph{1}414 & {147} & \ph{1}221 \\
11 & & & & \ph{1}538 & {248} & \ph{1}354 \\
12 & & & & \ph{1}650 & {361} & \ph{1}530 \\
13 & & & & \ph{1}911 & {592} & \ph{1}740 \\
14 & & & & 1142 & {842} & 1148 \\
15 & & & & 1531 & {1237} & 1835 \\
\hline
\end{tabular}
\end{center}
\end{footnotesize}
\end{table}
Of course, the computational time is not the only important factor, we also have to
consider the accuracy and reliability. In each step of the staircase algorithm a
rank of a matrix has to be estimated numerically, which is a delicate task.
After several steps it may happen that the gap between the important singular values
and the meaningful ones that should be zero in exact computation, virtually disappears.
In such case the algorithm fails and does not return any roots.
As the number of steps in the staircase algorithm increases with degree of the polynomials,
such problems occur more often for polynomials of large degree.
A heuristic that usually helps in such
cases is to apply the algorithm on a transformed system
\begin{align*}
\widetilde p &:= \ph{-}c p + s q = 0,\\
\widetilde q &:= -s p + c q = 0
\end{align*}
for random $c$ and $s$ such that $c^2+s^2=1$.
As this transformation does not change the conditioning of the roots, we
can conclude that the difficulties with the staircase algorithm are not directly related to
the conditioning. The trick does not work every time, and it seems that for
some systems the only way to make the determinantal representation approach to work is
to increase the machine precision.
We can apply the same approach to systems of polynomials in more than two variables.
However, since the size of the corresponding $\Delta$ matrices
is the product of sizes of all representations, this is competitive only
for $n=3$ and $d\le 3$. For a comparison, if we have a system of three polynomials
in three variables of degree $3$, then the size of the $\Delta$ matrices
is $343\times 343$. For degree $4$ the size increases to $1000\times 1000$
and PHCpack is faster. Finally, for $n=4$ and the smallest nontrivial $d=2$
we already get $\Delta$ matrices of size $625\times 625$ and the method
is not efficient.
\section{Outlook} \label{sec:Outlook}
We have introduced {\em uniform determinantal representations}, which
rather than representing a single polynomial as the determinant of a
matrix of affine-linear forms, represent all polynomials of degree at
most $d$ in $n$ variables as such a determinant. We have seen that in
the bivariate case, these determinantal representations are useful for
numerically solving bivariate systems of equations; and in the general
multivariate case we have determined, up to constants, the asymptotic
behaviour of $N^*(n,d)$, the minimal size of such a representation,
for $n$ fixed and $d \to \infty$.
We now summarise several results that have been shown in the paper.
\begin{itemize}
\item For fixed $n$ and $d \to \infty$, $N^*(n,d) \sim d^{n/2}$, see
Theorem~\ref{thm:Main}.
This is a noticeable improvement on \cite{Quarez}, where an asymptotic rate
of $N^*(n,d) \sim d^n$ is shown, with the remark that the representation
in \cite{Quarez} are symmetric. However, symmetry currently cannot be exploited
by methods that compute roots of multivariate polynomial systems.
\item For fixed odd $d$ and $n \to \infty$, $N^*(n,d) \sim n^{(d-1)/2}$, which
is the same rate as Quarez \cite{Quarez}, who also manages to get symmetric representations.
For fixed even $d$ and $n \to \infty$, $N^*(n,d) \lsim n^{d/2}$, which
again is the same rate as in \cite{Quarez}. However, we have a slightly
smaller lower bound for the asymptotic rate of $n^{(d-1)/2}$.
\item Tables~\ref{tab:minnd} and \ref{tab:wv} give constructions
for the smallest representations that we have been able to find for some small values of $n$ and $d$.
\item $N^*(n,2) \le n+1$; cf.~Table~\ref{tab:minnd}.
\item $N^*(n,3) \le 2n+1$; cf.~Table~\ref{tab:minnd}.
\item $N^*(n,4) \le \frac{1}{6} n^2+O(n)$; see Example~\ref{ex:turan}.
\item $N^*(2,d) \le 2d-1$; cf.~Table~\ref{tab:minnd}.
Note that this result satisfies Dixon~\cite{Dixon}
up to an asymptotic factor 2 whereby no computations are necessary
for the determinantal representation.
In particular, it is a major improvement on the $\sim \frac{1}{4} d^2$
of \cite{Quarez, BorMichiel}.
\item Due to the smaller sizes of the representations, the numerical
approach for bivariate polynomials ($n=2$) is competitive to (say)
Mathematica for degree $d$ up to $d \approx 15$ (see Section~\ref{sec:Numerics});
this in contrast to $d \approx 9$ as obtained in \cite{BorMichiel}.
\item Under some conditions, the results carry over to the case
of matrix coefficients (see Section~\ref{sec:MatPol}).
\end{itemize}
There are still many interesting open questions, both of intrinsic mathematical
interest and of relevance to polynomial system solving.
First of all, in a situation where the degree $d$ is fixed and the number
$n$ of variables grows, what is the asymptotic behaviour of $N^*(n,d)$?
Although we expect that the above inequalities are equalities,
whether this holds is still open.
For general fixed $d$, the proof of Theorem~\ref{thm:Main} yields a lower
bound which is a constant (depending on $d$) times $n^{(d-1)/2}$. For odd
$d$ we obtain a matching upper bound (with a different constant) by using
Proposition~\ref{prop:Cons2} with $V=W=F_{(d-1)/2}$. However, for even $d$
we only know how to obtain $O(n^{d/2})$. We remark that this latter is
(up to a constant) the same bound as obtained in \cite[Thm.~4.4]{Quarez}
for {\em symmetric} uniform representations.
Second, in the case of fixed $n$ and varying $d$ studied in this paper,
what are the best constants in Theorem~\ref{thm:Main}? More specifically,
for fixed $n$, does $\lim_{d \to \infty} \frac{N^*(n,d)}{d^{n/2}}$ exist,
and if so, what is its value?
Third, how can our techniques for upper bounds and lower bounds be further
sharpened? Can singular matrix spaces {\em other} than compression spaces
be used to obtain tighter upper bounds (constructions) on $N^*(n,d)$?
Can the action of the affine group be used more systematically to find
lower bounds on $N^*(n,d)$?
Fourth, is it true that each minimal uniform representation gives
rise to a representation of the corresponding matrix polynomial
(cf.~Section~\ref{sec:MatPol})?
Finally, we have restricted our attention to matrices that, apart from
being affine-linear in $x_1,\dots,x_n$, are also affine-linear in the
coefficients $c_\alpha$. Our proofs give the same asymptotic behaviour
(with different constants) if we require, in addition, that no quadratic terms $c_\alpha
x_i$ may occur in $M$. If, instead, we relax the condition that
$M$ be affine-linear in the $c_\alpha$ to a {\em polynomial} dependence on
the $c_\alpha$, then the same bounds still apply; see
Remark~\ref{re:Poly}. But what if we relax this to
{\em rational} dependence of $M$ on the $c_\alpha$? Given that $p_{n,d}$
is only linear in the $c_\alpha$ it seems unlikely that allowing $M$ to
be rational in the $c_\alpha$ we would gain anything, but we currently
do not know how to formalise this intuition. On the other hand, in
cases where a (non-uniform) determinantal representation of size $d$ is known to
exist for every (or sufficiently general) polynomials of degree $d$
in $d$ variables (e.g., in the case of plane curves), it follows that
this representation can be chosen to have entries {\em algebraic} in
the $c_\alpha$. This observation rules out approaches
aimed at proving lower bounds in a too general setting.
\vskip1in
| 75c5fe2c1eadf6ae77ba39cd9f0b86baebd4dba8 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
The dynamics of electrolytic solutions is essential in the description of most most processes in (bio)chemistry and soft matter physics.
This includes electrostatic screening via the formation of double layers, electro-osmotic flow\cite{anderson89a,paxton06a,ibele07a,reinmuller12a}, electrophoresis\cite{obrien78a,smoluchowski17a,hueckel24a,henry31a}, and self-electrophoresis\cite{paxton04a,moran11a,moran14a,brown14a,ebbens14a,sabass12b,brown15a}.
Industrial applications range from DNA sequencing\cite{woolley95a,mikheyev14a} and oil recovery\cite{pagonabarraga10a} to the detection\cite{castellanos01a}, separation\cite{schaegger87a}, and characterization\cite{blanche81a} of analyte molecules.
The coupled occurrence of diffusion, hydrodynamics, and electrostatics, often referred to as electrokinetics or electrohydrodynamics, gives rise to complex physical behavior.
Electrokinetic processes involve dynamics on vastly different length scales with double layers typically having a nanometer scale in aqueous solution and system geometries often ranging from several micrometers to millimeters.
The small length scales make it difficult to access the nanoscopic details in experiment, while the large discrepancy in length scales poses a challenge to simulation methods.
Existing methods that solve the electrokinetic equations numerically are based on a variety of algorithms for hydrodynamics, generally falling into two classes.
Firstly, particle-based algorithms, which include dissipative particle dynamics\cite{medina_efficient_2015,smiatek11a} (DPD) and multi-particle collision dynamics\cite{frank08a,frank09a} (MPCD).
Secondly, continuum (grid-based) algorithms, which encompass methods like finite elements\cite{moran11a,moran14a,brown15a,lewis72a} (FEM), the boundary element method\cite{allison05a,hoyt98a,house12a} (BEM), finite differences\cite{schmitz12a}, finite volumes\cite{jeong03a}, electrokinetics based on lattice Boltzmann (LB)\cite{capuani04a,warren97a}, and the smooth profile method \cite{nakayama08a} (SP).
For additional details, we refer to the overviews given in Refs.~\onlinecite{pagonabarraga10a,rotenberg13a}.
The particle-based methods solve the full time-dependent problem and are intrinsically able to include mobile charged colloids and macromolecules.
The downside of these methods is that solute ions, water, and macromolecules are resolved on the same scale.
This sets a low limit for the maximum size of the simulation domain, given current computational abilities.
Continuum solvers allow for the study of systems with much larger length scales because the ions are not resolved.
In general, introducing moving boundaries in numeric solvers for the continuum electrokinetic equations poses a more challenging problem than for their particle-based counterparts.
Tackling this issue, however, is well worth the effort, as it grants access to the wide variety of physical problems
beyond those that can be studied via Galilean transformation to the rest frame of the charged object, an approach that is not possible in general.
Moving particles will permit the study of interactions of multiple (self-)electrophoretically driven colloids, as well as their interactions with any stationary boundaries in the system.
When it comes to resolving large system sizes, the FEM method has proven itself to be a very efficient solver for the stationary electrokinetic equations.
FEM has been successfully applied in arbitrary geometries and for a diverse range of physical systems, including: nanopores\cite{white_ion_2008,lan_nanoparticle_2011,rempfer16b}, self-electrophoresing colloids\cite{moran11a,moran14a,kreissl16a,brown15a,qian08b,wang13b,chiang14a}, and electro-osmotic pumping\cite{mengeaud02a,bianchi00a}.
The strength of the FEM lies in its ability to adaptively increase the resolution in regions with high gradients, i.e., the electric double layers.
The downside is that the treatment of time-dependent boundary positions is computationally exceptionally expensive, due to the constant need for remeshing.
However, there are examples of its application to phoretic problems with moving boundaries\cite{yang16a}.
An alternative to FEM was proposed by \citet{capuani04a}. Their lattice electrokinetics method (EK) combines finite differences and finite volumes to solve the diffusion-advection of solute and the electrostatics problem.
For hydrodynamics, they exploit the computational efficiency of the LB method\cite{mcnamara88a,duenweg09a}.
Unlike earlier methods that aimed to achieve a similar level of description\cite{warren97a}, EK propagates the solute by considering solute fluxes between neighboring lattice cells instead of concentrations in lattice cells, thus ensuring local mass conservation and reducing discretization artifacts\cite{capuani04a}.
Furthermore, EK, like LB, is a local algorithm, with the exception of the electrostatics solver, making it generally more efficient than FEM and other solvers, which are fully nonlocal.
Of specific interest is EK's ability to resolve the particles' surfaces as boundary conditions on the grid.
This allows for the study of extended objects and straightforward incorporation of local variation of the surface properties, which is, e.g., useful in studying self-propelled particles \cite{scagliarini16a}.
Extending this method further with dynamically-adaptive grid refinement, a common practice in LB\cite{schornbaum16a}, will constitute an algorithm fully capable of directly modeling microfluidic systems.
The EK method has previously been used to study problems such as electrophoresis\cite{giupponi11a} and sedimentation\cite{capuani04a} of single charged spherical colloids.
Other applications include charge transport in porous media\cite{rotenberg08a,obliger14a} and the translocation of DNA molecules through nanopores\cite{reboux06a}, where this method can tremendously reduce computational effort when compared to particle-based methods, and thus enable one to reach experimentally relevant length and time scales.
In this manuscript, we present an extension of the lattice electrokinetics method proposed by \citet{capuani04a} to include moving boundaries.
For the underlying LB method, a moving boundary scheme\cite{ladd94a,aidun98a} is already widely employed for moving particles.
To adapt the EK method for moving boundaries, we first introduce solute mass and charge conservation by displacing solute from cells that are claimed by a moving particle.
We subsequently add a partial volume smoothing to reduce the effect of a cell being claimed in a single time step.
We do so by incorporating a term in the solute flux calculation that depends on the amount of volume in a cell that is overlapped by a particle.
This allows for a gradual expulsion of solute from a cell as it is increasingly overlapped.
We validate our method by considering the problem of electrophoresis of spherical colloids in an external field.
We observe that the moving boundary method is capable of reproducing results using the Capuani method with fixed boundaries (in the co-moving frame).
Moreover, we observe that the smoothing term is necessary to prevent strong variations in the particle velocity due to the grid discretization --- it reduces the oscillations by more than a factor of $10$.
The added computational cost due to the moving boundaries amounts to approximately a $10\%$ increase per time step, compared to the stationary boundary algorithm of \citet{capuani04a}.
The method we describe retains the flexibility and efficiency of the LB and EK methods, while granting the EK method access to an entire class of problems where the relative motion of multiple particles is of relevance.
The remainder of this paper is laid out as follows:
In \cref{sec:equations}, we give the governing equations.
In \cref{sec:numerical}, we introduce the numerical methods, including a novel method for coupling moving particles with EK.
In \cref{sec:validation}, we validate the new method by comparing equivalent electrophoretic and electro-osmotic systems.
We conclude in \cref{sec:conclusion}.
\section{The Electrokinetic Equations}
\label{sec:equations}
In this section, we introduce the electrokinetic equations for bulk electrolytes.
These describe the motion of a charged solute in a solvent fluid by a diffusion-advection process.
Boundary conditions will be discussed in \cref{sec:numerical}.
For each solute species $k$, the solute flux is given by
\begin{align}
\vec{j}_k(\vec{r},t)=&
\underbrace{-D_k\vec{\nabla}\rho_k(\vec{r},t)
-\frac{D_k}{k_\text{B} T} z_ke \rho_k(\vec{r},t) \vec{\nabla} \Phi(\vec{r},t)}_{\vec{j}^\text{diff}_k} \nonumber\\
&+ \underbrace{\rho_k(\vec{r},t) \vec{u}(\vec{r},t)}_{\vec{j}^\text{adv}_k}
\label{eq:flux}
\end{align}
and obeys the continuity equation
\begin{equation}
\frac{\partial \rho_k}{\partial t}(\vec{r},t) = - \vec{\nabla} \cdot \vec{j}_k(\vec{r},t)
\label{eq:continuity}
.
\end{equation}
Here, $\rho_k(\vec{r},t)$ is the concentration field and $\vec{j}_k(\vec{r},t)$ is the associated flux.
$\Phi(\vec{r},t)$ refers to the electrostatic potential and $\vec{u}(\vec{r},t)$ to the fluid velocity.
$\vec{\nabla}\cdot$ is the divergence operator; $\partial/\partial t$ denotes the time derivative.
$D_k$ is the diffusion coefficient of the $k$th solute, $k_\text{B}$ Boltzmann's constant, $T$ the absolute temperature, $z_k$ is the $k$th solute's ionic valency, and $e$ is the unit charge.
The electrostatic potential is given by Poisson's equation
\begin{align}
\nabla^2 \Phi(\vec{r},t) &= -\frac{1}{\varepsilon} \sum\limits_k z_ke\rho_k(\vec{r},t) \nonumber \\ &= -\frac{4\pi \lambda_\text{B} k_\text{B}T}{e} \sum\limits_k z_k\rho_k(\vec{r},t)
,
\label{eq:poisson}
\end{align}
with the Bjerrum length
\begin{equation}
\lambda_\text{B}=\frac{e^2}{4\pi \varepsilon k_\text{B} T}
.
\end{equation}
A spatially homogeneous dielectric permittivity $\varepsilon$ is assumed here.
As the relevant length scales in colloidal systems are on the nano- and micrometer scale, we are in the low Reynolds number regime --- viscous friction dominates over inertial forces --- and the dynamics of the fluid is thus described by the stationary incompressible Stokes equations:
\begin{align}
\eta \nabla^2\vec{u}(\vec{r},t) &= \vec{\nabla} p(\vec{r},t) - \vec{f}_\text{ext}(\vec{r},t) , \label{eq:stokes} \\
\vec{\nabla}\cdot\vec{u}(\vec{r},t) &= 0 \label{eq:continuity-fluid}
,
\end{align}
with the kinematic viscosity $\eta$, the pressure density $p(\vec{r},t)$, and the applied force density $\vec{f}_\text{ext}(\vec{r},t)$.
The fluid's continuity \cref{eq:continuity-fluid} neglects the contribution of the solute to the total mass, which is reasonable at typical solute concentrations up to several mol/l \citep{degraaf15c}.
The force density couples the motion of the fluid to the dynamics of the solutes via the following equation\cite{rempfer16a}:
\begin{equation}
\vec{f}_\text{ext}(\vec{r},t)=k_BT\sum\limits_k \frac{\vec{j}_k^\text{diff}(\vec{r},t)}{D_k}
.
\label{eq:fluidcoupling}
\end{equation}
This represents a frictional coupling proportional to the relative velocities of fluid and ions.
\section{Numerical Methods}
\label{sec:numerical}
In this section, we introduce the numerical methods employed to solve the electrokinetic equations~\eqref{eq:flux}-\eqref{eq:fluidcoupling}.
\Cref{sec:lb,sec:ek} cover the LB method and the EK method, respectively, while \cref{sec:ladd} describes the moving boundary method for coupling particles into LB.
\Cref{sec:ek-moving} describes a new method for combining moving boundaries with EK.
\subsection{Lattice Boltzmann}
\label{sec:lb}
The LB method is based on solving the Boltzmann transport equation,
\begin{align}
\frac{\mathrm{d}f}{\mathrm{d}t}=
&\frac{\partial f}{\partial t} + \vec{v} \cdot \vec{\nabla}\!_x f + \frac{1}{m} \vec{F} \cdot \vec{\nabla}\!_v f
\left. \frac{\partial f}{\partial t} \right| _\text{collision}
\equiv \tens{\Omega} f
\label{eq:transport}
.
\end{align}
$\vec{\nabla}\!_a$ is the gradient operator with respect to $\vec{a}$.
$f=f(\vec{x},\vec{v}, t)$ is a system's single-particle phase space probability distribution function.
It gives the probability density of finding a particle at position $\vec{x}$ with a velocity $\vec{v}$ at time $t$.
$\vec{F}$ is the force acting on the particle.
The right-hand side is the collision operator.
The Boltzmann transport equation~\eqref{eq:transport} describes microscopic particle motion, while the Stokes equation~\eqref{eq:stokes} describes flow of a Newtonian continuum fluid.
However, both are based on the same conservation laws, those of mass and momentum,
and they can therefore be used interchangeably when studying mesoscopic hydrodynamics.
The LB method linearizes the collision operator $\tens{\Omega}$ on the right-hand side of \Cref{eq:transport} around the equilibrium distribution $f^\text{eq}$.
The total derivative on the left-hand side is discretized using finite differences\cite{duenweg09a} with a lattice spacing of $a_\text{grid}$ and a time step of $\tau$ to obtain
\begin{equation}
f_i(\vec{r}+\tau\vec{c}_i, t+\tau) = f_i(\vec{r},t)+\sum\limits_j \Omega_{ij} \underbrace{\left(f_j(\vec{r},t) - f^\text{eq}_j(\vec{r},t)\right)}_{f^\text{neq}_j(\vec{r},t)}
,
\label{eq:lb}
\end{equation}
with the lattice vectors $\vec{c}_i$.
The collision represents a linear relaxation of the populations towards their equilibrium Maxwell-Boltzmann distribution.
At each lattice site, there is a number of fluid populations $f_i$, e.g., 19 different ones for the widely used D3Q19 model applied in this work, corresponding to a set of velocity vectors $\vec{c}_i$ connecting the site to neighboring cells.
For D3Q19, these are the rest mode, the six face neighbors, and the twelve edge neighbors.
In each time step, a collision operation is performed that redistributes fluid between the 19 populations within one cell, i.e., \cref{eq:lb} is applied with only $\tens{\Omega}$ on the right-hand side.
Next, a streaming step is performed which transfers the populations to the neighboring cell in the direction of their respective velocity vector, i.e., \cref{eq:lb} is applied without $\tens{\Omega}$ on the right-hand side.
From the populations $f_i$, the macroscopic physical quantities of fluid density and flow velocity are obtained using
\begin{align}
\rho(\vec{r},t)&=\sum\limits_i f_i(\vec{r},t) , \\
\vec{u}(\vec{r},t)&=\frac{1}{\rho(\vec{r},t)}\sum\limits_i f_i(\vec{r},t)\vec{c}_i .
\end{align}
Throughout the manuscript, we choose a two relaxation time (TRT) LB\cite{ginzburg08a} collision operator for its high fidelity of reproducing no-slip (zero velocity) boundaries:
\begin{subequations}
\begin{align}
\sum\limits_j
\Omega_{ij} f^\text{neq}_j(\vec{r},t)
&= \lambda_\text{e} \left(f_i^+ - f_i^{\text{eq}+}\right) +
\lambda_\text{o} \left(f_i^- - f_i^{\text{eq}-}\right) , \\
f_i^{\pm}(\vec{r},t) &= \frac{1}{2} \left( f_i(\vec{r},t) \pm f_{\bar{i}}(\vec{r},t) \right) , \\
f_i^{\text{eq}\pm}(\vec{r},t) &= \frac{1}{2} \left( f_i^\text{eq}(\vec{r},t) \pm f_{\bar{i}}^\text{eq}(\vec{r},t) \right)
.
\end{align}
\end{subequations}
$f_i^{\pm}$ and $f_i^{\text{eq}\pm}$ are symmetric ($+$) and antisymmetric ($-$) combinations of populations and equilibrium populations, respectively.
The symmetric relaxation parameter $\lambda_\text{e}$ derives from the fluid viscosity, while the antisymmetric $\lambda_\text{o}=3/16$ is tuned to minimize boundary slip\cite{ginzburg08b}.
The index $\bar{i}$ is the one whose velocity vector points in the direction opposite to the one with index $i$, i.e. $\vec{c}_{\bar{i}}=-\vec{c}_i$.
We will have forces acting on the fluid that vary in time and space for moving boundaries, therefore we need to augment the collision operation to be $\Omega_{ij}f^\text{neq}_j+\Delta_i$ with the appropriate force term $\Delta_i$.
We utilize the one originally constructed by \citet{guo02a} and generalized to multiple relaxation times\cite{schiller08a,schiller14a}:
\begin{subequations}
\begin{align}
\Delta_i &= \frac{3w_ia_\text{grid}^2\tau^2}{\rho} \left[\vec{f}_\text{ext}\cdot\vec{c}_i \phantom{\frac{\tau^2}{a_\text{grid}^2}}\right. \nonumber \\
& \phantom{++}+ \left.\frac{3}{2} \mathrm{Tr}\left(
\tens{G} \left( \vec{c}_i\otimes\vec{c}_i \right)\frac{\tau^2}{a_\text{grid}^2} - \frac{1}{3}\tens{G}
\right)
\right] \label{eq:guo} , \\
\tens{G} &= \frac{2+\lambda_\text{e}}{2} \left( \vec{u}\otimes\vec{f}_\text{ext} + \vec{f}_\text{ext}\otimes\vec{u} \right)
,
\end{align}
\end{subequations}
where $\otimes$ denotes the tensor (dyadic) product of two vectors and $\mathrm{Tr}$ the trace of a matrix.
$w_i$ is the lattice weight factor associated with direction $i$,
$\vec{f}_\text{ext}$ is the force acting on a lattice cell, and $\vec{u}$ the flow velocity in that cell.
Until we introduce the moving boundaries in \cref{sec:ladd}, we can consider colloidal particles by requiring them to be fixed relative to the lattice.
This corresponds to cells inside the colloid being a no-slip boundary for the fluid (corresponding to the co-moving frame, if a Galilean transformation is permitted).
In that case, the streaming step is modified so that populations streaming into the boundary are reflected back.
A generalization to arbitrary boundary velocities $\vec{v}_\text{b}$ is possible\cite{zou97a} and leads to
\begin{equation}
f_i(\vec{r}_\text{b}+\tau\vec{c}_i,t+\tau)
=f_{\bar{i}}(\vec{r}_\text{b},t)
+ \frac{6 w_i\tau^2}{\rhoa_\text{grid}^2} \vec{c}_i\cdot \vec{v}_\text{b}
,
\end{equation}
where $\vec{r}_\text{b}$ is the boundary node position.
This is referred to as a velocity bounce-back boundary.
We do not include thermal fluctuations of the solvent in this work.
Neither do we do so for the solute, as will be discussed in \cref{sec:ek}.
A consistent thermalization for the pure LB algorithm is available\cite{duenweg07a},
and such fluctuations may be significant for nanoscale systems.
\subsection{Moving LB Boundaries}
\label{sec:ladd}
There are multiple methods that incorporate particles into an LB fluid\cite{duenweg09a},
ranging from point-particle descriptions\cite{ahlrichs98a} to particles that are resolved on the grid\cite{ladd94a,aidun98a}.
In this paper, we are interested in the latter, for which
the method by \citet{ladd94a} is the obvious choice to achieve coupling.
It considers all cells inside a particle as no-slip boundaries in the particle-co-moving frame.
In the lab frame, this corresponds to a velocity boundary condition with velocity $\vec{v}_\text{b}$ equal to the particle's surface velocity
\begin{equation}
\vec{v}_\text{b}(\vec{r}_\text{b},t)=\vec{v}(t) + \vec{\omega}(t) \times (\vec{r}_\text{b} - \vec{r}(t))
.
\end{equation}
Here, the particle is located at $\vec{r}$ and has linear velocity $\vec{v}$ and angular velocity $\vec{\omega}$.
The linear and angular momentum exchanged during the bounce-back is transferred to the particle by applying an appropriate force and torque,
\begin{subequations}
\label{eq:force}
\begin{align}
\vec{F}(t)&=a_\text{grid}^3\sum\limits_{\vec{r}_\text{b}} \sum\limits_{i=1}^{19} \vec{c}_i \left(f_i(\vec{r}_\text{b},t)+f_{\bar{i}}(\vec{r}_\text{f},t)\right) , \\
\vec{T}(t)&=a_\text{grid}^3\sum\limits_{\vec{r}_\text{b}} \sum\limits_{i=1}^{19} \left(\vec{r}_\text{b}-\vec{r}\right) \times \vec{c}_i \left(f_i(\vec{r}_\text{b},t)+f_{\bar{i}}(\vec{r}_\text{f},t)\right)
.
\end{align}
\end{subequations}
Here, the bounce-back is taking place between the boundary cell $\vec{r}_\text{b}$ and the fluid cell $\vec{r}_\text{f}$.
After determining these forces, the particle trajectory can be integrated, in our case using a symplectic Euler integrator.
These forces, along with any non-hydrodynamic forces applied, cause the particle to move across the lattice.
While it does so, the set of cells overlapped by the particle changes.
The original method by \citet{ladd94a} has some shortcomings, due to the presence of fluid inside the particle.
These include forces exerted on the particle by the internal fluid\cite{lowe95a} and the requirement that the particle's density is the same or larger than that of the fluid.
The modification by \citet{aidun98a} removes this unphysical fluid.
However, one now needs to destroy fluid in cells that are newly overlapped by a particle and create fluid in cells that are vacated.
The former is straightforward to do, while the latter requires choosing a value for each of the $f_i$ in the cell that becomes `new' fluid.
The initialization of the new fluid cell is problematic, as the D3Q19 LB model has more microscopic degrees of freedom than specified by the hydrodynamic boundary condition.
The typical choice is to use LB's equilibrium populations for a velocity equal to the particle's velocity and a density equal to the surrounding fluid cells' average density\cite{aidun98a}.
To conserve momentum, the created / destroyed fluid's momentum $\vec{p}_\text{f}$ is removed from / added to the particle as an additional force:
\begin{equation}
\vec{F}(t)=\pm\frac{1}{\tau}\vec{p}_\text{f}(\vec{r}_\text{f},t)
.
\end{equation}
Fluid mass is never conserved, as conserving it would violate the incompressibility of the fluid. However, the time-averaged fluid mass fluctuates around the correct value while the particle traverses the grid.
\subsection{Lattice Electrokinetics}
\label{sec:ek}
The method proposed by \citet{capuani04a} discretizes \cref{eq:flux} with a 19-point stencil on the same lattice that LB is performed on.
\citet{rempfer16a} have shown that the discretization originally proposed in Ref.~\onlinecite{capuani04a} introduces errors scaling exponentially with $E/a_\text{grid}$ where $E$ is the applied electric field.
We instead use
\begin{align}
j_{ki}^\text{diff}&(\vec{r}\rightarrow\vec{r}+\vec{c}_i,t) \nonumber\\
=& \frac{D_k}{a_\text{grid}}\left(\rho_k(\vec{r},t) - \rho_k(\vec{r}+\vec{c}_i,t)\right) \nonumber\\
&- \frac{D_k z_ke}{k_\text{B} Ta_\text{grid}(1+2\sqrt{2})} \frac{\rho_k(\vec{r},t)
+ \rho_k(\vec{r}+\vec{c}_i,t)}{2} \nonumber\\
&\times \left(\Phi(\vec{r},t) - \Phi(\vec{r}+\vec{c}_i,t)\right)
\label{eq:flux-disc}
\end{align}
for the diffusive term, as in Ref.~\onlinecite{rempfer16a}.
For the advective term $j_{ki}^\text{adv}$, a volume-of-fluid method is used\cite{capuani04a}: the solute in a cell at $\vec{r}$ is virtually displaced by a distance of $\vec{u}(\vec{r},t)\tau$, where $\vec{u}$ is the velocity of the LB fluid.
This displaced cell is intersected with all 26 neighboring cells to determine the amount of solute that needs to be transferred.
Once the fluxes have been calculated, they can be propagated using the discretized form of the continuity equation~\eqref{eq:continuity},
\begin{equation}
\frac{V_0}{\tau}\left(\rho_k(\vec{r},t+\tau)-\rho_k(\vec{r},t)\right) =
- A_0 \sum\limits_i j_{ki}(\vec{r},t)
,
\end{equation}
with $V_0$ the cell volume and $A_0$ an effective surface area chosen such that \cref{eq:flux-disc} recovers the correct (bulk) diffusive mean-square displacement of $6Dt$. This results in $A_0=1+2\sqrt{2}$ for the D3Q19 lattice\cite{capuani04a}.
The fluid coupling force is a direct discretization of \cref{eq:fluidcoupling} and is applied using \cref{eq:guo}:
\begin{equation}
\vec{f}_\text{ext}(\vec{r},t)=\frac{k_\text{B}Ta_\text{grid}}{\tau}\sum_k \sum_i \frac{j_{ki}^\text{diff}(\vec{r},t)}{D_k} \vec{c}_i
.
\end{equation}
Electrostatics \eqref{eq:poisson} is treated using any available lattice-based electrostatics solver.
For this work, one based on Fast Fourier Transforms (FFT) is used, but iterative solvers for linear equation systems, such as Successive Over-Relaxation (SOR) or Krylov subspace methods, may also be applied.
The non-FFT-based solvers are also capable of incorporating inhomogeneous dielectric coefficients; the assumption of homogeneity has only entered into the specific form of Poisson's equation \eqref{eq:poisson} used here.
Besides driving the migrative flux \eqref{eq:flux-disc}, electrostatics also results in a force acting on the colloidal particles, which is added to the hydrodynamic drag force \eqref{eq:force} before integrating the particle trajectory.
At this point, EK is only capable of handling stationary boundary conditions, such as walls and particles that are stationary with respect to the lattice.
They are simply mapped into the LB fluid as a no-slip boundary in all cells inside them and any EK solute fluxes into or out of these cells is set to zero.
Using normal fluxes other than zero allows for the incorporation of chemical reactions occurring on the surface that lead to phenomena such as charge regulation and self-electrophoresis\cite{degraaf15c,brown15a}.
Boundaries may be charged by considering their charge distribution $\rho_\text{b}(\vec{r},t)$ as an additional summand when solving \cref{eq:poisson}.
As mentioned before, thermal fluctuations are not included.
The addition of noise to the concentration fields --- while obtaining the proper fluctuation-dissipation relation for the total fluid --- is non-trivial and goes beyond the scope of the current manuscript.
Thermalization of the EK algorithm based on existing numerical methods for fluctuating hydrodynamics of electrolytes\cite{peraud16a,zudrop14a} shall be the subject of future study.
\subsection{Moving EK Boundaries}
\label{sec:ek-moving}
In \cref{sec:ladd}, the well-known moving boundary method for LB was introduced.
It is not mass-conserving, so a straightforward adaptation to EK would cause the amount of solute to vary over time and thus violate charge conservation.
Furthermore, as the solute charge is typically far less homogeneously distributed than the fluid mass, one can expect that charge would not even be conserved on average over long time scales.
A simulation undergoing such a net charge drift will typically not be able to produce physically correct results.
\subsubsection{Simple Charge Conservation Scheme}
\label{sec:simplescheme}
To avoid a net charge drift, any solute from cells claimed by a particle needs to be expelled and redistributed into surrounding cells.
Such a scheme is illustrated in \cref{fig:chargecons}.
Cells vacated behind the particle are refilled by taking the sum of the solute concentration of the surrounding $N_\text{f}$ non-boundary cells and dividing it by $N_\text{f}+1$ to account for the fact that this amount of solute will now be shared with one additional cell.
This means that the solute concentration
\begin{equation}
\rho_k(\vec{r}_\text{new},t+\tau)=\frac{1}{N_\text{f}+1}\sum\limits_{i=1}^{N_\text{f}} \rho_k(\vec{r}_\text{new}+\vec{c}_i,t)
\label{eq:refill}
\end{equation}
is put into the vacated cell.
To conserve total solute mass, that amount is then removed from the surrounding non-fluid cells in amounts proportional to their current solute concentration:
\begin{equation}
\rho_k(\vec{r}+\vec{c}_i,t+\tau) = \rho_k(\vec{r}+\vec{c}_i,t) -\frac{\rho_k(\vec{r}+\vec{c}_i,t)}{N_\text{f}+1}
\label{eq:refillneighbors}
.
\end{equation}
It should be noted that expulsion and vacation do not necessarily happen at the same time in the front and back of a moving particle, and thus the total number of boundary cells oscillates over time.
\begin{figure}[b]
\centering
\includegraphics[width=\linewidth]{mb_ek}
\caption{Illustration of the mass conservation modification to the Ladd boundary scheme to make it usable for EK. Cells whose center is inside the particle are considered to be boundary nodes. The arrows indicate how solute is drawn into vacated cells (panes 2, 3, and 6) and expelled from newly-overlapped cells (panes 4 and 5).}
\label{fig:chargecons}
\end{figure}
\subsubsection{Enhanced Partial Volume Scheme}
The scheme introduced in \cref{sec:simplescheme} has the disadvantage that it moves large amounts of solute at a few points in time.
Therefore, as we demonstrate in \cref{sec:validation}, the particle's velocity can vary quite strongly during the time steps after a cell has been claimed or vacated.
To reduce these effects, we propose a partial volume scheme, which is illustrated in \cref{fig:partialvolume}.
In the following, $\Psi(\vec{r},t)$ is a field describing the volume fraction of the cell at $\vec{r}$ that is overlapped by a particle, with $\Psi=1$ meaning that the cell is completely inside the particle and $\Psi=0$ completely outside.
In the calculation of the diffusive fluxes \eqref{eq:flux-disc}, the concentrations are replaced with ones that take into account that all solute resides in the non-overlapped part of the cells.
To prevent the resulting diffusive fluxes from diverging as $\Psi\rightarrow 1$, we renormalize them by scaling them with the volume.
This leads to the following modified expression for the flux:
\begin{align}
j_{ki}^\text{diff}&(\vec{r}\rightarrow\vec{r}+\vec{c}_i,t) \nonumber\\
=& \left[ \frac{D_k}{a_\text{grid}}\left(\frac{\rho_k(\vec{r},t)}{1-\Psi(\vec{r},t)} - \frac{\rho_k(\vec{r}+\vec{c}_i,t)}{1-\Psi(\vec{r}+\vec{c}_i,t)}\right) \right. \nonumber\\
&- \frac{D_k z_ke}{2 k_\text{B} Ta_\text{grid}(1+2\sqrt{2})} \left( \frac{\rho_k(\vec{r},t)}{1-\Psi(\vec{r},t)} \right. \nonumber\\
&+ \left.\frac{\rho_k(\vec{r}+\vec{c}_i,t)}{1-\Psi(\vec{r}+\vec{c}_i,t)} \right)
\times \left. \vphantom{\frac{1}{1}} \left(\Phi(\vec{r},t) - \Phi(\vec{r}+\vec{c}_i,t)\right) \right] \nonumber\\
& \times \left(1-\Psi(\vec{r},t)\right) \left(1-\Psi(\vec{r}+\vec{c}_i,t)\right)
\label{eq:flux-pv}
.
\end{align}
With this change, refilling vacated cells as per \cref{eq:refill,eq:refillneighbors} is no longer necessary. They can be set to zero concentration and will be filled up by the diffusive flux again as $\Psi$ increases.
We calculate $\Psi$ numerically by sub-dividing each cell into 8 equally-sized cells and determining how many of them are completely inside and completely outside the particle.
For those cells that are neither, the subdivision is recursively repeated up to a maximum depth of $4$.
Expelling solute from a cell that is claimed by a particle is, however, still necessary --- even with the modified expression for the flux --- as the cell is not necessarily completely empty by the time it is claimed, due to the discretized motion of the colloid.
The expelled amount of solute with \cref{eq:flux-pv} is much smaller than with \cref{eq:flux-disc} and thus the effect of this sudden change on the simulation is reduced to acceptable levels.
One further source of sudden variations in solute fluxes is the change in electrostatic potential when the volume across which a particle's charge is distributed varies due to the fluctuation in the number of boundary cells.
Therefore, when calculating the electrostatic potential, each particle's total charge $Q=Ze$ is distributed among all cells that are at least partially overlapped by that particle:
\begin{equation}
\rho_\text{b}(\vec{r},t) = Ze\frac{\Psi(\vec{r},t)}{V_\text{p}}
,
\end{equation}
with $V_\text{p}$ the particle's (non-discrete) volume.
Inhomogeneous charge distributions are also possible as long as the charge in a cell varies smoothly while the cell is slowly claimed or vacated by the colloidal particle.
\begin{figure}[b]
\centering
\includegraphics[width=\linewidth]{mb_ek_pv}
\caption{Illustration of the partial volume scheme for moving boundaries in EK. The shading of the cells inside the particle corresponds to the overlapped volume $\Psi$ to indicate how the particle's charge is distributed across the cell layer at its surface. In the calculation of the diffusive flux, the concentrations are scaled with $1-\Psi$ to determine the effective concentrations.}
\label{fig:partialvolume}
\end{figure}
\section{Validation}
\label{sec:validation}
We implement our new algorithm using the \textsf{waLBerla} framework\cite{godenschwager13a}.
It supports several lattice Boltzmann models, including the one introduced in \cref{sec:lb}, and correctly handles the moving LB boundaries described in \cref{sec:ladd}.
We already added an implementation of the EK model described in \cref{sec:ek}.
\textsf{waLBerla} provides excellent scaling on high-performance computing clusters and contains advanced features, such as grid refinement\cite{schornbaum16a}, that may prove useful for EK simulations of complex systems in the future.
\subsection{Electrophoresis of a Single Colloid}
\label{sec:singlecolloid}
To validate the moving boundary EK method described in \cref{sec:ek-moving}, we choose a simple system consisting of an electrophoresing sphere.
Specifically, we simulate a single, homogeneously charged, spherical colloid in a cubic box with periodic boundary conditions (PBCs) undergoing electrophoresis in a uniform external electric field.
This system has already been studied extensively using the EK method without moving boundaries by considering the equivalent problem of electro-osmotic flow\cite{giupponi11a}.
The parameters used are given in \cref{tab:parameters}.
They were chosen to match one of the data points of Ref.~\onlinecite{giupponi11a}.
A salt concentration of $c=\SI{1e-3}{\mole\per\litre}$ is experimentally relevant\cite{lobaskin07a,garbov04a,medebach04b,palberg04a} and the (effective) colloid charge of $Z=30$ is realistic for a particle of radius $R=\SI{4e-9}{\metre}$.
The resulting Debye length of $\lambda_\text{D}=\SI{10}{\nano\metre}$ for that concentration is neither in the thin ($\lambda_\text{D}\ll R$) nor thick ($\lambda_\text{D}\gg R$) double layer limit and should thus demonstrate the capability of the method to deal with these intermediate regimes.
The charge of $Z=30$ was chosen such that Debye-H\"uckel theory is not applicable --- the requirement being $e\zeta < k_\text{B}T$ with $\zeta=Zk_BT\lambda_\text{B}/(R+R^2/\lambda_\text{D})$ --- while for $Z=3$, it is.
\begin{table}
\centering
\begin{tabular}{c|ll}
Temperature & $T=\SI{298.15}{\kelvin}$ & \\
Relative permittivity & $\varepsilon=78.54$ & \\
Fluid density & $\rho=\SI{997.04}{\kilo\gram\per\metre\cubed}$ & \\
Viscosity & $\eta=\SI{0.8937e-3}{\pascal\second}$ & \\
\multicolumn{3}{c}{} \\
Salt concentration
& $c=\SI{1e-3}{\mole\per\litre}$ & \\
& $c=\SI{1e-4}{\mole\per\litre}$ & \\
Salt valency & $z_\pm=\pm 1$ & \\
Diffusion coefficient & $D=\SI{2}{\metre\squared\per\second}$ & \\
External field & $E=\SI{256.9e3}{\volt\per\metre}$ & \\
\multicolumn{3}{c}{} \\
Sphere radius & $R\approx\SIrange{3e-9}{8e-9}{\metre}$ & \\
Sphere charge & $Q=\SI{3}{\elementarycharge}=\SI{0.4807e-18}{\coulomb}$ & \\
& $Q=\SI{30}{\elementarycharge}=\SI{4.8807e-18}{\coulomb}$ & \\
Density of particle & $\rho_\text{p}=2\rho$ & \\
Box length & $L=\SI{64e-9}{\metre}$ & \\
\multicolumn{3}{c}{} \\
Length unit & $\Delta x = \SI{1e-9}{\metre}$ & \\
Mass unit & $\Delta m = \rho \Delta x^3=\SI{9.97e-25}{\kilo\gram}$ & \\
Energy unit & $\Delta E=k_\text{B}T=\SI{4.12e-21}{\joule}$ & \\
Time unit & $\Delta t=\sqrt{{\Delta m\Delta x^2}/{\Delta E}}$ & \\
& $\phantom{\Delta t}=\SI{1.56e-11}{\second}$ & \\
LB grid spacing & $a_\text{grid}=\Delta x = \SI{1e-9}{\metre}$ & \\
LB time step & $\tau=0.2\Delta t=\SI{3.1e-12}{\second}$ & \\
\end{tabular}
\caption{Parameters used in the simulation and their conversion to simulation units.
Note that our choice of $\rho_\text{p}=2\rho$ is arbitrary and irrelevant to the working of our algorithm, as verified for a single system where we used $\rho_\text{p}=\rho$.}
\label{tab:parameters}
\end{table}
The salt concentration $c$ and the grid spacing $a_\text{grid}$ need to be chosen such that one Debye length is resolved by a minimum of approximately 4 cells, as we will see during validation.
This precludes us from using a salt concentration of much higher than $c=\SI{1e-3}{\mole\per\litre}$ without increasing the grid resolution.
However, sufficiently resolving the double layer is a general requirement of the EK method\cite{capuani04a}, as well as for FEM and other algorithms, and not specific to the use of moving boundaries.
We will compare the results of the different simulation methods via a single number, the reduced electrophoretic mobility\cite{wiersema66a}:
\begin{equation}
\tilde\mu=\frac{6\pi \eta \lambda_\text{B}}{e}\frac{v}{E}
,
\end{equation}
where $v$ is the speed of the particle's motion relative to the bulk fluid and $E$ is the applied electric field.
\Cref{fig:bothmethods} compares results obtained for the same parameter set (particle radius $R\approx\SI{4e-9}{\metre}$, particle charge $Z=30$, and salt concentration $c=\SI{1e-3}{\mole\per\litre}$) for a simulation with a fixed particle, a moving particle without, and a moving particle with the partial volume scheme, respectively.
The effective particle radius in the moving scheme is obtained by counting the number of cells that are completely inside the particle, averaging that number over the simulation time, and setting it equal to the sphere volume $(4/3)\pi R^3$.
This allows us to choose $R$ such that the moving and stationary simulations study the same physical system.
One can see that both moving particle methods on average agree with the fixed boundary simulation's stationary mobility value to within $1.4\%$ and $2.4\%$, respectively.
Agreement of the full time evolution is not expected as the electro-osmotic (fixed boundary) and electrophoretic (moving boundary) problems are, in fact, only equivalent in the stationary state.
The method without partial volumes results in a mobility oscillating with an amplitude of $20.3\%$ around the mean and a period corresponding to the time it takes the particle to move forward by one grid cell.
The partial volume scheme on the other hand reduces the mobility oscillations to $1.7\%$, which is generally small enough to consider any instantaneous value to be a good estimate for the true value.
The period remains bound to the particle moving forward by one cell.
\begin{figure}[t]
\centering
\includegraphics[width=\linewidth]{compare}
\caption{Comparison of the transient behavior ($t$ denotes time) of the reduced electrophoretic mobility $\tilde\mu$ for the fixed particle (purple curve), the particle moving via the simple moving boundary scheme (blue curve), and the particle moving via the partial volume scheme (green curve).
Here, $R=\SI{4e-9}{\metre}$, $Z=30$, and $c=\SI{1e-3}{\mole\per\litre}$ were used.}
\label{fig:bothmethods}
\end{figure}
To ensure that the method does not adversely affect the shape of the electric double layer around the colloid, we examine the charge distribution around the colloid in \cref{fig:chargedist}.
We see there is a small deviation close to the colloid's surface, but we can attribute this to the slight difference in particle size, as discussed above.
\begin{figure}[b]
\centering
\includegraphics[width=\linewidth]{chargedist}
\caption{Comparison of the shape of the double layer around the colloid for the fixed and moving particles.
Here, $R\approx\SI{4e-9}{\metre}$, $Z=30$, and $c=\SI{1e-3}{\mole\per\litre}$ were used.
The shaded area refers to the space taken up by the colloid.
The dashed line shows the absolute difference between the two charge distributions.}
\label{fig:chargedist}
\end{figure}
We further validated the partial volume method by running additional simulations with various concentrations $c$, particle charges $Z$, and particle radii $R$.
The results are presented in \cref{fig:results} and are compared with reference results obtained by solving the electrokinetic equations with the FEM solver \textsf{COMSOL}.
Note that we use a finite but large simulation domain in the FEM model, while the EK model handles PBCs naturally; this may lead to slight differences in the results obtained with both methods.
We observe an excellent agreement between fixed and moving boundary simulations, as well as the FEM reference results.
The agreement improves as the particle size is increased, and thus the quality of the staircase approximation to its spherical shape.
For example, for $R\approx\SI{7e-9}{m}$, $Z=30$, and $c=\SI{1e-3}{\mole\per\litre}$, the oscillation amplitude drops to $1.3\%$.
The remaining difference in the mobility is easily explained by a slight mismatch in volume (and thus effective radius) between the fixed and the moving particle.
Finally, we examined the speed of our simulation compared to the method of \citet{capuani04a} for the geometries that we considered.
The moving boundary algorithm adds only 2\% to the simulation time per time step, when compared to the equivalent fixed boundary simulation.
Our smoothing results in a further increase of less than 10\% in time, when compared to the non-smoothed moving boundaries.
Together, this shows that our method does not incur an unreasonable computational cost and can therefore be straightforwardly applied to domain sizes that are currently accessible to the \citeauthor{capuani04a} method\cite{capuani04a,giupponi11a}.
\begin{figure}[t]
\centering
\includegraphics[width=\linewidth]{mobilities}
\caption{Comparison of the steady-state value of the electrophoretic mobility for different values of $R$, $c$, and $Z$.}
\label{fig:results}
\end{figure}
\subsection{Electrophoresis of a Colloidal Suspension}
In this section, we consider the electrophoresis of a system comprised of several spheres.
This allows us to verify our moving boundary method in a more physically complex system with interactions that the original \citeauthor{capuani04a} method could not access.
We pick one parameter set ($R\approx\SI{4e-9}{\metre}$, $Z=30$, and $c=\SI{1e-3}{\mole\per\litre}$) and repeat the simulation from \cref{sec:singlecolloid}.
At these parameters, the volume fraction is about 0.1\% and the double layers do not overlap noticeably, therefore only long-ranged hydrodynamic interactions are expected to mediate interactions between the colloids.
Instead of having one spherical colloid interact with only its periodic images, we now enlarge the simulation box and add additional identical spheres while keeping the colloidal volume fraction constant.
The colloids are either positioned randomly in a cubic box or in a regular lattice in a cuboidal box.
This kind of simulation is typically performed to ensure the results do not suffer from artifacts of the PBCs\cite{lobaskin07a}.
As opposed to the single colloid and its periodic images, which always maintain their relative positions, the colloids in this simulation are free to move relative to each other.
\Cref{fig:multiple} shows that the electrophoretic mobility is indeed mostly independent of the number of colloids, varying $0.9\%$ from the average and $1.8\%$ from the value for one colloid in PBCs.
This matches the findings of \citet{lobaskin07a} using colloids modeled with the Ahlrichs-D\"unweg coupling\cite{ahlrichs98a,fischer15a,degraaf15b}, where agreement to within a few percent was found.
The results of \cref{sec:singlecolloid} are therefore not artifacts of the periodicity of the simulation domain.
For the case of a cubic simulation domain, we furthermore observe that the alignment of the colloids on a regular body-centered cubic lattice is a stable configuration.
\begin{figure}[t]
\centering
\includegraphics[width=\linewidth]{multi}
\caption{Comparison of the steady-state value of the electrophoretic mobility for different numbers of colloids at constant volume fraction.
Here, $R\approx\SI{4e-9}{\metre}$, $Z=30$, and $c=\SI{1e-3}{\mole\per\litre}$ were used.
The error bars show the spread of values for different initial random placements of colloids in cubic boxes, while the squares show the value obtained for initial placement on a regular lattice.}
\label{fig:multiple}
\end{figure}
\section{Conclusion}
\label{sec:conclusion}
Summarizing, we have introduced a method to simulate electrokinetic phenomena in colloidal suspensions.
Our scheme builds upon the lattice electrokinetics algorithm of \citet{capuani04a},
which itself is capable of simulating moving colloids only by considering them as stationary boundary conditions in the Galilean-transformed (co-moving) frame.
Motion of the colloids relative to the lattice can be incorporated by employing a method similar to the moving boundary method for LB\cite{ladd94a}.
A key modification is required, however: We introduce mass conservation for the solute species in order to conserve charge.
This is accomplished by redistributing solute from and to cells neighboring the ones that were recently vacated and claimed by the particle, respectively.
The above procedure is, in principle, sufficient to enable the simulation of, for example, colloidal electrophoresis.
However, in practice, further improvements are desirable to allow for the simulation of colloids with as few LB cells as possible.
To reduce the effects of sudden and strong solute fluxes, when a cell is claimed or vacated by a particle, a smoothing scheme by partial volumes is introduced. That is, the electrokinetic equations are solved for an effective concentration that incorporates what fraction of a cell is actually accessible to the solute, i.e., not overlapped by a colloid.
This form of smoothing reduces the lattice artifacts by more than a decade in our test simulations.
We also showed that having smoothing and moving boundaries only slows down the simulation by about $10\%$ compared to the original Capuani algorithm.
We validated our method for the electrophoresis of spherical colloids.
For a single sphere, we find excellent agreement of the reduced steady-state electrophoretic mobility $\tilde\mu$ with the one obtained for the equivalent problem of electro-osmotic flow around a fixed sphere (co-moving frame).
One can also obtain a reliable estimate for the mobility without the smoothing via partial volumes, but in this case it is necessary to average $\tilde\mu$ over one period of its oscillations.
While this is a feasible solution for the system under consideration here, systems where the transient behavior of multiple particles is of interest would require averaging over many periods, thus requiring much longer overall simulation times.
Alternatively, the oscillations could be reduced by increasing the grid resolution.
This is again not desirable, as it comes with a steep increase in required computation time --- performing the same simulation at twice the grid resolution takes eight times as much computing time.
It is therefore clear that our smoothing is a prerequisite for the efficient study of electrokinetic moving boundary problems.
Our moving boundary EK method was further verified for the electrophoresis of multiple spheres that were free to move and interact.
We find the electrophoretic mobility of the spheres to be almost independent of their initial arrangement.
Such a simulation can only be performed with moving boundaries and thus demonstrates the power of the method presented.
Note that our moving boundary EK method imposes no limitations on simulation parameters beyond those already present in the original non-moving algorithm.
These requirements include sufficient discretization of the double layer and limitations on the choice of time step, diffusivity, and grid discretization\cite{capuani04a}.
Furthermore, it leaves the EK and LB algorithms almost unmodified, which makes it easy to incorporate into existing LB or EK simulation codes.
Finally, it also does not influence the scaling behavior of these algorithms as the work required to map a particle onto the lattice is linear in the number of lattice cells and particles.
In this investigation, we have primarily studied the case of external electrophoresis of a single colloid.
As we have shown, however, our method can also be employed to investigate cases where Galilean transformation to a co-moving frame for the study of an equivalent electro-osmotic is not possible.
These types of systems will be the subject of future work and could include a diverse range of systems.
For example, our method is readily applicable to study the translocation, characterization, concentration, separation, and transport of DNA, proteins, and other biochemical analytes; interactions between oppositely charged nanoparticles in an oscillatory electric field; collective dynamics of self-electrophoresing colloids, and many others.
Our work thus opens the door for the study of a wide range of physical systems that were previously inaccessible to continuum lattice-based methods.
\begin{acknowledgments}
MK, GR, JdG, and CH thank the DFG for funding through the SPP 1726 ``Microswimmers --- From Single Particle Motion to Collective Behavior''.
GR and CH acknowledge further funding through the SFB716, TP C.5.
JdG gratefully acknowledges financial support by an NWO Rubicon Grant (\#680501210)
and funding by a Marie Sk\l odowska-Curie Intra European Fellowship (G.A. No. 654916) within Horizon 2020.
We thank G. Davies and J. Harting for useful discussions.
\end{acknowledgments}
| 6ffd832729b8ef957ca010a7320769f6fd11ad69 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Unless stated otherwise, the graphs considered in this paper are simple, loopless and finite. A \emph{homomorphism} $h$ of a graph $G$ to a graph $H$ is a mapping $h:V(G)\to V(H)$ such that adjacency is preserved by $h$, that is, the images of two adjacent vertices of $G$ must be adjacent in $H$. If such a mapping exists, we note $G\to H$. For a fixed graph $H$, given an input graph $G$, the decision problem \PBCOL{$H$} (whose name is derived from the proximity of the problem to proper vertex-colouring) consists of determining whether $G\to H$ holds. Problems of the form \PBCOL{$H$} for some fixed graph $H$, are called \emph{homomorphism problems}. A classic theorem of Hell and Ne\v{s}et\v{r}il~\cite{HN90} states a \emph{dichotomy} for this problem: if $H$ is bipartite, \PBCOL{$H$} is polynomial-time solvable; otherwise, it is NP-complete.
\medskip
\noindent\textbf{Tropical graphs.} As an extension of graph homomorphisms, homomorphisms of edge-coloured graphs have been studied, see for example~\cite{AM98,Bthesis,B94,BFHN15,BH00}. In this paper, we consider the variant where the \emph{vertices} are coloured. We initiate the study of \emph{tropical graph homomorphism problems}, in which the vertex sets of the graphs are partitioned into colour classes. Formally, a \emph{tropical graph} $(G,c)$ is a graph $G$ together with a (not necessarily proper) vertex-colouring $c:V(G)\to C$ of $G$, where $C$ is a set of colours. If $|C|=k$, we say that $(G,c)$ is a \emph{$k$-tropical graph}. Given two tropical graphs $(G,c_1)$ and $(H,c_2)$ (where the colour set of $c_1$ is a subset of the colour set of $c_2$), a homomorphism $h$ of $(G,c_1)$ to $(H,c_2)$ is a homomorphism of $G$ to $H$ that also preserves the colours, that is, for each vertex $v$ of $G$, $c_1(v)=c_2(h(v))$. For a fixed tropical graph $(H,c)$, problem \PBCOL{$(H,c)$} asks whether, given an input tropical graph $(G,c_1)$, we have $(G,c_1)\to (H,c)$.
\medskip
\noindent\textbf{The homomorphism factoring problem.} Brewster and MacGillivray defined the following related problem in~\cite{BM97}. For two fixed graphs $H$ and $Y$ and a homomorphism $h$ of $H$ to $Y$, the \textsc{$(H,h,Y)$-Factoring} problem takes as an input, a graph $G$ together with a homomorphism $g$ of $G$ to $Y$, and asks for the existence of a homomorphism $f$ of $G$ to $H$ such that $f=h\circ g$.
The \PBCOL{$(H,c)$} problem corresponds to \textsc{$(H,c,K_{|C|}^+)$-Factoring} where $K_{|C|}^+$ is the complete graph on $|C|$ vertices with all loops (and with $C$ the set of colours used by $c$). (Note that in~\cite{BM97}, loops were not considered.)
\medskip
\noindent\textbf{Constraint satisfaction problems (CSPs).} Graph homomorphism problems fall into a more general class of decision problems, the \emph{constraint satisfaction problems}, defined for \emph{relational structures}. A relational structure $S$ over a \emph{vocabulary} (a vocabulary is a set of pairs $(R_i,a_i)$ of relation names and arities) consists of a \emph{domain} $V(S)$ of vertices together with a set of relations corresponding to the vocabulary, that is, $R_i\subseteq V(S)^{a_i}$ for each relation $R_i$ of the vocabulary. Given two relational structures $S$ and $T$ over the same vocabulary, a homomorphism of $S$ to $T$ is a mapping $h:V(S)\to V(T)$ such that each relation $R_i$ is preserved, that is, for each subset of $V(S)^{a_i}$ of $R_i$ in $S$, its image set in $T$ also belongs to $R_i$. For a fixed relational structure $T$, \textsc{$T$-CSP} is the decision problem asking whether a given input relational structure has a homomorphism to $T$.
Using this terminology, a graph $H$ is a relational structure over the vocabulary $\{(A,2)\}$ consisting of a single binary relation $A$ (adjacency). Hence, \PBCOL{$H$} is a CSP. Further, \PBCOL{$(H,c)$} is equivalent to the problem \textsc{$C(H,c)$-CSP}, where $C(H,c)$ is obtained from $H$ by adding a set of $k$ unary relations to $H$ (one for each colour class of the $k$-colouring $c$).
\medskip
\noindent\textbf{The Dichotomy Conjecture.} In their celebrated paper~\cite{FV98}, Feder and Vardi posed the following conjecture.
\begin{conjecture}[Feder and Vardi~\cite{FV98}]\label{conj:CSP-dicho}
For every fixed relational structure $T$, \textsc{$T$-CSP} is polynomial-time solvable or NP-complete.
\end{conjecture}
Conjecture~\ref{conj:CSP-dicho} became known as the \emph{Dichotomy Conjecture} and has given rise to extensive work in this area, see for example~\cite{B03,B02,FH98,FH98csp,FHH99,FHH02}. If the conjecture holds, it would imply a fundamental distinction between CSP and the whole class NP. Indeed, the latter is known (unless P$=$NP) to contain so-called \emph{NP-intermediate} problems that are neither NP-complete nor polynomial-time solvable~\cite{L75}.
The Dichotomy Conjecture
was motivated by several earlier dichotomy theorems for special cases, such as the one of Schaefer for binary structures~\cite{S78} or the one of Hell and Ne\v{s}et\v{r}il for undirected graphs, stated as follows.
\begin{theorem}[Hell and Ne\v{s}et\v{r}il Dichotomy \cite{HN90}]\label{thm:HN}
Let $H$ be an undirected graph. If $H$ is bipartite, then \PBCOL{$H$} is polynomial-time solvable. Otherwise, \PBCOL{$H$} is NP-complete.
\end{theorem}
\medskip
\noindent\textbf{Digraph homomorphisms.} Digraph homomorphisms are also well-studied in the context of complexity dichotomies. We will relate them to tropical graph homomorphisms. For a digraph $D$, \PBCOL{$D$} asks whether an input digraph admits a homomorphism to $D$, that is, a homomorphism of the underlying undirected graphs that also preserves the orientation of the arcs.
While in the case of undirected graphs, the \PBCOL{H} problem is only polynomial time for graphs whose core is either $K_1$ or $K_2$, in the case of digraphs the problem remains polynomial time for a large class of digraphs which are cores. The classification of such cores has been one of the difficulties of the conjecture. Such classifications are given for certain interesting subclasses, see for example~\cite{BH90,BHM88,BHM95,BKN09,F01}. A proof of a conjectured classification of the general case has been announced while this paper was under review (see~\cite{FKR17}). If valid, this would imply the truth of the Dichotomy Conjecture, as Feder and Vardi~\cite{FV98} showed the following (seemingly weaker) statement to be equivalent to it.
\begin{conjecture}[Equivalent form of the Dichotomy Conjecture, Feder and Vardi~\cite{FV98}]\label{conj:digraph-dicho}
For every bipartite digraph $D$, \PBCOL{$D$} is polynomial-time solvable or NP-complete.
\end{conjecture}
In Section~\ref{sec:dicho-CSP}, similarly to its above reformulation (Conjecture~\ref{conj:digraph-dicho}), we will show that the Dichotomy Conjecture has an equivalent formulation as a dichotomy for tropical homomorphisms problems. More precisely, we will show that the Dichotomy Conjecture is true if and only if its restriction to \PBCOL{$(H,c)$} problems, where $(H,c)$ is a $2$-tropical bipartite graph, also holds. In other words, one can say that the class of $2$-tropical bipartite graph homomorphisms is as rich as the whole class of CSPs.
For many digraphs $D$ it is known such that \PBCOL{$D$} is NP-complete. Such a digraph of order~$4$ and size~$5$ is presented in the book by Hell and Ne\v{s}et\v{r}il~\cite[page 151]{HNbook}. Such oriented trees are also known, see~\cite{HNZ96} or~\cite[page 158]{HNbook}; the smallest such known tree has order~$45$. A full dichotomy is known for oriented cycles~\cite{F01}; the smallest such NP-complete oriented cycle has order between~$24$ and~$36$~\cite{D16,F01}. Using these results, one can easily exhibit some NP-complete \PBCOL{$(H,c)$} problems. To this end, given a digraph $D$, we construct the $3$-tropical graph $T(D)$ as follows. Start with the set of vertices $V(D)$ and colour its vertices Blue. For each arc $\overrightarrow{uv}$ in $D$, add a path $ux_ux_vv$ of length~$3$ from $u$ to $v$ in $T(D)$, where $x_u$ and $x_v$ are two new vertices coloured Red and Green, respectively. The following fact is not difficult to observe.
\begin{proposition}\label{prop:digraph-reduc}
For any two digraphs $D_1$ and $D_2$, we have $D_1\to D_2$ if and only if $T(D_1)\to T(D_2)$.
\end{proposition}
By the above results on NP-complete \PBCOL{$D$} problems and Proposition~\ref{prop:digraph-reduc}, we obtain a $3$-tropical graph of order~$14$, a $3$-tropical tree of order~$133$, and a $3$-tropical cycle of order between~$72$ and~$108$ whose associated homomorphism problems are NP-complete. Nevertheless, in this paper, we exhibit (by using other reduction techniques) much smaller tropical graphs, trees and cycles $(H,c)$ with \PBCOL{$(H,c)$} NP-complete.
\medskip
\noindent\textbf{List homomorphisms.} Dichotomy theorems have also been obtained for a list-based extension of the class of homomorphism problems, the \emph{list-homomorphism problems}. In this setting, introduced by Feder and Hell in~\cite{FH98}, the input consists of a pair $(G,L)$, where $G$ is a graph and $L:V(G)\to 2^{V(H)}$ is a list assignment representing a set of allowed images for each vertex of $G$. For a fixed graph $H$, the decision problem \PBlistCOL{$H$} asks whether there is a homomorphism $h$ of $G$ to $H$ such that for each vertex $v$ of $G$, $h(v)\in L(v)$. Problem \PBlistCOL{$H$} can be seen as a generalization of \PBCOL{$H$}. Indeed, restricting \PBlistCOL{$H$} to the class of inputs where for each vertex $v$ of $G$, $L(v)=V(H)$, corresponds precisely to \PBCOL{$H$}. Therefore, if \PBCOL{$H$} is NP-complete, so is \PBlistCOL{$H$}. For this set of problems, a full complexity dichotomy has been established in a series of three papers~\cite{FH98,FHH99,FHH02}. We state the dichotomy result for simple graphs from~\cite{FHH99}, that is related to our work. (A circular arc graphs is an intersection graph of arcs on a cycle.)
\begin{theorem}[Feder, Hell and Huang~\cite{FHH99}]\label{thm:listhom-table}
If $H$ is a bipartite graph such that its complement is a circular arc graph, then \PBlistCOL{$H$} is polynomial-time solvable. Otherwise, \PBlistCOL{$H$} is NP-complete.
\end{theorem}
Given a tropical graph $(H,c)$, the problem \PBCOL{$(H,c)$} is equivalent to the restriction of \PBlistCOL{$H$} to instances $(G,L)$ where each list is the set of vertices in one of the colour classes of $c$. Next, we introduce a less restricted variant of \PBlistCOL{$H$} that is also based on tropical graph homomorphisms.
\medskip
\noindent\textbf{The \PBtropCOL{\boldmath{$H$}} problem.}
Given a fixed graph $H$, we introduce the decision problem \PBtropCOL{$H$}, whose instances consist of (1) a vertex-colouring $c$ of $H$ and (2) a tropical graph $(G,c_2)$. Then, \PBtropCOL{$H$} consists of deciding whether $(G,c_1)\to (H,c)$.
Alternatively, \PBtropCOL{$H$} is an instance restriction of \PBlistCOL{$H$} to instances with \emph{laminar lists}, that is, lists such that for each pair of distinct vertices $v_1, v_2 \in V(G)$, $L(v_1) = L(v_2)$ or $L(v_1) \cap L(v_2) =\emptyset$. (We remark that \PBtropCOL{$H$}, as well as \PBlistCOL{$H$}, can also be formulated as a CSP, where certain unary relations encode the list constraints: so-called \emph{full CSPs}, see~\cite{FH98csp} for details.)
Given the difficulty of studying \PBCOL{$(H,c)$} problems, as will be demonstrated in Section~\ref{sec:dicho-CSP}, the study of \PBtropCOL{$H$} problems will be the focus of the other parts of this paper. This study is directed by the following question.
\begin{question}\label{mainquestion}
For a given graph $H$, what is the complexity of \PBtropCOL{$H$}?
\end{question}
Clearly, \PBCOL{$(H,c)$} where each vertex receives the same colour, is computationally equivalent to \PBCOL{$H$}. Therefore, by the Hell-Ne\v{s}et\v{r}il dichotomy of Theorem~\ref{thm:HN}, if $H$ is non-bipartite, \PBtropCOL{$H$} is NP-complete. Furthermore, by the above formulation of \PBtropCOL{$H$} as an instance restriction of \PBlistCOL{$H$}, whenever \PBlistCOL{$H$} is polynomial-time solvable, so is \PBtropCOL{$H$}.
Thus, according to Theorems~\ref{thm:HN} and~\ref{thm:listhom-table}, all problems \PBtropCOL{$H$} where $H$ is not bipartite are NP-complete, and all problems \PBtropCOL{$H$} where $H$ is bipartite and its complement is a circular-arc graph are polynomial-time solvable. Thus, it remains to study \PBtropCOL{$H$} when $H$ belongs to the class of bipartite graphs whose complement is not a circular-arc graph. This class of graphs has been well-studied, and characterized by forbidden induced subgraphs~\cite{TM76}. It is a rich class of graphs that includes all cycles of length at least~$6$, all trees with at least one vertex from which there are three branches of length at least~$3$, and an many other graphs~\cite{TM76}.
Observe that for any induced subgraph $H'$ of a graph $H$, one can reduce \PBtropCOL{$H'$} to \PBtropCOL{$H$} by assigning, in the input colouring of $H$, a dummy colour to all the vertices of $H-H'$. Hence, if \PBtropCOL{$H$} is polynomial-time solvable, then \PBtropCOL{$H'$} is also polynomial-time solvable. Conversely, if \PBtropCOL{$H'$} is NP-complete, so is \PBtropCOL{$H$}. Therefore, to answer Question~\ref{mainquestion}, it is enough to consider minimal graphs $H$ such that \PBtropCOL{$H$} is NP-complete.
A first question is to study the case of minimal graphs $H$ for which \PBlistCOL{$H$} is NP-complete; such a list is known and it follows from Theorem~\ref{thm:listhom-table}. In particular, it contains all even cycles of length at least~$6$. In Section~\ref{sec:list_hom}, we show that for every even cycle $C_{2k}$ of length at least~$48$, \PBtropCOL{$C_{2k}$} is NP-complete. On the other hand,for every even cycle $C_{2k}$ of length at most~$12$, \PBtropCOL{$C_{2k}$} is polynomial-time solvable. Unfortunately, for each graph $H$ in the above-mentioned list that is not a cycle, \PBtropCOL{$H$} is polynomial-time solvable, and thus larger graphs will be needed in the quest of a similar characterization of NP-complete \PBtropCOL{$H$} problems.
In Section~\ref{sec:smallgraphs}, we show that for every bipartite graph $H$ of order at most~$8$, \PBtropCOL{$H$} is polynomial-time solvable, but there is a bipartite graph $H_9$ of order~$9$ such that \PBtropCOL{$H_9$} is NP-complete.
Finally, in Section~\ref{sec:trees}, we study the case of trees. We prove that for every tree $T$ of order at most~$11$, \PBtropCOL{$T$} is polynomial-time solvable, but there is a tree $T_{23}$ of order~$23$ such that \PBtropCOL{$T_{23}$} is NP-complete.
We remark that our NP-completeness results are finer than those that can be obtained from Proposition~\ref{prop:digraph-reduc}, in the sense that the orders of the obtained target graphs are much smaller. Similarly, we note that the results in~\cite{BM97} imply the existence of NP-complete \PBtropCOL{$H$} problems, and $H$ can be chosen to be a tree or a cycle. However, similarly as in Proposition~\ref{prop:digraph-reduc}, these results are also based on reductions from NP-complete \PBCOL{$D$} problems, where $H$ is obtained from the digraph $D$ by replacing each arc by a path (its length depends on $D$, but it is always at least~$3$). Thus, the NP-complete tropical targets obtained in~\cite{BM97} are trees of order at least~$133$ and cycles of order at least~$72$, which is much more than the ones exhibited in the present paper.
\shortpaper{
\medskip
To improve the presentation, some results in this paper are given without proof. The full paper, containing all proofs, is available online~\cite{fullversion}.
}
\section{Preliminaries and tools}\label{sec:prelim}
In this section we gather some necessary preliminary definitions and results.
\subsection{Isomorphisms, cores}
For tropical graph homomorphisms, we have the same basic notions and properties as in the theory of graph homomorphisms. A homomorphism of tropical graph $(G,c_1)$ to $(H,c_2)$ is an \emph{isomorphism} if it is a bijection and it acts bijectively on the set of edges.
\begin{definition}
The \emph{core} of a tropical graph $(G,c)$ is the smallest (in terms of the order) induced tropical subgraph $(G',c_{|G'})$ admitting a homomorphism of $(G,c)$ to $(G',c_{|G'})$.
\end{definition}
In the same way as for simple graphs, it can be proved that the core of a tropical graph is unique. A tropical graph $(G,c)$ is called a \emph{core} if its core is isomorphic to $(G,c)$ itself.
Moreover, we can restrict ourselves to studying only cores. Indeed it is not difficult to check that $(G,c_1)$ admits a homomorphism to $(H,c_2)$ if and only if the core of $(G,c_1)$ admits a homomorphism to the core of $(H,c_2)$.
\subsection{Formal definitions of the used computational problems}
We now formally define all the decision problems used in this paper.
\problemdec{\PBCOL{$H$}}{A (di)graph $G$.}{Does there exist a homomorphism of $G$ to $H$?}
\smallskip
\problemdec{\PBlistCOL{$H$}}{A graph $G$ and a list function $L:V(G)\to 2^{V(H)}$.}{Is there a homomorphism $f$ of $G$ to $H$ such that for every vertex $x$ of $G$, $f(x)\in L(x)$?}
\smallskip
\problemdec{\PBCOL{$(H,c)$}}{A tropical graph $(G,c_1)$.}{Does $(G,c_1)$ admit a homomorphism to $(H,c)$?}
\smallskip
\problemdec{\PBtropCOL{$H$}}{A vertex-colouring $c$ of $H$, and a tropical graph $(G,c_1)$.}{Does $(G,c_1)$ admit a homomorphism to $(H,c)$?}
\smallskip
\problemdec{\textsc{$T$-CSP}}{A relational structure $S$ over the same vocabulary as $T$.}{Does $S$ admit a homomorphism to $T$?}
\smallskip
\problemdec{\textsc{$k$-SAT}}{A pair $(X,C)$ where $X$ is a set of Boolean variables and $C$ is a set of $k$-tuples of literals of $X$, that is, variables of $X$ or their negation.}{Is there a truth assignment $A:X\to\{0,1\}$ such that each clause of $C$ contains at least one true literal?}
\smallskip
\problemdec{\textsc{NAE $k$-SAT}}{A pair $(X,C)$ where $X$ is a set variables and $C$ is a set of $k$-tuples of variables of $X$.}{Is there a partition of $X$ into two classes such that each clause of $C$ contains at least one variable in each class?}
It is a folklore result that \textsc{$2$-SAT} is polynomial-time solvable, a fact for example observed in~\cite{K67}. On the other hand, \textsc{$3$-SAT} is NP-complete~\cite{K72}, and \textsc{NAE $3$-SAT} is NP-complete as well~\cite{M98} (even if the input formula contains no negated variables).
\subsection{Bipartite graphs}
We now give several facts that are useful when working with homomorphisms of bipartite graphs.
\begin{observation}\label{obs:bipartite-hom}
Let $H$ be a bipartite graph with parts $A,B$. If $\phi:G\to H$ is a homomorphism of $G$ to $H$, then $G$ must be bipartite. Moreover, if $G$ and $H$ are connected, then $\phi^{-1}(A)$ and $\phi^{-1}(B)$ are the two parts of $G$.
\end{observation}
The next proposition shows that for bipartite target graphs, we may assume (at the cost of doubling the number of colours) that no two vertices from two different parts of the bipartition are coloured with the same colour.
\begin{proposition}\label{prop:bipartite-hom}
Let $(H,c)$ be a connected tropical bipartite graph with parts $A,B$, and assume that vertices in $A$ and $B$ are coloured by $c$ with colours in set $C_A$ and $C_B$, respectively. Let $c'$ be the colouring with colour set $(C_A\times 0)\cup (C_B\times 1)$ obtained from $c$ with $c'(x)=(c(x),0)$ if $x\in A$ and $c'(x)=(c(x),1)$ if $x\in B$. If \PBCOL{$(H,c')$} is polynomial-time solvable, then \PBCOL{$(H,c)$} is polynomial-time solvable.
\end{proposition}
\begin{proof}
Let $(G,c_1)$ be a bipartite tropical graph. We may assume $G$ is connected since the complexity of \PBCOL{$(H,c)$} and \PBCOL{$(H,c')$} stays the same for connected inputs. Let $c_1'$ and $c_1''$ be the colourings obtained from $c_1$ by performing a similar modification as for $c'$: $c_1'(x)=(c_1(x),0)$ if $x\in A$ and $c_1'(x)=(c_1(x),1)$ if $x\in B$, and $c_1''(x)=(c_1(x),1)$ if $x\in A$ and $c_1''(x)=(c_1(x),0)$ if $x\in B$. Now it is clear, by Observation~\ref{obs:bipartite-hom}, that $(G,c_1)\to (H,c)$ if and only if either $(G,c_1')\to (H,c')$ or $(G,c_1'')\to (H,c')$. Since the latter condition can be checked in polynomial time, the proof is complete.
\end{proof}
\subsection{Generic lemmas for polynomiality}
We now prove several generic lemmas that will be useful to prove that a specific \PBCOL{$(H,c)$} problem is polynomial-time solvable
\begin{definition
Let $(H,c)$ be a tropical graph. A vertex of $(H,c)$ is a
\emph{forcing vertex} if all its neighbours are coloured with distinct
colours.
\end{definition}
This is a useful concept since in any mapping of a tropical
graph $(G,c')$ to a target containing a forcing vertex $x$, if a
vertex of $G$ is mapped to $x$, then the mapping of all its neighbours
is forced. We have the following immediate application:
\begin{lemma}\label{lemm:all-forcing}
Let $(H,c)$ be a tropical graph. If all vertices of $H$ are
forcing vertices, then \PBCOL{$(H,c)$} is polynomial-time solvable.
\end{lemma}
\begin{proof}
Choose any vertex $x$ of the instance $(G,c_1)$, and map it to any
vertex of $(H,c)$ with the same colour. Once this choice is made, the
mapping for the whole connected component of $x$ is forced. Hence, try
all $O(|V(H)|)$ possibilities to map $x$, and repeat this for every
connected component of $G$. The tropical graph $(G,c_1)$ is a YES-instance if and only if
every connected component admits a mapping.
\end{proof}
\begin{lemma}[\textsc{$2$-SAT}]\label{lemm:2SAT}
Let $(H,c)$ be a tropical graph and let $\{S_1,\ldots,S_k\}$ be a collection of independent sets of $H$, each of size at most~$2$. Assume that for every tropical graph $(G,c_1)$ admitting a homomorphism
to $(H,c)$, there exists a partition $\mathcal P=P_1,\ldots,P_\ell$ of
$V(G)$ into $\ell\leq k$ sets and a homomorphism $f:(G,c_1)\to (H,c)$
such that for every $i\in\{1,\ldots,\ell\}$, there is a
$j=j(i)\in\{1,\ldots,k\}$ such that all vertices of $P_i$ map to
vertices of $S_j$. Then \PBCOL{$(H,c)$} is polynomial-time solvable.
\end{lemma}
\begin{proof}
We reduce \PBCOL{$(H,c)$} to \textsc{$2$-SAT}. For every set $S_i$, if
$S_i$ contains only one vertex $s$, $s$ represents TRUE. If $S_i$
contains two vertices $s,s'$, one of them represents TRUE, the other
FALSE (note that if some vertex belongs to two distinct sets $S_i$ and
$S_j$, it is allowed to represent, say, FALSE with respect to $S_i$
and TRUE with respect to $S_j$). Now, given an instance $(G,c_1)$ of
\PBCOL{$(H,c)$}, we build a \textsc{$2$-SAT} formula over variable set
$V(G)$ that is satisfiable if and only if $(G,c_1)\to (H,c)$, as
follows.
For every edge $xy$ of $G$, assume that in $f$, $x$ is mapped to a
vertex of $S_i$ and $y$ is mapped to a vertex of $S_j$ (necessarily if
$(G,c_1)\to (H,c)$ we have $i\neq j$ since $S_i,S_j$ induce
independent sets). Let $F_{xy}$ be a disjunction of conjunctive
2-clauses over variables $x,y$. For every edge $uv$ between a vertex
$u$ in $S_i$ and a vertex $v$ in $S_j$, depending on the truth value
assigned to $u$ and $v$, add to $F_{xy}$ the conjunctive clause that
would be true if $x$ is assigned the truth value of $u$ and $y$ is
assigned the truth value of $v$. For example: if $u=FALSE$ and
$v=TRUE$ add the clause $(\overline{x}\wedge y)$. When $F_{xy}$ is
constructed, transform it into an equivalent conjunction of
disjunctive clauses and add it to the constructed \textsc{$2$-SAT}
formula. Now, by the construction, if the formula is satisfiable we
construct a homomorphism by mapping every vertex $x$ to the vertex of
the corresponding set $S_i$ that has been assigned the same truth
value as $x$ in the satisfying assignment. By construction it is clear
that this is a valid mapping. On the other hand, if the formula is not
satisfiable, there is no homomorphism of $(G,c_1)$ to $(H,c)$
satisfying the conditions, and hence there is no homomorphism at all.
\end{proof}
As a corollary of Lemma~\ref{lemm:2SAT} and Proposition~\ref{prop:bipartite-hom}, we obtain the following lemma:
\begin{lemma}\label{lemm:2SAT-application}
If $(H,c)$ is a bipartite tropical graph where each colour is used at most twice, then \PBCOL{$(H,c)$} is polynomial-time solvable.
\end{lemma}
Given a set $S$ of vertices, the \emph{boundary} $B(S)$ is the set of vertices in $S$ that have a neighbour out of $S$
\begin{lemma}\label{lemm:distinct-vertex-boundary}
Let $(H,c)$ be a tropical graph containing a connected subgraph
$S$ of forcing vertices such that:\\
(a) every vertex in $B(S)$ is coloured with a distinct colour (let $C(S)$ be the set of colours given to vertices in $B(S)$), and
(b) no colour of $C(S)$ is present in $V(H)\setminus S$.\\
If \PBlistCOL{$(H-S)$} is polynomial-time solvable, then \PBCOL{$(H,c)$} is polynomial-time solvable.
\end{lemma}
\begin{proof}
Let $\overline{S}=V(H)\setminus S$. Let $(G,c_1)$ be an instance of
\PBCOL{$(H,c)$}. Consider an arbitrary vertex $v$ of $G$ with
$c_1(v)=i$. Then, $v$ must be mapped to a vertex
coloured~$i$. For every possible choice of mapping $v$, we will
construct one instance of \PBlistCOL{$(H-S)$}. To construct an
instance from such a choice, we first partition $V(G)$ into two sets:
the set $V_S$ containing the vertices that must map to vertices in $S$
(and their images are determined), and the set $V_{\overline{S}}$
containing the vertices that must map to vertices of $\overline{S}$. We
now distinguish two basic cases, that will be repeatedly applied during
the construction.
\noindent\textbf{Case~1: vertex \boldmath{$v$} is mapped to a vertex in \boldmath{$S$}.} If $v$ has been mapped to a vertex $x$ of $S$,
since $x$ is a forcing vertex, the mapping of all neighbours of $v$ is
determined (anytime there is a conflict we return NO for the specific
instance under construction). We continue to propagate the forced
mapping as much as possible (i.e. as long as the forced images belong to
$S$) within a connected set of $G$ containing $v$. This yields a
connected set $C_v$ of vertices of $G$ whose mapping is determined,
and whose neighbourhood $N_v=N(C_v)\setminus C_v$ consists of vertices
each of which must be mapped to a determined vertex of
$\overline{S}$. We add $C_v$ to $V_S$. We now remove the set $C_v$
from $G$ and repeat the procedure for all vertices of $N_v$ using
Case~2.
\noindent\textbf{Case~2: vertex \boldmath{$v$} is mapped to a vertex
in \boldmath{$\overline{S}$}.} We perform a BFS search on the
remaining vertices in $G$, until we have computed a maximal connected
set $C_v$ of vertices containing $v$ in which no vertex is coloured
with a colour in $C(S)$. Then, for every vertex $x$ of $C_v$ with a
neighbour $y$ that is coloured~$i$ ($i\in C(S)$), by Property~(a) we
know that $y$ must be mapped to a vertex in $B(S)$, and moreover the
image of $y$ is determined by colour~$i$. Hence the neighbourhood
$N_v=N(C_v)\setminus C_v$ has only vertices whose mapping is
determined. We add $C_v$ to set $V_{\overline{S}}$ and apply Case~1 to
every vertex in $N_v$.
\noindent\textbf{End of the procedure.} Once $V(G)$ has been
partitioned into $V_S$ and $V_{\overline{S}}$ (where the mapping of all
vertices in $V_S\cup N(V_S)$ is fixed), we
can reduce this instance to a corresponding instance of
\PBlistCOL{$(H-S)$}.
In total, $(G,c_1)$ is a YES-instance if and only if at least one of the
$O(|V(G)|)$ constructed instances of \PBlistCOL{$(H-S)$} is a YES-instance.
\end{proof}
The next lemma is similar to Lemma~\ref{lemm:distinct-vertex-boundary}
but now the boundary is distinguished using edges.
\begin{lemma}\label{lemm:distinct-edge-boundary}
Let $(H,c)$ be a tropical graph containing a connected subgraph
$S$ of forcing vertices with boundary $B=B(S)$ and $N=N(B)\setminus S$. Assume that the following properties hold:\\
(a) for every pairs $xy$, $x'y'$ of distinct edges of $B\times N$, we have $(c(x),c(y))\neq (c(x'),c(y'))$, and\\
(b) for every edge $xy$ of $B\times N$, there is no edge in $(H-S)\times (H-S)$ whose endpoints are coloured $c(x)$ and $c(y)$.
If \PBlistCOL{$(H-S)$} is polynomial-time solvable, then \PBCOL{$(H,c)$} is
polynomial-time solvable.
\end{lemma}
\begin{proof}
The proof is almost the same as the one of
Lemma~\ref{lemm:distinct-vertex-boundary}, except that now, while
computing an instance of \PBlistCOL{$(H-S)$}, the distinction between
$V_S$ and $V_{\overline{S}}$ is determined by the edges of $B\times
N$.
\end{proof}
The next lemma identify some unique features of a tropical graph to simplify the problem into a list-homomorphism problem.
\begin{definition
A \emph{Unique Tropical Feature} in a tropical graph $(H,c)$ is a vertex or an edge of $H$ that satisfies one of the following conditions.
\begin{itemize}
\item[Type 1.] A vertex $u$ of $H$ whose colour class is $\{u\}$.
\item[Type 2.] An edge $uv$ of $H$ such that there is no other edge in $H$ whose vertices are coloured $c(u)$ and $c(v)$, respectively.
\item[Type 3.] A vertex $u$ of $H$ such that $N(u)$ is monochromatic in $(H,c)$ with colour $s$, and every vertex coloured $s$ that does not belong to $N(u)$ has no neighbour coloured with $c(u)$.
\item[Type 4.] A forcing vertex $u$ of $H$ such that for each pair $v,w$ of distinct vertices in $N(u)$, there is no path $v'u'w'$ in $H-u$ with $c(v) = c(v')$, $c(u)=c(u')$ and $c(w)=c(w')$.
\end{itemize}
\end{definition}
\begin{definition}
Let $(H,c)$ be a tropical graph and $S$ a set of Unique Tropical Features of $(H,c)$. $S$ is partitioned into four sets as $S=S_1 \cup S_2 \cup S_3 \cup S_4$, where $S_i$ is the set of unique tropical features of type $i$ in $S$. We define $H(S)$ as follows : $V(H(S)) = (V(H)\cup \{u_v| u\in S_4, v\in N(u) \})\setminus (S_1 \cup S_3 \cup S_4)$ and $E(H(S)) = (E(H[V(H(S))])\setminus S_2)\cup \{u_vv| u\in S_4, v\in N(u)\}$.
\end{definition}
In other words, $H(S)$ is the graph obtained from $H$ by removing unique tropical features of type $1$, $2$, and $3$, and for each unique tropical feature $u$ of type $4$, replacing $N[u]$ by $d(u)$ pending edges.
\begin{lemma}\label{lemm:unique-feature}
Let $(H,c)$ be a tropical graph and $S$ a set of unique tropical features of $(H,c)$. If \PBlistCOL{$(H(S))$} is polynomial-time solvable, then \PBCOL{$(H,c)$} is polynomial-time solvable.
\end{lemma}
\begin{proof}
Let $(G,c')$ be an instance of \PBCOL{$(H,c)$}. We are going to construct a graph $G'$ and associate to each vertex of $G'$ a list of vertices of $H(S)$ such that there is a list-homomorphism from $G'$ to $H(S)$ (with respect to these lists) if and only if there is a tropical homomorphism of $(G,c')$ to $(H,c)$. We proceed with sequential modifications, by considering the unique tropical features of $S$ one by one.
First, we can see the instance $(G,c')$ of \PBCOL{$(H,c)$} as an instance of \PBlistCOL{$H$} by giving to each vertex $u$ in $G$ the list $L(u)$ of vertex in $H$ coloured $c'(u)$. If at any point in the following, we update the list of a vertex to be empty, we can conclude that there is no tropical homomorphism between $(G,c')$ and $(H,c)$.
For each unique tropical feature $u$ of type $1$ in $S$, there is a colour $s$ such that only the vertex $u$ is coloured $s$ in $(H,c)$. Every vertex in $(G,c')$ coloured $s$ must be mapped to $u$ and has a list of size at most one. For each vertex $v$ in $(G,c')$ coloured $s$, we update the list of each of its neighbours $w$ such that $L(w)$ becomes $L(w)\cap N(u)$. We can then delete $v$ from $(G,c')$ and forget $L(v)$ without affecting the existence of a list-homomorphism. Indeed, if a homomorphism exists, then it must map each neighbour of $v$ to a neighbour of $u$. Moreover, there is no other vertex of $(G,c')$ that can be mapped to $u$.
For each unique tropical feature $uv$ of type $2$ in $S$, there is no other edge than $uv$ in $H$ such that the colour of its vertices are $c(u)$ and $c(v)$. Every edge in $(G,c')$ whose vertices are coloured $c(u)$ and $c(v)$ must be mapped to $uv$. For each edge $xy$ in $(G,c')$ such that $c'(x)= c(u)$ and $c'(y)=c(v)$, we update the list of $x$ and $y$ such that $L(x)$ becomes $L(x)\cap \{u\}$ and $L(y)$ becomes $L(y)\cap \{v\}$. We can then delete the edge $uv$ from $(G,c')$ without changing the existence of a list-homomorphism. Indeed, if a homomorphism exists, it must map $x$ to $u$ and $y$ to $v$. Again, there is no other edge of $(G,c')$ that can be mapped to $uv$.
For each unique tropical feature $u$ of type $3$ in $S$, $N(u)$ is monochromatic in $(H,c)$ of colour $s$ and any vertex coloured $s$ with a neighour coloured $c(u)$ must belong to $N(u)$. Let $v$ be a vertex of $G$ such that $c(v)=c(u)$ and $N(v)$ is monochromatic in $(G,c')$ of colour $s$. Then, we can assume that $v$ is mapped to $u$. Indeed, in every tropical homomorphism of $(G,c')$ to $(H,c)$, if $v$ is not mapped to $u$, it is mapped to a vertex at distance~$2$ from $u$, and one obtains another valid tropical homomorphism by only changing the mapping of $v$ to $u$. For each such vertex $v$, we update the list of its neighbours $w$ such that $L(w)$ becomes $L(w)\cap N(u)$. We can then delete $v$ from $(G,c')$ without affecting the existence of a list-homomorphism. Indeed, if a homomorphism exists, it maps every neighbour of $v$ to a neighbour of $u$. Moreover, there no other vertex of $(G,c')$ can be mapped to $u$.
Finally, let $u$ be a vertex of type $4$ in $S$. Thus, by the definition of type $4$, for each $v, w \in N(u)$, there is no other path $v'u'w'$ in $H$ such that $c(v)=c(v')$, $c(u)=c(u')$ and $c(w)=c(w')$. Furthermore, since $u$ is a forcing vertex, we have $c(v)\neq c(w)$ for any two neighbours $v$ and $w$ of $u$.
Let $x$ be a vertex of $G$ such that $c'(x)=c(u)$ and such that at least two neighbours of $x$ are of colours $c(v)$ or $c(w)$, one of each. Then, as $x$ is of type $4$, any homomorphism of $(G, c')$ to $(H,c)$ must map all such vertices $x$ to $u$. Remove all such vertices from $G$ and let $(G', c')$ be the remaining tropical graph. For any vertex $y$ of $G'$ if it is of colour $c(u)$, it may then either map to another vertex of this colour, or all its neighbours must map a same neighbour of $u$. Let $(H_1, c)$ be a tropical graph obtained from $(H, c)$ by removing the vertex $u$, and then adding one new vertex for each vertex in $N_H(u)$ and assigning the colour $c(u)$ to it. It follows that $(G',c')$ admits a homomorphism to $(H',c)$ if and only
$(G,c')$ admits a homomorphism to $(H,c)$, proving our claim.
In conclusion, we have built an instance $(G', L)$ of \PBlistCOL{$H(S)$} that maps to $H(S)$ if and only if $(G,c')$ maps to $(H,c)$, thus proving our claim.
We remark, furthermore, that these changes used to introduced $(G', L)$ and $H(S)$ are compatible even between different types of vertices, thus we may allow $S$ to contain a combination of such vertices. However, in this work we will only consider sets $S$ whose elements are all of a same type.
\end{proof}
\section{\PBCOL{$(H,c)$} and the Dichotomy Conjecture}\label{sec:dicho-CSP}
Since each \PBCOL{$(H,c)$} problem is a CSP, the Feder--Vardi Dichotomy Conjecture (Conjecture~\ref{conj:CSP-dicho}) would imply a complexity dichotomy for the class of \PBCOL{$(H,c)$} problems. As we mentioned before a proof of the conjecture has been recently announced, thus every \PBCOL{$(H,c)$} is either polynomial time solvable or it is an NP-complete problem. Here we point out that an independent proof even on a very restricted set of $(H,c)$ would also prove the original conjecture.
Following the construction of Feder and Vardi (\cite[Theorem~10]{FV98}) and based on its exposition in the book by Hell and Ne\v{s}et\v{r}il~\cite[Theorem~5.14]{HNbook}, one can modify their gadgets to prove a similar statement for the class of $2$-tropical bipartite graph homomorphism problems.
\shortpaper{The proof being very similar to the proofs in~\cite{BFHN15,FV98,HNbook}, we skip it here and refer to our manuscript~\cite{fullversion} instead.
}
\begin{theorem}\label{thm:CSP}
For each CSP template $T$ there is a $2$-coloured graph $(H,c)$ such that \PBCOL{$(H,c)$}
and \textsc{$T$-CSP} are polynomially equivalent. Moreover, $(H,c)$ can be chosen to be bipartite
and homomorphic to a $2$-coloured forcing path.
\end{theorem}
\longpaper{
\begin{proof}
We follow the proof of Theorem~5.14 in the book~\cite{HNbook} proving a similar statement for digraph
homomorphism problems. The structure of the proof in~\cite{HNbook} is as follows. First, one shows that
for each CSP template $T$, there is a bipartite graph $H$ such that the \textsc{$T$-CSP} problem and
the \textsc{$H$-Retraction} problem are polynomially equivalent. Next, it is shown that for each
bipartite graph $H$ there is a digraph $H'$ such that \textsc{$H$-Retraction} and
\textsc{$H'$-Retraction} are polynomially equivalent. Finally it is observed that $H'$ is a core and
thus \textsc{$H'$-Retraction} and \PBCOL{$(H',c)$} are polynomially equivalent.
We adapt this proof to the case of $2$-tropical graph homomorphism problems.
The construction of $H'$ from $H$ in~\cite{HNbook} is through the use of so-called zig-zag paths.
In our case, we replace these zig-zag paths by specific $2$-coloured graphs that play the same role. This will allow us to construct
a $2$-coloured graph $H'$ from a bipartite graph $H$ such that \textsc{$H$-Retraction} and
\textsc{$H'$-Retraction} are polynomially equivalent. Our paths will have black vertices denoted by
$B$ and white vertices denoted by $W$. Hence the path $WB^4W^4B$ consists of one white vertex, four black vertices,
four white vertices and a black vertex. The maximal monochromatic subpaths are called \emph{runs}.
Thus the above path is the concatenation of four runs: the first and last of length~$1$, the middle two of length~$4$.
Given an odd integer $\ell$, we construct a path $P$ consisting of $\ell$ runs.
The first and the last run each consist of a single white vertex. The interior runs are
of length four. We denote that last (rightmost) vertex of $P$ by $0$. From $P$ we
construct $\ell-2$ paths $P_1, \dots, P_{\ell-2}$. Path $P_i$ ($i=1, 2, \dots, \ell-2$)
is obtained from $P$ by replacing the $i^{th}$ run of length four with a run of length~$2$.
We denote the rightmost vertex of $P_i$ by $i$.
Similarly, for an even integer $k$, we construct a second family of paths $Q$ and
$Q_j$, ($j=1, 2, \dots, k-2$). The leftmost vertex of $Q$ is $1$ and
the leftmost vertex of $Q_j$ is $j$. The paths are described below:
$$
\begin{array}{rclcrcl}
P & := & W\underbrace{B^4 W^4 \cdots W^4 B^4}_{\ell-2}W & \hspace{0.5cm} &
Q & := & W\underbrace{B^4 W^4 \cdots B^4 W^4}_{k-2}B \\
P_i & := & W\underbrace{B^4 \cdots W^4}_{i-1}B^2\underbrace{W^4 \cdots B^4}_{\ell-i-2}W \hspace{1em}\mbox{($i$ odd)} & &
Q_j & := & W\underbrace{B^4 \cdots W^4}_{j-1}B^2\underbrace{W^4 \cdots W^4}_{k-j-2}B \hspace{1em}\mbox{($j$ odd)} \\
P_i & := & W\underbrace{B^4 \cdots B^4}_{i-1}W^2\underbrace{B^4 \cdots B^4}_{\ell-i-2}W \hspace{1em}\mbox{($i$ even)} & &
Q_j & := & W\underbrace{B^4 \cdots B^4}_{j-1}W^2\underbrace{B^4 \cdots W^4}_{k-j-2}B \hspace{1em}\mbox{($j$ even)} \\
\end{array}
$$
We observe the following (c.f. page 156 of~\cite{HNbook}):
\begin{enumerate}
\item The paths $P$ and $P_i$ ($i=1,2,\dots \ell-2$) each admit a homomorphism \emph{onto} a \emph{$2$-colour forcing path} of length~$2\ell - 1$, (that is, a path consisting of one run of length~$1$, $\ell - 2$ runs each of length~$2$ and a final run of length~$1$: $WBBWWB\cdots W$).
\item The paths $Q$ and $Q_j$ ($j=1,2,\dots k-2$) each admit a homomorphism onto a $2$-colour forcing path of length~$2k-1$.
\item $P_i \to P_{i'}$ implies $i=i'$.
\item $Q_j \to Q_{j'}$ implies $j=j'$.
\item $P \to P_i$ for all $i$.
\item $Q \to Q_j$ for all $j$.
\item if $X$ is a $2$-tropical graph and $x$ is a vertex of $X$ such that $f: X \to P_i$ and $f': X \to P_{i'}$ for $i\neq i'$ with $f(x) = i$ and $f'(x) = i'$, then there is a homomorphism $F:X \to P$ with $F(x)=0$.
\item if $Y$ is a $2$-tropical graph and $y$ is a vertex of $Y$ such that $f: Y \to Q_j$ and $f': Y \to Q_{j'}$ for $j \neq j'$ with $f(y) = j$ and $f'(y) = j'$, then there is a homomorphism $F:Y \to Q$ with $F(y)=1$.
\end{enumerate}
We note that $2$-colour forcing paths in $2$-tropical graphs can be used to define \emph{height} analogously to height in directed acyclic graphs. More precisely, suppose $G$ is a connected $2$-tropical graph that admits a homomorphism onto a $2$-colour forcing path, say $FP$, of even length. Let the vertices of $FP$ be $h_0, h_1, \dots, h_{2t}$. Observe that each vertex in the path has at most one white neighbour and at most one black neighbour. Thus once a single vertex $u$ in $G$ is mapped to $FP$, the image of each neighbour of $u$ is uniquely determined and by connectivity, the image of all vertices is uniquely determined. In particular, as $G$ maps \emph{onto} $FP$, there is exactly one homomorphism of $G$ to the path. (More precisely, if the path has length congruent to 0 modulo 4, there is an automorphism that reverses the path. In this case there are two homomorphisms that are equivalent up to the reversing.) We then observe that if $g:G \stackrel{onto}{\to} FP$, $h:H \to FP$, and $f: G \to H$, then for all vertices $u \in V(G)$, $g(u) = h(f(u))$. This allows us to define the height of $u \in V(G)$ to be $h_i$ when $g(u)=h_i$. Specifically, vertices at height $h_i$ in $G$ must map to vertices at height $h_i$ in $H$.
For each problem $T$ in CSP there is a bipartite graph $H$ such that \textsc{$T$-CSP} and \textsc{$H$-Retraction} are equivalent~\cite{FV98,HNbook}. Let $H$ be a bipartite graph with parts $(A,B)$, with $A=\{a_1,\dots,a_{|A|}\}$ and $B=\{b_1,\dots,b_{|B|}\}$. Let $\ell$ (respectively $k$) be the smallest odd (respectively even) integer greater than or equal to $|A|$ (respectively $|B|$). To each vertex $a_i \in A$ attach a copy of $P_i$ identifying $i$ in $P_i$ with $a_i$ in $A$. Colour all original vertex of $H$ white. To each vertex $b_j \in B$ attach a copy of $Q_j$ identifying $j$ in $Q_j$ with $b_j$ in $B$. Call the resulting $2$-tropical graph $(H',c)$. See Figure~\ref{fig:rbtarget} for an illustration.
Let $G$ be an instance of \textsc{$H$-Retraction}. In particular, we may assume without loss of generality that $H$ is a subgraph of $G$, $G$ is connected, and $G$ is bipartite. We colour the original vertices of $G$ white. Let $(A',B')$ be the partite classes of $G$ where $A \subseteq A'$ and $B \subseteq B'$. To each vertex $v$ of $A' \backslash A$, we attach a copy of $P$, identifying $v$ and $0$. To the vertices of $A \cup B$, we attach paths $P_i$ and $Q_j$ as described above to create a copy of $H'$. Call the resulting $2$-tropical graph $(G',c')$. In particular, note that $(G',c')$ and $(H',c')$ both map onto a $2$-colour forcing path of length $2\ell+2k-1$. The (original) vertices of $G$ and $H$ are at height $2\ell-1$ and $2\ell$ for colour classes $A$ and $B$ respectively. In particular, by the eight above properties, under any homomorphism $f:G' \to H'$ the restriction of $f$ to $G$ must map onto $H$ with vertices in $A'$ mapping to $A$ and vertices in $B'$ mapping to $B$.
Using the eight properties of the paths above and following the proof of Theorem~5.14 in~\cite{HNbook}, we conclude that $G$ is a YES instance of \textsc{$H$-Retraction} if and only if $(G',c')$ is a YES instance of \textsc{$(H',c)$-Retraction}.
On the other hand, let $(G',c')$ be an instance of \textsc{$(H',c)$-Retraction}. We sketch the proof from~\cite{HNbook}. We observe that $(G',c')$ must map to a $2$-colour forcing path of length $2\ell+2k-1$. The two levels of $G'$ corresponding to $H$ induce a bipartite graph (with white vertices) which we call $G$. The components of $G'-E(G)$ fall into two types: those which map to lower levels and those that map to higher levels than $G$. Let $C_t$ be a component that maps to a lower level. After required identifications we may assume $C_t$ contains only one vertex from $G$ (say $v$) and $C_t$ must map to some $P_i$. If $P_i$ is the unique $P_i$ path to which $C_t$ maps, then we modify $G'$ by identifying $v$ and $i$. Otherwise, $C_t$ maps to two paths and (by the properties~$5$--$8$) hence to all paths. The resulting graph $(G',c')$ retracts to $(H',c)$ if and only if $G$ retracts to $H$.
\end{proof}
\tikzset{
bigblue/.style={circle, draw=blue!80,fill=blue!40,thick, inner sep=1.5pt, minimum size=5mm},
bigred/.style={circle, draw=red!80,fill=red!40,thick, inner sep=1.5pt, minimum size=5mm},
bigblack/.style={circle, draw=black!100,fill=black!40,thick, inner sep=1.5pt, minimum size=5mm},
bluevertex/.style={circle, draw=blue!100,fill=blue!100,thick, inner sep=0pt, minimum size=2mm},
redvertex/.style={circle, draw=red!100,fill=red!100,thick, inner sep=0pt, minimum size=2mm},
blackvertex/.style={circle, draw=black!100,fill=black!100,thick, inner sep=0pt, minimum size=2mm},
whitevertex/.style={circle, draw=black!100,fill=white!100,thick, inner sep=0pt, minimum size=2mm},
smallblack/.style={circle, draw=black!100,fill=black!100,thick, inner sep=0pt, minimum size=1mm},
smallwhite/.style={circle, draw=black!100,fill=white!100,thick, inner sep=0pt, minimum size=1mm},
}
\begin{figure}[ht!]
\begin{center}
\begin{tikzpicture}
\node[smallwhite] (p1) at (0,4) {};
\node[smallblack] (p2) at (0.66,4) {};
\node[smallblack] (p3) at (1.32,4) {};
\node[smallblack] (p4) at (0.66,4.5) {};
\node[smallblack] (p5) at (1.32,4.5) {};
\draw[black] (p1)--(p2)--(p3)--(p4)--(p5)--(1.485,4.5);
\node at (1.65,4.25) {$\cdots$};
\node[smallblack] (p6) at (1.98,4) {};
\node[smallblack] (p7) at (2.64,4) {};
\node[smallblack] (p8) at (1.98,4.5) {};
\node[smallblack] (p9) at (2.64,4.5) {};
\node[smallwhite] (p10) at (3.3,4.5) {};
\node[smallwhite] (p11) at (3.96,4.5) {};
\node[smallwhite] (p12) at (3.3,4) {};
\node[smallwhite] (p13) at (3.96,4) {};
\node[smallblack] (p14) at (4.62,4) {};
\node[smallblack] (p15) at (5.28,4) {};
\node[whitevertex] (pa1) at (6,4) {};
\draw[black] (1.815,4)--(p6)--(p7)--(p8)--(p9)--(p10)--(p11)--(p12)--(p13)--(p14)--(p15)--(pa1);
\begin{scope}[yshift=-1.5cm]
\node[smallwhite] (p1) at (0,4) {};
\node[smallblack] (p2) at (0.66,4) {};
\node[smallblack] (p3) at (1.32,4) {};
\node[smallblack] (p4) at (0.66,4.5) {};
\node[smallblack] (p5) at (1.32,4.5) {};
\draw[black] (p1)--(p2)--(p3)--(p4)--(p5)--(1.485,4.5);
\node at (1.65,4.25) {$\cdots$};
\node[smallblack] (p6) at (1.98,4) {};
\node[smallblack] (p7) at (2.64,4) {};
\node[smallblack] (p8) at (1.98,4.5) {};
\node[smallblack] (p9) at (2.64,4.5) {};
\node[smallwhite] (p10) at (3.3,4.5) {};
\node[smallwhite] (p11) at (3.96,4.5) {};
\node[smallblack] (p12) at (4.62,4.5) {};
\node[smallblack] (p13) at (5.28,4.5) {};
\node[smallblack] (p14) at (4.62,4) {};
\node[smallblack] (p15) at (5.28,4) {};
\node[whitevertex] (pa2) at (6,4) {};
\draw[black] (1.815,4)--(p6)--(p7)--(p8)--(p9)--(p10)--(p11)--(p12)--(p13)--(p14)--(p15)--(pa2);
\end{scope}
\begin{scope}[yshift=-3cm]
\node[smallwhite] (p1) at (0,4) {};
\node[smallblack] (p2) at (0.66,4) {};
\node[smallblack] (p3) at (1.32,4) {};
\node[smallblack] (p4) at (0.66,4.5) {};
\node[smallblack] (p5) at (1.32,4.5) {};
\draw[black] (p1)--(p2)--(p3)--(p4)--(p5)--(1.485,4.5);
\node at (1.65,4.25) {$\cdots$};
\node[smallblack] (p6) at (1.98,4) {};
\node[smallblack] (p7) at (2.64,4) {};
\node[smallwhite] (p8) at (3.3,4) {};
\node[smallwhite] (p9) at (3.96,4) {};
\node[smallwhite] (p10) at (3.3,4.5) {};
\node[smallwhite] (p11) at (3.96,4.5) {};
\node[smallblack] (p12) at (4.62,4.5) {};
\node[smallblack] (p13) at (5.28,4.5) {};
\node[smallblack] (p14) at (4.62,4) {};
\node[smallblack] (p15) at (5.28,4) {};
\node[whitevertex] (pa3) at (6,4) {};
\draw[black] (1.815,4)--(p6)--(p7)--(p8)--(p9)--(p10)--(p11)--(p12)--(p13)--(p14)--(p15)--(pa3);
\end{scope}
\begin{scope}[yshift=-0.75cm]
\node[whitevertex] (qb1) at (7,4) {};
\node[smallblack] (q11) at (7.66,4) {};
\node[smallblack] (q12) at (8.32,4) {};
\node[smallblack] (q13) at (7.66,4.5) {};
\node[smallblack] (q14) at (8.32,4.5) {};
\node[smallwhite] (q15) at (8.98,4.5) {};
\node[smallwhite] (q16) at (9.64,4.5) {};
\node[smallwhite] (q17) at (8.98,4) {};
\node[smallwhite] (q18) at (9.64,4) {};
\node[smallblack] (q19) at (10.3,4) {};
\node[smallblack] (q20) at (10.86,4) {};
\draw[black] (qb1)--(q11)--(q12)--(q13)--(q14)--(q15)--(q16)--(q17)--(q18)--(q19)--(q20)--(11.125,4);
\node at (11.29,4.25) {$\cdots$};
\node[smallwhite] (q21) at (11.62,4) {};
\node[smallwhite] (q22) at (12.28,4) {};
\node[smallwhite] (q23) at (11.62,4.5) {};
\node[smallwhite] (q24) at (12.28,4.5) {};
\node[smallblack] (q25) at (12.94,4.5) {};
\draw[black] (11.455,4)--(q21)--(q22)--(q23)--(q24)--(q25);
\end{scope}
\begin{scope}[yshift=-2.25cm]
\node[whitevertex] (qb2) at (7,4) {};
\node[smallblack] (q11) at (7.66,4) {};
\node[smallblack] (q12) at (8.32,4) {};
\node[smallblack] (q13) at (7.66,4.5) {};
\node[smallblack] (q14) at (8.32,4.5) {};
\node[smallwhite] (q15) at (8.98,4.5) {};
\node[smallwhite] (q16) at (9.64,4.5) {};
\node[smallblack] (q17) at (10.3,4.5) {};
\node[smallblack] (q18) at (10.86,4.5) {};
\node[smallblack] (q19) at (10.3,4) {};
\node[smallblack] (q20) at (10.86,4) {};
\draw[black] (qb2)--(q11)--(q12)--(q13)--(q14)--(q15)--(q16)--(q17)--(q18)--(q19)--(q20)--(11.125,4);
\node at (11.29,4.25) {$\cdots$};
\node[smallwhite] (q21) at (11.62,4) {};
\node[smallwhite] (q22) at (12.28,4) {};
\node[smallwhite] (q23) at (11.62,4.5) {};
\node[smallwhite] (q24) at (12.28,4.5) {};
\node[smallblack] (q25) at (12.94,4.5) {};
\draw[black] (11.455,4)--(q21)--(q22)--(q23)--(q24)--(q25);
\end{scope}
\draw[black] (pa1)--(qb1)--(pa2)--(qb2)--(pa3) (pa3)--(qb1);
\node at (6.5,4.5) {$H$};
\node at (6,3.25) {$\vdots$};
\node at (6,2) {$\vdots$};
\node at (7,2.5) {$\vdots$};
\end{tikzpicture}
\end{center}
\caption{Construction of a $2$-tropical target $H'$ from a \textsc{$H$-Retraction} problem.}
\label{fig:rbtarget}
\end{figure}
\section{Minimal graphs $H$ for NP-complete \PBlistCOL{$H$}}\label{sec:list_hom}
Recall the dichotomy theorem for list homomorphism problems of Feder, Hell and Huang (Theorem~\ref{thm:listhom-table}): \PBlistCOL{$H$} is polynomial-time solvable if $H$ is bipartite and its complement is a circular arc graph, otherwise NP-complete. Alternatively, the latter class of graphs was characterized by Trotter and Moore~\cite{TM76} in terms of seven families of forbidden induced subgraphs: six infinite ones and a finite one.
\shortpaper{One of these families is the family of even cycles of length at least~$6$. The only tree in the list of forbidden graphs, which is called $G_1$ in~\cite{FHH99}, is a claw where each edge is subdivided twice.}
\longpaper{See their descriptions in Table~\ref{table}, as reproduced from~\cite{FHH99}. To concisely describe these seven families, they employ the following notation: Let $\mathcal F = \{S_i : 1 \leq i \leq k\}$ be a family of subsets of $\{1,2,\ldots,\ell\}$. Define $H_F$ to be the bipartite graph $(X,Y)$ with $X = \{x_1,x_2,\ldots,x_\ell\}$ and $Y =\{y_1,y_2,\ldots,y_k\}$ such that $x_iy_j$ is an edge if and only if $i\in S_j$. The families $\mathcal C$, $\mathcal T$, $\mathcal W$, $\mathcal D$, $\mathcal M$, $\mathcal N$ and $\mathcal G$ in Table~\ref{table} are defined in this way. Note that the graph $C_i$ in $\mathcal C$ is the cycle of length~$i$. See Figure~\ref{fig:table} for an illustration of the other families from Table~\ref{table}. Also note that $G_1$, which is a claw where each edge is subdivided twice, is the only tree in the table.
Given the above characterization, we can reformulate Theorem~\ref{thm:listhom-table} as follows.
\begin{theorem}[Restatement of Theorem~\ref{thm:listhom-table}, Feder, Hell and Huang~\cite{FHH99}
If $H$ contains one of the graphs defined in Table~\ref{table} as an induced subgraph, then \PBlistCOL{$H$} is NP-complete. Otherwise, \PBlistCOL{$H$} is polynomial-time solvable.
\end{theorem}
\begin{table}[ht!]
\begin{tabular}{l}
\hline
$C_6 = \{\{1, 2\}, \{2, 3\}, \{3, 1\}\}$ \\
$C_8 = \{\{1, 2\}, \{2, 3\}, \{3, 4\}, \{4, 1\}\}$ \\
$C_{10} = \{\{1, 2\}, \{2, 3\}, \{3, 4\}, \{4, 5\}, \{5, 1\}\}$\\
\ldots\\
$T_1 = \{\{1, 2\}, \{2, 3\}, \{3, 4\}, \{2, 3, 5\}, \{5\}\}$\\
$T_2 = \{\{1, 2\}, \{2, 3\}, \{3, 4\}, \{4, 5\}, \{2, 3, 4, 6\}\{6\}\}$\\
$T_3 = \{\{1, 2\}, \{2, 3\}, \{3, 4\}, \{4, 5\}, \{5, 6\}, \{2, 3, 4, 5, 7\}, \{7\}\}$\\
\ldots\\
$W_1 = \{\{1, 2\}, \{2, 3\}, \{1, 2, 4\}, \{2, 3, 4\}, \{4\}\}$\\
$W_2 = \{\{1, 2\}, \{2, 3\}, \{3, 4\}, \{1, 2, 3, 5\}, \{2, 3, 4, 5\}, \{5\}\}$\\
$W_3 = \{\{1, 2\}, \{2, 3\}, \{3, 4\}, \{4, 5\}, \{1, 2, 3, 4, 6\}, \{2, 3, 4, 5, 6\}, \{6\}\}$\\
\ldots\\
$D_1 = \{\{1, 2, 5\}, \{2, 3, 5\}, \{3\}, \{4, 5\}, \{2, 3, 4, 5\}\}$\\
$D_2 = \{\{1, 2, 6\}, \{2, 3, 6\}, \{3, 4, 6\}, \{4\}, \{5, 6\}, \{2, 3, 4, 5, 6\}\}$\\
$D_3 = \{\{1, 2, 7\}, \{2, 3, 7\}, \{3, 4, 7\}, \{4, 5, 7\}, \{5\}, \{6, 7\}, \{2, 3, 4, 5, 6, 7\}\}$\\
\ldots\\
$M_1 = \{\{1, 2, 3, 4, 5\}, \{1, 2, 3\}, \{1\}, \{1, 2, 4, 6\}, \{2, 4\}, \{2, 5\}\}$\\
$M_2 = \{\{1, 2, 3, 4, 5, 6, 7\}, \{1, 2, 3, 4, 5\}, \{1, 2, 3\}, \{1\}, \{1, 2, 3, 4, 6, 8\}, \{1, 2, 4, 6\}, \{2, 4\}, \{2, 7\}\}$\\
$M_3 = \{\{1, 2, 3, 4, 5, 6, 7, 8, 9\}, \{1, 2, 3, 4, 5, 6, 7\}, \{1, 2, 3, 4, 5\}, \{1, 2, 3\}, \{1\}, \{1, 2, 3, 4, 5, 6, 8, 10\},
$\\
\hspace{240pt}$\{1, 2, 3, 4, 6, 8\}, \{1, 2, 4, 6\}, \{2, 4\}, \{2, 9\}\}$\\
\ldots\\
$N_1 = \{\{1, 2, 3\}, \{1\}, \{1, 2, 4, 6\}, \{2, 4\}, \{2, 5\}, \{6\}\}$\\
$N_2 = \{\{1, 2, 3, 4, 5\}, \{1, 2, 3\}, \{1\}, \{1, 2, 3, 4, 6, 8\}, \{1, 2, 4, 6\}, \{2, 4\}, \{2, 7\}, \{8\}\}$\\
$N_3 = \{\{1, 2, 3, 4, 5, 6, 7\}, \{1, 2, 3, 4, 5\}, \{1, 2, 3\}, \{1\}, \{1, 2, 3, 4, 5, 6, 8, 10\}, \{1, 2, 3, 4, 6, 8\},$\\
\hspace{250pt}$\{1, 2, 4, 6\}, \{2, 4\}, \{2, 9\}, \{10\}\}$\\
\ldots\\
$G_1 = \{\{1, 3, 5\}, \{1, 2\}, \{3, 4\}, \{5, 6\}\}$\\
$G_2 = \{\{1\}, \{1, 2, 3, 4\}, \{2, 4, 5\}, \{2, 3, 6\}\}$\\
$G_3 = \{\{1, 2\}, \{3, 4\}, \{5\}, \{1, 2, 3\}, \{1, 3, 5\}\}$\\
\hline
\end{tabular}
\caption{Six infinite families $\mathcal C$, $\mathcal T$, $\mathcal W$, $\mathcal D$, $\mathcal M$, $\mathcal N$ and family $\mathcal G$ of size~$3$ of forbidden induced subgraphs for polynomial-time \PBlistCOL{$H$} problems.}
\label{table}
\end{table}
\begin{figure}[ht!]
\centering
\subfigure[$T_i$]{
\scalebox{0.85}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=0.7]
\node[draw,shape=circle,fill](x1) at (2.5,1.5) {};
\path (x1)+(0.5,0) node {$x_1$};
\node[draw,shape=circle,fill](y1) at (2.5,0) {};
\path (y1)+(0,-0.5) node {$y_1$};
\node[draw,shape=circle,fill](x2) at (4,0) {};
\path (x2)+(-0.25,-0.5) node {$x_2$};
\node[draw,shape=circle,fill](y2) at (4.75,-2) {};
\path (y2)+(0,-0.5) node {$y_2$};
\node[draw,shape=circle,fill](x3) at (5.5,0) {};
\path (x3)+(-0.5,0) node {$x_3$};
\node[draw,shape=circle,fill](xi+1) at (7.5,0) {};
\path (xi+1)+(0.7,0) node {$x_{i+1}$};
\node[draw,shape=circle,fill](yi+1) at (8.25,-2) {};
\path (yi+1)+(0,-0.5) node {$y_{i+1}$};
\node[draw,shape=circle,fill](xi+2) at (9,0) {};
\path (xi+2)+(0.5,-0.5) node {$x_{i+2}$};
\node[draw,shape=circle,fill](yi+2) at (10.5,0) {};
\path (yi+2)+(0.25,-0.5) node {$y_{i+2}$};
\node[draw,shape=circle,fill](xi+3) at (10.5,1.5) {};
\path (xi+3)+(-0.7,0) node {$x_{i+3}$};
\node[draw,shape=circle,fill](yi+3) at (6.5,3) {};
\path (yi+3)+(0.6,0.25) node {$y_{i+3}$};
\node[draw,shape=circle,fill](xi+4) at (6.5,4.5) {};
\path (xi+4)+(0.7,0) node {$x_{i+4}$};
\node[draw,shape=circle,fill](yi+4) at (5,4.5) {};
\path (yi+4)+(-0.7,0) node {$y_{i+4}$};
\node at (6.5,0) {$\cdots$};
\draw[line width=1pt] (x1)--(y1)--(x2)--(y2)--(x3)--(yi+3)--(xi+1)--(yi+1)--(xi+2)--(yi+2)--(xi+3)
(x2)--(yi+3)--(xi+2) (yi+3)--(xi+4)--(yi+4)
(x3) -- ++(0.3,-0.5) (xi+1) -- ++(-0.3,-0.5)
(yi+3) -- ++(0.15,-1.2) (yi+3) -- ++(-0.15,-1.2);
\end{tikzpicture}}
}\qquad
\subfigure[$W_i$]{
\scalebox{0.85}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=0.7]
\node[draw,shape=circle,fill](x1) at (0,0) {};
\path (x1)+(-0.5,0.25) node {$x_1$};
\node[draw,shape=circle,fill](x2) at (1.5,0) {};
\path (x2)+(-0.5,0.25) node {$x_2$};
\node[draw,shape=circle,fill](y1) at (0.75,-2) {};
\path (y1)+(0,-0.5) node {$y_1$};
\node[draw,shape=circle,fill](xi) at (3.5,0) {};
\path (xi)+(0.5,0.25) node {$x_i$};
\node[draw,shape=circle,fill](yi) at (4.25,-2) {};
\path (yi)+(0,-0.5) node {$y_i$};
\node[draw,shape=circle,fill](xi+1) at (5,0) {};
\path (xi+1)+(0.6,0.25) node {$x_{i+1}$};
\node[draw,shape=circle,fill](yi+1) at (1.5,2) {};
\path (yi+1)+(-0.5,0.25) node {$y_{i+1}$};
\node[draw,shape=circle,fill](yi+2) at (3.5,2) {};
\path (yi+2)+(0.5,0.25) node {$y_{i+2}$};
\node[draw,shape=circle,fill](xi+2) at (2.5,2.5) {};
\path (xi+2)+(0,-0.5) node {$x_{i+2}$};
\node[draw,shape=circle,fill](yi+3) at (2.5,3.5) {};
\path (yi+3)+(0,0.5) node {$y_{i+3}$};
\node at (2.5,0) {$\cdots$};
\draw[line width=1pt] (x1)--(y1)--(x2)--(yi+1)--(xi)--(yi)--(xi+1)--(yi+2)--(x2)
(x1)--(yi+1)--(xi+2)--(yi+3) (xi+2)--(yi+2)--(xi)
(x2) -- ++(0.3,-0.5) (xi) -- ++(-0.3,-0.5)
(yi+1) -- ++(0.3,-1.2) (yi+1) -- ++(0.6,-1.2)
(yi+2) -- ++(-0.3,-1.2) (yi+2) -- ++(-0.6,-1.2);
\end{tikzpicture}}
}\qquad
\subfigure[$D_i$]{
\scalebox{0.85}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=0.7]
\node[draw,shape=circle,fill](x1) at (0,0) {};
\path (x1)+(0,0.4) node {$x_1$};
\node[draw,shape=circle,fill](x2) at (1.5,0) {};
\path (x2)+(-0.25,0.4) node {$x_2$};
\node[draw,shape=circle,fill](y1) at (0.75,-2) {};
\path (y1)+(-0.25,-0.3) node {$y_1$};
\node[draw,shape=circle,fill](y2) at (2.25,-2) {};
\path (y2)+(-0.25,-0.3) node {$y_2$};
\node[draw,shape=circle,fill](xi+1) at (3.5,0) {};
\path (xi+1)+(0.6,0.25) node {$x_{i+1}$};
\node[draw,shape=circle,fill](yi+1) at (4.25,-2) {};
\path (yi+1)+(0.7,0) node {$y_{i+1}$};
\node[draw,shape=circle,fill](xi+2) at (5,0) {};
\path (xi+2)+(0.5,0.35) node {$x_{i+2}$};
\node[draw,shape=circle,fill](yi+2) at (6.5,0) {};
\path (yi+2)+(-0.2,-0.5) node {$y_{i+2}$};
\node[draw,shape=circle,fill](xi+3) at (6.85,1) {};
\path (xi+3)+(0.5,0.3) node {$x_{i+3}$};
\node[draw,shape=circle,fill](yi+3) at (6.85,-1) {};
\path (yi+3)+(0.5,-0.4) node {$y_{i+3}$};
\node[draw,shape=circle,fill](xi+4) at (3.25,-4) {};
\path (xi+4)+(0,-0.5) node {$x_{i+4}$};
\node[draw,shape=circle,fill](yi+4) at (2.5,2) {};
\path (yi+4)+(0,0.5) node {$y_{i+4}$};
\node at (2.5,0) {$\cdots$};
\node at (3.25,-2) {$\cdots$};
\draw[line width=1pt] (x1)--(y1)--(xi+4)--(y2)--(x2)--(y1)
(x2)--(yi+4)--(xi+1)--(yi+1)
(xi+4)--(yi+1)--(xi+2)--(yi+2)
(xi+2)--(yi+4) .. controls +(-4,0) and +(-5.5,0.5) .. (xi+4)--(yi+3)--(xi+3)--(yi+4)
(y2) -- ++(0.25,0.5) (xi+1) -- ++(-0.25,-0.5)
(yi+4) -- ++(0.25,-1.2) (yi+4) -- ++(-0.25,-1.2)
(xi+4) -- ++(0.3,1.2) (xi+4) -- ++(-0.3,1.2);
\end{tikzpicture}}
}\qquad
\subfigure[$M_i$]{
\scalebox{0.85}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=0.7]
\node[draw,shape=circle,fill](x1) at (0,0) {};
\path (x1)+(-0.35,0.3) node {$x_1$};
\node[draw,shape=circle,fill](x2) at (1.5,0) {};
\path (x2)+(-0.5,0) node {$x_2$};
\node[draw,shape=circle,fill](x3) at (3,0) {};
\path (x3)+(-0.5,-0.1) node {$x_3$};
\node[draw,shape=circle,fill](x4) at (4.5,0) {};
\path (x4)+(-0.5,-0.1) node {$x_4$};
\node[draw,shape=circle,fill](x2i) at (6.5,0) {};
\path (x2i)+(0.5,0.25) node {$x_{2i}$};
\node[draw,shape=circle,fill](x2i+1) at (8,0) {};
\path (x2i+1)+(0.7,0.5) node {$x_{2i+1}$};
\node[draw,shape=circle,fill](x2i+2) at (9.5,0) {};
\path (x2i+2)+(0.5,0.25) node {$x_{2i+2}$};
\node[draw,shape=circle,fill](x2i+3) at (11,0) {};
\path (x2i+3)+(0.5,0.35) node {$x_{2i+3}$};
\node[draw,shape=circle,fill](x2i+4) at (-1,2.5) {};
\path (x2i+4)+(0,0.5) node {$x_{2i+4}$};
\node[draw,shape=circle,fill](y1) at (7,-2.5) {};
\path (y1)+(0,-0.5) node {$y_1$};
\node[draw,shape=circle,fill](y2) at (5.5,-2.5) {};
\path (y2)+(0,-0.5) node {$y_2$};
\node[draw,shape=circle,fill](yi+1) at (2,-2.5) {};
\path (yi+1)+(0,-0.5) node {$y_{i+1}$};
\node[draw,shape=circle,fill](yi+2) at (-1,0) {};
\path (yi+2)+(0,-0.5) node {$y_{i+2}$};
\node[draw,shape=circle,fill](yi+3) at (0.5,2.5) {};
\path (yi+3)+(0,0.5) node {$y_{i+3}$};
\node[draw,shape=circle,fill](y2i+3) at (7.5,2.5) {};
\path (y2i+3)+(0,0.5) node {$y_{2i+3}$};
\node[draw,shape=circle,fill](y2i+4) at (9,2.5) {};
\path (y2i+4)+(0,0.5) node {$y_{2i+4}$};
\node at (5.5,0) {$\cdots$};
\node at (3.25,-2.5) {$\cdots$};
\node at (2.5,2.5) {$\cdots$};
\draw[line width=1pt] (yi+2)--(x1)--(yi+1)--(x2)--(y2)--(x1)--(y1)--(x3)--(yi+1)
(x3)--(y2)--(x2i)--(yi+3)--(x2i+2)--(y1)--(x2i+1)--(y2)
(y2)--(x4)--(y1)--(x2i+3)--(y2i+4)--(x2)--(yi+3)--(x1)
(x2i+4)--(yi+3)--(x3) (x2)--(y1)--(x2i)
(yi+3)--(x4)--(y2i+3)--(x2)
(y1) -- ++(-0.5,1.5) (y1) -- ++(-0.8,1.5)
(y2) -- ++(-0.2,1.5) (y2) -- ++(0.2,1.5)
(yi+3) -- ++(2.3,-1.2) ;
\end{tikzpicture}}
}\qquad
\subfigure[$N_i$]{
\scalebox{0.85}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=0.7]
\node[draw,shape=circle,fill](x1) at (0,0) {};
\path (x1)+(-0.35,0.3) node {$x_1$};
\node[draw,shape=circle,fill](x2) at (1.5,0) {};
\path (x2)+(-0.5,0) node {$x_2$};
\node[draw,shape=circle,fill](x3) at (3,0) {};
\path (x3)+(-0.5,-0.1) node {$x_3$};
\node[draw,shape=circle,fill](x4) at (4.5,0) {};
\path (x4)+(-0.5,-0.1) node {$x_4$};
\node[draw,shape=circle,fill](x2i) at (6.5,0) {};
\path (x2i)+(0.5,0.25) node {$x_{2i}$};
\node[draw,shape=circle,fill](x2i+1) at (8,0) {};
\path (x2i+1)+(0.7,0.5) node {$x_{2i+1}$};
\node[draw,shape=circle,fill](x2i+2) at (9.5,0) {};
\path (x2i+2)+(0.5,0.25) node {$x_{2i+2}$};
\node[draw,shape=circle,fill](x2i+3) at (11,0) {};
\path (x2i+3)+(0.5,0.35) node {$x_{2i+3}$};
\node[draw,shape=circle,fill](x2i+4) at (-1,2.5) {};
\path (x2i+4)+(0,0.5) node {$x_{2i+4}$};
\node[draw,shape=circle,fill](y1) at (5.5,-2.5) {};
\path (y1)+(0,-0.5) node {$y_1$};
\node[draw,shape=circle,fill](yi) at (2,-2.5) {};
\path (yi)+(0,-0.5) node {$y_{i}$};
\node[draw,shape=circle,fill](yi+1) at (-1,0) {};
\path (yi+1)+(0,-0.5) node {$y_{i+1}$};
\node[draw,shape=circle,fill](yi+2) at (0.5,2.5) {};
\path (yi+2)+(0,0.5) node {$y_{i+2}$};
\node[draw,shape=circle,fill](y2i+2) at (7.5,2.5) {};
\path (y2i+2)+(0,0.5) node {$y_{2i+2}$};
\node[draw,shape=circle,fill](y2i+3) at (9,2.5) {};
\path (y2i+3)+(0,0.5) node {$y_{2i+3}$};
\node[draw,shape=circle,fill](y2i+4) at (-1,1.5) {};
\path (y2i+4)+(0,-0.5) node {$y_{2i+4}$};
\node at (5.5,0) {$\cdots$};
\node at (3.25,-2.5) {$\cdots$};
\node at (2.5,2.5) {$\cdots$};
\draw[line width=1pt] (yi+1)--(x1)--(yi)--(x2)--(y1)--(x1) (x3)--(yi)
(x3)--(y1)--(x2i)--(yi+2)--(x2i+2) (x2i+1)--(y1)
(y1)--(x4) (x2i+3)--(y2i+3)--(x2)--(yi+2)--(x1)
(y2i+4)--(x2i+4)--(yi+2)--(x3)
(yi+2)--(x4)--(y2i+2)--(x2)
(y1) -- ++(-0.2,1.5) (y1) -- ++(0.2,1.5)
(yi+2) -- ++(2.3,-1.2);
\end{tikzpicture}}
}\\
\subfigure[$G_1$]{
\scalebox{0.85}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=0.7]
\node[draw,shape=circle,fill](x1) at (0.25,0) {};
\path (x1)+(-0.5,0) node {$x_1$};
\node[draw,shape=circle,fill](y2) at (0.25,1) {};
\path (y2)+(-0.5,0) node {$y_2$};
\node[draw,shape=circle,fill](x2) at (0.25,2) {};
\path (x2)+(-0.5,0) node {$x_2$};
\node[draw,shape=circle,fill](x3) at (1.75,0) {};
\path (x3)+(0.5,0) node {$x_3$};
\node[draw,shape=circle,fill](y3) at (1.75,1) {};
\path (y3)+(0.5,0) node {$y_3$};
\node[draw,shape=circle,fill](x4) at (1.75,2) {};
\path (x4)+(0.5,0) node {$x_4$};
\node[draw,shape=circle,fill](y1) at (1,0) {};
\path (y1)+(0,0.5) node {$y_1$};
\node[draw,shape=circle,fill](x5) at (1,-1) {};
\path (x5)+(0.5,0) node {$x_5$};
\node[draw,shape=circle,fill](y4) at (1,-2) {};
\path (y4)+(0.5,0) node {$y_4$};
\node[draw,shape=circle,fill](x6) at (1,-3) {};
\path (x6)+(0.5,0) node {$x_6$};
\draw[line width=1pt] (x2)--(y2)--(x1)--(y1)--(x5)--(y4)--(x6)
(y1)--(x3)--(y3)--(x4);
\end{tikzpicture}}
}\qquad
\subfigure[$G_2$]{
\scalebox{0.85}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=0.7]
\node[draw,shape=circle,fill](y4) at (0,0) {};
\path (y4)+(-0.5,0) node {$y_4$};
\node[draw,shape=circle,fill](x6) at (0,1) {};
\path (x6)+(-0.5,0) node {$x_6$};
\node[draw,shape=circle,fill](x3) at (0,-1) {};
\path (x3)+(-0.5,0) node {$x_3$};
\node[draw,shape=circle,fill](y3) at (2,0) {};
\path (y3)+(0.5,0) node {$y_3$};
\node[draw,shape=circle,fill](x5) at (2,1) {};
\path (x5)+(0.5,0) node {$x_5$};
\node[draw,shape=circle,fill](x4) at (2,-1) {};
\path (x4)+(0.5,0) node {$x_4$};
\node[draw,shape=circle,fill](x2) at (1,0) {};
\path (x2)+(0,0.5) node {$x_2$};
\node[draw,shape=circle,fill](y2) at (1,-1) {};
\path (y2)+(0.4,0.3) node {$y_2$};
\node[draw,shape=circle,fill](x1) at (1,-2) {};
\path (x1)+(0.5,0) node {$x_1$};
\node[draw,shape=circle,fill](y1) at (1,-3) {};
\path (y1)+(0.5,0) node {$y_1$};
\draw[line width=1pt] (x6)--(y4)--(x2)--(y3)--(x5)
(y4)--(x3)--(y2)--(x4)--(y3)
(x2)--(y2)--(x1)--(y1);
\end{tikzpicture}}
}\qquad
\subfigure[$G_3$]{
\scalebox{0.85}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=0.7]
\node[draw,shape=circle,fill](x2) at (0,0) {};
\path (x2)+(-0.5,0) node {$x_2$};
\node[draw,shape=circle,fill](y4) at (0,1) {};
\path (y4)+(-0.5,0) node {$y_4$};
\node[draw,shape=circle,fill](x3) at (0,2) {};
\path (x3)+(-0.5,0) node {$x_3$};
\node[draw,shape=circle,fill](y2) at (0,3) {};
\path (y2)+(-0.5,0) node {$y_2$};
\node[draw,shape=circle,fill](x4) at (0,4) {};
\path (x4)+(-0.5,0) node {$x_4$};
\node[draw,shape=circle,fill](y1) at (1,0) {};
\path (y1)+(0.5,0) node {$y_1$};
\node[draw,shape=circle,fill](x1) at (1,1) {};
\path (x1)+(0.5,0) node {$x_1$};
\node[draw,shape=circle,fill](y5) at (1,2) {};
\path (y5)+(0.5,0) node {$y_5$};
\node[draw,shape=circle,fill](x5) at (1,3) {};
\path (x5)+(0.5,0) node {$x_5$};
\node[draw,shape=circle,fill](y3) at (1,4) {};
\path (y3)+(0.5,0) node {$y_3$};
\draw[line width=1pt] (x4)--(y2)--(x3)--(y5)--(x5)--(y3)
(x3)--(y4)--(x1)--(y5)
(y4)--(x2)--(y1)--(x1);
\end{tikzpicture}}
}
\caption{Illustration of the families defined in Table~\ref{table} (except the cycles in $\mathcal C$).}
\label{fig:table}
\end{figure}
\shortpaper{In fact, one can show the following result (see~\cite{fullversion} for a proof).
\begin{theorem}\label{thm:table}
Let $H$ be a graph in the characterization of forbidden induced subgraphs of~\cite{TM76} that is no an even cycle. Then, \PBtropCOL{$H$} is polynomial-time solvable.
\end{theorem}
}
In this section, we first turn our attention to the family of even cycles of length at least~$6$. We show that \PBtropCOL{$C_{2k}$} is polynomial-time solvable for any $k\leq 6$. On the other hand, for any $k\geq 24$, \PBtropCOL{$C_{2k}$} is NP-complete.
\longpaper{We then prove that for all other minimal graphs $H$ described in Table~\ref{table}, \PBtropCOL{$H$} is polynomial-time solvable.}
\subsection{Polynomial-time cases for even cycles}
We now prove that the tropical homomorphism problems for small even cycles are polynomial-time solvable.
\begin{theorem} \label{thm: C12}
For each integer $k$ with $2\leq k\leq 6$, \PBtropCOL{$C_{2k}$} is polynomial-time solvable.
\end{theorem}
\begin{proof}
Since \PBlistCOL{$C_4$} is polynomial-time solvable, \PBtropCOL{$C_{4}$} is polynomial-time solvable.
We will consider all cases $k\in\{3,4,5,6\}$ separately. But in each of those cases we note that
if $(C_{2k}, c)$ is not a core, then the core is path, and since the \PBlistCOL{$P_k$} is polynomial-time solvable for any $k\geq 1$, \PBtropCOL{$C_{2k}$} would also be polynomial-time solvable. Hence in the rest of the proof we always assume $(C_{2k}, c)$ is a core. Furthermore, by Proposition~\ref{prop:bipartite-hom}, we can assume that the colour sets of $c$ in $X$ and $Y$ are disjoint.
\medskip
First, assume $k=3$. There are three vertices in each part of the bipartition of $C_6$. If one vertex is coloured with a colour not present anywhere else in the part, Lemma~\ref{lemm:unique-feature} implies again that \PBCOL{$(C_6,c)$} is polynomial-time solvable. Hence, we can assume that each part of the bipartition is monochromatic. But then $(C_6,c)$ is not a core, a contradiction with our assumption.
\medskip
Suppose $k=4$. There are four vertices in each part of the bipartition $(X,Y)$ of $C_8$. If there is a vertex that, in $c$, is the only one coloured with its colour, since \PBlistCOL{$P_k$} is polynomial-time solvable for any $k\geq 1$, by Lemma~\ref{lemm:unique-feature} \PBCOL{$(C_8,c)$} is polynomial-time solvable. Hence we may assume that each colour appears at least twice, in particular each part of the bipartition is coloured with either one or two colours. If some part, say $X$, is coloured with only one colour (say Blue) then $(C_8,c)$ is not a core which again contradicts our assumption.
Hence, in each part, there are exactly two vertices of each colour. In this case we can use Lemma~\ref{lemm:2SAT} with $S_1$, $S_2$, $S_3$ and $S_4$ being the four sets of two vertices with the same colour. It follows that \PBCOL{$(C_8,c)$} is polynomial-time solvable.
\medskip
Assume that $k=5$, and let $c$ be a vertex-colouring of $C_{10}$. By similar arguments as in the proof of Theorems~\ref{thm:small_graph} and~\ref{thm:SmallTree}, using Lemma~\ref{lemm:unique-feature} and the fact that $(H,c)$ should not be homomorphic to a $P_2$- or $P_3$-subgraph, each part of the bipartition $(X,Y)$ contains exactly two vertices of one colour and three vertices of another colour, say $X$ has three vertices coloured~$1$ and two vertices coloured~$2$, and $Y$ has three vertices coloured~$a$ and two vertices coloured~$b$.
The cyclic order of the colours of $X$ can be either $1-1-1-2-2$ or $1-1-2-1-2$ (up to permutation of colours and other symmetries). If this order is $1-1-1-2-2$, then the vertex of $Y$ adjacent to the two vertices coloured~$2$ satisfies the hypothesis of Lemma~\ref{lemm:unique-feature} and hence \PBCOL{$(C_{10},c)$} is polynomial-time solvable. The same argument can be applied to $Y$, hence the cyclic order of the colours of $Y$ is $a-a-b-a-b$.
Hence, there is a unique vertex $y$ of $Y$ whose two neighbours are coloured~$1$. If $c(y)=b$, then the second vertex of $Y$ coloured~$b$ is in the centre of a $3$-vertex path coloured $1-b-2$ that satisfies the hypothesis of Lemma~\ref{lemm:unique-feature}, hence \PBCOL{$(C_{10},c)$} is polynomial-time solvable. Therefore, we have $c(y)=a$. By the same argument, the unique vertex of $X$ adjacent to two vertices of $Y$ coloured~$a$ must be coloured~$1$. Therefore, up to symmetries $c$ is one of the three colourings $1-a-1-a-2-b-1-a-2-b$, $1-a-1-b-2-a-1-a-2-b$ and $1-a-1-b-2-a-1-b-2-a$ (in the cyclic order).
We are going to use the Lemma~\ref{lemm:2SAT} to conclude the case $k=5$. In a homomorphism to $(C_{10},c)$, a vertex coloured~$2$ or~$b$ can only be mapped to the two vertices in $(C_{10},c)$ of the corresponding colour. A vertex $v$ coloured~$1$ adjacent to at least one vertex coloured~$b$ or a vertex coloured~$a$ adjacent to at least one vertex coloured~$2$ also can only be mapped to two vertices of $(C_{10},c)$ (the ones having the same properties as $v$). However, a vertex coloured~$1$ all whose neighbours are coloured~$a$ can be mapped to three different vertices in $(C_{10},c)$ (say $x_1$, $x_2$, $x_3$, the vertices coloured~$1$, that all have a neighbour coloured~$a$). But at least one of $x_1$, $x_2$, $x_3$, say $x_1$, has a common neighbour coloured~$a$ with one of the two other vertices (say $x_2$). Therefore, if there is a homomorphism $h$ of some tropical graph $(G,c_1)$ to $(C_{10},c)$ mapping a vertex $v$ of $G$ coloured~$1$ all whose neighbours are coloured~$a$ to $x_1$, we can modify $h$ so that $v$ is mapped to $x_2$ instead. In other words, there is a homomorphism of $(G,c_1)$ to $(C_{10},c)$ where none of the vertices coloured~$1$ all whose neighbours are coloured~$a$ is mapped to $x_1$. Therefore such vertices have two possible targets: $x_2$ and $x_3$. The same is true for vertices coloured~$a$ all whose neighbours are coloured~$1$. Thus, $(C_{10},c)$ satisfies the hypothesis of Lemma~\ref{lemm:2SAT} and \PBCOL{$(C_{10},c)$} is polynomial-time solvable.
\medskip
Finally, assume now that $k=6$. Again, using Lemma~\ref{lemm:unique-feature}, we can assume than each part of the bipartition has at most three colours, and each colour appears at least twice. Furthermore, if there are exactly three colours in each part, each colour appears exactly twice and hence \PBCOL{$(C_{12},c)$} is polynomial-time solvable by Lemma~\ref{lemm:2SAT}. If one part of the bipartition has one colour and the other has at most two colours, then $(C_{12},c)$ would not be a core. Therefore, the numbers of colours of the parts in the bipartition are either one and three, two and three, or two and two.
Assume that one part, say $X$, is monochromatic (say Red) and the other, $Y$, has three colours (thus two vertices of each colour). For the graph to be a core and not satisfy Lemma~\ref{lemm:unique-feature}, the three colours of $Y$ must form the cyclic pattern $x-y-z-x-y-z$. In this case, considering any vertex $v$ of colour Red in an input tropical graph $(G,c_1)$, in any homomorphism $(G,c_1)\to (C_{12},c)$, all the neighbours of $v$ with the same colour must be identified. Furthermore, no Red vertex in $(G,c_1)$ can have neighbours of three distinct colours. Therefore, the mapping of each connected component is forced after making a choice for one vertex. Since there are two choices per vertex, we have a polynomial-time algorithm for \PBCOL{$(C_{12},c)$}.
Assume now that one part, say $X$, contains two colours ($a$ and $b$) and the other, $Y$, contains three colours ($x$, $y$ and $z$). Note that there are exactly two vertices of each colour in $Y$. We are going to use Lemma~\ref{lemm:2SAT} to conclude this case. A vertex of some input graph $(G,c_1)$ coloured $x$, $y$ or $z$ can only be mapped to two possible vertices in $(C_{12},c)$. A vertex of $(G,c_1)$ coloured $a$ or $b$ (say $a$) and having all its neighbours of the same colour, say $x$, might be mapped to more than two vertices of $(C_{12},c)$. However, once again, there are always two of these vertices that, together, are adjacent to all the vertices of colour~$x$ (indeed, there are only two vertices of colour~$x$). These two vertices are the designated targets for Lemma~\ref{lemm:2SAT}. A vertex coloured~$a$ (or~$b$) with two different colours in its neighbourhood can only be mapped to two possible vertices if there is no pattern $x-a-y-a-x-a-y$ in the graph (up to permutation of colours). Hence, if there is no such pattern in the graph (up to permutation of colours), $(C_{12},c)$ satisfies the hypothesis of Lemma~\ref{lemm:2SAT} and \PBCOL{$(C_{12},c)$} is polynomial-time solvable. On the other hand, if there is a pattern $x-a-y-a-x-a-y$ in the graph, then there is a unique path coloured $a-x-b$ or $a-y-b$ in the graph and, by Lemma~\ref{lemm:unique-feature}, \PBCOL{$(C_{12},c)$} is polynomial-time solvable as well.
Therefore, we are left to consider the cases where there are exactly two colours in each part. We assume first that there are two vertices coloured~$a$ and four vertices coloured~$b$ in one part, say $X$. If the neighbours of vertices of colour~$a$ all have the same colour, say~$x$, then $(C_{12},c)$ is not a core because it can be mapped to its sub-path coloured $a-x-b-y$. We suppose without loss of generality that the coloured cycle contains a path coloured $y-a-x-b$. Then, if there is no other path coloured $y-a-x-b$, by Lemma~\ref{lemm:unique-feature} \PBCOL{$(C_{12},c)$} is polynomial-time solvable. Therefore, there is another such path in $(C_{12},c)$. If this other path is part of a path $x-a-y-a-x$, then the problem is polynomial-time solvable by applying Lemma~\ref{lemm:unique-feature} to the star $a-y-a$. Up to symmetry, we are left with two cases: $y-a-x-b-.-b-.-a-.-b-.-b$ or $y-a-x-b-.-b-.-b-.-a-.-b$ (where a dot could be colour $x$ or $y$). The first case must be $y-a-x-b-.-b-y-a-x-b-.-b$, because otherwise, $(C_{12},c)$ is not a core. Any placement of the remaining $x$'s and $y$'s yields a polynomial-time solvable case using Lemma~\ref{lemm:distinct-vertex-boundary}. Similarly, the second case must be $y-a-x-b-.-b-.-b-y-a-x-b$. Then, \PBCOL{$(C_{12},c)$} is polynomial-time solvable because of Lemma~\ref{lemm:distinct-edge-boundary}, with $a-x-b-y-a$ as forcing set and $x-b-.-b-.-b-y$, which contains no vertex coloured~$a$, as the other set.
Finally, we can assume, without loss of generality, that there are exactly three vertices for each of the two colours in each part. There are three possible configurations in each part: $a-a-a-b-b-b$, $a-a-b-b-a-b$ or $a-b-a-b-a-b$, up to permutations of colours. If one part of the bipartition is in the first configuration, then, either we have the pattern $a-x-b$ or $a-y-b$ that satisfies the hypothesis of Lemma~\ref{lemm:unique-feature}, or we have two paths $a-x-b$, in which case, there is a unique path $a-y-a$ or $b-y-b$ which satisfies the hypothesis of Lemma~\ref{lemm:unique-feature}. Suppose some part of the bipartition is in the second case. Then, if we have the pattern $a-x-a-.-b-x-b-.-a-.-b-.$, there is a unique path $a-x-b$ satisfying the hypothesis of Lemma~\ref{lemm:unique-feature}. Otherwise, we have the pattern $a-x-a-.-b-y-b-.-a-.-b-.$, in which case we can apply Lemma~\ref{lemm:2SAT} in a similar way as for $C_{10}$. Therefore, both parts of the bipartition must be in the third configuration. But then every vertex is a forcing vertex and we can apply Lemma~\ref{lemm:all-forcing}. This completes the proof.
\end{proof}
\subsection{NP-completeness results for even cycles}
We now show that \PBtropCOL{$C_{2k}$} is NP-complete whenever $k\geq 24$.
We present a proof using a specific $4$-tropical $48$-cycle. The proof holds similarly for any larger even cycle. It also works similarly for some $3$-tropical cycles $C_{2k}$ for $k\geq 24$ and for $2$-tropical cycles $C_{2k}$ for $k\geq 27$.
We use the colour set $\{G, B, R, Y\}$ (for Green, Blue, Red and Yellow).
We define $P_{x,y}$ to be a tropical path of length~$8$, with vertices
$x=x_0, x_1, \dots, x_7, x_8=y$ where $\{c(x),c(y)\}=\{G, B\}$, $c(x_5)=R$ and all others are coloured Yellow.
Thus, $P_{x,y}$ represents one of the two non-isomorphic tropical graphs from Figure~\ref{fig:Pxy}.
The distance of the only vertex of colour $R$ from the two ends defines an orientation from one end to another. Thus, in our figures, an arc between two vertices $u$ and $v$ is a $P_{uv}$ path.
\begin{figure}[ht!]
\centering
\scalebox{1}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=1]
\begin{scope}
\node[draw,shape=circle,color=black,fill](x0) at (0,0) {};
\path (x0)+(0,-0.3) node {$x$};
\path (x0)+(0,+0.3) node {$G$};
\node[draw,shape=star,color=black,fill](x1) at (1,0) {};
\path (x1)+(0,-0.3) node {$x_1$};
\path (x1)+(0,+0.3) node {$Y$};
\node[draw,shape=star,color=black,fill](x2) at (2,0) {};
\path (x2)+(0,-0.3) node {$x_2$};
\path (x2)+(0,+0.3) node {$Y$};
\node[draw,shape=star,color=black,fill](x3) at (3,0) {};
\path (x3)+(0,-0.3) node {$x_3$};
\path (x3)+(0,+0.3) node {$Y$};
\node[draw,shape=star,color=black,fill](x4) at (4,0) {};
\path (x4)+(0,-0.3) node {$x_4$};
\path (x4)+(0,+0.3) node {$Y$};
\node[draw,shape=rectangle,scale=1.4,color=black,fill](x5) at (5,0) {};
\path (x5)+(0,-0.3) node {$x_5$};
\path (x5)+(0,+0.3) node {$R$};
\node[draw,shape=star,color=black,fill](x6) at (6,0) {};
\path (x6)+(0,-0.3) node {$x_6$};
\path (x6)+(0,+0.3) node {$Y$};
\node[draw,shape=star,color=black,fill](x7) at (7,0) {};
\path (x7)+(0,-0.3) node {$x_7$};
\path (x7)+(0,+0.3) node {$Y$};
\node[draw,shape=diamond,color=black,fill](x8) at (8,0) {};
\path (x8)+(0,-0.3) node {$y$};
\path (x8)+(0,+0.3) node {$B$};
\draw[line width=1pt] (x0)--(x1)--(x2)--(x3)--(x4)--(x5)--(x6)--(x7)--(x8);
\end{scope}
\begin{scope}[yshift=-1.2cm]
\node[draw,shape=diamond,color=black,fill](x0) at (0,0) {};
\path (x0)+(0,-0.3) node {$x$};
\path (x0)+(0,+0.3) node {$B$};
\node[draw,shape=star,color=black,fill](x1) at (1,0) {};
\path (x1)+(0,-0.3) node {$x_1$};
\path (x1)+(0,+0.3) node {$Y$};
\node[draw,shape=star,color=black,fill](x2) at (2,0) {};
\path (x2)+(0,-0.3) node {$x_2$};
\path (x2)+(0,+0.3) node {$Y$};
\node[draw,shape=star,color=black,fill](x3) at (3,0) {};
\path (x3)+(0,-0.3) node {$x_3$};
\path (x3)+(0,+0.3) node {$Y$};
\node[draw,shape=star,color=black,fill](x4) at (4,0) {};
\path (x4)+(0,-0.3) node {$x_4$};
\path (x4)+(0,+0.3) node {$Y$};
\node[draw,shape=rectangle,scale=1.4,color=black,fill](x5) at (5,0) {};
\path (x5)+(0,-0.3) node {$x_5$};
\path (x5)+(0,+0.3) node {$R$};
\node[draw,shape=star,color=black,fill](x6) at (6,0) {};
\path (x6)+(0,-0.3) node {$x_6$};
\path (x6)+(0,+0.3) node {$Y$};
\node[draw,shape=star,color=black,fill](x7) at (7,0) {};
\path (x7)+(0,-0.3) node {$x_7$};
\path (x7)+(0,+0.3) node {$Y$};
\node[draw,shape=circle,color=black,fill](x8) at (8,0) {};
\path (x8)+(0,-0.3) node {$y$};
\path (x8)+(0,+0.3) node {$G$};
\draw[line width=1pt] (x0)--(x1)--(x2)--(x3)--(x4)--(x5)--(x6)--(x7)--(x8);
\end{scope}
\end{tikzpicture}}\caption{The two non-isomorphic graphs of type $P_{xy}$.}
\label{fig:Pxy}
\end{figure}
Similarly, $Q_{z,t}$ is defined to be a tropical path of length~$10$ with vertices
$z=z_0, z_1, \dots, z_9, z_{10}=t$ where $\{c(z),c(t)\}=\{G, B\}$, $c(z_5)=R$ and all others are coloured Yellow. See Figure~\ref{fig:QZt} for an illustration.
In this case, as the only vertex of colour $R$ is at the same distance from both ends, the two possible colourings of the end-vertices correspond to isomorphic graphs. Hence, in our figures, a dotted edge will be used to represent a $Q$-type path between two vertices.
\begin{figure}[ht!]
\centering
\scalebox{1}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=1]
\begin{scope}
\node[draw,shape=circle,color=black,fill](x0) at (0,0) {};
\path (x0)+(0,-0.3) node {$z$};
\path (x0)+(0,+0.3) node {$G$};
\node[draw,shape=star,color=black,fill](x1) at (1,0) {};
\path (x1)+(0,-0.3) node {$z_1$};
\path (x1)+(0,+0.3) node {$Y$};
\node[draw,shape=star,color=black,fill](x2) at (2,0) {};
\path (x2)+(0,-0.3) node {$z_2$};
\path (x2)+(0,+0.3) node {$Y$};
\node[draw,shape=star,color=black,fill](x3) at (3,0) {};
\path (x3)+(0,-0.3) node {$z_3$};
\path (x3)+(0,+0.3) node {$Y$};
\node[draw,shape=star,color=black,fill](x4) at (4,0) {};
\path (x4)+(0,-0.3) node {$z_4$};
\path (x4)+(0,+0.3) node {$Y$};
\node[draw,shape=rectangle,scale=1.4,color=black,fill](x5) at (5,0) {};
\path (x5)+(0,-0.3) node {$z_5$};
\path (x5)+(0,+0.3) node {$R$};
\node[draw,shape=star,color=black,fill](x6) at (6,0) {};
\path (x6)+(0,-0.3) node {$z_6$};
\path (x6)+(0,+0.3) node {$Y$};
\node[draw,shape=star,color=black,fill](x7) at (7,0) {};
\path (x7)+(0,-0.3) node {$z_7$};
\path (x7)+(0,+0.3) node {$Y$};
\node[draw,shape=star,color=black,fill](x8) at (8,0) {};
\path (x8)+(0,-0.3) node {$z_8$};
\path (x8)+(0,+0.3) node {$Y$};
\node[draw,shape=star,color=black,fill](x9) at (9,0) {};
\path (x9)+(0,-0.3) node {$z_9$};
\path (x9)+(0,+0.3) node {$Y$};
\node[draw,shape=diamond,color=black,fill](x10) at (10,0) {};
\path (x10)+(0,-0.3) node {$t$};
\path (x10)+(0,+0.3) node {$B$};
\draw[line width=1pt] (x0)--(x1)--(x2)--(x3)--(x4)--(x5)--(x6)--(x7)--(x8)--(x9)--(x10);
\end{scope}
\end{tikzpicture}}
\caption{The $Q$-type path $Q_{z,t}$.}
\label{fig:QZt}
\end{figure}
The following lemma is easy to observe.
\begin{lemma} \label{lem:arc_and_dotted_line}
The following is true.
\begin{enumerate}
\item $P_{x,y}$ admits a tropical homomorphism to $P_{u,v}$ if and only if $c(x)=c(u)$ and $c(y)=c(v)$.
\item $Q_{z,t}$ admits a tropical homomorphism to $P_{u,v}$ both in the case where $c(z)=c(u)$ and $c(t)=c(v)$, and in the case where $c(z)=c(v)$ and $c(t)=c(u)$.
\end{enumerate}
\end{lemma}
By Lemma~\ref{lem:arc_and_dotted_line}, in our abbreviated notation of arcs and dotted edges, a dotted edge can map to a dotted edge or to an arc as long as the colours of the end-vertices are preserved. However, to map an arc to another arc, not only the colours of the end-vertices must be preserved, but also the direction of the arc.
With our notation, the tropical directed $6$-cycle of Figure~\ref{fig:C48} corresponds to a $4$-tropical $48$-cycle, $(C_{48},c)$.
\begin{figure}[ht!]
\centering
\scalebox{1}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=1]
\begin{scope}
\draw node[color=black,fill,circle,label=0*360/6-90:$g_0$,label=0*360/6+90:$G$](g0) at (0*360/6-90:1.3cm) {};
\draw node[color=black,fill,diamond,label=1*360/6-90:$b_0$,label=1*360/6+90:$B$](b0) at (1*360/6-90:1.3cm) {};
\draw node[color=black,fill,circle,label=2*360/6-90:$g_1$,label=2*360/6+90:$G$](g1) at (2*360/6-90:1.3cm) {};
\draw node[color=black,fill,diamond,label=3*360/6-90:$b_1$,label=3*360/6+90:$B$](b1) at (3*360/6-90:1.3cm) {};
\draw node[color=black,fill,circle,label=4*360/6-90:$g_2$,label=4*360/6+90:$G$](g2) at (4*360/6-90:1.3cm) {};
\draw node[color=black,fill,diamond,label=5*360/6-90:$b_2$,label=5*360/6+90:$B$](b2) at (5*360/6-90:1.3cm) {};
\draw[->, line width=1pt] (g0)--(b0);
\draw[->, line width=1pt] (b0)--(g1);
\draw[->, line width=1pt] (g1)--(b1);
\draw[->, line width=1pt] (b1)--(g2);
\draw[->, line width=1pt] (g2)--(b2);
\draw[->, line width=1pt] (b2)--(g0);
\end{scope}
\end{tikzpicture}}
\caption{A short representation of the $4$-tropical $48$-cycle $(C_{48},c)$.}
\label{fig:C48}
\end{figure}
Our aim is to show that NAE $3$-SAT reduces (in polynomial time) to \PBCOL{$(C_{48},c)$}.
\begin{theorem} \label{thm:C48}
For any $k\geq 24$, \PBtropCOL{$C_{2k}$} is NP-complete.
\end{theorem}
\begin{proof}
We prove the statement when $k=24$ and observe that the same reduction holds for any $k\geq 24$. Indeed, one can make $P_{x,y}$ and $Q_{z,t}$ longer while still satisfying Lemma \ref{lem:arc_and_dotted_line}.
\PBCOL{$(C_{48},c)$} is clearly in NP. To show NP-hardness, we show that NAE $3$-SAT can be reduced in polynomial-time to \PBCOL{$(C_{48},c)$}.
Let $(X,C)$ be an instance of NAE $3$-SAT. To partition $X$ into two parts, it is enough to decide, for each pair of elements of $X$, whether they are in a same part or not. Thus, we are expected to define a binary relation among variables which satisfies the following conditions.
\begin{enumerate}
\item $X_p\sim X_q \wedge X_q\sim X_r \Rightarrow X_q\sim X_r$ (Partition)
\item $X_p\nsim X_q \wedge X_q\nsim X_r \Rightarrow X_p\sim X_r$ (Partition into two parts)
\end{enumerate}
To build our gadget, we start with a partial gadget associated to each pair of variables of $X$. To each pair $x_i, x_j\in X$, we associate the $4$-tropical $6$-cycle $(C_{x_ix_j},c)$ of Figure~\ref{fig:gadgetCx1x2}. Here, $U_G$ (coloured Green) is a common vertex of all such cycles, but all other vertices are distinct.
\begin{figure}[ht!]
\centering
\scalebox{1}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=1]
\begin{scope}
\draw node[color=black,fill,circle,label=0*360/6-90:$U_G$,label=0*360/6+90:$G$](g0) at (0*360/6-90:1.3cm) {};
\draw node[color=black,fill,diamond,label=1*360/6-90:$b^0_{x_ix_j}$,label=1*360/6+90:$B$](b0) at (1*360/6-90:1.3cm) {};
\draw node[color=black,fill,circle,label=2*360/6-90:$g^1_{x_ix_j}$,label=2*360/6+90:$G$](g1) at (2*360/6-90:1.3cm) {};
\draw node[color=black,fill,diamond,label=3*360/6-90:$b^1_{x_ix_j}$,label=3*360/6+90:$B$](b1) at (3*360/6-90:1.3cm) {};
\draw node[color=black,fill,circle,label=4*360/6-90:$g^2_{x_ix_j}$,label=4*360/6+90:$G$](g2) at (4*360/6-90:1.3cm) {};
\draw node[color=black,fill,diamond,label=5*360/6-90:$b^2_{x_ix_j}$,label=5*360/6+90:$B$](b2) at (5*360/6-90:1.3cm) {};
\draw[->, line width=1pt] (g0)--(b0);
\draw[->, line width=1pt] (b0)--(g1);
\draw[loosely dashed, line width=1pt] (g1)--(b1);
\draw[->, line width=1pt] (b1)--(g2);
\draw[loosely dashed, line width=1pt] (g2)--(b2);
\draw[loosely dashed, line width=1pt] (b2)--(g0);
\end{scope}
\end{tikzpicture}}
\caption{$(C_{x_ix_j},c)$}
\label{fig:gadgetCx1x2}
\end{figure}
We are interested in possible mappings of this partial gadget into our tropical $48$-cycle, $(C_{48}, c)$ of Figure~\ref{fig:C48}. By the symmetries of $(C_{48},c)$, we assume, without loss of generality, that $U_G$ maps to $g_0$. Having this assumed, we observe the following crucial fact.
\begin{claim}\label{claim:C48}
There are exactly two possible homomorphisms of $(C_{x_ix_j},c)$ to $(C_{48},c)$.
\begin{enumerate}
\item A mapping $\sigma$ given by $\sigma(U_G)=g_0$, $\sigma(b^0_{x_ix_j})=b_0$, $\sigma(g^1_{x_ix_j})=g_1$, $\sigma(b^1_{x_ix_j})=b_1$, $\sigma(g^2_{x_ix_j})=g_2$ and $\sigma(b^2_{x_ix_j})=b_2$
\item A mapping $\rho$ give by $\rho(U_G)=g_0$, $\rho(b^0_{x_ix_j})=b_0$, $\rho(g^1_{x_ix_j})=g_1$, $\rho(b^1_{x_ix_j})=b_0$,
$\rho(g^2_{x_ix_j})=g_1$ and $\rho(b^2_{x_ix_j})=b_0$
\end{enumerate}
\end{claim}
The main idea of our reduction lies in Claim~\ref{claim:C48}. After completing the description of our gadgets, we will have a $4$-tropical graph containing a copy of $C_{x_ix_j}$ for each pair $x_i,x_j$ of variables. If we find a homomorphism of this graph to $(C_{48},c)$, then its restriction to $C_{x_ix_j}$ is either a mapping of type $\sigma$, or of type $\rho$. A $\sigma$-mapping would correspond to assigning $x_i$ and $x_j$ to two different parts, and a $\rho$-mapping would correspond to assigning them to a same part of a partition of $X$.
\begin{observation}
It is never possible to map $b^2_{x_ix_j}$ to $b_1$ or to map $b^1_{x_ix_j}$ to $b_2$.
\end{observation}
To enforce the two conditions, partitioning $X$ into two parts by a binary relation, we add more structures. Consider the three partial gadgets $(C_{x_px_q},c)$, $(C_{x_qx_r},c)$ and $(C_{x_px_r},c)$.
Considering $b^1_{x_px_q}$ of $(C_{x_px_q},c)$, we choose vertices $b^2_{x_px_r}$ and $b^2_{x_qx_r}$
from $(C_{x_px_r},c)$ and $(C_{x_qx_r},c)$ respectively, and connect them by a tree as in Figure~\ref{fig:treeCx1x2x3}.
The internal vertices of these trees are all new and distinct.
\begin{figure}[ht!]
\centering
\scalebox{0.75}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=1]
\begin{scope}
\draw node[color=black,fill,diamond,label=0:$b^1_{x_px_q}$,label=180:$B$](b1x1x2) at (0,0) {};
\draw node[color=black,fill,diamond,label=-90:$b^2_{x_qx_r}$,label=90:$B$](b2x2x3) at (3,-1) {};
\draw node[color=black,fill,diamond,label=0:$b^2_{x_px_r}$,label=180:$B$](b2x1x3) at (0,-2) {};
\draw node[color=black,fill,circle,label=180:$G$](g0) at (0,-1) {};
\draw node[color=black,fill,circle,label=90:$G$](g1) at (2,-1) {};
\draw node[color=black,fill,diamond,label=90:$B$](b0) at (1,-1) {};
\draw[->, line width=1pt] (g1)--(b0);
\draw[->, line width=1pt] (b0)--(g0);
\draw[loosely dashed, line width=1pt] (b1x1x2)--(g0)--(b2x1x3);
\draw[loosely dashed, line width=1pt] (g1)--(b2x2x3);
\end{scope}
\end{tikzpicture}}
\caption{Tree connecting $b^1_{x_px_q}$, $b^2_{x_px_r}$ and $b^2_{x_qx_r}$}
\label{fig:treeCx1x2x3}
\end{figure}
We build similar structures on $(b^1_{x_px_r},b^2_{x_qx_r},b^2_{x_px_q})$ and on $(b^1_{x_qx_r},b^2_{x_px_q},b^2_{x_px_r})$, where the order corresponds to the structure. Let $(C_{x_px_qx_r},c)$ be the resulting partial gadget (see Figure \ref{fig:Cx1x2x3}).
\begin{figure}[ht!]
\centering
\scalebox{1}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=1]
\begin{scope}
\node[draw,circle,fill,label=-90:$U_G$,color=black,label=30:$G$](ug) at (0,0) {};
\begin{scope}[yshift=4cm]
\node[draw,color=black,fill,diamond,label=0*180/4+180:$B$](b0x12) at (0*180/4:1.5cm) {};
\node[draw,color=black,fill,circle,label=1*180/4+180:$G$](g1x12) at (1*180/4:1.5cm) {};
\node[draw,diamond,fill,label=2*180/4:$b^1_{x_px_q}$,color=black,label=2*180/4+180:$B$](b1x12) at (2*180/4:1.5cm) {};
\node[draw,color=black,fill,circle,label=3*180/4+180:$G$](g2x12) at (3*180/4:1.5cm) {};
\node[draw,diamond,fill,label=4*180/4:$b^2_{x_px_q}$,color=black,label=4*180/4+180:$B$](b2x12) at (4*180/4:1.5cm) {};
\draw[->, line width=1pt] (ug)--(b0x12);
\draw[->, line width=1pt] (b0x12)--(g1x12);
\draw[loosely dashed, line width=1pt] (g1x12)--(b1x12);
\draw[->, line width=1pt] (b1x12)--(g2x12);
\draw[loosely dashed, line width=1pt] (g2x12)--(b2x12);
\draw[loosely dashed, line width=1pt] (b2x12)--(ug);
\end{scope}
\begin{scope}[rotate=-120, yshift=4cm]
\node[draw,color=black,fill,diamond,label=0*180/4+60:$B$](b0x13) at (0*180/4:1.5cm) {};
\node[draw,color=black,fill,circle,label=1*180/4+60:$G$](g1x13) at (1*180/4:1.5cm) {};
\node[draw,diamond,fill,label=-120+2*180/4:$b^1_{x_px_r}$,color=black,label=2*180/4+60:$B$](b1x13) at (2*180/4:1.5cm) {};
\node[draw,color=black,fill,circle,label=3*180/4+60:$G$](g2x13) at (3*180/4:1.5cm) {};
\node[draw,diamond,fill,label=-120+4*180/4:$b^2_{x_px_r}$,color=black,label=4*180/4+60:$B$](b2x13) at (4*180/4:1.5cm) {};
\draw[->, line width=1pt] (ug)--(b0x13);
\draw[->, line width=1pt] (b0x13)--(g1x13);
\draw[loosely dashed, line width=1pt] (g1x13)--(b1x13);
\draw[->, line width=1pt] (b1x13)--(g2x13);
\draw[loosely dashed, line width=1pt] (g2x13)--(b2x13);
\draw[loosely dashed, line width=1pt] (b2x13)--(ug);
\end{scope}
\begin{scope}[rotate=120, yshift=4cm]
\node[draw,color=black,fill,diamond,label=0*180/4+300:$B$](b0x23) at (0*180/4:1.5cm) {};
\node[draw,color=black,fill,circle,label=1*180/4+300:$G$](g1x23) at (1*180/4:1.5cm) {};
\node[draw,diamond,fill,label=120+2*180/4:$b^1_{x_qx_r}$,color=black,label=2*180/4+300:$B$](b1x23) at (2*180/4:1.5cm) {};
\node[draw,color=black,fill,circle,label=3*180/4+300:$G$](g2x23) at (3*180/4:1.5cm) {};
\node[draw,diamond,fill,label=120+4*180/4:$b^2_{x_qx_r}$,color=black,label=4*180/4+300:$B$](b2x23) at (4*180/4:1.5cm) {};
\draw[->, line width=1pt] (ug)--(b0x23);
\draw[->, line width=1pt] (b0x23)--(g1x23);
\draw[loosely dashed, line width=1pt] (g1x23)--(b1x23);
\draw[->, line width=1pt] (b1x23)--(g2x23);
\draw[loosely dashed, line width=1pt] (g2x23)--(b2x23);
\draw[loosely dashed, line width=1pt] (b2x23)--(ug);
\end{scope}
\draw[loosely dashed, line width=1pt] (b1x12)--(b2x13) node[midway,color=black,fill,circle,label=30:$G$](g0x1){};
\draw[loosely dashed, line width=1pt] (b1x13)--(b2x23) node[midway,color=black,fill,circle,label=-90:$G$](g0x3){};
\draw[loosely dashed, line width=1pt] (b1x23)--(b2x12) node[midway,color=black,fill,circle,label=150:$G$](g0x2){};
\node[draw,color=black,fill,diamond,label=0*60:$B$](b0x1) at (0*60:2cm) {};
\node[draw,color=black,fill,circle,label=1*60:$G$](g1x2) at (1*60:2cm) {};
\node[draw,color=black,fill,diamond,label=2*60:$B$](b0x2) at (2*60:2cm) {};
\node[draw,color=black,fill,circle,label=3*60:$G$](g1x3) at (3*60:2cm) {};
\node[draw,color=black,fill,diamond,label=4*60:$B$](b0x3) at (4*60:2cm) {};
\node[draw,color=black,fill,circle,label=5*60:$G$](g1x1) at (5*60:2cm) {};
\draw[->, line width=1pt] (b0x1)--(g0x1);
\draw[->, line width=1pt] (g1x1)--(b0x1);
\draw[loosely dashed, line width=1pt] (b2x23)--(g1x1);
\draw[->, line width=1pt] (b0x2)--(g0x2);
\draw[->, line width=1pt] (g1x2)--(b0x2);
\draw[loosely dashed, line width=1pt] (b2x13)--(g1x2);
\draw[->, line width=1pt] (b0x3)--(g0x3);
\draw[->, line width=1pt] (g1x3)--(b0x3);
\draw[loosely dashed, line width=1pt] (b2x12)--(g1x3);
\end{scope}
\end{tikzpicture}}
\caption{$C_{x_px_qx_r}$}
\label{fig:Cx1x2x3}
\end{figure}
\begin{claim} \label{claim:nice_representation_of_equivalence}
In any mapping of $(C_{x_px_qx_r},c)$ to $(C_{48},c)$, an odd number of $(C_{x_ix_j},c)$ is mapped to $(C_{48},c)$ by a $\rho$-mapping. Furthermore, for any choice of an odd number of $(C_{x_ix_j},c)$ (that is either one or all three of them), there exists a mapping of $(C_{x_px_qx_r},c)$ to $(C_{48},c)$ which induces a $\rho$-mapping exactly on our choice.
\end{claim}
\emph{Proof of claim}
Indeed, each $(C_{x_ix_j},c)$ can be mapped to $(C_{48},c)$ only by $\sigma$ or $\rho$, which implies that there are eight ways to map the union of $(C_{x_px_q},c)$, $(C_{x_px_r},c)$ and $(C_{x_qx_r},c)$ to $(C_{48},c)$. Of these eight ways, four map an odd number of $(C_{x_ix_j},c)$ to $(C_{48},c)$ by a $\rho$-mapping. The four remaining ways are to map all $(C_{x_ix_j},c)$ to $(C_{48},c)$ by a $\sigma$-mapping, or to choose one of them to map by a $\sigma$-mapping and to map the two others by a $\rho$-mapping. One can check easily that the union of $(C_{x_px_q},c)$, $(C_{x_px_r},c)$, $(C_{x_qx_r},c)$ and the tree of Figure~\ref{fig:treeCx1x2x3} has six ways to be mapped to $(C_{48},c)$. Indeed, it is no longer possible to map all $(C_{x_ix_j},c)$ by $\sigma$ nor to map $(C_{x_px_r},c)$ by $\sigma$ and $(C_{x_px_q},c)$ and $(C_{x_qx_r},c)$ by $\rho$. By symmetry, this implies Claim~\ref{claim:nice_representation_of_equivalence}.~{\tiny ($\Box$)}
Finally, to complete the gadget, what remains is to forbid the possibility of a $\rho$-mapping for all three of $(C_{x_px_q},c)$, $(C_{x_px_r},c)$ and $(C_{x_qx_r},c)$ in the case where $(x_px_qx_r)$ is a clause in $C$. This is done by adding a $b^1_{x_px_q}b^2_{x_qx_r}$-path shown in Figure~\ref{fig:partialGadgetForClause}.
\begin{figure}[ht!]
\centering
\scalebox{1}{\begin{tikzpicture}[join=bevel,inner sep=0.5mm,scale=1]
\begin{scope}
\draw node[color=black,fill,diamond,label=-90:$b^1_{x_px_q}$,label=90:$B$](b1x1x2) at (0,0) {};
\draw node[color=black,fill,diamond,label=-90:$b^2_{x_qx_r}$,label=90:$B$](b2x2x3) at (4,0) {};
\draw node[color=black,fill,circle,label=90:$G$](x1) at (1,0) {};
\draw node[color=black,fill,diamond,label=90:$B$](x2) at (2,0) {};
\draw node[color=black,fill,circle,label=90:$G$](x3) at (3,0) {};
\draw[->, line width=1pt] (b1x1x2)--(x1);
\draw[->, line width=1pt] (x1)--(x2);
\draw[->, line width=1pt] (x2)--(x3);
\draw[loosely dashed, line width=1pt] (x3)--(b2x2x3);
\end{scope}
\end{tikzpicture}}
\caption{Partial clause gadget.}
\label{fig:partialGadgetForClause}
\end{figure}
Let $f(X,C)$ the final gadget we have just built. Assuming that there are $v$ variables and $c$ clauses, the $4$-tropical graph $f(X,C)$ has $1+53\times v^2+132\times v^3+33\times c$ vertices. To complete our proof we want to prove the following.
$(X,C)$ is a YES instance of NAE $3$-SAT if and only if the $4$-tropical graph $f(X,C)$ admits a homomorphism to $(C_{48},c)$.
It follows directly form our construction that if $f(X,C) \to (C_{48},c)$, then $(X,C)$ is a YES instance of \textsc{NAE $3$-SAT}.
We need to show that if $(X,C)$ is a YES instance, then there exists a homomorphism of $f(X,C)$ to $(C_{48},c)$.
Let $(X,C)$ be a YES instance of \textsc{NAE $3$-SAT}. There exists a partition $p: X \rightarrow \{A,B\}$ such that every clause in $C$ is not fully included in $A$ or $B$. We build a homomorphism of $f(X,C)$ to $(C_{48},c)$ in the following way. $U_G$ is mapped to $g_0$. For each pair of variables $x_i, x_j \in X$, we map $C_{x_ix_j}$ by a $\rho$-mapping if and only if $p(x_i)=p(x_j)$, and by a $\sigma$-mapping otherwise. For every triple of variable $x_p, x_q, x_r \in X$, there is an odd number of pairs $x_i,x_j$ of variables in $\{x_p, x_q, x_r\}$ such that $p(x_i)=p(x_j)$. It follows from Claim \ref{claim:nice_representation_of_equivalence} that one can extend the mapping to any $C_{x_px_qx_r}$. Moreover, as two such structures only intersect on $C_{x_ix_j}$, we can extend the mapping to every $C_{x_px_qx_r}$. It only remains to map the $b^1_{x_px_q}b^2_{x_qx_r}$-path added for the clause, shown in Figure~\ref{fig:partialGadgetForClause}. If $(x_p, x_q, x_r)$ is a clause in $C$, then $p(x_p) \neq p(x_q)$ or $p(x_q) \neq p(x_r)$. It follows that $C_{x_px_q}$ or $C_{x_qx_r}$ is mapped by a $\sigma$-mapping, in which case the $b^1_{x_px_q}b^2_{x_qx_r}$-path shown in Figure~\ref{fig:partialGadgetForClause} can also be mapped. We have shown that there is a homomorphism of $f(X,C)$ to $(C_{48},c)$. This concludes the proof.
\end{proof}
We observe that the proof could be slightly modified to obtain variations of Theorem~\ref{thm:C48}.
\begin{remark}\label{rem:C48}\
\begin{enumerate}
\item In the reduction from Theorem~\ref{thm:C48}, Red vertices are never in the same part of the bipartition as Blue and Green vertices. It follows that one could colour every Red vertex Blue, and Theorem \ref{thm:C48} would still hold, for $3$-tropical cycles.
\item The idea of this proof can also be extended for a $2$-tropical $54$-cycle. To do this we first insert a Red vertex between $x_5$ and $x_6$ in $P_{xy}$ and a Red vertex between $z_5$ and $z_6$ in $Q_{zt}$. We observe that the proof follows similarly. However, in this case, all blue vertices are in one part and all green vertices are on the other part of the bipartition. Thus, as in the previous claim, we can remove two colours now and use the natural bipartition to distinguish two sets of colours for each colour class.
\end{enumerate}
\end{remark}
\longpaper{
\subsection{Other families of minimal graphs}
Next, we show that for each of the minimal graphs $H$ from Table~\ref{table} (other than even cycles) that make \PBlistCOL{$H$} NP-complete, \PBtropCOL{$H$} is polynomial-time solvable.
\begin{theorem}\label{thm:table}f
For every graph $H$ belonging to one of the six families $\mathcal T$, $\mathcal W$, $\mathcal D$, $\mathcal M$, $\mathcal N$ and $\mathcal G$ described in Table~\ref{table}, \PBtropCOL{$H$} is polynomial-time solvable.
\end{theorem}
\begin{proof}
We assume for contradiction, that for some integer~$i$ and a family $\mathcal F$ among $\mathcal T$, $\mathcal W$, $\mathcal D$, $\mathcal M$, $\mathcal N$ and $\mathcal G$, there is a problem \PBCOL{$(F_i,c)$} that is not polynomial-time solvable.
\medskip
\noindent\textbf{Family \boldmath{$\mathcal{T}$}.} Suppose $x_{i+4}$ is coloured~$m$. Suppose $y_{i+3}$ is coloured~$a$, $a\neq m$ by Proposition~\ref{prop:bipartite-hom}. Then, $y_{i+4}$ cannot be coloured~$a$ (otherwise it can be folded onto $y_{i+3}$), so it is coloured~$b$. Because of Lemma~\ref{lemm:unique-feature}, there must be another $P_3$ coloured $amb$ on the graph, but for the graph to be a core, the vertex coloured $m$ of this $P_3$ must not be adjacent to $y_{i+3}$. However, note that $y_{i+3}$ is adjacent to every vertex of $X$ except for $x_1$ and $x_{i+3}$, both of which have degree~$1$ and cannot create a $3$-vertex path coloured $a$-$m$-$b$.
\medskip
\noindent\textbf{Family \boldmath{$\mathcal{W}$}.} Now, we consider $W_i$. We try to find a colouring $c$ of $W_i$ such that \PBCOL{$(W_i,c)$} is not polynomial-time solvable. Suppose $y_{i+2}$ is coloured with colour $a$. Suppose $x_{i+3}$ is coloured $m$. $y_{i+4}$ cannot be coloured~$a$, otherwise it can be folded onto $y_{i+2}$, so we may assume it is coloured~$b$. $y_{i+3}$ cannot be coloured~$b$, for otherwise $y_{i+4}$ can be folded onto it, so it is coloured~$a$ or~$d$. Suppose first that it is coloured~$d$. By Lemma~\ref{lemm:unique-feature}, there is another vertex coloured~$a$, and the only one which could not be folded onto $y_{i+2}$ is $y_{i+1}$, so it must be coloured~$a$. Similarly, $y_{1}$ is coloured~$d$. By Lemma~\ref{lemm:unique-feature}, there is another edge besides $x_{i+3},y_{i+4}$ with endpoints coloured~$m$ and~$b$, but for the graph to be a core, the edge $x_{i+3}y_{i+4}$ must not be able to fold onto it. However, it is easily verified that this is impossible. So, we must assume that $y_{i+3}$ is coloured~$a$. There is no other vertex in $Y$ coloured~$a$, otherwise it can be folded onto $y_{i+2}$ or $y_{i+3}$. We can assume, without loss of generality, that a connected subgraph of the source graph, coloured only with~$m$ and~$b$, and with only vertices of colour~$a$ at distance~$1$, will be sent to $x_{i+3}$ and $y_{i+4}$. Knowing this, we can contract each such subgraph to a single vertex, coloured with a new colour~$\omega$, and similarly replace $x_{i+3}$ and $y_{i+4}$ by a single vertex coloured~$\omega$, adjacent to $y_{i+2}$ and $y_{i+3}$. There will be a homomorphism between the source graph and $(W_i,c)$ if and only if there is one after such transformation. However, the graph obtained after such transformation will not contain any induced subgraph from the table above, which yields a contradiction.
\medskip
\noindent\textbf{Family \boldmath{$\mathcal{D}$}.} Now, consider $D_i$ and a colouring $c$ such that \PBCOL{$(D_i,c)$} is not polynomial-time solvable. Suppose $x_{i+4}$ is coloured~$m$ and $y_{i+4}$ is coloured~$a$. By Lemma~\ref{lemm:unique-feature}, there is another vertex coloured~$a$. We may assume that $y_1$ is such a vertex because it is the only one that cannot be folded on $y_{i+4}$. Then, $x_1$ cannot have colour~$m$, for otherwise it can be folded onto $x_{i+4}$, so it is coloured~$l$. By Lemma~\ref{lemm:unique-feature}, there is another vertex in $X$ coloured~$l$, say $v$. $y_1x_1$ can be folded onto $y_{i+4}v$, which yields a contradiction.
\medskip
\noindent\textbf{Family \boldmath{$\mathcal{M}$}.} Now, consider $M_i$ and a colouring $c$ such that \PBCOL{$(M_i,c)$} is not polynomial-time solvable. Suppose $x_2$ is coloured~$m$ and $y_{i+2}$ is coloured~$a$. By Lemma~\ref{lemm:unique-feature}, there is another vertex in $X$ coloured~$m$. The only vertex which can be coloured~$m$ without being able to be folded onto $x_2$ is $x_1$. This is because $y_{i+2}$ is adjacent only to $x_1$ and $x_2$ is adjacent to every vertex in $Y$ except for $y_{i+2}$. So we may assume $x_1$ is coloured~$m$. By Lemma~\ref{lemm:unique-feature}, there is another vertex in $Y$ coloured~$a$, say $v$. $x_1y_{i+2}$ can be folded onto $x_2v$ since $x_2$ is adjacent to every vertex in $Y$ except $y_{i+2}$, which yields a contradiction.
\medskip
\noindent\textbf{Family \boldmath{$\mathcal{N}$}.} Now, consider $N_i$ and a colouring $c$ such that \PBCOL{$(N_i,c)$} is not polynomial-time solvable. Suppose $x_2$ is coloured $m$. For $3\leq j\leq 2i+3$, $x_j$ cannot be coloured~$m$, for otherwise it can be folded onto $x_2$ since $N(x_j) \subset N(X_2)$. By Lemma~\ref{lemm:unique-feature}, $x_1$ or $x_{2i+4}$ must be coloured~$m$. Both $x_1$ and $x_{2i+4}$ have a neighbour of degree~$1$ (namely, $y_{i+1}$ and $y_{2i+4}$, respectively), which are the two only vertices in $Y$ not adjacent to $x_2$. By Lemma~\ref{lemm:unique-feature}, neither $x_1y_{i+1}$ nor $x_{2i+4}y_{2i+4}$ can be an edge of unique colour. Either exactly one of them is coloured~$ma$ and a neighbour $v$ of $x_2$ is coloured~$a$, in which case the graph is not a core because the edge can be folded on $x_2v$ (since $N(x_1) \setminus \{y_{i+1}\}$ and $N(x_{2i+4}) \setminus \{y_{2i+4}\}$ are both subsets of $N(x_2)$), or both $x_1y_{i+1}$ and $x_{2i+4}y_{2i+4}$ are coloured $ma$ and the graph is not a core because $x_{2i+4}y_{2i+4}$ can be folded on $x_1y_{i+1}$ since $N(x_{2i+4}) \setminus \{y_{2i+4} \} \subset N(x_1)$, yielding a contradiction.
\medskip
\noindent\textbf{Family \boldmath{$\mathcal{G}$}.} We try to find a colouring $c$ of $G_1$ such that \PBCOL{$(G_1,c)$} is not polynomial-time solvable. The colour of $y_1$ is, say, $a$. By Lemma~\ref{lemm:unique-feature}, colour~$a$ must be present somewhere else in $Y$. By symmetry, we can assume $y_2$ is coloured~$a$. The two neighbours of $y_2$ cannot be coloured with the same colour, for otherwise we can fold $x_2$ on $x_1$, implying that $(G_1,c)$ is not a core, a contradiction. Without loss of generality, $x_1$ and $x_2$ are coloured~$1$ and~$2$ respectively. By Lemma~\ref{lemm:unique-feature} applied to edge $y_2x_2$, there must be another edge coloured~$a2$. However, if a neighbour of $y_1$ is coloured~$2$, we can fold $y_2x_2$ onto $y_1$ and the graph is not a core, a contradiction. It follows that the other edge coloured~$a2$ is either $y_3x_4$ or $y_4x_6$. By symmetry, we can assume that $x_4$ is coloured~$2$ and $y_3$ is coloured~$a$. $x_3$ cannot be coloured~$1$ or~$2$, for otherwise $(G_1,c)$ is not a core. Therefore, $x_3$ is coloured with a third colour, say~$3$. At this point, $y_1x_1y_2x_2$ is coloured $a1a2$ and $y_1x_3y_3x_4$ is coloured $a3a2$. Consider the colour of $y_4$. It must be $a$ by Lemma~\ref{lemm:unique-feature}. There are only two uncoloured vertices, $x_5$ and $x_6$, which must be coloured~$1$ and~$3$ by Lemma~\ref{lemm:unique-feature}. The graph is not a core in both cases as we can either fold $x_6y_4$ onto $x_1y_1$ or the edge $x_6y_4$ onto $x_3y_1$, a contradiction.
Now, let $c$ be a colouring of $G_2$ such that \PBCOL{$(G_2,c)$} is not polynomial-time solvable. Suppose the vertex $y_2$ is coloured with $a$. Then, $y_1$ cannot be coloured~$a$, for otherwise it can be folded onto $y_2$, which yields a contradiction. Therefore, $y_1$ is coloured~$b$. Because of Lemma~\ref{lemm:unique-feature}, $y_3$ and $y_4$ must be coloured~$a$ and~$b$. By symmetry, we may assume $y_3$ is coloured~$a$ and $y_4$ is coloured~$b$. Suppose $x_5$ is coloured~$m$. Then $x_1$, $x_2$, $x_3$ and $x_4$ cannot be coloured~$m$, for otherwise $y_3x_5$ can be folded on $y_2$. Thus, $y_3x_5$ is the only edge coloured~$am$. Lemma~\ref{lemm:unique-feature} yields a contradiction.
Now, let $c$ be a colouring of $G_3$ such that \PBCOL{$(G_3,c)$} is not polynomial-time solvable. By Lemma~\ref{lemm:unique-feature}, there are at most two colours in each part of the bipartition. If $x_1$ and $x_2$ have the same colour, $x_2$ can be folded onto $x_1$, a contradiction. Similarly, if $y_1$ and $y_4$ have the same colour, $y_1$ can be folded onto $y_4$. Then $x_1$, $y_1$, $x_2$ and $y_4$ induce a complete bipartite graph with every colour of $c$, implying that $(G_3,c)$ is not a core, a contradiction.
\end{proof}
\section{Bipartite graphs of small order}\label{sec:smallgraphs}
In this section, we show that for each graph $H$ of order at most~$8$, \PBtropCOL{$H$} is polynomial-time solvable. On the other hand, there is a graph $H_9$ of order~$9$ such that \PBtropCOL{$H_9$} is NP-complete.
\begin{theorem} \label{thm:small_graph}
For any bipartite graph $H$ of order at most~$8$, \PBtropCOL{$H$} is polynomial-time solvable.
\end{theorem}
\begin{proof}
It suffices to prove that for each bipartite graph $H$ of order at most~$8$ and each colouring $c$ of $H$, \PBCOL{$(H,c)$} is polynomial-time solvable. In fact, by Proposition~\ref{prop:bipartite-hom} it suffices to show the statement for colourings of $H$ such that the colour sets in the two parts of the bipartition are disjoint. To prove that \PBCOL{$(H,c)$} is polynomial-time solvable it is enough to prove it for the core of $S(H,c)$, it is also enough to prove it for each connected component of $(H,c)$. Thus in the rest of the proof we always assume that $(H,c)$ is connected core. Let $(X,Y)$ be the bipartition of $H$.
Since the only graphs of order at most~$8$ in the characterization of minimal NP-complete graphs $H$ with \PBlistCOL{$H$} NP-complete are the cycles $C_6$ and $C_8$~\cite{FHH99} \longpaper{(see Table~\ref{table})}, by Theorem~\ref{thm:listhom-table}, if $H$ does not contain an induced $6$-cycle or an induced $8$-cycle, then \PBlistCOL{$H$} is polynomial-time solvable and therefore \PBtropCOL{$H$} is polynomial-time solvable. Therefore $H$ contains an induced $6$-cycle or an induced $8$-cycle.
If $H$ contains an induced copy of $C_8$, then $H$ is isomorphic to $C_8$ itself and hence we are done by Theorem~\ref{thm: C12}. Therefore, we can assume that $H$ contains an induced copy of $C_6$. Again by Theorem~\ref{thm: C12}, if $H$ is isomorphic to $C_6$, we are done.
Now, assume that $H$ is a bipartite graph of order~$7$ or~$8$ with an induced copy of $C_6$. If one part, say $X$, is of order~$3$, then all its vertices belong to each $6$-cycle of $H$. Hence, for each $x\in X$, \PBlistCOL{($H-x$)} is polynomial-time solvable. Thus, if $X$ is not monochromatic, we can apply Lemma~\ref{lemm:unique-feature} and \PBCOL{$(H,c)$} is polynomial-time solvable. Therefore we may assume $X$ is monochromatic, say Blue. If $Y$ contains at most two colours, then $(H,c)$ contains as a subgraph the path on three vertices where the central vertex is Blue and the other vertices are coloured with the colours of $Y$.But then $(H,c)$ maps to this subgraph and, therefore, it is not a core, a contradiction. Hence, $Y$ contains at least three colours. If $|Y|=4$, then $Y$ contains two colours that are the unique ones coloured with their colour. Moreover, \PBlistCOL{$(H-\{x,y\})$} contains no $6$-cycle and, therefore, by Lemma~\ref{lemm:unique-feature} \PBCOL{$(H,c)$}, is polynomial-time solvable. Hence we can assume that $|Y|=5$. If $Y$ contains at least four colours, by the same argument we are done, therefore, we assume that $Y$ contains exactly three colours. If $(H,c)$ contains a star with a Blue centre and a three leaves of different colours, then $(H,c)$ is not a core. Therefore the neighbourhood of each vertex of $X$ contains at most two colours.
Assume that the three vertices $y_1$, $y_2$, $y_3$ of $Y$ in the $6$-cycle have three different colours. Let $y_4$, another element of $Y$ be of the same colour as $y_i$. By the previous observation, $y_4$ can only be adjacent to neighbours of $y_i$. But then mapping $y_4$ to $y_1$ is a homomorphism which means $(H,c)$ is not a core.
Therefore, we can assume that $c(y_1)=c(y_2)=1$ and $c(y_3)=2$. Then, the vertex coloured~$3$ has degree~$1$ and is adjacent to the common neighbour of $y_1$ and $y_2$. But then again, $(H,c)$ is not a core.
Therefore, $H$ is a bipartite graph of order~$8$ and $|X|=|Y|=4$. If there are at least three colours in one part of the bipartition (say $X$), then two vertices $x_1$, $x_2$ in $X$ form two colour classes of size~$1$. Moreover, $H-\{x_1,x_2\}$ has no $6$-cycle and therefore, by Lemma~\ref{lemm:unique-feature}, \PBCOL{$(H,c)$} is polynomial-time solvable. We may then assume that each part of the bipartition contains at most two colours. If one part, say $X$, contains exactly one colour (say Blue), then $(H,c)$ contains a path on three vertices with every colour of $c$ (the central vertex is Blue) and is not a core, a contradiction. Therefore each part of the bipartition contains exactly two colours. If in each part, each colour has exactly two vertices, we can apply Lemma~\ref{lemm:2SAT} to show that \PBCOL{$(H,c)$} is polynomial-time solvable. Therefore, we can assume that there is a colour, say Blue, where exactly three vertices of one part, say $x_1$, $x_2$, $x_3$ from part $X$, coloured Blue ($x_4$ is coloured Green). If $H-x_4$ contains no induced $6$-cycle (it cannot contain an $8$-cycle since it has order~$7$), then \PBlistCOL{$(H-x_4)$} is polynomial-time solvable and we can use Lemma~\ref{lemm:unique-feature} and \PBCOL{$(H,c)$} is polynomial-time solvable. Hence we may assume $H-x_4$ contains an induced $6$-cycle $C$. Note that $C$ must contain three vertices of $X$ and therefore contains all three of $x_1$, $x_2$, $x_3$. If the three other vertices $y_1$, $y_2$ an $y_3$ of $C$ are coloured with the same colour, then $(H,c)$ is not a core, a contradiction. Therefore assume, without loss of generality, that $c(y_1)=c(y_2)=1$ and $c(y_3)=2$. Then, in order for $(H,c)$ to be a core, we cannot have both $x_1$ and $y_1$ (respectively, $y_2$ and $x_3$) of degree~$3$. More precisely, either $d(y_1)=d(x_3)=2$ and $d(x_1)=d(y_2)=3$, or $d(y_1)=d(x_3)=3$ and $d(x_1)=d(y_2)=2$. In both cases, we have $d(y_3)=2$, for otherwise $(H,c)$ contains a $4$-cycle with all four colours, and $(H,c)$ is not a core. If $c(y_4)=1$, then $(H,c)$ contains a path on four vertices coloured $2$-Blue-$1$-Green; moreover there is no edge in $(H,c)$ whose endpoints are coloured Green and $2$, therefore $(H,c)$ is homomorphic to the above path and is not a core. If $c(y_4)=2$, then $(H,c)$ contains a $4$-coloured $4$-cycle and again $(H,c)$ is not a core, a contradiction. As no such tropical graph exists, we have shown that for all possible cases the $(H,c)$-colouring problem is polynomial-time solvable.
\end{proof}
Denote by $H_9$ the graph obtained from a $6$-cycle by adding a pendant degree~$1$-vertex to three independent vertices (see Figure~\ref{fig:G9}).
\begin{figure}[ht!]
\centering
\includegraphics[scale=0.7]{9graph.pdf}
\caption{The $4$-tropical graph $H_9$.}
\label{fig:G9}
\end{figure}
\begin{theorem}\label{thm:H_9}
\PBtropCOL{$H_9$} is NP-complete.
\end{theorem}
\begin{proof}
We show that \PBCOL{$(H_9,c)$} is NP-complete, where $c$ is the $4$-colouring of $H_9$ illustrated in Figure~\ref{fig:G9}. We describe a reduction from \PBlistCOL{$C_6$}, which is NP-complete~\cite{FHH99}. We label the vertices in $C_6$ from~$1$ to~$6$ sequentially. We also do that in the $C_6$ included in $H_9$. We assume without loss of generality that the vertex adjacent to the Red vertex is labelled~$1$, and the one adjacent to the Green one is labelled~$3$. It follows that the vertex adjacent to the Yellow vertex is labelled~$5$.
Let $(G,L)$ be an instance of \PBlistCOL{$C_6$}, where $L$ is the list-assignment function. If $G$ is not bipartite, then $G$ has no homomorphism to $C_6$, so we can assume that $G$ is bipartite. Since $G$ and $C_6$ are bipartite, we may assume that $\forall u \in V(G)$, either $L(u)\subseteq\{1,3,5\}$, or $L(u)\subseteq\{2,4,6\}$. Thus $|L(u)|\leq 3$.
From $(G,L)$, we build an instance $f(G,L)$ of \PBCOL{$(H_9,c)$} as follows. First, we consider a copy $G'$ of $G$, we let $G'\subset f(G,L)$ and colour every vertex of $G'$ Black. We call $u'$ the copy of vertex $u$ in $G'$. Then, for each vertex $u$ of $G$, we add a gadget $H_u$ to $f(G,L)$ that is attached to $u'$. The gadget is described below and depends only on $L(u)$.
\begin{itemize}
\item If $L(u)=\{1\}$ (respectively, $\{3\}$ or $\{5\}$), then $H_u$ is a single Red (respectively, Green or Yellow) vertex of degree~$1$ adjacent only to $u'$.
\item If $L(u)=\{2\}$ (respectively, $\{4\}$ or $\{6\}$), then $H_u$ consists of two $2$-vertex path: a Red--Black path and a Green--Black path (respectively, a Green--Black path and a Yellow--Black path or a Yellow--Black path and a Red--Black path) whose Black vertex is of degree~$2$ and is adjacent to $u'$ (the other vertex is of degree~$1$).
\item If $L(u)=\{2,4\}$ (respectively, $\{4,6\}$ or $\{2,6\}$), then $H_u$ is a $2$-vertex Green--Black (respectively, Yellow--Black or Red--Black) path whose Black vertex is of degree~$2$ and adjacent to $u'$ (the other vertex is of degree~$1$).
\item If $L(u)=\{1,3\}$ (respectively, $\{3,5\}$ or $\{1,5\}$), then $H_u$ is a $5$-vertex Red--Black--Black--Black--Green path (respectively, Green--Black--Black--Black--Yellow or Yellow--Black--Black--Black--Red) whose middle Black vertex is of degree~$3$ and adjacent to $u'$ (the endpoints of the path are of degree~$1$ and the other two vertices have degree~$2$).
\item If $L(u)=\{1,3,5\}$, then $H_u$ is a $3$-vertex Black--Black--Red path with the black leaf adjacent to $u'$.
\item If $L(u)=\{2,4,6\}$, then $H_u$ is a $4$-vertex Black--Black--Black--Red path with the black leaf adjacent to $u'$.
\end{itemize}
Let us prove that $G$ has a homomorphism to $C_6$ that fulfils the constraints of list $L$, if and only if $f(G,L)\to (H_9,c)$.
For the first direction, consider a list homomorphism $h$ of $G$ to $C_6$ with the list function $L$. We build a homomorphism $h'$ of $f(G,L)$ to $(H_9,c)$ as follows. First of all, each copy $v'$ of a vertex $v$ of $G$ with $h(v)=i$ is mapped to $i$ in $(H_9,c)$. It is clear that this defines a homomorphism of the subgraph $G'$ of $f(G,L)$ to the Black $6$-cycle in $(H_9,c)$. It is now easy to complete $h'$ into a homomorphism of $f(G,L)$ to $(H_9,c)$ by considering each gadget $H_u$ independently.
For the converse, let $h_T$ be a homomorphism of $f(G,L)$ to $(H_9,c)$. Then, we claim that the restriction of $h_T$ to the vertices of the subgraph $G'$ of $f(G,L)$ is a list homomorphism of $G$ to $C_6$ with list function $L$. Indeed, let $u'$ be a vertex of $G'$. If $H_u$ has one vertex (say a Red vertex), then $L(u)=\{1\}$. Then necessarily $u'$ is sent to a neighbour of a vertex coloured Red in $(H_9,c)$. Since the only such neighbour is vertex~$1$, $u'\in h_T(u)$. All the other cases follow from similar considerations.
\end{proof}
\section{Trees}\label{sec:trees}
We now consider the complexity of tropical homomorphism problems when the target tropical graph is a tropical tree
It follows from the results in Section~\ref{sec:list_hom} that for every tree $T$ of order at most~$10$, \PBtropCOL{$T$} is polynomial-time solvable. Indeed, such a tree needs to contain a minimal tree $T$ of order at most~$10$ for which \PBlistCOL{$T$} is NP-complete, and the only such tree is $G_1$, which has order~$10$~\cite{FHH99}. \longpaper{(See Table~\ref{table}.)} We proved in Theorem~\ref{thm:table} that \PBtropCOL{$G_1$} is polynomial-time solvable. With some efforts, one can extend this to trees of order at most~$11$.
\shortpaper{The proof is tedious and we omit it here, see~\cite{fullversion} for details.}
\begin{theorem}\label{thm:SmallTree}
For every tree $T$ of order at most~$11$, \PBtropCOL{$T$} is polynomial-time solvable.
\end{theorem}
\longpaper{
\begin{proof}
Let $G_1$ be the smallest tree such that \PBlistCOL{$G_1$} is NP-hard, as defined in Table~\ref{table} of Section~\ref{sec:list_hom} ($G_1$ has order $10$ and is obtained from a claw by subdividing each edge twice). We let $V(G_1)=\{c,x_1,y_1,z_1,x_2,y_2,z_2,x_3,y_3,z_3\}$, with edges $cx_i$, $x_iy_i$, $y_iz_i$ for $i=1,2,3$
Assume for a contradiction that there is a tree $T_0$ of order~$11$ such that \PBtropCOL{$T$} is not polynomial-time solvable. Then, $T_0$ is a connected core. Once again, by Proposition~\ref{prop:bipartite-hom}, we may assume that the colour sets of the two parts in the bipartition of $T_0$ are disjoint. By Theorem~\ref{thm:listhom-table}, for any tree $T$ which does not contain $G_1$ as an induced subgraph, \PBlistCOL{$T$} is polynomial-time solvable, and therefore \PBtropCOL{$T$} is polynomial-time solvable. Hence $G_1$ is a subtree of $T_0$.
There are four non-isomorphic trees of order~$11$ which contain $G_1$, depending on where we attach the additional vertex $a$. If in $T_0$, $a$ is adjacent to $c$, then the same arguments as in the proof of Theorem~\ref{thm:table} showing that \PBtropCOL{$G_1$} is polynomial-time solvable show that \PBtropCOL{$T_0$} is polynomial-time solvable, a contradiction.
Let $(A,B)$ be the bipartition of $T_0$ with $\{c,y_1,y_2,y_3\}\subseteq A$ and $\{x_1,x_2,x_3,z_1,z_2,z_3\}\subseteq B$. For the remainder, we may assume that no vertex (except $a$) is the only one with its colour, for otherwise, by Lemma~\ref{lemm:unique-feature}, \PBtropCOL{$T_0$} would be polynomial-time solvable. In particular, $A-a$ is coloured with at most two colours and $B-a$ is coloured with at most three colours.
Assume first that $a$ is adjacent to a vertex $x_i$ of $G_1$, say $x_1$. The colours of $x_1$ and $z_1$ must be distinct, otherwise $(T_0,c_0)$ is not a core. Without loss of generality, assume that $c_0(x_1)=1$ and $c_0(z_1)=2$. Without loss of generality the central vertex $c$ is Black. The supplementary vertex $a$ must be coloured with a different colour than $c$ and $y_1$ (say with colour Red), otherwise $(T_0,c_0)$ is not a core. Hence $y_1$ is not Red. Assume first that $y_1$ is Green. Then (without loss of generality), $y_2$ is Black and $y_3$ is Green, otherwise we could apply Lemma~\ref{lemm:unique-feature}. But by Lemma~\ref{lemm:unique-feature}, there must be two edges with endpoints $1$ and Green, and one with endpoints $2$ and Green. Hence $c_0(x_3)=1$ and $c_0(z_3)=2$ (if $c_0(x_3)=2$ and $c_0(z_3)=1$ then $(T_0,c_0)$ is not a core). But again by Lemma~\ref{lemm:unique-feature} we need another edge with endpoints Black and $1$, and one with endpoints Black and $2$. But in both cases $(T_0,c_0)$ is not a core, a contradiction. This shows that vertex $y_1$ must be Black. Then, since $(T_0,c_0)$ is a core, vertex $c$ has no neighbour coloured~$2$. But if there is no second edge with endpoints coloured $2$ and Black, then we could apply Lemma~\ref{lemm:unique-feature}. Hence one of $y_2$ and $y_3$, say $y_2$, must be Black, and $c_0(z_2)=2$. If $c_0(x_2)=1$, $(T_0,c_0)$ is not a core, therefore $c_0(x_2)=3$, and $c_0(x_3)\in\{1,3\}$. If $y_3$ is Black, then $(T_0,c_0)$ is not a core, hence $y_3$ is Red. But both neighbours of $y_3$ must have distinct colours, which means we can apply Lemma~\ref{lemm:unique-feature} to one of the edges incident with $y_3$, a contradiction.
Assume now that $a$ is adjacent to a vertex $y_i$ of $G_1$, say $y_1$. Then, the colours of $a$, $x_1$ and $z_1$ must be distinct, say $c_0(x_1)=1$, $c_0(z_1)=2$, $c_0(a)=3$. Without loss of generality the central vertex $c$ is Black. By Lemma~\ref{lemm:unique-feature}, there is another vertex coloured Black. If $y_1$ is Black, then by Lemma~\ref{lemm:unique-feature} we have two further edges with endpoints Black-$2$ and Black-$3$. But these edges cannot be both incident with $c$ (otherwise $(T_0,c_0)$ is not a core), hence there is another Black vertex. Then in fact, Lemma~\ref{lemm:unique-feature} implies that both $y_2$ and $y_3$ are Black. But then, any way to complete $c_0$ implies that $(T_0,c_0)$ is not a core, a contradiction. Therefore, $y_1$ is not Black (say it is Red) and we can assume that $y_2$ is Black, and since we need a second Red vertex, $y_3$ is Red. But one of the type of edges among Red-$1$, Red-$2$ and Red-$3$ will appear only once, and we can apply Lemma~\ref{lemm:unique-feature}, a contradiction.
We assume finally that $a$ is adjacent to a vertex $z_i$ of $G_1$, say $z_1$. Without loss of generality, vertex $a$ is Black, vertex $z_1$ is coloured~$1$, and vertex $y_1$ is Red (otherwise, $(T_0,c_0)$ is not a core). By Lemma~\ref{lemm:unique-feature}, there must be another $3$-vertex path coloured Black-$1$-Red. This path must be $cx_iy_i$ with $c$ Black, for otherwise $(T_0,c_0)$ is not a core. We can assume that $c_0(x_2)=1$ and $y_1$ is Red. Then $c_0(x_1)\neq 1$, assume $c_0(x_1)=2$. Then again by Lemma~\ref{lemm:unique-feature} there is another $3$-vertex path coloured Black-$2$-Red. The only possibility is that $c_0(x_3)=2$ and $y_3$ is Red. Then $c_0(z_3)\notin\{1,2\}$, otherwise $(T_0,c_0)$ is not a core. Hence we assume $c_0(z_3)=3$, which by Lemma~\ref{lemm:unique-feature} implies $c_0(z_2)=3$. But then there is a unique $3$-vertex path coloured $1$-Red-$3$, and by Lemma~\ref{lemm:unique-feature}, \PBCOL{$(T_0,c_0)$} is polynomial-time solvable, a contradiction. This completes the proof.
\end{proof}
Let $T_{23}$ be the tree of order~$23$ shown in Figure~\ref{target_23_tree}.
\begin{figure}[ht!]
\centering
\includegraphics[scale=0.7]{23tree.pdf}
\caption{The $7$-tropical tree $(T_{23},c)$}
\label{target_23_tree}
\end{figure}
\begin{theorem}\label{thm:T23}
\PBtropCOL{$T_{23}$} is NP-complete.
\end{theorem}
\begin{proof}
We give a reduction from \textsc{$3$-SAT} to \PBCOL{$(T_{23},c)$}, where $c$ is the colouring of Figure~\ref{target_23_tree}. Given an instance $(X,C)$ of \textsc{$3$-SAT}, we construct an instance $f(X,C)=(G_{X,C},c_{X,C})$ of \PBCOL{$(T_{23},c)$}
To construct the graph $G_{X,C}$, we first define the following building blocks. See Figure~\ref{building_block_tree} for illustrations.
\begin{itemize}
\item The block $S_{1,2}$ is a graph built from a $7$-vertex black-coloured path with vertex set $\{x_1,\ldots,x_7\}$ where a BlackCross leaf is attached to vertices $x_1$ and $x_7$, a RedDot leaf is attached to vertices $x_2$ and $x_6$, and a GreenDot leaf is attached to vertex $x_4$.
\item The block $S_{1,T}$ is a graph built from a $7$-vertex black-coloured path with vertex set $\{x_1,\ldots,x_7\}$ where a BlackCross leaf is attached to vertices $x_1$ and $x_7$, a RedDot leaf is attached to vertices $x_2$ and $x_6$, and a RedCross leaf is attached to vertex $x_4$.
\item The block $S_{1,T}$ is a graph built from a $7$-vertex black-coloured path with vertex set $\{x_1,\ldots,x_7\}$ where a BlackCross leaf is attached to vertices $x_1$ and $x_7$, a GreenDot leaf is attached to vertices $x_2$ and $x_6$, and a GreenCross leaf is attached to vertex $x_4$.
\item The \emph{NOT-block} is depicted in Figure~\ref{fig:NOT-block}.
\item The \emph{A-block} is depicted in Figure~\ref{fig:A-block}.
\end{itemize}
Illustrations of these blocks can be found in Figure~\ref{building_block_tree}.
\begin{figure}[ht!]
\centering
\subfigure[The blocks $S_{1,2}$, $S_{1,T}$ and $S_{2,T}$.]{\label{fig:S-blocks}\includegraphics[scale=0.7]{Sblocks.pdf}}\qquad
\subfigure[The variable gadget of $x$, essentially a NOT-block.]{\label{fig:NOT-block}\includegraphics[scale=0.7]{T23_variable.pdf}}\qquad
\subfigure[The $A$-block and its representation as an arrow.]{\label{fig:A-block}\includegraphics[scale=0.7]{Ablock.pdf}}
\caption{The building blocks of $G_{X,C}$.}
\label{building_block_tree}
\end{figure}
We now define gadgets for each variable of $X$ and each clause of $C$. The graph $G_{X,C}$ is formed by the set of all variable and clause gadgets.
\begin{itemize}
\item For a variable $x\in X$, the \emph{variable gadget of $x$} consists of the four vertices $x^0$, $x^1$, $\bar{x}^0$ and $\bar{x}^1$, coloured respectively BlackDot, BlackCross, BlackDot and BlackCross, joined by a NOT-block as described in Figure \ref{fig:NOT-block}. The image of $x^0$ and $x^1$ in $(T_{23},c)$ correspond to the truth-value of the litteral $x$. Similarly, the image of $\bar{x}^0$ and $\bar{x}^1$ correspond to the truth-value of the litteral $\bar{x}$. For a litteral $l$, we use the notation $l^0$ (resp. $l^1$) to describe either $x^0$ (resp. $x^1)$ when $l=x$ with $x\in X$, or $\bar{x}^0$ (resp. $\bar{x}^1$) when $l=\bar{x}$ with $x\in X$.
\item For each clause $c=(l_1,l_2,l_3)\in C$, there is a \emph{clause gadget of $c$} (as drawn in Figure~\ref{gadget_clause_tree}) connecting vertices $l_1^0$, $l_2^0$ and $l_3^0$.
\end{itemize}
\begin{figure}[ht!]
\centering
\includegraphics[scale=0.7]{GadgetClauseTree.pdf}
\caption{Example of a clause gadget of clause $(l_1,l_2,l_3)$. The full details of the $A$-blocks and $S_{1,2}$-blocks are represented in Figure~\ref{building_block_tree}.}
\label{gadget_clause_tree}
\end{figure}
We now show that $G_{X,C}\to (T_{23},c)$ if and only if $(X,C)$ is satisfiable.
\medskip
Assume first that there is a homomorphism $h$ of $G_{X,C}$ to $(T_{23},c)$. We first prove some properties of $h$.
\begin{claim}\label{clm:T_{23}}
The homomorphism $h$ satisfies the following properties.
\begin{itemize}
\item[(1)] For each literal $l$ of a variable of $X$, vertices $l^0$ and $l^1$ are mapped to the two vertices of one of the pairs $T$, $F_1$ or $F_2$. The same holds for the extremities of the blocks $S_{1,2}$, $S_{1,T}$, $S_{2,T}$ and $A$.
\item[(2)] The two extremities of each block $S_{1,2}$ are both mapped either to the vertices of $T$, or to vertices of $F_1\cup F_2$.
\item[(3)] The two extremities of each block $S_{1,T}$ are both mapped either to the vertices of $F_2$, or to vertices of $F_1\cup T$.
\item[(4)] The two extremities of each block $S_{2,T}$ are both mapped either to the vertices of $F_1$, or to vertices of $F_2\cup T$.
\item[(5)] For each variable $x$ of $X$, exactly one of $x^0$ and $\bar{x}^0$ is mapped to a vertex of $T$, and the other is mapped to a vertex of $F_1$ or $F_2$.
\item[(6)] In any $A$-block, either some extremity is mapped to $T$ (then the other extremity can be mapped to any of $F_1$, $F_2$ or $T$), or the left extremity is mapped to $F_2$ and the right extremity, to $F_1$.
\end{itemize}
\end{claim}
\noindent\emph{Proof of claim.}
(1) This is immediate since the only pairs in $(T_{23},c)$ consisting of two adjacent BlackDot and BlackCross vertices are the ones of $T$, $F_1$ and $F_2$.
\smallskip
(2)--(4) We only prove~(2), since the three proofs are not difficult and similar. By~(1), the extremities of $S_{1,2}$ are mapped to vertices of $T\cup F_1\cup F_2$. If one extremity is mapped to $T$, the remainder of the mapping is forced and the claim follows. If one extremity is mapped to $F_1\cup F_2$, one can easily complete it to a mapping where the other extremity is mapped to either $F_1$ or $F_2$.
\smallskip
(5) By (1), $x^0$ and $\bar{x}^0$ must be mapped to a vertex of $T\cup F_1\cup F_2$. Without loss of generality, we can assume that $x^0$ corresponds to the left extremity of the NOT-block $N_x$ connecting $x^0$ and $\bar{x}^0$. First assume that $x^0$ and $\bar{x}^0$ are mapped to the vertex of $T$ coloured BlackDot. Then, considering the vertices of $N_x$ from left to right, the mapping is forced and the degree~$3$-vertex of $N_x$ at distance~$2$ both of a RedDot and a RedCross vertex must be mapped to the vertex $c$ of $T_{23}$. But then, continuing towards the right of $N_x$, $\bar{x}^0$ cannot be mapped to a vertex of $T$. Therefore, we may assume that both $x^0$ and $\bar{x}^0$ are mapped to the BlackCross vertices of $F_1\cup F_2$. If $x^0$ is mapped to the BlackCross vertex in $F_1$, then again going through $N_x$ from left to right the mapping is forced; the central vertex of $N_x$ must be mapped to a vertex of $F_2$, and $\bar{x}^0$ must be mapped to a vertex of $T$, a contradiction. The same applied when $x^0$ is mapped to the BlackCross vertex in $F_2$, completing the proof of~(5).
\smallskip
(6) An $A$-block is composed of two parts: the upper part and the lower part. Observe that if the left extremity of an $A$-block is mapped to $F_1$, then using~(2) and~(4), the mapping of the upper part of the $A$-block is forced and the right extremity has to be mapped to $T$. Similarly, if the left extremity is mapped to $F_2$, by~(2) and~(3) the right extremity cannot be mapped to $F_2$. On the other hand, for all other combinations of mapping the extremities to $T$, $F_1$ or $F_2$ the mapping can be extended.~{\tiny ($\Box$)}
\medskip
We are ready to show how to construct the truth assignment $A(h)$. If $h(l^0)\in T$ for some literal $l$, we let $l$ be True and if $h(l^0)\in F_1\cup F_2$, we let $l$ be False. By Claim~\ref{clm:T_{23}}(5), this is a consistent truth assignment for $X$. For any clause $c=(l_1,l_2,l_3)$, in the clause gadget of $c$, we have three $A$-blocks forming a directed triangle. Hence, by Claim~\ref{clm:T_{23}}(6), there must be one of the three extremities of this triangle mapped to a vertex of $T$. Therefore, by Claim~\ref{clm:T_{23}}(2), at least one of the vertices $l_1^0$, $l_2^0$ and $l_3^0$ is mapped to $T$. This shows that $A(h)$ satisfies the formula $(X,C)$.
\medskip
Reciprocally, if there is a solution for $(X,C)$, one can build a homomorphism of $G_{X,C}$ to $(T_{23},c)$ by mapping, for each literal $l$, the vertices $l_0$ and $l_1$ to one of the vertex pairs $F_1$, $F_2$ and $T$ of $(T_{23},c)$ corresponding to the truth value of $l$ (if $l$ is False, we may choose one of $F_1$ and $F_2$ arbitrarily). Then, using Claim~\ref{clm:T_{23}}, one can easily complete this to a valid mapping.
\end{proof}
\section{Conclusion}
We have shown that the class of \PBCOL{$(H,c)$} problems has a very rich structure, since they fall into the classes of CSPs for which a dichotomy theorem would imply the truth of the Feder--Vardi Dichotomy Conjecture. Hence, we turned our attention to the class of \PBtropCOL{$H$} problems, for which a dichotomy theorem might exist. Despite some initial results in this direction, we have not been able to exhibit such a dichotomy, and leave this as the major open problem in this paper.
Towards a solution to this problem, we propose a simpler question. All bipartite graphs $H$ that we know with problem \PBtropCOL{$H$} being NP-complete contain, as an induced subgraph, either an even cycle of length at least~$6$ (for example cycles themselves or $H_9$), or the graph $G_1$\longpaper{ from Table~\ref{table}}, that is, a claw with each edge subdivided twice (this is the case for $T_{23}$). Hence, we ask the following. (A bipartite graph is \emph{chordal} if it contains no induced cycle of length at least~$6$.)
\begin{question} \label{conj:nocycle}
Is it true that for any chordal bipartite graph $H$ with no induced copy of $G_1$, \PBtropCOL{$H$} is polynomial-time solvable?
\end{question}
Note that Question~\ref{conj:nocycle} is not an attempt at giving an exact classification, since \PBtropCOL{$G_1$} and \PBtropCOL{$C_{2k}$} for $k\leq 6$ are polynomial-time solvable.
\smallskip
Another interesting question would be to consider the restriction of \PBtropCOL{$H$} to $2$-tropical graphs. Recall that by Remark~\ref{rem:C48}(2), one can slightly modify the gadgets from Theorem~\ref{thm:C48} and the colouring of the cycle, to obtain a $2$-colouring $c$ of $C_{54}$ such that \PBCOL{$(C_{54},c)$} is NP-complete.
\medskip
Finally, we relate our work to the \textsc{$(H,h,Y)$-Factoring} problem studied in~\cite{BM97} and mentioned in the introduction. Recall that \PBCOL{$(H,c)$} corresponds to \textsc{$(H,c,K_{|C|}^+)$-Factoring} where $K_{|C|}^+$ is the complete graph on $|C|$ vertices with all loops, and with $C$ the set of colours used by $c$. In~\cite{BM97}, the authors studied \textsc{$(H,h,Y)$-Factoring} when $Y$ has no loops. Using reductions from NP-complete \PBCOL{$D$} problems where $D$ is an oriented even cycle or an oriented tree, they proved that for any fixed graph $Y$ which is not a path on at most four vertices, there is an even cycle $C$ and a tree $T$ such that \textsc{$(C,h_C,Y)$-Factoring} and \textsc{$(T,h_T,Y)$-Factoring} are NP-complete (for some suitable homomorphisms $h_C$ and $h_T$). Note that $C$ and $T$ here are fairly large. We can strengthen these results as follows. Consider our reduction of Theorem~\ref{thm:C48} showing in particular that \PBtropCOL{$C_{48}$} is NP-complete. As noted in Remark~\ref{rem:C48}(1), the given colouring $c$ of $C_{48}$ can easily be made a proper colouring by separating the red vertices into two classes, according to which part of the bipartition of $C_{48}$ they belong to. Then, one can observe that $c$ is in fact a homomorphism to a tree $T_1$ obtained from a claw where one edge is subdivided once (the three vertices of degree~$1$ are coloured Blue, Black and Green, and the two other vertices are the two kinds of Red). Thus, for any graph $Y$ containing this subdivided claw as a subgraph, we deduce that \textsc{$(C_{48},c_{1|T_1},Y)$-Factoring} is NP-complete. We can use a similar approach for our result of Theorem~\ref{thm:T23}, that \PBtropCOL{$T_{23}$} is NP-complete. Note that the colouring $c_2$ we give is in fact a homomorphism to a tree $T_2$ which is obtained from a star with five branches by subdividing one edge once. Thus, for any graph $Y$ containing $T_2$ as a subgraph, \textsc{$(T_{23},c_{2|T_2},Y)$-Factoring} is NP-complete. Of course we can apply this argument by replacing $T_1$ and $T_2$ by the underlying graph of any loop-free homomorphic image of $(C_{48},c_1)$ and $(T_{23},c_2)$, respectively.
\vspace{0.5cm}
\noindent\textbf{Acknowledgements.} We thank Petru Valicov for initial discussions on the topic of this paper.
| 0e150619408a8131638d54f89645527b21d43ee1 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
A major goal of planetary exploration is to obtain a fundamental understanding
of planetary environments, both as they are currently and as they were in the past.
This knowledge
can be used to explore the questions of (a) how conditions for planetary habitability
arose and (b) the origins of life. Titan is a unique object of study in this quest. Other
than Earth itself, and Pluto \citep[which has also been observed to have
photochemically produced haze;][]{stern15,gladstone16},
Titan is the only solar system body demonstrated to have
complex organic chemistry occurring today. Its atmospheric properties---
(1) a thick N$_2$ atmosphere, (2) a reducing
atmospheric composition, (3) energy sources for driving disequilibrium
chemistry and (4) an aerosol layer for shielding the surface from
solar UV radiation---suggest it is a counterpart of the early Earth, before the latter's
reducing atmosphere was eradicated by the emergence and evolution of life
\citep{ct99,lunine05,lm08}.
A significant number of photochemical models have been developed to
investigate the distribution of hydrocarbons in Titan's atmosphere
\citep{strobel74,yap84,lara96,wa04,delahaye08,lavvas08a,lavvas08b,krasnopolsky09}.
Recently, more constraints have been placed on the abundance of
hydrocarbons and nitriles in the mesosphere of Titan (500 -- 1000 km)
from Cassini/UVIS stellar occultations \citep{koskinen11,kammer13}.
In combination with the updated version of Cassini/ CIRS
limb view \citep{vinatier10}, the complete profiles of C$_2$H$_2$, C$_2$H$_4$,
C$_6$H$_6$, HCN, HC$_3$N are revealed for the first time. C$_3$ compounds,
including C$_3$H$_6$, were modeled by \cite{li15}, and the agreement
with observations \citep{nixon13} is satisfactory.
The chemistry of many of these nitrogen molecules has recently been
modeled by \cite{loison15}.
In this paper we introduce our updated Titan chemical model that
includes the formation of such potentially astrobiologically important
molecules as acrylonitrile. In addition to the usual gas phase
chemistry, it also includes a numerically stable treatment of the
condensation and sublimation, allowing the formation and destruction
of ices in the lower atmosphere to be tracked. Haze formation is also
included in a parameterized fashion, allowing for the permanent
removal of molecules from the atmosphere. We present here
the effects of condensation on the nitrogen chemistry. The
interaction of hydrocarbons and nitrile species in the condensed
phase is complex and is beyond the scope of this paper \citep[see, for
example, Figures 1 and 2 of][]{anderson16}.
We begin with describing our updated model and in particular our
treatment of condensation and sublimation (Section~\ref{sec:model}).
We use this updated model to consider the chemistry in Titan's
atmosphere from the surface of the moon to an altitude of 1500 km. We
explore how condensation processes and haze formation affect the
predicted gas phase abundances of observable molecules
(Section~\ref{sec:results}). We also consider where the
condensates form within the atmosphere (Section~\ref{sec:cond}).
Section~\ref{sec:conc} presents our conclusions.
\section{The Model}\label{sec:model}
We use the Caltech/JPL photochemical model \citep[KINETICS;][]{allen81}
with a recently updated
chemical network, and with the addition of condensation and
sublimation processes to explore the atmospheric chemistry of Titan.
The 1-D model solves the mass continuity equation from the surface of
Titan to 1500 km altitude:
\begin{equation}
\frac{\partial n_i}{\partial t} + \frac{\partial \psi_i}{\partial z} = P_i - L_i
\end{equation}
where $n_i$ is the number density of species $i$, and $P_i$ and $L_i$
are its chemical production and loss rates respectively. $\psi_i$ is
the vertical flux of $i$ calculated from
\begin{equation}
\psi_i = - \frac{\partial n_i}{\partial z}(D_i + K_{zz}) - n_i \left( \frac{D_i}{H_i} + \frac{K_{zz}}{H_a} \right) - n_i \frac{\partial T}{\partial z} \frac{(1+\alpha_i)D_i + K_{zz}}{T}
\end{equation}
where $D_i$ and $H_i$ are the molecular diffusion coefficient and the
scale height for species $i$ respectively, $H_a$ is the atmospheric
scale height, $\alpha_i$ is the thermal diffusion coefficient of
species $i$, $T$ is the temperature and $K_{zz}$ is the eddy diffusion
coefficient. The eddy diffusion coefficient used here is taken from
\cite{li15} and can be summarized as
\begin{equation}
log K_{zz}(z) = \left\{
\begin{array}{ll}
log(3 \times 10^3) , & z < z_1\\
\\
log(3 \times 10^3) \dfrac{z_2 - z}{z_2 -z_1} + log(2 \times 10^7) \dfrac{z - z_1}{z_2 - z_1} , & z_1 \leq z < z_2 \\
\\
log(2 \times 10^7) \dfrac{z_3 - z}{z_3 - z_1} + log(2 \times 10^6) \dfrac{z - z_2}{z_3 - z_2} , & z_2 \leq z < z_3\\
\\
log(2 \times 10^6) \dfrac{z_4 - z}{z_4 - z_3} + log(4 \times 10^8) \dfrac{z - z_3}{z_4 - z_3} , & z_3 \leq z < z_4\\
\\
log(4 \times 10^8) , & z \geq z_4
\end{array}
\right.
\end{equation}
\noindent The atmospheric density and temperature profiles are also taken
from \cite{li15}, and are based on the T40 Cassini flyby
\citep{westlake11}.
Aerosols are included in our model, both for the absorption of UV
radiation and to provide surfaces onto which molecules can condense.
The aerosol properties are from \cite{lavvas10} who derived them
from a microphysical model
validated against Cassini/Descent Imager Spectral Radiometer (DISR) observations.
Their results provide the mixing
ratio and surface area of aerosol particles as a function of
altitude (Figure~\ref{fig:aerosol}). To calculate the absorption of UV
by dust we assume absorbing aerosols with
extinction cross-sections that are independent of wavelength
\citep{li14,li15}.
\begin{figure}
\includegraphics[width=0.5\linewidth]{rad.pdf}
\includegraphics[width=0.5\linewidth]{mr.pdf}
\caption{\label{fig:aerosol}Aerosol properties from
\cite{lavvas10} derived from a microphysical model validated against
DISR observations. {\it left:} The mean radius of
particles. {\it right:} The mixing ratio of particles.}
\end{figure}
\subsection{Boundary Conditions}
The lower boundary of our model is the surface of Titan and the upper
boundary is at 1500 km. For H and H$_2$ the flux at the lower
boundary is zero and at the top of the atmosphere these molecules are
allowed to escape with velocities of 2.4
$\times$ 10$^4$ cms$^{-1}$ and 6.1 $\times$ 10$^{3}$
cms$^{-1}$ respectively (equivalent to fluxes of 3.78 $\times$
10$^8$ H atoms cm$^{-2}$ and 6.2 $\times$ 10$^9$
H$_2$ molecules cm$^{-2}$). For all other gaseous species the
concentration gradient at the lower boundary is assumed to be zero,
while they have zero flux at the top boundary.
Observations suggest that CH$_4$ can escape from the top of the
atmosphere by sputtering \citep{delahaye07} but the same effect can be
generated in models by applying a larger eddy diffusivity
\citep{li15,li14,yelle08} which is the approach we have taken here.
Condensed species have zero flux at both the upper and lower boundaries.
Table~\ref{tab:molecules} provides a list of the molecules in
our model. The mixing ratio of N$_2$ is set according to the
observational data and held fixed, with values below 50 km taken from
the Huygens observations \citep{niemann05} and above 1000 km from
Cassini/UVIS data \citep{kammer13}. Between 50 and 1000 km the
mixing ratio is assumed to be 0.98.
The mixing ratio of CH$_4$ is fixed to the observed (super-saturated) values
\citep{niemann10} below the tropopause and allowed to vary above this.
\begin{table}
\begin{tabularx}{\linewidth}{l X}
\hline
Family & Molecule \\
\hline
& H, H$_2$\\
& \\
hydrocarbons & C CH CH$_2$ $^3$CH$_2$ CH$_3$ CH$_4$ C2 C$_2$H C$_2$H$_2$
C$_2$H$_3$ C$_2$H$_4$ C$_2$H$_5$ C$_2$H$_6$ C$_3$
C$_3$H C$_3$H$_2$ C$_3$H$_3$
C$_2$CCH$_2$ CH$_3$C$_2$H C$_3$H$_5$ C$_3$H$_6$
C$_3$H$_7$ C$_3$H$_8$ C$_4$H C$_4$H$_2$
C$_4$H$_3$ C$_4$H$_4$ C$_4$H$_5$ 1-C$_4$H$_6$
1,2-C$_4$H$_6$ 1,3-C$_4$H$_6$ C$_4$H$_8$
C$_4$H$_9$
C$_4$H$_{10}$ C$_5$H$_3$ C$_5$H$_4$ C$_6$H
C$_6$H$_2$ C$_6$H$_3$ C$_6$H$_4$ C$_6$H$_5$
$l$-C$_6$H$_6$
C$_6$H$_6$ C$_8$H$_2$\\
& \\
nitrogen-molecules & N NH NH$_2$ NH$_3$ N$_2$H N$_2$H$_2$
N$_2$H$_3$ N$_2$H$_4$
CN HCN HNC H$_2$CN CHCN CH$_2$CN CH$_3$CN
C$_2$H$_3$CN
C$_2$H$_5$CN C$_3$H$_5$CN C$_2$N$_2$
HC$_2$N$_2$ C$_3$N HC$_3$N HC$_4$N
CH$_3$C$_2$CN H$_2$C$_3$N C$_4$N$_2$ HC$_5$N
C$_6$N$_2$ CH$_2$NH CH$_2$NH$_2$
CH$_3$NH CH$_3$NH$_2$ \\
& \\
condensed molecules & C$_2$H$_2^c$ C$_2$H$_4^c$ C$_2$H$_6^c$ CH$_2$CCH$_2^c$
CH$_3$C$_2$H$^c$
C$_3$H$_6^c$ C$_3$H$_8^c$ C$_4$H$_2^c$ C$_4$H$_4^c$
1-C$_4$H$_6$$^c$ 1,2-C$_4$H$_6$$^c$ 1,3-C$_4$H$_6$$^c$
C$_4$H$_8^c$ C$_4$H$_{10}^c$ C$_5$H$_4^c$
$l$-C$_6$H$_6^c$ C$_6$H$_6^c$ HCN$^c$ HNC$^c$ CH$_3$CN$^c$
C$_2$H$_3$CN$^c$ C$_2$H$_5$CN$^c$
C$_3$H$_5$CN$^c$ C$_2$N$_2^c$ C$_4$N$_2^c$ C$_6$N$_2^c$
HC$_3$N$^c$ HC$_5$N$^c$ CH$_3$C$_2$CN$^c$
CH$_2$NH$^c$ CH$_3$NH$_2^c$ NH$_3^c$
N$_2$H$_2^c$ N$_2$H$_4^c$ \\
\end{tabularx}
\caption{\label{tab:molecules}The species included in the model. A
superscript of $^c$ indicates the molecule is condensed. }
\end{table}
\subsection{Condensation and Sublimation\label{sec:cond}}
Condensation occurs when the saturation ratio, $S$, of a molecule is greater than 1.
S is defined as n($x$)/n$_{sat}$(s), where n($x$) is the
gas phase mixing ratio of species $x$ and n$_{sat}$($x$) is its saturated density
derived from the saturated vapor pressure. For $S$ $<$ 1, condensation is
switched off and sublimation of any adsorbed molecules can occur.
The abrupt change in behavior at $S$ = 1 can lead to numerical
instabilities where the system oscillates between the condensation and
sublimation regimes. In previous Titan
models various methods have been used to smooth out the transition
and prevent such instabilities.
For example, \cite{yap84}
parameterized the condensation rate in terms of $S$:
\begin{equation}
\hbox{Loss rate} \propto - \frac{S-1}{S}
\end{equation}
This results in a relatively constant loss rate as a function of $S$.
A more complicated expression was used by \cite{lavvas08a}
to ensure that the loss rate increases with increasing saturation ratios:
\begin{equation}
\hbox{Loss rate} \propto -(S-1) \frac{exp(-0.5/ln(S+1)^2)}{ln(S+1)^2} \hbox{~~~for S $>$ 1}
\end{equation}
Other expressions that have been invoked include
\begin{equation}
\hbox{Loss rate} \propto -ln S
\end{equation}
\citep{krasnopolsky09}.
Here we use a numerically stable method to determine the net
condensation rate. The rate at which molecules condense on to
a pre-existing aerosol particle is given by
the collision rate with the particle:
\begin{equation}
k_{c} = \alpha_x \sigma v_x n(x) \hbox{~~~molecules s$^{-1}$} \label{eq:freeze}
\end{equation}
where $\alpha$ is the sticking coefficient of molecule $x$ (where $\alpha_x$ $\leq$ 1),
$\sigma$ is the collisional cross-section of the particles,
$v_x$ is the gas phase
velocity of $x$, and $n(x)$ is its number density.
For a pure ice the saturated vapor pressure is measured when the
condensation and sublimation processes are in equilibrium. In this scenario
\begin{equation}
k_c n_{sat}(x) = k_s \Theta_x \label{eq:sub}
\end{equation}
where $k_s$ is the sublimation rate and $\Theta_x$ is the surface
coverage of molecule $x$. In the case of a pure ice, $\Theta_x$ = 1,
and hence the sublimation rate, $k_s$ = $k_c n_{sat}(x)$. The net
condensation rate, $J_c$ is therefore
\begin{equation}
J_c = \alpha_x \sigma v_x [n(x) - n_{sat}(x)\Theta_x] \hbox{~~~molecules s$^{-1}$}
\end{equation}
When sublimation is taking place from a mixture of ices (rather than from pure ice)
$\Theta$ will be less than 1 and the resulting gas phase abundance will
be lower than the saturated value.
$\Theta$ is calculated from
\begin{equation}
\Theta = n(x^c)/\Sigma_y n(y^c)
\end{equation}
where $n(x^c)$ is the number density of $x$ in the condensed phase,
$\Sigma_y n(y^c)$ is the total number density of all molecules condensed on to the
grain surface. We assume that the ices are well-mixed, so that the
composition of the surface from which sublimation occurs reflects that
of the bulk of the ice.
To determine the saturated densities used in this paper
we use the expressions for the saturated vapor pressure given in
Table~\ref{tab:ant}.
The values from these fits are extrapolated as necessary to provide
saturation vapor pressures over a wider range of temperatures.
Figure~\ref{fig:comp} compares the predicted mixing ratios of HCN and C$_2$H$_2$
with the value predicted directly from the saturated vapor pressure. It can be seen that the model
produces good agreement with the saturation vapor pressure in regions where the gas
is saturated.
\begin{table}
\centering
\scriptsize
\begin{tabular}{lccl}
\hline
Molecule & Expression for log P$_{sat}$& Temp range & Notes \\
& (mmHg) & (K) \\
\hline
CH$_4$ & 6.84570 - 435.6214/(T-1.639) & 91 -- 189 & \cite{yaws}\\
C$_2$H$_2$ & 6.09748 - (1644.1/T) + 7.42346 log(1000./T) & 80 -- 145 &\cite{moses92}\\
& 7.3147 - 790.20947/(T-10.141) & 192 -- 208 & \cite{lara96}\\
C$_2$H$_4$ & 1.5477 - 1038.1 (1/T - 0.011) + 16537./(1/T - 0.011)$^2$ & 77 -- 89 & \cite{moses92} \\
& 8.724 - 901.6/(T-2.555) & 89 --104 & \cite{moses92}\\
& 50.79 - 1703./T - 17.141 log(T) & 104 -- 120 & \cite{moses92}\\
& 6.74756 - 585./(T-18.18) & 120 -- 155 & \cite{moses92}\\
C$_2$H$_6$ & 10.01 - 1085./(T - 0.561) & 30 -- 90 & \cite{lara96} \\
& 6.9534 - 699.10608/(T-12.736) & 91 -- 305 & \cite{yaws}\\
CH$_3$C$_2$H & 6.78485 - 803.72998/(T-43.92) & 183 -- 267 & \cite{yaws}\\
CH$_2$CCH$_2$ & 6.62555 - 684.69623/(T-55.658) & 144 -- 294 & \cite{yaws}\\
C$_3$H$_6$ & 6.8196 - 785./(T-26.) & 161 -- 241 & \cite{yaws}\\
C$_3$H$_8$ & 7.0189 - 889.8642/(T-15.916) & 85 -- 176 & \cite{yaws}\\
C$_4$H$_2$ & 5.3817 - 3300.5/T + 16.63415 log$_{10}$(1000./T) & 127--237 & \cite{lara96}\\
& 6.5326 - 761.68429/(T-74.732) & 237 -- 478 & \cite{yaws}\\
C$_4$H$_4$ & 6.6633 - 826.0438/(T-59.712) & 181 -- 454 & \cite{yaws}\\
1-C$_4$H$_6$ & 6.98198 - 988.75(T-39.99) & 205 -- 300 & \cite{yaws}\\
1,2-C$_4$H$_6$ & 6.99383 - 1041.117/(T-30.726) & 247 -- 303 & \cite{yaws}\\
1,3-C$_4$H$_6$ & 6.84999 - 930.546/(T-34.146) & 215 -- 287 & \cite{yaws}\\
C$_4$H$_8$ & 6.8429 - 926.0998/(T-33.) & 192 -- 286 & \cite{yaws}\\
C$_4$H$_{10}$ & 7.0096 - 1022.47681/(T-24.755) & 135 -- 425 & \cite{yaws}\\
C$_5$H$_4$ & 7.986 - 1509.98716/(T-32.226) & 234 -- 367 & \cite{yaws}\\
l-C$_6$H$_6$ & 7.95508 - 1773.77625/(T-52.937) & 341 -- 449 & \cite{yaws}\\
C$_6$H$_6$ & 6.814 - 1090.43115/(T-75.852) & 233 -- 562 & \cite{yaws}\\
NH$_3$ & 7.5874 - 1013.78149/(T-24.17) & 196 -- 405 & \cite{yaws}\\
HCN & 11.41 - 2318./T & 132 -- 168 & \cite{lara96} \\
& 8.0258 - 1608.28491/(T-286.893) & 260 -- 456 & \cite{yaws}\\
HNC & 11.41 - 2318./T & 132 -- 168 & same as HCN \\
& 8.0258 - 1608.28491/(T-286.893) & 260 -- 456 & same as HCN\\
C$_2$N$_2$ & 6.9442 - 779.237/(T-60.078) & 146 -- 400 & \cite{yaws}\\
C$_4$N$_2$ & 8.269 - 2155./T & 147 -- 384 & \cite{yaws}\\
C$_6$N$_2$ & 8.269 - 2155./T & 147 -- 384 & same as C$_4$N$_2$ \\
HC$_3$N & 6.2249 - 714.01178/(T-101.55) & 214 -- 315 & \cite{yaws}\\
HC$_5$N & 6.2249 - 714.01178/(T-101.55) & 214 -- 315 & same as HC$_3$N\\
C$_2$H$_3$CN & 7.8376 - 1482.7653/T-25.) & 189 -- 535 & \cite{yaws}\\
C$_2$H$_5$CN & 7.0414 - 1270.41907/(T-65.33) & 204 -- 564 & \cite{yaws}\\
C$_3$H$_5$CN & 7.0406 - 1617.87915/(T-34.032) & 186 -- 583 & \cite{yaws}\\
N$_2$H$_2$ & 7.8288 - 1698.58081/(T-43.21) & 270 -- 653 & same as N$_2$H$_4$\\
N$_2$H$_4$ & 7.8288 - 1698.58081/(T-43.21) & 270 -- 653 & \cite{yaws}\\
CH$_3$NH$_2$ & 7.3638 - 1025.39819/(T-37.938) & 180 -- 430 & \cite{yaws}\\
CH$_3$CN & 6.8376 - 995.2049/(T-80.494) & 266 -- 518 & \cite{yaws}\\
CH$_3$C$_2$CN & 6.2249 - 714.01178/(T-101.855) & 214 -- 315 & \cite{yaws}\\
CH$_2$NH & 8.0913 - 1582.91077/(T-33.904) & 175 --512 & From \cite{yaws} value for CH$_3$OH\\
\hline
\end{tabular}
\caption{\label{tab:ant} Expressions used to calculate the saturated vapor pressures. In the absence
of laboratory data we assume that the saturated vapor pressure of HNC is the same as HCN, and that
of N$_2$H$_2$ is the same as N$_2$H$_4$. We follow \cite{loison15} in using the vapor pressure of
CH$_3$OH for CH$_2$NH and in using H$_3$CN for H$_5$CN.}
\end{table}
\begin{figure}
\includegraphics[width=0.5\linewidth]{HCNsatdens_crop.pdf}
\includegraphics[width=0.5\linewidth]{C2H6satdens_crop.pdf}
\caption{\label{fig:comp}A comparison of the model results and the
calculated mixing ratio under saturation conditions for (a) HCN and
(b) C$_2$H$_2$. The saturated
value (black line) is calculated assuming a pure ice and using the vapor pressure expressions
given in Table~\ref{tab:ant}. The model is shown in
red. The two lines coincide in the region where the model
calculations reach the saturated value and deviate where the
calculated mixing ratios are below the saturated values.}
\end{figure}
\subsection{\label{sec:sed}Sedimentation and Haze Formation}
We assume that the abundance, size and location of the
aerosol particles is fixed. In reality the particles do not remain at the same altitude but
rather sediment out towards the surface of Titan, taking any condensates
with them. To mimic this effect we have included a loss process for
condensed molecules which removes them from the model atmosphere with a rate
coefficient of 10$^{-10}$ s$^{-1}$. All condensed
species are assumed to be lost at the same rate. The assumed size of this
reaction rate is somewhat arbitrary and to test the sensitivity of our results
to its value we
also considered a loss rate of 10$^{-12}$
molecules s$^{-1}$. Changing the rate was found to have no effect on the predicted
gas phase mixing ratios.
In addition to the condensation of ice or liquids onto existing aerosols,
molecules can also be incorporated into new or existing aerosols. In
this scenario the molecules are then unavailable for return to the gas
via sublimation and are permanently removed from the gas
\citep{liang07}. This process is simulated using rates that
are proportional to the collision rates between aerosols (assuming mean radii
provided by \cite{lavvas10}) and molecules.
We simulate this by adding reactions that remove the molecules from the gas with
\begin{equation}
\hbox{X} + haze = haze \hbox{~~~k = $\beta v \sigma n_g$ s$^{-1}$}
\label{eq:aero}
\end{equation}
where $n_g$ is the mixing ratio of aerosol particles, and
$\beta$ is an efficiency factor ranging from 0.01 to 10 depending
on the molecule. The value of $\beta$ was chosen for each molecule
to maximize the agreement of the models with the observations. The
molecules to be removed in this way are HCN
($\beta$ = 0.01), C$_2$H$_3$CN ($\beta$ = 0.1), HC$_3$N and HNC ($\beta$ = 10)
C$_2$H$_5$CN ($\beta$ = 1). Other molecules
are assumed not to condense in this way -- for these molecules
agreement of the models with observations is sufficiently good
without invoking an additional loss mechanism such as haze formation.
\section{Results}\label{sec:results}
\subsection{The Effect of Condensation Processes}
We present the results of three models with different assumptions
about the condensation and sublimation. Model A is a gas phase only
model, with no condensation. Model B includes condensation and
sublimation processes as outlined in Section~\ref{sec:cond}, and the
sedimentation of aerosol particles and their condensates. Model C
extends Model B to include
the removal of molecules from the gas by haze formation.
The model parameters are summarized in
Table~\ref{tab:model}.
\begin{table}
\begin{tabular}{cccc}
\hline
Model & Condensation & Sedimentation & Haze \\
& & & formation \\
\hline
A & No & No & No \\
B & Yes & Yes & No \\
C & Yes & Yes & Yes \\
\hline
\end{tabular}
\caption{\label{tab:model}Summary of model assumptions. Condensation
and sublimation rates are discussed in
Section~\ref{sec:cond}. Sedimentation and haze formation rates are
discussed in Section~\ref{sec:sed}.}
\end{table}
The largest effects are seen for the biggest molecules and in particular for those that
contain nitrogen. The addition of sedimentation increases the
rate of removal of these species from the gas in the lower atmosphere and
improves agreement with the observations.
However, some molecules are still found to be over-abundant.
Further improvement is
achieved between 200 and 600 km for HCN, HNC,
HC$_3$N and C$_2$H$_5$CN if these molecules are assumed to be incorporated
into haze particles.
Below we discuss the chemistry of several species in more detail.
\subsection{Distribution of Nitrogen Molecules}
\begin{figure}
\includegraphics[width=0.5\linewidth]{NH3_crop.pdf}
\includegraphics[width=0.5\linewidth]{HCN_crop.pdf}
\includegraphics[width=0.5\linewidth]{HNC_crop.pdf}
\includegraphics[width=0.5\linewidth]{HC3N_crop.pdf}
\caption{\label{fig:nchem1}Distribution of some nitrogen-bearing
molecules. (red line = Model A, blue line = Model B, green line = Model C).
Cassini/INMS ($\bullet$ \cite{cui09}; $\triangleleft$ \cite{magee09},
$\Leftcircle$ \cite{vuitton07}), Cassini/UVIS (+
\cite{koskinen11}, $\vartriangle$
\cite{shemansky05}, $*$ \cite{kammer15}),
Keck \citep[$\blacktriangledown$][]{kim05}),
Voyager occultation observations ($\square$ \cite{vervack04}),
Cassini/CIRS ($\times$ \cite{vinatier10}, $\ocircle$\cite{nixon10},
$\Diamondblack$ \cite{teanby13}),
\citep[$\blacktriangle$][]{moreno11}, IRAM
($\diamond$ \cite{marten02}), ALMA ($\oplus$ \cite{cordiner14}). }
\end{figure}
\begin{figure}
\includegraphics[width=0.5\linewidth]{C2N2_crop.pdf}
\includegraphics[width=0.5\linewidth]{CH3CN_crop.pdf}
\includegraphics[width=0.5\linewidth]{C2H3CN_crop.pdf}
\includegraphics[width=0.5\linewidth]{C2H5CN_crop.pdf}
\caption{\label{fig:nchem2}Abundances of more nitrogen-bearing
species. Cassini/INMS ($\bullet$ \cite{cui09},
$\triangleleft$ \cite{magee09}, $\Leftcircle$ \cite{vuitton07}), Cassini/CIRS
($\Diamondblack$ \cite{teanby13}),
IRAM 30m ($\diamond$ \cite{marten02}),
ALMA data ($\oplus$ \cite{cordiner15}).}
\end{figure}
\subsubsection{NH$_3$}
In the lower atmosphere upper limits of the NH$_3$ abundance are
provided by Herschel/SPIRE measurements \cite[65 -- 100
km;][]{teanby13} and from
CIRS/Cassini limb observations \citep[110 -- 250 km;][]{nixon10}.
In the upper atmosphere the abundance is derived from Cassini/INMS of
NH$_4^+$ \citep{vuitton07} at 1100 km.
\cite{cui09} claim a detection of NH$_3$ in the ionosphere
between 950 and 1200km. Their value is an order of magnitude larger than that
derived by Vuitton et al. and its origin is a matter of debate.
It is possible that this high value is due to spent hydrazine fuel \citep{magee09}.
Our model abundances in the upper atmosphere are a factor of 10 lower than
the observations of \cite{vuitton07} (Figure~\ref{fig:nchem1}). Below 250 km our models are
considerably lower (but consistent with) the upper limits derived by \cite{teanby13} and \cite{nixon10}.
The main formation processes for NH$_3$ are
\begin{align}
\hbox{C$_2$H$_3$} + \hbox{NH$_2$} & \longrightarrow \hbox{NH$_3$} +
\hbox{C$_2$H$_2$} & \hbox{$<$ 800
km}
\end{align}
with destruction by photodissociation.
As discussed by \cite{loison15} the formation of NH$_3$ via
neutral-neutral reactions depends on the presence of NH$_2$ which is
not efficiently produced in Titan's
atmosphere. The inclusion of ion-molecule chemistry may lead to higher
abundances of NH$_3$.
\subsubsection{HCN}
Observations of HCN have been made from 100 km to 1000 km. The
millimeter observations of \cite{marten02} covered the whole disk and
were mainly sensitive to the mid-latitude and equatorial regions.
Observations from Cassini/CIRS \citep{vinatier07,vinatier10}, UVIS
\citep{koskinen11,shemansky05,kammer15}, and INMS \citep{magee09}
provide abundance information between 400 and 1000 km.
Abundances in the lower atmosphere are also provided by \cite{kim05}
from Keck observations \citep{geballe03}. \cite{vervack04} used
Voyager 1 Ultraviolet Spectrometer measurements to determine
abundances between 500 and 900km, although the inferred abundances are
much higher than other estimates. The differences between the Voyager
1 HCN abundances and those from Cassini may be due to solar cycle
variations. Investigating such differences is beyond the scope of
this work.
Overall our models are in good agreement with the observational data (Figure~\ref{fig:nchem1}).
We find that condensation and sublimation are important for HCN below 500 km. The
best fit to the observations is obtained with Model C
(Figure~\ref{fig:nchem1}), where sedimentation and haze formation
reduce the abundance of HCN below 500 km.
The main formation processes are
\begin{align}
\hbox{HNC} + \hbox{H} & \longrightarrow \hbox{HCN} + \hbox{H} & \hbox{300 -- 800 km}\\
\hbox{CN} + \hbox{CH$_4$} & \longrightarrow \hbox{HCN} + \hbox{CH$_3$} & \hbox{200 -- 600 km}\\
\hbox{N} + \hbox{CH$_2$} & \longrightarrow \hbox{HCN} + \hbox{H} & \hbox{600 -- 900 km}\\
\hbox{C$_2$H$_3$CN} + h\nu & \longrightarrow \hbox{HCN} + \hbox{C$_2$H$_2$} & \hbox{$<$ 1000 km}\\
\hbox{H$_2$CN} + \hbox{H} & \longrightarrow \hbox{HCN} + \hbox{H}_2 & \hbox{900 - 1300 km}
\end{align}
Photodissociation plays a role in both the formation of HCN (via
photodisssociation of C$_2$H$_3$CN above 1000 km) and in its
destruction (forming CN and H). Below 200 km destruction is by
\begin{equation}
\hbox{C$_2$H$_3$} + \hbox{HCN} \longrightarrow \hbox{C$_2$H$_3$CN} + \hbox{H}
\end{equation}
\subsubsection{HNC}
The first observations of HNC in Titan were made using Herschel/HIFI
by \cite{moreno11}. These measurements do not allow the exact
vertical abundance profile to be determined. Several
possible profiles can fit the data depending on the mixing ratio and
the cut-off altitude assumed. \cite{loison15} suggest two possible
profiles: one where the mixing ratio of HNC is 1.4 $\times$ 10$^{-5}$ above 900
km (shown in Figure~\ref{fig:nchem1}) and another where the mixing
ratio is 6 $\times$ 10$^{-5}$ above 1000 km. Our models fall between these two ranges.
More recently \cite{cordiner14} used ALMA to detect HNC. They found
that the emission mainly originates at altitudes above 400 km and that
there are two emission peaks that are not symmetrical in longitude.
We are able to match their best fit profile reasonably well with model C
(green line; Figure~\ref{fig:nchem1}), where HNC
forms haze providing the best agreement with the data at lower altitudes.
The main formation channels of HNC are
\begin{align}
\hbox{C$_2$H$_3$CN} + h\nu & \longrightarrow \hbox{HNC} + \hbox{C$_2$H$_2$} &
\hbox{$<$
900 km}\\
\hbox{H$_2$CN} + \hbox{H} & \longrightarrow \hbox{HNC} + \hbox{H$_2$} & \hbox{$>$
900 km}\\
\hbox{N} + \hbox{CH$_3$} & \longrightarrow \hbox{HNC} + \hbox{H} & \hbox{$>$ 900 km}
\end{align}
The main destruction process is by reaction with H atoms forming HCN.
This reaction has an activation barrier. In the literature the value for the activation barrier
ranges from 800 to 2000 K \citep{te96,sn98,petrie02,kida}. Here we are using the rate
from the KIDA database \citep{kida} which has the highest activation barrier of 2000 K.
\cite{loison15} used the lowest value (800 K), resulting in more efficient HCN production
and consequently a lower gas phase abundance of HNC
than we see here. We find that reducing the activation barrier does indeed reduce the
mixing ratio of HNC but does not result in a good fit to the ALMA observations in this region
(Figure~\ref{fig:HNCcomp}).
\begin{figure}
\centering
\includegraphics[width=0.5\linewidth]{HNC_ebcomp.pdf}
\caption{\label{fig:HNCcomp}How the HNC abundance depends on the
activation barrier of the reaction H + HNC $\longrightarrow$ HCN + H.
{\it green} E$_A$ = 2000 K \citep{kida}, {\it blue} E$_A$ = 1400 K, {\it red}
E$_A$ = 800 K \citep{loison15}. The lower
activation barrier results in more HNC being converted into HCN
but does not result in a better agreement with the altitude distribution seen in the
ALMA observations.}
\end{figure}
\subsubsection{HC$_3$N}
HC$_3$N has been observed at altitudes from 200 to 1000 km
\citep{marten02,vervack04,teanby06,vuitton07,cui09,magee09,vinatier10,cordiner14}.
Below 500 km our models are in excellent agreement with the
observations if it is assumed that HC$_3$N forms aerosols and thus is
permanently removed from the gas
(Figure~\ref{fig:nchem1}{\it ~(bottom left)}). Condensation and sublimation alone
result in an over-estimate of the abundance compared to the
observations in this region.
Good agreement is also seen for all models between 500 km
and 700 km. Above this our models tend to under predict the HC$_3$N
abundance.
Below 100 km the mixing ratio follows the
saturation level, so that below this
altitude the mixing ratio is much reduced compared to the gas only model. Better
agreement with the observations below 400 km is obtained in the
haze formation model where condensed molecules are assumed to be
incorporated into aerosol particles and removed from the gas.
The main formation process below 1000 km is
\begin{equation}
\hbox{C$_3$N} + \hbox{CH$_4$} \longrightarrow \hbox{HC$_3$N} + \hbox{CH}_3
\end{equation}
and above 800 km by
\begin{align}
\hbox{CN} + \hbox{C$_2$H$_2$} & \longrightarrow \hbox{HC$_3$N} + \hbox{H} \\
\hbox{C$_2$H$_3$CN} + h\nu & \longrightarrow \hbox{HC$_3$N} + \hbox{H}_2
\end{align}
Destruction is by photodissociation
\begin{equation}
\hbox{HC$_3$N} + h\nu \longrightarrow \hbox{C$_3$N} + \hbox{H}
\end{equation}
and by reaction with H atoms
\begin{equation}
\hbox{H} + \hbox{HC$_3$N} + \hbox{M} \longrightarrow \hbox{H$_2$C$_3$N} + \hbox{M}
\end{equation}
The observations show a sharp decrease in the abundance of HC$_3$N below
400 km. In our models this can be accounted for if HC$_3$N is incorporated
into haze particles (Model C). An alternative explanation of
meridional circulation and condensation in the polar regions has been
suggested \citep{loison15,hourdin04}.
\subsubsection{C$_2$N$_2$}
Observations of C$_2$N$_2$ have been made by Cassini/CIRS \citep{teanby06,teanby09} and by
Cassini/INMS \citep{cui09,magee09}. The models with condensation are
in very good agreement with both of these datasets (Figure~\ref{fig:nchem2}). Without
condensation the abundance in the lower atmosphere is over-estimated.
The main formation route for C$_2$N$_2$ is by the reaction of CN and HNC:
\begin{equation}
\hbox{CN} + \hbox{HNC} \longrightarrow \hbox{C$_2$N$_2$} + \hbox{H}
\end{equation}
with destruction via photodissociation forming CN or by
\begin{equation}
\hbox{H} + \hbox{C$_2$N$_2$} + \hbox{M} \longrightarrow \hbox{H$_2$CN$_2$} + \hbox{M} \hbox{~~~~$<$ 400 km}
\end{equation}
\subsubsection{CH$_3$CN}
Submillimeter observations with the IRAM 30 m telescope detected
the CH$_3$CN (12-11) rotational line providing a disk average vertical
profile up to 500 km, dominated by the equatorial region \citep{marten02}.
Cassini/CIRS \citep{nixon13} and Cassini/INMS
\citep{vuitton07,cui09} provide estimates of the abundance above 1000 km.
All models are in good agreement with the observations below 800 km, although all predict
slightly lower abundances than observed between 500 and 600 km. The predicted mixing ratio
at 1100 km is a factor of 10 lower than the observed value of 3 $\times$ 10$^{-5}$ \cite{cui09}.
The main formation processes are
\begin{align}
\hbox{CN} + \hbox{CH}_4 & \longrightarrow \hbox{CH$_3$CN} + \hbox{H} & \hbox{$<$ 900 km}\\
\hbox{NH} + \hbox{C$_2$H$_4$} & \longrightarrow \hbox{CH$_3$N} + \hbox{H}
& \hbox{$>$ 900 km}\\
\hbox{CH$_2$NH} + \hbox{CH} & = \hbox{CH$_3$CN} + \hbox{H} & \hbox{$<$ 900 km}
\end{align}
with destruction by
\begin{align}
\hbox{CH$_3$CN} + \hbox{H} & \longrightarrow \hbox{CN} + \hbox{CH}_4 & \hbox{$<$ 1200 km}
\end{align}
\subsubsection{C$_2$H$_3$CN}
Several observations have placed upper limits on the abundance of
C$_2$H$_3$CN. \cite{marten02} used the IRAM 30m telescope to determine
upper limits between 100 and 300 km. Cassini/INMS has
provided an upper limit of 4 $\times$ 10$^{-7}$ at 1077 km \citep{cui09},
while Cassini/INMS \cite{magee09} determined a mixing ratio of 3.5 $\times$
10$^{-7}$ at 1050 km and \cite{vuitton07} found 10$^{-5}$ at 1100 km
from observations of ions. Cordiner et al. (private communication)
have detected C$_2$H$_3$CN in the submillimeter and found an average
abundance of 1.9 x 10$^{-9}$ above 300 km.The model abundance of C$_2$H$_3$CN in the
upper atmosphere is within a factor of 2 of the \cite{vuitton07} value
but 50 times higher than \cite{magee09} and \cite{cui09}.
None of our models have a constant mixing ratio with altitude between
100 and 300 km as derived from the IRAM observations (Figure~\ref{fig:nchem2}). Model B and C
(which include condensation) are consistent with the derived mixing
ratio at a particular altitude, but neither reproduce the constant
value between 100 and 300 km. In the upper atmosphere all models
predict mixing ratios within a factor a 3 of the \cite{magee09} result
but are over-abundant compared to the other measurement in this region.
The main production mechanism is by
reaction of CN with C$_2$H$_4$:
\begin{align}
\hbox{C$_2$H$_3$} + \hbox{HCN} & \longrightarrow \hbox{C$_2$H$_5$CN} + \hbox{H} & \hbox{$<$ 200 km}\\
\hbox{CN} + \hbox{C$_2$H$_4$} & \longrightarrow \hbox{C$_2$H$_3$CN} + \hbox{H} &
\hbox{400 -- 800
km}\\
\hbox{H$_2$C$_3$N} + \hbox{H} & \longrightarrow \hbox{C$_2$H$_3$CN} & \hbox{$>$ 400 km}
\end{align}
Gas phase destruction processes are
\begin{align}
\hbox{C$_2$H$_3$CN} + h\nu & \longrightarrow \left\{
\begin{array}{ll} \hbox{HCN} + \hbox{C$_2$H$_2$}\\
\hbox{HNC} + \hbox{C$_2$H$_2$}\\
\hbox{HC$_3$N} + \hbox{H}_2
\end{array}
\right.
\end{align}
Below 400 km haze formation and sedimentation of aerosol particles
play an important role in determining the gas mixing ratio in Model C.
\subsubsection{C$_2$H$_5$CN}
Upper limits for the abundance of C$_2$H$_5$CN have been determined between 100 and 300
km from IRAM 30m observations \cite{marten02}, with abundances in the upper
atmosphere provided by Cassini/INMS data \citep{vuitton07,cui09,magee09}.
More recently
\cite{cordiner15} detected this molecule above 200 km using ALMA.
Our models over-estimate the abundance of C$_2$H$_5$CN in the upper
atmosphere (Figure~\ref{fig:nchem2}), probably because we do not include ion chemistry \citep[for
a discussion of this point see][]{loison15}. Below 700 km, Model C
is in excellent agreement with the ALMA data of Cordiner et al.
The main formation process is
\begin{equation}
\hbox{CH$_2$CN} + \hbox{CH$_3$} + \hbox{M} \longrightarrow \hbox{C$_2$H$_5$CN} + \hbox{M}
\end{equation}
Destruction is by photodissociation
\begin{align}
\hbox{C$_2$H$_5$CN} + h\nu & \longrightarrow \left\{
\begin{array}{ll}
\hbox{C$_2$H$_4$} + \hbox{HCN} \\
\hbox{CH}_3 + \hbox{CH$_2$CN} \\
\hbox{C$_2$H$_3$CN} + \hbox{H$_2$}
\end{array}
\right.
\end{align}
or by reaction with CH, C$_2$H$_3$ or C$_2$H.
\begin{align}
\hbox{C$_2$H$_5$CN} + \hbox{CH} & \longrightarrow \hbox{CH$_2$CN} + \hbox{C$_2$H$_4$}\\
\hbox{C$_2$H$_5$CN} + \hbox{C$_2$H} & \longrightarrow \hbox{HC$_3$N} + \hbox{C$_2$H$_5$} & \hbox{$<$ 300 km}\\
\hbox{C$_2$H$_3$CN} + \hbox{C$_2$H$_3$} & \longrightarrow \hbox{C$_2$H$_3$CN} + \hbox{C$_2$H$_5$} & \hbox{$<$ 300 km}
\end{align}
\subsection{\label{sec:flux}Condensates in Titan's Atmosphere}
We find several layers at which condensates are abundant with the location
being molecule dependent. The first condensate layer
is in the lower atmosphere around the tropopause. Here we find
condensates of C$_2$H$_2$, C$_2$H$_4$, C$_2$H$_6$, C$_3$H$_8$ among others.
A little further up in the atmosphere around 65 -- 80 km several molecules have peaks in
condensation e.g.\ HCN, C$_4$H$_8$, C$_4$H$_2$, C$_2$H$_3$CN, C$_2$N$_2$,
CH$_3$C$_2$H. Another layer of C$_2$H$_3$CN, CH$_3$CN, C$_2$H$_5$CN
and CH$_3$C$_2$H forms around 110 -- 130 km. Several molecules also
have high condensation levels between 600 and 900 km
e.g. CH$_3$C$_2$H, HC$_5$N, HC$_3$N, CH$_3$CN, and
C$_2$H$_5$CN. Figure~\ref{fig:HCN_vap} shows the condensation layers for
HCN and HC$_3$N. Both these molecules have high condensate abundances
between 70 and 100 km, but HC$_3$N has a further peak around 500 km where
the atmospheric temperature dips, and the gas phase abundance of this molecule is high.
\begin{figure}
\includegraphics[width=0.5\linewidth]{HCN_sat_crop.pdf}
\includegraphics[width=0.5\linewidth]{HC3N_sat_crop.pdf}
\caption{\label{fig:HCN_vap}The calculated mixing ratios of HCN (left)
and HC$_2$N (right) in the
gas (red line) and in condensed form (blue line). Also shown is the
saturated mixing ratio (green line). The mixing ratio of HCN$^c$
peak at $\sim$ 65 km. The mixing ratio of HCN in the gas around this
altitude is slightly below the saturated value, even though there
is condensed HCN available to be sublimated. The reason for this is
that the condensates are not pure. Since the sublimation rate
depends on the surface coverage of the condensed molecule
sublimation from a mixed condensate is less efficient than from a
pure condensate, leading to lower gas abundances. A similar effect
is seen for other condensed molecules. For HC$_3$N the peak
condensate mixing ratio occurs around 500 km where there is a dip in
temperature corresponding to a high local abundance of HC$_3$N. Other
nitrogen-bearing molecules show similar behavior in this region.}
\end{figure}
The net flux of material falling on to the surface of Titan can be
calculated from the difference between the atmospheric formation and
destruction.
Table~\ref{tab:flux} presents our predictions of the surface flux of
nitrogen molecules. These are in solid form and if evenly distributed
across Titan's surface would create a layer 4.4 m deep over a timescale
of 1 Gyr. This amount of ``fixed nitrogen'' could be of biological importance.
\begin{table}
\small
\centering
\begin{tabular}{lcrrc}
\hline
Molecule & Flux & Flux & Depth (m) & Solid/Liquid \\
& (molecules cm$^{-2}$ s$^{-1}$) &(g cm$^{-2}$/Gyr) & & at 95 K \\
\hline
HCN$^c$ & -1.2 $\times$ 10$^8$ & -170 & 2.12 & S \\
HNC$^c$ & -8.1 $\times$ 10$^6$ & -11.0 & 0.14 & S\\
HC$_3$N$^c$ & -2.9 $\times$ 10$^7$ & -77.4 & 1.0 & S \\
HC$_5$N$^c$ & -3.4 \ 10$^6$ & -13.5 & 0.17\\
C$_2$N$_2^c$ & -5.8 $\times$ 10$^5$ & -15.8 & 0.02 & S \\
CH$_3$CN$^c$ & -2.8 $\times$ 10$^5$ & -0.6 & 0.01 & S \\
CH$_3$C$_2$CN & -4.5 $\times$ 10$^6$ & -15.3 & 0.19 & \\
C$_2$H$_5$CN$^c$ & -6.4 $\times$ 10$^6$ & -18.4 & 0.23 & S \\
C$_2$H$_3$CN$^c$ & -1.5 $\times$ 10$^7$ & -41.6 & 0.52 & S \\
\hline
Total N & & & 4.4 m & \\
\hline
\end{tabular}
\caption{\label{tab:flux}Flux of condensed molecules onto
Titan's surface. Also shown are the estimated deposit thickness
(depths) calculated
assuming an average density of 0.8 g cm$^{-3}$.}
\end{table}
\section{Discussion and Conclusions}\label{sec:conc}
The removal of molecules by condensation plays an important role in determining
the gas phase composition of Titan's atmosphere, as well as creating
new aerosols. Condensates are found throughout the atmosphere.
For the majority of molecules, condensation is most efficient below the tropopause.
Larger molecules, and in particular nitrogen-bearing molecules
have another condensation peak between 200 and 600 km.
Relatively high abundances of condensates can also be present above
500 km if the gas phase abundance of a given molecule is high, e.g.\
HC$_3$N, HC$_5$N, CH$_3$CN and C$_2$H$_5$CN. These molecules
condense in the region where Titan's haze forms. The
effect is enhanced if it is assumed that some molecules can be
permanently removed from the gas by being incorporated into aerosol
particles. This mechanism was able to bring the abundances of
HC$_3$N, HCN, HNC, CH$_3$CN and C$_2$H$_5$CN into good
agreement with the observations below 600 km.
Although Titan possesses a rich organic chemistry it is unclear whether this could
lead to life. Photochemically produced compounds on Titan,
principally acetylene, ethane and organic solids, would release energy
when consumed with atmospheric hydrogen, which is also a photochemical
product. \cite{mckaysmith05} speculate on the possibility of widespread
methanogenic life in liquid methane on Titan. On Earth fixed nitrogen is often
a limiting nutrient. Our work shows that an abundant supply of fixed nitrogen,
including species of considerable complexity, is available from atmospheric photochemistry.
Creating the kinds of lipid membranes that form the basis of lie on Earth
depends on the presence of liquid water. Titan's atmosphere contains little oxygen
and the surface temperature is well below that at which liquid water can survive.
Instead surface liquids are hydrocarbons \citep[e.g.][]{hayes16}. Therefore any
astrobiological processes, if present, are likely to be quite different to those on Earth.
A recent paper by \cite{slc15} suggests that as alternative to lipids, membranes could
be formed from small nitrogen-bearing organic molecules such as acrylonitrile (C$_2$H$_3$CN).
Stevenson et al.\ calculate that a membrane composed of acrylonitrile molecules would
be thermodynamically stable at cryogenic temperatures and would have a high energy barrier
to decomposition.
All of our models predict abundances of C$_2$H$_3$CN that are in agreement with observations
above 500 km. Below this condensation and incorporation into haze are required to bring the
predicted mixing ratios down to the values inferred from observations \cite{cordiner15}. If
acrylonitrile were to be involved in life formation it needs to reach the surface of Titan.
Our predicted flux of this molecule onto Titan's surface is 1.5 $\times$ 10$^7$ molecules cm$^{-2}$ s$^{-1}$,
or $\sim$ 41.5 gcm$^{-2}$/Gyr, a quantity that is potentially of biological importance.
\section*{Acknowledgements}
This research was conducted at the Jet Propulsion
Laboratory, California Institute of Technology under contract with the
National Aeronautics and Space Administration.
Support was provided by the NASA Astrobiology Institute/Titan as a Prebiotic
Chemical System. YLY was supported in part by the Cassini UVIS program via NASA grant
JPL.1459109 to the California Institute of Technology. The
authors thank Dr. Run-Lie Shia for his assistance with the KINETICS code
and Dr. Panyotis Lavvas for providing the aerosol data used in these models.
| 21527556408329e3545a39d9d8ce0c8ee80d1a17 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
The present paper concerns a time operator in quantum theory which is
defined, in a first stage of cognition, as a symmetric operator canonically conjugate to a Hamiltonian if it exists.
The uncertainty relation which is derived from the canonical commutation relation of a time operator and a Hamiltonian may be
interpreted as a mathematically rigorous form of {\em time-energy uncertainty relation}.
Moreover time operators may play important roles in quantum phenomena \cite{miy01,ara05,ara08b,am08b,mme08,mme09}.
To explain motivations for studying time operators, we begin with a brief historical review on time and time operator
in quantum theory (cf. also \cite[Chapter 1]{mme08}).
\subsection{Historical backgrounds}
In the old quantum theory, N. Bohr assumed that the interaction of the electrons
in an atom with an electromagnetic field causes transitions among the allowed electron orbits in such a way that the transitions are accompanied by the absorption or the emission of electromagnetic radiations by the atom.
In this hypothetical theory, however,
no principle for the timing of occurrence of these transitions was shown.
The new quantum theory presented
by Heisenberg (1925), Born-Heisenberg-Jordan (1926) and Schr\"odinger (1926)
provides a method of calculating the transition probabilities, but the question of the timing at which
the events occur was not addressed explicitly.
Heisenberg introduced two kinds of uncertainty relations, i.e., the uncertainty relation for position and momentum,
and that for time and energy.
He argued (\cite[p.179, equation (2)]{hei27}) that the imprecision within which the instant of transition is specifiable is given by $\Delta t$ obeying
the uncertainty relation
\eq{uto}
(\Delta t) (\Delta E)\sim \hbar=\frac {h}{2\pi}
\end{equation}
with the change $\Delta E$ of energy in the quantum jump, where $h$ is the Planck constant.
Although many of the issues involved in the uncertainty principle for position and momentum have been clarified so far,
similar clarity has not yet been achieved on the uncertainty principle for time and energy.
For example, in \cite{aa90,mt45}, uncertainty relation \kak{uto}
is derived, but $\Delta t$ is not considered an imprecision of measurement on time;
interpretations such as \lq\lq{a measurement act of the time gives an unexpected change to an energy level}" or \lq\lq{it dishevels a clock to have been going to measure energy exactly}" may be invalid unless any restrictions are imposed
depending on measurement setups.
In addition, the definition of $\Delta t$ seems to vary from case to case.
\subsection{Time in quantum theory and time operator}\label{subsec-time}
It is said that there exists a three-fold role of
time in quantum theory ( see, e.g., \cite{bus01} and \cite[Chapter 3]{mme08}).
Firstly time is identified as the parameter entering the Schr\"odinger equation,
which is a differential equation describing
the causal continuous
change of states of a quantum system,
and measured by a laboratory clock. Time in this sense is called the external time.
The external time measurement is carried out with clocks that are not dynamically connected with objects investigated in experiments.
By contrast, time as a dynamical one can be defined by
the dynamical behavior of quantum objects. A dynamical time is defined and measured in terms of a physical system undergoing changes. Examples include
the linear uniform motion of a free particle and the oscillation of the atoms in an atomic clock.
Finally, time can be considered as a quantum object which forms a canonical pair with a Hamiltonian in a suitable sense.
As already mentioned, time in this sense is called a time operator in its
simplest form. There is in fact a hierarchy of time operators as is shown below.
The main purpose of the present paper is to analyze this hierarchy mathematically and
to establish abstract existence theorems on time operators
in relation to the hierarchy with applications to Schr\"odinger Hamiltonians.
A simple example of time operator is given as follows.
A non-relativistic quantum particle with mass $m>0$ under the action of a constant force $F\in {\mathbb{R}}\setminus\{0\}$
in the one-dimensional space ${\mathbb{R}}$
is governed by the Hamiltonian
$$
H_F=\frac{1}{2m} P^2-FQ
$$
acting in
$L^2({\mathbb{R}})$, the Hilbert space of square integrable Borel measurable functions on ${\mathbb{R}}$, with the momentum operator $P=-iD_x$ ($D_x$ is the generalized differential operator in the variable $x\in{\mathbb{R}}$)\footnote{We use the physical unit system where $\hbar=1$.} and
$Q$ being the multiplication operator by $x$.
It is shown that
$H_F$ is essentially self-adjoint on $C_0^{\infty}({\mathbb{R}})$, the space of
infinitely differentiable functions on ${\mathbb{R}}$ with compact support, and hence its closure $\overline{H}_F$ is self-adjoint
(but, note that $\overline{H}_F$ is neither bounded from below nor from above).
The self-adjoint operator
$T_F=P/F $
satisfies
the canonical commutation relation (CCR)
$$
[H_F,T_F]=-i\bbbone
$$
on a dense domain $\mathcal D$ (e.g., $\mathcal D=C_0^{\infty}({\mathbb{R}})$),
where $\bbbone$ denotes identity and $[A,B]:=AB-BA$.
This shows that $T_F$ is a canonical conjugate operator to the Hamiltonian $H_F$ and hence a time operator of $H_F$.
From the CCR, one can derive
the uncertainty relation
of Heisenberg type
$$
(\Delta H_F)_{\psi}(\Delta T_F)_{\psi}\geq \frac{1}{2}
$$
for all unit vectors $\psi\in \mathcal D$, where $(\Delta A)_{\psi}$
denotes the uncertainty of $A$ with respect to the state vector $\psi$
(see (\ref{Delta-A}) for its definition). This inequality may be interpreted as a form of time-energy uncertainty relation
in the present model.
As for time operator, however, there is a long history of confusion and controversy\footnote{A germ of the notion of time
operator is found already in Heisenberg's paper \cite[pp.177--179]{hei27} in 1927.}.
The origin of this may come from the statement of Pauli made in 1933 (\cite[p.63, footnote 2]{pau33}) that
the introduction of a time observable $T$ satisfying the CCR
\eq{to}
[H, T]=-i\bbbone
\end{equation}
with a self-adjoint operator $H$ having a discrete eigenvalue
is basically forbidden. Although there are no explicit arguments for this statement in the cited literature
(only reference to Dirac's textbook), a formal (false in fact) argument leading to the statement may be as follows:
let $\phi$ be an eigenvector of $H$ with a discrete eigenvalue $E$:$H\phi=E\phi$.
Then, using (\ref{to}) formally, one obtains
$He^{i\varepsilon T}\phi=(E+\varepsilon)e^{i\varepsilon T}\phi\cdots (*)$ for all $\varepsilon\in{\mathbb{R}}$.
Hence $e^{i\varepsilon T}\phi$ is an eigenvector of $H$ with eigenvalue $E+\varepsilon$.
Since $\varepsilon\in{\mathbb{R}}$ is arbitrary, it follows that each point in ${\mathbb{R}}$ is an eigenvalue of $H$.
But this obviously contradicts the discreteness of eigenvalues of $H$.
It should be noted, however,
that this argument is very formal, in particular, no attention was paid to the domain of the operators involved. E.g.,
if $\phi$ is not in the domain of $T^n$ for some $n\in \mathbb{N}$, then
the expansion $e^{i\varepsilon T}\phi=\sum_{n=0}^{\infty}(i\varepsilon)^nT^n\phi/n!$ is meaningless; even in the case
where $\phi$ is in the domain of $T^n$ for all $n\in\mathbb{N}$,
$\sum_{n=0}^{\infty}(i\varepsilon)^nT^n\phi/n!$ is not necessarily convergent ; moreover,
$e^{i\varepsilon T}\phi$ is not necessarily in the domain of $H$ and, if $e^{i\varepsilon T}\phi$ is not in the domain of $H$, then
$(*)$ is meaningless.
It is well known \cite[p.2]{pu67} that at least one of $T$ and $H$ satisfying the CCR (\ref{to}) on a dense domain is an unbounded operator and, for unbounded operators, their domain must be carefully considered.
As a matter of fact, the above argument is incorrect and so is the Pauli's statement too.
Indeed, one can construct a self-adjoint operator $H$ which is bounded from below with purely discrete spectrum and a self-adjoint operator $T$ such that \kak{to} holds
on a dense domain. This was pointed out in \cite[p.4]{bus01} and mathematically rigorous constructions
of such time operators $T$ have been done in \cite{gal02,am08b}.
The history of studies on time operators as well as on representations of CCR
suggests that there may be a hierarchy of time operators and this indeed is the case as is shown below in the present paper.
It is important to distinguish each class from the others in the hierarchy.
In our words, the time observable $T$ such that the above formal argument
may take a rigorous form is an {\em ultra-strong} time operator (see Remark \ref{rem-u-strong} below), since the {\em operator equality} $e^{-i\varepsilon T}He^{i\varepsilon T}=H+\varepsilon,\,\varepsilon\in{\mathbb{R}}\cdots (\dagger)$
is tacitly assumed in the above argument in fact, which, however,
is {\em not equivalent} to \kak{to} in the mathematically rigorous sense \cite{fug67},
and, if $H$ is self-adjoint, then $(\dagger)$ is equivalent to the Weyl relation $e^{i\varepsilon T}e^{itH}=e^{-it\varepsilon}e^{itH}e^{i\varepsilon T},\,t,\varepsilon\in{\mathbb{R}}\cdots (\dagger\dagger)$, meaning that $T$ is an ultra-strong time operator of $H$ (but, if
$H$ is not essentially self-adjoint, then $(\dagger)$ does not imply $(\dagger\dagger)$ with $H$ replaced by the closure
$\overline{H}$ of $H$, because, in this case, \lq\lq{\,$e^{it\overline{H}}$\,}" is meaningless as a unitary operator).
If a self-adjoint operator $H$ has an ultra-strong time operator $T$, then
$T$ is a strong time operator of $H$ (see (\ref{hierarchy}) below) and hence
$H$ is absolutely continuous (see Proposition \ref{time} below) so that $H$ has no eigenvalues. Therefore,
in this case, the above argument becomes meaningless.
Moreover, if $H$ is semi-bounded, then no strong time operator $T$ of $H$ is essentially self-adjoint
(\cite{miy01}, \cite[Theorem 2.8]{ara05}) and
hence \lq\lq{\,$e^{i\varepsilon \overline{T}}$\,}" makes no sense as a unitary operator. In this sense too,
the above argument is meaningless.
It has been absurd that studies on time observables have been ruled out for so many years due to the
Pauli's statement without any questions. If one could have carefully examined the Pauli's statement
with mathematically rigorous thinking, then one could have found incorrectness of it.
\subsection{Rough description of main results}
As already mentioned, a time operator $T$ of a self-adjoint operator $H$ is
defined to be a symmetric operator satisfying CCR~\kak{to} on a suitable dense domain (we shall
give a more detailed description of time operators in Section \ref{sec2}).
Another approach to consider time operators as observables is an application of
positive operator valued measures (POVM) \cite[Chapter 10]{mme08}. In this paper, however,
we take an operator-theoretical approach to classify time operators
and to construct a time operator
for a given self-adjoint operator without invoking POVM.
Consequently we are led to extend the conventional notion of time operator.
Indeed, commutation relation \kak{to} can be weakened in at least two manners
and we find a time operator $T$ for each weakened form.
As we have learned from the formal argument on the Pauli's statement,
taking care of domains of $T$ and $H$ is crucial not to be led to incorrect
conclusions. Thus
the domain of time operators is one key ingredient to study them.
We now outline main results obtained in the present paper in (1)--(5) below (rigorous statements of assumptions and results will be given from Section \ref{sec2}).
Let $H$ be a self-adjoint operator acting in a complex Hilbert space ${\mathcal H}$.
\medskip
\noindent
{\bf (1) Ultra-weak time operators and a hierarchy of time operators}.
It has so far been known that there are at least three classes of time operators \cite{ara08b,ara08c}, i.e.,
time operators as canonical conjugates of a Hamiltonian in the conventional sense, which may be called
{\em ordinary time operators} to distinguish
them from other classes of time operators, {\em strong time operators} and {\em weak time operators}.
In the present paper, in addition to these classes of time operators,
we introduce a new concept on time operator, which we call {\it ultra-weak time operator},
and study it. An ultra-weak time operator, however, is not an operator in general, but defined to be a
sesquilinear form ${\mathfrak{t}}:\mathcal D_1\times \mathcal D_2\to{\mathbb{C}}$ with non-zero subspaces $\mathcal D_1$ and $\mathcal D_2$ of ${\mathcal H}$ such that
$$
{\mathfrak{t}}[ H\phi,\psi] -{\mathfrak{t}}[H\psi,\phi]^*=-i(\phi,\psi),\quad \psi,\phi\in\ms E,
$$
where, for $z\in{\mathbb{C}}$, $z^*$ denotes the complex conjugate of $z$, $(\,\, ,\,\,)$ is the inner product of ${\mathcal H}$ (linear in the second variable) and
$\ms E$ is a non-zero subspace of ${\mathcal H}$ (for the rigorous definition of ${\mathfrak{t}}$, see Definition \ref{def-uwt}).
The class of ultra-weak time operators may be compared to
the space of distributions in the context of theory of functions (as there exists a distribution which is not a function,
there may exist an ultra-weak time operator which is not an operator).
For convenience, we also introduce the concept of ultra-strong time operator which has been already mentioned above.
These five classes of time operators form a hierarchy in the following sense:
\begin{align}
\{\mbox{ultra-strong t.o.}\}\!\subset\! \{\mbox{strong t.o.}\}
\!&\subset\! \{\mbox{t.o.}\}\!\nonumber\\
&\subset\! \{\mbox{weak t.o.}\}\!\subset\!
\{\mbox{ultra-weak t.o.}\}, \label{hierarchy}
\end{align}
where t.o. is abbreviation of \lq\lq{time operators}".
See Section \ref{sec2} below for more details.
Generally speaking, it is expected that each class in the hierarchy of time operators
has proper roles in connection with quantum phenomena.
In this paper, we particularly concentrate our attention on
strong time operators, time operators and ultra-weak time operators.
As a possible physical aspect of ultra-weak time operators, a weak form of uncertainty relation
is given (see Proposition~\ref{U1}).
\medskip
\noindent
{\bf (2) Existence of strong time operators in an abstract framework}.
A strong time operator $T$ of a self-adjoint operator $H$ is defined through
the weak Weyl relation (see Definitions~\ref{wwr} and \ref{1100}).
It is known that \kak{to} is satisfied on a dense domain and the spectrum
$\sigma}%{\rm Sp}(H)$ of $H$ must be purely absolutely continuous. Hence, if
$H$ has an eigenvalue, no strong time operator of $H$ exists.
Then a natural question is to ask the existence of a strong time operator for an absolutely continuous self-adjoint operator.
We introduce a class $S_0({\mathcal H})$ of self-adjoint operators on ${\mathcal H}$ in Definition \ref{S1} and prove the following theorem (Theorem \ref{thm-ac-st}):
\begin{theorem}\label{M0} Assume that ${\mathcal H}$ is separable and that $H\in S_0({\mathcal H})$.
Then $H$ has a strong time operator.
\end{theorem}
It may be interesting to consider extensions of this theorem to a more general class of absolutely continuous
self-adjoint operators. But this will be done elsewhere.
In this paper, we
next proceed to construction of a time operator for a self-adjoint operator
which has point spectra (eigenvalues).
\medskip
\noindent
{\bf (3) Existence of time operators of a self-adjoint operator with point spectra}.
As for general existence of time operators of a self-adjoint operator $H$ with
point spectra, only limited classes of $H$
have been found \cite{gal02,am08a,am08b,ara09}.
In this paper we extend these results (Theorem \ref{thm-d-top}):
\begin{theorem}\label{M1}
Let $\sigma}%{\rm Sp} (H)=\{E_n\}_{n=1}^\infty$,
$E_1<E_2<\cdots $ and $\lim_{n\to\infty}E_n=\infty$.
Then there exists a time operator $T$ of $H$.
\end{theorem}
In \cite{gal02,am08a,am08b}, time operators of $H$ having purely discrete spectrum are constructed, but
the growth condition $\sum_{n=1}^\infty{1}/{E_n^2}<\infty$ for $\{E_n\}_n$ is imposed. This condition seems to be artificial. An important point
in Theorem \ref{M1} is that this condition is not required.
We show in Subsection \ref{subsec-III} that
the noncommutative harmonic oscillator Hamiltonian \cite{iw07} and the Rabi Hamiltonian
\cite{rab36,rab37,bra11,mps14} are included in this class of Hamiltonians as concrete examples.
\begin{remark}{\rm After submitting the first version of the present paper, we have learned
that Teranishi \cite{te16} has proved a theorem essentially same as Theorem \ref{M1} by
a method different from ours.}
\end{remark}
\medskip
\noindent
{\bf (4) Ultra-weak time operators}.
We also establish a theorem on the existence of ultra-weak time operators
for a general class of self-adjoint operators with infinitely many discrete eigenvalues but the accumulation point is not $\infty$ (Theorem \ref{thm-uwt}).
\begin{theorem}
Suppose that
$\sigma}%{\rm Sp} (H)\setminus\{0\}=\{E_j\}_{j=1}^\infty$,
$E_1<E_2<\cdots <0$, $\lim_{j\to\infty}E_j=0$, and $0$ is not an eigenvalue of $H$.
Then there exists an ultra-weak time operator ${\mathfrak{t}}$ of
$H$.
\end{theorem}
It will be seen in Subsection \ref{4.1}
that ${\mathfrak{t}}[\phi, \psi]=(\phi,A\psi)$ formally with some operator $A$.
The crucial point is that $A$ is of the form
$$A=-\frac{1}{2}(T_{-1}H^{-2}+H^{-2}T_{-1}),$$
where $T_{-1}$ denotes a time operator of $H^{-1}$.
It is difficult to show, however, that ${\rm D}(A)\not=\{0\}$ and ${\rm D}(HA)\cap {\rm D}(AH)\not=\{0\}$.
This is the reason why the introduction of an ultra-weak time operator ${\mathfrak{t}}$
as a sesquilinear form is needed and may be even natural.
\medskip
\noindent
{\bf (5) Ultra-weak time operators for Schr\"odinger operators}.
Finally, by applying the results described in (1)--(4) above, we construct an ultra-weak time operator for
a class of Schr\"odinger operators, including the Hamiltonian of the hydrogen atom.
It is shown in Theorem \ref{main2} that, for
a class of potentials $V:{\mathbb{R}}^d\to{\mathbb{R}}$,
the $d$-dimensional Schr\"odinger operator
$$
{H_V}:=-\frac 1{2m}\Delta+V
$$
acting in $L^2({\mathbb{R}}^d)$ has an ultra-weak time operator,
where $\Delta$ is the $d$-dimensional generalized Laplacian.
Below are some examples of ${H_V}$ having an ultra-weak time operator (see Subsection \ref{ex}
for more details).
(i) Let $U\in L^\infty ({\mathbb{R}}^3)$ and
$$V (x):=\frac{U (x)}{ (1+|x|^2)^{{\half} +\varepsilon}}.$$
Suppose that
$U$ is negative, continuous, spherically symmetric and
satisfies that
$U (x)=-1/|x|^\alpha$ for $|x|>R$
with $0<\alpha<1$ and $R>0$.
For each $\alpha$, we
can choose $\varepsilon>0$ such that
$2\varepsilon+\alpha<1$.
Then
$H_V$ has an ultra-weak time operator.
See Example \ref{47}.
(ii) Let
$$H_{\rm hyd}:=-\frac 1{2m}\Delta-\frac{\gamma}{|x|}$$ be
the $3$-dimensional hydrogen Schr\"odinger operator with a constant $\gamma> 0$. Then
$H_{\rm hyd}$
has an ultra-weak time operator. See Example \ref{ex-hyd}.
(iii)
Suppose that ${H_V}$ has an ultra-weak time operator.
Then, under some conditions, we can
show that the following operators $f({H_V})$ also have an ultra-weak time operator (see Theorem \ref{thm-uwt-HS}):
\begin{itemize}
\item[(a)]
$f ({H_V})=e^{-\beta {H_V}}$ for $\beta\in{\mathbb{R}}\setminus\{0\}$;
\item[(b)]
$f ({H_V})=\sum_{j=0}^N a_j H_V^j$ ($a_j\in{\mathbb{R}}, N\in{\mathbb N}$);
\item[(c)]
$f ({H_V})=\sin (2\pi \beta {H_V})$ for $\beta\in{\mathbb{R}}\setminus\{k/2E_j|k\in{\mathbb Z}, j\in{\mathbb N}\}$, where $\{E_j\}_{j\in{\mathbb N}}$ denotes the discrete spectrum of ${H_V}$.
\end{itemize}
See Examples \ref{exp}, \ref{poly} and \ref{411}.
In the next section we give definitions of terminology used in this paper and remarks
from mathematical point of view.
\section{Mathematical Backgrounds of Time Operators}\label{sec2}
\subsection{A review on mathematical analysis on time operators}
Mathematical analysis on time operators has been developed
in the papers \cite{miy01,gal02,gcb04,ara05, ara07,ara08a,ara08b,ara08c,am08a,am08b,ara09,hkm09}.
Let $A$ and $B$ be linear operators on a complex Hilbert space ${\mathcal H}$, satisfying
the canonical commutation relation
\begin{equation}
[A, B]=-i\bbbone
\label{CCR}
\end{equation}
on a non-zero subspace $\mathcal D\subset {\rm D}(AB) \cap {\rm D}(BA)$, where, for a linear operator $L$ on ${\mathcal H}$, ${\rm D}(L)$ denotes the domain
of $L$. We call $\mathcal D$ a {\it CCR-domain} for the pair $(A,B)$.
It is well known \cite[p.2]{pu67} that, if $\mathcal D$ is dense in ${\mathcal H}$,
then (\ref{CCR})
implies that ${\mathcal H}$ has to be infinite dimensional and at least one of $A$ and $B$ is unbounded. We call this property the {\it unbounded property} of CCR.
It is easy to see that, if $\mathcal D$ is an invariant subspace of $A$ and $B$, then
$\mathcal D$ has to be infinite dimensional and hence at least one of $A$ and $B$ as linear operators on $\overline{\mathcal D}$
(the closure of $\mathcal D$) with domain $\mathcal D$ is unbounded.
From representation theoretic point of view,
$({\mathcal H}, \mathcal D, \{A,B\})$ is called a {\it representation of the CCR with one degree of freedom} (usually $\mathcal D$ is assumed to be
a dense invariant subspace of $A$ and $B$, but, here, we do not require this property).
W denote by $(f,g)_{{\mathcal H}}$ ($f,g\in {\mathcal H}$) and $\|\cdot \|_{{\mathcal H}}$
the scalar (inner) product of ${\mathcal H}$, linear in $g$ and antilinear in $f$, and the norm of ${\mathcal H}$ respectively.
But we sometimes omit the subscript \lq\lq{${\mathcal H}$}" in $(f,g)_{{\mathcal H}}$ and $\|\cdot\|_{{\mathcal H}}$ if there is no
danger of confusions.
The CCR (\ref{CCR}) implies an physically important inequality:
if $A$ and $B$ in (\ref{CCR}) are symmetric operators on ${\mathcal H}$, then (\ref{CCR}) yields the {\it uncertainty relation}
of Heisenberg type \cite[Chapter III, \S 4]{vn32}:
\begin{equation}
(\Delta A)_{\psi}(\Delta B)_{\psi}\geq \frac{1}{2}\label{UR}
\end{equation}
for all $\psi\in \mathcal D$ with $\|\psi\|=1$, where
\begin{equation}
(\Delta A)_{\psi}:=\|(A-(\psi, A\psi)_{\mathcal H})\psi\|_{\mathcal H},\,\psi\in{\rm D}(A), \,\, \|\psi\|=1, \label{Delta-A}
\end{equation}
the {\it uncertainty} of $A$ with respect to
$\psi$.\footnote{Inequality (\ref{UR}) can be derived also from a weak version of (\ref{CCR}):
$(A\psi,B\phi)-(B\psi,A\phi)=-i(\psi,\phi),\,\psi,\phi \in \mathcal D_{\rm w}$, where $\mathcal D_{\rm w}$ is a non-zero
subspace of ${\rm D}(A)\cap{\rm D}(B)$.}
The concept of representation of the CCR with one degree of freedom
can be extended to the case of finite degrees of freedom.
Let $A_j$ and $B_j$ ($j, k=1,\ldots, d, d\in \mathbb{N}$) be symmetric operators on ${\mathcal H}$ and $\mathcal D$ be
a non-zero subspace of ${\mathcal H}$ such that $\mathcal D \subset \cap_{j,k=1}^d[{\rm D}(A_jB_k)\cap {\rm D}(B_kA_j)\cap {\rm D}(A_jA_k)\cap {\rm D}(B_jB_k)]$.
Then the triple $({\mathcal H},\mathcal D, \{A_j,B_j|j=1,\ldots,d\})$
is called a {\it representation of the CCR's with $d$ degrees of freedom}
if the CCR's with $d$ degrees of freedom
\begin{equation}
[A_j,B_k]=-i\delta_{jk}\bbbone,\quad [A_j,A_k]=0,\quad [B_j,B_k]=0,\quad j,k=1,\ldots,d\label{CCR-d}
\end{equation}
hold on $\mathcal D$, where $\delta_{jk}$ is the Kronecker delta. The subspace $\mathcal D$ is called a {\it CCR-domain}
for $\{A_j,B_j|j=1,\ldots,d\}$.
There is a stronger version of representation of the CCR's with $d$ degrees of freedom.
A set $\{A_j,B_j|j=1,\ldots,d\}$ of self-adjoint operators on ${\mathcal H}$ is
called a {\it Weyl representation of the CCR's with $d$ degrees of freedom}
if the {\it Weyl relations}
\begin{equation}
e^{-is A_j} e^{-itB_{k}}=e^{ist\delta_{jk}}e^{-itB_{k}}e^{-isA_j},\quad j,k=1,\ldots,d,\, s,t\in{\mathbb{R}}\label{Weyl}
\end{equation}
hold.
The Weyl relations (\ref{Weyl}) imply that there exists a dense invariant domain $\mathcal D$ of $A_j$ and $B_j$
($j=1,\ldots,d$) such that (\ref{CCR-d}) holds on $\mathcal D$ \cite[Theorem 4.9.1]{pu67}.
Hence the Weyl representation $\{A_j,B_j|j=1,\ldots,d\}$
is a representation of the CCR's with $d$ degrees of freedom.
But {\it the converse is not true} (e.g., \cite{fug67,sch83b,ara98}).
A Weyl representation $ \{A_j, B_{k}| j, k=1,\ldots, d\}$ of the CCR's with $d$ degrees of freedom
is said to be {\it irreducible} if any subspace $\mathcal D$ of ${\mathcal H}$ left invariant
by $e^{-itA_j}$ and $e^{-itB_j}$ for all $t\in {\mathbb{R}}$ and $j=1,\ldots,d$ is $\{0\}$ or ${\mathcal H}$.
In quantum mechanics on the $d$-dimensional space
$${{\mathbb{R}}^d}=\{x=(x_1,\ldots,x_d)|x_j\in {\mathbb{R}}\},$$
the momentum operator $P:=(P_1,\ldots,P_d)$ and the position operator $Q:=(Q_1,\ldots,Q_d)$
are defined by $P_j:=-iD_j$ ($D_j$ is the generalized partial differential operator
in $x_j$) and
$Q_j:=M_{x_j} (\mbox{the multiplication operator by }x_j)$, \, $j=1,\ldots, d$.
For all $j=1,\ldots,d$, $P_j$ and $Q_j$ are
self-adjoint operators on the Hilbert space ${L^2(\BR)}$, satisfying the CCR's with $d$ degrees of freedom:
\eq{ccr2}
[P_j, Q_{k}]=-i\delta_{jk}\bbbone, \quad
[P_j, P_{k}]=0, \quad
[Q_j, Q_{k}]=0
\end{equation}
on the domain $\cap_{j,k=1}^d[{\rm D} (P_jQ_{k})\cap {\rm D} (Q_{k}P_j) \cap {\rm D}(Q_jQ_{k})\cap {\rm D}(Q_{k}Q_j) ]$.
Hence $({L^2(\BR)}$, $C_0^{\infty}({{\mathbb{R}}^d})$, $\{P_j,Q_j|j=1,\ldots,d\})$
is a representation of the CCR's with $d$ degrees of freedom, where $C_0^{\infty}({{\mathbb{R}}^d})$ is
the space of infinitely differentiable functions on ${{\mathbb{R}}^d}$ with compact support.
This representation of CCR's is called the {\it Schr\"odinger representation of the CCR}
(or the {\it Schr\"odinger system} \cite{pu67}) {\it with $d$ degrees of freedom}.
By an application of (\ref{UR}), one obtains
the {\it position-momentum
uncertainty relations}
\begin{equation}
(\Delta P_j)_{\psi}( \Delta Q_j)_{\psi} \geq \frac{1}{2},\quad j=1,\ldots,d\label{UR1}
\end{equation}
for all $\psi\in {\rm D}(P_jQ_j)\cap {\rm D}(Q_jP_j)$ with $\|\psi\|=1$, basic inequalities in quantum
mechanics which show a big difference between quantum mechanics and classical mechanics.\footnote{
Inequality (\ref{UR1}) holds also for all $\psi \in {\rm D}(P_j)\cap{\rm D}(Q_j)$ with $\|\psi\|=1$.}
The Schr\"odinger representation $ \{P_j, Q_{k}|j, k=1,\ldots, d\}$
is an irreducible Weyl representation (\cite[Theorem 4.5.1]{pu67}; \cite[Theorem 3.12]{ara06}).
Conversely it is known as the von Neumann uniqueness theorem (e.g.,
\cite[Theorem 4.11.1]{pu67}) that,
if ${\mathcal H}$ is separable and $ \{A_j, B_{k}|j, k=1,\ldots,d\}$ is an irreducible Weyl representation
of the CCR's with $d$ degrees of freedom,
then
$$
{\mathcal H}\cong {L^2(\BR)}, \quad
A_j\cong P_j, \quad B_j\cong Q_j, \quad j=1,...,d. $$
Here
$\cong$ denotes a unitary equivalence.
Usually models of quantum mechanics in ${{\mathbb{R}}^d}$ are constructed from the Sch\"odinger
representation of the CCR's with $d$ degrees of freedom. In this case,
physical quantities, which are required to be represented by self-adjoint operators on ${L^2(\BR)}$, are made from $P_j$ and $Q_j$, $j=1,\ldots,d$.
Among others, the Hamiltonian of a model, which describes the total energy of the quantum system under consideration, is important.
The classical Hamiltonian of a non-relativistic particle of mass $m$ in a potential $V:{{\mathbb{R}}^d} \to {\mathbb{R}}$ is
given by $ H_{\rm cl}(p,x)=p^2/2m+V(x),\, (p,x)\in {{\mathbb{R}}^d}\times {{\mathbb{R}}^d}$.
Then the corresponding quantum Hamiltonian
is given by
the Schr\"odinger operator (or the Schr\"odinger Hamiltonian)
$$
{H_V}:=H_{\rm cl}(P, Q):=\frac 1{2m} \sum_{j=1}^d P_j^2+V (Q)=-\frac 1{2m}\Delta+V(Q)
$$
on
${L^2(\BR)}$, where
$V (Q)$ is defined by the functional calculus
using the joint spectral measure of $Q_1, \cdots, Q_d$ (note that $(Q_1,\ldots,Q_d)$
is a set of strongly commuting self-adjoint operators\footnote{
A set $\{A_1,\ldots,A_n\}$ of self-adjoint operators on a Hilbert space
is said to be {\it strongly commuting} if the spectral measure $E_{A_j}$ of $A_j$
commutes with $E_{A_k}$ for all $j,k=1,\ldots,n, j\not=k$ (i.e.,
for all Borel sets $J,K\subset {\mathbb{R}}$, $E_{A_j}(J)E_{A_k}(K)=
E_{A_k}(K)E_{A_j}(J)$).}) and $\Delta:=\sum_{j=1}^dD_j^2$ is the $d$-dimensional
generalized Laplacian. It is shown in fact that $V(Q)$ is the multiplication operator by
the function $V$ . Hence one simply denotes $V(Q)$ by $V$. Thus
\begin{equation}
{H_V}=H_0+V, \label{HV}
\end{equation}
where
\begin{equation}
H_0:=-\frac 1{2m}\Delta.\label{H0}
\end{equation}
In general, according to an axiom of quantum mechanics due to von Neumann,
the time evolution of the quantum system whose Hamiltonian is given by a self-adjoint operator $H$ on
a Hilbert space ${\mathcal H}$ is described by the unitary operator $e^{-itH}$ with time parameter
$t\in{\mathbb{R}}$ in such a way that, if $\phi\in {{\mathcal H}}$ is a state vector at $t=0$, then
the state vector at time $t$ is given by $\phi_t=e^{-itH}\phi$, provided that no measurement is made for
the quantum system under consideration in the time
interval $[0,t]$.
If $\phi\in {\rm D}(H)$, then
$\phi_t$ is strongly differentiable in $t$, $\phi_t\in D(H)$ for all $t\in{\mathbb{R}}$, and
obeys the abstract Schr\"odinger equation
$$
i\frac{d \phi_t}{dt}= H\phi_t.
$$
Here time $t$ is usually treated as a parameter, not as an operator.
It is the external time mentioned in Subsection \ref{subsec-time}.
In relativistic classical mechanics, the energy variable is regarded as the variable canonically conjugate to
the time variable
as so is the momentum variable to the position variable and this may be extended to non-relativistic classical mechanics
as a limit of relativistic one.
From this point of view (or in view of the time-energy uncertainty relation
proposed by Heisenberg), one may infer that a quantum Hamiltonian $H$
may have a symmetric operator $T$ corresponding to time, satisfying CCR
\begin{equation}
[H,T]=-i\bbbone\label{CCR2}
\end{equation}
on a non-zero subspace $\mathcal D_{H,T}$ included in ${\rm D}(HT)\cap {\rm D}(TH)$.
Such an operator $T$ is called a {\it time operator} of $H$ (some authors
use the form $[H,T]=i\bbbone$ instead of (\ref{CCR2}), but this is not essential, just a convention).
From a purely mathematical point of view (apart from the context of quantum physics), this definition
applies to any pair $(H,T)$ of a self-adjoint operator $H$ and a symmetric operator $T$
obeying (\ref{CCR2}) on a non-zero subspace included in ${\rm D}(HT)\cap {\rm D}(TH)$.
\begin{remark}\label{rem-top}{\rm It is obvious that, if $T$ is a time operator of $H$, then,
for all $\alpha\in {\mathbb{R}}\setminus\{0\}$, $\alpha^{-1}T$ is a time operator of $\alpha H$.}
\end{remark}
The uncertainty relation
\begin{equation}
(\Delta H)_{\psi}(\Delta T)_{\psi}\geq \frac 12,\quad \psi\in \mathcal D_{T,H},\,\|\psi\|=1\label{TE-ur}
\end{equation}
implied by (\ref{CCR2}) may be interpreted as a form of {\it time-energy
uncertainty relation}. The time operator $T$ is physical
in the sense that it gives a lower bound for the uncertainty $(\Delta H)_{\psi}$
of $H$ with respect to the state $\psi\in \mathcal D_{T,H}$.
In the physics literature, formal (heuristic) constructions of \lq\lq{time operators}" have been done
for special classes of Schr\"odiner Hamiltonians (e.g., \cite{ab61,fuj80,fwy80,gys81,bau83}).
But, since the theory of CCR's with dense CCR-domains involves unbounded operators as remarked above,
formal manipulations are questionable and results based on them remain vague and inconclusive.
In fact, mathematically rigorous considerations lead one to
distinguish some classes of time operators as recalled below. These classes correspond
to different types of representations of CCR's (see, e.g., \cite{fug67,jm80,sch83a,sch83b,dor84}).
It should be noted that
there exist representations of CCR's which are inequivalent to Schr\"odinger ones (e.g., \cite{fug67},
\cite{sch83b}, \cite{ara98}) and, interestingly enough, some of them are connected with characteristic
physical phenomena such as the so-called Aharonov-Bohm effect (see \cite{ara98} and references therein).
Mathematically rigorous studies on time operators, including
general theories of time operators (not necessarily restricted to time operators of Schr\"odinger operators), have been made by some authors (e.g.,
\cite{miy01, gal02, gcb04, ara05, ara07, ara08a, ara08b, am08a, am08b, ara09, hkm09} and references therein; see also \cite{jm80, sch83a, sch83b,dor84} for earlier studies from purely mathematical points of view).
The present paper is a continuation of those studies, in particular, concentrating
on constructions of time operators
{\it in a generalized sense} associated with a class of Schr\"odinger
operators which contains the Hamiltonian of the hydrogen atom.
Let $H$ be a self-adjoint operator on ${\mathcal H}$ and bounded from below. Then
the von Neumann uniqueness theorem tells us that
there exists no self-adjoint operator $T$
such that pair $ (H, T)$ satisfies the Weyl relation (\ref{Weyl}) with $d=1$,
since $\sigma}%{\rm Sp} (P)={\mathbb{R}}$ and then $H\not\cong P$, where, for a linear operator $L$, $\sigma}%{\rm Sp}(L)$ denotes the spectrum of $L$.
Thus, to treat such a case, it is natural to introduce a weaker version
of the Weyl representation with one degree of freedom to define a class of
time operators.
\bd{wwr}\TTT{weak Weyl relation}
{\rm
A pair $ (A, B)$ consisting of a self-adjoint operator $A$ and symmetric operator $B$ on ${\mathcal H}$
is called a {\it weak Weyl representation} with one degree of freedom
if
$e^{-itA} {\rm D} (B)\subset {\rm D} (B)$ for all $t\in {\mathbb{R}}$ and the {\it weak Weyl relation}
\begin{equation}
B e^{-itA}\psi =e^{-itA} (B+t)\psi \label{w-Weyl}
\end{equation}
holds for all $\psi\in {\rm D} (B)$ and all $t\in{\mathbb{R}}$.
}
\end{definition}
Studies on this class of representations from purely mathematical points of view have been
done in \cite{jm80, sch83a, sch83b,dor84}.
It is easy to see that a Weyl representation $\{A,B\}$
is a weak Weyl representation and that
the weak Weyl relation (\ref{w-Weyl}) implies the CCR (\ref{CCR}) on ${\rm D}(AB)\cap {\rm D}(BA)$.
But one should note that a weak Weyl representation $(A,B)$
with both $A$ and $B$ being self-adjoint is not necessarily a Weyl representation.
\bd{1100}
\TTT{strong time operator}
{\rm
A symmetric operator
$T$ on ${\mathcal H}$ is called a {\it strong time operator} of
a self-adjoint operator $H$ on ${\mathcal H}$ if
$ (H, T)$ is a weak Weyl representation.
}
\end{definition}
\begin{remark}\label{rem-u-strong}{\rm (1) In relation to strong time operators,
it may be convenient to give a name to a self-adjoint operator $T$ on ${\mathcal H}$
such that $(H,T)$ is a Weyl representation of the CCR
with one degree of freedom. We call such an operator $T$ an {\it ultra-strong time operator} of $H$.
It follows that an ultra-strong time operator is a strong time operator. But the converse is not true.
If ${\mathcal H}$ is separable, then, by the von Neumann uniqueness theorem, $(H,T)$ is unitarily equivalent to the
direct sum of the Schr\"odinger representation $(P,Q)$ with $d=1$.
(2) It is well known or easy to see that, if $(H,T)$ is a Weyl representation of the CCR with one degree of
freedom, then $\sigma(H)=\sigma(T)={\mathbb{R}}$ (for this fact, separability of ${\mathcal H}$ is not assumed).
Hence a semi-bounded self-adjoint operator (i.e. a self-adjoint operator which is
bounded from below or above) has no ultra-strong time operators.
}
\end{remark}
As far as we know,
a firm mathematical investigation of a strong time operator was initiated by \cite{miy01}, although
the name \lq\lq{strong time operator}" is not used in \cite{miy01} (it was introduced first in
\cite{ara08b} to
distinguish different classes of time operators). Further investigations and generalizations on strong time operators
were done in
\cite{ara05, ara07}. See also
\cite{am08a, am08b, hkm09,rt09}. It is known that, if $ (H, T)$ satisfies the weak Weyl relation,
then $\sigma}%{\rm Sp}(H)$ is purely absolutely continuous \cite{sch83a}.
Hence, if $H$ has an eigenvalue, then $H$ has no strong time operator.
In the context of quantum physics, in addition to time-energy uncertainty relation (\ref{TE-ur}),
a strong time operator $T$ of a Hamiltonian $H$ may have properties richer
than those of time operators of $H$. For example,
it controls decay rates in time $t\in{\mathbb{R}}$ of transition probabilities $|(\phi, e^{-itH}\psi)|^2$ ($\phi,\psi\in {\mathcal H},
\|\phi\|=\|\psi\|=1$) in the following form \cite[Theorem 8.5]{ara05}: for each natural number
$n\in \mathbb{N}$ and all unit vectors $\phi,\psi\in {\rm D}(T^n)$, there exists a constant $d_n^T(\phi,\psi)\geq 0$ such that,
for all $t\in {\mathbb{R}}\setminus\{0\}$,
$$
|(\phi,e^{-itH}\psi)|^2\leq \frac{d_n^T(\phi,\psi)^2}{|t|^{2n}}.
$$
This shows a very interesting correspondence between decay rates in time of
transition probabilities and regularities of state vectors $\phi, \psi$.\footnote{
Here we mean by \lq\lq{regularity}" of a vector $\psi$ the number $n$ such that $\psi\in {\rm D}(T^n)$.
} It tells us also the importance of domains of time operators.
In \cite{gal02,am08b}, a time operator of a self-adjoint operator whose spectrum is purely discrete
with a growth condition is constructed. In \cite{ara09}, necessary and sufficient conditions for a self-adjoint
operator with purely discrete spectrum to have a time operator were given.
From these investigations, it is suggested that the concept of time operator
should be weakened for a self-adjoint operator (a Hamiltonian in the context of quantum mechanics)
whose spectrum is not purely absolutely continuous and whose discrete spectrum does not satisfy
conditions formulated in \cite{ara09}.
One of weaker versions of time operator is defined as follows:
\bd{weakdef}\TTT{weak time operator}
{\rm A symmetric operator $T$ on ${\mathcal H}$ is called
a {\it weak time operator} of
a self-adjoint operator $H$ on ${\mathcal H}$ if there exists a non-zero subspace $\mathcal D_{\rm w}
\subset {\rm D}(T)\cap{\rm D}(H)$ such that the {\it weak CCR} on $\mathcal D_{\rm w}$ holds:
\begin{equation}
(H\phi, T\psi)- (T\phi, H\psi)=-i (\phi, \psi),\quad \phi,\psi\in \mathcal D_{\rm w}. \label{w-CCR}
\end{equation}
We call $\mathcal D_{\rm w}$ a {\it weak-CCR domain} for the pair $(H,T)$.
}
\end{definition}
It is obvious that a time operator $T$ of $H$ is a weak time operator
of $H$ with $\mathcal D_{\rm w}=\mathcal D_{H,T}$.
We remark that (\ref{w-CCR}) implies the time-energy uncertainty relation (\ref{TE-ur})
with $\psi\in\mathcal D_{\rm w}$ ($\|\psi\|=1$).
One should keep in mind the following fact:
\begin{proposition}\label{prop-w-time} Let $T$ be a weak time operator of a self-adjoint operator $H$ and
$\mathcal D_{\rm w}$ be a weak-CCR domain for $(H,T)$. Then $H$ has no eigenvectors in $\mathcal D_{\rm w}$.
\end{proposition}
\proof Let $H\psi=E\psi$ with $\psi\in \mathcal D_{\rm w}$ and $E\in {\mathbb{R}}$.
Taking $\phi $ in (\ref{w-CCR}) to be $\psi$, we see that the left hand side is equal to 0. Hence
$\|\psi\|^2=0$, implying $\psi=0$.\qed
\begin{remark}{\rm Unfortunately we do not know whether or not there exists a weak time operator which cannot
be a time operator. We leave this problem for future study.
}
\end{remark}
\subsection{Ultra-weak time operator}\label{subsec-uwt}
Proposition \ref{prop-w-time} implies that, if
a self-adjoint operator $H$ with an eigenvalue $E$ has a weak time operator,
then all the eigenvectors of $H$ with eigenvalue $E$ are out of any weak-CCR domain for $(H,T)$.
On the other hand, $H$ may have a complete set of eigenvectors so that
the subspace algebraically spanned by the eigenvectors of $H$ is dense in ${\mathcal H}$.
This suggests that
such a self-adjoint operator may have tendency not to have a weak time operator.
Taking into account this possibility and in the spirit of seeking ideas as general as possible,
we generalize the concept of weak time operator:
\begin{definition}[ultra-weak time operator]\label{def-uwt}{\rm Let $H$ be a self-adjoint operator on ${\mathcal H}$ and
$\mathcal D_1$ and $\mathcal D_2$ be non-zero subspaces of ${\mathcal H}$.
A sesquilinear form ${\mathfrak{t}}:\mathcal D_1 \times \mathcal D_2\to{\mathbb{C}}$ ($\mathcal D_1\times\mathcal D_2\ni (\phi,\psi)\mapsto {\mathfrak{t}}[\phi,\psi]\in{\mathbb{C}}$)
with domain ${\rm D}({\mathfrak{t}})=\mathcal D_1\times \mathcal D_2$
(${\mathfrak{t}}[\phi,\psi]$ is antilinear in $\phi$ and linear in $\psi$)
is called {\it an ultra-weak time operator} of $H$ if
there exist non-zero subspaces $\mathcal D$ and
$\ms E$ of $\mathcal D_1\cap\mathcal D_2$ such that
the following (i)--(iii) hold:
\begin{list}{}{}
\item[(i)] $\ms E \subset D(H)\cap \mathcal D$.
\item[(ii)] (symmetry on $\mathcal D$) ${\mathfrak{t}}[\phi,\psi]^*={\mathfrak{t}}[\psi,\phi],\, \phi,\psi\in \mathcal D$.
\item[(iii)] (ultra-weak CCR) $H\ms E\subset \mathcal D_1$ and, for all $\psi, \phi\in \ms E$,
\begin{equation}
{\mathfrak{t}}[H\phi,\psi] -{\mathfrak{t}}[H\psi,\phi]^*=-i(\phi,\psi)\label{TH}
\end{equation}
\end{list}
We call $\ms E$ an {\it ultra-weak CCR-domain} for $(H,{\mathfrak{t}})$
and $\mathcal D$ a {\it symmetric domain} of ${\mathfrak{t}}$.
}
\end{definition}
\begin{remark}\label{rem-uwt}{\rm
(1) As far as we know, the concept \lq\lq{ultra-weak time operator}" introduced here is new.
(2) Although there may be no operators associated with the sesquilinear form ${\mathfrak{t}}$
in the above definition,
we use, by abuse of word, \lq\lq{ultra-weak time operator}" to indicate
that it is a concept weaker than that of weak time operator as shown below.
Let $T$ be a weak time operator of $H$ with a weak CCR-domain $\mathcal D_{\rm w}$.
Then one can define a sesquilinear form ${\mathfrak{t}}_T:{\mathcal H} \times {\rm D}(T)
\to {\mathbb{C}}$ by
$$
{\mathfrak{t}}_T[\phi,\psi]:=(\phi,T\psi),\quad \phi\in{\mathcal H}, \psi\in {\rm D}(T).
$$
Then it is easy to see that
${\mathfrak{t}}_T[ \phi,\psi]^*={\mathfrak{t}}_T[ \psi,\phi],\,\psi,\phi\in {\rm D}(T)$ and, for all $\phi,\psi\in \mathcal D_{\rm w}$,
${\mathfrak{t}}_T[ H\phi,\psi]-{\mathfrak{t}}_T[ H\psi,\phi]^*=-i(\psi,\phi)$. Hence ${\mathfrak{t}}_T$ is an ultra-weak time operator
of $H$ with $\mathcal D_{\rm w}$ being an ultra-weak CCR-domain and ${\rm D}(T)$ a symmetry domain. Therefore the concept of ultra-weak time operator is a generalization of weak time
operator.
(3) If $H\psi\in \mathcal D$ in (\ref{TH}), then, by the symmetry of ${\mathfrak{t}}[\cdot,\cdot]$ on $\mathcal D$,
(\ref{TH}) takes the following form:
$$
{\mathfrak{t}}[ H\phi,\psi] -{\mathfrak{t}}[ \phi,H\psi]=-i(\phi,\psi)
$$
}
\end{remark}
For a sesquilinear form ${\mathfrak{t}}:\mathcal D_1\times \mathcal D_2 \to {\mathbb{C}}$ and a constant $a\in{\mathbb{R}}$, we define a sesquilinear form ${\mathfrak{t}}-a:\mathcal D_1\times \mathcal D_2\to{\mathbb{C}}$
by
$$
({\mathfrak{t}}-a)[\phi,\psi]:={\mathfrak{t}}[\phi,\psi]-a(\phi,\psi),\quad \phi\in \mathcal D_1,\psi\in \mathcal D_2.
$$
In the case of the pair $(H,{\mathfrak{t}})$
in Definition \ref{def-uwt}, the uncertainty relation (\ref{UR}) associated with CCR
is generalized as follows:
\begin{proposition}[uncertainty relation for $(H,{\mathfrak{t}})$] \label{U1}
Assume that $H$ has an ultra-weak time operator ${\mathfrak{t}}$ as
in Definition \ref{def-uwt}.
Then, for all $a,b\in{\mathbb{R}}$ and a unit vector $\psi\in\ms E$,
\begin{equation}
\left|({\mathfrak{t}}-a)[ (H-b)\psi,\psi]\right|\geq \frac 12.\label{TH1}
\end{equation}
\end{proposition}
\proof Using (\ref{TH}), we have $\Im \left\{({\mathfrak{t}}-a)[ (H-b)\psi,\psi]\right\}
=-\frac 1{2}$.
Since $|z|\geq |\Im z|$ for all $z\in{\mathbb{C}}$, (\ref{TH1}) follows.
\qed
In summary, we have seen that there exist five classes of time operators
with inclusion relation (\ref{hierarchy}).
\subsection{Outline of the present paper}
Having introduced the new concept \lq\lq{ultra-weak time operator},
we now outline the contents of the present paper.
In Section \ref{sec3}, we review an abstract theory of time operators
and give new additional results.
Among others, we prove an existence theorem
on a strong time operator of an absolutely continuous self-adjoint operator
(Theorem \ref{thm-ac-st}).
Sections \ref{sec4} is devoted to showing the existence of time operators of self-adjoint operators with purely discrete spectra.
This includes an extension of existence theorems on time operators
in \cite{gal02,am08b}.
In Section \ref{sec5}, we introduce a class $S_1({\mathcal H})$ of self-adjoint operators on ${\mathcal H}$ (see Definition \ref{def-S1}) such that
each element of $S_1({\mathcal H})$ has an ultra-weak time operator with a dense ultra-weak CCR-domain (Theorem \ref{thm-uwt1}).
Moreover, for a class of Borel measurable functions $f:{\mathbb{R}}\to{\mathbb{R}}$, we formulate
sufficient conditions for $f(H)$ to have an ultra-weak time operator (Corollary \ref{main3}).
In Section \ref{sec6}, we discuss applications of the abstract results to
the Schr\"odinger operator ${H_V}$.
We find classes of potentials $V$ for which
${H_V}$ has an ultra-weak time operator with a dense ultra-weak CCR-domain (Theorem \ref{thm-uwt2}).
Also we show that the Hamiltonian of the hydrogen atom
(i.e. the case where $V(x)=-\gamma/|x|, \, x\in {\mathbb{R}}^3\setminus\{0\}$ with a constant
$\gamma>0$) has an ultra-weak time operator with a dense ultra-weak CCR-domain (Example \ref{ex-hyd2}).
In the last section, for a class of $f$, an existence theorem on an ultra-weak time operator of $f({H_V})$
is proved (Theorem \ref{thm-uwt-HS}) and some examples are given.
\section{Abstract Theory of Time Operators--Review with Additional Results}
\label{sec3}
\subsection{A general structure of time operators}
We first note an elementary fact:
\begin{proposition}\label{prop-uni} Let $H$ be a self-adjoint operator on a Hilbert space ${\mathcal H}$ and $T$ be a time
operator of $H$ with a CCR-domain $\mathcal D$ for $(H,T)$.
Let $H'$ be a self-adjoint operator on a Hilbert space ${\mathcal H}'$ such that $UHU^{-1}=H'$ for a unitary
operator $U:{\mathcal H}\to{\mathcal H}'$. Then $T':=UTU^{-1}$
is a time operator of $H'$ with a CCR-domain $U\mathcal D$ for $(H',T')$.
\end{proposition}
\proof An easy exercise.
\qed
In what follows, $H$ denotes a self-adjoint operator on a complex Hilbert space ${\mathcal H}$.
As is well known (e.g., \cite[\S 10.1]{ka76}, \cite[Theorem VII.24]{rs1}), ${\mathcal H}$ has the orthogonal decomposition
\begin{equation}
{\mathcal H}={\mathcal H}_{\rm ac}(H)\oplus{\mathcal H}_{\rm sc}(H)\oplus {\mathcal H}_{\rm p}(H),
\end{equation}
where ${\mathcal H}_{\rm ac}(H)$ (resp. ${\mathcal H}_{\rm sc}(H)$, ${\mathcal H}_{\rm p}(H)$) is
the subspace of absolute continuity (resp. of singular continuity, of discontinuity) with respect to $H$,
and $H$ is reduced by each subspace ${\mathcal H}_{\#}(H)$ ($\#=$ ac, sc, p).
We denote the reduced part of $H$ to ${\mathcal H}_{\#}(H)$ by $H_{\#}$ and set
$$
\sigma_{\rm ac}(H):=\sigma(H_{\rm ac}),\quad \sigma_{\rm sc}(H):=\sigma(H_{\rm sc}),
$$
which are called the absolutely continuous spectrum and the singular continuous spectrum of $H$ respectively.
We denote by $\sigma_{\rm p}(H)$ the set of all eigenvalues of $H$. We remark
that $\sigma(H_{\rm p})=\overline{\sigma_{\rm p}(H)}$, the closure of $\sigma_{\rm p}(H)$.
We have
\begin{equation}
H=H_{\rm ac}\oplus H_{\rm sc}\oplus H_{\rm p}\label{H-decomp}
\end{equation}
and
$$
\sigma(H)=\sigma_{\rm ac}(H)\cup\sigma_{\rm sc}(H)\cup \overline{\sigma_{\rm p}(H)}.
$$
An eigenvalue of $H$ is called a discrete eigenvalue of $H$ if it is an isolated eigenvalue of $H$ with a finite
multiplicity. The set $\sigma_{\rm disc}(H)$ of all the discrete eigenvalues of $H$ is called the
discrete spectrum of $H$.
The following proposition shows that the problem of constructing time operators
of $H$ is reduced to that of constructing time operators of each $H_{\#}$.
\begin{proposition} Suppose that each $H_{\#}$ has a time operator $T_{\#}$ with a CCR-domain $\mathcal D_{\#}$. Then
the direct sum
$$
T:=T_{\rm ac}\oplus T_{\rm sc}(H)\oplus T_{\rm p}
$$
is a time operator of $H$ with a CCR-domain $\mathcal D_{\rm ac}\oplus \mathcal D_{\rm sc}\oplus \mathcal D_{\rm p}$.
\end{proposition}
\proof Since the direct sum of symmetric operators is again a symmetric operator in general,
it follows that $T$ is symmetric.
By the assumption, we have for all $\psi_{\#}\in \mathcal D_{\#}$
$$
[H_{\#},T_{\#}]\psi_{\#}=-i\psi_{\#}.
$$
Let $\psi=(\psi_{\rm ac},\psi_{\rm sc},\psi_{\rm p})\in \mathcal D_{\rm ac}\oplus \mathcal D_{\rm sc}\oplus \mathcal D_{\rm p}$.
Then, by (\ref{H-decomp}), $\psi \in D(HT)\cap D(TH)$ and
$$
[H,T]\psi=([H_{\rm ac},T_{\rm ac}]\psi_{\rm ac},[H_{\rm sc},T_{\rm sc}]\psi_{\rm sc},
[H_{\rm p},T_{\rm p}]\psi_{\rm p})=-i\psi.
$$
Hence $T$ is a time operator of $H$ with a CCR-domain $\mathcal D_{\rm ac}\oplus \mathcal D_{\rm sc}\oplus \mathcal D_{\rm p}$.
\qed
\subsection{Strong time operators}
\subsubsection{A summary of known results and additional results}
We summarize some basic facts on strong time operators of $H$.
\begin{proposition}\label{prop-st} A symmetric operator $T$ is a strong time operator of $H$ if and only if
operator equality $e^{itH}Te^{-itH}=T+t$ holds for all $t\in{\mathbb{R}}$.
\end{proposition}
\proof See \cite[Proposition 2.1]{ara05}.
\qed
Note that the operator equality given in this proposition
implies that, for all $t\in {\mathbb{R}}$, $e^{-itH}{\rm D}(T)={\rm D}(T)$.
\begin{proposition}\label{prop-st-uni} Let $T$ be a strong time operator of $H$
and $H'$ be a self-adjoint operator on a Hilbert space ${\mathcal H}'$ such that,
for a unitary operator $U:{\mathcal H}\to{\mathcal H}'$, $UHU^{-1}=H'$.
Then $T':=UTU^{-1}$ is a strong time operator of $H'$.
\end{proposition}
\proof By the functional calculus, for all $t\in {\mathbb{R}}$, $e^{itH'}=Ue^{itH}U^{-1}$. By this fact and
Proposition \ref{prop-st},
we have
$$
e^{itH'}T'e^{-itH'}=Ue^{itH}Te^{-itH}U^{-1}=U(T+t)U^{-1}=T'+t.
$$
Hence $T'$ is a strong time operator of $H'$.
\qed
\begin{proposition}[\cite{ara05}]
\label{time}
Suppose that $H$ has a strong time operator
$T$.
Then:
\begin{list}{}{}
\item{(1)} The closure $\overline{T}$ of $T$ is also a strong
time operator of $H$.
\item{(2)} If $H$ is semi-bounded, then $T$ is not essentially self-adjoint.
\item{(3)} The operator $H$ is absolutely continuous.
\end{list}
\end{proposition}
\begin{proposition}\label{prop-st1} Let $T_1,\ldots, T_n$ ($n\geq 2$) be strong time operators of $H$.
\begin{list}{}{}
\item{(1)} Let $S:=\sum_{k=1}^na_kT_k$ with $a_k\in{\mathbb{R}}$ ($k=1,\ldots,n$) satisfying \linebreak $\sum_{k=1}^na_k=1$.
Then, for all $t\in {\mathbb{R}}$, operator equality
\begin{equation}
e^{itH}Se^{-itH}=S+t\label{S}
\end{equation}
holds. In particular, if $\cap_{k=1}^nD(T_k)$ is dense, then $S$ is a strong time operator of $H$.
\item{(2)} For any pair $(k,\ell)$ with $k\not=\ell$ ($k,\ell=1,\ldots,n$),
$(T_k-T_{\ell})e^{itH}\psi=e^{itH}(T_k-T_{\ell})\psi$ for all $t\in {\mathbb{R}}$ and $\psi\in D(T_k)\cap D(T_{\ell})$.
\end{list}
\end{proposition}
\proof (1) By Proposition \ref{prop-st}, we have operator equalities
\begin{equation}
e^{itH}T_ke^{-itH}=T_k+t, \quad
t\in {\mathbb{R}}, k=1,\ldots,n. \label{Tk}
\end{equation}
Since $e^{itH}Se^{-itH}=\sum_{k=1}^ne^{itH}a_kT_ke^{-itH}$ (operator equality),
(\ref{Tk}) implies (\ref{S}).
If $\cap_{k=1}^d{\rm D}(T_k)$ is dense, then $S$ is a symmetric
operator and hence it is a strong time operator of $H$.
(2) This easily follows from (\ref{Tk}).
\qed
Proposition \ref{prop-st1}-(1) shows that any real convex combination $S$ of strong time operators of $H$
such that ${\rm D}(S)$ is dense is a strong time operator of $H$.
Let $\{H_1, \ldots, H_n\}$ be a set of strongly commuting self-adjoint operators
on ${\mathcal H}$. Then
$\sum_{j=1}^nH_j$
is essentially self-adjoint
and, for all $t\in {\mathbb{R}}$,
\begin{equation}
e^{it\overline{\sum_{j=1}^nH_j}}=\prod_{j=1}^ne^{itH_j},\label{expH}
\end{equation}
where the order of the product of
$e^{itH_1},\ldots, e^{itH_n}$ on the right hand side is arbitrary (this is due to the commutativity of
$e^{itH_j}$ and $e^{itH_k}$ ($j,k=1,\ldots,n$) which follows the strong commutativity of $\{H_1,\ldots,H_n\}$).
\begin{proposition}\label{prop-st2} Let $\{H_1,\ldots,H_n\}$ be as above and assume that, for some $j$,
$H_j$ has a strong time operator $T_j$ such that
$e^{itH_k}T_je^{-itH_k}=T_j$ for all $k\not=j$. Then $T_j$ is a strong time operator of $\overline{\sum_{j=1}^nH_j}$.
\end{proposition}
\proof By the present assumption and Proposition \ref{prop-st}, we have
operator equality $e^{itH_j}T_je^{-itH_j}=T_j+t$ for all $t\in {\mathbb{R}}$.
Hence, by (\ref{expH}) and the commutativity of the operators $e^{itH_k}, \, k=1,\ldots,n$,
we have
$$
e^{it\overline{\sum_{j=1}^nH_j}}T_je^{-it\overline{\sum_{j=1}^nH_j}}=
\left(\prod_{k\not=j}e^{itH_k}\right)(T_j+t)\left(\prod_{k\not=j}e^{-itH_k}\right)
=T_j+t.
$$
Thus the desired result follows.
\qed
Proposition \ref{prop-st2} may be useful
to find strong time operators of a self-adjoint operator which is given by the closure of the sum
of strongly commuting self-adjoint operators.
A variant of Proposition \ref{prop-st2} is formulated as follows.
Let $\{A_1,\ldots, A_n\}$ be a set of strongly commuting self-adjoint operators
on ${\mathcal H}$ such that each $A_j$ is injective. Suppose that each $A_j$ has a strong time operator $B_j$
such that, for all $j=1,\ldots,n$, ${\rm D}({B}_jA_j^{-1})\cap {\rm D}(A_j^{-1}{B}_j)$
is dense and, for all $t\in {\mathbb{R}}$,
$e^{itA_k}B_je^{itA_k}=B_j, \, k\not=j, k=1,\ldots,n$.
By the strong commutativity of $\{A_1,\ldots,A_n\}$,
the operator
$$
H_A:=\sum_{j=1}^nA_j^2
$$
is a non-negative self-adjoint operator.
For each $j=1,\ldots,n$, the operator
$$
T_j:=\frac 14 \left(\overline{B}_jA_j^{-1}+A_j^{-1}\overline{B}_j\right)
$$
is symmetric.
\begin{proposition}[\cite{ara05}]\label{prop-ara05}
For each $j=1,\ldots,n$, $T_j$ is a strong time operator of $H_A$.
\end{proposition}
A general scheme to construct strong time operators for a given pair $(H,T)$ of a weak Weyl representation
is described in \cite[\S 10]{ara05}. A generalization of this scheme is given as follows.
By the functional calculus, for any real-valued continuous function $f$ on ${\mathbb{R}}$,
$f (H)$ is a self-adjoint operator on ${\mathcal H}$.
Then a natural question is: does $f(H)$ has a strong time operator ?
A heuristic argument to answer the question is as follows. Let $f\in C^1({\mathbb{R}})$ and denote the derivative of $f$ by $f'$. We have
$[T, H]=+i\bbbone $, which intuitively implies that $T=+id/dH$.
Hence we may formally see that
$[T, f (H)]=if' (H)$(in \cite[Theorem 6.2]{ara05}, this is justified for
all $f\in C^1({\mathbb{R}})$ such that $f$ and $f'$ are bounded), and then
$Te^{-itf (H)}=e^{-itf (H)} (T+tf' (H))$ holds.
Multiplying
$f' (H) ^{-1}$ on the both sides, we may have
$T f' (H) ^{-1} e^{-itf (H)}=
e^{-itf (H)} (Tf' (H) ^{-1}+t)$, and, by symmetrizing $T f' (H) ^{-1}$,
we expect that
$\frac{1}{2} (T f' (H) ^{-1}+f' (H) ^{-1}T)$ is
a strong time operator of $f (H)$.
Actually this result is justified under some conditions:
\begin{proposition}[{\cite[Theorem 1.9]{hkm09}}]\label{prop-hkm}
\label{hkm09}
Let $K$ be a closed null subset of ${\mathbb{R}}$ with respect to the Lebesgue measure.
Assume that $f\in C^2 ({\mathbb{R}}\setminus K)$ and
$L:=\{\lambda\in {\mathbb{R}}\setminus K| f' (\lambda)=0\}$ is a null set with respect to the Lebesgue measure.
Suppose that $H$ has a strong time operator $T_H$ which is closed and let
$$
D:=\{g(H) {\rm D} (T_H)| g\in C_0^\infty ({\mathbb{R}}\setminus L\cup K)\}.
$$
Then
$$
T_{f (H)}:=\frac{1}{2}\overline{(T_Hf' (H)^{-1}+f' (H)^{-1}T_H) \lceil D}
$$
is a strong time operator of $f (H)$,
where, for a linear operator $L$ and a subspace $\mathcal D\subset {\rm D}(L)$,
$L\lceil \mathcal D$ denotes the restriction of $L$ to $\mathcal D$.
\end{proposition}
\begin{example}[Aharonov-Bohm time operator]\label{ex-AB}{\rm Let $m>0$ be a constant.
Then it is obvious that $\sqrt{2m}Q_j$ is a strong time operator of $P_j/\sqrt{2m}$
in the Hilbert space ${\mathcal H}=L^2({{\mathbb{R}}^d})$.
Consider the function $f(\lambda)=\lambda^2,\,\lambda\in {\mathbb{R}}$. Then $f'(\lambda)=2\lambda$.
Hence $\{\lambda\in{\mathbb{R}}|f'(\lambda)=0\}=\{0\}$.
Therefore the subspace $D$ in Proposition \ref{prop-hkm} takes the form
$D_{{\rm AB},j}:={\rm L.H.}\{g(P_j){\rm D}(Q_j)|g\in C_0^{\infty}({\mathbb{R}}\setminus \{0\})\}$.
Hence, letting
$$
T_{{\rm AB},j}:=\frac m2\left(Q_jP_j^{-1}+P_j^{-1}Q_j\right),
$$
the operator
$$
\widetilde T_{{\rm AB},j}:=\overline{T_{{\rm AB},j}\lceil D_{{\rm AB},j}}
$$
is a strong time operator of $P_j^2/2m$.
Since $(P_1,\ldots,P_d)$ is a set of strongly commuting self-adjoint operators,
it follows from Proposition \ref{prop-ara05} that
$\widetilde T_{{\rm AB},j}$ is a strong time operator of $H_0$.
There is another domain on which $T_{{\rm AB},j}$ becomes a strong time operator of $H_0$ \cite{ara07}.
Let
$$\Omega_j:=\{k\in {{\mathbb{R}}^d}| k_j\not=0\}, \quad D_{{\rm AB},j}':=\{f\in L^2({{\mathbb{R}}^d})|\hat f\in C_0^{\infty}(\Omega_j)\},$$
where $\hat f$ is the $L^2$-Fourier transform of $f$.
Then $D_{{\rm AB},j}'$ is dense. Moreover, by using the Fourier analysis, it is shown that
the operators $Q_j,P_j^{-1}, e^{itP_j^2/2m}$ and
$e^{itH_0}$ ($\forall t\in{\mathbb{R}}$) leave $D_{{\rm AB},j}'$ invariant and, for all $t\in {\mathbb{R}}$,
$e^{itH_0}T_{{\rm AB},j}e^{-itH_0}=T_{{\rm AB},j}+t$ on $D_{{\rm AB},j}'$.
Hence
$$
T_{{\rm AB},j}':=T_{{\rm AB},j}\lceil D_{{\rm AB},j}'
$$ is a strong time operator of
$H_0$. We note that ${\rm D}(Q_j)\supset D_{{\rm AB},j}'$. Hence, for each $g\in C_0^{\infty}({\mathbb{R}}\setminus \{0\})$,
$g(P_j){\rm D}(Q_j)\supset g(P_j)D_{{\rm AB},j}'$. For any $g\in C_0^{\infty}({\mathbb{R}}\setminus\{0\})$
such that $\hat g(k_j)>0,\forall k_j\in {\mathbb{R}}$, $g(P_j)D_{{\rm AB},j}'=D_{{\rm AB},j}'$. It is not so difficult to show
that such a function $g$ exists.
Therefore $D_{{\rm AB},j}\supset D_{{\rm AB},j}'$ in fact.
A time operator of $H_0$ obtained as a restriction of $T_{{\rm AB},j}$ to a subspace or its closure
is called an {\it Aharonov-Bohm time operator} \cite{ab61,miy01}.
}
\end{example}
\begin{example}{\rm As a generalization of Aharonov-Bohm time operators,
one can construct strong time operators of a self-adjoint operator $H$ of the form $H=F(P)$ with $F\in C^1({{\mathbb{R}}^d})$,
which includes the free relativistic Schr\"odinger Hamiltonian $(-\Delta +m^2)^{1/2}$ ($m>0$)
and its fractional version $(-\Delta+m^2)^{\alpha}$ ($\alpha>0$).
This approach can be applied also to constructions of strong time operators of Dirac type operators \cite{th92}.
See \cite[\S 11]{ara05}) for the details. }
\end{example}
\subsubsection{Existence of a strong time operator for a class of absolutely continuous self-adjoint
operators}
\label{4.1}
As already mentioned, a self-adjoint operator which has a strong time operator
is absolutely continuous. Then a natural question is: does an absolutely continuous self-adjoint
operator have a strong time operator ?
To our best knowledge, this question has not been answered
in an abstract framework.
In what follows, we give a partial affirmative answer to the question.
We recall an important concept.
For a linear operator $A$ on a Hilbert space ${\mathcal H}$,
a non-zero vector $\phi\in \cap_{n=1}^{\infty}{\rm D}(A^n)$
is called a {\it cyclic vector} for $A$ if
$${\rm L.H.}\{A^n\phi|n\in \{0\}\cup\mathbb{N}\}$$
is dense in ${\mathcal H}$, where, for a subset $\mathcal D$ of ${\mathcal H}$, ${\rm L.H.}\mathcal D$ denotes
the algebraic linear hull of vectors in $\mathcal D$.
We denote by $E_H$ the spectral measure of $H$.
For a non-zero vector $\psi\in{\mathcal H}$, a measure $\mu_{\psi}$ on ${\mathbb{R}}$ is defined by
$$
\mu_{\psi}(B):=\|E_H(B)\psi\|^2,\quad B\in \bm{B},
$$
where $\bm{B}$ is the family of Borel sets of ${\mathbb{R}}$.
We define a function $X$ on ${\mathbb{R}}$ by
$$
X(\lambda):=\lambda,\quad \lambda\in{\mathbb{R}}.
$$
We note the following fact:
\begin{lemma}\label{lem-c1} Assume that ${\mathcal H}$ is separable. Suppose that $H$ has a cyclic vector $\phi$.
Then there exists a unitary operator $U$ from ${\mathcal H}$ to $L^2({\mathbb{R}},d\mu_{\phi})$
such that $U\phi=1$ and $UHU^{-1}=M_X$, the multiplication operator by the function $X$ acting in
$L^2({\mathbb{R}},d\mu_{\phi})$. Moreover,
the subspace ${\rm L.H.}\{e^{itX}|t\in {\mathbb{R}}\}$ is dense in
$L^2({\mathbb{R}},d\mu_{\phi})$.
\end{lemma}
\proof The first half of the lemma follows from an easy extension of Lemma 1 in \cite[\S VII.2]{rs1}
to the case of unbounded self-adjoint operators \cite[Theorem 1.8]{ara06}.
To prove the second half of the lemma, we note that, by the cyclicity of $\phi$ for $H$,
${\rm L.H.}\{H^n\phi|n\in \{0\}\cup\mathbb{N}\}$ is dense in ${\mathcal H}$.
By the functional calculus, we have
$$
\lim_{t\to 0}(-i)^n\left(\frac{e^{itH}-1}{t}\right)^n\phi=H^n\phi.
$$
Hence it follows that ${\rm L.H.}\{e^{itH}\phi|t\in{\mathbb{R}}\}$ is dense in ${\mathcal H}$.
By the first half of the lemma, we have $Ue^{itH}\phi=e^{itX}$. Hence ${\rm L.H.}\{e^{itX}|t\in{\mathbb{R}}\}$
is dense in $L^2({\mathbb{R}},d\mu_{\phi})$.
\qed
Let $\psi \in {\mathcal H}$. If $\mu_{\psi}$ is absolutely continuous with respect to the Lebesgue measure on ${\mathbb{R}}$,
then we denote by $\rho_{\psi}$ the Radon-Nykod\'ym derivative of $\mu_{\psi}$: $\rho_{\psi}\geq 0$ and
$\mu_{\psi}(B)=\int_B\rho_{\psi}(\lambda)d\lambda,\, B\in \bm{B}$.
We introduce a class of self-adjoint operators on ${\mathcal H}$.
\begin{definition}\label{S1}
{\rm
We say that a self-adjoint operator $H$ on ${\mathcal H}$ is in the class $S_0({\mathcal H})$
if it satisfies the following (i) and (ii):
\begin{list}{}{}
\item[(i)] $H$ is absolutely continuous.
\item[(ii)] $H$ has a cyclic vector $\phi$ such that
$\rho_{\phi}$ is differentiable on ${\mathbb{R}}$ and
\begin{align*}
& \lim_{\lambda\to\pm\infty}\rho_{\phi}(\lambda)=0,\quad
\int_{\rho(\lambda)>0}\frac{\rho_{\phi}'(\lambda)^2}{\rho_{\phi}(\lambda)}d\lambda<\infty.
\end{align*}
\end{list}
}
\end{definition}
Let ${\mathcal H}$ be separable and $H\in S_0({\mathcal H})$ with a cyclic vector $\phi$
satisfying the above (ii) and
$$
W_{\phi}(\lambda):=\left\{
\begin{array}{ll}
\displaystyle \frac{\rho_{\phi}'(\lambda)}{\rho_{\phi}(\lambda)} & \ \mbox{\rm for $ \rho_{\phi}(\lambda)>0 $ } \\
0 & \ \mbox{\rm for $\rho_{\phi}(\lambda)=0 $}
\end{array}
\right..
$$
Then we define an operator $Y$ on $L^2({\mathbb{R}},d\mu_{\phi})$
as follows:
\begin{align*}
&D(Y):={\rm L.H.}\{e^{itX}|t\in {\mathbb{R}}\},\quad Y:=i\frac{d}{d\lambda}+\frac i2W_{\phi}.
\end{align*}
\begin{lemma} The operator $Y$ is a symmetric operator.
\end{lemma}
\proof By Lemma \ref{lem-c1}, $D(Y)$ is dense in $L^2({\mathbb{R}},d\mu_{\phi})$.
Using (ii) and integration by parts, we see that, for all $f,g\in D(Y)$,
$(f,Yg)_{L^2({\mathbb{R}},d\mu_{\phi})}=(Yf,g)_{L^2({\mathbb{R}},d\mu_{\phi})}$.
Hence $Y$ is a symmetric operator.
\qed
\begin{lemma}\label{lem-Y} The operator $Y$ is a strong time operator of $M_X$.
\end{lemma}
\proof It is obvious that, for all $t\in{\mathbb{R}}$, $e^{itM_X}D(Y)\subset D(Y)$.
Let $f(\lambda)=e^{is\lambda},\, s\in{\mathbb{R}}, \lambda \in {\mathbb{R}}$.
Then, using the fact that $if'(\lambda)=-sf(\lambda)$, we see that
$$
(e^{itM_X}Ye^{-itM_X}f)(\lambda)=e^{it\lambda}\left(i\frac{d}{d\lambda}+\frac i2 W_{\phi}
\right)e^{-i(t-s)\lambda}
=tf(\lambda)+(Yf)(\lambda).
$$
Thus $Y$ is a strong time operator of $M_X$.
\qed
\begin{theorem}\label{thm-ac-st} Assume that ${\mathcal H}$ is separable and that $H\in S_0({\mathcal H})$.
Then $H$ has a strong time operator.
\end{theorem}
\proof We have $U^{-1}M_XU=H$. By Lemma \ref{lem-Y}, $Y$ is a strong time operator of $M_X$.
Hence, by an application of Proposition \ref{prop-st-uni}, $U^{-1}YU$ is a strong time operator of $H$.\qed
Thus we have found a class $S_0({\mathcal H})$ of self-adjoint operators on a separable Hilbert space ${\mathcal H}$
which each have a strong time operator.
\subsection{Construction of strong time operators of a self-adjoint operator
from those of another self-adjoint operator}
We consider two self-adjoint operators $H$ and $H'$ acting in Hilbert spaces ${\mathcal H}$ and ${\mathcal H}'$ respectively.
If ${\mathcal H}={\mathcal H}'$, then $H'=H+(H'-H)$ on ${\rm D}(H)\cap {\rm D}(H')$
and hence $H'$ can be regarded as a perturbation of $H$.
We denote by $P_{\rm ac}(H)$ the orthogonal projection onto the absolutely continuous subspace ${\mathcal H}_{\rm ac}(H)$ of $H$.
For a linear operator $A$, we denote by ${\rm Ran}(A)$ the range of $A$.
\begin{lemma}\label{lem-wave} Assume the following (A.1)--(A.3):
\begin{list}{}{}
\item{(A.1)} The wave operators
$$
W_{\pm}:=\mbox{\rm s-}\lim_{t\to\pm\infty}e^{itH'}Je^{-itH}P_{\rm ac}(H)
$$
exist, where $\mbox{\rm s-}\lim$ means strong limit and
$J:{\mathcal H}\to{\mathcal H}'$ is a bounded linear operator.
\item{(A.2)} $\lim_{t\to\pm\infty}\|Je^{-itH}P_{\rm ac}(H)\psi\|=\|P_{\rm ac}(H)\psi\|,\quad \psi\in {\mathcal H}$.
\item{(A.3)}(completeness) ${\rm Ran}(W_{\pm})={\mathcal H}_{\rm ac}(H')$.
\end{list}
Let $U_{\pm}:=W_{\pm}\lceil {\mathcal H}_{\rm ac}(H)$.
Then $U_{\pm}$ are unitary operators from ${\mathcal H}_{\rm ac}(H)$ to ${\mathcal H}_{\rm ac}(H')$
such that
$$
H'_{\rm ac}=U_{\pm}H_{\rm ac}U_{\pm}^{-1}.
$$
\end{lemma}
\proof See textbooks of quantum scattering theory (e.g., \cite{ku79,rs3}).
\qed
\begin{theorem}\label{thm-st-uni1} Assume (A.1)--(A.3) in Lemma \ref{lem-wave}.
Suppose that $H_{\rm ac}$ has a strong time operator $T$.
Then $T'_{\pm}:=U_{\pm}TU_{\pm}^{-1}$ are strong time operators of $H_{\rm ac}'$.
\end{theorem}
\proof This follows from Lemma \ref{lem-wave} and an application of Proposition \ref{prop-st-uni}.
\qed
Theorem \ref{thm-st-uni1} can be used to construct strong time operators
of $H'$ from those of $H$.
\section{Time Operators of a Self-adjoint Operator with Purely Discrete Spectrum }\label{sec4}
\subsection{Case (I)}
\label{weak}
If $\sigma_{\rm disc}(H)\not=\emptyset$, then no strong time operator of $H$ exists by Proposition \ref{time}-(3).
But, even in that case, $H$ may have time operators or weak time operators.
We first recall basic results on this aspect.
\begin{proposition}[\cite{ara09, gal02}]\label{ag}
Suppose that
$\sigma}%{\rm Sp} (H)=\sigma}%{\rm Sp}_{\rm disc} (H)=\{E_n\}_{n=1}^\infty$ ($E_n\not=E_m$ for $n\not=m$),
each eigenvalue $E_n$ is simple,
and, for some $N\geq 1$, $E_n\not=0,\,n\geq N$,
$ \sum_{n=N}^\infty {1}/{E_n^2}<\infty$.
Let $e_n$ be a normalized eigenvector of $H$ with eigenvalue $E_n$: $He_{n}=E_ne_{n}$
and
define
\eq{tdef}
T\phi=i\sum_{n=1}^\infty
\left( \sum_{m\ne n}\frac{ (e_{m}, \phi)}{E_n-E_m}\right)
e_{n},\quad \phi\in {\rm D}(T)
\end{equation}
with domain
\eq{F}
{\rm D} (T):=\ms F
:={\rm L. H. }\{e_n|n\in{\mathbb N}\},
\end{equation}
Then $T$ is a symmetric operator and
$[H, T]=-i\bbbone$ holds on
$$\ms E
:={\rm L. H. }\{e_n-e_m|n, m\in{\mathbb N}\}.$$
Furthermore
$\ms E$ is dense.
\end{proposition}
This proposition shows that $T$ is a time operator of $H$ with a dense CCR-domain $\ms E$ and
hence $T$ is a weak time operator of $H$ too with a weak-CCR domain $\ms E$.
But ${\rm D}(T)={\rm D}(T)\cap {\rm D}(H)$ cannot be a weak-CCR domain for $(H,T)$, since
${\rm D}(T)$ contains an eigenvector of $H$ (see Proposition \ref{prop-w-time}) (note that $\ms E$ contains
no eigenvectors of $H$).
\begin{example}[1-dimensional quantum harmonic oscillator]\label{ex-1hos}
{\rm The Hamiltonian \, of a $1$-dimensional quantum
harmonic oscillator is given by
$$
H_{\rm osc}:=-\frac{1}{2}\Delta+\frac 12 \omega^2 x^2
$$
acting in $L^2({\mathbb{R}})$, where $\Delta$ is the $1$-dimensional generalized Laplacian and $\omega>0$ is a constant.
It is shown that $H_{\rm osc}$ is self-adjoint,
$\sigma}%{\rm Sp}(H_{\rm osc})=\sigma_{\rm disc}(H_{\rm osc})=\{\omega(n+\frac{1}{2} )\}_{n=0}^\infty$ and each eigenvalue $\omega(n+\frac{1}{2})$ is simple.
Since $\sum_{n=1}^{\infty}\frac{1}{(n+\frac{1}{2})^2}<\infty$, the assumption in Proposition \ref{ag} holds.
Hence $H_{\rm osc}$ has a time operator $T_{\rm osc}$ given by
$$
T_{\rm osc}f:=\frac{i}{\omega}\sum_{n=1}^{\infty}\left(\sum_{m\not=n}\frac{(e_m,f)}{n-m}\right)e_m,\quad f\in {\rm D}(T_{\rm osc}).
$$
One can show that $T$ is bounded and $\sigma(\overline{T})=[-\pi/\omega,\pi/\omega]$
(see \cite[Example 4.2]{am08a}).
}
\end{example}
\bc{tds}\label{cor-H-1}
Suppose that
$\sigma}%{\rm Sp} (H)\setminus\{0\}=\sigma}%{\rm Sp}_{\rm disc} (H)=\{E_n\}_{n=1}^\infty$,
each $E_n$ is simple,
$E_1<E_2<\cdots <0$, $0\not\in \sigma}%{\rm Sp}_{\rm p} (H)$, and
$\sum_{n=1}^\infty E_n^2<
\infty$.
Then the operator $T_{\rm d}$ defined by
\begin{align}
\label{td11}
T_{\rm d}}%\rm d} \phi:=i\sum_{n=1}^\infty
\left( \sum_{m\ne n}\frac{ (e_{m}, \phi)}{\frac{1}{E_n}-\frac{1}{E_m}}\right)
e_{n},\quad \phi\in {\rm D}(T_{\rm d}}%\rm d}):=\ms F
\end{align}
is a time operator of $H^{-1} $,
where $\ms F$ is given by \kak{F},
i.e.,
$[H^{-1}, T_{\rm d}}%\rm d}]=-i\bbbone$ on $\ms E$.
\end{corollary}
\proof
We see that $\sigma(H^{-1})=\sigma}%{\rm Sp}_{\rm disc} (H^{-1})=\{1/E_n\}_{n=1}^\infty$ and $\sum_{n=1}^{\infty}\frac{1}{(1/E_n)^2}<\infty$.
Hence the corollary follows from Proposition~\ref{ag}.
\qed
\subsection{Case (II)}
In Corollary \ref{tds}, condition
$\sum_{n=1}^\infty E_n^2<\infty$
is imposed to construct a time operator of $H^{-1}$, which is needed to apply Proposition \ref{ag} with $H$ replaced by
$H^{-1}$.
In this section, we show that
the condition $\sum_{n=1}^\infty E_n^2<\infty$ can be
removed. The idea is to decompose ${\mathcal H}$ into the direct sum
of appropriate mutually orthogonal closed subspaces \cite{sw14}.
\begin{lemma}\label{lem-an-p} Let $p>1$ and $\{a_n\}_{n=1}^{\infty}$ be a complex sequence such that \linebreak $\lim_{n\to\infty}a_n=0$
and $a_n\not=a_m$ for $n\not=m$, $n,m\in \mathbb{N}$.
Let $A:=\{a_n|n\in \mathbb{N}\}$ be the set corresponding to the sequence $\{a_n\}_{n=1}^{\infty}$. Then there exist an $N \in \mathbb{N}\cup \{\infty\}$
and subsequences $\{a_{kn}\}_{n=1}^{\infty}$ of $\{a_n\}_{n=1}^{\infty}$ ($k=1,\ldots,N$) such that
the sets
$A_k:=\{a_{kn}|n\in \mathbb{N}\}$, $k=1,\ldots,N$, have the following properties:
\begin{align*}
&A_k\cap A_l=\emptyset\mbox{ for } k\not=l,\, k,l=1,\ldots,N;\\
&A=\cup_{k=1}^NA_k;\\
&\sum_{n=1}^{\infty}|a_{kn}|^p<\infty,\quad k=1,\ldots,N.
\end{align*}
\end{lemma}
\proof For each $k\in\mathbb{N}$, let $J_k:=\{a_n| 1/(k+1)<|a_n|\leq 1/k\}\subset A$ and
$\{k|J_k\not=\emptyset\}=\{k_1,k_2, \ldots\}$ with $k_1<k_2<\ldots$, which is an infinite set by the condition
$\lim_{n\to\infty}a_n=0$. It is obvious that $A=\cup_{n=1}^{\infty}J_{k_n}$ and
$J_{k_n}\cap J_{k_m}=\emptyset$ for all $(n,m)$ with $n\not=m$.
Let $a_{1n}\in J_{k_n}$. Then $\sum_{n=1}^{\infty}|a_{1n}|^p \leq \sum_{n=1}^{\infty}1/k_n^p<\infty$.
Let $A_1:=\{a_{1n}|n\in\mathbb{N}\}$ and $A':=A\setminus A_1$.
Write $A'=\{b_n|n\in\mathbb{N}\}$
with $b_n\not=b_m$ ($n\not=m$).
Then we can apply the preceding procedure on $\{a_n\}_{n=1}^{\infty}$ to $\{b_n\}_{n=1}^{\infty}$ to conclude that
there exists a subsequence $\{a_{2n}\}_{n=1}^{\infty}$ of $\{b_n\}_{n=1}^{\infty}$
such that $\sum_{n=1}^{\infty}|a_{2n}|^p<\infty$. Hence we obtain a subset
$A_2:=\{a_{2n}|n\in\mathbb{N}\}$. Obviously $A_1\cap A_2=\emptyset$. Then we give a similar consideration to
$A'':=A'\setminus A_2=A\setminus (A_1\cup A_2)$. In this way,
by induction, we can show that, for each $k\in\mathbb{N}$,
there exists a subset $A_k$ which is empty or $A_k=\{a_{kn}|n\in\mathbb{N}\}\subset A$
such that $\sum_{n=1}^{\infty}|a_{kn}|^p<\infty$, $A_k\cap A_j=\emptyset, k\not=j$ and
$A=\cup_{k=1}^{\infty}A_k$ (if, for some $N\in\mathbb{N}$, $A=\cup_{k=1}^NA_k$, then
$A_k=\emptyset, k\geq N+1$). \qed
If a self-adjoint operator $S$ on a Hilbert space $\mathcal K$ is reduced by a closed subspace $\mathcal D$ of $\mathcal K$,
then we denote by $S_{\mathcal D}$ the reduced part of $S$ to $\mathcal D$, unless otherwise stated.
\begin{lemma}\label{sw}
Let
$\sigma}%{\rm Sp} (H)=\sigma}%{\rm Sp}_{\rm disc} (H)=\{E_n\}_{n=1}^\infty$,
$E_1<E_2<\cdots<0$, \linebreak $\lim_{n\to\infty} E_n=0$ and $0\not\in \sigma}%{\rm Sp}_{\rm p} (H)$.
Then there exist mutually orthogonal closed subspaces ${\mathcal H}_j$ of ${\mathcal H}$ ($j=1,\ldots,N,\, N\leq \infty$)
such that ${\mathcal H}$ is decomposed as
${\mathcal H}=\oplus_{j=1}^N {\mathcal H}_j$ ($N\leq \infty$)
and (1)--(3) below are satisfied.
\begin{list}{}{}
\item[(1)] Each ${\mathcal H}_j$ reduces $H$ and
$\sigma}%{\rm Sp} (H_j)\setminus\{0\}=\sigma}%{\rm Sp}_{\rm disc}(H_j))=\{F_{jk}\}_{k=1}^\infty$, where $H_j:=H_{{\mathcal H}_j}$.
\item[(2)]
Each eigenvalue $F_{jk}$ ($1\leq j\leq N, 1\leq k\leq \infty$) is simple.
\item[(3)] $\d \sum_{k=1}^\infty F_{jk}^2<\infty$ for each $1\leq j\leq N$.
\end{list}
\end{lemma}
\proof
Note that $0$ is the unique accumulation point of the set $\{E_n|n\in \mathbb{N}\}$.
Let $M_n$ be the multiplicity of $E_n$ (which is finite, since $E_n\in \sigma_{\rm disc}(H)$).
Let $\{e_n^i|i=1, \cdots, M_n\}$ be a complete orthonormal system (CONS) of $\ker(H-E_n)$:
$He_n^i=E_n e_n^i,\,i=1,\ldots,M_n$.
We set
$$
\sup_{n\geq 1} M_n=M\quad \mbox{ and } \quad \limsup_{n\to\infty}M_n=m.
$$
We consider two cases:
(A) $m=\infty$ and (B) $m<\infty$.
{\bf Case (A)}. Suppose that $m=\infty$.
In this case, $M=\infty$ and,
for each $k\geq 1$ and each $n$, there exists an $N\geq n$
such that $M_N\geq k$.
Using this fact, we see that, for each $k\geq 1$, the subspace
$$
\ms G_k={\rm L. H. } \{e_j^k| \ M_j\geq k\}
$$
is infinite dimensional and $\ms G_k$ is orthogonal to $\ms G_l$ for all $k,l$ with $k\not=l$.
Since $\{e_n^i|n\geq 1, i=1,\ldots,M_n\}$ is a CONS of ${\mathcal H}$, we have the orthogonal decomposition
\eq{gk3}
{\mathcal H}=\oplus_{k=1}^\infty \overline{\ms G_k}.
\end{equation}
Fix $k$ and consider $\ms G_k$.
Let $\ms A:=\sigma}%{\rm Sp}_{\rm disc} (H_{\ms G_k})=\{a_j|j\in\mathbb{N}\}$ ($=\{E_n|M_n\geq k\}$).
Then each eigenvalue $a_j$ is simple and $a_j\not=a_k$ for $j\not=k$, $\lim_{j\to\infty}a_j=0$.
Hence, we can apply Lemma \ref{lem-an-p} with $p=2$ to conclude that
there exist an $N_k\leq \infty$ and subsets $\ms A_l:=
\{a_j^l\in \ms A|j\in\mathbb{N}, \sum_{j=1}^\infty |a_j^l|^2<\infty\}$ of $\ms A$
such that $\ms A=\cup_{l=1}^{N_k}\ms A_l$ (a disjoint union).
Hence we can decompose $\overline{\ms G_k}$ as
\eq{gk1}
\overline{\ms G_k}=\oplus_{l=1}^{N_k} \overline{\ms G_k^l},
\end{equation}
where $\ms G_k^l:={\rm L.H}\{g_j\in \ms G_k
|Hg_j=a_j^lg_j, j\in \mathbb{N}\}$ (hence
$\sigma}%{\rm Sp}_{\rm disc} (H_{\ms G_k^l})=\ms A_l$).
Thus
\eq{gk4}
{\mathcal H}=\oplus _{k=1}^\infty \oplus_{l=1}^{N_k} \overline{\ms G_k^l}
\end{equation} and the lemma follows.
{\bf Case (B)}. Suppose that $m<\infty$. Then we have $m\leq M<\infty$.
Hence we need only to consider four cases $ (a)- (d)$ below.
{\bf (a) $M=m=1$}. In this case, ${\mathcal H}=\overline{\ms G_1}$
and
$\overline{\ms G_1}$ can be decomposed as \kak{gk1}.
Then the lemma follows.
{\bf (b) $M\geq 2$ and $M=m$}.
In this case, for all $k=1,\ldots,M$, $\ms G_k$ is infinite dimensional.
Hence, in the same way as in the case $m=\infty$
we can see that ${\mathcal H}=\oplus_{k=1}^M\overline{\ms G_k}$ and
$\overline{\ms G_k}$ can be decomposed as \kak{gk1}. Thus the lemma follows.
{\bf (c) $M\geq 2$ and $m=1$}.
In this case, there exists a $j_0\in \mathbb{N}$ such that for all $j\geq j_0$,
$M_j=1$.
Let $\ms B_k={\rm L. H. } \{e_j^k|j< j_0, k\leq M_j\}$, $k=1, \cdots, M$ and
$\ms C:={\rm L. H. } \{e_j^1|j\geq j_0\}$.
Then we can decompose $\ms C$ as
$\overline{\ms C}=\oplus_{k=1}^{M} \overline{\ms C_k}$,
where
$\ms C_k={\rm L.H.}\{e_{j_k}^1|j_k \geq j_0, j_k=j_0+k-1+Mr, \, r\in \{0\}\cup\mathbb{N}\}$ ($k=1,\ldots,M$).
Define
$\ms D_k=\ms B_k\oplus \overline{\ms C_k}$, $k=1, \cdots, M$. Then
we have ${\mathcal H}=\oplus_{k=1}^M \ms D_k$.
In the same way as in the case (A), we can decompose $\ms D_k$ like \kak{gk1}. Thus the lemma follows.
{\bf (d) $M\geq 2$, $M>m$ and $m\geq 2$}. In this case, $\{j|M_j=m\}$ is a countable infinite set.
Hence, for $j=1,\ldots,m$, $\ms G_j$ is infinite dimensional.
We have the orthogonal decomposition
\eq{gk2}
{\mathcal H}=\left(
\oplus_{j=1}^{m-1}\overline{\ms G_j}\right)
\oplus \ms K,
\end{equation}
where $\ms K=\left( \oplus_{j=1}^{m-1}\overline{\ms G_j}\right)^{\perp}$.
The closed subspace $\ms K$ reduces $H$.
Since $\ms G_m\subset \ms K$, it follows that
$\sigma}%{\rm Sp} (H_{\ms K})\setminus \{0\}=\sigma_{\rm disc}(H_{\ms K})$ is an infinite set. Let $\sigma_{\rm disc}(H_{\ms K})=\{b_j\}_{j=1}^\infty$
and $\beta_j$ be the multiplicity of eigenvalue $b_j$.
Then $\sup_j\beta_j=M-m+1$ and
$\sup_j \beta _j\geq \limsup _j\beta _j=1$. Hence
by
(a) and (c),
we can decompose $\ms K$
as $\ms K=\oplus_{j=1}^{M-m+1}\ms K_j$, where
$\ms K_j$ is an infinite dimensional closed subspace of $\ms K$.
Hence
\eq{gk333}
{\mathcal H}=
\left( \oplus_{j=1}^{m-1}\overline{\ms G_j}\right)
\oplus
\left( \oplus_{j=1}^{M-m+1}\ms K_j\right).
\end{equation}
In the same way as in the case (A), we can decompose $\overline{\ms G_j}$ and $\ms K_j$ like \kak{gk1}. Thus the lemma follows.
\qed
Combining Corollary \ref{cor-H-1} and Lemma \ref{sw}, we can prove the following lemma.
\bt{wada+sasaki}\TTT{time operator of $H^{-1}$}\label{thm-TH}
Suppose that
$\sigma}%{\rm Sp} (H)\setminus\{0\}=\sigma}%{\rm Sp}_{\rm disc} (H)=\{E_j\}_{j=1}^\infty$,
$E_1<E_2<\cdots <0$, $\lim_{j\to\infty}E_j=0$, and $0\not\in \sigma}%{\rm Sp}_{\rm p} (H)$.
Then there exists a time operator $T_{-1}$ of
$H^{-1}$ with a dense CCR-domain for $(H^{-1},T_{-1})$.
\end{theorem}
\proof
By Lemma \ref{sw}, ${\mathcal H}$ can be decomposed as
${\mathcal H}=\oplus_{j=1}^N {\mathcal H}_j$ with $N\leq \infty$.
By Proposition \ref{ag}, a time operator $S_j$ of $H^{-1}_j$ exists:
$$[H^{-1} _j, S_j]=-i\bbbone $$ on $\ms E_{j}:={\rm L. H. }\{e_n^j-e_m^j, n, m\in \mathbb N\}$, where $\{e_n^j\}_{n=1}^\infty $ denotes the eigenvectors of $H_j$ such that
$H_je_n^j=F_{jn}e_n^j$ and
${\rm D} (S_j)={\rm L. H. } \{e_n^j|n\in\mathbb N\}$.
Define $T_{-1}$ by
$T_{-1}:=\oplus_{j=1}^N S_j$ with
${\rm D} (T_{-1}):=\oplus_{j=1}^N {\rm D} (S_j)$ (algebraic direct sum).
Then $T_{-1}$ is a time operator of $H^{-1}$
with a CCR-domain given by $\oplus_{j=1}^N \ms E_j$ (algebraic direct sum), which is dense in ${\mathcal H}$.
\qed
\subsection{Case (III)}\label{subsec-III}
We next consider an extension of Proposition \ref{ag} to the case
where no restriction is imposed on the growth order of the discrete eigenvalues $\{E_n\}_{n=1}^{\infty}$
of $H$.
\begin{lemma}\label{tsw}
Suppose that
$\sigma}%{\rm Sp} (H)=\sigma}%{\rm Sp}_{\rm disc} (H)=\{E_n\}_{n=1}^\infty$
with $0<E_1<E_2<\cdots <E_n<E_{n+1}<\cdots $ and
$\lim_{n\to\infty} E_n=\infty$.
Then there exist mutually orthogonal closed subspaces ${\mathcal H}_j$ of ${\mathcal H}$ ($j=1,\ldots,N, N\leq \infty$) such that
${\mathcal H}=\oplus_{j=1}^N {\mathcal H}_j$ and
(1)--(3) below are satisfied:
\begin{list}{}{}
\item[(1)] Each ${\mathcal H}_j$ reduces $H$ and
$\sigma}%{\rm Sp} (H_{{\mathcal H}_j})=\sigma}%{\rm Sp}_{\rm disc}(H_{{\mathcal H}_j})=\{F_{jk}\}_{k=1}^\infty$.
\item[(2)]
Each $F_{jk}$ ($1\leq j\leq N, k\in \mathbb{N}$) is simple.
\item[(3)] $\d \sum_{k=1}^\infty \frac{1}{F_{jk}^2}<\infty$ for each $1\leq j\leq N$.
\end{list}
\end{lemma}
\proof
Let $K=H^{-1}$. Then $K$ is self-adjoint and
$\sigma}%{\rm Sp} (K)\setminus\{0\}=\sigma}%{\rm Sp}_{\rm disc} (K)=\{1/E_n\}_{n=1}^\infty$,
$1/E_1>1/E_2>\cdots>0$, $\lim_{j\to\infty} 1/E_j=0$ and $0\not\in \sigma}%{\rm Sp}_{\rm p} (K)$.
Hence, by applying Lemma \ref{sw} to the case where $H$ and $E_n$ there are replaced by $-K$ and
$-1/E_n$ respectively, we see that ${\mathcal H}$ has an orthogonal decomposition
${\mathcal H}=\oplus_{j=1}^N {\mathcal H}_j$ ($N\leq \infty$)
with closed subspaces ${\mathcal H}_j$ of ${\mathcal H}$
such that (1)--(3) above are satisfied.
\qed
\bt{27}\TTT{time operator of $H$}\label{thm-d-top}
Let
$\sigma}%{\rm Sp} (H)=\sigma}%{\rm Sp}_{\rm disc} (H)=\{E_n\}_{n=1}^\infty$,
$E_1<E_2<\cdots $ and $\lim_{n\to\infty}E_n=\infty$.
Then there exists a time operator $T$ of $H$ with a dense CCR-domain for $(H,T)$.
\end{theorem}
\proof
The method of proof is similar to that of Theorem \ref{wada+sasaki}.
By Lemma \ref{tsw}, ${\mathcal H}$ can be decomposed as
${\mathcal H}=\oplus_{j=1}^N {\mathcal H}_j$ with $N\leq \infty$.
By Proposition \ref{ag} a time operator $T_{j}$ of
$H_{{\mathcal H}_j}$ exists:
$[H _{{\mathcal H}_j}, T_{j}]=-i\bbbone $ on $\ms E_{j}={\rm L. H. }\{e_n^j-e_m^j, n, m\in \mathbb N\}$, where $\{e_n^j\}_{n=1}^\infty $ denotes the eigenvectors of $H_{\ms H_j}$ such that
$He_n^j=F_{jn}e_n^j$, and the domain of $T_j$ is given by
${\rm D} (T_j)={\rm L. H. } \{e_n^j|n\in\mathbb N\}$.
Define $T$ by
$T:=\oplus_{j=1}^N T_j$ with
${\rm D} (T)=\oplus_j^N {\rm D} (T_j)$ (algebraic direct sum).
Then $T$ is a time operator of $H$ with a CCR-domain $\oplus_{j=1}^N\ms E_j$ (algebraic direct sum), which is dense in ${\mathcal H}$.
\qed
\begin{example}[$d$-dimensional quantum harmonic oscillator]
{\rm Let $\omega_j>0$ ($j=1,\ldots,d$) be a constant and
$$
H_{{\rm osc},\,j}:=-\frac 12 D_j^2+\frac 12 \omega_j^2x_j^2
$$
acting in $L^2({{\mathbb{R}}^d})$ (see Example \ref{ex-1hos}).
Then the Hamiltonian of a $d$-dimensional quantum harmonic oscillator is given by
$$
H_{\rm osc}^{(d)}:=\sum_{j=1}^dH_{{\rm osc},\,j}
$$
acting in $L^2({{\mathbb{R}}^d})$.
It follows that $H_{\rm osc}^{(d)}$ is self-adjoint and
$$
\sigma(H_{\rm osc}^{(d)})=\sigma_{\rm disc}(H_{\rm osc}^{(d)})
=\left\{\sum_{j=1}^d\omega_j\left(n_j+\frac 12\right)|n_j\in \{0\}\cup\mathbb{N},\,j=1,\ldots,d\right\}.
$$
Hence, by Theorem \ref{thm-d-top}, $H_{\rm osc}^{(d)}$ has a time operator with a dense CCR-domain.
}
\end{example}
\begin{example}[non-commutative harmonic oscillator]
{\rm
Let $A$ and $J$ be $2\times 2$ matrices defined by
$$A=\begin{pmatrix} \alpha& 0\\ 0& \beta\end{pmatrix},\quad
\alpha,\beta\geq0,\quad J=\begin{pmatrix} 0& {-1}\\ 1& 0\end{pmatrix}.$$
Let $\alpha\beta>1$.
The Hamiltonian $H(\alpha,\beta)$ of the non-commutative harmonic oscillator \cite{p10} is defined by
the self-adjoint operator
\eq{ncho}
H(\alpha,\beta)=A\otimes (-\frac{1}{2} \Delta+\frac{1}{2} x^2)+J\otimes (xD+\frac{1}{2})
\end{equation}
on the Hilbert space
${\mathbb{C}}^2\otimes L^2({\mathbb{R}})$, where $D$ is the generalized differential operator in $x$.
It is shown in \cite{iw07} that
$\sigma}%{\rm Sp}(H(\alpha,\beta))=\sigma}%{\rm Sp}_{\rm disc}(H(\alpha,\beta))$ $=\{\lambda_n\}_{n=1}^\infty$ and
the multiplicity of each $\lambda_n$ is not greater than $2$ with $\lambda_n\to\infty$ ($n\to\infty$).
Hence, by Theorem \ref{thm-d-top}, $H(\alpha,\beta)$ has a time operator with a dense CCR-domain.
}\end{example}
\begin{example}[Rabi model]{\rm
Let $\mathbb{Z}_+:=\{0\}\cup\mathbb{N}$ be the set of nonnegative integers
and
$$
\ell^2(\mathbb{Z}_+):=\left\{\psi=\{\psi_n\}_{n=0}^{\infty}\big|
\psi_n\in{\mathbb{C}},n\geq 0,\,\sum_{n=0}^{\infty}|\psi_n|^2<\infty\right\}
$$
be the Hilbert space of absolutely square summable complex sequences indexed by $\mathbb{Z}_+$.
The Hilbert space $\ell^2(\mathbb{Z}_+)$ is in fact the boson Fock space $\ms F_{\rm b}({\mathbb{C}})$ over ${\mathbb{C}}$
(e.g. \cite[p.53, Example 2]{rs1} and \cite[\S X.7]{rs2}): $\ell^2(\mathbb{Z}_+)=\ms F_{\rm b}({\mathbb{C}})$.
We denote by $a$ the annihilation operator on $\ms F_{\rm b}({\mathbb{C}})$:
\begin{align*}
&(a\psi)_n:=\sqrt{n+1}\psi_{n+1}, n\geq 0, \\
&
\psi\in {\rm D}(a):=
\left\{\psi\in \ell^2(\mathbb{Z}_+)|\sum_{n=0}^{\infty}n|\psi_n|^2<\infty\right\}.
\end{align*}
We have $(a^*\psi)_0=0,\, (a^*\psi)_n=\sqrt{n}\psi_{n-1},\, n\geq 1$ for all $\psi\in {\rm D}(a^*)$.
The commutation relation $[a,a^*]=\bbbone$ holds on the dense subspace $\ell_0(\mathbb{Z}_+)
:=\{\psi\in \ell^2(\mathbb{Z}_+)|\exists n_0\in \mathbb{N} \,\,\mbox{\rm such that}\,\, \psi_n=0,\, \forall n\geq n_0\}$.
Let $\sigma_x,\sigma_y,\sigma_z$ be the Pauli matrices:
\begin{align*}
\sigma_x= \begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix},\quad
\sigma_y = \begin{pmatrix} 0 & -i \\ i & 0 \end{pmatrix},\quad
\sigma_z = \begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}
\end{align*}
and
\eq{rabi}
H_{\rm Rabi}:={\mu \sigma}%{\rm Sp}_z\otimes \bbbone +\omega\bbbone\otimes a^{\ast} a+
g \sigma}%{\rm Sp}_x\otimes (a+a^{\ast})}
\end{equation}
on ${\mathbb{C}}^2\otimes \ms F_{\rm b}({\mathbb{C}})$, where $\mu>0$, $\omega>0$ and $g\in {\mathbb{R}}$ are constants.
The model whose Hamiltonian is given by $H_{\rm Rabi}$ is called the Rabi model \cite{rab36,rab37,bra11}.
The matrix
$$
U:=\frac 1{\sqrt{2}}(\sigma_x+\sigma_z)
$$
is unitary and self-adjoint. By direct computations using the properties that $\sigma_j\sigma_k+\sigma_k\sigma_j=2\delta_{jk},
j,k=x,y,z$, we see that
$$
\widetilde H_{\rm Rabi}:=UH_{\rm Rabi}U^{-1}=\mu\sigma_x+H,
\quad H:=\left(
\begin{array}{cc}
H_+ &0\\
0 & H_-
\end{array}
\right),
$$
where $H_{\pm}:=\omega a^*a\pm g(a+a^*)$ and
we have used the natural identification ${\mathbb{C}}^2\otimes \ms F_{\rm b}({\mathbb{C}})=\ms F_{\rm b}({\mathbb{C}})\oplus \ms F_{\rm b}({\mathbb{C}})$.
It is well known that the operator $\pi_g:=(g/\omega)\overline{i(a-a^*)}$ is self-adjoint and
$$
e^{\pm i\pi_g}ae^{\mp i\pi_g}=a\mp \frac g{\omega}.
$$
Hence
$$
e^{\pm i\pi_g}H_{\pm}e^{\mp i\pi_g}=\omega a^*a - \frac {g^2}{\omega},
$$
implying that $\sigma(H_{\pm})=\sigma_{\rm disc}(H_{\pm})=\sigma(\omega a^*a-\frac{g^2}{\omega})=
\{\nu_n| n\in \mathbb{Z}_+\}$ with $\nu_n:= \omega n-\frac{g^2}{\omega}$. Hence
$\sigma(H)=\sigma_{\rm disc}(H)=\{\nu_n|n\in\mathbb{Z}_+\}$ with the multiplicity of each eigenvalue $\nu_n$ being two.
Since $\mu\sigma_x$ is bounded, it follows from the min-max principle that
$\widetilde H_{\rm Rabi}$ (and hence $H_{\rm Rabi}$) has purely discrete spectrum with $
\sigma(H_{\rm Rabi})=\sigma_{\rm disc}(H_{\rm Rabi})=
\sigma_{\rm disc}(\widetilde H_{\rm Rabi})=\{\nu_n'|n\in\mathbb{Z}_+\}$
satisfying $ \nu_n-\mu\leq \nu_{2n}'\leq \nu_n+\mu,\, n\geq 0$, where $\nu_0'\leq \nu_1'\leq \cdots \leq \nu_n'\leq
\nu_{n+1}'\leq \cdots$ counting multiplicities (see also \cite{bra11,mps14} for studies on spectral properties
of $H_{\rm Rabi}$).
Hence we can apply Theorem \ref{thm-d-top} to conclude that
$H_{\rm Rabi}$ has a time operator with a dense CCR-domain.
}
\end{example}
\section{Ultra-Weak Time Operators}\label{sec5}
\subsection{Ultra-weak time operators of a self-adjoint operator}
In this subsection, we consider the case where a self-adjoint operator $H$ obeys the assumption
of Theorem \ref{thm-TH} and ask if $H$ has a time operator.
We first give a formal heuristic argument.
By Theorem \ref{thm-TH}, we know that $H^{-1}$ has a time operator $T_{-1} $ with a dense CCR-domain
for $(H^{-1},T_{-1})$.
Since the unique accumulation point of $\sigma}%{\rm Sp} (H)$ is $0$, but not $\infty$,
it is not straightforward to apply Proposition \ref{ag}
to construct a time operator of $H$.
The key idea we use is
to regard $H$ as $H= (H^{-1})^{-1}$.
Let $f (x)=x^{-1}$. Then $H=f (H^{-1})$.
Since $f' (x)=-x^{-2}$, a formal application of
Proposition \ref{prop-hkm}
suggests that
$A=-\frac{1}{2} (T_{-1} H^{-2}+H^{-2} T_{-1})$
may be a time operator of $H$.
But, we note that no eigenvectors of $H$ are in ${\rm D}(H^{-2}T_{-1})$.
Hence it seems to be difficult to show that ${\rm D}(A)\not=\{0\}$ and
${\rm D}(HA)\cap {\rm D}(AH)\not=\{0\}$.
Thus we are led to consider a form version of $A$.
We use the notation in the proof of Theorem \ref{thm-TH}.
It is obvious that, for all $k\in \mathbb{Z}$,
${\rm D}(S_j)\subset {\rm D}(H_j^k)$. Hence we define a sesquilinear form ${\mathfrak{t}}_j:{\rm D}(S_j)\times {\rm D}(S_j)\to{\mathbb{C}}$ by
\begin{equation}
{\mathfrak{t}}_j[ \phi,\psi]:
=-\frac 12 \left\{(S_j\phi, H_j^{-2}\psi)+(H_j^{-2}\phi,S_j\psi)\right\},\quad
\phi,\psi\in {\rm D}(S_j).
\end{equation}
\begin{lemma}\label{lem-uwt} Let
$$
H_j^{-1}\ms E_j:=\{H_j^{-1}\psi |\psi\in \ms E_j\}={\rm L.H.}\left\{\frac 1{E_n}e_n^j-\frac 1{E_m}e_m^j)\big|n,m\in\mathbb{N}\right\}.
$$
Then, for all $\psi,\phi\in H_j^{-1}\ms E$, $H_j\phi$ and $H_j\psi$ are in
${\rm D}(S_j)$ and
\begin{equation}
{\mathfrak{t}}_j[ H_j\phi,\psi]-{\mathfrak{t}}_j[ \phi,H_j\psi]=-i(\phi,\psi).\label{TjH}
\end{equation}
\end{lemma}
\proof Since $H_j(H_j^{-1}\ms E_j)=\ms E_j\subset {\rm D}(S_j)$,
$H_j\phi\in{\rm D}(S_j)$ for all $\phi\in H_j^{-1}\ms E_j$.
\linebreak By direct computations, we have
\begin{align*}
{\mathfrak{t}}[ H_j\phi,\psi]-{\mathfrak{t}}[ \phi,H_j\psi]
=&-\frac 12
\bigg\{
(S_jH_j\phi, H_j^{-2}\psi)
-(S_j\phi,H_j^{-1}\psi)\\
&+(H_j^{-1}\phi, S_j\psi)-(H_j^{-2}\phi, S_jH_j\psi)\bigg\}.
\end{align*}
We can write $\phi=H_j^{-1}\eta$ and $\psi=H_j^{-1}\chi$ with
$\eta,\chi\in \ms E_j$.
Then we have
\begin{align*}
(S_jH_j\phi, H_j^{-2}\psi)
-(S_j\phi,H_j^{-1}\psi)
&=(H_j^{-1}S_j\eta,H_j^{-1}\psi)-(S_jH_j^{-1}\eta, H_j^{-1}\psi)\\
&=
(-i\eta,H_j^{-1}\psi)=i(\phi,\psi),
\end{align*}
where we have used that $S_j$ is a time operator of $H_j^{-1}$ with a CCR-domain $\ms E_j$.
Similarly we have
\begin{align*}
(H_j^{-1}\phi, S_j\psi)-(H_j^{-2}\phi, S_jH_j\psi)
&=(H_j^{-1}\phi, S_jH_j^{-1}\chi)-(H_j^{-1}\phi, H_j^{-1}S_j\chi)\\
&=i(\phi, \psi).
\end{align*}
Thus (\ref{TjH}) follows.\qed
Lemma \ref{lem-uwt} shows that ${\mathfrak{t}}_j$ is an ultra-weak time operator of $H_j$ with
$H_j^{-1}\ms E_j$ being an ultra-weak CCR-domain.
We introduce
$$
\widetilde{\ms E}:=\oplus_{j=1}^NH_j^{-1}\ms E_j \,\,\,\, (\mbox{algebraic direct sum}).
$$
Since $H_j^{-1}\ms E_j$ is dense in ${\mathcal H}_j$, $\widetilde {\ms E}$ is dense in
${\mathcal H}$ and $\widetilde {\ms E}\subset {\rm D}(H)$.
\begin{theorem}\label{thm-uwt} Under the same assumption as in Theorem \ref{thm-TH},
there exists an ultra-weak time operator ${\mathfrak{t}}_{\rm p}$ of
$H$ with $\widetilde {\ms E}$ being an ultra-weak CCR-domain.
\end{theorem}
\proof Let $T_{-1}$ be as in Theorem \ref{thm-TH} and define a sesquilinear form ${\mathfrak{t}}_{\rm p}:
{\rm D}(T_{-1})\times {\rm D}(T_{-1})\to{\mathbb{C}}$ by
$$
{\mathfrak{t}}_{\rm p}[ \psi,\phi]:=
\sum_{j=1}^N{\mathfrak{t}}[ \psi_j,\phi_j], \quad \psi=(\psi_j)_{j=1}^N, \phi=(\phi_j)_{j=1}^N\in {\rm D}(T_{-1}).
$$
We remark that, in the case $N=\infty$, $\psi_j=0$ for all sufficiently large $j$ and hence
the sum $\sum_{j=1}^N$ on the right hand side is over only finite terms, being well defined.
It follows from Lemma \ref{lem-uwt} that, for all $\psi,\phi\in \widetilde {\ms E}$,
$H\psi,H\phi\in{\ms E}\subset {\rm D}(T_{-1})$ and
$$
{\mathfrak{t}}_{\rm p}[ H\phi,\psi]-{\mathfrak{t}}_{\rm p}[ \phi, H\psi]=-i(\phi,\psi).
$$
This means that ${\mathfrak{t}}_{\rm p}$ is an ultra-weak time operator of $H$ with
$\widetilde {\ms E}$ being an ultra-weak CCR domain.
\qed
We now proceed to showing
existence of an ultra-weak time operator
of a self-adjoint operator in a general class.
\begin{definition}[class $S_1({\mathcal H})$]\label{def-S1}
{\rm A self-adjoint operator
$H$ on ${\mathcal H}$ is said to be in the class $S_1({\mathcal H})$ if it has the following properties (H.1)--(H.4):
\begin{list}{}{}
\item[(H.1)]
$\sigma}%{\rm Sp}_{\rm sc} (H )=\emptyset$.
\item[(H.2)]
$\sigma}%{\rm Sp}_{\rm ac} (H)=[0, \infty)$.
\item[(H.3)]
$\sigma}%{\rm Sp}_{\rm disc} (H)=\sigma_{\rm p}(H)=\{E_n\}_{n=1}^\infty$,
$E_1<E_2<\cdots<0$, $\lim_{n\to\infty}E_n=0$ (hence $0\not\in \sigma}%{\rm Sp}_{\rm p} (H)$).
\item[(H.4)]
There exists a strong time operator $T_{\rm ac}$
of $H_{\rm ac}$ in
$\ms H_{\rm ac}(H)$.
\end{list}
}
\end{definition}
Let $H\in S_1({\mathcal H})$. Then we have the orthogonal decomposition
$$
{\mathcal H}={\mathcal H}_{\rm ac}(H)\oplus {\mathcal H}_{\rm p}(H).
$$
By (H.3), we can apply Theorem \ref{thm-uwt}
to the case where $H$ is replaced by $H_{\rm p}$ to conclude that
$H_{\rm p}$ has an ultra-weak time operator ${\mathfrak{t}}_{\rm p}$ with a dense ultra-weak CCR domain $\ms E_{\rm p}$
such that
$$
{\mathfrak{t}}_{\rm p}[ H_{\rm p}\phi,\psi]-{\mathfrak{t}}_{\rm p}[ \phi,H_{\rm p}\psi]=-i(\phi,\psi),\quad \phi,\psi\in \ms E_{\rm p}.
$$
We denote by $\mathcal D_{\rm p}$ the subspace ${\rm D}(T_{-1})$ in the proof
of Theorem \ref{thm-uwt}. Hence ${\mathfrak{t}}_{\rm p}:\mathcal D_{\rm p}\times \mathcal D_{\rm p}\to{\mathbb{C}}$ with
$\ms E_{\rm p}\subset {\rm D}(H_{\rm p})\cap \mathcal D_{\rm p}$ and $H_{\rm p}\ms E_{\rm p}\subset \mathcal D_{\rm p}$.
By (H.4), there exists a dense CCR-domain $\mathcal D_{\rm ac}$ for $(H_{\rm ac}, T_{\rm ac})$.
Let $\widetilde {\ms E}_{\rm p}:=H_{\rm p}^{-1}{\ms E}_{\rm p}$ and
\begin{equation}
\mathcal D_H:=\mathcal D_{\rm ac}\oplus \widetilde{\ms E}_{\rm p},
\end{equation}
which is dense in ${\mathcal H}$.
We define a sesquilinear form
${\mathfrak{t}}_H :({\mathcal H}_{\rm ac}(H)\oplus \mathcal D_{\rm p})\times ({\rm D}(T_{\rm ac})\oplus \mathcal D_{\rm p})\to{\mathbb{C}}$
by
\begin{align}
{\mathfrak{t}}_H [\phi_1\oplus\phi_2, \psi_1\oplus\psi_2] =&
(\phi_1, T_{\rm ac}\psi_1)+{\mathfrak{t}}_{\rm p}[ \phi_2, \psi_2],\nonumber\\
& \quad \phi_1\in {\mathcal H}_{\rm ac}(H),
\psi_1\in {\rm D}(T_{\rm ac}), \phi_2,\psi_2\in \mathcal D_{\rm p}.\label{time1}
\end{align}
Now we are in the position to state and prove the main result
in this section.
\begin{theorem}[abstract ultra-weak time operator 1]\label{thm-uwt1}Let $H\in S_1({\mathcal H})$.
Then the sesquilinear form ${\mathfrak{t}}_H$ defined by (\ref{time1})
is an ultra-weak time operator of $H$ with $\mathcal D_H$ being an ultra-weak CCR-domain.
\end{theorem}
\proof Let ${\mathfrak{t}}_{\rm ac}:{\mathcal H}_{\rm ac}(H)\times {\rm D}(T_{\rm ac})\to{\mathbb{C}}$
by
$${\mathfrak{t}}_{\rm ac}[\phi,\psi]:=(\phi,T_{\rm ac}\psi),\,\phi\in {\mathcal H}_{\rm ac}(H),\psi\in{\rm D}(T_{\rm ac}).$$
Then, by Remark \ref{rem-uwt}-(2), ${\mathfrak{t}}_{\rm ac}$ is an ultra-weak time operator
of $H_{\rm ac}$ with $\mathcal D_{\rm ac}$ being an ultra-weak CCR-domain.
Then, in the same way as in the proof of Theorem \ref{thm-uwt},
one can show that ${\mathfrak{t}}_H$ is an ultra-weak time operator of $H$ with
$\mathcal D_H$ being an ultra-weak CCR-domain.
\qed
\subsection{Ultra-weak time operators of $f(H)$}
We can also construct an ultra-weak time operator of
$f(H)$ for some function $f:{\mathbb{R}}\to{\mathbb{R}}$.
A strong time operator of $f(H_{\rm ac})$ is already constructed in Proposition \ref{hkm09}.
Hence we need only to construct an ultra-weak time operator
of $f(H_{\rm p})$.
A set of conditions for that is as follows.
\begin{assumption}\label{hara}
{\rm Let $H\in S_1({\mathcal H})$.
\begin{list}{}{}
\item{(1)} The function $f:{\mathbb{R}}\to{\mathbb{R}}$ satisfies the same
assumption as in Proposition~\ref{hkm09}.
\item{(2)} The function $f$ is continuous at $x=0$.
\item{(3)} $f(\sigma}%{\rm Sp}_{\rm disc} (H))$ is an infinite set such that
the multiplicity of each point in $f(\sigma}%{\rm Sp}_{\rm disc}(H))$ as an eigenvalue of $f(H)$ is finite.
\item{(4)} $f (0)\not\in \sigma}%{\rm Sp}_{\rm p} (f (H))$.
\end{list}
}\end{assumption}
Suppose that Assumption \ref{hara} holds.
Then $f(H)$ is self-adjoint and reduced by ${\mathcal H}_{\rm ac}(H)$ and ${\mathcal H}_{\rm p}(H)$
(these properties follow from only the fact that $f:{\mathbb{R}}\to{\mathbb{R}}$, Borel measurable, and the general theory of functional calculus).
We denote the reduced part of $f(H)$ to ${\mathcal H}_{\rm ac}(H)$ and ${\mathcal H}_{\rm p}(H)$
by $f(H)_{\rm ac}$ and $f(H)_{\rm p}$ respectively.
By the functional calculus,
we have
$f (H)_{\rm ac}=f (H_{\rm ac})$, $f (H)_{\rm p}=f (H_{\rm p})$ and
$f (H)=f (H_{\rm ac})\oplus f (H_{\rm p})$.
This implies that
\begin{equation}
\sigma(f(H))=\sigma(f(H_{\rm ac}))\cup\sigma(f(H_{\rm p}))=\ov{f ([0, \infty))}\cup \ov{\{f (E_j)\}_{j=1}^\infty}.
\end{equation}
\bc{main3}
\TTT{abstract ultra-weak time operator 2}
Under Assumption \ref{hara}, there exists an ultra-weak time operator ${\mathfrak{t}}_{H}^f$ of
$f (H) $ with a dense ultra-weak CCR-domain.
\end{corollary}
\proof Let $T_{\rm ac}$ be a strong time operator of $H_{\rm ac}$.
Then the strong time operator of
$f (H_{\rm ac})$ is given by
$$T_{{\rm ac}}^f=\frac{1}{2}\ov{ (T_{\rm ac}f' (H_{\rm ac})^{-1}
+f' (H_{\rm ac})^{-1}T_{\rm ac}) \lceil D}$$ by
Proposition \ref{hkm09}, where
$D:=\{g(H_{\rm ac}) {\rm D} (T_{\rm ac})| g\in C_0^\infty ({\mathbb{R}}\setminus L\cup K)\}$.
Let $$\tilde f (x)=f (x)-f (0). $$
Then $\lim_{x\to 0}\tilde f (x)=0$. We can write
$\sigma}%{\rm Sp} (\tilde f (H_{\rm p}))=\{F_j\}_{j=1}^\infty$, where $F_j\not=F_k,\,j\not=k$ and
the multiplicity of each $F_j$ is finite. It follows that $\lim_{j\to\infty}F_j=0$
and
$0\not\in \sigma}%{\rm Sp}_{\rm p} (\tilde f (H_{\rm p}))$.
Hence, by a minor modification of the proof of Lemma \ref{lem-uwt},
we can show that
there is an ultra-weak time operator ${\mathfrak{t}}_{\rm p}^{\tilde f}:\mathcal D_{\rm p}^f\times \mathcal D_{\rm p}^f\to {\mathbb{C}}$ of
$\tilde f (H_{\rm p})$, where $\mathcal D_{\rm p}^f$ is a dense subspace in ${\mathcal H}_{\rm p}(H)$.
We define a sesquilinear form
${\mathfrak{t}}_{H}^f :({\mathcal H}_{\rm ac}(H)\oplus \mathcal D_{\rm p}^f)\times ({\rm D}(T_{\rm ac}^f)\oplus
\mathcal D_{\rm p}^f)\to{\mathbb{C}}$
by
\eq{time11}
{\mathfrak{t}}_H^f [\phi_1\oplus\phi_2, \psi_1\oplus\psi_2] = (\phi_1, T_{\rm ac}^f\psi_1)+{\mathfrak{t}}_{\rm p}^{\tilde f}
[ \phi_2, \psi_2]
\end{equation}
for
$\phi_1\in {\mathcal H}_{\rm ac}(H), \psi_1\in {\rm D} (T_{\rm ac}^f)$ and $\phi_2, \psi_2\in \mathcal D_{\rm p}^f$.
Note that $f (0)$ is a scalar. Then one can show that
${\mathfrak{t}}_H ^f$ is an ultra-weak time operator with
$\mathcal D_{\rm ac}\oplus\widetilde {\ms E}_{\rm p}^{\tilde f}$ being
a ultra-weak CCR-domain, where $\mathcal D_{\rm ac}$ is a dense CCR-domain for $(f(H_{\rm ac}), T_{\rm ac}^f)$.
and $\widetilde{\ms E}_{\rm p}^{\tilde f}$ is an ultra-weak CCR-domain for $(\tilde f(H_{\rm p}),{\mathfrak{t}}_{\rm p}^{\tilde f})$.
\qed
\section{Applications to Schr\"odinger Operators}\label{sec6}
\subsection{Ultra-weak time operators of Schr\"odinger operators}\label{ex}
In this subsection, we apply Theorem \ref{thm-uwt1}
to the Schr\"odinger operator ${H_V}$ given by (\ref{HV}) to show that,
for a general class of potentials $V$, ${H_V}$ has an ultra-weak time operator with a {\it dense} ultra-weak CCR-domain.
This is done by collecting known results on spectral properties of Schr\"odinger operators.
Suppose that $V$ is of the form
\eq{agmon}
V (x)=\frac{W (x)}{ (|x|^2+1)^{{\half} +\varepsilon}},
\end{equation}
where $\varepsilon>0$ and
$W:{{\mathbb{R}}^d}\to{\mathbb{R}}$ is a Borel measurable function such that
$W (-\Delta+i)^{-1}$ is a compact operator on ${L^2(\BR)}$.
Such a potential $V$ is called an {\it Agmon potential} (\cite[p.439]{rs3} or \cite[p.169]{rs4}).
It is easily shown that $V$ is relatively compact with respect to
the free Hamiltonian $H_0$ given by (\ref{H0}).
Hence, by a general fact \cite[p.113, Corollary 2]{rs4},
${H_V}$ is self-adjoint with ${\rm D}({H_V})={\rm D} (H_0)$ and
\begin{equation}
\sigma}%{\rm Sp}_{\rm ess} ({H_V} )=\sigma}%{\rm Sp}_{\rm ess} (H_0)=[0, \infty),\label{HV-ess}
\end{equation}
where, for a self-adjoint operator $S$, $\sigma}%{\rm Sp}_{\rm ess}(S)$ denotes the essential spectrum of $S$.
Following facts are known as
Agmon-Kato-Kuroda theorem:
\bp{akk}\TTT{absence of $\sigma}%{\rm Sp}_{\rm sc} (H)$, existence and completeness of wave operators}
Let $V$ be an Agmon potential.
Then:
\begin{list}{}{}
\item{(1)} $\sigma}%{\rm Sp}_{\rm sc} ({H_V} )=\emptyset$.
\item{(2)}
The set of positive eigenvalues of ${H_V} $ is a discrete subset of $ (0, \infty)$.
\item{(3)} The wave operators
$\d \Omega_{\pm} :={\rm s}\!-\!\lim_{t\to\pm\infty}
e^{it{H_V} }e^{-itH_0}$
exist and complete: ${\rm Ran}(\d\Omega_{\pm})={\mathcal H}_{\rm ac}({H_V})$.
In particular $\sigma}%{\rm Sp}_{\rm ac} ({H_V} )=[0,\infty)$.
\end{list}
\end{proposition}
\proof
See \cite[Theorem XIII. 33]{rs4}.
\qed
In order to construct an ultra-weak time operator of
$({H_V})_{\rm p}$, we need
the condition
$\#\sigma}%{\rm Sp}_{\rm disc}({H_V}) =\infty$.
For this purpose, we introduce an assumption.
\begin {assumption}\label{v2}
{\rm There are constants $R_0, a>0$ and $\delta>0$ such that
\eq{takusan}
V (x)\leq -\frac{a}{|x|^{2-\delta}}\quad \mbox{ for } |x|>R_0.
\end{equation}
}
\end{assumption}
\begin{lemma}\TTT{infinite number of negative eigenvalues}
Let $V$ be an Agmon potential. Then, under Assumption \ref{v2},
$\sigma}%{\rm Sp}_{\rm disc} ({H_V})\subset (-\infty, 0)$ and
$\sigma}%{\rm Sp}_{\rm disc} ({H_V})$ is an infinite set.
In particular, the point $0\in{\mathbb{R}}$ is the unique accumulation point of $\sigma}%{\rm Sp}_{\rm disc}({H_V})$.
\end{lemma}
\proof Let $\mu_1:=\inf_{\psi\in {\rm D}({H_V});\|\psi\|=1}(\psi,{H_V}\psi)$ and
$$
\mu_n:=\sup_{\phi_1,\ldots,\phi_{n-1}\in {L^2(\BR)}}\inf_{\substack{\psi\in D({H_V});\|\psi\|=1\\
\psi\in \{\phi_1,\ldots,\phi_{n-1}\}^{\perp} }}(\psi,{H_V}\psi),\quad n\geq 2.
$$
In the case $d=3$, it is already known that $\mu_n<0$ for all $n\in \mathbb{N}$ \cite[Theorem XIII.6(a)]{rs4}.
It is easy to see that the method of the proof of this fact is valid also in the case of arbitrary $d$.
Hence we have $\mu_n<0$ for all $n\in \mathbb{N}$.
Then (\ref{HV-ess}) and the min-max principle imply the desired results.
\qed
\begin{assumption}\label{v3}
{\rm The potential $V$ is spherically symmetric, $V=V(|x|)$, and
\eq{nai}
\int_a^\infty |V (r)| dr<\infty
\end{equation}
for some $a>0$.
}\end{assumption}
\begin{lemma}\TTT{absence of strictly positive
eigenvalues} Let $V$ be an Agmon potential. Then,
under Assumption \ref{v3},
${H_V}$
has no strictly positive eigenvalues.
\end{lemma}
\proof Since ${\rm D}(V)\supset {\rm D}(H_0)\supset C_0^{\infty}({{\mathbb{R}}^d})$,
it follows that $V\in L_{\rm loc}^2({{\mathbb{R}}^d}\setminus \{0\})$.
Hence we can apply \cite[Theorem XIII.56]{rs4} to
derive the desired result.\qed
\bt{main2}\label{thm-uwt2}
Let $V$ be an Agmon potential such that $0\not\in \sigma}%{\rm Sp}_{\rm p} ({H_V}) $.
Suppose that Assumptions \ref{v2} and \ref{v3} hold.
Then ${H_V}$ has an ultra-weak time operator with a dense ultra-weak CCR-domain.
\end{theorem}
\proof
By Proposition \ref{akk}, $\sigma}%{\rm Sp}_{\rm sc} ({H_V})=\emptyset$ and
the wave operators $\Omega_{\pm}$ exist and are complete.
Hence, by Theorem \ref{thm-st-uni1},
$$
T_{{\rm ac}, j\pm}:=\Omega_{\pm} \tau_j\Omega_{\pm}^{-1}
P_{\rm ac}({H_V})\quad (j=1,\ldots,d)
$$
are strong time operators of $({H_V})_{\rm ac}$, where
$\tau_j:=\widetilde T_{{\rm AB},j}$ or $T_{{\rm AB},j}'$ denotes the Aharonov-Bohm time operators in Example \ref{ex-AB}.
Under Assumptions \ref{v2} and \ref{v3},
we can see that
$\sigma}%{\rm Sp} ({H_V} )=\{E_j\}_{j=1}^\infty\cup[0, \infty)$,
$E_1<E_2<\cdots<0$, $\lim_{n\to\infty}E_n=0$,
$\sigma}%{\rm Sp}_{\rm disc}({H_V} )=\{E_n\}_{n=1}^{\infty}$ and
$\sigma}%{\rm Sp}_{\rm ac} ({H_V} )=[0, \infty)$. Hence ${H_V}\in S_1(L^2({{\mathbb{R}}^d}))$.
Thus, by Theorem \ref{thm-uwt1}, we obtain the desired result.
\qed
Finally we consider conditions for
the absence of zero eigenvalue of
${H_V}$.
\bp{v4}\TTT{absence of zero eigenvalue}
Assume the following (1) and (2):
\begin{list}{}{}
\item{(1)} $d\geq 3$, $V\in L_{\rm loc}^{d/2}({{\mathbb{R}}^d})$.
\item{(2)} $V$ can be written as $V=V_1+V_2$, where $V_1$ and $V_2$ are real-valued
Borel measurable functions on ${{\mathbb{R}}^d}$ satisfying the following conditions:
\begin{list}{}{}
\item[(i)] There exists a constant $R>0$ such that $V_1$ and $V_2$ are locally bounded on $S_R=\{x\in{{\mathbb{R}}^d} | |x|>R\}$
and $V_1$ is strictly negative on
$S_R$,
\item[(ii)]
Let $S^{d-1}:=
\{w\in{{\mathbb{R}}^d}|\,|w|=1\}$, the $(d-1)$-dimensional unit sphere.
Then $V_1(rw)$ ($r=|x|$) is differentiable in $r>R$ and
there exist a constant $s\in (0,1)$ and a positive differentiable function $h$
on $[R,\infty)$ such that
$$
\d \sup_{w\in S^{d-1}} \frac{d}{dr}(r^{s+1} V_1(rw))\leq -r^{s} h(r)^2,\quad r>R.
$$
\item[(iii)] $\displaystyle\lim_{r\to\infty}\d \frac{r^{-1}+r \sup_{w\in S^{d-1}} |V_2(rw)|}{h(r)}=0$.
\item[(iv)] There exists a constant $C>0$ such that $\d \frac{d}{dr} h(r)\leq Ch^2(r)$ on $S_R$.
\item[(v)] For all all $f\in D({H_V})$,
$$\int_{S_R} h^2(|x|)|f(x)|^2 dx<\infty,\quad
\int_{S_R} |V_1(x)| |f(x)|^2 dx<\infty.
$$
\end{list}
\end{list}
Then $0\not\in \sigma}%{\rm Sp}_{\rm p}({H_V})$.
\end{proposition}
\proof
This is due to \cite[Theorem 2.4]{fs04} and \cite{jk85}. Also see
\cite{uch87}.
\qed
A key fact to prove Proposition \ref{v4} is as follows. Condition
$d\geq3$ and
$V\in L_{\rm loc}^{d/2} ({{\mathbb{R}}^d})$ imply
that, if a solution $f$ of partial differential equation
$-\Delta f+Vf=0$ satisfies that $f(x)=0$ for all $x\in S_R$ with some $R>0$,
then $f (x)=0$ for all $x\in {{\mathbb{R}}^d}$ by the unique continuation proven
in \cite{jk85}.
\begin{example}\label{ex-hyd}{\rm Let $d\geq 3$ and $V (x)=-1/|x|^{2-\varepsilon}$ with
$0<\varepsilon<2$. Then it is easy to check that the potential $V$ satisfies
conditions (1) and (2) in Proposition \ref{v4} (take $V_1=V$, $V_2=0$ and $h(r)=
\sqrt{s-1+\varepsilon}\,r^{(\varepsilon-2)/2}, \, r>0$ with $1-\varepsilon<s<1$). Hence, by Proposition \ref{v4}, ${H_V} $ has
no zero eigenvalue.
In particular,
the hydrogen Schr\"odinger operator
\begin{equation}
H_{\rm hyd}:= H_0-\frac{\gamma}{|x|} \label{H-hyd}
\end{equation}
for $d=3$ with a constant $\gamma>0$
has no zero eigenvalue.}
\end{example}
\begin{example}\label{47}
{\rm Let $d\geq 3$.
Suppose that $U\in L^\infty ({\mathbb{R}}^3)$.
Then
$$V (x)=\frac{U (x)}{ (1+|x|^2)^{{\half} +\varepsilon}}$$ is an Agmon potential for all $\varepsilon>0$.
Suppose that
$U$ is negative, continuous, spherically symmetric and
satisfies that
$U (x)=-1/|x|^\alpha$ for $|x|>R$
with $0<\alpha<1$ and $R>0$.
For each $\alpha$, we
can choose $\varepsilon>0$ such that
$2\varepsilon+\alpha<1$.
Hence $V$ satisfies \kak{takusan} and \kak{nai}.
Moreover it is easy to see that $V$ satisfies (1) and (2) in Proposition \ref{v4}
with
$$
V_1(x):=-\frac {\chi_{[R,\infty)}(|x|)}{|x|^{1+2\varepsilon+\alpha}},\quad
V_2(x):=\frac{U(x)}{(1+|x|^2)^{{\half}+\varepsilon}}+\frac {\chi_{[R,\infty)}(|x|)}{|x|^{1+2\varepsilon+\alpha}},
$$
where $\chi_{[R,\infty)}$ is the characteristic function of the interval $[R,\infty)$.
Hence, by Proposition \ref{v4}, $0\not\in \sigma}%{\rm Sp}_{\rm p}({H_V} )$.
Thus, by Theorem \ref{thm-uwt2},
${H_V}$ has an ultra-weak time operator with
a dense ultra-weak CCR-domain.
}
\end{example}
\begin{example}
\TTT{hydrogen atom}\label{ex-hyd2}
\label{main4}
{\rm
It is known that the hydrogen Schr\"odinger operator
$H_{\rm hyd}$ given by (\ref{H-hyd}) is self-adjoint with
${\rm D}(H_{\rm hyd})=D(H_0)$.
It is easy to see that the Coulomb potential $-\gamma/|x|$ with $d=3$ is not an Agmon potential.
Hence we can not apply Theorem \ref{thm-uwt2} to the case ${H_V}=H_{\rm hyd}$.
But we can show that $H_{\rm hyd}$ has an ultra-weak time operator in the following way.
The spectral properties of $H_{\rm hyd}$ are also well known:
$$
\sigma}%{\rm Sp}(H_{\rm hyd})=\sigma_{\rm p}(H_{\rm hyd})\cup \sigma_{\rm ac}(H_{\rm hyd}), \quad \sigma_{\rm sc}(H_{\rm hyd})=\emptyset
$$
with
$$
\sigma_{\rm p}(H_{\rm hyd})=\sigma_{\rm disc}(H_{\rm hyd})=
\left\{-\frac {m\gamma^2}{2n^2}|n\in\mathbb{N}\right\},\quad \sigma_{\rm ac}(H_{\rm hyd})=[0,\infty).
$$
The fact that
$0\not\in \sigma}%{\rm Sp}_{\rm p}(H_{\rm hyd})$ follows from Example \ref{ex-hyd} and Proposition \ref{v4}.
It is shown that the modified wave operators
$\mbox{\rm s-}\lim_{t\rightarrow \pm\infty} e^{itH_{\rm hyd}}Je^{-itH_0}$
with some unitary operator $J$ exist and are complete \cite[Theorems XI. 71 and XI.72]{rs3}.
These facts imply that $H_{\rm hyd} \in S_1(L^2({\mathbb{R}}^3))$.
Thus, by Theorem \ref{thm-uwt1}, $H_{\rm hyd}$ has an ultra-weak time operator with a dense ultra-weak CCR-domain.
}\end{example}
\subsection{Ultra-weak time operators of $f({H_V})$}
In this subsection, we assume that ${H_V}\in S_1(L^2({{\mathbb{R}}^d}))$ (see Definition \ref{def-S1})
and give some examples of functions $f:{\mathbb{R}}\to{\mathbb{R}}$ such that $f({H_V})$ has a ultra-weak-time operator
with a dense ultra-weak CCR-domain.
We first give a sufficient condition for (4) in Assumption \ref{hara} to hold.
\begin{lemma}\label{lem-f0} Let ${H_V}\in S_1(L^2({{\mathbb{R}}^d}))$ and $f:{\mathbb{R}}\to{\mathbb{R}}$, Borel measurable.
Suppose that, for all $n\in \mathbb{N}$, $f(E_n)\not=f(0)$ and $f(x)\not=f(0),\,{\rm a.e.}x\geq 0$.
Then $f(0)\not\in\sigma_{\rm p}({H_V})$.
\end{lemma}
\proof Let $\psi\in D(f({H_V}))$ such that $f({H_V})\psi=f(0)\psi$. Then
$$\|(f({H_V})-f(0))\psi\|^2=0$$ which is equivalent to
$\int_{{\mathbb{R}}}|f(\lambda)-f(0)|^2d\|E(\lambda)\psi\|^2=0$, where $E(\cdot)$ is the spectral measure of ${H_V}$.
We can decompose $\psi$ as $\psi=(\psi_{\rm ac},\psi_{\rm p})\in {\mathcal H}_{\rm ac}({H_V})\oplus{\mathcal H}_{\rm p}({H_V})$.
We denote by $\rho$ the Radon-Nykod\'ym derivative of the absolutely continuous
measure $\|E(\cdot)\psi_{\rm ac}\|^2$ with respect to the Lebesgue measure on ${\mathbb{R}}$.
Then we have
\begin{align*}
\int_{{\mathbb{R}}}|f(\lambda)-f(0)|^2d\|E(\lambda)\psi\|^2&=
\sum_{n=1}^{\infty}|f(E_n)-f(0)|^2\|E(\{E_n\})\psi_{\rm p}\|^2\\
& \quad +
\int_{[0,\infty)}|f(\lambda)-f(0)|^2\rho(\lambda)d\lambda.
\end{align*}
Hence, by the present assumption, $\|E(\{E_n\})\psi_{\rm p}\|^2=0\cdots (*)$ for all $n\in \mathbb{N}$ and
$\int_{[0,\infty)}|f(\lambda)-f(0)|^2\rho(\lambda)d\lambda=0\cdots(**)$.
Equation $(*)$ implies that $E(\{E_n\})\psi_{\rm p}=0,\,\forall n\geq 1$.
Since ${H_V}$ is $S_1(L^2({{\mathbb{R}}^d}))$, it follows that $\psi_{\rm p}\in {\mathcal H}_{\rm p}({H_V})^{\perp}$.
Hence $\psi_{\rm p}=0$. On the other hand, $(**)$ implies that $\rho(\lambda)=0$ a.e.$\lambda\in [0,\infty)$, from
which it follows that $\psi_{\rm ac}=0$. Thus $\psi=0$.
\qed
\begin{theorem}\label{thm-uwt-HS} Let ${H_V}\in S_1(L^2({{\mathbb{R}}^d}))$ and $f:{\mathbb{R}}\to{\mathbb{R}}$, Borel measurable.
Assume the following (1)--(4):
\begin{list}{}{}
\item{(1)} The function $f:{\mathbb{R}}\to{\mathbb{R}}$ satisfies the same
assumption as in Proposition~\ref{hkm09}.
\item{(2)} The function $f$ is continuous at $x=0$.
\item{(3)} $f(\sigma}%{\rm Sp}_{\rm disc} ({H_V}))$ is an infinite set such that
the multiplicity of each point in $f(\sigma}%{\rm Sp}_{\rm disc}({H_V}))$ as an eigenvalue of $f({H_V})$ is finite.
\item{(4)} For all $n\in \mathbb{N}$, $f(E_n)\not=f(0)$ and $f(x)\not=f(0),\,{\rm a.e.}x\geq 0$.
\end{list}
Then $f({H_V})$ has an ultra-weak time operator with a dense ultra-weak CCR-domain.
\end{theorem}
\proof By Lemma \ref{lem-f0}, property (4) in Assumption \ref{hara}
is satisfied. Hence, by Corollary \ref{main3}, the desired result is derived.
\qed
In Examples below, we assume that ${H_V}\in S_1(L^2({{\mathbb{R}}^d}))$.
\begin{example}\TTT{$f ({H_V})=e^{-\beta {H_V}}$}\label{exp}
{\rm
Let $f (x)=e^{-\beta x}$, $\beta\in{\mathbb{R}}\setminus\{0\}$.
Then it is easy to see that the function $f$ satisfies the assumption in Theorem \ref{thm-uwt-HS}.
Hence $e^{-\beta {H_V}}$ has an ultra-weak time operator with a dense ultra-weak CCR-domain.
Note that, if $\beta>0$ (resp. $\beta <0$), $e^{-\beta {H_V}}$ is bounded (resp. unbounded).
In particular,
$e^{-\beta H_{\rm hyd}}$ has an ultra-weak time operator with a dense ultra-weak CCR-domain.
}\end{example}
\begin{example}\TTT{$f ({H_V})=\sum_{j=0}^N a_j H_V^j$}\label{poly}
{\rm
Let $f (x)=\sum_{j=0}^N a_j x^j$ be a real polynomial ($a_j\in {\mathbb{R}}, N\in \mathbb{N}, \,a_N\not=0$).
We have $f(0)=a_0$.
Suppose that, for all $n\in \mathbb{N}$, $\sum_{j=1}^Na_jE_n^j\not=0$ and $\sum_{j=1}^Na_jx^{j-1}\not=0,\,x\geq 0$.
Then one can show that $f$ satisfies the assumption in Theorem \ref{thm-uwt-HS}.
Hence $\sum_{j=0}^Na_j{H_V}^j$ has an ultra-weak time operator with a dense ultra-weak CCR-domain.
In particular,
$\sum_{j=0}^Na_jH_{\rm hyd}^j$ has an ultra-weak time operator with a dense ultra-weak CCR-domain.
}\end{example}
\begin{example}\TTT{$f ({H_V})=\sin (2\pi \beta {H_V})$}\label{411}
{\rm
Let $f (x)=\sin (2\pi \beta x) $, $\beta\in {\mathbb{R}}\setminus\{0\}$.
Then $f(0)=0$. Let $\beta\not\in \{k/2E_n|k\in \mathbb{Z}, n\in\mathbb{N}\}$. Then
$\sin (2\pi\beta E_n)\not=0$ for all $n\in \mathbb{N}$ and hence
$f(E_n)\not=f(0)$. It is obvious that $f(x)\not=f(0)$ for a.e.$x\geq 0$.
Moreover $\Lambda:=\{\sin(2\pi\beta E_n)|n\in \mathbb{N}\}$ is an infinite set
and each point in $\Lambda$ as an eigenvalue of $\sin(2\pi\beta {H_V})$ is
in $\sigma_{\rm disc}(\sin(2\pi\beta{H_V}))$ (note that, for
$-1/4\beta\leq x<0$, $\sin(2\pi\beta x)$ is strictly monotone increasing).
In this way we can show that, in the present case, the assumption in Theorem \ref{thm-uwt-HS}
holds. Thus $\sin(2\pi\beta {H_V})$ has an ultra-weak time operator with a dense ultra-weak CCR-domain.
In particular,
$\sin(2\pi\beta H_{\rm hyd})$ has an ultra-weak time operator with a dense ultra-weak CCR-domain.
}\end{example}
In the same manner as above, one can find many concrete functions $f$ such that
$f({H_V})$ has an ultra-weak time operator with a dense ultra-weak CCR-domain.
\subsection*{Acknowledgments}
F. H. thanks Atsushi Inoue and Konrad Schm\"udgen for helpful discussions in a workshop held in Fukuoka University at March of 2014, and Jun Uchiyama and Erik Skibsted for sending papers \cite{uch87, fs04} to him, respectively.
He also thanks Aarhus University in Denmark and Rennes I University in France for kind hospitality. This work was partially done at these universities.
| 3c7728b52e43cb80f2d2ff333c4ec4c6d091f34b | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}\label{se:intro}
In this paper, we consider mixed integer minimization problem as follows:
\begin{equation}\label{p:primal}
(\mathcal{P}_b):~~~~\min \{P(x,v)= Q(x)+W(x)-f^Tv~|~(x,v)\in \x_v\}
\end{equation}
where $Q(x)=\frac{1}{2}x^T A x-c^Tx$, $W(x)= \frac{1}{2}\lpa\frac{1}{2}x^T B x-\alpha \rpa^2$, $ A\in\mathbb{R}^{n\times n}$ is a symmetric matrix and $ B\in\mathbb{R}^{n\times n}$ is a symmetric positive semi-definite matrix, $\alpha>0$ is a real number, $c,f\in \mathbb{R}^n$ are two given vectors, $v\in\{0,1\}^n$ means fixed cost variable, and
\begin{equation}
\x_v=\{(x,v)\in \mathbb{R}^n\times \{0,1\}^n~|~-v\leq x \leq v \}.
\end{equation}
The constraint $x\in [-v,v]$ with $v\in\{0,1\}^n$ is called as fixed charge constraint \cite{Padberg1985}and problem $\mathcal{P}_b$ belongs to the class of fixed charge problems, which have attracted considerable attention and extensive study in the integer programming literature in recent years \cite{Aardal1998,Dukwon1999,Glover2005,Steffen2009,Mathieu2013,Agra2014,Adlakha2015,Housh2015,Gaocost2008}. This class of problem has
many practical applications in a variety of problems, including facility location\cite{Aardal1998}, network design \cite{Dukwon1999,Steffen2009,Agra2014}, lot-sizing application \cite{Glover2005,Mathieu2013}, and transportation \cite{Mathieu2013,Housh2015,Gaocost2008,Adlakha2015}.The literature provides only a few exact methods for solving fixed charge problems, such as branch-and-bound type
methods, then a good deal of effort has been devoted to
finding approximate solutions to fixed charge problems by heuristic methods. Although these problems can be written as mixed integer programs, their computational requirements usually increase exponentially with
the size of the problem due to the existence of fixed cost terms in its objective function.
Canonical duality theory, as a breakthrough methodological theory, was originally developed from infinite dimensional nonconvex systems \cite{Gaocost2000}. It has been applied successfully for solving a class of challenging problems in global optimization and nonconvex analysis, such
as quadratic problems \cite{FangSC2008,Gao2004,Gao2010}, box constrained problems problems \cite{Gaobox2007a}, polynomial optimization \cite{Zhou2014,Gaobox2007a,Gao2006}, transportation problems
\cite{Gaocost2008}, location problems \cite{Gao2012}, and integer programming problems \cite{FangSC2008,Wang2012}.
Under some certain conditions, these problems can be solved by canonical
dual method to obtain global minimums of the primal problems.
Moreover, problem $\mathcal{P}_b$ is related with some problems solved by canonical duality method. Particularly, if the fixed cost term $f^Tv=0$ and the feasible space is defined by $\x=\{x\in \mathbb{R}^n~|~\mathscr{L}^l\leq x \leq \mathscr{L}^u\}$ where $\mathscr{L}^l$ and $\mathscr{L}^u$ are two given vectors, then $\mathcal{P}_b$ changes to the nonconvex polynomial programming problem with box constraints in \cite{Gaobox2007a}. If $f^Tv=0$ and the feasible space is simply defined by $\x=\{x\in \mathbb{R}^n~|~x_i =0~or~1, i = 1, \cdots, n\}$, $\mathcal{P}_b$ converts to the simple 0-1 quadratic programming
problem in \cite{FangSC2008}. If $W(x)=0$, $\mathcal{P}_b$ becomes the fixed cost quadratic problem that the canonical duality theory is introduced to solve in \cite{Gaocost2008}, where the existence and uniqueness of global optimal solutions are proved.
In this paper, we design a canonical duality algorithm to solve a mixed-integer nonconvex fourth-order polynomial minimization problem with fixed cost terms. The presented method is inspired by the method introduced by David Y. Gao for fixed cost quadratic programs \cite{Gaocost2008} and provides a nontrivial extension of his work to polynomial minimization problem. Based on the geometrically admissible operators of \cite{Gaobox2007a,Gaocost2008}, we define a new geometrically admissible operator and transform the fourth-order polynomial minimization problem in $\mathbb{R}^{2n}$ into a continuous concave maximization dual problem in $\mathbb{R}^{n+1}$ without duality gap. We give the global optimality conditions and obtain the existence and uniqueness criteria. Furthermore, we discuss the application to decouple problem. Some numerical experiments are given to show
our method is effective.
This paper is organised as follows. In the next section, the canonical dual problem for $\mathcal{P}_b$ is formulated, and the corresponding complementary-dual principle is showed. The global optimality criteria are put forward in Section 3, and the existence and uniqueness criteria are proposed in Section 4. We apply our method to decoupled problem in Section 5 and illustrate the effectiveness of our method by some examples in Section 6. Some conclusions and further work are given in the last section.
\section{Canonical dual problem}\label{se:cano}
We propose and describe the canonical dual problem of $\mathcal{P}_b$ without
duality gap in this section. First, the box constraints $-v\leq x \leq v$, $v\in \{0,1\}^n$ in the primal problem can be rewritten as relaxed quadratic form:
\begin{equation}\label{constrains}
x\circ x\leq v,~~~~v\circ(v-e)\leq 0,
\end{equation}
where $e=\{1\}^n$ is an n-vector of all ones and notation $x\circ v= (x_1v_1,
x_2v_2,\cdots, x_nv_n)$ denotes the Hadamard product between any two vectors $x,v\in \mathbb{R}^n$. So $\mathcal{P}_b$ can be reformulated to the relaxed problem in the following:
\begin{equation}\label{p:relax}
(\mathcal{P}_r):~~~~\min \{P(x,v)= \frac{1}{2}x^T A x-c^Tx+\frac{1}{2}\lpa\frac{1}{2}x^T B x-\alpha \rpa^2-f^Tv~|~x\circ x\leq v,v\circ(v-e)\leq 0\}.
\end{equation}
In our paper we introduce a so-called geometrically admissible operator
\begin{equation}\label{operator}
y=\Lambda(x,v)=\bpmat \xi\\ \epsilon \epmat=\bpmat \xi\\ \epsilon_1 \\ \epsilon_2 \epmat=\bpmat
\frac{1}{2}x^T Bx-\alpha\\
x\circ x-v\\
v\circ v-v\\
\epmat
\in\mathbb{R}^{2n+1},
\end{equation}
and it is trivial to see that the constraints (\ref{constrains}) are equivalent to $\epsilon\leq 0$.
Define
\begin{equation}
V(y)=\frac{1}{2}\xi^2+\Psi(\epsilon),
\end{equation}
where
\begin{equation}
\Psi(\epsilon)=
\left\{\begin{array}{ll}
0 &~~ if~~\epsilon\leq 0\in\mathbb{R}^{2n},\\
+\infty &~~\wotherwise,
\end{array}\right.
\end{equation}
and let $U(x,v)=-Q(x)+f^Tv$, then unconstrained canonical form of relaxed primal problem $(\mathcal{P}_r)$
can be obtained:
\begin{equation}\label{pc}
(\mathcal{P}_c):~~~~\min \{\Pi(x,v)= V(\Lambda(x,v))-U(x,v)~|~x\in \mathbb{R}^n,v\in \mathbb{R}^n\}.
\end{equation}
Set $y^*=\left(\begin{array}{l}
\varsigma\\
\sigma
\end{array}\right)=\left(\begin{array}{l}
\varsigma\\
\sigma_1\\
\sigma_2
\end{array}\right)$ be the dual vector of $y=\left(\begin{array}{l}
\xi\\
\epsilon
\end{array}\right)$ with $\epsilon\leq 0$, then sup-Fenchel conjugate of V (y) can be
defined as
\begin{eqnarray}\label{tianjian2002a}
V^\sharp(y^*)&=&\sup_{y\in\bbR^{1+2n}}\{y^Ty^*-V(y)\}\nonumber\\
&=&\sup_{\xi\geq -\alpha}\{\xi\varsigma-\frac{1}{2}\xi^2\}+\sup_{\epsilon_1\in\bbR^{n}}\sup_{\epsilon_2\in\bbR^{n}}\{\epsilon_1^T\sigma_1+\epsilon_2^T\sigma_2-\Psi(\epsilon)\}\nonumber\\
&=&\frac{1}{2}\varsigma^2+\Psi^\sharp(\sigma),~s.t.~\varsigma\geq -\alpha,
\end{eqnarray}
where
\begin{equation}\label{tianjian2002}
\Psi^\sharp(\sigma)=
\left\{\begin{array}{ll}
0 &~~ if~~\sigma_1\geq 0\subseteq\mathbb{R}^{n},\sigma_2\geq 0\subseteq\mathbb{R}^{n}\\
+\infty &~~\wotherwise.
\end{array}\right.
\end{equation}
Now the extended canonical duality
relations can be given from the theory of convex analysis:
\begin{equation}\label{tianjia2004}
y^*\in \partial V(y)~\Leftrightarrow~y\in \partial V^\sharp(y^*)~\Leftrightarrow~V(y)+V^\sharp(y^*)=y^Ty^*,
\end{equation}
and when $\epsilon\leq 0$ and $\sigma\geq 0$ we have
$$\epsilon_1^T\sigma_1+\epsilon_2^T\sigma_2=\frac{1}{2}(\xi-\varsigma)^2\geq 0,$$
then it yields that
\begin{equation}\label{complementarity}
\epsilon_1^T\sigma_1=0~and~\epsilon_2^T\sigma_2=0.
\end{equation}
Apparently it holds from $\epsilon_2^T\sigma_2=\sigma_2^T(v\circ v-v)=0$ in (\ref{complementarity}) that $v\circ v-v=0~for~\forall~\sigma_2>0$.
By replacing $V(\Lambda(x,v))$ in (\ref{pc}) by the Fenchel-Young equality
$V(\Lambda(x,v))=\Lambda(x,v)^Ty^*-V^\sharp(y^*)$, the
total complementary function $\vX(x,v,\varsigma,\sigma)$ corresponding to $(\mathcal{P}_c)$ can be defined in the following:
\begin{eqnarray}\label{duiouhanshu}
\vX(x,v,\varsigma,\sigma)&=&\Lambda(x,v)^Ty^*-V^\sharp(y^*)-U(x,v)\nonumber\\
&=&\frac{1}{2}x^T G(\varsigma,\sigma_1)x-c^Tx+v^T Diag(\sigma_2)v-(f+\sigma_1+\sigma_2)^Tv\nonumber\\
&&-\frac{1}{2}\varsigma^2-\alpha\varsigma-\Psi^\sharp(\sigma),
\end{eqnarray}
where
\begin{eqnarray}\label{tjre001}
G(\varsigma,\sigma_1)=A+\varsigma B+2Diag(\sigma_1),
\end{eqnarray}
and $Diag(\sigma_1)$ denotes a diagonal matrix with diagonal elements ${(\sigma_1)}_i$,
$i = 1, \cdots, n$. Then we get the canonical dual function:
\begin{eqnarray}
\vP^d(\varsigma,\sigma)=\sta\lbc\vX(x,v,\varsigma,\sigma)~|~x\in\mbb{R}^n,v\in\mbb{R}^n\rbc=U^{\Lambda}(\varsigma,\sigma)-V^\sharp(\varsigma,\sigma),
\end{eqnarray}
where $U^{\Lambda}(\varsigma,\sigma)=\sta\lbc\Lambda(x,v)^Ty^*-U(x,v)~|~x\in\mbb{R}^n,v\in\mbb{R}^n\rbc$
is the $\Lambda$-conjugate transformation and the notation $\sta\{\cdot\}$ represents calculating stationary point with respect to $(x,v)$. Correspondingly, introducing a dual feasible space
\begin{equation}\label{inequality}
S_\sharp=\{(\varsigma,\sigma)\in \mathbb{R}^1\times \mathbb{R}^{2n}~|~\varsigma\geq -\alpha,\sigma_1\geq 0,\sigma_2>0, c\in\mathcal{C}_{ol}(G(\varsigma,\sigma_1))\},
\end{equation}
where $\mathcal{C}_{ol}(G)$ denotes the column space of $G$, we can construct the canonical dual function as follows:
\begin{eqnarray}
\vP^d(\varsigma,\sigma)&=&U^{\Lambda}(\varsigma,\sigma)-\frac{1}{2}\varsigma^2\nonumber\\
&=&-\frac{1}{2}c^T G^+(\varsigma,\sigma_1)c-\frac{1}{4}\sum_{i=1}^n\frac{1}{{(\sigma_2)}_i}\lpa f_i+{(\sigma_1)}_i+{(\sigma_2)}_i\rpa^2\nonumber\\
&&-\frac{1}{2}\varsigma^2-\alpha\varsigma,~~\forall (\varsigma,\sigma)\in S_\sharp,\label{2000}
\end{eqnarray}
where $G^+$ stands for the Moore-Penrose generalized inverse of $G$. Then it leads to the canonical dual problem for the primal problem $(\mcal{P}_b)$ as
\begin{equation}\label{p:dual}
(\mcal{P}^\sharp):~~~~\max\lbc\vP^d(\varsigma,\sigma)~|~(\varsigma,\sigma)\in S_\sharp\rbc.
\end{equation}
Next we present the complementary-dual principle and for simplicity we denote $t\oslash s=\{t_i/s_i\}^n$ for any given n-vectors $t=\{t_i\}^n$ and $s=\{s_i\}^n$.
\begin{theorem}\label{th:Complementary-Dual}
If $(\bar\varsigma,\bar\sigma)\in S_\sharp$ is a KKT point of $\vP^d(\varsigma,\sigma)$, then the vector $(\bar x,\bar v)$ is feasible to the primal problem ($\mathcal{P}_b$) and we have
\beq\label{eq:nogap}
P(\bar x, \bar v)=\vX(\bar x,\bar v,\bar \varsigma,\bar \sigma)=\vP^d(\bar \varsigma,\bar \sigma),
\eeq
where
\beq\label{eq:solvedx1}
\bar{x}=\bar{x}(\bar\varsigma,\bar\sigma_1)=G^+(\bar\varsigma,\bar\sigma_1)c,
\eeq
\beq\label{eq:solvedx2}
\bar{v}=\bar{v}(\bar\sigma)=\frac{1}{2}(f+\bar\sigma_1+\bar\sigma_2)\oslash \bar\sigma_2.
\eeq
\end{theorem}
\begin{proof}
Introducing lagrange multiplier $\epsilon=(\epsilon_0,\epsilon_1,\epsilon_2)\in \mathbb{R}^{1}\times\mathbb{R}^{n}\times\mathbb{R}^{n}$ with the respective three inequalities in (\ref{inequality}), we have the lagrangian function $\Theta$ for problem ($\mathcal{P}^\sharp$):
\begin{equation}
\Theta(\varsigma,\sigma,\epsilon_0,\epsilon_1,\epsilon_2)=\vP^d(\varsigma,\sigma)-\epsilon_0(\varsigma+\alpha)
-\epsilon_1^T\sigma_1-\epsilon_2^T\sigma_2.
\end{equation}
It follows from the criticality conditions
$$\nabla_{\sigma_1}\Theta(\bar\varsigma,\bar\sigma,\epsilon_0,\epsilon_1,\epsilon_2)=0,~~~~\nabla_{\sigma_2}\Theta(\bar\varsigma,\bar\sigma,\epsilon_0,\epsilon_1,\epsilon_2)=0$$
that
\begin{equation}\label{2001}
\epsilon_1=\nabla_{\sigma_1}\vP^d(\bar\varsigma,\bar\sigma)=\bar{x}(\bar\varsigma,\bar\sigma_1)\circ \bar{x}(\bar\varsigma,\bar\sigma_1)-\bar{v}(\bar\sigma)
\end{equation}
\begin{equation}\label{2002}
\epsilon_2=\nabla_{\sigma_2}\vP^d(\bar\varsigma,\bar\sigma)=\bar{v}(\bar\sigma)\circ \bar{v}(\bar\sigma)-\bar{v}(\bar\sigma),
\end{equation}
where $\bar{x}(\bar\varsigma,\bar\sigma_1)=G^+(\bar\varsigma,\bar\sigma_1)c$ and $\bar{v}(\bar\sigma)=\frac{1}{2}(f+\bar\sigma_1+\bar\sigma_2)\oslash \bar\sigma_2$.
Then the corresponding KKT conditions include
\begin{equation}\label{2003}
\bar\sigma_1^T\epsilon_1=0~and~\bar\sigma_2^T\epsilon_2=0,
\end{equation}
where $\bar\sigma_1\geq 0$, $\bar\sigma_2>0$, $\epsilon_1\leq 0$ and $\epsilon_2\leq 0$.
By $\epsilon_1\leq 0$ and (\ref{2001}), we have $\bar x\circ \bar x\leq \bar v$. Clearly, together with $\bar\sigma_2>0$ and (\ref{2002}), $\bar\sigma_2^T(\bar v\circ \bar v-\bar v)=0$ in (\ref{2003}) implies that $\bar v\circ \bar v=\bar v$. So when $(\bar\varsigma,\bar\sigma)$ is a KKT point of the problem $\vP^d(\varsigma,\sigma)$, $(\bar x,\bar v)$ is feasible point of ($\mathcal{P}_b$).
Furthermore,
$\bar{v}=\frac{1}{2}(f+\bar \sigma_1+\bar \sigma_2)\oslash \bar \sigma_2$ implies $$\sum_{i=1}^n{(\bar \sigma_2)}_i\lpa\frac{1}{2}\frac{f_i+{(\bar \sigma_1)}_i+{(\bar \sigma_2)}_i}{{(\bar \sigma_2)}_i} \rpa^2=\bar \sigma_2^T(\bar v\circ \bar v)=\bar v^T Diag(\bar \sigma_2)\bar v$$ and $$2 \bar v^T Diag(\bar \sigma_2)\bar v=\bar v^T(f+\bar \sigma_1+\bar \sigma_2),$$
then with $\bar{x}=G^+(\bar \varsigma,\bar \sigma_1)c$ and $\Psi^\sharp(\bar\sigma)=0$, from (\ref{2000}) we get
\begin{eqnarray}\label{2005}
\vP^d(\bar \varsigma,\bar \sigma)
&=&\frac{1}{2}x^T G(\bar \varsigma,\bar \sigma_1)\bar x-c^T \bar x-\bar v^T Diag(\bar \sigma_2)\bar v-\frac{1}{2}\bar \varsigma^2-\alpha\bar \varsigma\nonumber\\
&=&\frac{1}{2}\bar x^T G(\bar \varsigma,\bar \sigma_1)\bar x-c^T \bar x+\bar v^T Diag(\bar \sigma_2)\bar v-\bar v^T(f+\bar\sigma_1+\bar\sigma_2)-\frac{1}{2}\bar \varsigma^2-\alpha\bar \varsigma\\
&=&\vX(\bar x,\bar v,\bar\varsigma,\bar\sigma),\nonumber
\end{eqnarray}
where $(\bar\varsigma,\bar\sigma)\in S_\sharp$. By (\ref{tjre001}), (\ref{2005}),
$\bar v^T Diag(\bar \sigma_2)\bar v=\bar \sigma_2^T(\bar v\circ \bar v)$ and $\bar x^T Diag(\bar \sigma_1)\bar x=\bar \sigma_1^T(\bar x\circ \bar x)$,
we have
\begin{eqnarray}\label{2006}
\vP^d(\bar \varsigma,\bar \sigma)
&=&\vX(\bar x,\bar v,\bar\varsigma,\bar\sigma)\nonumber\\
&=&\frac{1}{2}\bar x^T \Big(A+\bar \varsigma B+2Diag(\bar \sigma_1)\Big)\bar x-c^T \bar x+\bar \sigma_2^T(\bar v\circ \bar v)-\bar v^T(f+\bar \sigma_1+\bar \sigma_2)\nonumber\\
&&-\frac{1}{2}\bar \varsigma^2-\alpha\bar \varsigma\nonumber\\
&=&\frac{1}{2}\bar x^T A\bar x-c^T \bar x-f^T\bar v+\bar \sigma_1^T(\bar x\circ \bar x-\bar v)+\bar \sigma_2^T(\bar v\circ \bar v-\bar v)\nonumber\\
&&+\frac{1}{2}\bar \varsigma \bar x^T B\bar x-\frac{1}{2}\bar \varsigma^2-\alpha\bar \varsigma.
\end{eqnarray}
By substituting $\bar \sigma_1^T(\bar x\circ \bar x-\bar v)=0$ and $\bar\sigma_2^T(\bar v\circ \bar v-\bar v)=0$ into (\ref{2006}), we obtain
\begin{eqnarray}\label{2008}
\vP^d(\bar \varsigma,\bar \sigma)
=\vX(\bar x,\bar v,\bar\varsigma,\bar\sigma)=\frac{1}{2}\bar x^T A\bar x-c^T \bar x-f^T\bar v+\frac{1}{2}\bar \varsigma \bar x^T B\bar x-\frac{1}{2}\bar \varsigma^2-\alpha\bar \varsigma.
\end{eqnarray}
From $\nabla_{\varsigma}\Theta(\bar\varsigma,\bar\sigma,\epsilon_0,\epsilon_1,\epsilon_2)=0$,
together with $\bar{x}=G^+(\bar\varsigma,\bar\sigma_1)c$ it follows that
\begin{equation}\label{tianjia2001}
\frac{1}{2}\bar x^T B \bar x-\bar \varsigma -\alpha-\epsilon_0=0,
\end{equation}
and the accompanying KKT conditions include
\begin{equation}
\bar\varsigma +\alpha\geq 0,~\epsilon_0\leq 0,~\epsilon_0(\bar\varsigma +\alpha)=0.
\end{equation}
Suppose $\epsilon_0< 0$, it holds from $\epsilon_0(\bar\varsigma +\alpha)=0$ that $\bar\varsigma +\alpha=0$, then due to (\ref{tianjia2001}) we have
$$\frac{1}{2}\bar x^T B \bar x-\epsilon_0=0,$$
on the other hand, since $B$ is a symmetric positive semi-definite matrix and $\epsilon_0< 0$, we acquire
$$\frac{1}{2}\bar x^T B \bar x-\epsilon_0>0,$$
which is a contradiction, then $\epsilon_0=0$. Thereby from (\ref{tianjia2001}), we have
\begin{equation}
\frac{1}{2}\bar x^T B \bar x-\bar \varsigma -\alpha=0,
\end{equation}
which implies that $\alpha=\frac{1}{2}\bar x^T B \bar x-\bar \varsigma$
and $\bar \varsigma=\frac{1}{2}\bar x^T B \bar x-\alpha$, then the following equality holds:
$$\frac{1}{2}\bar \varsigma^2+\alpha\bar \varsigma=\frac{1}{2}\bar \varsigma^2+\Big(\frac{1}{2}\bar x^T B \bar x-\bar \varsigma\Big)\bar \varsigma=\frac{1}{2}\bar \varsigma \bar x^T B\bar x-\frac{1}{2}\bar \varsigma^2=\frac{1}{2}\bar \varsigma \bar x^T B\bar x-\frac{1}{2}\lpa\frac{1}{2}\bar x^T B \bar x-\alpha \rpa^2.$$
Thus it is apparent from (\ref{2008}) that
\begin{eqnarray}
\vP^d(\bar \varsigma,\bar \sigma)
=\vX(\bar x,\bar v,\bar\varsigma,\bar\sigma)=\frac{1}{2}\bar x^T A\bar x-c^T \bar x+\frac{1}{2}\lpa\frac{1}{2}\bar x^T B \bar x-\alpha \rpa^2-f^T\bar v
=P(\bar x, \bar v).
\end{eqnarray}
The proof is completed.
\hfill\qed
\end{proof}
\section{Global Optimality Criteria}
The global optimality conditions for
problem $(\mathcal{P}_b)$ are developed in this section. Firstly, we introduce the following useful feasible space:
\begin{equation}\label{3001}
S^+_\sharp=\{(\varsigma,\sigma)\in \mathbb{R}^1\times \mathbb{R}^{2n}~|~\varsigma\geq -\alpha,\sigma_1\geq 0,\sigma_2>0, G(\varsigma,\sigma_1)\succ 0\},
\end{equation}
where $G(\varsigma,\sigma_1)\succ 0$ means that $G(\varsigma,\sigma_1)$ is a positive definite matrix. It is easy to prove that $S^+_\sharp$ is a convex set. In the following, we use $G$ for short to denote $G(\varsigma,\sigma_1)$.
For convenience, we give the first and second derivatives of function $\vP^d(\varsigma,\sigma)$:
\begin{eqnarray}
\nabla \vP^d(\varsigma,\sigma)
&=&\left(\begin{array}{l}
\frac{1}{2}c^T G^{-1} \frac{\partial G}{\partial\varsigma} G^{-1}c-\varsigma-\alpha\\
\\
\lbc\frac{1}{2}c^T G^{-1} \frac{\partial G}{\partial(\sigma_1)_i} G^{-1}c-\frac{1}{2}\frac{f_i+(\sigma_1)_i+(\sigma_2)_i}{(\sigma_2)_i}\rbc_{i=1}^n\\
\\
\lbc-\frac{1}{2}\frac{f_j+(\sigma_1)_j+(\sigma_2)_j}{(\sigma_2)_j}
+\frac{1}{4}\lpa\frac{f_i+(\sigma_1)_i+(\sigma_2)_i}{(\sigma_2)_i}\rpa^2\rbc_{j=1}^n
\end{array}\right),\label{3004}\\
\nabla^2\vP^d(\varsigma,\sigma)&=&-J_1-J_2-J_3,\label{3005}
\end{eqnarray}
in which $J_1$, $J_2$ and $J_3\in\mbb{R}^{(2n+1)\times (2n+1)}$ are defined as
$$
J_1=\begin{bmatrix} 1 & 0& \cdots& 0\\0 & 0& \cdots& 0\\ \vdots& \vdots& \ddots & \vdots\\ 0 & 0& \cdots& 0 \end{bmatrix}, ~~
J_2=\begin{bmatrix} Z^TG^{-1}Z & \bm 0_{(n+1)\times n}\\\bm 0_{n\times(n+1)} & \bm 0_{n\times n} \end{bmatrix}, \wand
J_3=\begin{bmatrix} 0 & \bm 0_{1\times n}& \bm 0_{1\times n}\\\bm 0_{n\times 1}&H_{2\sigma_1^2} & H_{\sigma_1\sigma_2} \\\bm 0_{n\times 1}&H_{\sigma_2\sigma_1} & H_{\sigma_2^2}\end{bmatrix},
$$
where
\begin{eqnarray*}
&& Z=
\begin{bmatrix}
\frac{\partial G}{\partial\varsigma}G^{-1}c, \frac{\partial G}{\partial(\sigma_1)_1}G^{-1}c, \frac{\partial G}{\partial(\sigma_1)_2}G^{-1}c,\ldots, \frac{\partial G}{\partial(\sigma_1)_n}G^{-1}c
\end{bmatrix},\\
&&H_{2\sigma_1^2}= Diag\lbc-\frac{1}{2(\sigma_2)_i}\rbc,\\
&& H_{\sigma_1\sigma_2} =H_{\sigma_2\sigma_1} = Diag\lbc\frac{(\sigma_1)_i+f_i}{2(\sigma_2)^2_i}\rbc ,
\\
&& H_{\sigma_2^2}= Diag\lbc-\frac{((\sigma_1)_i+f_i)^2}{2(\sigma_2)^3_i}\rbc.
\end{eqnarray*}
\begin{lemma}\label{lemma2}
The canonical dual function $\vP^d(\varsigma,\sigma)$ is concave on $S^+_\sharp$.
\end{lemma}
\begin{proof}
For any given non-zero vector $W=
\left(\begin{array}{l}
r\\
s\\
t
\end{array}\right)\in \mbb{R}^{2n+1}$, where $r\in \mbb{R}^{1}$, $s\in \mbb{R}^{n}$, $t\in \mbb{R}^{n}$,
let $Z_0=\left(\begin{array}{l}
r\\
s
\end{array}\right)$, by (\ref{3005}), with $G\succ 0$ and $\sigma_2>0$ we have
\begin{eqnarray}\label{3006}
W^T\nabla^2\vP^d(\varsigma,\sigma)W&=&-W^TJ_1W-W^TJ_2W-W^TJ_3W\nonumber\\
&=&-r^2-(ZZ_0)^TG^{-1}(ZZ_0)+\sum_{i=1}^n-\frac{1}{2(\sigma_2)_i}\lpa s_i-t_i\frac{(\sigma_1)_i+f_i}{(\sigma_2)_i} \rpa^2\nonumber\\
&\leq&0,
\end{eqnarray}
so the canonical dual function $\vP^d(\varsigma,\sigma)$ is concave on $S^+_\sharp$.
\hfill\qed
\end{proof}
\begin{theorem}\label{theorem4}
Suppose that the vector $\bar{y}^*=(\bar \varsigma,\bar \sigma)=(\bar \varsigma,\bar \sigma_1,\bar \sigma_2)\in S^+_\sharp$ is a critical point
of the dual function $\vP^d(\varsigma,\sigma)$, then $\bar{y}^*$ is a global maximizer of $\vP^d(\varsigma,\sigma)$ on $S^+_\sharp$. Let $(\bar{x}, \bar v)=\lpa G^{-1}(\bar\varsigma,\bar\sigma_1)c,\frac{1}{2}(f+\bar\sigma_1+\bar\sigma_2)\oslash \bar\sigma_2)\rpa$, the $(\bar{x}, \bar v)$ is a global minimum of $P(x,v)$ on $\x_v$(i.e., the $(\bar{x}, \bar v)$ is a global solution of $(\mathcal{P}_b)$), and
\begin{equation}\label{3007}
P(\bar{x}, \bar v) =\min_{(x,v)\in\x_v} P(x,v)=\max_{(\varsigma,\sigma)\in S^+_\sharp} \vP^d(\varsigma,\sigma)=\vP^d(\bar \varsigma,\bar \sigma).
\end{equation}
\end{theorem}
\begin{proof}
From Lemma \ref{lemma2} the dual function $\vP^d(\varsigma,\sigma)$ is concave on $S^+_\sharp$, so
$\bar{y}^*=(\bar \varsigma,\bar \sigma)$ is a global maximizer of $\vP^d(\varsigma,\sigma)$ on $S^+_\sharp$, i.e.,
\begin{equation}\label{3009}
\vP^d(\bar \varsigma,\bar \sigma)=\max_{(\varsigma,\sigma)\in S^+_\sharp} \vP^d(\varsigma,\sigma),
\end{equation}
and this critical point
of $\vP^d(\varsigma,\sigma)$ is a KKT point of $\vP^d(\varsigma,\sigma)$. Then by Theorem \ref{th:Complementary-Dual},
the vector $(\bar{x}, \bar v)$ defined
by (\ref{eq:solvedx1})(now $G^+(\bar\varsigma,\bar\sigma_1)=G^{-1}(\bar\varsigma,\bar\sigma_1)$) and (\ref{eq:solvedx2}) is a feasible solution to problem $(\mathcal{P}_b)$ and
\begin{equation}\label{3008}
P(\bar x, \bar v)=\vX(\bar x,\bar v,\bar \varsigma,\bar \sigma)=\vP^d(\bar \varsigma,\bar \sigma).
\end{equation}
As $G\succ 0$ and $\sigma_2>0$, we have $\nabla^2_{(x,v)}\vX(x,v,\varsigma,\sigma)\succ 0$
and
$\nabla^2_{(\varsigma,\sigma)}\vX(x,v,\varsigma,\sigma)\preceq 0$. So $\vX(x,v,\varsigma,\sigma)$ is convex in $(x, v)\in \mbb{R}^{2n}=\mbb{R}^{n}\times \mbb{R}^{n}$ and concave in $y^*=(\varsigma,\sigma)\in S^+_\sharp$, then
\begin{equation}\label{3010}
\vP^d(\varsigma,\sigma)=\min_{(x, v)\in \mbb{R}^{2n}} \vX(x,v,\varsigma,\sigma),
\end{equation}
and $(\bar x,\bar v,\bar{y}^*)=(\bar x,\bar v,\bar \varsigma,\bar \sigma)$ is a saddle point of the total complementary function $\vX(\bar x,\bar v,\bar \varsigma,\bar \sigma)$ on $\mbb{R}^{2n}\times S^+_\sharp$, thereby the saddle min-max duality theory holds
\begin{equation}\label{3011}
\max_{(\varsigma,\sigma)\in S^+_\sharp} \min_{(x, v)\in \mbb{R}^{2n}}\vX(x,v,\varsigma,\sigma)= \min_{(x, v)\in \mbb{R}^{2n}}\max_{(\varsigma,\sigma)\in S^+_\sharp}\vX(x,v,\varsigma,\sigma).
\end{equation}
Combining (\ref{3008}), (\ref{3009}), (\ref{3010}) and (\ref{3011}), we get
\begin{equation}\label{3012}
P(\bar x, \bar v)= \min_{(x, v)\in \mbb{R}^{2n}}\max_{(\varsigma,\sigma)\in S^+_\sharp}\vX(x,v,\varsigma,\sigma),
\end{equation}
where $(\bar{x}, \bar v)\in \x_v$.
By (\ref{operator}),
$\Lambda(x,v)=\bpmat \xi\\ \epsilon \epmat=\bpmat \xi\\ \epsilon_1 \\ \epsilon_2 \epmat =\bpmat
\frac{1}{2}x^T Bx-\alpha\\
x\circ x-v\\
v\circ v-v\\
\epmat$.
When $(\varsigma,\sigma)\in S^+_\sharp$, together with (\ref{tianjian2002a}) and (\ref{tianjian2002}), $V^\sharp(y^*)=\frac{1}{2}\varsigma^2$. From (\ref{duiouhanshu}) and $U(x,v)=-Q(x)+f^Tv$, when $(\varsigma,\sigma)\in S^+_\sharp$ we obtain
\begin{eqnarray}\label{3013}
\vX(x,v,\varsigma,\sigma)&=&\Lambda(x,v)^Ty^*-V^\sharp(y^*)-U(x,v)\nonumber\\
&=&\bpmat \xi\\ \epsilon_1 \\ \epsilon_2 \epmat^T\bpmat \varsigma\\ \sigma_1 \\ \sigma_2 \epmat-\frac{1}{2}\varsigma^2-(-Q(x)+f^Tv)\nonumber\\
&=&Q(x)-f^Tv-\frac{1}{2}\varsigma^2+\xi\varsigma+\epsilon_1^T\sigma_1+\epsilon_2^T\sigma_2,
\end{eqnarray}
(a) Consider $\max_{(\varsigma,\sigma)\in S^+_\sharp}(-\frac{1}{2}\varsigma^2+\xi\varsigma)$.
\begin{equation}\label{3014aa}
\max_{(\varsigma,\sigma)\in S^+_\sharp}(-\frac{1}{2}\varsigma^2+\xi\varsigma)=\frac{1}{2}\xi^2= \frac{1}{2}\lpa\frac{1}{2}x^T B x-\alpha \rpa^2=W(x).
\end{equation}
(b) Consider $\max_{(\varsigma,\sigma)\in S^+_\sharp}(\epsilon_1^T\sigma_1+\epsilon_2^T\sigma_2)$.
Noting $\bpmat \epsilon_1 \\ \epsilon_2 \epmat =\bpmat
x\circ x-v\\
v\circ v-v\\
\epmat$, by (\ref{constrains}) we define the relaxed quadratic form region
\begin{equation}\label{3014}
\bar\x_v=\{(x,v)\in \mathbb{R}^n\times \mathbb{R}^n~|~x\circ x\leq v,~v\circ(v-e)\leq 0 \},
\end{equation}
then we have $\x_v\subset\bar\x_v$ and
\begin{equation}\label{3015}
\bpmat \epsilon_1 \\ \epsilon_2 \epmat \leq0 \Longleftrightarrow (x,v)\in \bar\x_v.
\end{equation}
It follows from $(\varsigma,\sigma)\in S^+_\sharp$ that $\sigma_1\geq 0$ and $\sigma_2>0$, thus with (\ref{3015}) we have
\begin{equation}\label{3016}
\max_{(\varsigma,\sigma)\in S^+_\sharp}(\epsilon_1^T\sigma_1+\epsilon_2^T\sigma_2)=
\left\{\begin{array}{ll}
0 &~~ if~~(x,v)\in \bar\x_v,\\
+\infty &~~\wotherwise(~i.e.,~if~(x,v)\notin \bar\x_v).
\end{array}\right.
\end{equation}
Taking (\ref{3013}), (\ref{3014aa}) and (\ref{3016}) into consideration,
we find
\begin{eqnarray}\label{3017}
\max_{(\varsigma,\sigma)\in S^+_\sharp}\vX(x,v,\varsigma,\sigma)&=&Q(x)-f^Tv+\max_{(\varsigma,\sigma)\in S^+_\sharp}(-\frac{1}{2}\varsigma^2+\xi\varsigma)+\max_{(\varsigma,\sigma)\in S^+_\sharp}(\epsilon_1^T\sigma_1+\epsilon_2^T\sigma_2)\nonumber\\
&=&
\left\{\begin{array}{ll}
Q(x)+W(x)-f^Tv=P(x,v) &~~ if~~(x,v)\in \bar\x_v,\\
+\infty &~~\wotherwise.
\end{array}\right.
\end{eqnarray}
Then it holds form(\ref{3012}) and (\ref{3017}) that
\begin{equation}\label{3018}
P(\bar x, \bar v)= \min_{(x, v)\in \mbb{R}^{2n}}\max_{(\varsigma,\sigma)\in S^+_\sharp}\vX(x,v,\varsigma,\sigma)=\min_{(x, v)\in \bar\x_v}P(x,v).
\end{equation}
Because $\x_v\subset\bar\x_v$, it yields
$$\min_{(x, v)\in \bar\x_v}P(x,v)\leq \min_{(x, v)\in \x_v}P(x,v),$$
which with (\ref{3018}) lead to
$$P(\bar x, \bar v)\leq \min_{(x, v)\in \x_v}P(x,v),$$
which with $(\bar{x}, \bar v)\in \x_v$ implies
\begin{equation}\label{3019}
P(\bar{x}, \bar v) =\min_{(x,v)\in\x_v} P(x,v).
\end{equation}
Due to the fact that (\ref{3019}), (\ref{3008}) and (\ref{3009}), we get
$$P(\bar{x}, \bar v) =\min_{(x,v)\in\x_v} P(x,v)=\max_{(\varsigma,\sigma)\in S^+_\sharp} \vP^d(\varsigma,\sigma)=\vP^d(\bar \varsigma,\bar \sigma).$$
\hfill\qed
\end{proof}
Theorem \ref{theorem4} shows that our fourth-order polynomial mixed-integer minimization
problem $(\mathcal{P}_b)$ is canonically dual to the concave maximization
problem as follows:
\begin{equation}\label{3022}
(\mcal{P}^\sharp_+):~~~~\max\lbc\vP^d(\varsigma,\sigma)~|~(\varsigma,\sigma)\in S^+_\sharp\rbc.
\end{equation}
Noted that $\vP^d(\varsigma,\sigma)$ is a continuous concave function over a convex feasible space
$S^+_\sharp$, if $(\bar\varsigma,\bar\sigma)\in S^+_\sharp$ is a critical point of $\vP^d(\varsigma,\sigma)$, then it must be a global maximum point of problem $(\mcal{P}^\sharp_+)$, and $(\bar{x}, \bar v)=\lpa G^{-1}(\bar\varsigma,\bar\sigma_1)c,\frac{1}{2}(f+\bar\sigma_1+\bar\sigma_2)\oslash \bar\sigma_2)\rpa$ should be
a global minimum point of problem $(\mathcal{P}_b)$.
Using ${(\sigma_2)}_i=|f_i+{(\sigma_1)}_i|$ is the solution of $\min_{{(\sigma_2)}_i>0}\frac{1}{{(\sigma_2)}_i}\lpa f_i+{(\sigma_1)}_i+{(\sigma_2)}_i\rpa^2$,
we have
\begin{equation}\label{3028}
\max_{{(\sigma_2)}_i>0}-\frac{1}{4}\frac{1}{{(\sigma_2)}_i}\lpa f_i+{(\sigma_1)}_i+{(\sigma_2)}_i\rpa^2=-(f_i+{(\sigma_1)}_i)^+.
\end{equation}
So for a fixed $(\varsigma,\sigma_1)$, let
\begin{equation}\label{3025}
\vP^g(\varsigma,\sigma_1)=\max_{\sigma_2>0}\vP^d(\varsigma,\sigma)=-\frac{1}{2}c^T G^{-1}c-\sum_{i=1}^n\lpa f_i+{(\sigma_1)}_i\rpa^+
-\frac{1}{2}\varsigma^2-\alpha\varsigma,~~(\varsigma,\sigma_1)\in S^+_{\varsigma\sigma_1},
\end{equation}
where
\begin{equation}\label{3026}
S^+_{\varsigma\sigma_1}=\{(\varsigma,\sigma_1)\in \mathbb{R}^1\times \mathbb{R}^{n}~|~\varsigma\geq -\alpha,\sigma_1\geq 0, G(\varsigma,\sigma_1)\succ 0, f_i+{(\sigma_1)}_i\neq 0, \forall i=1,\cdots,n\}.
\end{equation}
Then we can write the canonical dual problem $(\mcal{P}^\sharp_+)$ to a simple
form:
\begin{equation}\label{3027}
(\mcal{P}^g_+):~~~~\max\lbc\vP^g(\varsigma,\sigma_1)~|~(\varsigma,\sigma_1)\in S^+_{\varsigma\sigma_1}\rbc.
\end{equation}
Moreover, set $\delta(t)^+=\{\delta_i(t_i)^+\}^n\in \mathbb{R}^n$, where
\begin{equation}\label{3030}
\delta_i(t_i)^+=
\left\{\begin{array}{ll}
1 &~~ if~~t_i>0,\\
0 &~~if~~t_i<0,
\end{array}\right. ~~i=1,\cdots,n.
\end{equation}
By Theorem 3, it can be easily to get next theorem about analytic solution to primal problem $(\mathcal{P}_b)$.
\begin{theorem}\label{theorem7}Given
$ A\in\mathbb{R}^{n\times n}$, $ B\succeq 0\in\mathbb{R}^{n\times n}$, $c,f\in \mathbb{R}^n$,
if $(\bar\varsigma,\bar\sigma_1)\in S^+_{\varsigma\sigma_1}$ is a critical point of $\vP^g(\varsigma,\sigma_1)$, then the vector
\begin{equation}\label{3031}
(\bar{x}, \bar v)=\lpa G^{-1}(\bar\varsigma,\bar\sigma_1)c,\delta(f+\bar\sigma_1)^+\rpa
\end{equation}
is a global minimum point of $(\mathcal{P}_b)$.
\end{theorem}
\section{Existence and Uniqueness Criteria}
In this section we study certain
existence and uniqueness conditions for the canonical dual problem to have a
critical point. We first let
\begin{equation}\label{4001}
\bar S^+_\sharp=\{(\varsigma,\sigma)\in \mathbb{R}^1\times \mathbb{R}^{2n}~|~\varsigma\geq -\alpha,\sigma_1\geq 0,\sigma_2\geq0, G\succeq 0\},
\end{equation}
\begin{equation}\label{4001.5}
\partial \bar S^+_\sharp=\{(\varsigma,\sigma)\in \bar S^+_\sharp~|~\det (G)=0\}.
\end{equation}
\begin{equation}\label{4002}
\mathfrak{g}_a=\{(\varsigma,\sigma)\in \mathbb{R}^1\times \mathbb{R}^{2n}~|~\sigma_2\geq0, \det (G)=0\}.
\end{equation}
\begin{lemma}\label{lemma7}If $\mathfrak{g}_a\subset \bar S^+_\sharp$, it holds that
$\partial \bar S^+_\sharp=\mathfrak{g}_a$.
\end{lemma}
\begin{proof}
It is obvious that $\bar S^+_\sharp$ is a closed convex set. From $\mathfrak{g}_a\subset \bar S^+_\sharp$, we have
\begin{equation}\label{4005}
\{(\varsigma,\sigma_1)\in \mathbb{R}^1\times \mathbb{R}^{n}~|~\det (G)=0\}\subset\{(\varsigma,\sigma_1)\in \mathbb{R}^1\times \mathbb{R}^{n}~|\varsigma\geq -\alpha,\sigma_1\geq 0, G\succeq 0\}\nonumber
\end{equation}
which shows
\begin{equation}\label{4006}
\{(\varsigma,\sigma_1)\in \mathbb{R}^1\times \mathbb{R}^{n}~|~G\succeq 0\}\subset\{(\varsigma,\sigma_1)\in \mathbb{R}^1\times \mathbb{R}^{n}~|\varsigma\geq -\alpha,\sigma_1\geq 0\}.\nonumber
\end{equation}
Then it holds
\begin{eqnarray}\label{4005}
\bar S^+_\sharp
=\{(\varsigma,\sigma)\in \mathbb{R}^1\times \mathbb{R}^{2n}~|~\sigma_2\geq0, G\succeq 0\},\nonumber
\end{eqnarray}
thus together with (\ref{4001.5}) we have
\begin{equation}\label{4005}
\partial \bar S^+_\sharp=\{(\varsigma,\sigma)\in \mathbb{R}^1\times \mathbb{R}^{2n}~|~\sigma_2\geq0, \det (G)=0\}=\mathfrak{g}_a.
\end{equation}
\hfill\qed
\end{proof}
Motivated by the existence and uniqueness criteria given in \cite{Gao2010,Gaobox2009b}, we have the following theorem about existence and uniqueness criteria.
\begin{theorem}\label{theorem9} Given $A\in\mathbb{R}^{n\times n}$ and a symmetric positive semi-definite matrix $B\in\mathbb{R}^{n\times n}$, $\alpha>0$, $c,f\in \mathbb{R}^n$, such that $S^+_\sharp\neq\emptyset$ and $\mathfrak{g}_a\subset \bar S^+_\sharp$. If for any given $(\varsigma_0,\sigma_0)\in \mathfrak{g}_a$ and $(\varsigma,\sigma)\in S^+_\sharp$,
\begin{equation}\label{4007}
\lim_{t\rightarrow 0^+}\vP^d(\varsigma_0+t\varsigma,\sigma_0+t\sigma)=-\infty,
\end{equation}
then the canonical dual problem $(\mcal{P}^g_+)$ has at least one critical point $(\bar\varsigma,\bar\sigma_1)\in S^+_{\varsigma\sigma_1}$ and the vector
$(\bar{x}, \bar v)=\lpa G^{-1}(\bar\varsigma,\bar\sigma_1)c,\delta(f+\bar\sigma_1)^+\rpa$
is a global optimizer of the primal problem $(\mathcal{P}_b)$. Furthermore, if $A$, $B$ are two diagonal matrices and
$c_i\neq 0,~\forall i=1,\cdots,n$, then the vector $(\bar{x}, \bar v)$ is a unique global minimum point of $(\mathcal{P}_b)$.
\end{theorem}
\begin{proof}
It holds from $\mathfrak{g}_a\subset \bar S^+_\sharp$ and Lemma \ref{lemma7} that
$\partial \bar S^+_\sharp=\mathfrak{g}_a$. By the definition, $\bar S^+_\sharp$ is a closed convex set and its interior is $S^+_\sharp$. If for any given $(\varsigma,\sigma)\in S^+_\sharp$
\begin{eqnarray}\label{4008}
\lim_{t\rightarrow +\infty}\vP^d(t\varsigma,t\sigma)
&=&\lim_{t\rightarrow +\infty}(-\frac{1}{2}c^T G^{-1}(t\varsigma,t\sigma_1)c-\frac{1}{4}\sum_{i=1}^n\frac{1}{t{(\sigma_2)}_i}\lpa f_i+t{(\sigma_1)}_i+t{(\sigma_2)}_i\rpa^2\nonumber\\
&&-\frac{1}{2}t^2\varsigma^2-\alpha t\varsigma)=-\infty,
\end{eqnarray}
which with (\ref{4007}) implies that the function $\vP^d(\varsigma,\sigma)$ is coercive on the open convex
set $S^+_\sharp$. Therefore, $(\mcal{P}^\sharp_+)$ has at least one critical point $(\bar\varsigma,\bar\sigma)\in S^+_\sharp$ and the vector $(\bar{x}, \bar v)=(G^{-1}(\bar\varsigma,\bar\sigma_1)c,\frac{1}{2}(f+\bar\sigma_1+\bar\sigma_2)\oslash \bar\sigma_2))$ is
a global minimum point of problem $(\mathcal{P}_b)$. So accordingly $(\mcal{P}^g_+)$ has at least one critical point $(\bar\varsigma,\bar\sigma_1)\in S^+_{\varsigma\sigma_1}$ and the vector $(\bar{x}, \bar v)=\lpa G^{-1}(\bar\varsigma,\bar\sigma_1)c,\delta(f+\bar\sigma_1)^+\rpa$
is a global minimum point of $(\mathcal{P}_b)$.
Next we will prove the canonical dual function $\vP^d(\varsigma,\sigma)$ is strict concave on $S^+_\sharp$.
From (\ref{3006}), we have
\begin{equation}\label{4009}
W^T\nabla^2\vP^d(\varsigma,\sigma)W=-r^2-(ZZ_0)^TG^{-1}(ZZ_0)+\sum_{i=1}^n-\frac{1}{2(\sigma_2)_i}\lpa s_i-t_i\frac{(\sigma_1)_i+f_i}{(\sigma_2)_i} \rpa^2.
\end{equation}
(a) If $Z_0=\left(\begin{array}{l}
r\\
s
\end{array}\right)=0$, $W\neq 0$ leads to $t\neq 0$, then with ${(\sigma_2)}_i>0$ and $f_i+{(\sigma_1)}_i\neq 0$, we get $$W^T\nabla^2\vP^d(\varsigma,\sigma)W=\sum_{i=1}^n-\frac{1}{2(\sigma_2)_i}\lpa 0-t_i\frac{(\sigma_1)_i+f_i}{(\sigma_2)_i} \rpa^2=\sum_{i=1}^n-\frac{t_i^2}{2((\sigma_2)_i)^3}\lpa f_i+(\sigma_1)_i \rpa^2<0.$$
(b) If $Z_0=\left(\begin{array}{l}
r\\
s
\end{array}\right)\neq 0$, there are two cases.
(b.1) If $r\neq 0$, then $W^T\nabla^2\vP^d(\varsigma,\sigma)W\leq -r^2<0$.
(b.2) If $r=0$, then $s\neq 0$ and $Z_0=\left(\begin{array}{l}
0\\
s
\end{array}\right)$, thus
\begin{equation}\label{4010}
-(ZZ_0)^TG^{-1}(ZZ_0)=-4s^TDiag(x(\varsigma,\sigma_1))^TG^{-1}Diag(x(\varsigma,\sigma_1))s.
\end{equation}
Since $A$ and $B$ are two diagonal matrices, then from (\ref{tjre001}) we know $G$ is a diagonal matrix, which with $G\succ 0$ shows that $G^{-1}$ is a diagonal matrix whose all diagonal elements are positive. Then it follows from $x(\varsigma,\sigma_1) = G^{-1}c$ and $c_i\neq 0,~\forall i=1,\cdots,n$ that
$$x(\varsigma,\sigma_1)_i\neq 0,~\forall i=1,\cdots,n.$$
By $G^{-1}\succ 0$, we find $Diag(x(\varsigma,\sigma_1))^TG^{-1}Diag(x(\varsigma,\sigma_1))\succ 0$, then it is observed from (\ref{4010}) that
$$-(ZZ_0)^TG^{-1}(ZZ_0)<0,$$
hence
$$W^T\nabla^2\vP^d(\varsigma,\sigma)W\leq -(ZZ_0)^TG^{-1}(ZZ_0)<0.$$
From above two cases, we can see $\vP^d(\varsigma,\sigma)$ is strictly concave on $S^+_\sharp$.
Thereby $(\mcal{P}^\sharp_+)$ has a unique critical point in $S^+_\sharp$, which implies $(\mcal{P}^g_+)$ has
a unique critical point in $S^+_{\varsigma\sigma_1}$ and the primal problem has a unique global
minimum.
\hfill\qed
\end{proof}
\section{Application to Decoupled Problem}
In this section, we discuss the application of presented theory to the decoupled problems. Consider the decoupled minimization problem as follows:
\begin{equation}\label{5001}
\min \left\{P(x,v)= \frac{1}{2}\Big(\frac{1}{2}\sum_{i=1}^nb_ix_i^2-\alpha \Big)^2+\sum_{i=1}^n\lpa\frac{1}{2}a_ix_i^2-c_ix_i-f_iv_i\rpa\right\}
\end{equation}
\begin{equation}\label{5002}
s.t.~~-v_i\leq x_i\leq v_i,v_i\in\{0,1\}, i=1,\ldots,n.
\end{equation}
For simplicity, we may take $A = Diag (a)$
and $B = Diag (b)$ stand for two diagonal matrices with diagonal elements $a = \{a_i\} \in \mathbb{R}^{n}$ and $b = \{b_i\} \in \mathbb{R}^{n}$
respectively. The canonical dual function of above problem has the following simple form
\begin{equation}\label{5003}
\vP^d(\varsigma,\sigma)=-\frac{1}{2}\sum_{i=1}^n\lpa\frac{c_i^2}{a_i+\varsigma b_i+2{(\sigma_1)}_i}+\frac{\lpa f_i+{(\sigma_1)}_i+{(\sigma_2)}_i\rpa^2}{2{(\sigma_2)}_i}\rpa-\frac{1}{2}\varsigma^2-\alpha\varsigma.
\end{equation}
Due to $\nabla\vP^d(\varsigma,\sigma)=0$, then when ${(\sigma_2)}_i\neq0$ and $c_i\neq 0$, we can obtain the critical points of $\vP^d(\varsigma,\sigma)$ in the following
\begin{equation}\label{50041}
\varsigma=\frac{1}{2}\sum_{i=1}^nb_i-\alpha,~~{(\sigma_1)}_i\in M_i=\left\{-\frac{1}{2}\Big( a_i+\Big(\frac{1}{2}\sum_{i=1}^nb_i-\alpha\Big)b_i\pm c_i\Big)\right\},
\end{equation}
\begin{equation}\label{50042}
{(\sigma_2)}_i\in N_i=\left\{f_i-\frac{1}{2}\Big( a_i+\Big(\frac{1}{2}\sum_{i=1}^nb_i-\alpha\Big)b_i\pm c_i\Big)\right\},~~\forall i=1,\ldots,n.
\end{equation}
For ${(\sigma_2)}_i>0$, using Theorem \ref{th:Complementary-Dual}, we can derive the accompanying primal solution
\begin{equation}\label{5005}
(x_i,v_i)=\lpa-\frac{c_i}{a_i+(\frac{1}{2}\sum_{i=1}^nb_i-\alpha) b_i+2{(\sigma_1)}_i},\frac{ f_i+{(\sigma_1)}_i+{(\sigma_2)}_i}{2{(\sigma_2)}_i}\rpa,~~\forall i=1,\ldots,n.
\end{equation}
According to the fact that there are two possible solutions for each component of $\sigma=(\sigma_1,\sigma_2)\in \mathbb{R}^{2n}$,
together with (\ref{50041}) and (\ref{50042}), it follows that the canonical dual function $\vP^d$ has $2^n$ critical points.
Then from Theorem \ref{theorem4}, it is easy to show the global minimum of the primal problem will be arrived at by next theorem.
\begin{theorem}\label{theorem10}
Given $A=Diag(a)\in\mathbb{R}^{n\times n}$ and a symmetric positive semi-definite matrix $B=Diag(b)\in\mathbb{R}^{n\times n}$, $\alpha>0$, $c,f\in \mathbb{R}^n$,
$c_i\neq 0$ for $\forall i$, if
\begin{equation}
\max M_i>0~~and~~\max N_i>0,~~\forall i=1,\ldots,n.
\end{equation}
then $\vP^d(\varsigma,\sigma)$ has a unique critical point
\begin{eqnarray}
(\bar \varsigma,\bar \sigma)=(\bar \varsigma,\bar \sigma_1,\bar \sigma_2)
=\Bigg(\frac{1}{2}\sum_{i=1}^nb_i-\alpha,\big\{\max M_i,i=1,\ldots,n\big\},\big\{\max N_i,i=1,\ldots,n\big\}\Bigg)\in S^+_\sharp,\nonumber
\end{eqnarray}
which is a global maximizer of $\vP^d(\varsigma,\sigma)$ on $S^+_\sharp$, and
$$(\bar x,\bar v)=\lpa\left\{\frac{c_i}{|c_i|} \right\}, e \rpa$$
is a global minimum of $P(x, v)$ on $\x_v$.
\end{theorem}
\section{Examples}
Now we give a summary of numerical experiments to illustrate our method, where diagonal matrices $A$ and $B$, vectors $f$ and $c$ are chosen and taken at random. These examples are classified into three cases and in every case we give several representatives. In the first case we consider the decoupled problems satisfying the conditions of Theorem \ref{theorem10}, whose results are consistent with Theorem \ref{theorem10} and show our method is promising for decoupled problems. In the second case some general decoupled problems not satisfying Theorem \ref{theorem10} are computed by our method and the global solution are also obtained. In the last case, our method is tested for some general problems. All of performed examples show our method is very effective.
\subsection{Case 1: Decoupled Problems satisfying Theorem \ref{theorem10}}
In the following three decoupled examples, we can verify the conditions of Theorem \ref{theorem10} are satisfied, so
a unique critical point $(\bar\varsigma,\bar\sigma)$ of $\vP^d(\varsigma,\sigma)$ on $S^+_\sharp$ and a global minimum point $(\bar x,\bar v)$ of $P(x, v)$ on $\x_v$ are obtained by Theorem \ref{theorem10}. For simplicity we denote $\lambda_{min}(\bar\varsigma,\bar\sigma_1)$ be the smallest eigenvalue of $G$.
\newtheorem{example}{{Example}}
\begin{example}
Set $\alpha=10$, $A=Diag (1,-1,1,5,2)$, $B=Diag (2,4,1,4,2)$, $f=(20,12,-1,$ $1,13)$ and $c=(-8,-9,10,9,-5)$, in which $n=5$.
\end{example}
\begin{example}
Set $\alpha=20$, $A=Diag (7,9,6,-5,4,10,9,8)$, $B=Diag (3,5,4,3,1,7,5,7)$, $f=(13,-3,3,11,10,16,16,14)$ and $c=(7,-6,8,-1,-5,8,-8,7)$, in which $n=8$.
\end{example}
\begin{example}
Set $\alpha=25$, $A=Diag (2,8,7,3,6,14,10,1,-6,9)$, $B=Diag (9,1,2,1,6,8,5,$ $3,9,6)$, $f=(6,1,4,13,6,15,17,20,3,16)$ and $c=(19,14,-9,-9,-8,17,-22,-14,-8,18)$, in which $n=10$.
\end{example}
The values of $\frac{1}{2}\sum_{i=1}^nb_i-\alpha$, $M_i$ and $N_i$ from (\ref{50041}) and (\ref{50042}) are first needed to be computed and are listed
in Table 1 as follows. Then it can be found that conditions of Theorem \ref{theorem10} are all satisfied.
\begin{table}[H]
\centering
\scalebox{0.8}[0.8]
\begin{tabular}{|c|c|l|l|l|l|}
\hline
Experiments &$\frac{1}{2}\sum_{i=1}^nb_i-\alpha$ &$M_i$ & $Max~M_i$ & $N_i$& $Max~N_i$\\
\hline
\multirow{5}{*}{Example 1}&\multirow{5}{*}{-3.5} &$M_1=\{7,-1\}$ & 7 & $N_1=\{27,19\}$ & 27\\
&& $M_2=\{12,3\}$ & 12 & $N_2=\{24,15\}$& 24\\
& &$M_3=\{-3.75,6.25\}$ & 6.25 &$N_3=\{-4.75,5.25\}$ & 5.25\\
& &$M_4=\{0,9\}$ & 9 &$N_4=\{1,10\}$ & 10\\
& &$M_5=\{5,0\}$ & 5 & $N_5=\{18,13\}$& 18\\
\hline
\multirow{8}{*}{Example 2}&\multirow{8}{*}{-2.5} &$M_1=\{-3.25,3.75\}$ & 3.75& $N_1=\{9.75,16.75\}$ & 16.75 \\
&& $M_2=\{4.75,-1.25\}$ & 4.75 & $N_2=\{1.75,-4.25\}$& 1.75\\
& &$M_3=\{-2,6\}$ & 6 &$N_3=\{1,9\}$ & 9\\
& &$M_4=\{6.75,5.75\}$ & 6.75 &$N_4=\{17.75,16.75\}$ & 17.75\\
& &$M_5=\{1.75,-3.25\}$ & 1.75 & $N_5=\{11.75,6.75\}$& 11.75\\
& &$M_6=\{-0.25,7.75\}$ & 7.75 &$N_6=\{15.75,23.75\}$ & 23.75\\
& &$M_7=\{5.75,-2.25\}$ & 5.75 &$N_7=\{21.75,13.75\}$ & 21.75\\
& &$M_8=\{1.25,8.25\}$ & 8.25 & $N_8=\{15.25,22.25\}$& 22.25\\
\hline
\multirow{10}{*}{Example 3}&\multirow{10}{*}{0} &$M_1=\{-10.5,8.5\}$ & 8.5 &$N_1=\{-4.5,14.5\}$ & 14.5\\
&& $M_2=\{-11,3\}$ & 3 & $N_2=\{-10,4\}$& 4\\
& &$M_3=\{1,-8\}$ & 1 &$N_3=\{5,-4\}$ & 5\\
& &$M_4=\{3,-6\}$ & 3 &$N_4=\{16,7\}$ & 16\\
& &$M_5=\{1,-7\}$ & 1 & $N_5=\{7,-1\}$& 7\\
& &$M_6=\{-15.5,1.5\}$ & 1.5 &$N_6=\{-0.5,16.5\}$ & 16.5\\
& &$M_7=\{6,-16\}$ & 6 &$N_7=\{23,1\}$ & 23\\
& &$M_8=\{6.5,-7.5\}$ & 6.5 & $N_8=\{26.5,12.5\}$& 26.5\\
& &$M_9=\{7,-1\}$ & 7 &$N_9=\{10,2\}$ & 10\\
& &$M_{10}=\{-13.5,4.5\}$ & 4.5 & $N_{10}=\{2.5,20.5\}$& 20.5\\
\hline
\end{tabular}}
\caption{Conditions of Theorem \ref{theorem10} are satisfied in Examples 1-3.}
\end{table}
On the one hand, by Theorem \ref{theorem10} we know $\varsigma=\frac{1}{2}\sum_{i=1}^nb_i-\alpha$, $(\bar \sigma_1)_i=\max M_i$, $(\bar \sigma_2)_i=\max N_i$ and $(\bar x,\bar v)=\lpa\left\{\frac{c_i}{|c_i|} \right\}, e \rpa$,
so from Table 1 we can easily get the corresponding $(\bar\varsigma,\bar\sigma)$ and $(\bar x,\bar v)$ for Examples 1-3 listed in Table 2.
\begin{table}[!hbp]
\centering
\scalebox{0.78}[0.78]
\begin{tabular}{|c|l|l|l|}
\hline
Experiments &$\bar\varsigma$ &$\bar\sigma$ & Primal solution $(\bar x, \bar v)$ \\
\hline
\multirow{2}{*}{Example 1}&$\bar\varsigma=-3.5$ &$\bar\sigma_1=(7,12,6.25,9,5)$, &$\bar x=(-1,-1,1,1,-1)$, \\
& & $\bar\sigma_2=(27,24,5.25,10,18)$.& $\bar v=(1,1,1,1,1)$. \\
\hline
\multirow{3}{*}{Example 2} & $\bar\varsigma=-2.5$ &$\bar\sigma_1=(3.75,4.75,6,6.75,1.75,7.75,5.75,8.25)$, &$\bar x=(1,-1,1,-1,-1,1,-1,1)$\\
& & $\bar\sigma_2=(16.75,1.75,9,17.75,11.75,23.75,21.75,22.25)$.&$\bar v=(1,1,1,1,1,1,1,1)$. \\
\hline
\multirow{3}{*}{Example 3} &$\bar\varsigma=0$ &$\bar\sigma_1=(8.5,3,1,3,1,1.5,6,6.5,7,4.5)$, &$\bar x=(1,1,-1,-1,-1,1,-1,-1,-1,1)$, \\
& & $\bar\sigma_2=(14.5,4,5,16,7,16.5,23,26.5,10,20.5)$.& $\bar v=(1,1,1,1,1,1,1,1,1,1)$. \\
\hline
\end{tabular}}
\caption{Results of Theorem \ref{theorem10} in Examples 1-3.}
\end{table}
On the other hand, we can also get
$(\bar\varsigma,\bar\sigma)$ by solving the canonical dual problem $(\mcal{P}^\sharp_+)$, then $(\bar x,\bar v)$ can be computed by $(\bar{x}, \bar v)=\lpa G^{-1}(\bar\varsigma,\bar\sigma_1)c,\frac{1}{2}(f+\bar\sigma_1+\bar\sigma_2)\oslash \bar\sigma_2)\rpa$. The corresponding results are listed below:
\begin{table}[!hbp]
\centering
\scalebox{0.7}[0.7]
\begin{tabular}{|c|l|c|c|l|c|}
\hline
Experiments &Dual solution $(\bar\varsigma,\bar\sigma)$ &$\vP^d(\bar\varsigma,\bar\sigma)$ &$\lambda_{min}(\bar\varsigma,\bar\sigma_1)$ & Primal solution $(\bar x, \bar v)$& $P(\bar x, \bar v)$ \\
\hline
\multirow{3}{*}{Example 1}&$\bar\varsigma=-3.5$, &\multirow{3}{*}{-75.875} & \multirow{3}{*}{5}&$\bar x=(-1,-1,1,1,-1)$,& \multirow{3}{*}{-75.875} \\
& $\bar\sigma_1=(7,12,6.25,9,5)$,& & & $\bar v=(1,1,1,1,1)$. & \\
& $\bar\sigma_2=(27,24,5.25,10,18)$.& & & & \\
\hline
\multirow{3}{*}{Example 2} & $\bar\varsigma=-2.5$, &\multirow{3}{*}{-102.875} & \multirow{3}{*}{1}&$\bar x=(1,-1,1,-1,-1,1,-1,$& \multirow{3}{*}{-102.875} \\
& $\bar\sigma_1=(3.75,4.75,6,6.75,1.75,7.75,5.75,8.25)$, & && $1)$, & \\
& $\bar\sigma_2=(16.75,1.75,9,17.75,11.75,23.75,21.75,22.25)$. & &&$\bar v=(1,1,1,1,1,1,1,1)$. & \\
\hline
\multirow{3}{*}{Example 3} &$\bar\varsigma=0$, &\multirow{3}{*}{-212}&\multirow{3}{*}{8} &$\bar x=(1,1,-1,-1,-1,1,-1,$ & \multirow{3}{*}{-212} \\
& $\bar\sigma_1=(8.5,3,1,3,1,1.5,6,6.5,7,4.5)$, & & &$-1,-1,1)$, & \\
& $\bar\sigma_2=(14.5,4,5,16,7,16.5,23,26.5,10,20.5)$. & & &$\bar v=(1,1,1,1,1,1,1,1,1,1)$. & \\
\hline
\end{tabular}}
\caption{Results by solving problem $(\mcal{P}^\sharp_+)$ in Examples 1-3.}
\end{table}
From Tables 2 and 3, it can be seen that the results from Theorem \ref{theorem10} are consistent with the ones by our canonical dual method. And
the fact that $\bar\varsigma\geq -\alpha,\bar\sigma_1\geq 0,\bar\sigma_2\geq0, G(\bar\varsigma,\bar\sigma_1)\succeq 0$ in every example indicates
$(\bar \varsigma,\bar \sigma)\in S^+_\sharp$. By Theorem \ref{theorem10}, the solution of primal problem is obtained. It is verified that our method is promising for decoupled problems when the conditions of Theorem \ref{theorem10} are satisfied for the decoupled problems.
\subsection{Part 2: Decoupled Problems where Theorem \ref{theorem10} are not satisfied}
It needs to say that there are many decoupled problems not to satisfy Theorem \ref{theorem10}. Here we choose three of them to indicate the details. Let $n=5$ and $\alpha=10$, diagonal matrices $A$ and $B$, vectors $f$ and $c$ are chosen and taken at random.
\begin{example}
Set $A=Diag (6,3,9,9,2)$, $B=Diag (2,4,5,4,3)$, $f=(5,4,4,20,9)$ and $c=(1,-9,-6,3,-5)$.
\end{example}
\begin{example}
Set $A=Diag (1,-1,1,4,4)$, $B=Diag (1,1,1,4,5)$, $f=(1,-51,-1,-11,-61)$ and $c=(3,0,1,-2,0)$.
\end{example}
\begin{example}
Set $A=Diag (5,-1,2,5,1)$, $B=Diag (5,2,2,1,4)$, $f=(3,-35,-1,11,15)$ and $c=(7,0,4,-6,10)$.
\end{example}
\begin{table}[H]
\centering
\scalebox{0.8}[0.8]
\begin{tabular}{|c|c|l|l|l|l|}
\hline
Experiments &$\frac{1}{2}\sum_{i=1}^nb_i-\alpha$ &$M_i$ & $Max~M_i$ & $N_i$& $Max~N_i$\\
\hline
\multirow{5}{*}{Example 4}&\multirow{5}{*}{-1} &$M_1=\{-2.5,-1.5\}$ & -1.5 & $N_1=\{2.5,3.5\}$ & 3.5\\
&& $M_2=\{5,-4\}$ & 5 & $N_2=\{9,0\}$& 9\\
& &$M_3=\{1,-5\}$ & 1 &$N_3=\{5,-1\}$ & 5\\
& &$M_4=\{-4,-1\}$ & -1 &$N_4=\{16,19\}$ & 19\\
& &$M_5=\{3,-2\}$ & 3 & $N_5=\{12,7\}$& 12\\
\hline
\end{tabular}}
\caption{Conditions of Theorem \ref{theorem10} are not all satisfied in Examples 4.}
\end{table}
From Table 4, we find that $Max~M_1=-1.5<0$ which makes the conditions of Theorem \ref{theorem10} are not satisfied. We solve the simple form problem $(\mcal{P}^g_+)$ instead of the canonical dual problem $(\mcal{P}^\sharp_+)$ and use
Theorem \ref{theorem7} to obtain the analytic solution $(x, v)$ to primal problem $(\mathcal{P}_b)$. The corresponding results are listed below:
\begin{table}[!hbp]
\centering
\scalebox{0.8}[0.8]
\begin{tabular}{|c|l|c|c|l|c|}
\hline
Experiments &Dual solution $(\bar\varsigma,\bar\sigma_1)$ &$\vP^g(\bar\varsigma,\bar\sigma_1)$ &$\lambda_{min}(\bar\varsigma,\bar\sigma_1)$ & Primal solution $(\bar x, \bar v)$& $P(\bar x, \bar v)$ \\
\hline
\multirow{2}{*}{Example 4}&$\bar\varsigma=-1.82$, &\multirow{2}{*}{-51.7281} & \multirow{2}{*}{2.3593}& $\bar x=(0.424,-1,-1,1,-1)$,& \multirow{2}{*}{-51.7281} \\
& $\bar\sigma_1=(0,6.641,3.051,0.641,4.231)$.& & & $\bar v=(1,1,1,1,1)$. & \\
\hline
\end{tabular}}
\caption{Results by solving problem $(\mcal{P}^g_+)$ in Examples 4.}
\end{table}
It is obvious that there exists some $c_i=0$ in Examples 5 and 6, which does not satisfy the conditions of Theorem \ref{theorem10}. We also solve them by the simple form problem $(\mcal{P}^g_+)$, whose results are illustrated in Table 6.
\begin{table}[!hbp]
\centering
\scalebox{0.85}[0.85]
\begin{tabular}{|c|l|c|c|l|c|}
\hline
Experiments &Dual solution $(\bar\varsigma,\bar\sigma_1)$ &$\vP^g(\bar\varsigma,\bar\sigma_1)$ &$\lambda_{min}(\bar\varsigma,\bar\sigma_1)$& Primal solution $(\bar x, \bar v)$& $P(\bar x, \bar v)$ \\
\hline
\multirow{2}{*}{Example 5}&$\bar\varsigma=-7$, &\multirow{2}{*}{32.5} &\multirow{2}{*}{1} &$\bar x=(1,0,1,-1,0)$,& \multirow{2}{*}{32.5} \\
& $\bar\sigma_1=(4.5,34.987,3.5,13,54.367)$.& & & $\bar v=(1,0,1,1,0)$. & \\
\hline
\multirow{2}{*}{Example 6} & $\bar\varsigma=-4$, &\multirow{2}{*}{-40.5 } & \multirow{2}{*}{2.22424 }&$\bar x=(1,0,1,-1,1)$,& \multirow{2}{*}{-40.5 } \\
& $\bar\sigma_1=(11,5.612,5,2.5,12.5)$. & && $\bar v=(1,0,1,1,1)$. & \\
\hline
\end{tabular}}
\caption{Results by solving problem $(\mcal{P}^g_+)$ in Examples 5 and 6.}
\end{table}
From Tables 5-6, we have $\bar\varsigma\geq -\alpha,\bar\sigma_1\geq 0, G(\bar\varsigma,\bar\sigma_1)\succ 0, f_i+{(\bar\sigma_1)}_i\neq 0, \forall i=1,\cdots,n$ in every example, so
$(\bar \varsigma,\bar \sigma_1)\in S^+_{\varsigma\sigma_1}$. By Theorem \ref{theorem7}, the solution of primal problem is obtained. Thus when the conditions of Theorem \ref{theorem10} are not all satisfied, our method is also effective for the decoupled problem.
\subsection{Part 3: General Nonconvex Problems}
For general nonconvex problems in this part, we solve the simple form problem $(\mcal{P}^g_+)$ and use
Theorem \ref{theorem7} to obtain the analytic solution $(x, v)$ to primal problem $(\mathcal{P}_b)$. Two general nonconvex examples not decoupled are tested and the corresponding results are listed below:
\begin{example}
Set $$
A=\begin{bmatrix} 4 & 0& 1 \\
0 & -4& -6 \\
1 & -6& 4 \end{bmatrix}~~\wand~~
B=\begin{bmatrix} 7 & -3& -4\\
-3 & 8& 2\\
-4 & 2& 10 \end{bmatrix},
$$ $\alpha=8$, $f=(3,2,3)$ and $c=(10,6,7)$.
\end{example}
\begin{example} Set $$
A=\begin{bmatrix} 15 & 3& -3& -2& -4\\
3 & 21& -5& 0& 2\\
-3 & -5& 12& 0& 2\\
-2 & 0& 0& 14& 3\\
-4 & -2& 2& 3& 6 \end{bmatrix}~~\wand~~
B=\begin{bmatrix} 13 & 2& -4& 4& -6\\
2 & 6& -4& 1& -2\\
-4 & -4& 6& 0& -3\\
4 & 1& 0& 7& -7\\
-6 & -2& -3& -7& 21 \end{bmatrix},
$$ $\alpha=4$, $f=(6,-2,5,4,10)$ and $c=(7,-3,10,-4,-3)$.
\end{example}
The corresponding results are listed below:
\begin{table}[H]
\centering
\scalebox{0.8}[0.8]
\begin{tabular}{|c|l|c|c|l|c|}
\hline
Experiments &Dual solution $(\bar\varsigma,\bar\sigma_1)$ &$\vP^g(\bar\varsigma,\bar\sigma_1)$ &$\lambda_{min}(\bar\varsigma,\bar\sigma_1)$ & Primal solution $(\bar x, \bar v)$& $P(\bar x, \bar v)$ \\
\hline
\multirow{2}{*}{Example 7}&$\bar\varsigma=-0.5$, &\multirow{2}{*}{-33.875} & \multirow{2}{*}{1.58694}& $\bar x=(1,1,1)$,& \multirow{2}{*}{-33.875} \\
& $\bar\sigma_1=(2.5,9.75,6)$.& & & $\bar v=(1,1,1)$. & \\
\hline
\multirow{2}{*}{Example 8}&$\bar\varsigma=0.088$, &\multirow{2}{*}{-32.8777} & \multirow{2}{*}{5.54327}& $\bar x=(0.556,0,0.978,-0.174,-0.225)$,& \multirow{2}{*}{-32.8777} \\
& $\bar\sigma_1=(0,1.994,0,0,0)$.& & & $\bar v=(1,0,1,1,1)$. & \\
\hline
\end{tabular}}
\caption{Results by solving problem $(\mcal{P}^g_+)$ in Examples 7 and 8.}
\end{table}
From Table 7, it holds that $\bar\varsigma\geq -\alpha,\bar\sigma_1\geq 0, G(\bar\varsigma,\bar\sigma_1)\succ 0, f_i+{(\bar\sigma_1)}_i\neq 0, \forall i=1,\cdots,n$ in every example, so
$(\bar \varsigma,\bar \sigma_1)\in S^+_{\varsigma\sigma_1}$. By Theorem \ref{theorem7}, the solution of primal problem is obtained. So our method is also effective for the general problems.
\section{Conclusions and further work}\label{se:concl}
In this paper we propose a canonical duality method for solving a mixed-integer nonconvex fourth-order polynomial minimization problem
with fixed cost terms. By rewriting the box constraints in a relaxed quadratic form, a relaxed reformulation of the primal
problem $(\mathcal{P}_b)$ is obtained, then the canonical dual problem $(\mcal{P}^\sharp)$ is defined and the complementary-dual principle is proved.
The primal problem $(\mathcal{P}_b)$ is canonically dual to a concave maximization
problem $(\mcal{P}^\sharp_+)$ where a useful space $S^+_\sharp$ is introduced. This constrained nonconvex problem $(\mathcal{P}_b)$ in $\mathbb{R}^{2n}$ can be transformed into a continuous concave maximization dual problem $(\mcal{P}^g_+)$ in $\mathbb{R}^{n+1}$ without duality gap. The global optimality conditions are proposed and the existence and uniqueness criteria are illustrated. Application to the decoupled mixed-integer
problem is illustrated and analytic solution for a global minimum is obtained under some suitable conditions. Several examples are given to show our method is effective. Canonical duality theory is a potentially powerful methodology, which can be used to model complex
systems to a wide class of discrete and continuous problems in global optimization
and nonconvex analysis. The ideas and results with canonical duality theory presented in this paper can be used or generalized for solving other type of problems in the future. \\
| 86cae5348e969d7a40d917e3e5db0651091023a8 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\subsection{LSTM Equations}
\input{lstm_equations}
\subsection{Parameter reduction}
\label{subsec:am_parameter_reduction}
In order to reduce memory consumption and computation, we compress the previously described acoustic model using projection layers that sit between the outputs of an LSTM layer and both the recurrent and non-recurrent inputs to same and subsequent layers~\cite{SakSeniorBeaufays14}. In our system, we introduce projection matrices of rank 100 for the first four layers, and a projection matrix of rank 200 for the fifth hidden layer, as we have noticed the weight matrices' ranks increases with depth.
Of crucial importance, however, is that when a significant rank reduction is applied, it is not sufficient to initialize the projection layer's weight matrix randomly for training with the CTC criterion. In section~\ref{sec:training} we explore a number of options for better, faster and more stable training convergence for our model. They range from adding a multiplicative factor that along with the global learning rate scales the gradient of the projection layers' weights during backpropagation, to exploying a clever initialization proposed in~\cite{prabhavalkar2016} that uses the larger `uncompressed' model without the projection layer and jointly factorize its recurrent and (non-recurrent) inter-layer weight matrices at each hidden layer using a form of singular value decomposition to determine a shared projection layer.
\subsection{Quantization basics}
Given floating point values, our goal is to represent them as 8-bit integers.
Generally speaking we could use either a uniform or a non-uniform quantizer, or
even an optimal quantizer for a given value distribution~\cite{bovik2005}.
However, with simplicity and performance in mind, and validated by previous work
in the area~\cite{vanhoucke2011}, we settled on a uniform linear quantizer that
assumes a uniform distribution of the values within a given range. As a result
we do not need a decompression step at inference time.
\textbf{Quantizing.}
Given a set of float values $\mathbf{V}=\{V_x\}$, and a desired scale $S$
(e.g. we use 255 for 8-bits), we compute a factor $Q$ that produces a
scaled version of the original: $\mathbf{V}'=\{0 \le V_{x}' \le S\}$.
To maximize the use of $S$ we determine the range of values
$R = (V_\text{max} - V_\text{min})$ we want to quantize, i.e., squeeze into the
new scale $S$. Thus, the quantization factor can be expressed as $Q = \frac{S}{R}$,
and the quantized values as $V_{x}' = Q * (V_{x} - V_\text{min})$.
\textbf{Recovery.}
A quantized value $V_{x}'$ can be recovered (i.e. transformed back into its
approximate high-precision value) by performing the inverse of the quantization
operation, which implies computing a recovery factor $Q^{-1} = \frac{R}{S}$.
The recovered value is then expressed as $V_{x} = V_{x}' * Q^{-1} + V_\text{min}$.
\textbf{Quantization error and bias.}
Quantization is a lossy process, with two sources of error. The first is the difference
between the input value and its quantized-then-recovered value (precision loss).
The second is the result of discrepancies in quantization-recovery operations
that introduce a bias in the computed value (bias error)~\cite{Tan08}. Of the two sources
of error, the precision loss is theoretically and practically unavoidable but, on average,
has a smaller impact on what the original data represents (e.g. the difference in
the variances of $\mathbf{V}$ and $\mathbf{V}'$ is very small~\cite{gersho1992}). The bias, however,
is theoretically avoidable by paying close attention to how the quantized values are
manipulated, thus avoiding the introduction of inconsistencies. The latter is very important
since bias problems have a big impact on the quantization error. Consequently we pay particular
attention in eliminating bias error in Section~\ref{subsec:quantized_inference}.
\subsection{Quantized inference}
\label{subsec:quantized_inference}
The approach we follow during neural network inference is to treat each layer independently,
receiving and producing floating point values: inputs get quantized on-the-fly, while network parameters
offline. Internally, layers operate on 8-bit integers for
the matrix multiplications (typically the most computationally intensive operations), and their
product is recovered to floating point, as depicted in Figure \ref{fig:quantization_scheme_diagram}.
This simplifies the implementation of complex activation functions, and allows mixing integer layers
with float layers, if desired.
\begin{figure}
\centering
\includegraphics[width=\columnwidth]{quantization_scheme}
\footnotesize
\caption{\footnotesize Execution of typical inference $y = W X + B$ : weights $\mathbf{W}$ are
already quantized, and inputs $\mathbf{X}$ are quantized $Q(\cdot)$ on-the-fly before
performing multiplication $Mult(\cdot)$; the product is then recovered $R(\cdot)$ to
apply the biases $\mathbf{B}$ and the activation function $F(\cdot)$.}
\label{fig:quantization_scheme_diagram}
\end{figure}
\textbf{Multiplication of quantized values.}
In order to perform most of the multiplication,
$V_\text{c} = V_\text{a} * V_\text{b}$, in the 8-bit domain (though using 32-bit accumulators),
we must first apply the offset, $V_\text{min}$, such that
$V_\text{x} = \frac{V_\text{x}''}{Q}$, where $V_\text{x}''=V_\text{x}' + Q V_\text{min}$.
We then apply the recovery factor on the result, which for the multiplication of
two independently quantized values $V_\text{a}'$ and $V_\text{b}'$, is the inverse
product of their quantization factors $Q_\text{a}$ and $Q_\text{b}$ :
\begin{equation}
\label{eq:mult}
\footnotesize
V_\text{c} = \frac{V_\text{a}'' * V_\text{b}''}{Q_\text{a} * Q_\text{b}}\quad\quad\quad\quad\quad\text{(} Mult(\cdot) \text{ and } R(\cdot) \text{ in Figure 1)}
\end{equation}
\textbf{Integer multiplication: effects on quantization and recovery.}
In eq.~\eqref{eq:mult}, each factor $V_\text{x}''$ is of integer type. This
means in the $V_\text{x}''$ formulation: 1) $V_\text{x}'$ is already an integer;
2) $Q V_\text{min}$ is a float that will be rounded to an integer, and thus
introduce an error: $E = \text{float}(Q V_\text{min}) - \text{integer}(Q V_\text{min})$.
This requires that quantization be performed in a way that is consistent
with this formulation in order to avoid introducing bias error. Thus we
introduce a rounding operation $\text{round}(\cdot)$:
\begin{equation}
\label{eq:quant}
\footnotesize
V_\text{x}' = \text{round}(Q V_\text{x}) - \text{round}(Q V_\text{min})\quad\quad\text{(} Q(\cdot) \text{ in Figure 1)}
\end{equation}
Thus precision errors in the quantization and multiplication are
consistent and cancel each other. This also means that recovery needs to
be consistent with eq.~\eqref{eq:quant}:
\begin{equation}
\footnotesize
V_\text{x} = \frac{V_\text{x}' + \text{round}(Q V_\text{min})}{Q}\quad\quad\quad\quad\quad\quad\text{(} R(\cdot) \text{ in Figure 1)}
\end{equation}
\textbf{Efficient implementation.} The proposed quantization scheme
benefits from the reduced memory bandwidth of accessing 8-bit values, and enables
squeezing more values into any fast cache available, thus reducing power consumption and
access time. Furthermore, it allows better use of optimized SIMD instructions
by fitting in more values per operation, which offers a performance advantage over their
floating point counterparts. The overhead of the quantization and recovery
operations is typically negligible, and also parallelizable via SIMD. We do not cover any
specific implementation since whereas the previous benefits are generally applicable, the
details hinge on the targeted hardware, and that is beyond the scope of this paper. However,
in our previous work ~\cite{mcgraw16} we recorded a significant speed up over unquantized
floating point inference.
Logically, our scheme can be applied at a given level of granularity, subdividing groups of
values into sub-groups for better precision. This means parameter matrices
at different NN layers can be quantized independently, or
even further broken down into individually quantized sub-matrices.
We set the granularity at the level of the weight
matrices (e.g. the parameters associated with individual gates in an LSTM).
This results in a relatively small loss in final inference accuracy
(see Table~\ref{tbl:results}).
\subsection{Quantization aware training}
\label{subsec:quantization-aware_training}
\input{training}
\begin{table*}
\footnotesize
\centering
\begin{tabular}{|c|c|c|c|c||c|c|c|c|}
\hline
\textbf{System (Params.)} & \multicolumn{4}{|c||}{\textbf{WER (\%) on Clean Eval Set}} & \multicolumn{4}{|c|}{\textbf{WER (\%) on Noisy Eval Set}} \\
\hline
& \textbf{match} & \textbf{mismatch} & \textbf{quant} & \textbf{quant-all} & \textbf{match} & \textbf{mismatch} & \textbf{quant} & \textbf{quant-all} \\
\hline
\hline
$4\times300$ ($\sim$2.9M) & 13.6 & 14.3 (5.1\%) & 13.5 (-0.7\%) & 13.6 (0.0\%) & 26.3 & 28.2 (7.2\%) & 26.5 (0.8\%) & 26.5 (0.8\%) \\
\hline
$5\times300$ ($\sim$3.7M) & 12.5 & 13.1 (4.8\%) & 12.6 (0.8\%) & 12.7 (1.6\%) & 24.6 & 26.6 (8.1\%) & 24.8 (0.8\%) & 25.0 (1.6\%) \\
\hline
$4\times400$ ($\sim$5.0M) & 12.1 & 12.5 (3.3\%)& 12.3 (1.7\%) & 12.3 (1.7\%) & 23.2 & 25.0 (7.8\%) & 23.7 (2.2\%) & 23.8 (2.6\%) \\
\hline
$5\times400$ ($\sim$6.3M) & 11.4 & 11.7 (2.6\%) & 11.5 (0.9\%) & 11.7 (2.6\%) & 22.3 & 23.5 (5.4\%) & 22.6 (1.3\%) & 22.7 (1.8\%) \\
\hline
$4\times500$ ($\sim$7.7M) & 11.7 & 12.0 (2.6\%) & 11.7 (0.0\%) & 11.7 (0.0\%) & 22.6 & 23.6 (4.4\%) & 22.6 (0.0\%)& 22.7 (0.4\%) \\
\hline
$5\times500$ ($\sim$9.7M) & 10.9 & 11.1 (1.8\%) & 11.2 (2.8\%) & 11.1 (1.8\%) & 20.9 & 21.7 (3.8\%) & 21.4 (2.4\%) & 21.5 (2.9\%) \\
\hline
\hline
$P=100$ ($\sim$2.7M) & 11.6 & 12.1 (4.3\%) & 11.8 (1.7\%) & 11.9 (2.6\%) & 22.6 & 23.8 (5.3\%) & 23.1 (2.2\%) & 23.3 (3.1\%) \\
\hline
$P=200$ ($\sim$4.8M) & 10.6 & 10.8 (1.9\%) & 10.6 (0.0\%) & 10.7 (0.9\%) & 20.5 & 21.4 (4.4\%) & 20.6 (0.5\%) & 20.7 (1.0\%) \\
\hline
$P=300$ ($\sim$6.8M) & 10.3 & 10.5 (1.9\%) & 10.5 (1.9\%) & 10.6 (2.9\%) & 19.8 & 20.3 (2.5\%) & 20 (1.0\%) & 20.4 (3.0\%) \\
\hline
$P=400$ ($\sim$8.9M) & 10.3 & 10.5 (1.9\%) & 10.3 (0.0\%) & 10.5 (1.9\%) & 19.6 & 20.2 (3.1\%) & 19.8 (1.0\%) & 19.9 (1.5\%) \\
\hline
\hline
Avg. Relative Loss & - & 3.0\% & 0.9\% & 1.6\% & - & 5.2\% & 1.2\% & 1.9\% \\
\hline
\end{tabular}
\caption{\footnotesize Word error rates on `clean' and `noisy' evaluation sets for various
model architectures. Numbers in parentheses represent the loss relative to
the \textbf{`matched'} condition where models are trained and evaluated using
floating point arithmetic.}
\label{tbl:results}
\end{table*}
\section{Introduction}
\label{sec:intro}
\input{intro}
\section{Related work}
\label{sec:related-work}
\input{related-work}
\section{Quantization Scheme}
\label{sec:deepq}
\input{deepq}
\section{Experimental Setup}
\label{sec:experiments-setup}
\input{experiments}
\section{Results}
\label{sec:results}
\input{results}
\section{Conclusions}
\label{sec:conclusions}
\input{conclusions}
\newpage
\eightpt
\bibliographystyle{IEEEtran}
\subsection{Acoustic Model Architectures}
We evaluate architectures which vary along two main dimensions: the total number
of parameters in the model, and whether the architecture uses projection
layers~\cite{SakSeniorBeaufays14} or not.
We train RNN-based acoustic models with 4 or 5 layers of LSTM cells; the number
of LSTM cells, $N$, is kept the same in all of the layers. We use $N=300,
400, 500$ in our experiments for a total of 6 configurations. In addition, we
train models with 5 layers of 500 LSTM cells, but insert a projection
layer of $P$ units after each of the 5 LSTM layers to reduce
the rank of the recurrent and inter-layer weight matrices~\cite{SakSeniorBeaufays14}.
In this work, unlike our previous work~\cite{prabhavalkar2016}, we keep the
size of the projection layers the same across all layers. We consider
$P=100, 200, 300, 400$, thus adding 4 more configurations.
We utilize the same frontend as our previous work~\cite{prabhavalkar2016}:
standard 40-dimensional log mel-filterbank energies over the 8kHz range,
computed every 10ms on 25ms windows of input speech.
Following~\cite{SakSeniorRaoEtAl15b}, we stack features together from 8
consecutive frames (7 frames of right context) and only present every third
stacked frame as input to the network. In addition to stabilizing CTC training,
this reduces computation since the network is only evaluated once every 30ms.
In order to minimize the delay between the acoustics and the output labels produced
by the network, we constrain the set of CTC alignments to be within 100ms of the
locations determined by a forced-alignment~\cite{Senior15}.
Our decoding setup is identical to that presented in our previous
work~\cite{mcgraw16, prabhavalkar2016}.
Following~\cite{lei2013}, we generate a much smaller first-pass language model (LM)
(69.5K n-grams; mostly unigrams) which is composed with the lexicon transducer to
generate the decoder graph; models are re-scored on-the-fly with a larger
5-gram LM.
Our systems are trained on anonymized hand-transcribed utterances extracted
from Google voice-search ($\sim$3M utterances) and dictation ($\sim$1M
utterances) traffic. To improve robustness, we create `multi-style' training
data by synthetically distorting the utterances, simulating the effect of
background noise and reverberation. 20 distorted utterances are created for
each input utterance; noise samples used in this process are extracted from
environmental recordings of everyday events and Youtube videos. Results are
reported on a set of 13.3K hand-transcribed anonymized utterances (135K words)
extracted from Google traffic from an open-ended dictation domain. We also
report results on a `noisy' version of the evaluation set, created
synthetically using a noise distribution with similar characteristics as the
one used to train the model.
\section{Experiments}
\label{sec:experiments}
In pilot experiments, we found that quantization aware CTC training did not produce
models with a better word error rate (WER) performance than `standard' float trained models.
Therefore, in all of our experiments we use float CTC training, and then
apply quantization aware sMBR training.
\subsection{CTC Training of LSTM AMs with Projection Layers}
\begin{figure}
\centering
\includegraphics[width=0.92\columnwidth]{ler_plot.pdf}
\caption{\footnotesize Label error rates on held-out set as a function of training time
during CTC training of model with 200 projection layer nodes
$(P = 200)$ using different learning rate schedules.}
\label{fig:proj_layer_ler}
\end{figure}
As reported in our previous work~\cite{prabhavalkar2016}, we find that
training LSTMs to optimize the CTC criterion is somewhat unstable for models
with projection layers. One solution to this problem, which we proposed
in~\cite{prabhavalkar2016}, is to first train an `uncompressed' model without
any projection layers. This model is used to initialize the projection layer
matrices through a truncated singular value decomposition (SVD) of the
recurrent weight matrices. While this stabilizes the training
process, and has the benefit of providing a principled procedure for setting
the number of nodes in each of the projection layers, it has the drawback
that it requires a two-stage training process that increases
overall training time. Therefore, in the present work, we propose an alternative
strategy that stabilizes CTC training without requiring an expensive two-stage training
process, yet results in better convergence. As a representative example,
in Figure~\ref{fig:proj_layer_ler}, we plot CI-phoneme label error rates (LERs)
for the model with 200 projection nodes in
each layer $(P = 200)$, on a held-out development set. In all cases, we use an
exponentially decaying global learning rate (LR): $\eta_g(t) = c_g 10^{-
\frac{t}{T_g}}$, where $t$ is the total training time ($c_g = 1.5 \times
10^{-4}$ and $T_g = 20 \text{ days}$ in our experiments). The SVD-based
initialization~\cite{prabhavalkar2016} appears as `SVD initialization' in the
figure.
The most straightforward technique to stabilize training is to set the initial
global learning rate as high as possible while avoiding divergence
($c_g = 1.5 \times 10^{-7}$, in our experiments; `Low LR' in
Figure~\ref{fig:proj_layer_ler}). Although this stabilizes training, this leads
to extremely slow convergence since the learning
rate is many orders of magnitude smaller, and significantly worse LER than the
SVD-based initialization.
As an alternative, we propose using a lower learning rate for parameters in
the projection layer, by defining a separate projection learning rate
multiplier $\eta_p(t)$ which multiplies the global learning rate (i.e., the effective learning rate
for projection layer parameters is $\eta_g(t)\eta_p(t)$). We find that we can
stabilize training by using a lower effective learning rate for the projection
layer parameters relative to the rest of the system by using an exponentially
increasing projection learning rate multiplier that gradually scales the
effective learning rate multipler towards the global learning rate (`Scheduled
Projection LR' in Figure~\ref{fig:proj_layer_ler}): $\eta_{p}(t) = c_{p}^{\left(1 -
\min\left\{\frac{t}{T_{p}}, 1\right\}\right)}$ ($c_p = 10^{-3}$ and $T_p = 0.6 \text{
days}$ in our experiments). Note that, $\eta_p(t) \rightarrow 1 \text{ as } t
\rightarrow T_p$, and thus the same effective learning rate is used for all
parameters for $t > T_p$. As can be seen in Figure~\ref{fig:proj_layer_ler},
although the SVD-based initialization outperforms using a single low global
learning rate, the scheduled projection learning rate schedule results in the
fastest convergence, while avoiding the need for the two-stage training
required by the SVD-based initialization. Therefore, we employ the projection
learning rate schedule for CTC training of models with projection layers.
\subsection{Quantization aware sMBR Training of AMs}
\label{sec:am-smbr-qtrain}
Once models have been trained under the CTC criterion, we
sequence-train them to optimize the sMBR criterion.
In order to mitigate the instability encountered during sMBR
training of models with projection layers, we find that it is sufficient to
use a constant learning rate multiplier for projection layer nodes:
$\eta_p(t) = c^{\text{sMBR}}_{p}$ (we set,
$c^{\text{sMBR}}_{p} = 0.5$ and the global LR parameter,
$c_g = 1.5\times10^{-5}$, in our experiments).
| 767299e52c4683d16b7949e1f7f98d866b81e595 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
\IEEEPARstart
{T}{he} latest advantages in deep learning technologies has led to explosive growth in machine learning and computer vision for building systems that have shown significant improvements in a huge range of applications such as image classification \cite{krizhevsky2012imagenet}, \cite{vggnet} and object detection \cite{girshick2015fast}. The fully convolutional neural networks (FCN) \cite{long2015fully} permit end-to-end training and testing for image labeling; holistically-nested edge detector (HED) \cite{xie15hed} learns hierarchically embedded multi-scale edge fields to account for the low-, mid-, and high- level information for contours and object boundaries; Faster R-CNN \cite{ren2015faster} is a state-of-the-art object detection method depending on region proposal algorithms to predict object locations. FCN performs image-to-image training and testing, a factor that has become crucial in attaining a powerful modeling and computational capability of complex natural images and scenes.
\begin{figure}[!t]
\centering
\includegraphics[width=3.4in]{fig1}
\caption{Gland Haematoxylin and Eosin (H\&E) stained slides and ground truth labels. Images in the first row exemplify different glandular structures. Characteristics such as heterogeneousness and anisochromasia can be observed in the image. The second row shows the ground truth. To achieve better visual effects, each color represents an individual glandular structure.}
\end{figure}
The well-suited solution to image labeling/segmentation in which each pixel is assigned a label from a pre-specified set are FCN family models \cite{long2015fully,xie15hed}. However, when it concerns the problem where individual objects need to be identified, they fails to be directly applied to. This is a problem called instance segmentation. In image labeling, two different objects are assigned with the same label as long as they belong to the same class; while in instance segmentation, in addition to obtaining their class labels, it is also demanded that objects belonging to the same class are identified individually.
Exited in most organ systems as important structures, glands fulfill the responsibility of secreting proteins and carbohydrates. However, adenocarcinomas, the most prevalent type of cancer, arises form glandular epithelium. The precise instance segmentation of glands in histopathological images is essential for morphology assessment, which has been proven to be not only a valuable tool for clinical diagnosis but also the prerequisite of cancer differentiation. Nonetheless, the task of segmenting gland instances is very challenging due to the striking dissimilarity of glandular morphology in different histologic grades.
In computer vision, in spite of the promising results for instance segmentation that a recently developed progress \cite{dai2015instance} shows, it is suited for segmenting individual objects in natural scenes. With the proposal of fully convolutional network (FCN) \cite{long2015fully}, the "end-to-end" learning strategy has strongly simplified the training and testing process and achieved state-of-the-art results in solving the segmentation problem back at the time. Krahenbuh \emph{et al.} \cite{krahenbuhl2012efficient} and Zheng \emph{et al.} \cite{zheng2015conditional} integrate Conditional Random Fields (CRF) with FCN to achieve finer partitioning result of FCN. However, their inability of distinguishing different objects leads to the failure in instance segmentation problem.
The attempt to partition the image into semantically meaningful parts while classifying each part into one of pre-determined classes is called semantic segmentation and has already been well studied in computer vision. One limitation of semantic segmentation is its inability of detecting and delineating different instances of the same class while segmentation at the instance level being an important task in medical image analysis. The quantitative morphology evaluation as well as cancer grading and staging requires the instance segmentation of the gland histopathological slide images \cite{kainz2015semantic}.
Current semantic segmentation method cannot meet the demand of medical image analysis.
The intrinsic properties of medical image pose plenty of challenges in instance segmentation \cite{dimopoulos2014accurate}. First of all, objects being in heterogeneous shapes make it difficult to use mathematical shape models to achieve the segmentation. As Fig.1 shows, the cytoplasm being filled with mucinogen granule causes the nucleus being extruded into a flat shape whereas the nucleus appears as a round or oval body after secreting. Second, variability of intra- and extra- cellular matrix is often the culprit leading to anisochromasia. Therefore, the background portion of medical images contains more noise like intensity gradients, compared to natural images. Several problems arose in our exploration of analyzing medical image: 1) some objects lay near the others thus one can only see the tiny gaps between them when zooming in the image on a particular area; or 2) one entity borders another letting their edges adhesive with each other. We call this an issue of \emph{'coalescence'}. If these issues are omitted during the training phase, even there is only one pixel coalescing with another then segmentation would be a total disaster.
In this paper, we aim to developing a practical system for instance segmentation in gland histology images. We make use of multichannel learning, region, boundary and location cues using convolutional neural networks with deep supervision, and solve the instance segmentation issue in the gland histology image. Our algorithm is evaluated on the dataset provided by MICCAI 2015 Gland Segmentation Challenge Contest \cite{sirinukunwattana2016gland, sirinukunwattana2015stochastic} and achieves state-of-the-art performance among all participants and other popular methods of instance segmentation. We conduct a series of experiments in comparison with other algorithms and proves the superiority of the proposed framework.
This paper is arranged as follows. In section \ref{related}, a review of previous work in relative area is presented. In section \ref{method}, the complete methodology of the proposed framework of gland instance segmentation is described. In section \ref{exp}, a detailed evaluation on this method is demonstrated. In section \ref{con}, we give our conclusion.
\section{Related Work}
\label{related}
In this section, a retrospective introduction about instance segmentation will be delivered. Then, to present related work about our framework as clear as possible, information about channels will be delivered respectively, preceded by an overall review of multi-channel framework.
\subsection{Instance segmentation}
Instance segmentation, a task requires distinguishing contour, location, class and the number of objects in image, is attracting more and more attentions in image processing and computer vision. As a complex problem hardly be solved by traditional algorithms, more deep learning approaches are engaged to solve it. For example, SDS \cite{hariharan2014simultaneous} precedes with a proposal generator and then two parallel pathways for region foreground and bounding box are combined as outcome of instance segmentation. Hypercolumn \cite{hariharan2015hypercolumns}, complishes instance segmentation by utilizing hypercolumn features instead of traditional feature maps. MCNs \cite{dai2015instance} category predicted pixels via the result of object detection. In DeepMask \cite{pinheiro2015learning} and SharpMask \cite{pinheiro2016learning} two branches for segmentation and object score are engaged. Different form DeepMask, InstanceFCN \cite{dai2016instance} exploits local coherence rather than high-dimensional features to confirm instances. DCAN \cite{chen2016dcan}, the winner of 2015 MICCAI who shares the same dataset with us, combine contour and region for instance segmentation. To sum up, most of the models mentioned above make contributions to instance segmentation by integrating more than one CNN models to provide proposals and do segmentation.
\subsection{Multichannel Model}
Inspired by models mentioned above, since more than one model should engaged to solve instance segmentation problem, in another words, more than one kind of information is required, then building up a multichannel framework is also a plausible method. Multichannel model usually utilized to integrate features of various kinds to achieve more satisfying result. As far as we know, multichannel framework is rarely seen in instance segmentation of medical images. It can be seen in grouping features \cite{lee2014multi}, face recognizing \cite{chen2015learning} and image segmentation \cite{scharwachter2013efficient}, which leverage a bag-of-feature pipeline to improve the performance. In our multichannel framework, three channels aim at segmentation, object detection and edge detection are fused together. Related work about them are introduced respectively as follows.
\subsubsection{Image Segmentation}
Image segmentation aim at producing pixelwise labels to images. In neural network solution, the fully convolutional neural network \cite{long2015fully} takes the role of a watershed. Before that, patchwise training is common. Ciresan \emph{et al.} \cite{ciresan2012deep} utilize DNN to segment images of electron microscope. Farabet \emph{et al.} \cite{farabet2013learning} segment natural scene and label them. Liu \emph{et al.} \cite{liu2015crf} extract features of different patches from superpixel. Then CRF is trained to provide ultimate segmentation result. FCN \cite{long2015fully} puts forward a more efficient model to train end-to-end. After that, fully convolutional network attracts people’s attention. U-net \cite{ronneberger2015u} preserve more context information by maintaining more feature channels at up-sampling part compared to FCN. Dai \emph{et al.} \cite{dai2016instance} improve FCN model to solve instance segmentation problems.
We leverage the FCN model to produce the information of probability masks, as the region channel in our framework.
\subsubsection{Object Detection}
Object detection problem requires a system locate objects of different classes within an image. A common deep convolutional neural network solution is usually running a classier on candidate proposals and many models have been arose and improved on the basis of it. R-CNN \cite{girshick2014rich} is a representative approach that proposals are generated by an unsupervised algorithm and classified by SVMs by features extracted by DNN. To accelerate R-CNN, fast R-CNN \cite{girshick2015fast} and faster R-CNN \cite{ren2015faster} are put forward one after another. Different from R-CNN, DeepMultiBox \cite{erhan2014scalable} generate proposals by using DNN. The network for proposal producing in OverFeat \cite{sermanet2013overfeat} share weights with network designed for classification tasks and reconcile results of classification and proposals as the ultimate result. YOLO \cite{redmon2015you} is an end-to-end model that predict proposals and class probabilities simultaneously by regard object detection problem as a single regression problem.
\subsubsection{Edge detection}
Approaches to computational edge detection play a fundamental role in the history of image processing. In recent years, solutions of neural network, at once flourishing and effective, bring about a new access towards solving complicated edge detection problems. HED \cite{xie15hed} earns hierarchically embedded multi-scale edge fields to account for the low-, mid-, and high- level information for contours and object boundaries. DeepEdge \cite{bertasius2015deepedge} also utilize multi-scale of image to solve this problem by using deep convolutional neural network. Ganin \emph{et al.} \cite{ganin2014n} propose an approach to solve edge detection problem by integrate the neural network with the nearest neighbor search. Shen \emph{et al.} \cite{shen2015deepcontour} propose a new loss function and improve the accuracy of counter detection via a neural network.
\subsection{Previous work}
Earlier conference version of our approach were presented in Xu \emph{et al.}\cite{xu2016gland}. Here we further illustrate that: (1) we add another channel - detection channel - in this paper, due to the reason that the region channel and the detection channel complement each other; (2) this framework achieves state-of-the-art results; (3) to address the problem of images with rotation invariance, we find a new data augmentation strategy that is proven to be effective. (4) ablation experiments are carried out to corroborate the effectiveness of the proposed framework.
\section{Method}
\label{method}
In this section, we will introduce details about our framework (as shown in Fig.~\ref{model1}).
By integrating the information generated from different channels, our multichannel framework is capable of instance segmentation. Aiming at solving this problem, we select three channels, foreground segmentation channel for image segmentation, object detection channel for gland detection and edge detection channel. The reason of choosing these three channels is based on the fact that information of region, contour and location contributes receptively and complimentarily to our ultimate purpose and the joint effort of them will perform much better than each of them alone. In our framework, effects of different channels are distinct. The foreground segmentation channel in our framework distinguishes the foreground and background of images. Targeted regions in pathological contains complex morphological features. It is common that glands grow close to one another. This, however, will bring about a negative effect for algorithm that the distance between two adjacent glands are too diminutive to distinguish by computer. Machines tend to conflate two glands as a whole even there do exist a gap between them.
Therefore, the object detection channel is in demand of designating the bounding box of each gland to which foreground pixels in that box are belonged.
In regard to the overlapping area of bounding boxes, glands boundaries are predicted by the edge detection channel.
As for the area that glands are close, edge detection fails to precisely predict boundaries of glands and requires the assistance of object detection channel.
Only under the joint effort of various kinds of information, can instance segmentation problem be properly solved.
\begin{figure}[!t]
\centering
\includegraphics[width=3.4in]{model1}
\caption{This illustrates a brief structure of the proposed framework. The foreground segmentation channel distinguishes glands from the background. The object detection channel detects glands and their region in the image. The edge detection channel outputs the result of boundary detection. A convolution neural network concatenates features generated by different channels and produces segmented instances.}
\label{model1}
\end{figure}
\begin{figure*}[!t]
\centering
\includegraphics[width=7in]{model2}
\caption{This illustrates the structure of this framework. We fuse outputs of three channels to achieve instance segmentation. For all the channels in this framework, FCN for region channel, Faster RCNN for object channel and HED for edge channel, are all based on VGG16 model, we present this classical five pooling structure in details by "Conv Net" at the top of the figure and show it briefly by a rectangular block named "Conv Net". Especially, in region and object channels, arrows pointing from "Conv Net" denotes the output of the "Conv Net", while in edge channel they represent output of deep supervisions. In the region channel, strides of the last two pooling layers of "Conv Net" are set as 1; atrous convolution being applied to convolution layers leads to the higher resolution of feature maps (as annotated in brackets). }
\label{model2}
\end{figure*}
\subsection{Foreground Segmentation Channel}
The foreground segmentation channel distinguishes glands from the background.
With the arising of FCN, image segmentation become more effective thanks to the end-to-end training strategy and dense prediction on attribute-sized images. FCN replace the fully-connected layer with a convolutional layer and upsample the feature map to the same size as the original image through deconvolution thus an end-to-end training and prediction is guaranteed. Compared to the previous prevalent method sliding window in image segmentation, FCN is faster and simpler. FCN family models \cite{long2015fully,xie15hed} have achieved great accomplishment in labeling images. Usually, the FCN model can be regarded as the combination of a feature extractor and a pixel-wise predictor. Pixel-wise predictor predict probability mask of segmented images. The feature extractor is able to abstract high-level features by down-sampling and convolution. Though, useful high-level features are extracted, details of images sink in the process of max-pooling and stride convolution. Consequently, when objects are adjacent to each other, FCN may consider them as one. It is natural having FCN to solve image segmentation problems. However, instance segmentation is beyond the ability of FCN. It requires a system to differentiate instances of the same class even they are extremely close to each other. Even so, the probability mask produced by FCN still performs valuable support on solving instance segmentation problems.
To compensate the resolution reduction of the feature map due to downsampling, FCN introduce the skip architecture to combine deep, semantic information and shallow, appearance information. Nevertheless, DeepLab \cite{chen2016deeplab} proposes the FCN with atrous convolution that empowers the network with the wider receptive field without downsampling. Less downsampling layer means less space-invariance brought by downsampling which is benefit to the enhancement of segmentation precision.
Our foreground segmentation channel is a modified version of FCN-32s \cite{long2015fully} of which strides of pool4 and pool5 are 1 and subsequent convolution layers enlarge the receptive field by the atrous convolution.
Given an input image $X$ and the parameter of FCN network is denoted as ${w}_{s}$, thus the output of FCN is
\begin{equation}
{P}_{s}\left(Y^{*}_{s}=k \mid X;w_{s}\right) = {\mu}_{k}\left(h_{s}\left(X,w_{s}\right)\right),
\end{equation}
where $\mu(\cdot)$ is the softmax function. $\mu_{k}(\cdot)$ is the output of the $k$th category and $h_{s}(\cdot)$ outputs the feature map of the hidden layer.
\subsection{Object Detection Channel}
The object detection channel detects glands and their locations in the image.
The location of object is helpful on counting number and identifying the range of objects. According to some previous works on instance segmentation, such as MNC \cite{dai2016instance}, confirmation of the bounding-box is usually the first step towards instance segmentation. After that, segmentation and other options are carried out within bounding boxes. Though this method is highly approved, the loss of context information caused by limited receptive fields and bounding-box may exacerbate the segmentation result. Consequently, we integrate the information of location to the fusion network instead of segmenting instances within bounding boxes.
To achieve the location information, Faster-RCNN, a state-of-the-art object detection model, is engaged to solve this problem. In this model, convolutional layers are proposed to extract feature maps from images. After that, Region Proposal Network (RPN) takes arbitrary-sized feature map as input and produces a set of bounding-boxes with probability of objects. Region proposals will be converted into regions of interest and classified to form the final object detection result.
Filling is operated in consonance with other two channel and to annotate the overlapping area. Sizes of various channels should be the same before gathering into the fusion network. To guarantee the output size of object detection channel being in accordance with other channels, we reshape it and change the bounding box into another formation. The value of each pixel in region covered by the bounding box equals to the number of bounding boxes it belongs to. For example, if a pixel is in the public area of three bounding boxes, then the value of that pixel will be three.
We denote $w_{d}$ as the parameter of Faster-RCNN and $\phi$ represents the filling operation of bounding box. The output of this channel is
\begin{equation}
P_{d}\left(X,w_{d}\right) = \phi\left(h_{d}\left(X,w_{d}\right)\right).
\end{equation}
$h_{d}\left(\cdot\right)$ is the predicted coordinate of the bounding box.
\subsection{Edge Detection Channel}
The edge detection channel detects boundaries between glands.
The combination of merely the probability mask predicted by FCN and the location of glands tend to fuzzy boundaries of glands, especially between adjacent objects, consequently it is tough to distinguish different objects. To receive precise and clear boundaries, the information of edge is crucial which has also been proved by DCAN \cite{chen2016dcan}. The effectiveness of edge in our framework can be concluded into two aspects. Firstly, edge compensate the information loss caused by max-pooling and other operations in FCN. As a result, the contours become more precise and the morphology become more similar to the ground truth. Secondly, even if the location and the probability mask are confirmed, it is unavoidable that predicted pixel regions of adjacent objects are still connected. Edge, however, is able to differentiate them apart.
As expected, the synergies among region, location and edge finally achieve the state-of-the art result. The edge channel in our model is based on Holistically-nested Edge Detector (HED) \cite{xie15hed}. It is a CNN-based solution towards edge detection. It learns hierarchically embedded multi-scale edge fields to account for the low-, mid-, and high- level information for contours and object boundaries. In edge detection task, pixels of labels are much less than pixels of back ground. The imbalance may decrease the convergence rate or even cause the non-convergence problem. To solve the problem, deep supervision \cite{lee2015deeply} and balancing of the loss between positive and negative classes are engaged. In total, there are five side supervisions which are established before each down-sampling layers.
We denote $w_{e}$ as the parameter of HED, thus the $m$th prediction of deep supervision is
\begin{equation}
P^{(m)}_{e}(Y^{(m)*}_{e}=1 \mid X;w_{e})=\sigma(h^{(m)}_{e}(X,w_{e}).
\end{equation}
$\sigma(\cdot)$ denotes sigmoid function - the output layer of HED. $h^{(m)}_{e}$ represents the output of the hidden layer that relative to $m$th deep supervision. The weighted sum of M outputs of deep supervision is the final result of this channel and the weighted coefficient is $\alpha$. This process is delivered through the convolutional layer. The back propagation enables the network to learn relative importances of edge predictions under different scales.
\begin{equation}
P_{e}(Y^{*}_{e}=1 \mid X;w_{e},\alpha) = \sigma(\sum_{m=1}^{M}\alpha^{(m)}\cdot h^{(m)}_{e}(X,w_{e})).
\end{equation}
\subsection{Fusing Multichannel}
Merely receiving the information of these three channels is not the ultimate purpose of our algorithm. Instance segmentation is. As a result, a fusion system is of great importance to maximize synergies of these three kinds of information above. It is hard for a non-learning algorithm to recognize the pattern of all these information. Naturally, a CNN based solution is the best choice.
After obtaining outputs of these three channels, a shallow seven-layer convolutional neural network is used to combine the information and yield the final result. To reduce the information loss and ensure sufficiently large reception field, we once again replace downsampling with the atrous convolution.
We denote $w_{f}$ as the parameter of this network and $h_{f}$ as the hidden layer. Thus the output of the network is
\begin{equation}
P_{f}\left(Y_{f}^{*}=k\mid P_{s},P_{d},P_{e};w_{f}\right)=\mu_{k}\left(h_{f}\left(P_{s},P_{d},P_{e},w_{f}\right)\right).
\end{equation}
\section{Experiment}
\label{exp}
\begin{figure*}[!t]
\centering
\includegraphics[width=7in]{result}
\caption{From left to right: original image, ground truth, results of FCN, results of FCN with atrous convolution and results of the proposed framework. Compared to FCN, most of adjacent glandular structures are separated apart which indicates that our framework accomplishes the instance segmentation goal. However, few glands with small sizes or filled with red blood cells escape the detection of our model. The bad performance in the last row is due to the fact that in most samples, the white area is recognized as cytoplasm while in this sample, the white area is the background.
}
\label{result}
\end{figure*}
\begin{table*}[t]
\centering
\resizebox{\textwidth}{!}{
\begin{threeparttable}[b]
\caption{Performance in Comparison to Other Methods}
\begin{tabular}{c|c|c|c|c|c|c|c|c|c|c|c|c|c|c}
\hline
\multirow{3}{*}{Method}&
\multicolumn{4}{c|}{F1 Score}&
\multicolumn{4}{c|}{ObjectDice}&
\multicolumn{4}{c|}{ObjectHausdorff}&
\multirow{3}{*}{Rank Sum}&
\multirow{3}{*}{Weighted Rank Sum}\\
\cline{2-13}
&\multicolumn{2}{c|}{Part A}&
\multicolumn{2}{c|}{Part B}&
\multicolumn{2}{c|}{Part A}&
\multicolumn{2}{c|}{Part B}&
\multicolumn{2}{c|}{Part A}&
\multicolumn{2}{c|}{Part B}& \\
\cline{2-13}
& Score & Rank & Score & Rank & Score & Rank & Score & Rank & Score & Rank & Score & Rank & \\
\hline
FCN & 0.788 & 11 & 0.764 & 4 & 0.813 & 11 & 0.796 & 4 & 95.054 & 11 & 146.2478 & 4 & 45 & 27.75 \\
\hline
FCN with atrous convolution \cite{chen2016deeplab} & 0.854 & 9 & 0.798 & 2 & 0.879 & 6 & 0.825 & 2 & 62.216 & 9 & 118.734 & 2 & 30 & 19.5 \\
\hline\hline
\textbf{\cellcolor[rgb]{.9,.9,.9}Ours} & \cellcolor[rgb]{.9,.9,.9}0.893 & \cellcolor[rgb]{.9,.9,.9}3 & \textbf{\cellcolor[rgb]{.9,.9,.9}0.843} & \textbf{\cellcolor[rgb]{.9,.9,.9}1} & \textbf{\cellcolor[rgb]{.9,.9,.9}0.908} & \textbf{\cellcolor[rgb]{.9,.9,.9}1} & \textbf{\cellcolor[rgb]{.9,.9,.9}0.833} & \textbf{\cellcolor[rgb]{.9,.9,.9}1} & \textbf{\cellcolor[rgb]{.9,.9,.9}44.129} & \textbf{\cellcolor[rgb]{.9,.9,.9}1} & \textbf{\cellcolor[rgb]{.9,.9,.9}116.821} & \textbf{\cellcolor[rgb]{.9,.9,.9}1} & \cellcolor[rgb]{.9,.9,.9}8 & \cellcolor[rgb]{.9,.9,.9}4.5\\
\hline
CUMedVision2 \cite{chen2016dcan} & \textbf{0.912} & \textbf{1} & 0.716 & 6 & 0.897 & 2 & 0.781 & 8 & 45.418 & 2 & 160.347 & 9 & 28 & 9.5\\
\hline
ExB3 & 0.896 & 2 & 0.719 & 5 & 0.886 & 3 & 0.765 & 9 & 57.350 & 6 & 159.873 & 8 & 33 & 13.75\\
\hline
ExB2 & 0.892 & 4 & 0.686 & 9 & 0.884 & 4 & 0.754 & 10 & 54.785 & 3 & 187.442 & 11 & 41 & 15.75\\
\hline
ExB1 & 0.891 & 5 & 0.703 & 7 & 0.882 & 5 & 0.786 & 5 & 57.413 & 7 & 145.575 & 3 & 32 & 16.5\\
\hline
Frerburg2 \cite{ronneberger2015u} & 0.870 & 6 & 0.695 & 8 & 0.876 & 7 & 0.786 & 6 & 57.093 & 4 & 148.463 & 6 & 37 & 17.75\\
\hline
Frerburg1 \cite{ronneberger2015u} & 0.834 & 10 & 0.605 & 11 & 0.875 & 8 & 0.783 & 7 & 57.194 & 5 & 146.607 & 5 & 46 & 23\\
\hline
CUMedVision1 \cite{chen2016dcan} & 0.868 & 7 & 0.769 & 3 & 0.867 & 10 & 0.800 & 3 & 74.596 & 10 & 153.646 & 7 & 40 & 23.5\\
\hline
CVIP Dundee & 0.863 & 8 & 0.633 & 10 & 0.870 & 9 & 0.715 & 11 & 58.339 & 8 & 209.048 & 13 & 59 & 27.25\\
\hline
LIB & 0.777 & 12 & 0.306 & 14 & 0.781 & 12 & 0.617 & 13 & 112.706 & 13 & 190.447 & 12 & 76 & 37.5\\
\hline
CVML & 0.652 & 13 & 0.541 & 12 & 0.644 & 14 & 0.654 & 12 & 155.433 & 14 & 176.244 & 10 & 75 & 39.25\\
\hline
vision4GlaS & 0.635 & 14 & 0.527 & 13 & 0.737 & 13 & 0.610 & 14 & 107.491 & 12 & 210.105 & 14 & 80 & 39.5\\
\hline
\end{tabular}\end{threeparttable}}
\label{table}
\end{table*}
\subsection{Dataset}
Our method is evaluated on the dataset provided by MICCAI 2015 Gland Segmentation Challenge Contest \cite{sirinukunwattana2016gland, sirinukunwattana2015stochastic}. The dataset consists of 165 labeled colorectal cancer histological images scanned by Zeiss MIRAX MIDI. The resolution of the image is approximately 0.62μm per pixel. 85 images belong to training set and 80 affiliate to test sets (test A contains 60 images and test B contains 20 images). There are 37 benign sections and 48 malignant ones in training set, 33 benign sections and 27 malignant ones in testing set A and 4 benign sections and 16 malignant ones in testing set B.\subsection{Data augmentation and Processing}
We first preprocess data by performing per channel zero mean. The next step is to generate edge labels from region labels and perform dilation to edge labels afterward. Whether pixel is edge or not is decided by four nearest pixels (over, below, right and left) in region label. If all four pixels in region channel belongs to foreground or all of them belongs to background, then this pixel is regarded as edge. To enhance performance and combat overfitting, copious training data are needed. Given the circumstance of the absence of large dataset, data augmentation is essential before training. Two strategies of data augmentation has been carried out and the improvement of results is a strong evidence to prove the efficiency of data augmentation. In Strategy \uppercase\expandafter{\romannumeral1}, horizontal flipping and rotation operation ($0^\circ$, $90^\circ$, $180^\circ$, $270^\circ$) are used in the training images. Besides operations in Strategy \uppercase\expandafter{\romannumeral1}, Strategy \uppercase\expandafter{\romannumeral2} also includes sinusoidal transformation, pin cushion transformation and shear transformation. Deformation of original images is beneficial to the increasement of robustness and the promotion of final result. After data augmentation, a $400 \times 400$ region is cropped from the original image as input.
\subsection{Evaluation}
Evaluation method is the same as the competition goes. Three indicators are involved to evaluate performance on test A and test B. Indicators assess detection result respectively, segmentation performance and shape similarity. Final score is the summation of six rankings and the smaller the better. Since image amounts of test A and test B are of great difference, we not only calculate the rank sum as the host of MICCAI 2015 Gland Segmentation Challenge Contest demands, but we also list the weighted rank sum. The weighted rank sum is calculated as:
\begin{equation}
Weighted RS=\frac{3}{4}\sum test A Rank+\frac{1}{4}\sum test B Rank.
\end{equation}
The program for evaluation is given by MICCAI 2015 Gland Segmentation Challenge Contest \cite{sirinukunwattana2016gland, sirinukunwattana2015stochastic}.
The first criterion for evaluation reflets the accuracy of gland detection which is called F1score. The segmented glandular object of True Positive (TP) is the object that shares more than 50\% areas with the ground truth. Otherwise, the segmented area will be determined as False Positive (FP). Objects of ground truth without corresponding prediction are considered as False Negative (FN).
\begin{equation}
F1score = \frac{2\cdot Precision\cdot Recall}{Precision + Recall}
\end{equation}
\begin{equation}
Precision = \frac{TP}{TP+FP}
\end{equation}
\begin{equation}
Recall=\frac{TP}{TP+FN}
\end{equation}
Dice is the second criterion for evaluating the performance of segmentation. Dice index of the whole image is
\begin{equation}
D(G,S)=\frac{2(\mid G\cap S\mid)}{\mid G\mid +\mid S\mid},
\end{equation}
of which G represents the ground truth and S is the segmented result. However, it is not able to differentiate instances of same class. As a result, object-level dice score is employed to evaluate the segmentation result. The definition is as follows:
\begin{equation}
D_{object}(G,S)=\frac{1}{2}\left[\sum_{i=1}^{n_{S}}w_{i}D(G_{i},S_{i})+\sum_{j=1}^{n_{G}}\widetilde{w}_{j}D(\widetilde{G}_{i},\widetilde{S}_{i})\right],
\end{equation}
\begin{equation}
w_{i}=\frac{\mid S_{i}\mid}{\sum_{j=1}^{n_{S}}\mid S_{j}\mid},
\end{equation}
\begin{equation}
\widetilde{w}_{i}=\frac{\mid \widetilde{G}_{i}\mid}{\sum_{j=1}^{n_{G}}\mid \widetilde{G}_{j} \mid}.
\end{equation}
$n_{S}$ and $n_{G}$ are the number of instances in the segmented result and ground truth.
Shape similarity reflects the performance on morphology likelihood which plays a significant role in gland segmentation. Hausdorff distance is exploited to evaluate the shape similarity. To assess glands respectively, the index of Hausdorff distance deforms from the original formation:
\begin{equation}
H(G,S)=\mathrm{max}\left\{\underset{x\epsilon G}{sup} \underset{y\epsilon S}{inf}\left\|x-y\right\|,\underset{y\epsilon S}{sup} \underset{x\epsilon G}{inf}\left\|x-y\right\|\right\},
\end{equation}
to the object-level formation:
\begin{equation}
H_{object}(S,G)=\frac{1}{2}\left[\sum_{i=1}^{n_{s}}w_{i}H(G_{i},S_{i})+\sum_{i=1}^{n_{G}}\widetilde{w}_{i}H(\widetilde{G}_{i},\widetilde{S}_{i})\right],
\end{equation}
where
\begin{equation}
w_{i}=\frac{|S_{i}|}{\sum_{j=1}^{n_{S}}|S_{j}|},
\end{equation}
\begin{equation}
\widetilde{w}_{i}=\frac{|\widetilde{G}_{i}|}{\sum_{j=1}^{n_{G}}|\widetilde{G}_{j}|}.
\end{equation}
Similar to object-level dice index $n_{S}$ and $n_{G}$ represents instances of segmented objects and ground truth.
\subsection{Result and Discussion}Our framework performs well on datasets provided by MICCAI 2015 Gland Segmentation Challenge Contest and achieves the state-of-the-art result (as listed in Table \uppercase\expandafter{\romannumeral1}) among all participants \cite{sirinukunwattana2016gland}.
We rearrange the scores and ranks in this table. Our method outranks FCN and other participants \cite{sirinukunwattana2016gland} based on both rank sum and weighted rank sum.
Compared to FCN and FCN with atrous convolution, our framework obtains better score which is a convincing evidence that our work is more effective in solving instance segmentation problem in histological images. Though, FCN with atrous convolution performs better than FCN, for atrous convolution process less poolings and covers larger receptive fields, our framework combines information of region, location and edge to achieve higher score in the dataset. The reason why our framework rank higher, is because most of the adjacent glandular structures have been separated apart, so that more beneficial to meet the evaluation index of instance segmentation, while in FCN and FCN with atrous convolution they are not. Results of comparision are illustrated in Fig.~\ref{result}.
\begin{table*}[t]
\centering
\caption{Comparison with instance segmentation methods}
\begin{tabular}{c|c|c|c|c|c|c}
\hline
\multirow{2}{*}{Method} &
\multicolumn{2}{c|}{F1 Score} &
\multicolumn{2}{c|}{ObjectDice} &
\multicolumn{2}{c}{ObjectHausdorff} \\
\cline{2-7}
& Part A & Part B & Part A & Part B & Part A & Part B\\
\hline
HyperColumn \cite{hariharan2015hypercolumns} & 0.852 & 0.691 & 0.742 & 0.653 & 119.441 & 190.384\\
\hline
MNC \cite{dai2015instance} & 0.856 & 0.701 & 0.793 & 0.705 & 85.208 & 190.323\\
\hline
SDS \cite{hariharan2014simultaneous} & 0.545 & 0.322 & 0.647 & 0.495 & 116.833 & 229.853\\
\hline
BOX-$>$FCN with atrous convolution+EDGE3 & 0.807 & 0.700 & 0.790 & 0.696 & 114.230 & 197.360\\
\hline
\cellcolor[rgb]{.9,.9,.9}OURS & \cellcolor[rgb]{.9,.9,.9}\textbf{0.893} & \cellcolor[rgb]{.9,.9,.9}\textbf{0.843} & \cellcolor[rgb]{.9,.9,.9}\textbf{0.908} & \cellcolor[rgb]{.9,.9,.9}\textbf{0.833} & \cellcolor[rgb]{.9,.9,.9}\textbf{44.129} & \cellcolor[rgb]{.9,.9,.9}\textbf{116.821}\\
\hline
\end{tabular}
\end{table*}
\begin{table*}
\centering
\caption{Data Augmentation Strategy comparison}
\begin{tabular}{c|c|c|c|c|c|c|c}
\hline
\multirow{2}{*}{Strategy} &
\multirow{2}{*}{Method} &
\multicolumn{2}{c|}{F1 Score} &
\multicolumn{2}{c|}{ObjectDice} &
\multicolumn{2}{c}{ObjectHausdorff} \\
\cline{3-8}
& & Part A & Part B & Part A & Part B & Part A & Part B\\
\hline
\multirow{2}{*}{Strategy \uppercase\expandafter{\romannumeral1}} &
FCN & 0.709 & 0.708 & 0.748 & 0.779 & 129.941 & 159.639\\
\cline{2-8}
& FCN with atrous convolution \cite{chen2016deeplab} & 0.820 & 0.749 & 0.843 & 0.811 & 79.768 & 131.639\\
\hline
\multirow{2}{*}{Strategy \uppercase\expandafter{\romannumeral2}} &
FCN & 0.788 & 0.764 & 0.813 & 0.796 & 95.054 & 146.248\\
\cline{2-8}
& FCN with atrous convolution \cite{chen2016deeplab} & \textbf{0.854} & \textbf{0.798} & \textbf{0.879} & \textbf{0.825} & \textbf{62.216} & \textbf{118.734}\\
\hline
\end{tabular}
\end{table*}
\begin{table*}[t]
\centering
\caption{Plausibility of Channels. We denote AMC as fusion network with atrous convolution and MC as fusion network without atrous convolution. EDGE1 represents that edge label are not dilated while EDGE3 signifies edge label dilated by a disk filter with radius 3. BOX detnotes the bounding box.}
\begin{tabular}{c|c|c|c|c|c|c}
\hline
\multirow{2}{*}{Method} &
\multicolumn{2}{c|}{F1 Score} &
\multicolumn{2}{c|}{ObjectDice} &
\multicolumn{2}{c}{ObjectHausdorff} \\
\cline{2-7}
& Part A & Part B & Part A & Part B & Part A & Part B\\
\hline
MC: FCN + EDGE1 + BOX & 0.863 & 0.784 & 0.884 & 0.833 & 57.519 & 108.825\\
\hline
MC: FCN + EDGE3 + BOX & 0.886 & 0.795 & 0.901 & 0.840 & 49.578 & 100.681\\
\hline
MC: FCN with atrous convolution + EDGE3 + BOX & 0.890 & 0.816 & \textbf{0.905} & 0.841 & 47.081 & 107.413\\
\hline
\hline
AMC: FCN + EDGE3 + BOX & \textbf{0.893} & 0.803 & 0.903 & \textbf{0.846} & 47.510 & \textbf{97.440}\\
\hline
AMC: FCN with atrous convolution + EDGE3 + BOX & \textbf{0.893} & \textbf{0.843} & 0.908 & 0.833 & \textbf{44.129} & 116.821\\
\hline
AMC: FCN with atrous convolution + EDGE1 + BOX & 0.876 & 0.824 & 0.894 & 0.826 & 50.028 & 123.881\\
\hline
\hline
AMC: FCN with atrous convolution + BOX & 0.876 & 0.815 & 0.893 & 0.808 & 50.823 & 132.816\\
\hline
AMC: FCN with atrous convolution + EDGE3 & 0.874 & 0.816 & 0.904 & 0.832 & 46.307 & 109.174\\
\hline
\end{tabular}
\end{table*}
Ranks of test A are higher than test B in general due to the inconsistency of data distribution. In test A, most images are the normal ones while test B contains a majority of cancerous images which are more complicated in shape and lager in size. Hence, a larger receptive field is required in order to detect cancerous glands. However, before we exploit the atrous convolution algorithm, the downsampling layer not only gives the network larger receptive field but also make the resolution of the feature map decreases thus the segmentation result becomes worse. The atrous convolution algorithm empower the convolutional neural network with larger receptive field with less downsampling layers. Our multichannel framework enhances the performance based on the FCN with atrous convolution by adding two channels - edge detection channel and object detection channel.
Since the differences between background and foreground in histopathological image are small (3th row of Fig.~\ref{result}), FCN and FCN with atrous convolution sometimes predict the background pixel as gland thus raise the false positive rate. The multichannel framework abates the false positive by adding context of pixel while predicting object location.
Compared to CUMedVision1 \cite{chen2016dcan}, CUMedVision2 \cite{chen2016dcan} add the edge information thus results of test A improve yet those of test B deteriorate. But our method improves both results of test A and test B after combine the context of edge and location.
However, white regions in gland histopathological images are of two kinds: 1) cytoplasm; and 2) there is no cell or tissue (background). The difference between these two kinds is that cytoplasm usually appears surrounded by nuclei or other stained tissue. In the image of the last row in Fig.~\ref{result}, glands encircles white regions without cell or tissue causing that the machine mistakes them as cytoplasm. As for images of the 4th and 5th row in Fig.~\ref{result}, glands are split when cutting images, which is the reason that cytoplasm is mistaken as the background.
\textbf{Comparison with instance segmentation methods}
Currently, methods suitable for instance segmentation of images of natural scenes predict instances based on detection or proposal, such as SDS \cite{hariharan2014simultaneous}, Hypercolumn \cite{hariharan2015hypercolumns} and MNC \cite{dai2015instance}. One defect of this logic is its dependence on the precision of detection or proposal. If the object escapes the detection, it would evade the subsequent segmentation as well. Besides, the segmentation being restricted to a certain bounding box would have little access to context information hence impact the result. Under the condition of bounding boxes overlapping one another, which instance does the pixel in the overlapping region belongs to cannot be determined. The overlapping area falls into the category of the nearest gland in our experiment.
To further demonstrate the defect of the cascade architecture, we designed a baseline experiment. We first perform gland detection then segment gland instances inside bounding boxes. There is a shallow network (same as the fusion network) combines the information of foreground segmentation and edge detection to generate the final result. Configurations of all experiments are set the same as our method. Results are showed in Table \uppercase\expandafter{\romannumeral2} and prove to be less effective than the proposed framework.
\subsection{Ablation Experiment}
\subsubsection{Data Augmentation Strategy}
To enhance performance and combat overfitting, data augmentation is essential before training. We observe through experiments that adequate transformation of gland images is beneficial to training. This is because that glands are in various shape naturally and cancerous glands are more different in morphology. Here we evaluate the effect on results of foreground segmentation channel using Strategy \uppercase\expandafter{\romannumeral1} and Strategy \uppercase\expandafter{\romannumeral2}. We present the results in Table \uppercase\expandafter{\romannumeral3}.
\subsubsection{Plausibility of Channels}
In the convolutional neural network, the main purpose of downsampling is to enlarge the receptive field yet at a cost of decreased resolution and information loss of original data. Feature maps with low resolution would increase the difficulty of upsample layer training. The representational ability of feature maps is reduced after upsampling and further lead to inferior segmentation result. Another drawback of downsampling is the space invariance it introduced while segmentation is space sensitive. The inconsistence between downsampling and image segmentation is obvious. The atrous convolution algorithm empower the convolutional neural network with larger receptive field with less downsampling layers.
The comparison between segmentation performances of FCN with and without the atrous convolution shows the effectiveness of it in enhancing the segmentation precision. The foreground segmentation channel with the FCN with atrous convolution improves the performance of the multichannel framework. So does the fusion stage with the atrous convolution.
Pixels belonged to edge occupy a extremely small proportion of the whole image. The imbalance between edge and non-edge poses a severe threat to the network training that may lead to non convergence. Edge dilation can alleviate the imbalance in a certain way and improve the edge detection precision.
To prove that these three channels truly improve the performance of instance segmentation, we conduct the following two baseline experiments: a) merely launch foreground segmentation channel and edge detection channel; b) merely launch foreground segmentation channel and object detection channel. The results is in favor of the three-channel framework with no surprise.
Results of experiments mentioned above are presented in Table \uppercase\expandafter{\romannumeral4}.
\section{Conclusion}
\label{con}
We propose a new framework called deep multichannel neural networks which achieves state-of-the-art results in MICCAI 2015 Gland Segmentation Challenge. The universal framework extracts features of edge, region and location then concatenate them together to generate the result of instance segmentation.
In future work, this algorithm can be expanded in instance segmentation of medical images.
\section*{Acknowledgment}
This work is supported by Microsoft Research under the eHealth program, the Beijing National Science Foundation in China under Grant 4152033, the Technology and Innovation Commission of Shenzhen in China under Grant shenfagai2016-627, Beijing Young Talent Project in China, the Fundamental Research Funds for the Central Universities of China under Grant SKLSDE-2015ZX-27 from the State Key Laboratory of Software Development Environment in Beihang University in China.
\ifCLASSOPTIONcaptionsoff
\newpage
\fi
\bibliographystyle{IEEEtran}
| 49932c21a22ad57c549090228fee6665c532e335 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Multifractal analysis is the study of the concentration of a measure on level sets. In particular, we use measures for a hyperbolic map to study the multifractal spectrum. Multifractal analysts historically modeled an expanding map with a finite state shift. Their main result is the expression of the multifractal spectrum, which is concave and analytic. However, the multifractal spectrum is not analytic everywhere in the case of countable state Markov shifts. Iommi \cite{iommi2005multifractal} proves an expression for the multifractal spectrum in the setting of a countable Markov shift. Using Iommi's result, our paper proves that the multifractal spectrum has up to infinitely many phase transitions or non-analytic points.
\par Before outlining our results, we give some background and definitions in the area of multifractal analysis. Consider the countable state shift $\Sigma$ satisfying topological mixing and the Big Images Property. We take the measure $\mu$ on our countable shift to be a Gibbs state and $\alpha \in \mathbb{R}.$ The sequence $x$ is said to have pointwise dimension $\alpha$ if
\[
d_{\mu}(x):=\lim\limits_{m \rightarrow \infty}\frac{\log\mu([x_{1},...,x_{m}])}{\log|[x_{1},...,x_{m}]|}=\alpha.\]
Let the set of sequences $x$ in $\Sigma$ such that $d_{\mu}(x)$ does not exist be denoted as $X'$ and let the set of sequences with pointwise dimension $\alpha$ be denoted as $X_{\alpha}^{s}.$ Hence, we can decompose $\Sigma$ into a disjoint union called the multifractal decomposition based on the pointwise dimension of each $x$ in our countable shift. We will consider a function that gives the Hausdorff dimension of sets that have a pointwise dimension. The multifractal spectrum with respect to $\mu$ is the function \[f_{\mu}(\alpha)=\dim_{H}(X_{\alpha}^{s}).\]
We remark that the multifractal spectrum is similarly defined for finite state Markov shifts.
\par We first discuss historical results about the multifractal spectrum in the setting of finite state Markov shifts. Rand \cite{rand1989singularity} considers a cookie-cutter, which is a uniformly hyperbolic map. He uses thermodynamic formalism on the finite state Markov shift to prove that the multifractal spectrum is everywhere analytic. Next, we discuss the work of Cawley and Mauldin \cite{cawley1992multifractal}. They consider a fractal constructed by taking an iterated function system on a self-similar set (which yields a self-similar measure). Falconer \cite{falconer2004fractal} gives more details on the construction of such a fractal. Cawley and Mauldin essentially model an iterated function system, based on countably many contractions, with a finite state Markov shift. Their methodology to prove that the multifractal spectrum is everywhere analytic involves geometric arguments. \par
Pesin and Weiss \cite{pesin1997multifractal} consider a uniformly expanding map and prove an expression for the multifractal spectrum. They use a combination of thermodynamic formalism and a covering argument to prove that the multifractal spectrum is analytic everywhere. Analysing the multifractal spectrum involves using thermodynamic formalism differently in the case of measures on the countable full shift $\Sigma.$ Iommi \cite{iommi2005multifractal} proves a general formula for the multifractal spectrum with respect to a measure on the countable shift and analyses when the multifractal spectrum is non-analytic. Hanus, Mauldin, and Urba\'nski also consider the multifractal spectrum in the setting of the countable conformal iterated function system modeled by a countable Markov shift. Complementary to Iommi's result about the multifractal spectrum's non-analyticity, their paper \cite{hanus2002thermodynamic} gives additional conditions to prove that the multifractal spectrum is analytic.
\par
Sarig has developed the thermodynamic formalism on countable Markov shifts $\Sigma.$ His work \cite{sarig2003existence} has established criteria for the existence of Gibbs and equilibrium states for potentials. This is significant because the existence of equilibrium states, which are important for topological pressure, is not guaranteed for potentials on countable Markov shifts. These measures are critical in our paper to prove results about the non-analyticity points of $f_{\mu}(\alpha)$ in the setting of a countable Markov shift. For a thorough discussion of Sarig's work, please refer to Sarig's survey \cite{sarig2015thermodynamic}. \par
We remark that Iommi and Jordan's work analyses the phase transitions of the pressure function and the multifractal spectrum in a similar setting to our paper. Their paper \cite{iommi2015multifractal}, assumes that the two potentials $\phi$ and $\psi$ are bounded and $\lim\limits_{x \rightarrow 0}\frac{\psi(x)}{\log|T'(x)|}=\infty$ (in contrast, we assume that the limit is equal to $1$). Their result is that the multifractal spectrum has $0$ to $2$ phase transitions. In the paper \cite{iommi2013phase}, they take $g$ to be a continuous function defined on the range of the suspension flow. Then, they prove that the map $t \mapsto \mathcal{P}(tg)$ has $0$ to $1$ phase transition when the roof function dominates the floor function. Using these results, Iommi and Jordan prove that the multifractal spectrum has $0$ to $2$ phase transitions in their paper \cite{iommi2013phase}. \par
Their paper \cite{iommi2010multifractal}, considers level sets of $[0,1]$ generated by Birkhoff averages and expanding interval maps that have countably many branches. This paper uses results from \cite{iommi2013phase} and \cite{iommi2015multifractal}. Their results include a variational characterisation of the multifractal spectrum and the existence of $0$ to $2$ phase transitions for $f_{\mu}(\alpha)$ when $\alpha_{\lim},$ which is a ratio involving the potentials $\phi$ and $\psi,$ equals $0.$ In contrast, our paper assumes that $0 < \alpha_{\lim} \le \infty$ or does not exist. Their paper also proves results on the multifractal analysis of suspension flows. Let $T$ be an expanding interval map and $g$ be a continuous function defined on the range of the suspension flow. Iommi and Jordan prove that the Birkhoff spectrum with respect to $g$ has two phase transitions if the roof function dominates the geometric potential $\log|T'|.$ \par
We briefly discuss different examples of the multifractal spectrum's phase transitions in other settings. Researchers have studied phase transitions for non-uniformly expanding interval maps that have neutral fixed points. They use thermodynamical formalism and they respectively prove explicit formulae for the multifractal spectrum. Olivier's paper \cite{olivier2000structure} considers a cookie cutter on $[0,1]$ and in turn, takes an induced map defined on a Cantor set generated by this cookie cutter. Nakaishi's paper \cite{nakaishi2000multifractal} considers piecewise interval maps on $[0,1]$ and an induced transformation generated by these maps. We note that Nakaishi's paper is related to a paper by Pollicott and Weiss \cite{pollicottmultifractal}. The multifractal spectrum with respect to Bernoulli convolutions for algebraic paramters has been analysed in an example in Feng and Olivier's paper \cite{feng2003multifractal}.
Now that we have given some background into multifractal analysis, we outline our paper's results and methodology to obtain these results. This paper is about multifractal analysis in the setting of the countable Markov shift $\Sigma$ and the interval $[0,1].$ In our analysis of $f_{\mu}(\alpha),$ we use thermodynamic formalism. We take the locally H\"older potential functions $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ with Gibbs measure $\mu$ and the metric potential $\psi:\Sigma \rightarrow \mathbb{R}^{+}$ such that $\psi=\log|G'|$ for some expanding map $G.$
We use results from Sarig about Gibbs states on countable Markov shifts to prove the existence of Gibbs states for potential functions related to $\phi$ and $\psi.$ These Gibbs states help us show that the multifractal spectrum has phase transitions. Proving the following theorem about the analyticity of the multifractal spectrum requires using results from Sarig, Mauldin, Urba\'nski, and Iommi. We define $\alpha_{\lim}$ as follows:
Let $\bar{i} \in \Sigma$ be such that $\bar{i}=(i,i,...).$ Take
\[\alpha_{\lim}:=\lim\limits_{i \rightarrow \infty}\frac{\phi(\bar{i})}{-\psi(\bar{i})}\]
if the limit exists. We also let $\alpha_{\inf}:=\inf\{d_{\mu}(x): x \in \Sigma\}$ and $\alpha_{\sup}:=\sup\{d_{\mu}(x): x \in \Sigma\}.$ We remark that we will introduce a different definition of $\alpha_{\inf}$ and $\alpha_{\sup}$ in our paper and then prove the equivalence of both definitions of $\alpha_{\inf}$ and $\alpha_{\sup}.$
\begin{theorem}\label{thm:M1}
Let $\phi: \Sigma \rightarrow \mathbb{R}^{-}$ be a potential with Gibbs measure $\mu$ such that $P(\phi)<\infty.$ and $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ be a metric potential. Assume that $\phi$ and $\psi$ are non-cohomologous locally H\"older potentials such that $\alpha_{\lim}<\infty.$
\begin{enumerate}
\item There exist intervals $A_{i}$ such that $f_{\mu}(\alpha)$ is analytic on each of their interiors.
\item The interval $(\alpha_{\inf}, \alpha_{\sup})=\cup_{i=1}^{j}A_{i}$ such that $j=\{1,2,3,4\}.$
\item The multifractal spectrum is concave on $(\alpha_{\inf}, \alpha_{\sup}),$ has its maximum at $\alpha(0),$ and has zero to three phase transitions.
\end{enumerate}
\end{theorem}
We prove a theorem about the behaviour of the multifractal spectrum when $\alpha_{\lim}$ is infinite. Since some of the earlier results about the analyticity of the mutlfractal spectrum are still true for $\alpha_{\lim}=\infty,$ we prove a theorem stating that the multifractal spectrum has $0$ to $1$ phase transition as provided below.
\begin{theorem}\label{thm:M2}
Let $\phi: \Sigma \rightarrow \mathbb{R}^{-}$ be a potential with Gibbs measure $\mu$ such that $\mathcal{P}(\phi)<\infty$ and $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ be a metric potential. Assume that $\phi$ and $\psi$ are non-cohomologous locally H\"older potentials such that $\alpha_{\lim}=\infty.$
\begin{enumerate}
\item There exist intervals $A_{i}$ such that $f_{\mu}(\alpha)$ is analytic on each of their interiors.
\item The interval $(\alpha_{\inf}, \alpha_{\sup})=\cup_{i=1}^{j}A_{i}$ such that $j=\{1,2\}.$
\item The multifractal spectrum is concave on $(\alpha_{\inf}, \alpha_{\sup}),$ is equal to its maximum $f_{\mu}(\alpha(0))$ on $(\alpha(0),\alpha_{\sup}),$ and has zero to one phase transition.
\end{enumerate}
\end{theorem}
Finally, we apply Theorem \ref{thm:M1} and Theorem \ref{thm:M2} to the Gauss map $G$ by defining our locally H\"older potential as $\psi(x)=\log|G'(\pi(x))|$ with respect to the coding map $\pi:\Sigma \rightarrow [0,1].$ We provide examples that apply Theorems \ref{thm:M1} and \ref{thm:M2} to show that the multifractal spectrum has up to three phase transitions. Also, we provide an example in which the multifractal spectrum has infinitely many phase transitions. In these examples, we will use locally H\"older potentials to estimate $\phi$ and $\psi.$ Now, we will define some notation and concepts from thermodynamic formalism.
\subsection*{Acknowledgements}\hspace*{\fill} \par
The author thanks Dr.\ Thomas Jordan for his direction and guidance in the writing of this paper and Dr.\ Godofredo Iommi for his helpful comments.
\section{Thermodynamic Formalism}
We now introduce some concepts from thermodynamic formalism.
\subsection{Introductory Definitions from Thermodynamic Formalism}\hspace*{\fill} \par
Before proving our result about the phase transitions of the multifractal spectrum, we introduce some definitions from thermodynamic formalism. Denote $S$ as our countable state space and $A=(a_{lm})_{SxS}$ as our transition matrix of zeroes and ones. We will be treating $S$ as $\mathbb{N}.$ Note that $A$ can be represented by a directed graph. We let $$\Sigma_{A}:=\{x \in S^{\mathbb{N}}: a_{x_{i}x_{i+1}}=1 \text{ for every } i \ge 1\}.$$
We take $\sigma: \Sigma_{A} \rightarrow \Sigma_{A}$ to be the standard left shift. We now define the topology for our countable state Markov shift $\Sigma_{A}.$
\begin{definition}
Given $x_{1}, ..., x_{n}$ symbols in $S,$ define a {\em cylinder set} in $\Sigma_{A}$ as
$$[x_{1},...,x_{n}]=\{y \in \Sigma_{A}: y_{i}=x_{i} \text{ for } 1 \le i \le n\}.$$
\end{definition}
These cylinder sets form the topology for $\Sigma_{A}.$ Two important assumptions for our countable Markov shift are defined below.
\begin{definition}
The shift space $\Sigma_{A}$ satisfies the {\em big images and pre-images property} if there is a finite set $\{c_{1},c_{2},...,c_{m}\}$ from our alphabet $S$ such that for each $d \in S,$ there are $i,j \in \{1,...,m\}$ such that
$$a_{c_{i}d}a_{dc_{j}}=1.$$
\end{definition}
\begin{definition}
$\sigma:\Sigma_{A} \rightarrow \Sigma_{A}$ is said to be {\em topologically mixing} if for all $a,b \in S,$ there exists $N_{ab} \in \mathbb{N}$ such that for all $n>N_{ab},$
$$[a] \cap \sigma^{-n}[b] \not= \emptyset.$$
\end{definition}
We provide the following remark. Since $\Sigma_{A}$ is a non-compact shift space, the existence of topological mixing alone for $\Sigma$ is not enough for a Gibbs measure to exist. For this reason, Sarig \cite{sarig2003existence} proves that the combination of topologically mixing and the BIP (big images and pre-images property) is sufficient and necessary for a Gibbs measure to exist. Denote $\Sigma_{A}$ as $\Sigma.$ We assume that $\Sigma$ satisfies the big images and pre-images property and it is topological mixing. Now, we must define a property for functions on $\Sigma.$
\begin{definition}
Let $\psi: \Sigma \rightarrow \mathbb{R}.$ The {\em$n$-th variation} of $\psi$ is given by
$$V_{n}(\psi):=\sup\limits_{[x_{1},...,x_{n}] \subset \Sigma_{A}}\sup\limits_{x,y \in [x_{1},...,x_{n}]}|\psi(x)-\psi(y)|.$$
\end{definition}
\begin{definition}
Let $\psi: \Sigma \rightarrow \mathbb{R}.$ $\psi$ is said to be {\em locally H\"older continuous} if there exists $C>0$ and $\theta \in (0,1)$ such that for each $n \in \mathbb{N},$
$$V_{n}(\psi) \le C\theta^{n}.$$
\end{definition}
Local H\"older continuity is important in the definition of our metric for $\Sigma.$ Consider the following metric $d$ on the countable shift $\Sigma.$ Take the sequences $x=(x_{1},x_{2},...), y=(y_{1},y_{2},...) \in \Sigma.$ Find the first common, starting subword in which $x$ and $y$ agree. \par
In other terms, take $x \wedge y=x_{1},...,x_{k}$ such that $k=\max\{m \in \mathbb{N}:x_{i}=y_{i} \text{ for all } 1 \le i \le m\}.$ For each $i \in \mathbb{N},$ let $0<r_{i}<1.$
Then, given $x$ and $y,$ we define our metric as
$$d(x,y)=r_{x_{1}} \cdots r_{x_{k}}.$$
We can define a potential $\psi$ connected to this metric. In many of our examples, we will define a potential $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ as a locally constant function:
\[\psi(x)=\log{r_{x_{1}}^{-1}}=-\log{r_{x_{1}}}.\]
Then,
\begin{equation}\label{eq:B}
0 \le \prod_{j=0}^{n-1}(\exp(\psi(\sigma^{j}(x))))^{-1}=(r_{x_{1}}^{-1}r_{x_{2}}^{-1}\cdots r_{x_{n}}^{-1})^{-1}=r_{x_{1}}r_{x_{2}} \cdots r_{x_{n}}.
\end{equation}
Given our metric,
\begin{equation}\label{eq:C}
|[x_{1},...,x_{n}]|=\sup\limits_{x,y \in [x_{1},...,x_{n}]}d(x,y)=r_{y_{1}}r_{y_{2}} \cdots r_{y_{n}}.
\end{equation}
Therefore, by Equations (\ref{eq:B}) and (\ref{eq:C}),
\[\prod_{j=0}^{n-1}(\exp(\psi(\sigma^{j}(x))))^{-1} =r_{x_{1}}r_{x_{2}} \cdots r_{x_{n}}.\]
In fact, the equation above relates the diameter of the cylinder set $[x_{1},...,x_{n}]$ to the potential $\psi.$ Potentials satisfying such an inequality are called metric potentials, which are defined as follows.
\begin{definition}
A positive, locally H\"older potential $\psi$ with the property that there exists a constant $C>0$ such that
$$\frac{1}{C}\prod_{j=0}^{n-1}(\exp(\psi(\sigma^{j}(y))))^{-1} \le |[x_{1},x_{2},...,x_{n}]| \le C\prod_{j=0}^{n-1}(\exp(\psi(\sigma^{j}(y))))^{-1},$$
for every $y=(x_{1},x_{2},...,x_{n},y_{n+1},...) \in [x_{1},...,x_{n}],$ is said to be a {\em metric potential}.
\end{definition}
Thus, $\psi(x)=\log{r_{x_{1}}^{-1}}$ is a metric potential with respect to the chosen metric $d$ on $\Sigma$ and the constant $C=1.$ We can define more general metric potentials on different metrics on $\Sigma.$ We remark that it is important for $\psi$ to be a metric potential defined by a hyperbolic map. For instance, take a general expanding map $T:[a,b] \rightarrow [a,b]$ for $b > a \ge 0$ and then, assuming that $\psi(x)=\log|T'(\pi(x))|,$ we find that
\[-C\exp\left(\sum_{j=0}^{n-1}-\log|T'(\pi(x))|\right) \le |[x_{1},x_{2},...,x_{n}]| \le C\exp\left(\sum_{j=0}^{n-1}-\log|T'(\pi(x))|\right).\]
Hence, given the expanding map $T$ and metric potential $\psi,$ we are able to relate the diameter of cylinder sets on $\Sigma$ to the radii of balls on an interval. Defining our metric by using our potential also gives an additional expression for the local dimension of sets in $\Sigma.$ Furthermore, we would be able to calculate the Hausdorff dimension of sets with local dimension $\alpha,$ so we must give the following definition.
\begin{definition}
Let $\psi: \Sigma \rightarrow \mathbb{R}$ be locally H\"older. We define the {\em topological pressure} as follows
$$\mathcal{P}(\psi)= \lim\limits_{n \rightarrow \infty}\frac{1}{n}\log\sum_{[x_{1},...,x_{n}] \subset \Sigma}\exp\sup\limits_{y \in [x_{1},...,x_{n}]}\left(\sum_{i=0}^{n-1}\psi(\sigma^{i}(y))\right).$$
\end{definition}
We also define another form of pressure.
\begin{definition}
Let $\psi: \Sigma \rightarrow \mathbb{R}$ be locally H\"older. We define the {\em Gurevich pressure} as follows
$$\mathcal{P}_{G}(\psi)= \lim\limits_{n \rightarrow \infty}\frac{1}{n}\log\sum_{\sigma^{n}x=x}\exp\left(\sum_{i=0}^{n-1}\psi(\sigma^{i}(x))\right)\mathbbm{1}_{[x_{1}]}$$
such that $\mathbbm{1}_{[x_{1}]}$ is the indicator function on the cylinder $[x_{1}].$
\end{definition}
Because $\Sigma$ is topologically mixing, the Gurevich pressure does not depend on $x_{1}$. Sarig proves that the Gurevich pressure is equivalent to topological pressure when $\Sigma$ is BIP. We provide this result by Sarig \cite{sarig1999thermodynamic} (Pg 1571, Theorem 3), Mauldin, and Urba\'nski \cite{mauldin2003graph} (Pg 11, Theorem 2.1.8) below.
\begin{proposition}
Let $\Sigma$ be topologically mixing and $\psi:\Sigma \rightarrow \mathbb{R}$ be locally H\"older such that $\sup\psi<\infty.$ Let $M_{\sigma}(\Sigma)$ be the set of $\sigma-$invariant measures. Then,
\[\mathcal{P}(\psi)=\sup_{\mu \in M_{\sigma}(\Sigma)}\left\{\int \phi\, \mathrm{d}\mu + h(\mu)\right\}=\mathcal{P}_{G}(\psi).\]
\end{proposition}
We remark that Iommi, Jordan, and Todd \cite{iommijordantodd} (Pg 8, Theorem 2.10) proved that $\sup\psi<\infty$ is an unnecessary condition for the previous proposition. We will provide a result that gives another way to calculate the topological pressure of a potential. Let $\psi$ and $\psi$ be locally H\"older. We can approximate the topological pressure of a potential $q\phi-t\psi$ (with $q,t \in \mathbb{R}$) on the shift $\Sigma$ by restricting the potential to a compact, invariant subset $K \subset \Sigma.$ For such a set $K,$ denote
$$\mathcal{P}_{K}(q\phi-t\psi):=P((q\phi-t\psi)|_{K})$$
as the restriction of the topological pressure of $q\phi-t\psi$ to $K.$ Sarig \cite{sarig1999thermodynamic} (Pg 1570, Theorem 2), Mauldin, and Urba\'nski \cite{mauldin2003graph} (Pg 8, Theorem 2.1.5) have proven the following proposition.
\begin{proposition}
Let $\phi$ and $\psi$ be locally H\"older. If $\mathcal{K}:=\{K \subset \Sigma: K \text{ compact and }\\
\sigma\text{-invariant, } K \not= \emptyset\},$ then
$$\mathcal{P}(q\phi-t\psi)=\sup\limits_{K \in \mathcal{K}}\mathcal{P}_{K}(q\phi-t\psi).$$
\end{proposition}
Compact subsets of our countable Markov shift $\Sigma$ include finite state Markov shifts $\Sigma_{n}.$ We will later use a nested sequence of finite state full shifts to approximate the topological pressure of $q\phi-t\psi$ on $\mathbb{N}^{\mathbb{N}}.$
\begin{definition}
A probability measure $\mu$ is said to be a {\em Gibbs measure} for the potential $\phi: \Sigma \rightarrow \mathbb{R}$ if there exist two constants $M$ and $P$ such that, for each cylinder $[x_{1},x_{2},...,x_{m}]$ and every $x \in [x_{1},x_{2},...,x_{m}],$
$$\frac{1}{M} \le \frac{\mu([x_{1},x_{2},...,x_{m}])}{\exp(-mP+\sum_{j=0}^{m-1}\phi(\sigma^{j}(x)))} \le M.$$
\end{definition}
In fact, Mauldin and Urba\'nski \cite{mauldin2003graph} (Pg 13, Proposition 2.2.2) proved that $P=\mathcal{P}(\phi).$ Now, we will define the locally H\"older potentials needed for our analysis of the multifractal spectrum.
\subsection{The Potentials $\phi$ and $\psi$}\hspace*{\fill} \par
Throughout this paper, let $\phi: \Sigma \rightarrow \mathbb{R}^{-}$ be a locally H\"older potential such that $0 \le \mathcal{P}(\phi)<\infty.$ We assume that $\mu$ is the Gibbs measure for $\phi.$ We will show the existence of $\mu$ later. Also, let $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ be a locally H\"older, metric potential with respect to an appropriate metric.
\begin{definition}
Two functions $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ and $\psi:\Sigma \rightarrow \mathbb{R}^{+}$ are {\em cohomologous} in a class $\mathcal{H}$ if there exists a function $u: \Sigma \rightarrow \mathbb{R}$ in the class $\mathcal{H}$ such that
$$\phi-\psi=u-u\circ{\sigma}.$$
\end{definition}
We assume that $\phi$ and $\psi$ are non-cohomologous to each other. For brevity, we will instead state that $\phi$ and $\psi$ are non-cohomologous. For $q, t \in \mathbb{R},$ consider the family of potentials $q\phi-t\psi.$ Given this family of potentials, we will analyse the behaviour and phase transitions of a function $T(q)$ dependent on $q.$ This analysis is needed for our result about the multifractal spectrum's phase transitions.
\subsection{The Functions $T(q)$ and $\tilde{t}(q)$ and the Limit $\alpha_{\lim}$}\hspace*{\fill} \par
We will later prove that the multifractal spectrum's phase transitions are closely related to the behaviour of a family of potentials $q\phi-t\psi.$ To follow this argument, we must define the following function.
\begin{definition}
For each $q \in \mathbb{R},$ the {\em temperature function} $T(q)$ is defined as
$$T(q):= \inf\{t \in \mathbb{R}:\mathcal{P}(q\phi-t\psi) \le 0\}$$
\end{definition}
Proposition 4.3 on Pg 1892 of Iommi \cite{iommi2005multifractal} states the following important result.
\begin{proposition}\label{prop:CD1}
$T(q)$ is a convex and decreasing function.
\end{proposition}
We also define a function $\tilde{t}(q)$ which is similar to $T(q).$
\begin{definition}
For each $q \in \mathbb{R},$ the {\em function} $\tilde{t}(q)$ is defined as
$$\tilde{t}(q):=\inf\{t \in \mathbb{R}:\mathcal{P}(q\phi-t\psi) < \infty\}.$$
\end{definition}
We will explain the significance of the family of potentials $q\phi-\tilde{t}(q)\psi$ later. We need the following limit to get an expression for $\tilde{t}(q).$
\begin{definition}\label{def:lim}
Let $\bar{i} \in \Sigma$ be such that $\bar{i}=(i,i,...).$ Let
\[\alpha_{\lim}=\lim\limits_{i \rightarrow \infty}\frac{\phi(\bar{i})}{-\psi(\bar{i})}\]
if the limit exists.
\end{definition}
Assume that
\[t_{\infty}:=\inf\{t \in \mathbb{R}: \mathcal{P}(-t\psi)<\infty\}\]
exists and it is finite. When $\alpha_{\lim}$ exists, we will prove that
$$\tilde{t}(q)=-\alpha_{\lim}q+t_{\infty}.$$
\begin{definition}
Let $f:\Sigma \rightarrow \mathbb{R}$ be a potential. We define the \text{\em nth partition function} to be $$Z_{n}(f)=\sum_{[x_{1},...,x_{n}] \subset \Sigma}\exp\sup\limits_{y \in [x_{1},...,x_{n}]}\left(\sum_{i=0}^{n-1}f\left(\sigma^{i}(y)\right)\right).$$
\end{definition}
The following lemma is a modified version of Proposition 2.1.9 on Pg 11 in Mauldin and Urba\'nski \cite{mauldin2003graph}.
\begin{lemma}\label{lemma:MU1}
If $\Sigma$ has the BIP property and $f:\Sigma \rightarrow \mathbb{R}$ is locally H\"older, then $\mathcal{P}(f)<\infty$ if and only if $Z_{1}(f)<\infty.$
\end{lemma}
We use this lemma in the proof of the following proposition.
\begin{proposition}\label{prop:L1}
Let $\phi: \Sigma \rightarrow \mathbb{R}^{-}$ and $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ be non-cohomologous locally H\"older potentials. Assume that $\alpha_{\lim}<\infty.$ Let $t_{\infty}=\inf\{t \in \mathbb{R}:\mathcal{P}(-t\psi)<\infty\}$ be finite. Then,
$$\tilde{t}(q)=-\alpha_{\lim}q+t_{\infty}.$$
\end{proposition}
\begin{proof}
Take $N \in \mathbb{N}$ large and an arbitrary $i \in \mathbb{N}.$ Consider the one periodic sequences $\bar{n}=(n,n,...)$ for $n>N$ and $\bar{i}=(i,i,...).$ Given $\phi$ and $\psi$ locally H\"older potentials, we obtain the following estimate if $x \in [i]:$
\begin{equation}\label{eq:Variation}
\phi(\bar{i}) \le \phi(x)+V_{1}(\phi) \text{ and }
\psi(\bar{i}) \le \psi(x)+V_{1}(\psi)
\end{equation}
such that $V_{1}(\phi), V_{1}(\psi) \ge 0.$ Since $\Sigma$ has the BIP property and $\phi$ and $\psi$ are locally H\"older, Lemma~\ref{lemma:MU1} gives us that
$$Z_{1}(q\phi-t\psi)<\infty \text{ if and only if } \mathcal{P}(q\phi-t\psi)<\infty.$$
We have that for every $x=(x_{1},x_{2},...) \in \Sigma,$
\[Z_{1}(q\phi-t\psi)=\sum_{x_{1}=1}^{\infty}\exp\sup\limits_{x\in[x_{1}]}(q\phi-t\psi)(x).\]
By Equation (\ref{eq:Variation}), we get that for each $\bar{i}=(i,i,...)$ and $x=(i,x_{2},...),$
\[\sup\limits_{x\in[i]}(q\phi-t\psi)(x) \le (q\phi-t\psi)(\bar{i})+V_{1}(q\phi-t\psi).\]
Hence, our calculations for $\bar{i}$ above give us that
$$Z_{1}(q\phi-t\psi) \le \sum_{i=1}^{\infty}\exp((q\phi-t\psi)(\bar{i}))+V_{1}(q\phi-t\psi).$$
Fix an arbitrary $q \in \mathbb{R}.$ Thus, to prove that $Z_{1}(q\phi-t\psi)<\infty$ for some $t \in \mathbb{R},$ it suffices to prove that
\begin{equation}\label{eq:3}
\sum_{i=1}^{\infty}\exp((q\phi-t\psi)(\bar{i}))<\infty.
\end{equation}
Let $\varepsilon>0$ and take our one periodic sequence $\bar{n}$ introduced earlier. By definition~\ref{def:lim}, we immediately get that
\begin{equation}\label{eq:4}
\frac{\phi(\bar{n})}{-\psi(\bar{n})}-\varepsilon \le \alpha_{\lim} \le \frac{\phi(\bar{n})}{-\psi(\bar{n})}+\varepsilon.
\end{equation}
If $\mathcal{P}(-t\psi)<\infty,$ then $Z_{1}(-t\psi) \le \sum_{i=1}^{\infty}\exp((-t\psi)(i)+V_{1}(-t\psi))<\infty$
by our estimates above and Lemma~\ref{lemma:MU1}.
Furthermore, $\sum_{n=N}^{\infty}\exp((-t\psi)(\bar{n}))<\infty$ if and only if $\sum_{i=1}^{\infty}\exp((-t\psi)(\bar{i}))<\infty.$ To prove Inequality~\ref{eq:3} for $\bar{i},$ we will show that \[\sum_{n=1}^{\infty}\exp((q\phi-t\psi)(\bar{n}))<\infty\]
for our fixed $q \in \mathbb{R}$ and some $t \in \mathbb{R}.$
When $t=t_{\infty}-q\alpha_{\lim}$ and $\varepsilon \rightarrow 0,$
\begin{eqnarray}
\sum_{n=N}^{\infty}\exp((q\phi-t\psi)(\bar{n})) &=& \sum_{n=N}^{\infty}\exp((q\phi+-(t_{\infty}+q(\frac{\phi(\bar{n})}{\psi(\bar{n})}+\varepsilon)\psi)(\bar{n})) \nonumber\\
&=&\sum_{n=N}^{\infty}\exp((-t_{\infty}-q\varepsilon)\psi(\bar{n}))<\infty \nonumber
\end{eqnarray}
by definition of $t_{\infty}.$ Let $t=t_{\infty}-q\frac{\phi(\bar{n})}{-\psi(\bar{n})}+K$ such that $K \in \mathbb{R}.$
Then, if $K>0$ and $\varepsilon \rightarrow 0,$
$$\sum_{n=N}^{\infty}\exp((q\phi-t\psi)(\bar{n}))=\sum_{n=N}^{\infty}\exp(-(t_{\infty}+K)\psi(\bar{n}))<\infty$$
and if $K<0,$
$$\sum_{n=N}^{\infty}\exp((q\phi-t\psi)(\bar{n}))=\sum_{n=N}^{\infty}\exp(-(t_{\infty}+K)\psi(\bar{n}))=\infty$$
by definition of $t_{\infty}.$ \par
Thus,
\[\inf\{t \in \mathbb{R}: \sum_{n=1}^{\infty}\exp((q\phi-t\psi)(\bar{n}))<\infty\}=t_{\infty}-q\alpha_{\lim}.\]
Therefore, for each $q \in \mathbb{R},$
$$\inf\{t \in \mathbb{R}:\mathcal{P}(q\phi-t\psi)<\infty\}=-\alpha_{\lim}q+t_{\infty}$$
by Equation~\ref{eq:4}. By definition, it follows that
\[\tilde{t}(q)=-\alpha_{\lim}q+t_{\infty}.\]
It follows that $\tilde{t}(q)$ is a decreasing line because $\alpha_{\lim}>0.$
\end{proof}
The function $\tilde{t}(q)$ is connected to the following set $Q.$
\subsection{The Set $Q$ and The Function $\alpha(q)$}\hspace*{\fill} \par
Let $Q=\{q \in \mathbb{R}:T(q)=\tilde{t}(q)\}.$ Since $T(q)$ is strictly convex and $\tilde{t}(q)$ is linear, the following proposition is immediate.
\begin{proposition}\label{prop:Q1}
$Q$ can be a closed interval, half-open infinite interval, a point, or the empty set.
\end{proposition}
The following proposition immediately follows from Propositions \ref{prop:CD1}, \ref{prop:L1}, and \ref{prop:Q1}.
\begin{proposition}\label{prop:AAA1}
Let $\phi: \Sigma \rightarrow \mathbb{R}^{-}$ and $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ be non-cohomologous locally H\"older potentials. $T(q)$ has at most two phase transitions.
\end{proposition}
We remark that Hanus, Mauldin, and Urba\'nski \cite{hanus2002thermodynamic} considered the families of potentials $f^{(i)},$ which is strongly H\"older, and $\log|\phi_{i}^{'}|$, which is defined by a regular, conformal iterated function system $\{\phi_{i}\}$ that satisfy the open set condition (both the terms regular conformal iterated function system and the open set condition are defined in Chapter 4 of \cite{mauldin2003graph}). Their potentials give that $Q=\emptyset.$ Hence, the multifractal spectrum was analytic in their case. An important function used in multifractal analysis is $\alpha(q),$ which is connected to $Q.$ By Proposition~\ref{prop:Q1}, $Q=[q_{0}, q_{1}]$ for some $q_{0},q_{1} \in \mathbb{R} \cup \{-\infty,\infty\}.$
\begin{definition}
Let $Q=[q_{0},q_{1}]$ for some $q_{0},q_{1} \in \mathbb{R}.$ Then, we have a function $\alpha(q)$ such that
$$\alpha(q) =
\left\{
\begin{array}{ll}
-T'(q) & \mbox{if } q \in Q^{\complement} \\
\alpha_{\lim}& \mbox{if } q \in (q_{0},q_{1}) \\
\alpha^{-}=\lim\limits_{q \rightarrow q_{0}^{-}}\alpha(q)& \mbox{if } q=q_{0}>-\infty \\
\alpha^{+}=\lim\limits_{q \rightarrow q_{1}^{+}}\alpha(q)& \mbox{if } q=q_{1}<\infty. \\
\end{array}
\right.$$
If $q_{0}=-\infty,$ $\alpha^{-}=\alpha(q_{0})=\alpha_{\lim}$ and if $q_{1}=\infty,$ $\alpha^{+}=\alpha(q_{1})=\alpha_{\lim}.$
\end{definition}
If $Q$ is a singleton, then our preceding definition applies for $q_{0}=q_{1}<\infty.$ In that case, $\alpha^{-}=\alpha^{+}=\alpha(q_{0}).$ Our analysis of $\alpha(q)$ also depends on its extreme values. We remark that the extreme values of $\alpha(q)$ are the supremum and infimum of the possible local dimension, which we will define later, of any sequence in $\Sigma.$
\begin{definition}
We let
$\alpha_{\inf}:=\inf\limits_{q \in \mathbb{R}}\alpha(q)\text{ and } \alpha_{\sup}:=\sup\limits_{q \in \mathbb{R}}\alpha(q).$
\end{definition}
We prove the following formula connecting $\alpha_{\inf}$ and $\alpha_{\sup}$ to a ratio involving $\phi$ and $\psi.$
\begin{lemma}
Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ and $\psi:\Sigma \rightarrow \mathbb{R}^{+}$ be locally H\"older and $\Sigma$ have the BIP property. Then,
\[\alpha_{\sup}=\sup\limits_{\nu \in M(\Sigma, \sigma)}\frac{\int \phi\, \mathrm{d}\nu}{-\int \psi\, \mathrm{d}\nu} \text{ and } \alpha_{\inf}=\inf\limits_{\nu \in M(\Sigma, \sigma)}\frac{\int \phi\, \mathrm{d}\nu}{-\int \psi\, \mathrm{d}\nu}.\]
\end{lemma}
\begin{proof}
By the variational principle,
\[\mathcal{P}(q\phi-T(q)\psi)=\sup_{\nu \in M(\sigma,\sigma)}\{q\int \phi\, \mathrm{d}\nu -T(q)\int \psi\, \mathrm{d}\nu + h(\nu)\} \le 0.\]
Because entropy is non-negative, we get that
\[\sup_{\nu \in M(\sigma,\sigma)}\{q\int \phi\, \mathrm{d}\nu -T(q)\int \psi\, \mathrm{d}\nu\} \le 0.\]
Then, for any $\nu \in M(\Sigma, \sigma),$
\[q\int \phi\, \mathrm{d}\nu-T(q)\int \psi\, \mathrm{d}\nu \le 0.\]
With a bit of rearrangement,
\[\frac{q\int \phi\, \mathrm{d}\nu}{\int \psi \mathrm{d}\nu} \le T(q).\]
Taking the derivatives of both sides with respect to $q,$ we get that
\[\frac{\int \phi\, \mathrm{d}\nu}{-\int \psi\, \mathrm{d}\nu} \ge -T'(q).\]
It follows that
\[\sup\limits_{\nu \in M(\Sigma, \sigma)}\frac{\int \phi\, \mathrm{d}\nu}{-\int \psi\, \mathrm{d}\nu} \ge -T'(q).\]
We remark that
\[\sup\limits_{\nu \in M(\Sigma, \sigma)}\frac{\int \phi\, \mathrm{d}\nu}{-\int \psi\, \mathrm{d}\nu} \ge \alpha_{\lim}\] by definition of $\alpha_{\lim}$ and the approximation of locally H\"older potentials with locally constant functions.
Therefore,
\[\sup\limits_{\nu \in M(\Sigma, \sigma)}\frac{\int \phi\, \mathrm{d}\nu}{-\int \psi\, \mathrm{d}\nu} = \alpha_{\sup}.\]
Using a similar argument, the result for the infimum of $\alpha(q)$ also follows:
\[\inf\limits_{\nu \in M(\Sigma, \sigma)}\frac{\int \phi\, \mathrm{d}\nu}{-\int \psi\, \mathrm{d}\nu} = \alpha_{\inf}.\]
\end{proof}
Note that $\alpha_{\lim} \in [\alpha_{\inf},\alpha_{\sup}]$ if it exists. Since $Q=[q_{0},q_{1}]$ for some $q_{0},q_{1} \in \mathbb{R},$ we immediately get the following decomposition from the definition of $\alpha(q):$
\[(\alpha_{\inf},\alpha_{\sup})=\{\alpha(q):q\in Q^{\complement}\} \cup (\alpha^{+},\alpha_{\lim}) \cup \{\alpha_{\lim}\} \cup (\alpha_{\lim},\alpha^{-}) \cup \{\alpha^{-}\} \cup \{\alpha^{+}\}.\]
The function $\alpha(q)$ is always positive for every $q \in \mathbb{R}$ because Iommi \cite{iommi2005multifractal} (Pg 1892, Proposition 4.3) proved that $T(q)$ is a decreasing function of $q.$ We will soon notice that the multifractal spectrum is connected to the functions $T(q)$ and $\alpha(q).$
\subsection{The Multifractal Spectrum}\hspace*{\fill} \par
Before defining the multifractal spectrum we define symbolic dimension and the set $X_{\alpha}^{s}.$ We use the Gibbs measure $\mu$ for $\phi.$
\begin{definition}
A word $x=(x_{1},x_{2},...,x_{m},...) \in \Sigma$ has {\em symbolic dimension} $\alpha$ provided that
\[
d_{\mu}(x):=\lim\limits_{m \rightarrow \infty}\frac{\log\mu([x_{1},x_{2},...,x_{m}])}{\log|[x_{1},x_{2},...,x_{m}]|}=\alpha
\]
such that $\alpha_{\inf} \le \alpha \le \alpha_{\sup}.$
\end{definition}
We now consider sets with local dimension $\alpha$ as follows.
\begin{definition}
For each fixed $\alpha \in [\alpha_{\inf}, \alpha_{\sup}],$ we have the set
$$X_{\alpha}^{s}=\{x \in \Sigma: \lim\limits_{m \rightarrow \infty}\frac{\log\mu([x_{1},x_{2},...,x_{m}])}{\log|[x_{1},x_{2},...,x_{m}]|}=\alpha \}.$$
\end{definition}
Similarly, the local dimension of $x \in \Sigma$ is the limit
\[\lim\limits_{r \rightarrow 0}\frac{\log\mu(B(x,r))}{\log{r}}.\] We now prove a proposition that the pointwise and symbolic dimensions are equal a.e. $x \in \Sigma.$
\begin{proposition}\label{prop:AAA}
Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ be locally H\"older with Gibbs state $\mu$ and let $\mathcal{P}(\phi)<\infty.$
Take $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ as a locally H\"older metric potential. Assume that $\mu_{q}$ is the Gibbs state for $q\phi-T(q)\psi$ and $\phi,\psi \in \mathcal{L}^{1}(\mu_{q}).$ Then, for $\mu_{q}-$a.e. $x \in \Sigma,$
\begin{equation}\label{eq:Symbolic}
\frac{\int \phi \mathrm{d}\mu_{q}}{-\int \psi \mathrm{d}\mu_{q}}= \lim\limits_{m \rightarrow \infty}\frac{\log\mu([x_{1},x_{2},...,x_{m}])}{\log|[x_{1},x_{2},...,x_{m}]|}=\alpha(q).
\end{equation}
Furthermore, the pointwise and local dimension are equal a.e. $\Sigma.$
\end{proposition}
\begin{proof}
Consider the set
\[
\bar{X}:=\{x \in \Sigma: \lim\limits_{m \rightarrow \infty}\frac{\psi(\sigma^{m}(x))}{\sum_{n=0}^{m-1}\psi(\sigma^{m}(x))}=0\}=\{x \in \Sigma: \lim\limits_{m \rightarrow \infty}\frac{\sum_{n=0}^{m}\psi(\sigma^{n}(x))}{\sum_{n=0}^{m-1}\psi(\sigma^{n}(x))}=1\}.
\]
We will prove that $\mu(\bar{X})=1.$
First, we show that for any $x \in \bar{X},$
\[\lim\limits_{m \rightarrow \infty}\frac{\log\mu([x_{1},x_{2},...,x_{m}])}{\log|[x_{1},x_{2},...,x_{m}]|}=\lim\limits_{r \rightarrow 0}\frac{\log\mu(B(x,r))}{\log{r}}.\]
For each $x \in \bar{X},$ we can find $m \in \mathbb{N}$ large and $r>0$ such that the following construction holds. If $|[x_{1},...,x_{m},x_{m+1}]| \le r \le |[x_{1},...,x_{m}]|,$ we find that $\mu(B(x,r))=\mu([x_{1},x_{2},...,x_{m+1}])$ for some $m \in \mathbb{N}.$ We get the following inequality:
\[\lim\limits_{m \rightarrow \infty}\frac{\log\mu([x_{1},..,x_{m},x_{m+1}])}{\log|[x_{1},...,x_{m}]|} \le \lim\limits_{r \rightarrow 0}\frac{\log\mu(B(x,r))}{\log{r}} \le \lim\limits_{m \rightarrow \infty}\frac{\log\mu([x_{1},....,x_{m},x_{m+1}])}{\log|[x_{1},...,x_{m},x_{m+1}]|}\]
Hence, we will prove that
\[\lim\limits_{m \rightarrow \infty}\frac{\log\mu([x_{1},..,x_{m},x_{m+1}])}{\log|[x_{1},...,x_{m}]|}=\lim\limits_{m \rightarrow \infty}\frac{\log\mu([x_{1},....,x_{m},x_{m+1}])}{\log|[x_{1},...,x_{m},x_{m+1}]|}.\]
These limits are equal if
\[\lim\limits_{m \rightarrow \infty} \frac{\log|[x_{1},...,x_{m}]|}{\log|[x_{1},...,x_{m},x_{m+1}]|}=1.\]
For each $x \in \bar{X},$ there exists $C>0$ such that
\[\frac{1}{C} \le \frac{|[x_{1},...,x_{m}]|}{\prod_{n=0}^{m-1}(\exp(\psi(\sigma^{n}(x))))^{-1}} \le C \text{ and } \frac{1}{C} \le \frac{|[x_{1},...,x_{m},x_{m+1}]|}{\prod_{n=0}^{m}(\exp(\psi(\sigma^{n}(x))))^{-1}} \le C\]
because $\psi$ is a metric potential.
Then,
\[-\log{C} \le \log\left(\frac{|[x_{1},...,x_{m}]|}{\prod_{n=0}^{m-1}(\exp(\psi(\sigma^{n}(x))))^{-1}}\right) \le \log{C}\]
and
\[-\log{C} \le \log\left(\frac{|[x_{1},...,x_{m},x_{m+1}]|}{\prod_{n=0}^{m}(\exp(\psi(\sigma^{n}(x))))^{-1}}\right) \le \log{C}.\]
Hence, we have that
\[\log\left(\frac{|[x_{1},...,x_{m}]|}{\prod_{n=0}^{m-1}(\exp(\psi(\sigma^{n}(x))))^{-1}}\right)-\log\left(\frac{|[x_{1},...,x_{m+1}]|}{\prod_{n=0}^{m}(\exp(\psi(\sigma^{n}(x))))^{-1}}\right)=0.\]
This gives us that
\[\log\left(\frac{|[x_{1},...,x_{m}]|}{|[x_{1},...,x_{m},x_{m+1}]|}\right)-\log\left(\frac{\prod_{n=0}^{m}\exp(\psi(\sigma^{n}(x)))}{\prod_{n=0}^{m-1}\exp(\psi(\sigma^{n}(x)))}\right)=0,\]
so
\[\lim\limits_{m \rightarrow \infty}\log\left(\frac{|[x_{1},...,x_{m}]|}{|[x_{1},...,x_{m},x_{m+1}]|}\right)-\log\left(\frac{\prod_{n=0}^{m}\exp(\psi(\sigma^{n}(x)))}{\prod_{n=0}^{m-1}\exp(\psi(\sigma^{n}(x)))}\right)=0.\]
We find that
\[\lim\limits_{m \rightarrow \infty}\log\left(\frac{\prod_{n=0}^{m}\exp(\psi(\sigma^{n}(x)))}{\prod_{n=0}^{m-1}\exp(\psi(\sigma^{n}(x)))}\right)=\lim\limits_{m \rightarrow \infty}\sum_{n=0}^{m}\exp(\psi(\sigma^{n}(x)))-\sum_{n=0}^{m-1}\exp(\psi(\sigma^{n}(x)))=0\]
because $x \in \bar{X}.$
It follows that
\[\lim\limits_{m \rightarrow \infty} \frac{\log|[x_{1},...,x_{m}]|}{\log|[x_{1},...,x_{m},x_{m+1}]|}=1.\]
Hence, we find that
\[\lim\limits_{m \rightarrow \infty}\frac{\log\mu([x_{1},..,x_{m},x_{m+1}])}{\log|[x_{1},...,x_{m}]|} = \lim\limits_{r \rightarrow 0}\frac{\log\mu(B(x,r))}{\log{r}} = \lim\limits_{m \rightarrow \infty}\frac{\log\mu([x_{1},....,x_{m},x_{m+1}])}{\log|[x_{1},...,x_{m},x_{m+1}]|}.\]
Therefore, the pointwise and local dimension for each $x \in \bar{X}$ are equal:
\[\lim\limits_{r \rightarrow 0}\frac{\log\mu(B(x,r))}{\log{r}} = \lim\limits_{m \rightarrow \infty}\frac{\log\mu([x_{1},....,x_{m}])}{\log|[x_{1},...,x_{m}]|}.\]
Additionally, we find that $x \in \bar{X}$ satisfy
\[\sum_{k=m}^{\infty}\psi(\sigma^{k}(x))<\infty, \text{ so } \sum_{k=1}^{\infty}\psi(\sigma^{k}(x))<\infty\]
for large $m \in \mathbb{N}.$
Since $\mu$ is Gibbs for $\phi,$ $\psi$ is a metric potential, and the local and pointwise dimension for $x \in \bar{X}$ are equal, we immediately find that
\[\lim\limits_{r \rightarrow 0}\frac{\log\mu(B(x,r))}{\log{r}}=\lim\limits_{m \rightarrow \infty}\frac{\log\mu([x_{1},....,x_{m}])}{\log|[x_{1},...,x_{m}]|}=\lim\limits_{m \rightarrow \infty}\frac{\sum_{j=0}^{m-1}\phi(\sigma^{j}(x))}{-\sum_{j=0}^{m-1}\psi(\sigma^{j}(x))}=\frac{\int \phi \mathrm{d}\mu_{q}}{-\int \psi \mathrm{d}\mu_{q}}\]
for each $x \in \bar{X}$ by the Birkhoff ergodic theorem. Note that Iommi \cite{iommi2005multifractal} proved this result for a.e. $x \in \Sigma.$
It immediately follows that $\mu(\bar{X})=1.$ Hence, the pointwise and local dimension are equal a.e.
\end{proof}
We give the following result as an alternate characterisation of $\alpha_{\inf}$ and $\alpha_{\sup}$ because Proposition \ref{prop:AAA} can be applied to all $q \in \mathbb{R}.$
\begin{lemma}
Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ be locally H\"older with Gibbs state $\mu$ and let $\mathcal{P}(\phi)<\infty.$
Take $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ as a locally H\"older metric potential. Assume that $\mu_{q}$ is the Gibbs state for $q\phi-T(q)\psi$ and $\phi,\psi \in \mathcal{L}^{1}(\mu_{q}).$ Then,
\[\alpha_{\inf}=\inf\{d_{\mu}(x): x \in \Sigma\} \text{ and } \alpha_{\sup}=\sup\{d_{\mu}(x): x \in \Sigma\}.\]
\end{lemma}
\begin{proof}
The result follows from Proposition \ref{prop:AAA} and the definition of $\alpha_{\inf}$ and $\alpha_{\sup}.$
\end{proof}
As Proposition \ref{prop:AAA} infers, we can consider the set
\[X_{\alpha}:=\{x \in \Sigma: \lim\limits_{r \rightarrow 0}\frac{\log\mu(B(x,r))}{\log{r}}=\alpha\}.\] Since $X_{\alpha}=X_{\alpha}^{s}$ a.e., we use $X_{\alpha}^{s}$ in our definition of the multifractal spectrum.
\begin{definition}
For each $\alpha \in (\alpha_{\inf}, \alpha_{\sup}),$ the {\em multifractal spectrum} is the function $f_{\mu}(\alpha)$ defined by
\begin{equation}\label{eq:E}
\alpha \mapsto \dim_{H}(X_{\alpha}^{s}).
\end{equation}
\end{definition}
We remark that the multifractal spectrum depends on the measure $\mu.$ Iommi's work \cite{iommi2005multifractal} has a theorem stating that the multifractal spectrum is a Legendre transform. Hence, we will define the concepts of Fenchel and Legendre transforms.
\begin{definition}
Let $h$ be a convex function. $(h,g)$ is called a {\em Fenchel pair} if
\[g(p)=\sup\limits_{x}\{px-h(x)\}.\]
Alternatively, we say that $g$ is the {\em Fenchel transform} of $h.$ If $h$ is a convex, twice-differentiable function, then $g$ is called a {\em Legendre transform}.
\end{definition}
We provide Iommi's theorem from Pg 11, Theorem 4.1 of \cite{iommi2005multifractal}, which proves that the multifractal spectrum is a Legendre transform.
\begin{theorem}
Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ be locally H\"older with Gibbs state $\mu$ and let $\mathcal{P}(\phi)<\infty.$
Take $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ as a locally H\"older metric potential. The multifractal spectrum $f_{\mu}$ is the Fenchel transform of $T.$
\end{theorem}
We provide the following remark. Iommi proves that $T(q)$ is a convex function. If we take $x=q,$ $h(q)=T(q),$ $h'(q)=-\alpha(q),$ and $p=-\alpha,$
\[f_{\mu}(\alpha)=g(p)=\sup\limits_{x \in \mathbb{R}}\{px-h(x)\}=\sup\limits_{q \in \mathbb{R}}\{-q\alpha-T(q)\}=\inf\limits_{q\in\mathbb{R}}\{T(q)+q\alpha\}.\]
Hence, $(T,f_{\mu})$ form a Fenchel pair and $f_{\mu}(\alpha)$ is a Lengendre transform. Thus, we will use the following form of Iommi's theorem to prove Theorem \ref{thm:M1}.
\begin{theorem}\label{thm:I1}
Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ be locally H\"older with Gibbs state $\mu$ and let $\mathcal{P}(\phi)<\infty.$
Take $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ as a locally H\"older metric potential. For each $\alpha \in (\alpha_{\inf}, \alpha_{\sup}),$
\[f_{\mu}(\alpha)=\inf\limits_{q \in \mathbb{R}}\{T(q)+q\alpha\}=\dim_{H}(X_{\alpha}^{s}).\]
\end{theorem}
The following proposition is immediate.
\begin{proposition}\label{prop:AA1}
Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ be locally H\"older with Gibbs state $\mu$ and let $\mathcal{P}(\phi)<\infty.$
Take $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ as a locally H\"older metric potential. Furthermore, assume that $\phi$ and $\psi$ are non-cohomologous. If $T(q)$ is analytic over $\mathbb{R},$ then $f_{\mu}(\alpha(q))$ is analytic over $(\alpha_{\inf}, \alpha_{\sup}).$
\end{proposition}
Our paper is on the multifractal spectrum's phase transitions; however, we gave the preceding proposition for completeness. In order to analyse the phase transitions of the multifractal spectrum, we must analyse the phase transitions of $T(q).$ To do this, we must give criteria for the existence of special measures for our potential.
\subsection{Measures for Our Potentials}\hspace*{\fill} \par
In this section, we provide criteria for the existence and uniqueness of special measures for $\phi$ and $q\phi-T(q)\psi.$ We have the following (modified) result by Sarig \cite{sarig2003existence} (Pg 2, Theorem 1) and Mauldin and Urba\'nski \cite{mauldin2003graph} (Pg 14, Theorem 2.2.4).
\begin{theorem}\label{thm:S2}
Let $(\Sigma,\sigma)$ be topologically mixing and $\gamma: \Sigma \rightarrow \mathbb{R}$ is locally H\"older. Then, $\gamma$ has a unique invariant Gibbs state if and only if the transition matrix $A$ has the BIP property and $\mathcal{P}(\gamma)<\infty.$
\end{theorem}
Thus, by Sarig, Mauldin, and Urba\'nski, $\phi$ has a corresponding unique Gibbs measure $\mu.$ We now consider the family of potentials $q\phi-T(q)\psi.$
\begin{theorem}\label{thm:MR1}
Assume that $\Sigma$ satisfies the BIP property. Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ be locally H\"older with Gibbs state $\mu$ and let $\mathcal{P}(\phi)<\infty.$
Take $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ as a locally H\"older metric potential. For each $q \in Q^{\complement},$ there exists a unique, ergodic Gibbs state $\mu_{q}$ for $q\phi-T(q)\psi.$ Furthermore, $\mu_{q}(X_{\alpha(q)}^{s})=1.$ If we also have that $\int q\phi-T(q)\psi\, \mathrm{d}\mu_{q}>-\infty,$ $\mu_{q}$ is the unique equilibrium state for $q\phi-T(q)\psi.$
\end{theorem}
\begin{proof}
We have that $\Sigma$ is topologically mixing and BIP, the transition matrix $A$ satisfies the BIP property, and $q\phi-T(q)\psi$ is assumed to be locally H\"older. Hence, a unique Gibbs state for $q\phi-T(q)\psi$ exists by Theorem \ref{thm:S2}. Since $\Sigma$ is BIP, $\phi$ and $\psi$ are locally H\"older, and $q\phi-T(q)\psi$ has a Gibbs state, we have the unique, invariant, and ergodic Gibbs state $\mu_{q}$ for $q\phi-T(q)\psi$ by Theorem \ref{thm:S2}. It follows that $\mu_{q}(X_{\alpha(q)}^{s})=1$ because $\mu_{q}$ can be normalised on $X_{\alpha(q)}^{s}.$ Thus, if $-q\phi+T(q)\psi$ is integrable, we have that $\mu_{q}$ is the unique equilibrium state for $q\phi-T(q)\psi.$
\end{proof}
An important assumption for our potentials $-\phi,\psi$ is that they are in $\mathcal{L}^{1}(\mu_{q})$ for each $q \in Q^{\complement}.$ Before we provide a corollary complementing Theorem \ref{thm:MR1}, we give a definition from Pg 6 of Sarig \cite{sarig2003existence}. This definition will help us form a corollary about the function $\alpha(q)$ with respect to $\mu_{q}.$
\begin{definition}
Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ be a locally H\"older potential such that $\mathcal{P}(\phi)<\infty.$ Denote $Dir(\phi)$ as the collection of all $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ such that there exists $C_{\psi}>0, r \in (0,1)$ and $\varepsilon>0$ such that
\begin{enumerate}
\item $V_{n}(\psi)<C_{\psi}r^{n} \text{ for all } n \ge 1$
\item $\mathcal{P}(q\phi-t\psi)<\infty \text{ for all } t \in (T(q)-\varepsilon, T(q)+\varepsilon).$
\end{enumerate}
\end{definition}
Since $\psi$ is locally H\"older and there exist $\tilde{t}(q)<t<T(q)$ such that $\mathcal{P}(q\phi-t\psi)<\infty$ for $q \in Q^{\complement},$ it follows that $\psi \in Dir(\phi).$ Now, we provide a proposition (which is a modified version of Corollary 4 on Pg 6 of Sarig \cite{sarig2003existence}).
\begin{proposition}\label{cor:AN1}
If $\Sigma$ satisfies the BIP property, $\mathcal{P}(\phi)<\infty,$ and there exist $B>0$ and $\theta \in (0,1)$ such that $V_{n}(\phi)<B\theta^{n}$ for all $n \ge 1,$ then for every $\psi \in Dir(\phi)$ and fixed $q \in Q^{\complement},$ there exists $\varepsilon_{0}>0$ for which $t \mapsto \mathcal{P}(q\phi-t\psi)$ is real analytic on $(T(q)-\varepsilon_{0}, T(q)+\varepsilon_{0}).$
\end{proposition}
Hence, we have that $t \mapsto \mathcal{P}(q\phi-t\psi)$ is real analytic on $(T(q)-\varepsilon_{0}, T(q)+\varepsilon_{0}).$ We will use this important fact in the next section. Note that we can also apply Sarig's results to prove that $\mu$ is a Gibbs measure for $\phi.$ Furthermore, by Theorem \ref{thm:S2}, $\mu$ is the unique, ergodic Gibbs state for $\phi.$ Since we now have definitions and necessary results from thermodynamic formalism, we can now prove Theorem \ref{thm:M1} and Theorem \ref{thm:M2}, which are about the phase transitions of the multifractal spectrum.
\section{Proof of Theorems 1.1 and 1.2}
We will prove that the multifractal spectrum is analytic by taking advantage of a decomposition for $(\alpha_{\inf},\alpha_{\sup}).$ We take these steps in our proof.
\begin{enumerate}[I]
\item Let $Q=[q_{0},q_{1}]$ for some $q_{0},q_{1} \in \mathbb{R} \cup \{-\infty,\infty\}.$ We prove that the functions $t \mapsto \mathcal{P}(q\phi-t\psi),$ $T(q),$ and $\alpha(q)$ are analytic on open subintervals of $Q^{\complement}.$ Then, we prove that the multifractal spectrum is analytic on open subintervals of $\{\alpha(q): q \in Q^{\complement}\}.$
\item We prove that the multifractal spectrum is analytic on $(\alpha^{+},\alpha_{\lim})$ and $(\alpha_{\lim},\alpha^{-}).$
\item Finally, we assume that $Q=[q_{0},q_{1}]$ such that $0<q_{0}<q_{1}<\infty$ and $0< \alpha_{\lim} \le \infty$ exists. Our result about multifractal spectrum's phase transitions follows.
\end{enumerate}
\subsection{The Set $Q^{\complement}$ and Its Connection to $T(q)$}\hspace*{\fill} \par
First, we recall the definition of $Q:$
\[Q:=\{q \in \mathbb{R}: T(q)=\tilde{t}(q)\}.\]
We will prove that the functions $f_{\mu}(\alpha),$ $\alpha(q),$ and $T(q)$ are analytic on sets related to $Q^{\complement}.$
Fix $q \in Q^{\complement}.$ By Proposition \ref{cor:AN1}, $\mathcal{P}(q\phi-t\psi)$ is an analytic function of $t$ in a neighbourhood around $T(q).$ Hence, we can now prove that $\alpha(q)$ is analytic on open subintervals of $Q^{\complement}.$
\begin{proposition}\label{prop:B1}
Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ be locally H\"older with Gibbs state $\mu$ and let $\mathcal{P}(\phi)<\infty.$
Take $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ as a locally H\"older metric potential. Furthermore, assume that $\phi$ and $\psi$ are non-cohomologous. $T(q)$ is strictly convex, well defined, and analytic on open subintervals $Q^{\complement}.$ Furthermore, $\alpha(q)$ is well defined and analytic on open subintervals of $Q^{\complement}.$
\end{proposition}
\begin{proof}
Fix an arbitrary $q \in Q^{\complement}.$ By Theorem \ref{thm:MR1}, $\mu_{q}$ is Gibbs for $q\phi-T(q)\psi.$ Let $\varepsilon>0.$ Then, $t \mapsto \mathcal{P}(q\phi-t\psi)$ is an analytic function of $t$ in an $\varepsilon$-neighbourhood around $T(q)$ by Corollary \ref{cor:AN1}. Consider $\mathcal{P}(q\phi-t\psi)$ for $t \in (T(q)-\varepsilon, T(q)+\varepsilon).$
Since $-\phi,\psi \in \mathcal{L}^{1}(\mu_{q}),$ $\int -(q\phi-T(q)\psi)\, \mathrm{d}\mu_{q}<\infty.$ Hence, $\mu_{q}$ is the equilibrium state for $q\phi-T(q)\psi.$ We will denote $\mathcal{P}(q\phi-t\psi)$ as $\mathcal{P}(q,t).$ Then, by Proposition 2.6.13 on Pg 47 of Mauldin and Urba\'nski \cite{mauldin2003graph}, we can take the derivative of $\mathcal{P}(q\phi-t\psi)$:
\begin{equation}\label{eq:F}
\frac{\partial P(q,t)}{\partial t}=-\int_{\Sigma} \psi\, \mathrm{d}\mu_{q}.
\end{equation}
Since $\psi(x)>0$ for every $x \in \Sigma,$ Equation (\ref{eq:F}) gives us that
$$\frac{\partial P(q,t)}{\partial t}<0 \text{ and } \mathcal{P}(q\phi-T(q)\psi)=0.$$
Hence, $T(q)$ is well defined and analytic by the implicit function theorem. Since $T(q)$ is strictly decreasing and $\phi$ and $\psi$ are non-cohomologous to each other, $T(q)$ is strictly convex. Furthermore, $\alpha(q)$ is well defined and analytic because $T(q)$ is analytic and strictly convex.
\end{proof}
We now prove some results about the multifractal spectrum.
\begin{lemma}
Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ be locally H\"older with Gibbs state $\mu$ and let $\mathcal{P}(\phi)<\infty.$
Take $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ as a locally H\"older metric potential. Furthermore, assume that $\phi$ and $\psi$ are non-cohomologous. For each $\alpha \in \{\alpha(q): q \in Q^{\complement}\},$
\begin{equation}\label{eq:G}
f_{\mu}(\alpha)=f_{\mu}(\alpha(q))=T(q)+q\alpha(q).
\end{equation}
\end{lemma}
\begin{proof}
Take $\alpha=\alpha(q)$ for some $q \in Q^{\complement}.$ By Theorem \ref{thm:I1},
$$f_{\mu}(\alpha)=\inf\limits_{q \in \mathbb{R}}\{T(q)+q\alpha\}.$$
Then, since $T(q)$ is analytic in a neighbourhood of our $q \in Q^{\complement},$
$\frac{\mathrm{d}}{\mathrm{d}q}(T(q)+q\alpha)=0$ when $T'(q)+\alpha=0.$ We exactly have that $\alpha=-T'(q)=\alpha(q).$ Hence,
$$f_{\mu}(\alpha)=T(q)+q\alpha(q)$$
for our $q \in Q^{\complement}.$
\end{proof}
This lemma gives us a formula we need for the proof of the following proposition.
\begin{proposition}\label{prop:A1}
Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ be locally H\"older with Gibbs state $\mu$ and let $\mathcal{P}(\phi)<\infty.$
Take $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ as a locally H\"older metric potential. Furthermore, assume that $\phi$ and $\psi$ are non-cohomologous. The multifractal spectrum is analytic and strictly concave on any open subinterval $S$ of $\{\alpha(q): q \in Q^{\complement}\}.$
\end{proposition}
\begin{proof}
Let $\alpha=\alpha(q)$ s.t. $q \in Q^{\complement}.$ Remember that $\alpha(q)$ is well defined and analytic on $Q^{\complement}.$ By Equation (\ref{eq:G}), $f_{\mu}(\alpha(q))=T(q)+q\alpha(q).$ To prove that $f_{\mu}(\alpha)$ is analytic as a function of $\alpha,$ we will invert $\alpha(q).$ We now take the derivative of $f_{\mu}(\alpha(q)),$ which is also used in the proof of Lemma 6.17 in Pg 89 of Barreira \cite{barreira2008dimension}:
\begin{equation}\label{eq:H}
\frac{\mathrm{d}}{\mathrm{d}q}f_{\mu}(\alpha(q))= \frac{d}{d\alpha(q)}(f_{\mu}(\alpha(q)))\alpha'(q)=q\alpha'(q).
\end{equation}
Then, the derivative with respect to $\alpha(q)$ of the multifractal spectrum is
\[\frac{\mathrm{d}}{\mathrm{d}\alpha(q)}(f_{\mu}(\alpha(q)))=q.\]
Because we took the derivative in terms of $\alpha(q),$ $q$ is a function of $\alpha,$ i.e., $q=q(\alpha).$ Since $T(q)$ is strictly convex on $Q^{\complement},$ $-T''(q)<0$ for each $q \in Q^{\complement}.$ Because $\alpha'(q)=-T''(q)<0,$ $\alpha(q)$ and $q(\alpha)$ are invertible. Hence, since $\alpha(q)$ is analytic, $\frac{\mathrm{d}}{\mathrm{d}\alpha(q)}(f_{\mu}(\alpha(q)))=q(\alpha)$ is analytic. Thus, since $T(q(\alpha))$ and $q(\alpha)$ are analytic, $f_{\mu}(\alpha)$ is analytic as a function of $\alpha.$
To prove the strict concavity of $f_{\mu}(\alpha),$ we take further derivatives of the multifractal spectrum with respect to $q.$ Then, it follows that
$$1=\frac{\mathrm{d}^2}{\mathrm{d}q^2}(f_{\mu}(\alpha(q)))=\frac{\mathrm{d}^{2}}{\mathrm{d}\alpha^{2}}(f_{\mu}(\alpha(q)))\alpha'(q).$$
We have proven that
$$\frac{\mathrm{d}^2}{\mathrm{d}\alpha^2}(f_{\mu}(\alpha(q)))=\frac{1}{\alpha'(q)}=\frac{-1}{T''(q)}<0$$
because $\phi$ and $\psi$ are not cohomologous to each other (hence, $T(q)$ is strictly convex). Thus, the multifractal spectrum is strictly concave on any open subinterval $S \subset \{\alpha(q): q \in Q^{\complement}\}.$
\end{proof}
The proof of Proposition \ref{prop:A1} also establishes two further results about $\alpha(q)$ and $f_{\mu}(\alpha)$ respectively.
\begin{lemma}
Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ be locally H\"older with Gibbs state $\mu$ and let $\mathcal{P}(\phi)<\infty.$
Take $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ as a locally H\"older metric potential. Furthermore, assume that $\phi$ and $\psi$ are non-cohomologous. $\alpha(q)$ is a strictly decreasing function on open subintervals of $Q^{\complement}.$
\end{lemma}
\begin{proof}
Since $\alpha'(q)<0$ on $Q^{\complement},$ the lemma follows from Proposition \ref{prop:A1}.
\end{proof}
\begin{proposition}\label{prop:ID1}
Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ be locally H\"older with Gibbs state $\mu$ and let $\mathcal{P}(\phi)<\infty.$
Take $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ as a locally H\"older metric potential. Furthermore, assume that $\phi$ and $\psi$ are non-cohomologous. The multifractal spectrum $f_{\mu}(\alpha)$
\begin{enumerate}
\item increases on open subintervals $S$ of $\{\alpha(q): q<0\} \cap \{\alpha(q): q \in Q^{\complement}\}$
\item decreases on open subintervals of $\{\alpha(q): q>0\} \cap \{\alpha(q): q \in Q^{\complement}\}.$
\end{enumerate}
\end{proposition}
\begin{proof}
Equation (\ref{eq:H}) gives us that
$$\frac{\mathrm{d}}{\mathrm{d}\alpha}(f_{\mu}(\alpha(q))=q\alpha'(q)=-qT''(q).$$
Since $-T''(q)<0$ on $Q^{\complement},$ we get the following.
$\text{If } q<0, \frac{\mathrm{d}}{\mathrm{d}\alpha}(f_{\mu}(\alpha(q))>0.$ $\text{If } q>0, \frac{\mathrm{d}}{\mathrm{d}\alpha}(f_{\mu}(\alpha(q))<0.$ Hence, the increasing and decreasing behaviour on open subintervals of $\{\alpha(q):q\in Q^{\complement}\}$ is immediate.
\end{proof}
In summary, we proved that if $T(q)$ is analytic on open subintervals of $Q^{\complement},$ then $\alpha(q)$ is analytic on open subintervals of $Q^{\complement}.$ In turn, this gives us that $f_{\mu}(\alpha)$ is analytic on open subintervals of $\{\alpha(q):q\in Q^{\complement}\}.$ We also find that increasing and decreasing behaviour on open subintervals of $\{\alpha(q):q\in Q^{\complement}\}$ is based on the sign of each $q\in Q^{\complement}.$
\subsection{The Intervals $(\alpha^{+},\alpha_{\lim})$ and $(\alpha_{\lim},\alpha^{-})$}\hspace*{\fill} \par
Now that we have proven that the multifractal spectrum is analytic on open subintervals of $\{\alpha(q):q\in Q^{\complement}\},$ we must consider the behaviour of the multifractal spectrum on other open subintervals of $(\alpha_{\inf},\alpha_{\sup}).$ Without loss of generality, we assume that $Q=[q_{0},q_{1}]$ for $-\infty<0<q_{0}<q_{1}<\infty.$ As stated earlier, $\alpha^{-}=\alpha(q_{0})$ and $\alpha^{+}=\alpha(q_{1}).$ Using our decomposition of $(\alpha_{\inf},\alpha_{\sup}),$ we know that we must prove that the multifractal spectrum is analytic on $(\alpha^{+},\alpha_{\lim})$ and $(\alpha_{\lim},\alpha^{-}).$
\begin{proposition}\label{prop:A2}
Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ be locally H\"older with Gibbs state $\mu$ and let $\mathcal{P}(\phi)<\infty.$
Take $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ as a locally H\"older metric potential. Furthermore, assume that $\phi$ and $\psi$ are non-cohomologous.
\begin{enumerate}
\item The function $f_{\mu}(\alpha)=T(q_{1})+q_{1}\alpha$ on $(\alpha^{+},\alpha_{\lim})$ and $f_{\mu}(\alpha)=T(q_{0})+q_{0}\alpha$ on $(\alpha_{\lim},\alpha^{-}).$
\item The multifractal spectrum is an increasing linear function on $(\alpha^{+},\alpha_{\lim})$ and $(\alpha_{\lim},\alpha^{-})$ because $q_{0},q_{1}>0.$
\item Furthermore, the sign of $q_{0}$ and $q_{1}$ determine the increasing or decreasing behaviour of the multifractal spectrum on $(\alpha^{+},\alpha_{\lim})$ and $(\alpha_{\lim},\alpha^{-}).$
\end{enumerate}
\end{proposition}
\begin{proof}
Let $\alpha \in (\alpha(q_{1}), \alpha_{\lim}).$ For each $\alpha \in (\alpha(q_{1}), \alpha_{\lim}),$ we have that
$$f_{\mu}(\alpha)= \inf\limits_{q \in \mathbb{R}}\{T(q)+q\alpha\}=T(q_{1})+q_{1}\alpha.$$
Hence, $$\frac{\mathrm{d}}{\mathrm{d}\alpha}f_{\mu}(\alpha)=q_{1}>0.$$
Thus, $f_{\mu}(\alpha) $ is an increasing linear function with slope $q_{1}$ on the interval $(\alpha^{+}, \alpha_{\lim}).$ Assume that $\alpha \in (\alpha_{\lim}, \alpha(q_{0})).$ For each $\alpha \in (\alpha_{\lim}, \alpha(q_{0})),$
$f_{\mu}(\alpha)= \inf \limits_{q \in \mathbb{R}}\{T(q)+q\alpha\}=T(q_{0})+q_{0}\alpha.$ Hence, $$\frac{\mathrm{d}}{\mathrm{d}\alpha}f_{\mu}(\alpha)=q_{0}>0.$$
Thus, $f_{\mu}(\alpha)$ is an increasing linear function with slope $q_{0}$ on the interval $(\alpha_{\lim}, \alpha^{-}).$
\end{proof}
The multifractal spectrum is linear on $(\alpha_{\lim},\alpha^{-})$ and $(\alpha^{+},\alpha_{\lim})$ because of the endpoints of $Q.$ We can finally prove our main theorems.
\subsection{Phase Transitions- Intervals and Points of $Q$}\hspace*{\fill} \par
Now that we have proven results relating the analyticity of $T(q)$ to the analyticity of $f_{\mu}(\alpha),$ we use Propositions \ref{prop:Q1}, \ref{prop:A1}, and \ref{prop:A2} to prove that the multifractal spectrum has $0$ to $3$ phase transitions when $\alpha_{\lim}$ exists and $0<\alpha_{\lim}<\infty.$ Furthermore, we prove a complementary result when $\alpha_{\lim}=\infty.$ We provide the subsequent proposition to remind the reader about the forms $Q$ can take.
\begin{proposition}
$Q$ can be a closed interval, a half-open infinite interval, a point, or the empty set.
\end{proposition}
Analysing $Q$ gives us information about the number of phase transitions for the multifractal spectrum. We give a thorough discussion of the multifractal spectrum's phase transitions in the case that $Q$ is a closed interval with positive endpoints.
\subsubsection{Positive Closed Interval}\hspace*{\fill} \par
We assume that $Q=[q_{0}, q_{1}]$ for some $q_{0}, q_{1} \in \mathbb{R}$ such that $0<q_{0}<q_{1}<\infty.$
\begin{proposition}\label{prop:PCD1}
Let $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ be locally H\"older with Gibbs state $\mu$ and let $\mathcal{P}(\phi)<\infty.$
Take $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ as a locally H\"older metric potential. Furthermore, assume that $\phi$ and $\psi$ are non-cohomologous. We have four possible types of behaviour for the multifractal spectrum as follows. We call them cases 1 to 4 with respect to the order below.
\begin{enumerate}
\item Let $\alpha(q_{0})>\alpha_{\lim}>\alpha(q_{1}).$ The function $f_{\mu}(\alpha)$ is analytic on $(\alpha_{\inf}, \alpha(q_{1})),$ $(\alpha(q_{1}), \alpha_{\lim}),$ $(\alpha_{\lim}, \alpha(q_{0})),$ and $(\alpha(q_{0}), \alpha_{\sup}).$ There are three phase transitions for the multifractal spectrum at $\alpha(q_{0}), \alpha_{\lim},$ and $\alpha(q_{1}).$
\item Let $\alpha(q_{0})> \alpha_{\lim}=\alpha(q_{1}).$ The function $f_{\mu}(\alpha)$ is analytic on $(\alpha_{\inf}, \alpha_{\lim}),$ $(\alpha_{\lim}, \alpha(q_{0})),$ $(\alpha(q_{0}), \alpha_{\sup}).$ The multifractal spectrum has two phase transitions at $\alpha(q_{0})$ and $\alpha_{\lim}.$
\item Let $\alpha(q_{0})=\alpha_{\lim}>\alpha(q_{1}).$ The function $f_{\mu}(\alpha)$ is analytic on $(\alpha_{\inf}, \alpha(q_{1})),$ $(\alpha(q_{1}), \alpha_{\lim}),$ $(\alpha_{\lim},\alpha_{\sup}).$ The multifractal spectrum two phase transitions at $\alpha(q_{1})$ and $\alpha_{\lim}.$
\item Let $\alpha(q_{0})=\alpha_{\lim}=\alpha(q_{1}).$ The function $f_{\mu}(\alpha)$ is analytic on $(\alpha_{\inf}, \alpha_{\lim}),$ $(\alpha_{\lim},\alpha_{\sup}).$ The multifractal spectrum has a possible phase transition at $\alpha_{\lim}.$
\end{enumerate}
\end{proposition}
\begin{proof}
The method to prove cases 2 to 4 is similar the proof for case 1, so we will only provide the proof for case 1. Remember that $\alpha(q_{1})=\alpha^{+}$ and $\alpha(q_{0})=\alpha^{-}.$ We use the decomposition of $[\alpha_{\inf},\alpha_{\sup}]$ in the following way. Each $\alpha \in (\alpha_{\inf}, \alpha(q_{1}))$ satisfies $\alpha=\alpha(q)$ for $q > q_{1}.$ Proposition \ref{prop:A2} gives us that $f_{\mu}(\alpha)$ equals $T(q_{1})+q_{1}\alpha$ on $(\alpha^{+},\alpha_{\lim})$ and $T(q_{0})+q_{0}\alpha$ on $(\alpha_{\lim},\alpha^{-}).$ Each $\alpha \in (\alpha(q_{0}), \alpha(0))$ satisfies $\alpha=\alpha(q)$ for a unique $0< q \le q_{0}.$ Each $\alpha \in (\alpha(0), \alpha_{\sup})$ satisfies $\alpha=\alpha(q)$ for a unique $q<0.$
Hence, by Propositions \ref{prop:A1} and \ref{prop:A2}, the multifractal spectrum is analytic on $(\alpha_{\inf},\alpha^{+}),$ $(\alpha^{+},\alpha_{\lim}),$ $(\alpha_{\lim},\alpha^{-}),$ and $(\alpha^{-},\alpha_{\sup}).$ The increasing and decreasing behaviour of the multifractal spectrum is immediate from Proposition \ref{prop:ID1} and Lemma \ref{prop:A2}. Thus, it follows that the multifractal spectrum has phase transitions at $\alpha(q_{1}),$ $\alpha_{\lim},$ and $\alpha(q_{0}).$
\end{proof}
The proofs for the following other cases for $Q=[q_{0}, q_{1}]$ are proved in the same way as Proposition \ref{prop:PCD1}. In all of the following cases for $Q,$ the multifractal spectrum ranges from having no phase transitions if $\alpha^{-}=\alpha_{\lim}=\alpha^{+}$ to three phase transitions if $\alpha^{-}>\alpha_{\lim}>\alpha^{+}.$ Using the same techniques as the proof to Proposition \ref{prop:PCD1}, we get the behaviour as outlined below.
\begin{enumerate}
\item If $Q$ is a closed interval, then the multifractal spectrum has $0$ to $3$ phase transitions.
\item If $Q$ is a point $q \in \mathbb{R},$ then $\alpha^{+}=\alpha^{-}.$ Hence, the multifractal spectrum has $0$ to $2$ phase transitions.
\item If $Q$ is the half-open interval $(-\infty,q_{1}),$ then $\alpha_{\lim} \ge \alpha^{+}.$
The multifractal spectrum would then have $0$ to $1$ phase transition.
\item If $Q$ is the half-open interval $(q_{0},\infty),$ then $\alpha^{-} \ge \alpha_{\lim}.$
The multifractal spectrum would then have $0$ to $1$ phase transition.
\item If $Q$ is the open interval $(-\infty,\infty),$ then $\alpha^{+}=\alpha^{-}=\alpha_{\lim}=\alpha_{\inf}=\alpha_{\sup}.$ Then, the multifractal spectrum is constant because $f_{\mu}(\alpha)=t_{\infty}.$
\end{enumerate}
\begin{proposition}\label{prop:NO1}
Let $Q=\emptyset.$ This yields no phase transitions for the multifractal spectrum.
\end{proposition}
\begin{proof}
Since $Q=\emptyset,$ $T(q)>\tilde{t}(q)$ for all $q \in \mathbb{R}.$ Hence, $\mathcal{P}(q\phi-t\psi)$ is analytic as a function of $t$ on an $\varepsilon-$neighbourhood of $T(q).$ This gives us that $T(q)$ is analytic on all of $\mathbb{R}$ (as we proved earlier). Thus, by Proposition \ref{prop:AA1} the proposition follows.
\end{proof}
Therefore, we have proven Theorem \ref{thm:M1}. Theorem \ref{thm:M1} tells us that when $0<\alpha_{\lim}<\infty,$ the multifractal spectrum has $0$ to $3$ phase transitions. We repeat the theorem for completeness:
\begin{theorem}
Let $\phi: \Sigma \rightarrow \mathbb{R}^{-}$ be a potential with Gibbs measure $\mu$ such that $P(\phi)<\infty.$ and $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ be a metric potential. Assume that $\phi$ and $\psi$ are non-cohomologous locally H\"older potentials such that $\alpha_{\lim}<\infty.$
\begin{enumerate}
\item There exist intervals $A_{i}$ such that $f_{\mu}(\alpha)$ is analytic on each of their interiors.
\item The interval $(\alpha_{\inf}, \alpha_{\sup})=\cup_{i=1}^{j}A_{i}$ such that $j=\{1,2,3,4\}.$
\item The multifractal spectrum is concave on $(\alpha_{\inf}, \alpha_{\sup}),$ has its maximum at $\alpha(0),$ and has zero to three phase transitions.
\end{enumerate}
\end{theorem}
We now analyse the case when $\alpha_{\lim}=\infty.$
\begin{theorem}
Let $\phi: \Sigma \rightarrow \mathbb{R}^{-}$ be a potential with Gibbs measure $\mu$ such that $\mathcal{P}(\phi)<\infty$ and $\psi: \Sigma \rightarrow \mathbb{R}^{+}$ be a metric potential. Assume that $\phi$ and $\psi$ are non-cohomologous locally H\"older potentials such that $\alpha_{\lim}=\infty.$
\begin{enumerate}
\item There exist intervals $A_{i}$ such that $f_{\mu}(\alpha)$ is analytic on each of their interiors.
\item The interval $(\alpha_{\inf}, \alpha_{\sup})=\cup_{i=1}^{j}A_{i}$ such that $j=\{1,2\}.$
\item The multifractal spectrum is concave on $(\alpha_{\inf}, \alpha_{\sup}),$ is equal to its maximum $f_{\mu}(\alpha(0))$ on $(\alpha(0),\alpha_{\sup}),$ and has zero to one phase transition.
\end{enumerate}
\end{theorem}
\begin{proof}
Apply Propositions \ref{prop:ID1}, \ref{prop:A2}, and \ref{prop:NO1} because they do not need $\alpha_{\lim}<\infty.$
\end{proof}
We give a map that generates examples of phase transitions for $f_{\mu}(\alpha).$
\section{Adaptation For the Gauss Map}
Up until this point, we have only considered the multifractal spectrum with respect to a locally H\"older function, defined by a general expanding map. We give a specific expanding map as follows.
\subsection{The Gauss Map}
\begin{definition}
The {\em Gauss map} $G: [0,1]\setminus{\mathbb{Q}} \rightarrow [0,1]\setminus{\mathbb{Q}}$ is defined by
$$G(x)=\frac{1}{x} \mod{1}.$$
\end{definition}
The inverse branches of $G$ are similar to a translated Gauss map.
\begin{definition}
Let $I_{b}:=[\frac{1}{b+1},\frac{1}{b}]\setminus{\mathbb{Q}}.$ Define $G_{b}:[0,1]\setminus{\mathbb{Q}} \rightarrow I_{b}$ as the {\em inverse branch} $(G|_{I_{b}})^{-1}$ of the Gauss map, which is
$$G_{b}(x)=\frac{1}{x+b}$$
for $x \in [0,1]\setminus{\mathbb{Q}}.$ For each $\tilde{b}=(b_{1},b_{2},...,b_{n}) \in \mathbb{N}^{n},$ the composition of these inverse branches is
$$G_{\tilde{b}}:=G_{b_{1}} \circ G_{b_{2}} \circ \cdots \circ G_{b_{n}}.$$
\end{definition}
From this point, we use the full shift $\Sigma=\mathbb{N}^{\mathbb{N}}.$ The coding map we take between $\Sigma$ and $[0,1]$ is the continued fraction map.
\begin{definition}
The {\em coding map},
$\pi: \Sigma \rightarrow [0,1]\setminus{\mathbb{Q}},$ is defined as follows. For each sequence $a \in \Sigma$ such that $a=(a_{1}(x),a_{2}(x),...),$ the map $\pi$ is
$$\pi(a):=\frac{1}{a_{1}(x)+\frac{1}{a_{2}(x)+\frac{1}{a_{3}(x)+\cdots}}}$$
such that
\begin{enumerate}
\item $x=\pi(a)$
\item for each $i,k \in \mathbb{N},$ $G^{i-1}(x) \in (\frac{1}{k+1},\frac{1}{k})$ yields that $a_{i}(x)=k.$
\end{enumerate}
\end{definition}
We use the Gauss map to define the potential $\psi.$
\subsection{Thermodynamic Formalism}\hspace*{\fill} \par
By symbolic coding, we provide definitions for our locally H\"older potentials $\phi$ and $\psi.$
\begin{definition}
Define the locally H\"older potential $\psi:\Sigma \rightarrow \mathbb{R}^{+}$ on each word $x \in \Sigma$ by
$$\psi(x)=\log|G'(\pi(x))|.$$
\end{definition}
Again, we assume that $\phi: \Sigma \rightarrow \mathbb{R}^{-}$ is a locally Holder potential on $\Sigma$ such that $0 \le \mathcal{P}(\phi)<\infty.$ Again, we assume that $\phi$ and $\psi$ are non-cohomologous to each other. Using the potential $\psi,$ the following lemma relates the Gauss map to the diameter of any cylinder in $\Sigma.$ We use the following result to prove our main theorem.
\begin{lemma}\label{lemma:D1}
Let $\mu$ be an ergodic $G-$invariant measure on $[0,1],$ $\nu$ be an ergodic $\sigma-$invariant measure on $\Sigma,$ and let $\psi(x)=\log|G'(\pi(x))|$ for each word $x \in \Sigma.$ Then, for $\mu-$a.e. $z=\pi(x) \in [0,1]$ such that $x=(x_{1},x_{2},...) \in \Sigma,$
$$-\int_{\Sigma} \psi\, \mathrm{d}\nu=-\int_{0}^{1} \log{|G'|}\, \mathrm{d}\mu=\lim\limits_{n \rightarrow \infty}-\frac{1}{n}\sum_{i=0}^{n-1}\log{|G'(G^{i}(z))|} $$
$$\text{and} \lim\limits_{n \rightarrow \infty}-\frac{1}{n}\sum_{i=0}^{n-1}\log{|G'(G^{i}(z))|}=\lim\limits_{n \rightarrow \infty}-\frac{1}{n}\log|(G^{n})'(z)|=\lim\limits_{n \rightarrow \infty}\frac{1}{n}\log{|[x_{1},..,x_{n}]|}.$$
Furthermore, our choice of $\psi$ is a metric potential.
\end{lemma}
\begin{proof}
By the mean value theorem, there exists $z \in \pi([x_{1},...,x_{n}])$ such that $(G^{n})'(z)=(|[x_{1},...,x_{n}]|)^{-1}.$
By the chain rule, for $z \in [0,1],$ it follows that \[(G^{n})'(z)=G'(G^{n-1}(z))G'(G^{n-2}(z)) \cdots G'(z).\]
Combining both the diameter and chain rule equalities for $(G^{n})'(z),$ we get that
\begin{equation}\label{eq:1}
\sum_{i=0}^{n-1}\psi(\sigma^{i}(x))=\log{|(G^{n})'(z)|}=-\log{|[x_{1},...,x_{n}]|}=\sum_{i=0}^{n-1}\log{|G'(G^{i}(z))|}
\end{equation}
Hence,
\[|[x_{1},...,x_{n}]|=\exp\left(-\sum_{i=0}^{n-1}\psi(\sigma^{i}(x))\right).\]
Then, for $\mu-$a.e.$z=\pi(x) \in [0,1],$
the result follows from the Birkhoff ergodic theorem and $\psi$ is a metric potential.
\end{proof}
Since we have an expression for $\log{|[x_{1},...,x_{n}]|},$ let us again consider subsets of $\Sigma$ with symbolic dimension $\alpha.$
\section{The Multifractal Spectrum and $X_{\alpha}$}
We now revisit the set $X_{\alpha}^{s}$ and recall the definition of symbolic dimension. Again, we take $\mu$ as the Gibbs measure for the potential $\phi: \Sigma \rightarrow \mathbb{R}^{-}.$ There is also a measure $\bar{\mu}=\mu \circ \pi^{-1}$ on $[0,1].$
\begin{definition}
For each $\alpha \in [\alpha_{\inf},\alpha_{\sup}],$ there exists a word $x \in [x_{1},...,x_{n}] \subset \Sigma$ has \text{\em symbolic dimension} $\alpha$ if
\[
d_{\mu}(x):=\lim\limits_{n \rightarrow \infty}\frac{\log{\mu([x_{1},...,x_{n}])}}{\log{|[x_{1},...,x_{n}]|}}=\alpha.
\]
\end{definition}
Define the level set
$$\pi(X_{\alpha}^{s})=\pi\left(\left\{x \in \Sigma: \lim\limits_{n \rightarrow \infty}\frac{\log{\mu([x_{1},...,x_{n}])}}{\log{|[x_{1},...,x_{n}]|}}=\alpha\right\}\right).$$
There is a similar type of local dimension for $[0,1].$
\begin{definition}
For $r>0,$ let $B(x,r)$ be a ball centered at $x \in [0,1].$ Then, for each $\alpha \in [\alpha_{\inf},\alpha_{\sup}],$ $x \in [0,1]$ has \text{\em local dimension} $\alpha$ if
\[
\lim\limits_{r \rightarrow 0}\frac{\log{\bar{\mu}(B(x,r))}}{\log{r}}=\alpha.
\]
\end{definition}
Define the set
$$X_{\alpha}=\{x \in [0,1]\setminus{\mathbb{Q}}: \lim\limits_{r \rightarrow 0}\frac{\log{\bar{\mu}(B(x,r))}}{\log{r}}=\alpha\}.$$
We now give an important result by Iommi \cite{iommi2005multifractal} (Page 1891, Theorem 3.7).
\begin{proposition}\label{prop:I2}
For each $\alpha \in (\alpha_{\inf}, \alpha_{\sup}),$ $\pi(X_{\alpha}^{s})$ satisfies
$$\dim_{H}(\pi(X_{\alpha}^{s}))=f_{\mu}(\alpha)=\inf\limits_{q \in \mathbb{R}}\{T(q)+q\alpha\}.$$
\end{proposition}
We have a similar result for $X_{\alpha}.$
\begin{proposition}\label{thm:G1}
For each $\alpha \in (\alpha_{\inf}, \alpha_{\sup}),$ we have that
$$f_{\mu}(\alpha)=\dim_{H}(X_{\alpha}).$$
\end{proposition}
We outline the proof of Proposition \ref{thm:G1} as follows. To prove $\dim_{H}(X_{\alpha}) \ge f_{\mu}(\alpha),$ we first notice that the geometric structure of the Gauss map makes it difficult to cover cylinder sets with neighbourhoods and vice versa. Define $T_{n}(q)$ as the temperature function in the setting of $\Sigma_{n}.$ Take $\mu_{q_{n}}$ as the equilibrium state for $q\phi-T_{n}(q)\psi.$ We provide a condition involving the hitting times of a $\mu_{q_{n}}-$typical cylinder set. This gives us that $X_{\alpha}^{s} \subset \pi^{-1}(X_{\alpha}).$ Then, we use an inequality proven by Pesin and Weiss \cite{pesin1997multifractal} involving the pointwise dimension of $x \in \Sigma_{n}$ and the multifractal spectrum $f_{\mu,n}(\alpha).$ To apply these results to $\Sigma,$ we approximate pressure in $\Sigma$ with pressure in $\Sigma_{n}.$ This gives us a monotone convergence argument for $T_{n}(q)$ and it induces a similar argument involving Pesin and Weiss's inequality. Hence, the result follows.\par
To prove $\dim_{H}(X_{\alpha}) \le f_{\mu}(\alpha),$ we again use the behaviour of the Gauss map. Neighbouring $m-$cylinders nearly have equal diameters for large $m \in \mathbb{N}.$ We exploit this fact by slightly increasing the size of cylinders around each $x \in X_{\alpha}^{s}.$ This creates a Hausdorff cover for $\pi^{-1}(X_{\alpha}),$ which we use in an argument for bounding the Hausdorff measure of $X_{\alpha}.$ This gives the appropriate upper bound for $\dim_{H}(X_{\alpha}).$
Through Proposition \ref{thm:G1}, we can apply the results of Theorems \ref{thm:M1} and \ref{thm:M2} to the potential $\psi=\log|G'|.$ Now, we consider examples in the next section.
\section{Examples of Phase Transitions for the Gauss Map}
We consider the geometric potential $\psi=\log|G'|$ and provide examples in which the multifractal spectrum has up to infinitely many phase transitions.
Also, we apply the results of Theorems \ref{thm:M1} and \ref{thm:M2} as well as provide an example when $\alpha_{\lim}$ does not exist. Furthermore, we remark that by estimating $\psi(x)$ for $x=(x_{1},x_{2},...)$ with a locally H\"older potential $2\log(x_{1}),$ we find that $\psi$ is unbounded. Now, we consider the first of our three cases: $\alpha_{\lim}<\infty.$
\subsection{The Case $\alpha_{\lim}<\infty$}\hspace*{\fill} \par
As reflected by Theorem \ref{thm:M1}, we provide examples of potentials such that $f_{\mu}(\alpha)$ has zero to three phase transitions. The easiest way to show this is by approximating $\phi$ and $\psi$ with locally constant potentials.
\subsubsection{Approximation with Locally Constant Potentials}\hspace*{\fill} \par
Since $\phi$ and $\psi$ are locally H\"older, we will give some of our estimates in terms of one periodic sequences. Now, we provide a critical technique used in our examples. Locally H\"older functions can be approximated using locally constant functions as follows. Let $\Sigma=\mathbb{N}^{\mathbb{N}},$ take $\psi(x)=\log|G'(\pi(x))|$ for each $x \in \Sigma,$ and assume that $\tilde{\psi}(x)$ is a locally constant function.
Then, consider $\psi-\bar{\psi}.$ This potential is locally H\"older because it is a difference of locally H\"older functions. For any arbitrary $n-$cylinder $[x_{1},...,x_{n}]$ in $\Sigma$ and $x \in [x_{1},...,x_{n}],$
\[0 \le |\psi(x)-\bar{\psi}(x)| \le V_{n}(\psi-\tilde{\psi}) \le C\theta^{n}\]
for some $C>0$ and $\theta \in (0,1).$
Let $\varepsilon>0.$ Since $C>0$ is fixed, we have that for $n \ge N$ such that $N \in \mathbb{N}$ is large,
\[0 \le |\psi(x)-\bar{\psi}(x)| \le V_{n}(\psi-\bar{\psi}) \le C\theta^{N} \le \varepsilon.\]
It follows that we can approximate $\psi$ with a locally H\"older and locally constant potential $\bar{\psi}(x)=-\log(\frac{6}{\pi^{2}x_{1}^{2}}).$
Furthermore, we provide a formula used to calculate the topological pressure of $q\bar{\phi}-t\bar{\psi}.$ Let $0<p_{i}<1$ and $0<s_{i}<1$ for each $i \in \mathbb{N}.$ We will usually take $\bar{\phi}(x)=\log{p_{x_{1}}}$ and $\bar{\psi}(x)=\log{s_{x_{1}}^{-1}}.$ For most of our examples, $s_{i}=\frac{6}{\pi^{2}i^{2}}.$
Then, for $\varepsilon>0,$
\[\log\left(\sum_{i=1}^{\infty}p_{i}^{q}s_{i}^{t}\right) - \varepsilon \le \mathcal{P}(q\bar{\phi}-t\bar{\psi}) \le \log\left(\sum_{i=1}^{\infty}p_{i}^{q}s_{i}^{t}\right) + \varepsilon.\]
We now proceed with our examples.
\subsubsection{Example of Zero Phase Transitions}\hspace*{\fill} \par
For each $i \in \mathbb{N},$ let $0<p_{i}<1.$ Take $p_{i}=\frac{C}{(i+1)^{3}}$ with $C \approx 4.9491$ chosen so that $\sum_{i=1}^{\infty}p_{i}=1.$
For each $x=(x_{1},...) \in \Sigma,$ define the potentials $\phi:\Sigma\rightarrow\mathbb{R}^{-}$ and $\psi:\Sigma\rightarrow\mathbb{R}^{+}$ as follows:
\[
\phi(x)=\log{p_{x_{1}}}=\log\left(\frac{4.9491}{(x_{1}+1)^{3}}\right)
\]
and
\[
\psi(x)=\log|G'(\pi(x))|.
\]
Throughout this example, we will estimate $\psi(x)$ with $\tilde{\psi}(x)=-\log(\frac{6}{\pi^{2}(x_{1})^{2}})$ and take $\tilde{\phi}=\phi.$
We now consider the potential $q\tilde{\phi}-t\tilde{\psi}.$
As noticed earlier, the number of phase transitions is determined by the relationship between $T(q)$ and $\tilde{t}(q).$ We found that $\tilde{t}(q)=-\alpha_{\lim}q+t_{\infty}.$
The calculations for $\alpha_{\lim}$ are as follows:
Let $\bar{j}=(j,j,...)$ such that $j \in \mathbb{N}.$ Then,
\begin{equation}\label{eq:K}
\alpha_{\lim}=\lim\limits_{j\rightarrow\infty}\frac{\tilde{\phi}(\bar{j})}{-\tilde{\psi}(\bar{j})}=\lim\limits_{j\rightarrow\infty}\frac{-3\log{j}}{-2\log{j}}=\frac{3}{2}.
\end{equation}
The value $t_{\infty}$ can be found using Lemma~\ref{prop:L1}. \par
Remember that
\[
t_{\infty}=\inf\{t\in\mathbb{R}:\mathcal{P}(-t\tilde{\psi})<\infty\}= \inf\{t\in\mathbb{R}:Z_{1}(-t\tilde{\psi})<\infty\}.
\]
Let $j \in \mathbb{N}.$ We let $V_{1,j}(\tilde{\psi})$ be the H\"older constant for $\tilde{\psi}$ such that for every $i,x \in [j],$
\[|\tilde{\psi}(i)-\tilde{\psi}(x)| \le V_{1,j}(\tilde{\psi}).\]
Thus, we must find the infimum of all $t\in\mathbb{R}$ such that
\begin{eqnarray}\label{eq:L}
Z_{1}(-t\tilde{\psi}) &=& \sum_{x_{1}=1}^{\infty}\exp\sup\limits_{x\in[x_{1}]}\left(-2t\log{x_{1}}-t\log\left(\frac{6}{\pi^{2}}\right)\right) \nonumber \\
&\le&
\sum_{j=1}^{\infty}\exp\left(\log(j^{-2t})-t\log\left(\frac{6}{\pi^{2}}\right)+tV_{1,j}(\tilde{\psi})\right)<\infty
\end{eqnarray}
for $t>\frac{1}{2}$ because $V_{1,j}(\tilde{\psi}) \rightarrow 0$ as $j \rightarrow \infty.$ Hence, $t_{\infty}=\frac{1}{2}.$ Thus, by Equations (\ref{eq:K}) and (\ref{eq:L}),
\[
\tilde{t}(q)=-\frac{3}{2}q+\frac{1}{2}.
\]
By definition, we must have that $Z_{n}(q\tilde{\phi}-t\tilde{\psi})<1$ in order for $\mathcal{P}(q\tilde{\phi}-t\tilde{\psi})<0.$ As noticed earlier, we get $Q$ by considering the values of $q\in\mathbb{R}$ satisfy $T(q)=\tilde{t}(q).$ Hence, we must find $q\in\mathbb{R}$ such that $Z_{n}(q\tilde{\phi}-\tilde{t}(q)\tilde{\psi})<1.$ It is enough to consider $q\in\mathbb{R}$ such that $Z_{1}(q\tilde{\phi}-\tilde{t}(q)\tilde{\psi})<1.$ \par
Let $j \in \mathbb{N}.$ We let $V_{1,j}(q\tilde{\phi}-\tilde{t}(q)\tilde{\psi})$ be the H\"older constant for $q\tilde{\phi}-\tilde{t}(q)\tilde{\psi}$ such that for every $i,x \in [j],$
\[|(q\tilde{\phi}-\tilde{t}(q)\tilde{\psi})(i)-(q\tilde{\phi}-\tilde{t}(q)\tilde{\psi})(x)| \le V_{1,j}(q\tilde{\phi}-\tilde{t}(q)\tilde{\psi}).\]
Furthermore, for each $x \in [j]$ and $\bar{j}=(j,j,j....),$
\[|\sup\limits_{x\in[x_{1}]}q\tilde{\phi}-\tilde{t}(q)\tilde{\psi})(x)-(q\tilde{\phi}-\tilde{t}(q)\tilde{\psi})(\bar{j})|=C_{1,j}(q\tilde{\phi}-\tilde{t}(q)\tilde{\psi})\]
such that $C_{1,j}(q\tilde{\phi}-\tilde{t}(q)\tilde{\psi})>0.$
Hence,
\begin{eqnarray}
Z_{1}(q\tilde{\phi}-\tilde{t}(q)\tilde{\psi}) &=& \sum_{x_{1}=1}^{\infty}\exp\sup\limits_{x\in[x_{1}]}(q\tilde{\phi}-\tilde{t}(q)\tilde{\psi})(x) \nonumber \\
&=&
\sum_{j=1}^{\infty}\exp((q\tilde{\phi}-\tilde{t}(q)\tilde{\psi})(\bar{j})+C_{1,j}(q\tilde{\phi}-\tilde{t}(q)\tilde{\psi})) \nonumber \\
&=&
\sum_{j=1}^{\infty}\frac{(4.9491)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{2}q+\frac{1}{2}}}{j^{3q+2\tilde{t}(q)}}\exp(C_{1,j}(q\tilde{\phi}-\tilde{t}(q)\tilde{\psi}))=\infty. \nonumber
\end{eqnarray}
Hence, $T(q)\not=\tilde{t}(q)$ for any $q\in\mathbb{R}.$ Thus, $Q=\emptyset.$
Therefore, by Proposition 6.3, the multifractal spectrum has no phase transitions.
\subsubsection{One Phase Transition}\hspace*{\fill} \par
We first provide pictures of the multifractal spectrum and $T(q)$ for the following example.
\begin{figure}[H]
\centering
\begin{subfigure}{.5\textwidth}
\centering
\includegraphics[width=1\linewidth]{Multifractal_Spectrum.jpg}
\caption{\text{Multifractal Spectrum on} $(0.6,1.13653)$}
\label{fig:sub1}
\end{subfigure}%
\begin{subfigure}{.5\textwidth}
\centering
\includegraphics[width=1\linewidth]{Temperature_Function.jpg}
\caption{Temperature Function on $(-150,150)$}
\label{fig:sub2}
\end{subfigure}
\caption{We observe that $\alpha_{\lim}=0.6$ and $T(q)$ has phase transition around $q=1.3.$ Most importantly, the multifractal spectrum has its phase transition around $\alpha \in (0.955,0.999)$ and its maximum is $f_{\mu}(\alpha(0))=1$ (such that $\alpha(0)=1.0068$).}
\label{fig:test}
\end{figure}
For each $i \in \mathbb{N},$ let $0<p_{i}<1.$ Take $p_{i}=\frac{C}{(i)^{\frac{6}{5}}(\log(i+2))^{2}}$ with $C \approx 0.67569$ chosen so that $\sum_{i=1}^{\infty}p_{i}=1.$
For each $x=(x_{1},...) \in \Sigma,$ define the potentials $\phi:\Sigma\rightarrow\mathbb{R}^{-}$ and $\psi:\Sigma\rightarrow\mathbb{R}^{+}$ as follows:
\[
\phi(x)=\log{p_{x_{1}}}=\log\left(\frac{0.67569}{(i)^{\frac{6}{5}}(\log(i+2))^{2}}\right)
\]
and
\[
\psi(x)=\log|G'(\pi(x))|.
\]
Hence, we define the locally constant potentials $\bar{\phi}$ and $\bar{\psi}$ as follows:
\[\bar{\phi}(x):=\log{p_{x_{1}}} \text{ and } \bar{\psi}(x):=-\log\left({\frac{6}{\pi^{2}{x_{1}}^{2}}}\right)\] such that
\[p_{i}=\frac{0.67569}{(i)^{\frac{6}{5}}(\log(i+2))^{2}}\]
for each $i \in \mathbb{N}.$
We will again find an explicit expression for $\tilde{t}(q).$ Let $\bar{j}=(j,j,...)$ for any $j \in \mathbb{N}.$ We have
\[
\alpha_{\lim}=\lim\limits_{j\rightarrow\infty}\frac{\bar{\phi}(\bar{j})}{-\bar{\psi}(\bar{j})}=\lim\limits_{j\rightarrow\infty}\frac{-\frac{6}{5}\log{j}}{-\log|G'(\pi(\bar{j}))|}=\lim\limits_{j\rightarrow\infty}\frac{-\frac{6}{5}\log{j}}{-2\log{j}}=\frac{3}{5}.
\]
Again, we have that $Z_{1}(-t\bar{\psi})<\infty$ for $t>\frac{1}{2},$ so
$t_{\infty}=\frac{1}{2}.$ Hence, we get that
\[
\tilde{t}(q)=-\frac{3}{5}q+\frac{1}{2}.
\]
We prove that it is possible for the set $Q$ to equal $[q_{0},\infty)$ for some $q_{0}>\frac{6}{5}.$ This involves using partial sums of $\mathcal{P}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})$ to bound $\mathcal{P}(q\phi-\tilde{t}(q)\psi)$ below for some $q<\frac{6}{5}$ and $Z_{1}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})$ to estimate $\mathcal{P}(q\phi-\tilde{t}(q)\psi)$ above for a fixed $q>\frac{6}{5}.$ First, we need to prove that there exists a $q>\frac{6}{5}$ such that $Z_{1}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})<1.$ Again, we let $V_{1,j}(q\bar{\psi}+-\tilde{t}(q)\bar{\psi})$ be the H\"older constant for $q\bar{\phi}-\tilde{t}(q)\bar{\psi}$ such that for every $i,x \in [j],$
\[|(q\bar{\phi}-\tilde{t}(q)\bar{\psi})(i)-(q\bar{\phi}-\tilde{t}(q)\bar{\psi})(x)| \le V_{1,j}(q\bar{\phi}-\tilde{t}(q)\bar{\psi}).\]
The preceding estimate gives us that
\begin{eqnarray}
Z_{1}(q\bar{\phi}-\tilde{t}(q)\bar{\psi}) &=& \sum_{x_{1}=1}^{\infty}\exp\sup\limits_{x\in[x_{1}]}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})(x) \nonumber \\
& \le &
\sum_{j=1}^{\infty}\exp(q\bar{\phi}-\tilde{t}(q)\bar{\psi})(\bar{j})+V_{1,j}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})) \nonumber \\
&<& \sum_{j=1}^{\infty}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}\exp(V_{1,j}(q\bar{\phi}-\tilde{t}(q)\bar{\psi}))). \nonumber
\end{eqnarray}
\par We notice the following: Let $\varepsilon>0$ and $N \in \mathbb{N}$ be large. Then,
\[\sum_{j=1}^{\infty}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}=\sum_{j=1}^{N-1}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}+\sum_{j=N}^{\infty}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}.\]
Let $f(q)=\sum_{j=N}^{\infty}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}=\sum_{j=N}^{\infty}\frac{((0.67569)(\frac{6}{\pi^{2}})^{-\frac{3}{5}})^{q}(\frac{6}{\pi^{2}})^{\frac{1}{2}}}{(j)(\log(j+2))^{2q}}.$
We have that $f'(q)=$
\[\sum_{j=N}^{\infty}\frac{((0.67569)(\frac{6}{\pi^{2}})^{-\frac{3}{5}})^{q}(\frac{6}{\pi^{2}})^{\frac{1}{2}}\log((0.67569)(\frac{6}{\pi^{2}})^{-\frac{3}{5}})}{(j)(\log(j+2))^{2q}}-2\sum_{j=N}^{\infty}\frac{((0.67569)(\frac{6}{\pi^{2}})^{-\frac{3}{5}})^{q}(\frac{6}{\pi^{2}})^{\frac{1}{2}}}{j\log\log(j+2)(\log(j+2))^{2q}}\]
which is less than $\varepsilon$ for a fixed $q > \frac{6}{5}.$ Hence, $\sum_{j=N}^{\infty}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}$ is a decreasing function with respect to $q.$ \par
Let $g(q)=\sum_{j=1}^{N-1}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}.$
Similar to what was shown earlier, $g'(q)<0.$
Hence,
\[\sum_{j=1}^{N-1}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}} \text{ and } \sum_{j=N}^{\infty}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}\]
are decreasing functions with respect to $q.$
Then, for a fixed $q>\frac{6}{5},$
\begin{equation}\label{eq:N}
\sum_{j=1}^{N-1}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}} \le \int_{1}^{N-1} \frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}\, \mathrm{d}\mu \le 1-\varepsilon
\end{equation}
\begin{equation}\label{eq:P}
\text{and} \sum_{j=N}^{\infty}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}} \le \int_{N}^{\infty} \frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}\, \mathrm{d}\mu \le \varepsilon.
\end{equation}
Therefore, by Equations (\ref{eq:N}) and (\ref{eq:P}),
\[Z_{1}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})<\sum_{j=1}^{\infty}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}\exp(V_{1}(q\bar{\phi}-\tilde{t}(q)\bar{\psi}))) \le 1\]
for a fixed $\widehat{q}>\frac{6}{5}$ (with $\widehat{q}$ not necessarily equal to $q_{0}$).
Since ${\bar{\phi}}$ and ${\bar{\psi}}$ approximate $\phi$ and $\psi$ respectively, we will prove that
there exists a value $1<q<\frac{6}{5}$ such that
\[\mathcal{P}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})>0.\]
Let $\varepsilon>0.$ It follows that
\[\mathcal{P}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})>\sum_{j=1}^{\infty}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}-\varepsilon\]
because $\bar{\psi}$ and $\bar{\phi}$ are locally constant. We will estimate $\mathcal{P}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})$ by using
\[\sum_{j=1}^{\infty}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}.\]
In particular, we get that
\[\sum_{j=1}^{\infty}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}>\sum_{j=1}^{25}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}>1\]
for $q=1.15.$ Since
\begin{equation}\label{eq:partialsum}
\sum_{j=1}^{N}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{-\frac{3}{5}q+\frac{1}{2}}}{(j)(\log(j+2))^{2q}}
\end{equation}
is a decreasing function with respect to $q$ and the value of sum (\ref{eq:partialsum}) increases
as $N \rightarrow \infty,$ there exists a value $1<q<\frac{6}{5}$ such that
\[\mathcal{P}(q{\bar{\phi}}+-\tilde{t}(q){\bar{\psi}})>0.\]
Since ${\bar{\phi}}$ and ${\bar{\psi}}$ approximate $\phi$ and $\psi$ respectively, we get that
there exists a value $1<q<\frac{6}{5}$ such that
\[\mathcal{P}(q\phi-\tilde{t}(q)\psi)>0.\]
This means that $T(q)=\tilde{t}(q)$ for all $\widehat{q}>\frac{6}{5}$ and $T(q)>\tilde{t}(q)$ for some $q<\frac{6}{5}.$ Thus, we have established that $Q=[q_{0},\infty)$ for $q_{0}>\frac{6}{5}.$ Now, we can consider $\alpha(q_{0})$ for $q_{0}>\frac{6}{5}.$ Using our earlier estimates, we have that
\begin{eqnarray}
\int -\phi\,\mathrm{d}\mu_{q} &\le&
\int -\bar{\phi}\,\mathrm{d}\mu_{q}=\sum_{j=1}^{\infty}-\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{T(q)}}{j^{\frac{6}{5}q+2T(q)}(\log(j+2))^{2q}}\log\left(\frac{0.67569}{j^{\frac{6}{5}}(\log(j+2))^{2}}\right)\nonumber \\
&<&\sum_{j=1}^{\infty}\frac{1}{j^{\frac{6}{5}q+2T(q)-1}(\log(j+2))^{2q-1}}<\infty \nonumber
\end{eqnarray}
and
\begin{eqnarray}
\int \psi\, \mathrm{d}\mu_{q} &\le&
\int \bar{\psi}\, \mathrm{d}\mu_{q} =-\sum_{j=1}^{\infty}\frac{(0.67569)^{q}(\frac{6}{\pi^{2}})^{T(q)}}{j^{\frac{6}{5}q+2T(q)}(\log(j+2))^{2q}}\log\left(\frac{6}{\pi^{2}j^{2}}\right) \nonumber \\
&<&\sum_{j=1}^{\infty}\frac{1}{j^{\frac{6}{5}q+2T(q)-1}(\log(j+2))^{2q-1}}<\infty \nonumber
\end{eqnarray}
when $q>\frac{6}{5}.$ We remark that both integrals are infinite if $q<\frac{5}{6}.$ Hence, if $q_{0}>q>\frac{6}{5},$
\[
\alpha(q_{0})=\lim\limits_{q\rightarrow q_{0}^{-}}\frac{\int \phi\, \mathrm{d}\mu_{q}}{\int -\psi\, \mathrm{d}\mu_{q}}>\alpha_{\lim}.
\]
Using the techniques from the proof of Theorem~\ref{thm:M1}, we get the following. Given that $q_{0}>\frac{6}{5},$ $\alpha(q_{0}) \ge \alpha_{\lim}=\alpha_{\inf}.$ In this case, $f_{\mu}(\alpha)$ is analytic on $(\alpha_{\inf}, \alpha(q_{0}))$ and $(\alpha(q_{0}),\alpha_{\sup}).$ The multifractal spectrum is an increasing linear function and equals $T(q_{0})+q_{0}\alpha$ on $(\alpha_{\inf},\alpha(q_{0})),$ is strictly concave on $(\alpha(q_{0}),\alpha_{\sup}),$ and has its maximum at $\alpha(0).$ The multifractal spectrum has its only phase transition at $\alpha(q_{0}).$
\subsubsection{Two and Three Phase Transitions}\hspace*{\fill} \par
For each $j \in \mathbb{N},$ let $0<p_{j}<1.$ Pick a $k\in\mathbb{N}$ and let \[p_{k}=\frac{C_{k}}{(k)^{\frac{6}{5}}(\log(k+2))^{2}}\] such that $C_{k}>1.$ For each $i \not= k \in \mathbb{N},$ take \[p_{i}=\frac{C}{(i)^{\frac{6}{5}}(\log(i+2))^{2}}\] with $0<C<1$ chosen so that $\sum_{i=1}^{\infty}p_{i}=1.$ For simplicity, we take $C=\frac{3}{5}$ because the introduction of $C_{k}$ must mean that $C<0.67569.$ Note that $p_{k}=\frac{C_{k}}{C}\frac{C}{k^{\frac{6}{5}}(\log(k+2))^{2}}.$ For each $x=(x_{1},...) \in \Sigma,$ define the potentials $\phi:\Sigma\rightarrow\mathbb{R}^{-}$ and $\psi:\Sigma\rightarrow\mathbb{R}^{+}$ as follows:
\[
\phi(x)=\log{p_{x_{1}}}=\log\left(\frac{C}{(i)^{\frac{6}{5}}(\log(i+2))^{2}}\right)
\]
if $x_{1}=i\not=k,$
\[
\phi(x)=\log{p_{x_{1}}}=\log\left(\frac{C_{k}}{(k)^{\frac{6}{5}}(\log(k+2))^{2}}\right)
\]
if $x_{1}=k,$ and
\[
\psi(x)=\log|G'(\pi(x))|.
\]
\par Let $\bar{j}=(j,j,...)$ for any $j \in \mathbb{N}.$
Again, we approximate $\phi$ and $\psi$ with locally constant functions $\bar{\phi}$ and $\bar{\psi}.$
We let $\bar{\phi}(x)=\phi(x)$ and
$\bar{\psi}(x)=-\log\left(\frac{6}{\pi^{2}x_{1}^{-2}}\right)$ for each $x \in \Sigma.$ We again have that $Z_{1}(-t\psi)<\infty$ for $t>\frac{1}{2},$ so
$\alpha_{\lim}=\frac{3}{5}$ and $t_{\infty}=\frac{1}{2}.$ Hence,
\[
\tilde{t}(q)=-\frac{3}{5}q+\frac{1}{2}.
\]
We will try to prove that it is possible for the set $Q$ to equal $[q_{0},q_{1}]$ for $q_{0}>\frac{5}{6}.$
Since the arguments are nearly identical to the previous example, we instead give an outline. This involves using $\bar{\phi}$ and $\bar{\psi}$ to estimate $\mathcal{P}(q\phi-\tilde{t}(q)\psi)$ below for $q<\frac{5}{6},$ $Z_{1}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})$ to estimate $\mathcal{P}(q\phi-\tilde{t}(q)\psi)$ above for $q_{0}>q>\frac{5}{6},$ and again $\bar{\phi}$ and $\bar{\psi}$ to estimate $\mathcal{P}(q\phi-\tilde{t}(q)\psi)$ below for $\frac{5}{6}<q<q_{1}<\infty.$ In the case that $Q=[q_{0},q_{1}]$ such that $q_{0}<\frac{5}{6}<q_{1},$ the only modification to the previous argument is the need to use $Z_{1}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})$ to estimate $\mathcal{P}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})$ above for $q<\frac{5}{6}.$ For some $q<\frac{5}{6},$ $Z_{1}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})<1$ so $\mathcal{P}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})<0.$
To prove that $T(q)$ has a second phase transition at $q_{1},$ we consider the following. Using simple analysis, we find that we need to satisfy
\[\frac{\log{p_{k}}}{-2\log(k)}>\alpha_{\lim}\] in order for $q_{1}$ to exist (because this would mean that there exists $q>q_{1}$ such that $\alpha(q)>\alpha_{\lim}$).
Let $\varepsilon>0.$ Since $\alpha_{\lim}=\lim\limits_{j \rightarrow \infty}\frac{\log{p_{j}}}{\log|G'(\pi(\bar{j}))|}=\lim\limits_{j \rightarrow \infty}\frac{\log{p_{j}}}{2\log(j)},$ it follows that for each $j \ge J$ for some large $J \in \mathbb{N},$
\[\alpha_{\lim}-\varepsilon \le \frac{\log{p_{j}}}{-2\log{j}} \le \alpha_{\lim}+\varepsilon.\]
Thus, for $j \ge k,$ we must satisfy
\[\frac{\log{p_{k}}}{-2\log(k)}> \frac{\log{p_{j}}}{-2\log{j}}.\]
Hence, choosing $C_{k}>1$ gives us the second phase transition for $T(q).$
Now, we must analyse the behaviour of $\alpha(q)$ at $q_{0}$ and $q_{1}.$ The work to show that \[\int -\bar{\phi}\, \mathrm{d}\mu_{q}<\infty \text{ and} \int \bar{\psi}\, \mathrm{d}\mu_{q}<\infty\] for $q>\frac{5}{6}$ is identical to the previous example. Hence, if $q<q_{0}\le \frac{5}{6},$ \[\int -\bar{\phi}\, \mathrm{d}\mu_{q}=\infty \text{ and} \int \bar{\psi}\, \mathrm{d}\mu_{q}=\infty.\]
The same results are true for $q_{1}.$ \par
Therefore, using the techniques from the proof of Theorem~\ref{thm:M1}, we get either two or three phase transitions for our chosen $\phi$ and $\psi,$ depending on the values of $q_{0}$ and $q_{1}:$
\begin{enumerate}
\item If $q_{0} \le \frac{5}{6} < q_{1},$
$\alpha(q_{0})=\alpha_{\lim}>\alpha(q_{1}).$ The multifractal spectrum is analytic on $(\alpha_{\inf},\alpha(q_{1})),$ $(\alpha(q_{1}),\alpha(q_{0})),$ and $(\alpha(q_{0}),\alpha_{\sup}).$ Furthermore, $f_{\mu}(\alpha)$ is strictly concave on $(\alpha_{\inf},\alpha(q_{1}))$ as well as $(\alpha(q_{0}),\alpha_{\sup}),$ is linear and equals $T(q_{1})+q_{1}\alpha$ on $(\alpha(q_{1}),\alpha(q_{0})),$ and has its maximum at $\alpha(0).$ The multifractal spectrum has phase transitions at $\alpha_{\lim}$ and $\alpha(q_{1})$ in this case.
\item If $\frac{5}{6}<q_{0}<q_{1},$
$\alpha(q_{0})>\alpha_{\lim}>\alpha(q_{1}).$ The multifractal spectrum is analytic on $(\alpha_{\inf},\alpha(q_{1})),$ $(\alpha(q_{1}),\alpha_{\lim}),$ $(\alpha_{\lim},\alpha(q_{0})),$ and $(\alpha(q_{0}),\alpha_{\sup}).$ Furthermore, $f_{\mu}(\alpha)$ is strictly concave on $(\alpha_{\inf},\alpha(q_{1}))$ as well as $(\alpha(q_{0}),\alpha_{\sup}),$ is linear and equals $T(q_{1})+q_{1}\alpha$ on $(\alpha(q_{1}),\alpha_{\lim}),$ is linear and equals $T(q_{0})+q_{0}\alpha$ on $(\alpha_{\lim},\alpha(q_{0})),$ and has its maximum at $\alpha(0).$ Then, the multfractal spectrum has phase transitions at $\alpha(q_{0}), \alpha_{\lim}, \text{ and } \alpha(q_{1})$ in this case.
\end{enumerate}
In fact, Proposition~\ref{prop:PCD1} gives these results.
\subsection{The Case $\alpha_{\lim}=\infty$}\hspace*{\fill} \par
We provide the following example in which $\alpha_{\lim}=\infty.$ Let $\Sigma=\mathbb{N}^{\mathbb{N}}.$ Define $\phi:\Sigma \rightarrow \mathbb{R}^{-}$ and $\psi:\Sigma \rightarrow \mathbb{R}^{+}$ as follows. For each $i \in \mathbb{N},$ let
$p_{i}=\left(\frac{1}{2}\right)^{i}.$ We remark that $p_{i}$ is exactly the Minkowski $?-$function. Clearly, $\sum_{i=1}^{\infty}p_{i}=1.$ Again, we approximate $\psi(x)=\log|G'(\pi(x))|$ with $\bar{\psi}(x)=-\log\left(\frac{6}{\pi^{2}x_{1}^{2}}\right)$ and we let
\[
\phi(x)=\bar{\phi}(x)=\log{p_{x_{1}}}=\log\left(\left(\frac{1}{2}\right)^{x_{1}}\right).
\]
for each $x \in \Sigma.$
Let $\bar{j}=(j,j,j,...)$ such that $j \in \mathbb{N}.$ Since $\bar{\phi}$ and $\bar{\psi}$ are locally constant, we can approximate $\mathcal{P}(q\phi-t\psi)$ with
\[\mathcal{P}(q\bar{\phi}-t\bar{\psi})=\log{Z_{1}(q\bar{\phi}-t\bar{\psi})}=\log\left(\sum_{j=1}^{\infty}\frac{(\frac{6}{\pi^{2}})^{t}}{2^{jq}(j)^{2t}}\right).\]
We notice that
\begin{enumerate}
\item For each $q<0,$ \[ \log\left(\sum_{j=1}^{\infty}\frac{(\frac{6}{\pi^{2}})^{t}}{2^{jq}(j)^{2t}}\right)=\infty\]
independent of the choice of $t\in\mathbb{R}.$ Then, $T(q)=\tilde{t}(q)=\infty$ for these $q.$
\item If $q=0,$
\[\log\left(\sum_{j=1}^{\infty}\frac{(\frac{6}{\pi^{2}})^{t}}{(j)^{2t}}\right)<\infty\]
for every $t>\frac{1}{2}\in\mathbb{R}.$ Hence, $\tilde{t}(0)=\frac{1}{2}<T(0).$ In fact, $T(0)=1.$
\item For each $q>0,$ \[\log\left(\sum_{j=1}^{\infty}\frac{(\frac{6}{\pi^{2}})^{t}}{2^{jq}(j)^{2t}}\right)<\infty\]
independent of the choice of $t\in\mathbb{R}.$ Then, $-\infty=\tilde{t}(q)<T(q)$ for these $q.$
\end{enumerate}
\par By Lemma~\ref{lemma:MU1}, we get that
$$\tilde{t}(q) =
\left\{
\begin{array}{ll}
\infty & \mbox{if } q < 0 \\
\frac{1}{2} & \mbox{if } q = 0 \\
-\infty & \mbox{if } q > 0.
\end{array}
\right.$$
It follows that
$$Q=\{q \in \mathbb{R}: T(q)=\tilde{t}(q)\}=(-\infty,0).$$
By convention,
$$\alpha(0)=\lim\limits_{q \rightarrow 0^{+}}\alpha(q).$$
Also, notice that
$\alpha_{\lim}=\lim\limits_{i \rightarrow \infty}\frac{\log{p_{i}}}{\log{s_{i}}}=\infty.$ Then, $\alpha(0)<\alpha_{\lim}=\infty=\alpha_{\sup}$ because $T(0)=1>\tilde{t}(0)$ gives us that \[\lim\limits_{q \rightarrow 0^{+}}\int-\phi\, \mathrm{d}\mu_{q} \le \lim\limits_{q \rightarrow 0^{+}}\sum_{j=1}^{\infty}\frac{(\frac{6}{\pi^{2}})^{T(q)}j\log(2)}{2^{jq}(j)^{2T(q)}} \in (-\infty,\infty)\] \[ \text{and} \lim\limits_{q \rightarrow 0^{+}}\int\psi\, \mathrm{d}\mu_{q}\le \lim\limits_{q \rightarrow 0^{+}}\sum_{j=1}^{\infty}\frac{(\frac{6}{\pi^{2}})^{T(q)}[2\log(j)-\log(\frac{6}{\pi^{2}})]}{2^{jq}(j)^{2T(q)}} \in (-\infty,\infty).\]
Hence, we have the following analysis for the behaviour of the multifractal spectrum.
\begin{proposition}
For this example, the multifractal spectrum is increasing and analytic on $(\alpha_{\inf},\alpha(0)).$ $f_{\mu}(\alpha)=T(0)$ on $(\alpha(0),\infty).$ Furthermore, the multifractal spectrum has a phase transition at $\alpha(0).$
\end{proposition}
\begin{proof}
Each $\alpha \in (\alpha_{\inf},\alpha(0))$ satisfies $\alpha=\alpha(q)$ for a unique $q>0.$ Since each of these $q$ are in $Q^{\complement},$ $f_{\mu}(\alpha)$ is analytic on $(\alpha_{\inf},\alpha(0)).$ The multifractal spectrum increases on $(\alpha_{\inf},\alpha(0))$ follows from Proposition~\ref{prop:ID1}. For $\alpha \in (\alpha(0),\alpha_{\lim}),$ $f_{\mu}(\alpha)=T(0).$ Hence, the multifractal spectrum is constant on $(\alpha(0),\alpha_{\lim}).$
\end{proof}
\subsection{The Case When $\alpha_{\lim}$ Does Not Exist}\hspace*{\fill} \par
When $\alpha_{\lim}$ does not exist, the multifractal spectrum has up to infinitely many phase transitions. We remark that Iommi and Jordan \cite{iommi2013phase} create a similar example in the setting of a suspension flow. Now, we roughly outline the procedure for creating such an example in our setting. First, we take the locally H\"older potentials $\phi$ and $\psi$ such that
\[\phi(x)=\log{p_{x_{1}}} \text{ and } \psi(x)=\log|G'(\pi(x))|\]
for each $x=(x_{1},x_{2},...)\in\Sigma.$
Then, to define the $p_{i}$ for each $i \in \mathbb{N},$ we partition the natural numbers as follows. Let $r_{0}=0$ and $r_{1}=1.$ Consider the infinite sequence $\{r_{k}\}_{k=2}^{\infty}$ of primes $\{2,3,5,7,11,...\}.$ We define the sets $\{I_{k}\}_{k \in \mathbb{N}_{0}}$ as follows:
\[I_{0}:=\{m \in \mathbb{N} \text{ such that } m \text{ cannot be written as any prime power of any } n \in \mathbb{N}\}.\]
\[I_{1}:=\{m \in \mathbb{N} \text{ that can be written as the } 2\text{nd power of some } n \in \mathbb{N} \}\]
In general,
\[I_{k}:=\{m \in \mathbb{N} \text{ that can be written as the } r_{k+1}\text{st power of some } n \in \mathbb{N} \}.\]
For each $m \in \mathbb{N},$ we get that
\[p_{m} =\frac{C_{k}}{m^{l_{k}}(\log(m+2))^{M_{k}}}\]
if $k \in\mathbb{N}_{0}$ and $m \in I_{k}.$ We have increasing sequences of constants $\{C_{k}\}_{k \in \mathbb{N}_{0}}$ and $\{M_{k}\}_{k \in \mathbb{N}_{0}}.$ The terms of both sequences are chosen such that $\sum_{m=1}^{\infty}p_{m}=1.$ For each $k \in \mathbb{N} \cup \{0\},$ we have a recursive relation for the values of the sequence $\{l_{k}\}$ stated in the expression for $p_{m}.$
We remind the reader that locally H\"older potentials can be approximated by locally constant potentials. For each $a_{1}(x)=m \in \mathbb{N},$ define
\[s_{m}:= \frac{6}{\pi^{2}m^{2}}\] for each $m \in \mathbb{N};$ hence, we can estimate $\psi$ with $\bar{\psi}(x)=\log{s_{x_{1}}^{-1}}.$ \par
For each $I_{k},$ we have a function $\tilde{t}_{k}(q)$ as follows:
\[\tilde{t}_{k}(q):=\inf\{t \in \mathbb{R}: \sum_{I_{k}}p_{m}^{q}s_{m}^{\tilde{t}_{k}(q)}<\infty\}.\]
We define
\[\tilde{t}(q)=\left\{\sup_{q \in \mathbb{R}}\tilde{t}_{k}(q): k\in \mathbb{N} \cup \{0\}\right\}.\]
By construction, $|\tilde{t}_{k}'|>|\tilde{t}_{k+1}'|$ and each $\tilde{t}_{k}$ is linear. We get that the phase transitions for $\tilde{t}(q)$ occur at values of $q$ such that $\tilde{t}_{k}(q)=\tilde{t}_{k+1}(q).$ Proceeding with the computation of these $q,$ we find that they occur at each $q \in \mathbb{N}.$ Hence, $\tilde{t}(q)$ has infinitely many phase transitions.
Finally, we prove that $Z_{1}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})<1$
for each $q \in (k,k+1) \cup (a,1)$ for some $a \in \mathbb{R}$ and for each $k \in \mathbb{N}.$ This gives us that $\mathcal{P}(q\bar{\phi}-\tilde{t}(q)\bar{\psi})<1$
for those $q;$ hence, $T(q)=\tilde{t}(q)$ for all $q \ge \bar{q} \ge a$ (for some $1> \bar{q} \ge a$). Using techniques from the previous example, we get results for a possible phase transition for the multifractal spectrum at $\alpha(\bar{q}).$ Without loss of generality, let us assume that there is no phase transition at $\alpha(\bar{q}).$ Hence, we have the following behaviour. $f_{\mu}(\alpha)$ is analytic on $(\alpha(0),\alpha_{\sup}),$ $(\alpha(1),\alpha(0)),$ $(\alpha(2),\alpha(1)),$...,$(\alpha(N),\alpha(N-1)),$.... The phase transitions for $f_{\mu}(\alpha)$ are at $\alpha(1),$ $\alpha(2),$...$,\alpha(N),$.... The multifractal spectrum increases and is piecewise linear on $(\alpha_{\inf},\alpha(N)),$..., equals $T(3)+3\alpha$ and increases on $(\alpha(3),\alpha(2)),$ equals $T(2)+2\alpha$ and increases on $(\alpha(2),\alpha(1)),$ equals $\alpha$ and increases on $(\alpha(1),\alpha(0)),$ and finally, decreases on $(\alpha(0),\alpha_{\sup}).$ Finally, we remark that the existence of $\alpha_{\lim}$ is absolutely necessary for Theorems~\ref{thm:M1} and \ref{thm:M2} to be true.
| af8d63ef8e2294c31f5a06944b1f8343d55c8aba | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Modified gravity theories may provide an attractive alternative to the standard explanations of the present day observations that have shaken the well-established foundations of the theoretical physics. Astronomical observations have confirmed that our Universe does not according to standard general relativity, as derived from the Hilbert-Einstein action, $S=\int{\left(-R/2\kappa ^2+L_m\right)\sqrt{-g}d^4x}$, where $R$ is the Ricci scalar, $\kappa $ is the gravitational coupling constant, and $L_m$ is the matter Lagrangian, respectively. Extremely successful on the Solar System scale, somehow unexpectedly, general relativity faces, on a fundamental theoretical level, two important challenges, the dark energy and the dark matter problems, respectively. Several high precision astronomical observations, with the initial goal of improving the numerical values of the basic cosmological parameters by using the properties of the distant
Type Ia Supernovae, have provided the result that the Universe underwent recently a transition to an accelerating, de Sitter type phase \cite{1n,2n,3n,4n,acc}. The necessity of explaining the late time acceleration lead to the formulation of a new paradigm in theoretical physics and cosmology, which postulates that the explanation of the late time acceleration is the existence of a mysterious component of the Universe, called
dark energy (DE), which can describe the
late time dynamics of the Universe \cite{PeRa03,Pa03}, and can explain all the observed features of the recent (and future) cosmological evolution. However, in order to close the matter-energy balance of the Universe, a second, and equally mysterious component, called Dark Matter, is required. Dark Matter, assumed to be a non-baryonic and non-relativistic (cold) component of the Universe, is necessary for explaining the dynamics of the hydrogen clouds rotating around galaxies, and having flat rotation curves, as well as
the virial mass discrepancy in clusters of galaxies \cite{dm1,dm2}. The direct detection/observation of the dark matter is extremely difficult due to the fact that it interacts only gravitationally with the baryonic matter. After many decades of intensive observational and experimental efforts there is no direct evidence on the particle nature of the dark matter.
One of the best theoretical descriptions that fits almost perfectly the observational data is based on the simplest theoretical extension of general relativity, which includes in the gravitational field equations the cosmological constant $\Lambda $ \cite{Wein,Wein1}. Based on this theoretical formalism the basic paradigm of modern cosmology has been formulated as the $\Lambda $CDM-$\Lambda $ Cold Dark Matter Model, in which the dark energy is nothing but the simple constant introduced almost one hundred years ago by Einstein. Even that the $\Lambda $CDM model fits the data well, it raises some fundamental theoretical questions about the possibility of explaining it. There is no theoretical explanation for the physical/geometrical nature of $\Lambda $, and, moreover, general relativity cannot give any hints on why it is so small, and why it is so fine tuned \cite{Wein,Wein1}. Therefore, the possibility that dark energy can be explained as an intrinsic property of a {\it generalized gravity theory}, going beyond general relativity, and its Hilbert-Einstein gravitational action, cannot be rejected a priori. In this context a large number of modified gravity models, all trying to extend and generalize the standard Einsteinian theory of gravity, have been proposed. Historically, in going beyond Einstein gravity, the first, and most natural step, was to extend {\it the geometric part of the Hilbert-Einstein action}. One of the first attempts in this direction is represented by the $f(R)$ gravity theory, in
which the gravitational action is generalized to be an arbitrary function of the Ricci scalar, so that $S = \frac{1}{2\kappa ^2}\int {f(R) \sqrt{-g}d^4 x} +\int {L_m \sqrt{-g}d^4 x}$ \cite{Bu701,re4,Bu702,Bu703,re1,Bu704,Fel}. However, this as well as several other modifications of the Hilbert-Einstein action focussed only on the geometric part of the gravitational action, by explicitly postulating that the matter Lagrangian plays {\it a subordinate and passive role} only, as compared to the geometry \cite{Mat}. From a technical point of view such an approach implies a minimal coupling between matter and geometry. But a fundamental theoretical principle, forbidding an arbitrary coupling between matter and geometry, has not been formulated yet, and perhaps it may simply not exist. On the other hand, if general matter-geometry couplings are introduced, a large number of theoretical gravitational models, with extremely interesting physical and cosmological properties, can be easily constructed.
The first of the modified gravity theory with arbitrary geometry-matter coupling was the $f\left( R,L_{m}\right) $ modified gravity theory \cite{fL1,fL2,fL3,fL4}, in which the total gravitational action takes the form $S = \frac{1}{2\kappa ^2}\int {f\left(R,L_m\right) \sqrt{-g}d^4 x} $. In this kind of theories matter is essentially indistinguishable from geometry, and plays an active role in generating the geometrical properties of the space-time. A different geometry-matter coupling is introduced in the
$f(R,T)$ \cite{fT1,fT2} gravity theory, where matter and geometry are coupled via $T$, the trace of the energy-momentum tensor. The gravitational action of the $f(R,T)$ theory is given by $S=\int{\left[f(R,T)/2\kappa ^2+L_m\right]\sqrt{-g}d^4x}$. A recent review of the generalized $f\left(R,L_{m}\right)$ and $f(R,T)$ type
gravitational theories with non-minimal curvature-matter couplings can be found in \cite{Revn}. Several other gravitational theories involving geometry-matter couplings have also been proposed, and extensively studied, like, for example, the
Weyl-Cartan-Weitzenb\"{o}ck (WCW) gravity theory \cite{WCW}, hybrid metric-Palatini
f(R,\mathcal{R})$ gravity \cite{HM1,HM2,Revn1}, where $\mathcal{R}$ is the Ricci scalar
formed from a connection independent of the metric,
f\left(R,T,R_{\mu \nu }T^{\mu \nu }\right)$ gravity theory, where $R_{\mu \nu
} $ is the Ricci tensor, and $T_{\mu \nu }$ the matter energy-momentum
tensor, respectively \cite{Har4,Odin}, or $f(\tilde{T},\mathcal{T})$ gravity \cite{HT}, in which a
coupling between the torsion scalar $\tilde{T}$, essentially a geometric quantity, and the trace $T$ of the matter
energy-momentum tensor is introduced. Gravitational models with higher derivative matter fields were investigated in \cite{HDM}.
One of the interesting (and intriguing) properties of the gravitational theories with geometry-matter coupling is the non-conservation of the matter energy-momentum tensor, whose four-divergence is usually different of zero, $\nabla _{\mu}T^{\mu \nu}\neq 0$. This property can be interpreted, from a thermodynamic point of view, by using the formalism of open thermodynamic systems \cite{fT2}. Hence one can assume that the generalized conservation equations in these gravitational theories describe {\it irreversible matter creation processes}. Thus the non-conservation of the energy-momentum tensor describes an irreversible energy flow from the gravitational field to the newly created matter constituents, with the second law of thermodynamics requiring that space-time transforms into matter. In \cite{fT2} the equivalent particle number creation rates, the creation pressure and the entropy production rates were obtained for both $f\left(R,L_m \right)$ and $f(R,T)$ gravity theories. The temperature evolution laws of the newly created particles was also obtained, and studied. Due to the non-conservation of the energy-momentum tensor, which is a direct consequence of the geometry--matter coupling, during the cosmological evolution of the Universe a large amount of comoving entropy could be also produced.
The prediction of the production of particles from the cosmological vacuum is one of the remarkable results of the quantum field theory in curved space-times \cite{P1,Z1,P2,Full,P3}. Particles creation processes are supposed to play a fundamental role in the quantum field theoretical approaches to gravity, where they naturally appear. It is a standard result of quantum field theory in curved spacetimes that quanta of the minimally-coupled scalar field are created in the expanding Friedmann-Robertson-Walker Universe \cite{P3}. Therefore, the presence of particle creation processes in both quantum theories of gravity and modified gravity theories with geometry-matter coupling may suggest that a deep connection between these two, apparently very different physical theories, may exist. And, interestingly enough, such a connection has been found in \cite{re11}, where it was pointed out that by using a nonperturbative approach for the quantization of the metric, proposed in \cite{re8,re9,re10}, a particular type of $f(R,T)$ gravity, with Lagrangian given by $L=\left[(1-\alpha )R/2\kappa ^2+\left(L_m-\alpha T/2\right)\right]\sqrt{-g}$, where $\alpha $ is a constant, naturally emerges as a result of the quantum fluctuations of the metric. This result suggests that an equivalent microscopic quantum description of the matter creation processes in $f(R,T)$ or $f\left(R,L_m\right)$ gravity is possible, and such a description could shed some light on the physical mechanisms leading to particle generation via gravity and matter geometry coupling. Such mechanisms do indeed exist, and they can be understood, at least qualitatively, in the framework of some quantum/semiclassical gravity models.
It is the goal of the present paper to further investigate the cosmological implications of modified gravities induced by the quantum fluctuations of the gravitational metric, as initiated in \cite{re11,re8,re9,re10}. As a starting point we assume that a general quantum metric can be decomposed as the sum of the classical and of a fluctuating part, the latter being of quantum (or stochastic) origin. If such a decomposition is possible, the corresponding Einstein quantum gravity generates at the classical level modified gravity models with a nonminimal interaction between geometry and matter, as previously considered in \cite{fL1, fT1,Har4}. After assuming that the expectation value of the quantum correction can be generally expressed in terms of an arbitrary second order tensor $K_{\mu \nu}$, which can be constructed from the metric and from the thermodynamic quantities characterizing the matter content of the Universe, we derive from the first order quantum gravitational action the (classical) gravitational field equations in their general form. We analyze in detail the cosmological models obtained from the quantum fluctuations of the metric in tqo cases. First we assume that the quantum correction tensor $K_{\mu \nu}$ is given by the coupling of a scalar field and of a scalar function to the metric tensor, respectively. As a second case we consider that $K_{\mu \nu}$ is given by a term proportional to the matter energy-momentum tensor. The first choice gives a particular version of the $f(R,T)$ gravity model \cite{fT1}, while the second choice corresponds to specific case of the modified gravity theory of the form $f\left(R,T,R_{\mu \nu}T^{\mu \nu},T_{\mu \nu}T^{\mu \nu}\right)$ \cite{Har4}. For each considered model we obtain the gravitational field equations, and the generalized Friedmann equations for the case of a flat homogeneous and isotropic geometry. In some of these models the divergence of the matter energy-momentum tensor is non-zero, indicating a process of matter creation. From a physical point of view a non-zero divergence of the energy-momentum tensor can be interpreted as corresponding to an irreversible energy flow from the gravitational field to the matter fluid. Such an irreversible thermodynamic process is the direct consequence of the nonminimal curvature-matter coupling, induced in the present case by the quantum fluctuations of the metric \cite{fT2, creat, Pavon}. The cosmological evolution equations of these modified gravity models induced by the quantum fluctuations of the metric are investigated in detail by using both analytical and numerical methods. As a result of this analysis we show that a large variety of cosmological models can be constructed. Depending on the numerical values of the model parameters, these cosmological models can exhibit both late time accelerating, or decelerating behaviors.
The present paper is organized as follows. In Section~\ref{sect2} we derive the general set of field equations induced by the quantum fluctuations of the metric. The relation between this approach and the standard semi-classical formulation of quantum gravity is also briefly discussed. In Section~\ref{sect3} we investigate in detail the cosmological implications of the quantum fluctuations induced modified gravity models with the fluctuation tensor proportional to the metric. Two cases are considered, in which the fluctuation couples to the metric via a scalar field, and a scalar function, respectively. Modified gravity models induced by quantum metric fluctuations proportional to the energy-momentum tensor are investigate in Section~\ref{sect4}. Finally, we discuss and conclude our results in Section~\ref{sect5}. The details of the derivation of the gravitational field equations for an arbitrary metric fluctuation tensor and for a fluctuation tensor proportional to the matter energy-momentum tensor are presented in Appendices~\ref{App1} and \ref{App2}, respectively. In the present paper we use a system of units with $c=1$.
\section{Modified gravity from quantum metric fluctuations}\label{sect2}
In the present Section we will briefly review the fluctuating metric approach to quantum gravity, we will discuss its relation with standard semiclassical gravity, and we will point out the quantum mechanical origins of the modified gravity models with geometry-matter coupling. Moreover, we derive the general set of field equations induced by the quantum fluctuations of the metric for an arbitrary form of the tensor $K_{\mu \nu}$.
\subsection{Modified gravity as the semiclassical approximation of quantum gravity}
In the standard quantum mechanics physical (or geometrical) quantities must be represented by operators. Therefore, in a full non-perturbative quantum approach the Einstein gravitational field equations must take an operator form, given by \cite{re8,re9,re10}
\begin{eqnarray}
\hat{R}_{\mu\nu}-\frac{1}{2}R\hat{g}_{\mu\nu}=\frac{8\pi G}{c^4}\hat{T}_{\mu\nu}.
\end{eqnarray}
As shown in \cite{re8,re9,re10}, in order to extract meaningful physical information from the Einstein operator equations one must average it over all possible products of the metric operators $\hat {g}\left(x_1\right)...\hat{g}\left(x_n\right)$, and thus to solve the infinite set of equations
\begin{eqnarray*}
<Q|\hat{g}(x_1)\hat{G}_{\mu\nu}|Q>&=&<Q|\hat{g}(x_1) \hat{T}_{\mu\nu}|Q>,\\
<Q|\hat{g}(x_1)\hat{g}(x_2) \hat{G}_{\mu\nu}|Q>&=&<Q|\hat{g}(x_1)\hat{g}(x_2) \hat{T}_{\mu\nu}|Q>,\\
\dots&=&\dots,
\end{eqnarray*}
for the Green functions $\hat{G}_{\mu\nu}$. In the above equations \(|Q>\) is quantum state that might not be the ordinary vacuum state. These equations cannot be solved analytically, and hence we have to use some approximations \cite{re8}-\cite{re10}. In \cite{re8} it was suggested to decompose the metric operator into the sum of an average metric $g_{\mu \nu}$, and a fluctuating part $\delta\hat{g}_{\mu\nu}$, according to
\begin{eqnarray}\label{c1}
\hat{g}_{\mu\nu}=g_{\mu\nu}+\delta\hat{g}_{\mu\nu}.
\end{eqnarray}
Assuming that
\be\label{c2}
<\delta \hat{g}_{\mu\nu}>=K_{\mu\nu}\ne 0,
\ee
where $K_{\mu\nu}$ is a classical tensor quantity, and ignoring higher order fluctuations, the gravitational Lagrangian will be modified into \cite{re8}
\begin{eqnarray}\label{2}
\mathcal{L}&=&\mathcal{L}_g\left(\hat{g}_{\mu\nu}\right)+
\sqrt{-g}\mathcal{L}_m\left(\hat{g}_{\mu\nu}\right)\approx \mathcal{L}_g+\frac{\delta\mathcal{L}_g}{\delta g^{\mu\nu}}\delta\hat{g}^{\mu\nu}+\nonumber\\
&&\sqrt{-g}\mathcal{L}_m+
\frac{\delta(\sqrt{-g}\mathcal{L}_m)}{\delta g^{\mu\nu}}\delta\hat{g}^{\mu\nu}=-\frac{1}{2\kappa ^2}\sqrt{-g}\times \nonumber\\ &&\left(R+G_{\mu\nu}\delta\hat{g}^{\mu\nu}\right)+
\sqrt{-g}\Bigg(\mathcal{L}_m+\frac{1}{2}T_{\mu\nu}\delta\hat{g}^{\mu\nu}\Bigg),
\end{eqnarray}
where \(\kappa ^2=8\pi G/c^4\), and where we have defined the matter energy-momentum tensor as
\begin{eqnarray}
T_{_\mu\nu}=\frac{2}{\sqrt{-g}}\frac{\delta \left(\sqrt{-g}\mathcal{L}_m\right)}{\delta g^{\mu\nu}}.
\end{eqnarray}
Even that the above formalism starts from a full quantum approach of gravity, after performing the decomposition of the metric we are still considering semiclassical theories. In this paper we will consider several functional forms of the so called quantum perturbation tensor \(K^{\mu\nu}\), we will obtain the field equations of the corresponding gravity theory, and we will investigate their cosmological implications, respectively.
The gravitational field equations corresponding to the first order corrected quantum Lagrangian (\ref{2}) are given, in a general form, by
\bea\label{genfe}
G_{\mu\nu}=\kappa^2T_{\mu\nu}-
\Bigg\{\frac{1}{2}g_{\mu\nu}G_{\alpha\beta}K^{\alpha\beta}+\frac{1}{2}\times\nn
\Bigg(\Box K_{\mu\nu}+\nabla_{\alpha}\nabla_{\beta}K^{\alpha\beta}g_{\mu\nu}-
\nabla_{\alpha}\nabla_{(\mu}K_{\nu)}^\alpha\Bigg)+\nn
\gamma^{\alpha\beta}_{\mu\nu}R_{\alpha\beta}-\frac{1}{2}\Bigg[RK_{\mu\nu}+KR_{\mu\nu}+\gamma^{\alpha\beta}_{\mu\nu}(R g_{\alpha\beta})+\nn
\nabla_\mu\nabla_\nu K+g_{\mu\nu}\Box K\Bigg]\Bigg\}+\kappa^2\Bigg\{-\frac{1}{2}g_{\mu\nu}T_{\alpha\beta}K^{\alpha\beta}+\nn
\Bigg[\gamma^{\alpha\beta}_{\mu\nu}T_{\alpha\beta}+K^{\alpha\beta}\Bigg(2\frac{\delta^2 \mathcal{L}_m}{\delta g^{\mu\nu}\delta g^{\alpha\beta}}-\frac{1}{2}g_{\alpha\beta}T_{\mu\nu}\nn
-\frac{1}{2}g_{\mu\nu}g_{\alpha\beta}\mathcal{L}_m-\mathcal{L}_m\frac{\delta{g_{\alpha\beta}}}{\delta g^{\mu\nu}}\Bigg)\Bigg]\Bigg\},
\eea
where \(K=g_{\mu\nu}K^{\mu\nu}\)and \(A_{\alpha\beta}\delta K^{\alpha\beta}=\delta g^{\mu\nu}(\gamma^{\alpha\beta}_{\mu\nu}A_{\alpha\beta})\). Here \(A_{\alpha\beta}\) is either $R_{\alpha \beta}$ or $T_{\alpha \beta }$, and \(\gamma^{\alpha\beta}_{\mu\nu}\) is an algebraic tensor, an operator, or the combination of them. The detailed derivation of Eqs.~(\ref{genfe}) is presented in Appendix~\ref{App1}.
It is interesting to compare the formalism based on the quantum metric fluctuation proposal to the standard semiclassical gravity approach, which also leads to particle production via the non-conservation of the matter energy-momentum tensor \cite{Pavon}. Semiclassical gravity is constructed from the basic assumption that the gravitational field is still {\it classical}, while the {\it classical matter (bosonic) fields} $\phi $ are taken as {\it quantized}. The direct coupling of the quantized matter fields to the classical gravitational fields is performed via the replacement of the quantum energy momentum tensor $\hat{T}_{\mu \nu}$ by its expectation value $\left <\hat{T}_{\mu \nu}\right >$, obtained with respect to some quantum state $\Psi $. Therefore the effective semiclassical Einstein equation can be {\it postulated} as \cite{Carl},
\be\label{gr1}
R_{\mu \nu}-\frac{1}{2}g_{\mu \nu}R=\frac{8\pi G}{c^4}\left<\Psi \right |\hat{T}_{\mu \nu}\left |\Psi \right>.
\ee
From Eq.~(\ref{gr1}) it follows immediately that the classical energy-momentum tensor $T_{\mu \nu}$ of the gravitating system is obtained from its quantum counterpart through the definition $\left<\Psi \right |\hat{T}_{\mu \nu}\left |\Psi \right>=T_{\mu \nu}$.
The semiclassical Einstein equations ~(\ref{gr1}) can be derived from the variational principle \cite{Kibble}
\be\label{quantac}
\delta \left(S_g+S_{\psi}\right)=0,
\ee
where $S_g=\left(1/16\pi G\right)\int{R\sqrt{-g}d^4x}$ is the standard general relativistic classical action of the gravitational field, while the quantum part of the action is given by
\be\label{9}
S_{\Psi}=\int{\left[{\rm Im}\left \langle \dot{\Psi}|\Psi\right \rangle-\left \langle \Psi |\hat{H}|\Psi \right \rangle +\alpha \left(\left \langle \Psi |\Psi \right \rangle -1\right) \right]dt}.
\ee
In Eq.~(\ref{9}) $\hat{H}$ is the Hamiltonian operator of the gravitating system, and $\alpha $ is a Lagrange multiplier. The variation of Eq.~(\ref{quantac}) with respect to the wave function leads to the normalization condition for the quantum wave function $\left<\Psi|\Psi\right>=1$, to the Sch\"odinger equation for $\Psi $,
\be\label{Sch}
i\left|\dot{\Psi}(t)\right>=\hat{H}(t)\left|\Psi (t)\right>-\alpha (t)\left|\Psi (t)\right>,
\ee
and to the semiclassical Einstein equations ~(\ref{gr1}), respectively. Hence, in this simple phenomenological approach to semiclassical gravity, the Bianchi identities still require the conservation of the effective energy-momentum tensor, $\nabla _{\mu}\left<\Psi \right|\hat{T}^{\mu \nu}\left|\Psi \right>=0$.
A more general set of semiclassical Einstein equations, having many similarities with the modified gravity models, can be derived by introducing a new coupling between the quantum fields and the classical curvature scalar of the space-time. One such model was proposed in \cite{Kibble}, where the total action containing the geometry-quantum matter coupling term was assumed to be of the form
\be
\int{RF\left(\left<f(\phi)\right>\right)_{\Psi}\sqrt{-g}d^4x},
\ee
where $F$ and $f$ are arbitrary functions, and $\left(\left<f(\phi)\right>\right)_{\Psi}=\left<\Psi (t)\right|f[\phi (x)]\left|\Psi (t)\right>$. Such a geometry-matter coupling term modifies the Hamiltonian $H(t)$ in the Schr\"odinger equation ~(\ref{Sch}) into \cite{Kibble}
\be
\hat{H}(t)\rightarrow \hat{H}_{\Psi}=\hat{H}(t)-\int{N F'\left(\left<f(\phi)\right>\right)_{\Psi}f(\phi)\sqrt{\varsigma}d^3\xi},
\ee
where $N$ is the lapse function, while $\xi ^i$ are intrinsic space-time coordinates, chosen in such a way that the normal vector to a space-like surface is time-like on the entire space-time manifold. The scalar function $\varsigma $ can be obtained as $\varsigma = {\rm det} \;\varsigma _{rs}$, where $\varsigma _{rs}$ is the metric induced on a space-like surface $\sigma( t )$. The surface $\sigma (t)$ globally slices the space-time manifold into
space-like surfaces. Then, by taking into account the effect of the geometry-quantum matter coupling, the effective semiclassical Einstein equations become \cite{Kibble}
\bea\label{123}
R_{\mu \nu}-\frac{1}{2}Rg_{\mu \nu}&=&16\pi G\Big[\left< \hat{T}_{\mu \nu}\right> _{\Psi}+G_{\mu \nu}F-\nonumber\\
&& \nabla _{\mu}\nabla _{\nu} F+
g_{\mu \nu}\Box F\Big].
\eea
In the semi-classical gravitational model introduced through Eq.~(\ref{123}), the matter energy-momentum tensor is not conserved anymore, since $\nabla _{\mu}\left< \hat{T}^{\mu \nu}\right> _{\Psi}\neq 0$. Thus, describes an effective particle generation process, in which there is a quantum-mechanically induced energy transfer from space-time to matter. Eq.~(\ref{123}). Eq.~(\ref{123}) also gives an effective semiclassical description of the quantum processes in a gravitational field, which are intrinsically related to the matter and energy non-conservation.
It is interesting to compare Eqs.~(\ref{genfe}) and (\ref{123}), both based on some assumptions on the quantum nature of gravity. While Eq.~(\ref{genfe}) is derived through a fist order approximation to quantum gravity, Eq.~(\ref{123}) postulates the existence of a quantum coupling between geometry and matter. While the coupling in Eq.~(\ref{123}) is introduced via a scalar function, the quantum effects are introduced in Eq.~(\ref{genfe}) through the fluctuations of the metric, having a tensor algebraic structure. However, in both models, the matter energy-momentum tensor is generally not conserved, indicating the possibility of the energy transfer between geometry and matter. Therefore, the physical origin of the modified gravity theories with geometry-matter coupling, which "automatically" leads to matter creation processes, may be traced back to the semiclassical approximation of the quantum field theory in a Riemannian curved space-time geometry.
\subsection{The cosmological model}
For cosmological applications we adopt the Friedmann-Robertson-Walker metric,
\be
ds^2=c^2dt^2-a^2(t)\left(\frac{1}{1-Kr^2}dr^2+r^2d\theta^2+r^2sin^2\theta d\phi^2\right),
\ee
where $a(t)$ is the scale factor. The components of the Ricci tensor for this metric are
\begin{eqnarray}
R_{00}=-\frac{3\ddot{a}}{a},
R_{ij}=-\frac{2K+2\dot{a}^2+a\ddot{a}}{a^2}g_{ij}, i,j=1,2,3.
\end{eqnarray}
We define the Hubble function as $H=\dot{a}/a$. As an indicator of the possible accelerated expansion we consider the deceleration
parameter $q$, defined as
\begin{equation}
q=\frac{d}{dt}\frac{1}{H}-1.
\label{deccparam}
\end{equation}
Negative values of $q$ indicate accelerating evolution, while positive ones correspond to decelerating expansion. In
order to perform the comparison between the observational and
theoretical results, instead of the time variable $t$ we introduce the redshift $z$,
defined as
\begin{equation}
1+z=\frac{1}{a},
\label{redshoftfedin}
\end{equation
where we have adopted for the scale factor $a\left( z\right) $ the normalization
$a(t_0)=1 $, where $t_0$ is the present age of the Universe. Hence as a function of the redshift the time derivative operator can be expressed as
\begin{equation}
\frac{d}{dt}= -H(z)(1+z)\frac{d}{dz}.
\label{timeredshiftrel}
\end{equation}
\section{Modified gravity from quantum perturbation of the metric proportional to the classical metric, \(K^{\mu\nu}=\alpha(x)g^{\mu\nu}\)}\label{sect3}
As a first example of cosmological evolution in modified gravity models induced by the quantum fluctuations of the metric we consider the simple case in which the expectation value of the quantum fluctuation tensor is proportional to the classical metric \cite{re11,re8},
\be\label{K1}
K^{\mu\nu}=\alpha(x)g^{\mu\nu},
\ee
where \(\alpha (x)\) is an arbitrary function of the space-time coordinates $x=x^{\mu}=\left(x^0,x^1,x^2,x^3\right)$. The case $\alpha ={\rm constant}$ was investigated in \cite{re11} and \cite{re8}, respectively. This condition suggest that we have an additional part of the metric, due to the quantum perturbation effects, which is proportional to the classical one.
In the following we will consider two distinct cases, by assuming first than $\alpha (x)$ is a {\it scalar field}, with a specific self-interaction potential. As a second model we will assume that $\alpha (x)$ is a simple scalar function.
\subsection{Scalar field-metric coupling}
If \(\alpha (x)\) is a scalar field, we add an additional Lagrangian
\bea
L_{\alpha}=\sqrt{-g}\left[\frac{1}{2}\nabla _{\mu }\alpha \nabla ^{\mu}\alpha -V(\alpha)\right],
\eea
as the matter source into the general quantum perturbed Lagrangian (\ref{2}). Hence we obtain the Lagrangian of our model as
\bea\label{Lalpha}
L_{total}=&&-\frac{1}{2\kappa^2}\sqrt{-g}\left[(1-\alpha)R\right]+\sqrt{-g}\left(\mathcal{L}_m+\frac{1}{2}\alpha T\right)+\nonumber\\
&&\sqrt{-g}\left[\frac{1}{2}\nabla _{\mu }\alpha \nabla ^{\mu}\alpha -V(\alpha)\right].
\eea
By taking the variation of the Lagrangian with respect to the field $\alpha $, the Euler-Lagrange equation gives the generalized Klein-Gordon equation for the scalar field, which has the form
\bea\label{KGa}
\Box\alpha-\frac{1}{2\kappa^2}R-\frac{1}{2}T+\frac{\partial V}{\partial \alpha}=0.
\eea
By taking the variation of Eq.~(\ref{Lalpha}) with respect to the metric tensor we obtain the Einstein gravitational field equations as
\bea\label{fealpha}
&&R_{\mu\nu}-\frac{1}{2}R g_{\mu\nu}=\frac{2\kappa ^2}{1-\alpha}\Bigg\{\frac{1+\alpha}{2}T_{\mu\nu}
-\frac{1}{4}\alpha T g_{\mu\nu}+\nonumber\\
&&
\frac{1}{2}\alpha \theta_{\mu\nu}
-\frac{1}{4}g_{\mu\nu}(\nabla _{\beta }\alpha \nabla ^{\beta}\alpha -2V)+
\frac{1}{2}\nabla_\mu \alpha \nabla_\nu \alpha\Bigg\},\nonumber\\
\eea
where
\bea
\theta_{\mu\nu}= g^{\alpha\beta}\frac{\delta T_{\alpha\beta}}{\delta g^{\mu\nu}}
= -g_{\mu\nu}\mathcal{L}_m-2T_{\mu\nu}.
\eea
After contraction of the Einstein field equations we obtain
\begin{eqnarray}\label{Ralpha}
R=-\kappa ^2\left[T+\frac{\alpha\theta}{1-\alpha}-\frac{1}{1-\alpha}\left(\nabla _{\mu }\alpha \nabla ^{\mu}\alpha -4V\right)\right],
\end{eqnarray}
and thus we can reformulate the gravitational field equations as
\bea
R_{\mu\nu}=&&\frac{\kappa^2}{2(1-\alpha)}\Bigg\{2(1+\alpha)T_{\mu\nu}-Tg_{\mu\nu}+2\alpha\theta_{\mu\nu}\nonumber\\
&&-\alpha\theta g_{\mu\nu}-2Vg_{\mu\nu}+2\nabla_\mu\alpha\nabla_\nu\alpha\Bigg\}.
\eea
The divergence of the matter energy-momentum tensor can be obtained from Eq.~(\ref{fealpha}), and is given by
\bea\label{ex1}
\nabla^\nu T_{\mu\nu}&=&-\frac{1}{2(1+\alpha)} \Bigg\{\alpha\Bigg[2\nabla^\nu\theta_{\mu\nu}-g_{\mu\nu}\nabla^\nu T\Bigg]+ \nonumber\\
&&\frac{\nabla^\nu \alpha}{1-\alpha}\Bigg[4T_{\mu\nu}-Tg_{\mu\nu}+2\theta_{\mu\nu}+\nonumber\\
&&g_{\mu\nu}\left(2V+2(1-\alpha)\Box\alpha-\nabla_\beta \alpha \nabla^\beta \alpha \right)+\nonumber\\
&&2\nabla_\mu \alpha \nabla_\nu \alpha\Bigg]+2\nabla^\nu Vg_{\mu\nu}\Bigg\}.
\eea
In the following we will adopt for the matter energy-momentum tensor the perfect fluid form
\begin{eqnarray}\label{28}
T_{\mu\nu}=\left(\rho +p\right)u_\mu u_\nu -g_{\mu\nu}p,
\end{eqnarray}
where $\rho$ is the matter energy density, $p$ is the thermodynamic pressure, and $u^{\mu}$ is the matter four velocity, satisfying the normalization condition $u_{\mu}u^{\mu}=1$. For this choice of the energy-momentum tensor we have
\bea
\theta_{\mu\nu}&=& g^{\alpha\beta}\frac{\delta T_{\alpha\beta}}{\delta g^{\mu\nu}}
= -g_{\mu\nu}\mathcal{L}_m-2T_{\mu\nu}=\nonumber\\
&&g_{\mu\nu}p-2\left(\rho +p\right)u_\mu u_\nu .
\eea
To obtain the above equation we have adopted for the matter Lagrangian the representation \(\mathcal{L}_m=p\). For the scalar $\theta $ we obtain
\be
\theta=2\left(p-\rho \right).
\ee
With the use of Eq.~(\ref{Ralpha}) the Klein-Gordon equation now reads
\bea
\Box\alpha+\frac{1}{2(1-\alpha)}\left(\alpha\theta-\nabla _{\mu }\alpha \nabla ^{\mu}\alpha +4V\right)+\frac{\partial V}{\partial \alpha}=0.
\eea
\subsubsection{Cosmological applications}
In the following we will restrict our analysis to the case of the flat Friedmann-Robertson-Walker metrics, and hence we set \(K=0\) in the gravitational field equations. The Friedmann and the Klein-Gordon equations describing our generalized gravity model obtained from a fluctuating metric take the form
\be\label{f1alpha}
3H^2=\frac{\kappa ^2}{1-\alpha}\Bigg\{\rho-\frac{1}{2}\alpha(3\rho-p)+\frac{1}{2}\dot{\alpha}^2+V\Bigg\},
\ee
\be\label{f2alpha}
2\dot{H}+3H^2=\frac{-\kappa ^2}{1-\alpha}\Bigg\{p+\frac{\alpha}{2}(\rho -3p)+\frac{1}{2}\dot{\alpha}^2-V\Bigg\},
\ee
\bea\label{f3alpha}
\ddot{\alpha}+3H\dot{\alpha} +\frac{1}{2(1-\alpha)}\left[2\alpha \left(p-\rho \right)-\dot{\alpha}^2+4V\right]\frac{\partial V}{\partial \alpha}=0,\nonumber\\
\eea
where
\bea\label{35}
\hspace{-0.05cm}\Box\alpha=\nabla_\nu\nabla^\nu\alpha&=&\frac{1}{\sqrt{-g}}\frac{\partial(\sqrt{-g}\nabla^\nu\alpha)}{\partial x^\nu}
=\Bigg(\ddot{\alpha}+\dot{\alpha}\frac{3\dot{a}}{a}\Bigg).\nonumber\\
\eea
\paragraph{The energy conservation equation}
By multiplying Eq.~(\ref{f1alpha}) with \(a^3\), and taking the time derivative of its both sides, we obtain
\bea
3H^2+6\frac{\ddot{a}}{a}=\frac{\kappa^2}{1-\alpha}\Bigg\{3\Bigg[\rho-\frac{\alpha}{2}(3\rho-p)+\frac{\dot{\alpha}^2}{2}+V\Bigg]\nonumber\\
+\frac{a}{\dot{a}}\Bigg[\dot{\rho}-\frac{\dot{\alpha}}{2}(3\rho -p)-\frac{\alpha}{2}(3\dot{\rho}-\dot{p})
+\dot{\alpha}\ddot{\alpha}+\frac{\partial V}{\partial \alpha}\dot{\alpha}\Bigg]\nonumber\\
+\frac{a}{\dot{a}}\frac{\dot{\alpha}}{1-\alpha}\Bigg[\rho -\frac{1}{2}\alpha(3\rho -p)+\frac{1}{2}\dot{\alpha}^2+V\Bigg]\Bigg\}.\nonumber\\
\eea
With the use of Eq.~(\ref{f2alpha}) we obtain
\bea\label{encons1}
\Bigg[(1-\alpha)(\rho+p)+\dot{\alpha}^2\Bigg]\frac{d}{dt}a^3+\left(1-\frac{3}{2}\alpha\right)a^3\frac{d}{dt}\rho\nonumber\\
=-\frac{a^3}{2}\Bigg\{\alpha\dot{p}-\dot{\alpha}(3\rho-p)+2\dot{\alpha}\ddot{\alpha}+2\frac{\partial V}{\partial \alpha}\dot{\alpha}\nonumber\\
+\frac{2\dot{\alpha}}{1-\alpha}\Bigg[\rho-\frac{1}{2}\alpha(3\rho-p)+\frac{1}{2}\dot{\alpha}^2+V\Bigg]\Bigg\}.\nonumber\\
\eea
Eq.~(\ref{encons1}) can be rewritten in the equivalent form
\bea\label{constan1}
\frac{d}{dt}[(1-\alpha)\rho a^3]+(1-\alpha)p\frac{d}{dt}a^3=\frac{\alpha}{2}a^3(\dot{\rho}-\dot{p})\nonumber\\
-\frac{a^3}{2}\frac{\dot{\alpha}}{1-\alpha}\Bigg[p-\rho+2V+\dot{\alpha}^2\Bigg]-a^3\frac{\partial V}{\partial \alpha}\dot{\alpha}-\nonumber\\
a^3(\dot{\alpha}\ddot{\alpha}+\dot{\alpha}^2\frac{3\dot{a}}{a})-a^3\rho\dot{\alpha}.
\eea
The same conservation equation can be obtained directly from the field equations (\ref{ex1}). By taking into account the explicit form of the matter energy-momentum tensor as given by Eq.~(\ref{28}), and that
\bea
\theta_{\mu\nu}=g_{\mu\nu}p-2(\rho+p)u_\mu u_\nu,
\eea
we have
\bea
\nabla^\nu T_{\mu\nu}=\dot{\rho}+3(\rho+p)\frac{\dot{a}}{ca},
\eea
and
\bea
\nabla^\nu \theta_{\mu\nu}=-2\dot{\rho}-\dot{p}-6(\rho+p)\frac{\dot{a}}{ca},
\eea
respectively. Hence Eq.~(\ref{28}) takes the form
\bea
-2(1-\alpha)\left(\rho+p\right)\frac{3\dot{a}}{a}-2\left(1-\frac{3}{2}\alpha\right)\dot{\rho}-\alpha\dot{p}=2\frac{\partial V}{\partial \alpha}\dot{\alpha}\nonumber\\
+\frac{\dot{\alpha}}{1-\alpha}\Bigg[p-\rho+2V+2(1-\alpha)\Box\alpha+\dot{\alpha}^2\Bigg].\nonumber\\
\eea
By taking into account Eq.~(\ref{35}), after eliminating $\Box\alpha$ from the above equation, we reobtain again the conservation equation Eq.~(\ref{constan1}).
\paragraph{The dimensionless form of the cosmological evolution equation}
In order to simplify the mathematical formulation of the cosmological model we rescale first the field $\alpha $ and its potential $V$ as $\alpha \rightarrow \kappa \alpha$, and $V\rightarrow \kappa ^2V$, respectively. Next, we introduce a set of dimensionless variables $\left(\tau, h,r,P,v\right)$, defined as
\be
\tau =H_0t,H=H_0h,\rho =\frac{3H_0^2}{8\pi G}, p=\frac{3H_0^2}{8\pi G}P,v=\frac{1}{H_0^2}V,
\ee
where $H_0$ is the present day value of the Hubble function. Then the cosmological evolution equation take the dimensionless form
\be\label{d1}
3h^2=\frac{1}{1-\alpha}\left[r-\frac{1}{2}\alpha (3r-P)+\frac{1}{2}\left(\frac{d\alpha }{d\tau}\right)^2+v\right],
\ee
\bea\label{d2}
2\frac{dh}{d\tau}+3h^2&=&\frac{1}{1-\alpha }\Bigg\{-\left[P+\frac{\alpha}{2}\left(r-3P\right)\right]-\nonumber\\
&&\left[\frac{1}{2}\left(\frac{d\alpha }{d\tau}\right)^2-v\right]\Bigg\},
\eea
\bea\label{d3}
&&\frac{d^2\alpha }{d\tau ^2}+3h\frac{d\alpha }{d\tau}+\frac{1}{2(1-\alpha)}\Bigg[2\alpha (P-r)-\left(\frac{d\alpha }{d\tau}\right)^2+\nonumber\\
&&4V\Bigg]+
\frac{\partial v}{\partial \alpha}=0.
\eea
In order to close the system of Eqs.~(\ref{d1})-{\ref{d3}) we must specify the equation of state of the matter $P=P(r)$, and the functional form of the self-interaction potential of the scalar field $v$. In the following we will restrict our analysis to the case of the dust, with $P=0$. Then, by denoting $u=d\alpha /d\tau$, from Eq.~(\ref{d1}) we obtain for the dimensionless energy density $r$ the expression
\be
r=\frac{1}{1-3\alpha /2}\left[3(1-\alpha )h^2-\left(\frac{1}{2}u^2+v\right)\right].
\ee
Then, by introducing the redshift $z$ as an independent variable, it follows that the cosmological evolution is described by the following system of equations,
\be\label{daz1}
\frac{d\alpha }{dz}=-\frac{1}{1+z}\frac{u}{h},
\ee
\be\label{dhz1}
\frac{dh}{dz}=\frac{1}{2(1+z)h}\left[\frac{u^2}{1-\alpha }-\frac{6 (\alpha -1) h^2+u^2+2 v}{2\left(1-3 \alpha /2\right)}\right],
\ee
\bea\label{duz1}
\frac{du}{dz}&=&\frac{3}{1+z}u-\frac{4 (\alpha -1) \left(2 v-3 \alpha h^2\right)+(2-5 \alpha ) u^2}{4(1+z)(1-\alpha)(1-3 \alpha/2)h}+\nonumber\\
&&\frac{1}{(1+z)h}\frac{\partial v}{\partial \alpha}.
\eea
For the deceleration parameter we obtain the expression
\bea
q&=&(1+z)\frac{1}{h}\frac{dh}{dz}-1=\nonumber\\
&&\frac{1}{h^2}\left[\frac{u^2}{1-\alpha }-\frac{6 (\alpha -1) h^2+u^2+2 v}{2\left(1-3 \alpha /2\right)}\right]-1.
\eea
For the scalar field self-interaction potential we adopt a Higgs type form, so that
\be\label{potH1}
v(\alpha)=\frac{\mu ^2}{2}\alpha ^2-\frac{\lambda }{4}\alpha ^4,
\ee
where $\mu ^2>0$ and $\lambda >0$ are constants.
In the following we will consider two cases. In the first case we assume that the Universe dominated by the quantum fluctuations of the metric evolves in the minimum of the Higgs potential, so that $\partial v/\partial \alpha =0$, implying $\alpha =\pm \sqrt{\mu ^2/\lambda}$, and $v(\alpha)=\mu ^4/4\lambda ={\rm constant}$. Secondly, we will investigate the evolution of the Universe in the presence of a time varying "full" Higgs type scalar field self-interaction potential (\ref{potH1}). Once the form of the potential is fixed, the system of differential equations Eqs.~(\ref{daz1})-(\ref{duz1}) must be integrated with the initial conditions $\alpha (0)=\alpha _0$, $h(0)=1$, and $u(0)=u_0$, respectively.
\paragraph{Cosmological evolution of the Universe in the minimum of the Higgs potential}
In Figs.~\ref{fig1}-\ref{fig4} we present the results of the numerical integration of the system of cosmological evolution equations Eqs.~(\ref{daz1})-(\ref{duz1}), for different values of the constant self-interaction potential $v=v_0={\rm constant}$. The initial values of $\alpha $ and $u$ used to integrate the system are $\alpha (0)=0.01$ and $u(0)=0.1$, respectively.
\begin{figure}[tbp]
\includegraphics[width=7.5cm, angle=0]{fig1_Liu.eps}
\caption{Variation with respect to the redshift of the dimensionless Hubble function for the Universe in the modified gravity model induced by the coupling between the metric and a Higgs type scalar field, in the presence of a constant self-interaction potential $v=v_0$, for different values of $v_0$: $v_0=2.1$ (solid curve), $v_0=2.3$ (dotted curve), $v_0=2.5$ (short dashed curve), $v_0=2.7$ (dashed curve), and $v_0=2.9$ (long dashed curve), respectively.}
\label{fig1}
\end{figure}
\begin{figure}[tbp]
\includegraphics[width=7.5cm, angle=0]{fig2_Liu.eps}
\caption{Variation with respect to the redshift of the dimensionless matter energy density for the Universe in the modified gravity model induced by the coupling between the metric and a Higgs type scalar field, in the presence of a constant self-interaction potential $v=v_0$, for different values of $v_0$: $v_0=2.1$ (solid curve), $v_0=2.3$ (dotted curve), $v_0=2.5$ (short dashed curve), $v_0=2.7$ (dashed curve), and $v_0=2.9$ (long dashed curve), respectively. }
\label{fig2}
\end{figure}
\begin{figure}[tbp]
\includegraphics[width=7.5cm, angle=0]{fig3_Liu.eps}
\caption{Variation with respect to the redshift of the scalar field $\alpha$ for the Universe in the modified gravity model induced by the coupling between the metric and a Higgs type scalar field, in the presence of a constant self-interaction potential $v=v_0$, for different values of $v_0$: $v_0=2.1$ (solid curve), $v_0=2.3$ (dotted curve), $v_0=2.5$ (short dashed curve), $v_0=2.7$ (dashed curve), and $v_0=2.9$ (long dashed curve), respectively.}
\label{fig3}
\end{figure}
\begin{figure}[tbp]
\includegraphics[width=7.5cm, angle=0]{fig4_Liu.eps}
\caption{Variation with respect to the redshift of the deceleration parameter $q$ for the Universe in the modified gravity model induced by the coupling between the metric and a Higgs type scalar field, in the presence of a constant self-interaction potential $v=v_0$, for different values of $v_0$: $v_0=2.1$ (solid curve), $v_0=2.3$ (dotted curve), $v_0=2.5$ (short dashed curve), $v_0=2.7$ (dashed curve), and $v_0=2.9$ (long dashed curve), respectively. }
\label{fig4}
\end{figure}
The Hubble function, represented in Fig.~\ref{fig1}, is a monotonically increasing function of the redshift (a monotonically decreasing function of the cosmological time), indicating an expansionary evolution of the Universe. Its variation is basically independent on the adopted numerical values of $v_0$, and, at $z\in [0,0.10]$, $h$ becomes approximately constant, indiating that the Universe has entered in a de Sitter type phase. The matter energy density, depicted in Fig.~\ref{fig2}, is a monotonically increasing function of the redshift, and its variation show a strong dependence on the numerical value of $v_0$. In the considered range of values of $v_0$ at the present time the matter density can either reach values of the order of the critical density, with $r(0)\approx 1$, or become negligibly small. The function $\alpha $, shown in Fig.~\ref{fig3}, monotonically decreases with the redshift, and has negative numerical values. For large redshifts, the variation of $\alpha $ has a strong dependence on the numerical values of $v_0$, However, for $z\leq 0.2$, the changes in $\alpha $ due to the variation of $v_0$ become negligible, and for $z\in [0,0.05]$ $\alpha $ becomes a constant. The deceleration parameter $q$, plotted in Fig.~\ref{fig4}, shows a complex behavior. The Universe starts its evolution at a redshift $z=1$ in a decelerating phase, with $q\approx 1$. The Universe begins to accelerates, and it enters in a marginally accelerating phase, with $q=0$, at a redshift of around $z\approx 0.3$. The variation of the deceleration parameter strongly depends on the numerical values of $v_0$, and, depending on this numerical value, the evolution of the Universe at the present time can either be de Sitter, with $q=-1$, or with higher values of $q$, of the order of $q\approx -0.5$.
\paragraph{Cosmological evolution in the presence of the Higgs potential}
The evolution of the cosmological and physical parameters of a Universe filled with a scalar field with Higgs potential (\ref{potH1}), coupled to the fluctuating quantum metric, are presented in Figs.~\ref{fig5}-\ref{fig8}. To numerically integrate the gravitational field equations (\ref{daz1})-(\ref{duz1}) in the redshift range $z\in [0,2]$ we have adopted the initial conditions $\alpha (0)=0.1$ and $u(0)=0.1$, respectively. We have fixed the value of the coefficient $\lambda $ in the Higgs potential as $\lambda =150$, and we have varied the mass $\mu ^2>0$ of the Higgs field.
\begin{figure}[tbp]
\includegraphics[width=7.5cm, angle=0]{fig5_Liu.eps}
\caption{Variation with respect to the redshift of the dimensionless Hubble function for the Universe filled with a Higgs type scalar field coupled to the fluctuating quantum metric for $\lambda =150$, and for different values of $\mu ^2$: $\mu ^2=250$ (solid curve), $\mu ^2=300$ (dotted curve), $\mu ^2=350$ (short dashed curve), $\mu ^2=400$ (dashed curve), and $\mu ^2=450$ (long dashed curve), respectively.}
\label{fig5}
\end{figure}
\begin{figure}[tbp]
\includegraphics[width=7.5cm, angle=0]{fig6_Liu.eps}
\caption{Variation with respect to the redshift of the potential $v$ in the modified gravity model induced by the coupling between the metric and a Higgs type scalar field for $\lambda =150$, and for different values of $\mu ^2$: $\mu ^2=250$ (solid curve), $\mu ^2=300$ (dotted curve), $\mu ^2=350$ (short dashed curve), $\mu ^2=400$ (dashed curve), and $\mu ^2=450$ (long dashed curve), respectively. }
\label{fig6}
\end{figure}
\begin{figure}[tbp]
\includegraphics[width=7.5cm, angle=0]{fig7_Liu.eps}
\caption{Variation with respect to the redshift of the scalar field $\alpha$ in the modified gravity model induced by the coupling between the metric and a Higgs type scalar field, for $\lambda =150$, and for different values of $\mu ^2$: $\mu ^2=250$ (solid curve), $\mu ^2=300$ (dotted curve), $\mu ^2=350$ (short dashed curve), $\mu ^2=400$ (dashed curve), and $\mu ^2=450$ (long dashed curve), respectively.}
\label{fig7}
\end{figure}
\begin{figure}[tbp]
\includegraphics[width=7.5cm, angle=0]{fig8_Liu.eps}
\caption{Variation with respect to the redshift of the deceleration parameter $q$ in the modified gravity model induced by the coupling between the metric and a Higgs type scalar field, for a $\lambda =150$, and for different values of $\mu ^2$: $\mu ^2=250$ (solid curve), $\mu ^2=300$ (dotted curve), $\mu ^2=350$ (short dashed curve), $\mu ^2=400$ (dashed curve), and $\mu ^2=450$ (long dashed curve), respectively.}
\label{fig8}
\end{figure}
The Hubble function, plotted in Fig.~\ref{fig5}, is an increasing function of the redshift, indicating an expansionary evolution. However, $h$ has a complex behavior, which is strongly dependent on the numerical value of $\mu ^2$. The variation with $z$ of the Higgs potential is represented in Fig.~\ref{fig6}. The potential has a damped harmonic oscillator type behavior, with the amplitude of the oscillations decreasing while we are approaching the present day moment, with $z=0$. The same damped oscillator type pattern can be observed in the redshift evolution of the scalar field $\alpha $, depicted in Fig.~\ref{fig7}. An oscillatory behavior is also characteristic for the variation with respect to $z$ of the deceleration parameter $q$, presented in Fig.~\ref{fig8}. The behavior is strongly dependent on the numerical values of $\mu ^2$, so that at $z=2$ the Universe can be in either a decelerating ($q.0$), or in an accelerating phase, with $q<0$. There is an alternation of the accelerating and decelerating epochs, but, (almost) independently of the values of $\mu ^2$, the Universe enters a very rapid accelerating phase at $z\approx 0.10$, which brings down in a very short cosmological time interval the deceleration parameter from $q\approx 1.5$ to $q\approx -1$. Hence the de Sitter solution is also an attractor of this model.
\subsection{Scalar function coupling to the metric}
As a second example of modified gravity model induced by a quantum perturbation tensor of the form (\ref{K1}) we assume that $alpha $ is a "pure" function of the coordinates, and there is no specific physical field associated to it. Then,
by using the least action principle, from Eq.~(\ref{2}) we obtain the gravitational field equations as
\bea\label{4}
R_{\mu\nu}-\frac{1}{2}R g_{\mu\nu}&=&\frac{2\kappa ^2}{1-\alpha(x)}\Bigg\{\frac{1}{2}\left[1+\alpha(x)\right]T_{\mu\nu}-\nonumber\\
&&\frac{1}{4}\alpha (x)T g_{\mu\nu}+\frac{1}{2}\alpha (x) \theta_{\mu\nu}\Bigg\},
\eea
where we have denoted $\theta_{\mu\nu}= g^{\alpha\beta}\left(\delta T_{\alpha\beta}/\delta g^{\mu\nu}\right)$, and
\be
\frac{\delta T}{\delta g^{\mu\nu}}=T_{\mu\nu}+\theta_{\mu\nu},
\ee
respectively. As usual, \(T=T_{\mu\nu}g^{\mu\nu}\) is the trace of the energy-momentum tensor, and we have denoted \(\theta =\theta_{\mu\nu}g^{\mu\nu}\). After contraction Eq~(\ref{4}) reads
\begin{eqnarray}\label{7}
R=-\kappa ^2\left[T+\frac{\alpha(x)}{1-\alpha (x)}\theta\right].
\end{eqnarray}
Combining Eqs.~(\ref{4}) and (\ref{7}) we obtain the field equations in the form
\begin{eqnarray}
R_{\mu\nu}&=&\frac{2\kappa ^2}{1-\alpha (x)}\Bigg[\frac{1}{2}(1+\alpha (x))T_{\mu\nu}-\frac{1}{4} T g_{\mu\nu}+ \nonumber\\
&&\frac{1}{2}\alpha (x) \theta_{\mu\nu}-\frac{1}{4}\alpha (x)\theta g_{\mu\nu}\Bigg].
\end{eqnarray}
Assuming that the Lagrangian density of the matter \(\mathcal{L}_m\) depends only on the metric tensor but not on its derivatives, we obtain
\begin{eqnarray}
T_{\mu\nu}=2\frac{\partial \mathcal{L}_m}{\partial g^{\mu\nu}}-g_{\mu\nu}\mathcal{L}_m,
\end{eqnarray}
and
\begin{eqnarray}
\theta_{\mu\nu}=2g^{\alpha\beta}\frac{\partial^2 \mathcal{L}_m}{\partial g^{\alpha\beta}\partial g^{\mu\nu}}-g_{\mu\nu}\mathcal{L}_m-2T_{\mu\nu},
\end{eqnarray}
respectively. Different kinds of matter Lagrangians may lead to different \(\theta_{\mu\nu}\), and therefore to different gravitational theories.
By taking into account the mathematical identity \(\nabla^\nu G_{\mu\nu}=0\), after taking the covariant divergence of Eq.~(\ref{4}) we obtain
\begin{eqnarray}
&&0=\frac{\kappa ^2\left(\nabla^\nu \alpha\right)}{2(1-\alpha)^2} \left[2(1+\alpha)T_{\mu\nu}-\alpha T g_{\mu\nu}+2\alpha \theta_{\mu\nu}\right]+ \nonumber\\
&&\frac{\kappa ^2}{2(1-\alpha)}\Bigg[ 2\left(\nabla^\nu \alpha\right) T_{\mu\nu}+2(1+\alpha)\nabla^\nu T_{\mu\nu}-\left(\nabla^\nu\alpha\right) \times \nonumber\\
&&T g_{\mu\nu}-\alpha (\nabla^\nu T) g_{\mu\nu}+2\left(\nabla^\nu\alpha\right) \theta_{\mu\nu}+2\alpha (\nabla^\nu\theta_{\mu\nu})\Bigg]=\nonumber\\
&&\left(\nabla^\nu \alpha\right)\Bigg[\frac{2(1+\alpha)T_{\mu\nu}-\alpha T g_{\mu\nu}+2\alpha \theta_{\mu\nu}}{1-\alpha}+2T_{\mu\nu}-\nonumber\\
&&Tg_{\mu\nu}+2\theta_{\mu\nu}\Bigg]+\alpha\Bigg[2\nabla^\nu T_{\mu\nu}-g_{\mu\nu}\nabla^\nu T+2\nabla^\nu\theta_{\mu\nu}\Bigg]+\nonumber\\
&&2\nabla^\nu T_{\mu\nu}.
\end{eqnarray}
Now it is easy to check that the divergence of the matter energy-momentum tensor is
\begin{eqnarray}\label{emtc}
\nabla^\nu T_{\mu\nu}&=&-\frac{1}{2(1+\alpha)} \Bigg\{\left(\nabla^\nu \alpha\right)\Bigg[\frac{4T_{\mu\nu}-T g_{\mu\nu}+2\theta_{\mu\nu}}{1-\alpha}\Bigg]+\nonumber\\
&&\alpha\Bigg[-g_{\mu\nu}\nabla^\nu T+2\nabla^\nu\theta_{\mu\nu}\Bigg]\Bigg\}.
\end{eqnarray}
In the following discussion, we consider a perfect fluid described by
$\rho $, the matter energy density, and by $p$, the thermodynamic pressure. In the comoving frame with $u^{\mu}=(1,0,0,0)$ the components of the energy-momentum tensor have the components $T^{\mu}_{\nu}={\rm diag}\left(\rho,-p,-p,-p\right)$.
Moreover, we will consider that matter obeys an equation of state of the form $p=\omega \rho$, where $\omega ={\rm constant}$, and $0\leq \omega \leq 1$. Then the generalized Friedmann equations that follow from the gravitational field equations Eqs.~(\ref{4}) are
\be\label{f1c}
H^2=\frac{\kappa ^2}{3}\left[\frac{2-(3-\omega)\alpha}{2\left(1-\alpha\right)}\right]\rho,
\ee
\be\label{f2c}
\frac{\ddot{a}}{a}=-\frac{\kappa ^2}{6}\left[\frac{1+(3-4\alpha)\omega}{1-\alpha}\right]\rho,
\ee
or, equivalently,
\be\label{dda}
\frac{\ddot{a}}{a}=-\frac{1+(3-4\alpha)\omega}{2-(3-\omega)\alpha}H^2.
\ee
Eq.~(\ref{f2c}) can be alternatively written as
\be\label{f3c}
2\dot{H}+3H^2=-\kappa ^2\left[\frac{2\omega+\alpha(1-3\omega)}{2\left(1-\alpha\right)}\right]\rho.
\ee
Hence
\be\label{f4c}
\dot{H}=-\frac{\kappa ^2}{2}\left(1+\omega \right)\rho =\frac{3 (\alpha -1) (\omega +1)}{\alpha (\omega -3)+2} H^2.
\ee
For the deceleration parameter we obtain
\be
q=\frac{-\alpha (7 \omega +3)+6 \omega +4}{\alpha (\omega -3)+2}.
\ee
The presence of an accelerated expansion requires $\ddot{a}>$, which imposes on the function $\alpha (t)$ the condition
\be
\frac{\alpha (t)}{1-\alpha (t)}>\frac{1+3\omega }{4\omega }\frac{1}{1-\alpha (t)}, \forall t\geq t_a.
\ee
\subsubsection{Conservative models-the \(\nabla^\nu T_{\mu\nu}=0\) case}
In standard general relativity theory, the energy-momentum tensor is conserved, (\(\nabla^\nu T_{\mu\nu}=0\)), while in modified theories of gravity the classical-defined energy-momentum tensor is not always conserved. However, in the modified gravity theory induced by the quantum fluctuations of the metric, with expectation value of the fluctuations proportional to the metric tensor, thanks to the addition of the function $\alpha $, we can maintain the conservation of the energy and momentum by constraining the new function. In the following we assume for the matter Lagrangian the form \(L_m=p\), and, since the second derivatives of \(L_m\) with respect to the metric tensor are zero, we obtain for the tensor
$\theta _{\mu \nu}$ the expression
\be
\theta_{\mu\nu}=-g_{\mu\nu}L_m-2T_{\mu\nu}.
\ee
Demanding that \(\nabla^\nu T_{\mu\nu}=0\) we obtain
\begin{eqnarray}\label{20}
&&0=\left(\nabla^\nu \alpha\right)\Bigg[\frac{2(1+\alpha)T_{\mu\nu}-\alpha T g_{\mu\nu}+2\alpha \theta_{\mu\nu}}{1-\alpha}+2T_{\mu\nu}-\nonumber\\
&&Tg_{\mu\nu}+2\theta_{\mu\nu}\Bigg]
+\alpha\Bigg[2\nabla^\nu\theta_{\mu\nu}-g_{\mu\nu}\nabla^\nu T\Bigg].
\end{eqnarray}
Multiplying by \(u^{\mu}\) both sides of Eq~(\ref{20}) we obtain
\begin{eqnarray}
\dot{\alpha}\left[\frac{-\rho +p}{1-\alpha}\right]+
\alpha\left[-5\dot{\rho}+\dot{p}\right]=0,
\end{eqnarray}
giving
\begin{eqnarray}
\frac{\dot{\alpha}}{\alpha(1-\alpha)}&=&\frac{-5\dot{\rho}+\dot{p}}{\rho-p}.
\end{eqnarray}
For the linear barotropic equation of state with $p=\omega \rho $, for $\omega \neq 1$ we obtain
\be
\frac{\dot{\alpha}}{\alpha(1-\alpha)}=-\left(\frac{5-\omega}{1-\omega}\right)\frac{\dot{\rho}}{\rho},
\ee
which gives the density as a function of $\alpha $ as
\be\label{24}
\rho =\rho_0 \left(\frac{\alpha}{1-\alpha}\right)^{-\frac{1-\omega}{5-\omega}},
\ee
where $\rho _0$ is an arbitrary constant of integration. On the other hand the conservation of the energy-momentum tensor gives the equation
\be
-3\frac{\dot{a}}{a}(1+\omega)=\frac{\dot{\rho}}{\rho},
\ee
which give for the matter density the standard relation
\be\label{26}
\rho=\rho_0^{\prime}a^{-3(1+\omega)},
\ee
where $\rho_0^{\prime}$ is an arbitrary constant of integration. From Eqs.~(\ref{24}) and (\ref{26}) we obtain the scale factor dependence of the function $\alpha $ as
\be
\frac{\alpha }{1-\alpha }=\alpha _0a^{3n},
\ee
where $n=(1+\omega)(5-\omega)/(1-\omega)$, and $\alpha _0=\left(\rho_0^{\prime}/\rho _0\right)^{-(5-\omega)/(1-\omega)}$.
The the first Friedmann equation Eq.~(\ref{f1c}) gives
\be
\frac{\dot{a}}{a}=\frac{\kappa}{\sqrt{6}} \sqrt{\rho _0 a^{-3 (\omega +1)} \left(\alpha _0 (\omega -1)
a^{3 n}+2\right)},
\ee
from which we obtain
\bea
\frac{\kappa }{\sqrt{6}}\left( t-t_{0}\right) &=&\frac{\sqrt{2\alpha
_{0}(\omega -1)a^{3n}+4}}{3(\omega +1)}\times \nonumber\\
&&\frac{\,_{2}F_{1}\left( \frac{1}{2}
\frac{\omega +1}{2n};\frac{\omega +1}{2n}+1;-\frac{1}{2}a^{3n}\alpha
_{0}(\omega -1)\right) }{\sqrt{\rho _{0}a^{-3(\omega +1)}\left( \alpha
_{0}(\omega -1)a^{3n}+2\right) }},\nonumber\\
\eea
where $\,_{2}F_{1}\left(a,b;c,z\right)$ is the hypergeometric function $\,_{2}F_{1}\left(a,b;c,z\right)=\sum _{k=0}^{\infty}{\left[(a)_k(b)_k/(c)_k\right]\left(z^k/k!\right)}$, and $t_0$ is an arbitrary constant of integration. For the deceleration parameter we obtain the expression
\be
q(a)=\frac{1}{2} \left\{n \left[\frac{6}{\alpha _0 (\omega -1) a^{3 n}+2}-3\right]+3 \omega +1\right\}.
\ee
In the limit of small values of the scale factor $\alpha _0 (\omega -1) a^{3 n}<<2$, we obtain $q\approx (1+3\omega)/2>0$, indicating a decelerating expansion during the early stages of the cosmological evolution. For $\alpha _0 (\omega -1) a^{3 n}>>2$, and for very large values of $a$, $q\approx \left[3(\omega -n)+1\right]/2=4(\omega +2)/(1-\omega )$. Since generally for any realistic cosmological matter equation of state $\omega <1$, it follows that in both small and large time limits the time evolution of the Universe is decelerating.
\subsubsection{Non-conservative cosmological models with \(\nabla^\nu T_{\mu\nu}\neq 0\)}
By taking into account that \(T=\rho-3p\), from Eq.~(\ref{emtc}) we immediately obtain
\bea
-2(1+\alpha)\nabla^\nu T_{\mu\nu}&=&\nabla^\nu\alpha\frac{4T_{\mu\nu}-Tg_{\mu\nu}+2\theta_{\mu\nu}}{1-\alpha}+\nonumber\\
&&\alpha\left(2\nabla^\nu \theta_{\mu\nu}-g_{\mu\nu}T\right),
\eea
or, equivalently,
\begin{eqnarray}\label{35}
-2(1-\alpha)\nabla^\nu T_{\mu\nu}&=&\nabla^\nu\alpha\frac{g_{\mu\nu}\left(2L_m-T\right)}{1-\alpha}+\nonumber\\
&&\alpha g_{\mu\nu}\nabla^\nu \left(2L_m-T\right).
\end{eqnarray}
After multiplying Eq.~(\ref{35}) by the four-velocity vector $u^{\mu}$, defined in the comoving reference frame, we obtain
\begin{eqnarray}
u^\mu\nabla^\nu T_{\mu\nu}=\frac{1}{2(1-\alpha)}\left[\dot{\alpha}\frac{\rho-p}{1-\alpha}+\alpha(\dot{\rho}-\dot{p})\right].
\end{eqnarray}
By taking into account the mathematical identity
\be
u^\mu\nabla ^\nu T_{\mu\nu}=\dot{\rho}+3H(\rho+p)
\ee
we immediately find
\begin{eqnarray}\label{38}
\hspace{-0.7cm}\dot{\rho}+3H(\rho+p)&=&\frac{1}{2(1-\alpha)}\times \nonumber\\
&&\left[\dot{\alpha}(\rho-p)\left(\frac{1}{1-\alpha}\right)+\alpha(\dot{\rho}-\dot{p})\right].
\eea
For a general linear barotropic equation of state of the for $p=\omega (t)\rho$, Eq.~(\ref{38}) becomes
\bea
\left(1-\frac{3}{2}\alpha+\frac{1}{2}\omega\alpha\right)\dot{\rho}+\frac{1}{2}\alpha\rho\dot{\omega}
&=&3(\alpha-1)(1+\omega)H\rho+\nonumber\\
&&\frac{\dot{\alpha}(\omega-1)}{2(\alpha-1)}\rho.
\end{eqnarray}
For $\omega ={\rm constant}$ we have
\begin{eqnarray}\label{43}
\left(1-\frac{3}{2}\alpha+\frac{1}{2}\omega\alpha\right)\dot{\rho}
&=&3(\alpha-1)(1+\omega)H\rho+\nonumber\\
&&\frac{\dot{\alpha}}{2(\alpha-1)}(\omega-1)\rho .
\eea
After taking the derivative of Eq.~(\ref{f1c}) with respect to the time, and after substituting $\dot{H}$ with the use of Eq.~(\ref{f4c}), we obtain for the time deriative of the density the equation
\be\label{44}
\dot{\rho}=-\frac{6 H^2 \left[6 (\omega +1)(\alpha -1)^2 H +(\omega -1) \dot{\alpha }
\right]}{\kappa ^2 \left[(\omega -3) \alpha +2\right]^2}.
\ee
Substituting Eq.~(\ref{44}) into Eq.~(\ref{43}) gives the equation
\bea\label{45}
&&\left[ 6(\omega +1)H(\alpha -1)^{2}+(\omega -1)\dot{\alpha}\right] \times \nonumber\\
&&\left\{
6H^{2}(\alpha -1)+\kappa ^{2}\rho \left[ (\omega -3)\alpha +2\right]
\right\} =0.
\eea
Due to the first generalized Friedmann equation (\ref{f1c}), Eq.~(\ref{45}) is identically satisfied during the cosmological evolution. However, a second solution of the field equation can be obtained by also imposing the condition that the first term in Eq.~(\ref{45}0 also vanishes identically.
\paragraph{The case $\alpha =1-\frac{1-\omega}{6(1+\omega)}\frac{1}{\ln\left( a/a_0\right)}$}
By requiring that the first term in the left-hand side of Eq.~(\ref{45}) also vanishes, we obtain for the scalar function $\alpha (t)$ the differential equation
\be
\dot{\alpha }= \frac{6 (1+\omega ) (1-\alpha )^2}{1-\omega}H,
\ee
from which we obtain
\be
\alpha (t)=1-\frac{1-\omega}{6(1+\omega)}\frac{1}{\ln \left[a(t)/a_0\right]},
\ee
where $a_0$ is an arbitrary constant of integration. By substituting this expression of $\alpha $ into Eq.~(\ref{dda}), we obtain the following second order differential equation describing the time evolution of the scale factor,
\be
\frac{\ddot{a}}{\dot{a}}=\left[1-\frac{3 (\omega +1)}{6 (\omega +1) \ln (Ca)+\omega -3}\right]\left(\frac{\dot{a}}{a}\right).
\ee
By integration we first obtain
\be
H=\frac{\dot{a}}{a}= \frac{\zeta}{\sqrt{6 (\omega +1) \ln \left(a /a_0 \right)+\omega -3}},
\ee
where $\zeta$ is an arbitrary integration constant. Hence for the time variation of the scale factor we obtain
\be
a(t)=a_0e^{-\frac{\omega -3}{6(1+\omega)}}e^{\frac{3^{1/3}\zeta ^{2/3}}{2(1+\omega)^{1/3}}\left(t-t_0\right)^{2/3}},
\ee
where $t_0$ is an arbitrary constant of integration. For the time variation of the Hubble function we obtain
\be
H(t)=\frac{\zeta ^{2/3}}{3^{2/3} \sqrt[3]{\omega +1} \sqrt[3]{t-t_0}},
\ee
while the deceleration parameter $q$ of this model is given by
\be
q(t)=\frac{(1+\omega)^{1/3}}{3^{1/3} \zeta ^{2/3}
\left(t-t_0\right)^{2/3}}-1.
\ee
In the limit of large times $t\rightarrow \infty $, we have $q\rightarrow -1$, and therefore the Universe ends in an (approximately de Sitter) accelerating phase. For the time variation of the function $\alpha $ we find
\be
\alpha (t)=\frac{1-\omega }{-3^{4/3} \zeta ^{2/3} (\omega +1)^{2/3}
\left(t-t_0\right)^{2/3}+\omega -3}+1.
\ee
In the limit of large times $\alpha (t)$ tends to a constant, $\lim_{t\rightarrow \infty}\alpha (t)=1$.
\section{Modified gravity from quantum fluctuations proportional to the matter energy-momentum tensor--\(K^{\mu\nu}=\alpha T^{\mu\nu}\)}\label{sect4}
Many extensions of standard general theory of relativity are based on the assumption that in certain physical situations space-time and matter may couple to each other \cite{fL1,fT1,Har4}. Hence it is natural to also consider the case in which the average of the quantum fluctuations of the metric is proportional to the matter energy-momentum tensor, \(K^{\mu\nu}=\alpha T^{\mu\nu}\), where \(\alpha\) is a constant. This approach suggests that the quantum perturbations of the space-time may also be strongly influenced by the presence of the classical matter.
\subsection{The gravitational field equations}
With the choice \(K^{\mu\nu}=\alpha T^{\mu\nu}\) of the classical form of the average of the quantum fluctuations of the metric tensor we obtain for the first order quantum corrected gravitational Lagrangian the expression
\begin{eqnarray}\label{fT1}
\mathcal{L}&=&-\frac{1}{2k^2}\sqrt{-g}\left[R+\alpha \left(R_{\mu\nu}-\frac{1}{2}Rg_{\mu\nu}\right)T^{\mu\nu}\right]+\nonumber\\
&&\sqrt{-g}\left[L_m+\frac{1}{2}\alpha T_{\mu\nu}T^{\mu\nu}\right]=\nonumber\\
&&-\frac{1}{2k^2}\sqrt{-g}\left[R\left(1-\frac{1}{2}\alpha T\right)+\alpha R_{\mu\nu}T^{\mu\nu}\right]\nonumber\\
&&+\sqrt{-g}\left[\mathcal{L}_m+\frac{1}{2}\alpha T_{\mu\nu}T^{\mu\nu}\right].
\end{eqnarray}
By varying the gravitational action given by Eq.~(\ref{fT1}) with respect to the metric tensor $g^{\mu \nu}$ it follows that the classical gravitational field equations corresponding to the gravitational action (\ref{fT1}) are given by (for the full details of the derivation see Appendix~\ref{App2})
\begin{eqnarray}\label{ff1}
&&G_{\mu\nu}\left(1-\frac{1}{2}\alpha T\right)=\Bigg[\frac{1}{2}\alpha R\left(T_{\mu\nu}+\theta_{\mu\nu}\right)+\nonumber\\
&&\frac{1}{2}\alpha g_{\mu\nu}R_{\alpha\beta}T^{\alpha\beta}-\left(g_{\mu\nu}\Box-\nabla_\mu\nabla_\nu\right)\left(1-\frac{1}{2}\alpha T\right)\Bigg]-\nonumber\\
&&\kappa ^2\left[\frac{1}{2}\alpha g_{\mu\nu} T_{\alpha\beta}T^{\alpha\beta}-T_{\mu\nu}\right]+\nonumber\\
&&\kappa ^2\alpha\Bigg[T_{\alpha(\mu}T^\alpha_{\nu)}-T(g_{\mu\nu}\mathcal{L}_m+T_{\mu\nu})+2\mathcal{L}_m T_{\mu\nu}\Bigg]-\nonumber\\
&&\alpha\left[R_{\alpha(\mu}T^\alpha_{\nu)}-\frac{1}{2}R\left(g_{\mu\nu}\mathcal{L}_m+T_{\mu\nu}\right)+\mathcal{L}_m R_{\mu\nu}\right.+\nonumber\\
&&\left.\frac{1}{2}\left(\Box T_{\mu\nu}+\nabla_{\alpha}\nabla_{\beta}T^{\alpha\beta}g_{\mu\nu}-
\nabla_{\alpha}\nabla_{(\mu}T_{\nu)}^\alpha\right)\right],
\end{eqnarray}
where we have canceled the terms containing the second derivatives of \(\mathcal{L}_m\) with respect to the metric tensor, since in most cases of physical interest they vanish. After contraction of Eq.~(\ref{ff1}) we find
\begin{eqnarray}
R\left(1-\frac{1}{2}\alpha T\right)-\alpha R\mathcal{L}_m+\alpha\left(\Box T-\nabla_\mu\nabla_\nu T^{\mu\nu}\right)\nonumber\\
+\kappa ^2(T-2\alpha T\mathcal{L}_m-\alpha T^2)=0.
\end{eqnarray}
Eq.~(\ref{ff1}) can then be rewritten as
\begin{eqnarray}
&&R_{\mu\nu}\left(1-\frac{1}{2}\alpha T+\alpha \mathcal{L}_m \right)+\alpha R_{\alpha(\mu}T_{\nu)}^\alpha- \frac{\alpha}{2} g_{\mu\nu}R_{\alpha\beta}T^{\alpha\beta}=\nonumber\\
&&-\frac{1}{2}g_{\mu\nu}\frac{1-\frac{1}{2}\alpha T}{1-\frac{1}{2}\alpha T-\alpha \mathcal{L}_m}\Bigg[\alpha\left(\Box T-\nabla_\alpha\nabla_\beta T^{\alpha\beta}\right)+\nonumber\\
&&2\kappa ^2\left(-\alpha TL_m-\frac{1}{2}\alpha T^2+\frac{1}{2}T\right)\Bigg]-\frac{1}{2}\alpha\times \nonumber\\
&&\Bigg(\Box T_{\mu\nu}+g_{\mu\nu}\nabla_\alpha\nabla_\beta T^{\alpha\beta}-\nabla_\alpha\nabla_{(\mu} T^{\alpha}_{\nu)}\Bigg)+\nonumber\\
&&\frac{1}{2}\alpha\left(g_{\mu\nu}\Box-\nabla_\mu\nabla\nu \right)T+2\kappa ^2 \times \nonumber\\
&&\left[\frac{1}{2}\alpha\left(T_{\alpha(\mu}T_{\nu)}^\alpha-g_{\mu\nu}T\mathcal{L}_m-T_{\mu\nu}T+2\mathcal{L}_mT_{\mu\nu}\right)\right.-\nonumber\\
&&\left.\frac{1}{4}\alpha g_{\mu\nu}T_{\alpha\beta}T^{\alpha\beta}+\frac{1}{2}T_{\mu\nu}\right].
\end{eqnarray}
\subsection{The divergence of the energy-momentum tensor}
By taking the divergence of the gravitational field equations (\ref{ff1}) we obtain first
\bea
\Big[\kappa^2\left(\alpha T-2\alpha \mathcal{L}_m-1\right)-\alpha R\Bigg]\nabla^\nu T_{\mu\nu}=\frac{1}{2}\alpha \times\nonumber\\
\Bigg[G_{\mu\nu} \nabla^\nu T+T_{\mu\nu}\nabla^\nu R+\nabla_\mu(R_{\alpha\beta}T^{\alpha\beta}-\kappa^2 T_{\alpha\beta}T^{\alpha\beta})\Bigg]\nonumber\\
+\kappa^2\alpha\Bigg[\nabla^\nu(T_{\alpha(\mu}T_{\nu)}^\alpha)-\nabla_\mu (T\mathcal{L}_m)-T_{\mu\nu}\times\nonumber\\
\nabla^\nu (T-2\mathcal{L}_m)\Bigg]-\alpha\Bigg[\nabla^\nu(R_{\alpha(\mu}T_{\nu)}^\alpha)-\frac{1}{2}\nabla_\mu(R\mathcal{L}_m)-\nonumber\\
\frac{1}{2}T_{\mu\nu}\nabla^\nu R+\nabla^\nu(R_{\mu\nu}\mathcal{L}_m)+\frac{1}{2}(\nabla^\nu\Box T_{\mu\nu}\nonumber\\
+\nabla_\mu\nabla_\alpha\nabla_\beta T^{\alpha\beta}-\nabla^\nu\nabla_\alpha\nabla_{(\mu}T_{\nu)}^\alpha)\Bigg].\nonumber\\
\eea
Hence for the divergence of the matter energy-momentum tensor in the modified gravity model induced by the quantum fluctuations of the metric proportional to the matter energy-momentum tensor we find
\bea
\nabla^\nu T_{\mu\nu}=\frac{1}{\kappa^2(\alpha T-2\alpha L_m-1)-\alpha R}\Bigg\{\frac{1}{2}\alpha \times\nonumber\\
\Bigg[T_{\mu\nu}\nabla^\nu R+\nabla_\mu(R_{\alpha\beta}T^{\alpha\beta}-\kappa^2 T_{\alpha\beta}T^{\alpha\beta})\Bigg]\nonumber\\
+\kappa^2\alpha\Bigg[\nabla^\nu(T_{\alpha(\mu}T_{\nu)}^\alpha)-\nabla_\mu (T\mathcal{L}_m)-T_{\mu\nu}\times\nonumber\\
\nabla^\nu (T-2\mathcal{L}_m)\Bigg]-\alpha\Bigg[\nabla^\nu(R_{\alpha(\mu}T_{\nu)}^\alpha)-\frac{1}{2}\nabla_\mu(R\mathcal{L}_m)-\nonumber\\
\frac{1}{2}T_{\mu\nu}\nabla^\nu R+\nabla^\nu(R_{\mu\nu}\mathcal{L}_m)+\frac{1}{2}(\nabla^\nu\Box T_{\mu\nu}\nonumber\\
+\nabla_\mu\nabla_\alpha\nabla_\beta T^{\alpha\beta}-\nabla^\nu\nabla_\alpha\nabla_{(\mu}T_{\nu)}^\alpha)\Bigg]\Bigg\}\nonumber\\
+\frac{1}{2}\alpha\frac{2\nabla^\nu T}{2-\alpha T}\Bigg\{\Bigg[\frac{1}{2}\alpha R(T_{\mu\nu}+\theta_{\mu\nu})\nonumber\\
+\frac{1}{2}\alpha g_{\mu\nu}R_{\alpha\beta}T^{\alpha\beta}-(g_{\mu\nu}\Box-\nabla_\mu\nabla_\nu)(1-\frac{1}{2}\alpha T)\Bigg]\nonumber\\
-\kappa ^2\left[\frac{1}{2}\alpha g_{\mu\nu} T_{\alpha\beta}T^{\alpha\beta}-T_{\mu\nu}\right]\nonumber\\
+\kappa ^2\alpha\Bigg[T_{\alpha(\mu}T^\alpha_{\nu)}-T(g_{\mu\nu}\mathcal{L}_m+T_{\mu\nu})+2\mathcal{L}_m T_{\mu\nu}\Bigg]\nonumber\\
-\alpha\left[R_{\alpha(\mu}T^\alpha_{\nu)}-\frac{1}{2}R(g_{\mu\nu}\mathcal{L}_m+T_{\mu\nu})+\mathcal{L}_m R_{\mu\nu}\right.\nonumber\\
\left.+\frac{1}{2}\left(\Box T_{\mu\nu}+\nabla_{\alpha}\nabla_{\beta}T^{\alpha\beta}g_{\mu\nu}-
\nabla_{\alpha}\nabla_{(\mu}T_{\nu)}^\alpha\right)\right]\Bigg\}. \nonumber\\
\eea
The above results show that generally in this class of models the matter energy-momentum tensor is not conserved. The non-conservation of $T_{\mu \nu}$ can be related to particle production processes that takes place due to the quantum fluctuations of the space-time metric.
\subsection{Cosmological applications}
For simplicity in the following we consider a spatially flat space-time, in which \(K=0\). By taking into account the intermediate results
\begin{eqnarray}
\Box T&=&(\ddot{\rho}-3\ddot{p})+(\dot{\rho}-3\dot{p})\frac{3\dot{a}}{a},\nonumber\\
\nabla^\alpha\nabla^\beta T_{\alpha\beta}&=&\frac{3\dot{a}}{a}\Bigg[(2\dot{\rho}+\dot{p})+2(\rho+p)\frac{\dot{a}}{a}\Bigg]+\nonumber\\
&&\ddot{\rho}+3(\rho+p)\frac{\ddot{a}}{a},
\end{eqnarray}
\begin{eqnarray}
\Box T_{\mu\nu}&=&\ddot{\rho}+\dot{\rho}\frac{3\dot{a}}{a},
\nabla_\mu \nabla_\nu T=\ddot{\rho}-3\ddot{p},
\nabla_\alpha\nabla_{(\mu} T^{\alpha}_{\nu)})\nonumber\\
&=&2\ddot{\rho}+2\dot{\rho}\frac{3\dot{a}}{a}, \mu=\nu=0,
\end{eqnarray}
we obtain first from the 00 component of Eq.~(\ref{ff1})
\begin{eqnarray}\label{111}
\frac{3\ddot{a}}{a}\Bigg[1+\alpha(\rho+3p)\Bigg]+\frac{\dot{a}^2}{a^2}3\alpha p=\nonumber\\
\frac{1}{2}\frac{1-\frac{1}{2}\alpha(\rho-3p)}{1-\frac{1}{2}\alpha(\rho- p)}\Bigg[-\alpha\frac{3\dot{a}}{a}\Bigg(\dot{\rho}+4\dot{p}+2(\rho+p)\frac{\dot{a}}{a}\Bigg)\nonumber\\
-3\alpha\ddot{p}-3\alpha(\rho+p)\frac{\ddot{a}}{a}+\kappa^2(\rho-3p)(1-\alpha(\rho-p))\Bigg]\nonumber\\
+\frac{\alpha}{2}\Bigg\{3(\rho+p)\frac{\ddot{a}}{a}+\frac{3\dot{a}}{a}\Bigg[4\dot{p}+2(\rho+p)\frac{\dot{a}}{a}\Bigg]\Bigg\}\nonumber\\
-\kappa^2\left[\rho-\alpha\left(\frac{1}{2}\rho^2+4\rho p+\frac{3}{2}p^2\right)\right].
\end{eqnarray}
For \(\mu=\nu=1\) we obtain
\begin{eqnarray}
\Box T_{\mu\nu}=a^2(\ddot{p}+\dot{p}\frac{3\dot{a}}{a}),
\nabla_\mu \nabla_\nu T=0,
\nabla_\alpha\nabla_{(\mu} T^{\alpha}_{\nu)})=0.\nonumber\\
\end{eqnarray}
Hence the 11 component of Eq.~(\ref{ff1}) gives
\begin{eqnarray}\label{112}
\frac{\dot{a}^2}{a^2}(2-\alpha(\rho-5p))+\frac{\ddot{a}}{a}(1-\alpha(2\rho-3p))+2\alpha\dot{p}\frac{\dot{a}}{a}=\nonumber\\
\frac{1}{2}\frac{1-\frac{1}{2}\alpha(\rho-3p)}{1-\frac{1}{2}\alpha(\rho- p)}\Bigg[-\alpha\frac{3\dot{a}}{a}\Bigg(\dot{\rho}+4\dot{p}+2(\rho+p)\frac{\dot{a}}{a}\Bigg)\nonumber\\
-3\alpha\ddot{p}-3\alpha(\rho+p)\frac{\ddot{a}}{a}+\kappa^2(\rho-3p)(1-\alpha\rho+\alpha p)\Bigg]\nonumber\\
+\frac{1}{2}\alpha\Bigg\{2\ddot{p}+3(\rho+p)\frac{\ddot{a}}{a}+\frac{3\dot{a}}{a}\Bigg[\dot{\rho}+3\dot{p}+2(\rho+p)\frac{\dot{a}}{a}\Bigg]\Bigg\}\nonumber\\
+\kappa^2\left(\frac{1}{2}\alpha(\rho^2+3p^2)+p\right).\nonumber\\
\end{eqnarray}
From Eqs.~(\ref{111}) and (\ref{112}) we obtain
\begin{eqnarray}
\frac{\ddot{a}}{a}=\frac{\Gamma \cdot \Delta-\Theta \cdot \Lambda}{\Xi},\\
\frac{\dot{a}^2}{a^2}=-\frac{\Pi\cdot \Gamma -\Sigma\cdot k}{\Xi}.\\
\end{eqnarray}
Therefore the generalized Friedmann equations for this gravity model take the form
\bea
3H^2=-3\frac{\Pi\cdot \Gamma-\Sigma\cdot \Theta}{\Xi},\\
2\dot{H}+3H^2=\frac{\Gamma\cdot (2\Delta-\Pi)+(\Sigma-l2)\cdot \Theta}{\Xi},
\eea
where we have denoted
\bea
\lambda=\frac{1-\frac{1}{2}\alpha(\rho-3p)}{1-\frac{1}{2}\alpha(\rho- p)},
\eea,
\bea
\Gamma=\frac{3}{2} \alpha \lambda \left( \frac{\dot{a}}{a} \dot{\rho}+\ddot{p}\right)+6 \alpha \frac{\dot{a}}{a} (\lambda-1) \dot{p}+\nonumber\\
\kappa^2 \left[\rho -\alpha \left(\frac{3 p^2}{2}+4 p \rho +\frac{\rho ^2}{2}\right)\right]\nonumber\\
-\frac{1}{2} \lambda \kappa^2 (\rho -3 p) \left(-\alpha \rho +\alpha p+1\right),
\eea
\bea
\Delta=2-\alpha ( \rho-5p) +3 \alpha (\lambda -1) (p+\rho ),
\eea
\bea
\Theta=\frac{3}{2} \alpha (\lambda-1) \frac{\dot{a}}{a} \dot{\rho}+6 \alpha \frac{\dot{a}}{a} \lambda \dot{p}-\frac{5 \alpha \dot{p}}{2}\frac{\dot{a}}{a} \nonumber\\
-\kappa^2\left(\frac{\alpha \rho ^2}{2}+\frac{3 \alpha p^2}{2}+p\right)+\frac{3 \alpha \lambda \ddot{p}}{2}-\nonumber\\
\alpha \ddot{p}-\frac{1}{2} \lambda \kappa^2 (\rho -3 p) (-\alpha \rho +\alpha p+1),
\eea
\bea
\Lambda=3 \alpha \lambda ( \rho+p) -3 \alpha \rho ,
\eea
\bea
\Xi=3\Bigg[\frac{\alpha ^2 \rho ^2}{2}(11-9\lambda)-3 \alpha \lambda ( \rho+p)\nonumber\\
+ \alpha (2 \rho-7p) -(5+7\lambda ) \alpha ^2 p^2\nonumber\\
+\frac{\alpha ^2 }{2}p \rho(15-23\lambda)-2\Bigg],
\eea
\bea
\Pi=\frac{3 \alpha \lambda }{2}(\rho +p)+\frac{ \alpha}{2}(3p-7\rho )+1,
\eea
\bea
\Sigma=3\Bigg[\frac{\alpha}{2} ( \lambda-1) (p+\rho )+\alpha (3 p+\rho )+1\Bigg].
\eea
For the deceleration parameter we obtain the expression
\be
q=\frac{d}{dt}\frac{1}{H}-1=-\frac{\ddot{a}}{aH^2}=\frac{\Gamma\cdot \Delta-\Theta\cdot \Lambda}{\Pi\cdot \Gamma-\Sigma\cdot \Theta}.
\ee
For \(\alpha=0\) we have
$\Gamma=\kappa^2(\rho c^2+3p)/2$, $\Delta=2$, $\Theta=-\kappa^2(\rho c^2+p)/2$, $\Lambda=0$,
$\Xi=-6$, $\Pi=1$, and $\Sigma=3$, respectively. Hence in this limit we recover the standard Friedmann equations of general relativity.
\subsection{ Dust cosmological models with $p=0$}
By assuming that the matter content of the Universe consists of pressureless dust with $p=0$, we obtain immediately $\lambda=1$. Then the field equations can be written as
\bea
\frac{\ddot{a}}{a}=-\frac{3 \alpha H\dot{\rho}+\kappa^2 \rho }{6 (\alpha \rho +1)},
\eea
and
\bea\label{128}
H^2=-\frac{3 \alpha H \dot{\rho}(1-2\alpha\rho)+\kappa^2 \rho(4+\alpha\rho) }{6 (\alpha \rho -2) (\alpha \rho +1)},
\eea
respectively. Thus we obtain the generalized Friedmann equations of the present model as
\bea\label{129}
3H^2=-\frac{3 \alpha H \dot{\rho}(1-2\alpha\rho)+\kappa^2 \rho(4+\alpha\rho) }{2 (\alpha \rho -2) (\alpha \rho +1)},
\eea
\bea\label{130}
2\dot{H}+3H^2=-\frac{\alpha (\kappa^2 \rho ^2-3H\dot{\rho})}{2 (\alpha \rho -2) (\alpha \rho +1)}.
\eea
For the deceleration parameter we obtain
\bea
q=\frac{(3 \alpha H\dot{\rho}+k^2 \rho)(\alpha \rho -2)}{3 \alpha H \dot{\rho}(1-2\alpha\rho)+\kappa^2 \rho(4+\alpha\rho)}.
\eea
By multiplying with \(a^3\) both sides of Eq.~(129), taking its time derivative, and considering Eq.~(130), we obtain the time evolution of the matter density as
\bea
\ddot{\rho}H=-\dot{\rho}\frac{\ddot{a}}{a}+\frac{1}{-2\alpha+3\alpha^2\rho+3\alpha^3\rho^2-2\alpha^4\rho^3}\times\nonumber\\
\Bigg\{\dot{\rho}\Bigg[\frac{\kappa^2}{3}(8+4\alpha\rho+5\alpha^2\rho^2)+\alpha H^2(10-3\alpha\rho-\nonumber\\
9\alpha^2\rho^2+4\alpha^3\rho^3)-H\dot{\rho}(5\alpha^2-2\alpha^3\rho+2\alpha^4\rho^2)\Bigg]\nonumber\\
+\kappa^2H(8\rho+\alpha\rho^2-4\alpha^2\rho^3)\Bigg\}.
\eea
\subsubsection{de Sitter type evolution of the dust Universe}
By assuming that $H=H_0={\rm constant}$, the cosmological evolution equations (\ref{129}) and (\ref{130}) give for the time evolution of the density the first order differential equation
\be
\dot{\rho} (t)= \frac{2 \kappa ^2 \rho (t)}{3 \alpha H_0 [\alpha
\rho (t)-1]},
\ee
with the general solution given by
\be
\alpha \rho -\ln (\rho )= \frac{2\kappa ^2}{3\alpha H_0}\left(t-t_0\right),
\ee
where $t_0$ is an arbitrary constant of integration. When $\alpha \rho >>\ln (\rho)$, the matter energy density linearly increases in time as $\rho (t)\propto t$, indicating that the late time de Sitter expansion of the Universe is triggered by (essentially quantum) particle creation processes.
\subsubsection{Cosmological evolution of the dust Universe}
From Eq.~(\ref{128}) we obtain for the time evolution of the matter density the differential equation
\be\label{135}
\dot{\rho }=\frac{6 H^2 (\alpha \rho -2) (\alpha \rho
+1)+\kappa ^2 \rho (\alpha \rho +4)}{3 \alpha H (2 \alpha \rho
-1)}.
\ee
Substitution of this equation into Eq.~(\ref{130}) gives the evolution equation of the Hubble function as
\be\label{136}
2\dot{H}=-3H^2+\frac{3 H^2-\kappa ^2 \rho }{2 \alpha \rho -1}.
\ee
By introducing the set of dimensionless variables $\left(\tau, h, \alpha _0,r\right)$, defined as
\be
\tau =H_0t,H=H_0h,\alpha =\alpha _0\frac{\kappa ^2}{3H_0^2}, \rho =\frac{3H_0^2}{\kappa ^2}r,
\ee
where $H_0$ is the present day value of the Hubble function, and after changing the independent time variable from $\tau $ to the redshift $z$, Eqs.~(\ref{135}) and (\ref{136}) take the form
\be\label{137}
\frac{dr}{dz}=-\frac{2h^2\left(\alpha _0r-2\right)\left(\alpha _0r+1\right)+r\left(\alpha _0r+4\right)}{\alpha _0(1+z)h^2\left(2\alpha _0r-1\right)},
\ee
\be\label{138}
\frac{dh}{dz}=\frac{3}{2}\frac{h}{1+z}-\frac{3\left(h^2-r\right)}{2(1+z)h\left(2\alpha _0r-1\right)}.
\ee
The deceleration parameter is given by
\be
q=\frac{1}{2}-\frac{3\left(h^2-r\right)}{2h^2\left(2\alpha _0r-1\right)}.
\ee
The system of differential equation must be integrated with the initial conditions $h(0)=h_0$ and $r(0)=r_0$, respectively. It is interesting to note that for
$h(0)=r(0)=1$, the present day deceleration parameter takes the value $q(0)=1/2$. Thus, in order to obtain accelerating models we will adopt an initial condition for the matter energy density so that $r(0)=0.67$, while for $h$ we adopt the initial condition $h(0)=1$. The variation with respect to the redshift $z$ of the dimensionless Hubble function $h$, of the dimensionless matter energy density $r$, and of the deceleration parameter $q$ are represented for $z\in [0,1]$ in Figs.~\ref{fig9} - \ref{fig11}, respectively.
\begin{figure}[tbp]
\includegraphics[width=7.5cm, angle=0]{fig9_Liu.eps}
\caption{Variation with respect to the redshift of the dimensionless Hubble function $h(z)$ for the modified gravity model with fluctuating quantum metric with $K_{\mu \nu} =\alpha T_{\mu \nu}$ for $h(0)=1$, $r(0)=0.67$, and for different values of $\alpha _0$: $\alpha _0=1$ (solid curve), $\alpha _0=1.1$ (dotted curve), $\alpha _0=1.2$ (short dashed curve), $\alpha _0=1.3$ (dashed curve), and $\alpha _0=1.4$ (long dashed curve), respectively.}
\label{fig9}
\end{figure}
\begin{figure}[tbp]
\includegraphics[width=7.5cm, angle=0]{fig10_Liu.eps}
\caption{Variation with respect to the redshift of the dimensionless matter energy density $r(z)$ for the modified gravity model with fluctuating quantum metric with $K_{\mu \nu} =\alpha T_{\mu \nu}$ for $h(0)=1$, $r(0)=0.67$, and for different values of $\alpha _0$: $\alpha _0=1$ (solid curve), $\alpha _0=1.1$ (dotted curve), $\alpha _0=1.2$ (short dashed curve), $\alpha _0=1.3$ (dashed curve), and $\alpha _0=1.4$ (long dashed curve), respectively.}
\label{fig10}
\end{figure}
\begin{figure}[tbp]
\includegraphics[width=7.5cm, angle=0]{fig11_Liu.eps}
\caption{Variation with respect to the redshift of the deceleration parameter $q(z)$ for the modified gravity model with fluctuating quantum metric with $K_{\mu \nu} =\alpha T_{\mu \nu}$ for $h(0)=1$, $r(0)=0.67$, and for different values of $\alpha _0$: $\alpha _0=1$ (solid curve), $\alpha _0=1.1$ (dotted curve), $\alpha _0=1.2$ (short dashed curve), $\alpha _0=1.3$ (dashed curve), and $\alpha _0=1.4$ (long dashed curve), respectively.}
\label{fig11}
\end{figure}
The Hubble function, depicted in Fig.~\ref{fig9}, is a monotonically increasing function of $z$, indication an expansionary evolution of the Universe. The numerical values of $h$ in the chosen redshift range show a very mild dependence on the numerical values of the parameter $\alpha _0$. The energy density of the matter $r$, presented in Fig.~\ref{fig10}, is also a monotonically increasing function of the redshift, indicating a time decrease of $r$ during the cosmological evolution. The numerical values of $r$ depend strongly on $\alpha _0$. Finally, the deceleration parameter $q$, shown in Fig.~\ref{fig11}, also has a strong dependence on $\alpha _0$. In the range $z\in (0.14,1]$, the Universe is in a marginally decelerating phase, with $q\approx 0$. In this redshift range the evolution is independent on the numerical values of $\alpha _0$. For $z<0.14$, the Universe enters in an accelerating phase, and, depending on the values of $\alpha _0$, a large range of accelerating models can be constructed. The numerical values of $q$ rapidly increase with increasing $\alpha _0$, so that for $\alpha _0=1$, $q(0)\approx -1$, while for $\alpha _0=1.4$, the deceleration parameter $q$ is constant in all range $z\in [0,1]$, $q\approx 0$.
\section{Discussions and final remarks}\label{sect5}
In the present paper we have considered the cosmological properties of some classes of modified gravity models that are obtained from a first order correction of the quantum metric, as proposed in \cite{re8,re11}. By assuming that the quantum metric can be decomposed into two components, and by substituting the fluctuating part by its (classical) average value $K_{\mu \nu}$, a large class of modified gravity models can be obtained. As a first step in our study we have derived the general Einstein equations corresponding to an arbitrary $K_{\mu \nu}$. An important property of this class of models is the non-conservation of the matter energy-momentum tensor, which can be related to the physical processes of particle creation due to the quantum effects in the curved space-time. By assuming that $K_{\mu \nu}\propto g_{\mu \nu}$, a particular class of the modified $f(R,T)$ gravity theory is obtained. We have investigated in detail the cosmological properties of these models, by assuming that the coupling coefficient between the metric and the average value of the quantum fluctuation tensor is a scalar field with a non-vanishing self-interaction potential, and a simple scalar function. The scalar field self-interaction potential was assumed to be of Higgs type \cite{Aad}, which plays a fundamental role in elementary particle physics as describing the generation of mass in the quantum world.
We have investigated two cosmological models, in which the scalar field is in the minimum of the Higgs potential, and the case of the "complete" Higgs potential. In both approaches in the large time limit the Universe enters an accelerating phase, with the accelerating de Sitter solution acting as an attractor for these cosmological models. However, in the case of the "complete" Higgs potential the redshift evolution of the deceleration parameter $q$ indicates at low redshifts an extremely complex, oscillating behavior. Such a dynamics, as well as the corresponding cosmological evolution may play a significant role in the inflationary/post inflationary reheating phase in the history of the Universe. By assuming that the coupling between the metric and the quantum fluctuations is given by a scalar function, two distinct cases of cosmological models can be obtained. By imposing the conservativity of the energy-momentum tensor for a Universe filled with matter obeying a linear barotropic equation of state, a decelarating cosmological model is obtained. Thus a model could be useful to describe the evolution of the high density Universe at large redshifts. An alternative model with matter creation can be also constructed, by imposing a specific equation of evolution for the time evolution of the coupling function $\alpha $. This model leads to an approximately de Sitter type expansion, with the deceleration parameter tending to minus one in the large time limit,
A second modified gravity model can be obtained by assuming that the average value of the quantum fluctuation tensor is proportional to the matter energy-momentum tensor, $K_{\mu \nu}\propto T_{\mu \nu}$. This choice leads to an extension of the $f\left(R,T,R_{\mu \nu}T^{\mu \nu}\right)$ gravity theory \cite{Har4,Odin}, with the quantum corrected action including an extra term $T_{\mu \nu}T^{\mu \nu}$. Hence by considering the effects of the quantum fluctuations of the metric proportional to the matter energy-momentum tensor a particular case of a general $f\left(R,T,R_{\mu \nu}T^{\mu \nu},T_{\mu \nu}T^{\mu \nu}\right)$ modified gravity theory is obtained. The numerical analysis of the cosmological evolution equations for this model show that, depending on the numerical values of the coupling constant $\alpha $, for a Universe filled with dust matter a large variety of cosmological behaviors can be obtained at low redshifts, with the deceleration parameter varying between a constant (approximately) zero value on a large redshift range, and a de Sitter phase reached at $z=0$.
Quantum gravity represents the greatest challenge present day theoretical physics faces. Since no exact solutions for this problem are known, resorting to some approximate methods for studying quantum effects in gravity seems to be the best way to follow. A promising path may be represented by the inclusion of some tensor fluctuating terms in the metric, whose quantum mechanical origin can be well understood. Interestingly enough, such an approach leads to classical gravity models involving geometry-matter coupling, as well as non-conservative matter energy-momentum tensors, and, consequently, to particle creation processes. Hence even the study of the gravitational models with first order quantum corrections can lead to a better understanding of the physical foundations of the modified gravity models with geometry-matter coupling. In the present paper we have investigated some of the cosmological implications of these models, and we have developed some basic tools that could be used to further investigate the quantum effects in gravity,
\section*{Acknowledgments}
T. H. would like to thank the Yat Sen School of the Sun Yat - Sen University in Guangzhou, P. R. China, for the kind hospitality offered during the preparation of this work. S.-D. L. would like to thank to the Natural Science Funding of Guangdong Province for financial support (2016A030313313).
| 670923c0d89ff63ea686a0001a03ada630be19b4 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Stronger Hypotheses}\label{sec_StrongerHypotheses}
In this section we shall prove Theorems \ref{thm_MainTheorem_two},\ref{thm_MainTheorem_three} and \ref{thm_MainTheorem_four}. Let us start with Theorem \ref{thm_MainTheorem_two}. Let $l$,$m$,$\<S_n : n > m + 1\>$, $\<n_k: k < \omega\>$ be as in its statement. As before we can find a stationary set $T$ of $X \subseteq H_{\aleph_\omega}$ s.t $\sup(X \cap \aleph_n) \in S_n$ for all $n > m + 1$ and in the coiteration of $K_X$ with $K$, which we can assume to be linear in this case, the $K$-side of the iteration drops and the $K_X$-side is trivial.
As before iteration indices are cofinal in $\beta^X_\omega$ and hence $\theta_X$ is a limit for all $X$. So we can fix an $n^*$ such that whenever $\nu^X_i \geq \beta^X_{n^*}$ then there is no drop between $i$ and $\theta_X$. Also remember that whenever $k > n^*$ and $j < l$, then $\mu_{n_k} = \mu_{n_k + j}$. For such $i$ that $\mu^X_i \geq \beta^X_{n^*}$ let us call the degree of elementarity of $\pi^X_{i,i+1}$ at that point $m^*$ and let us refer to the - constant in $i$ - cofinality of $\rho_{m^*}(N^X_i)$ as $\lambda_X$. Then there exist infinitely many $k > n^*$ s.t. $\mu_{n_k} \neq \lambda_X$.
An important difference is that we can no longer prove iteration indices to be cofinal in $\beta_{n_k}$ even if $\mu_{n_k} \neq \lambda_X$. In fact, we will show that this is not the case! This is because our extenders might now have many generators.
\begin{observation}\label{obs_succcof}
Let $k \geq m + 1$. Let $\alpha \in \left[\beta^X_{n_k},\beta^X_{n_k + l}\right)$ be s.t. $K^X \models \exists \gamma: \alpha = \gamma^+$. Then $\op{cof}(\alpha) = \mu_{n_k}$.
\end{observation}
\begin{proof}
If $\alpha = \beta^X_{n_k + j}$ for some $j < l$ then this is by choice of our sequence. If not, then $\alpha$ is properly in between say $\beta^X_{n_k + j}$ and $\beta^X_{n_k + j + 1}$ and thus by weak covering $\op{cof}(\sigma_X(\alpha)) = \aleph_{n_k + j}$. W.l.o.g $X$ is closed under some function witnessing this. But this easily gives $\op{cof}(\alpha) = \op{cof}(\beta^X_{n_k + j}) = \mu_{n_k + j} = \mu_{n_k}$.
\end{proof}
\begin{lemma}
Let $k > n^*$ be s.t. $\mu_{n_k} \neq \lambda_X$. Then there exist an $i < \theta_X$ s.t. $\kappa^X_i < \beta^X_{n_k} \leq \nu^X_i$.
\end{lemma}
\begin{proof}
Assume not. Because iteration indices are cofinal in $\beta^X_\omega$ there is some least $i$ s.t $\nu^X_i > \beta^X_{n_k}$. By assumption we have $\kappa^X_i \geq \beta^X_{n_k}$. Then by coherence and the fact that there is no drop in between $i$ and $\theta_X$ means that $((\beta^X_{n_k})^+)^{K_X} = ((\beta^X_{n_k})^+)^{M^X_i}$ is a regular cardinal of $N^X_i$. Furthermore, because by minimality of $i$ all generators of the iteration up to this point are less than $\beta^X_{n_k}$, $N^X_i$ is sound above $\beta^X_{n_k}$. Lastly, $((\beta^X_{n_k})^+)^{K_X} \leq \rho_{m}(N^X_i)$ because there is no drop at $i$. So, Lemma \ref{lem_teclemma} applies and gives us that $\op{cof}(((\beta^X_{n_k})^+)^{K_X}) = \lambda_X$. On the other hand by Observation \ref{obs_succcof} $\op{cof}(((\beta^X_{n_k})^+)^{K_X}) = \mu_{n_k}$. Contradiction!
\end{proof}
Note here that by the same proof we have that $\op{cof}(((\kappa^X_i)^+)^{N^X_i}) = \lambda_X$ and thus it should be easy to see that $((\kappa^X_i)^+)^{N^X_i} < \beta^X_{n_k}$.
So for any $k > n^*$ s.t. $\mu_{n_k} \neq \lambda_X$ we can fix some $i^X_k$ with $\kappa^X_{i^X_k} < \beta^X_{n_k}$ and $\nu^X_{i^X_k} \geq \beta^X_{n_k}$. To simplify our notation we shall henceforth refer to $\kappa^X_{i^X_k}$ as $\eta^X_k$, to $\nu^X_{i^X_k}$ as $\zeta^X_k$, to the model $N^X_{i^X_k}$ as $M^X_k$ and to the extender $E^X_{i^X_k}$ as $F^X_k$.
\begin{lemma}\label{lemma_nuplus}
Let $k > n^*$ be s.t. $\mu_{n_k} \neq \lambda_X$. Then $((\zeta^X_k)^+)^{K_X} \geq \beta^X_{n_k + l}$.
\end{lemma}
\begin{proof}
Consider $M^* := \op{Ult}(M^X_k,F^X_k)$. In $M^*$, $((\zeta^X_k)^+)^{N^*}$, which equals $((\zeta^X_k)^+)^{K_X}$ by coherence, is certainly regular and $M^*$ is sound above $\zeta^X_k$. Notice also that $\rho_{m^*}(M^*) \geq ((\zeta^X_k)^+)^{M^*}$. So, by Lemma \ref{lem_teclemma} $\op{cof}(((\zeta^X_k)^+)^{M^*}) = \lambda_X$. On the other hand $((\zeta^X_k)^+)^{K_X}$ is a successor. card. in $K_X$; if it were in the interval $\left[\beta^X_{n_k},\beta^X_{n_k + l}\right)$, by Observation \ref{obs_succcof} it's cofinality would equal $\mu_{n_k}$. So, we can conclude that $((\zeta^X_k)^+)^{K_X} \geq \beta^X_{n_k + l}$.
\end{proof}
We can immediately conclude that $\op{o}^{K_X}(\eta^X_k) \geq \beta^X_{n_k + (l - 1)}$ for every $k > n^*$ s.t. $\mu_{n_k} \neq \lambda_X$. By elementarity then $\op{o}^K(\sigma_X(\eta^X_k)) \geq \aleph_{n_k + (l - 1)}$. Also, we know there exists infinitely many such $k$. So this concludes the proof of Theorem \ref{thm_MainTheorem_two}.
The proof gives a slightly stronger conclusion!
\begin{fact}\label{thefanciestoffactums}
Assume that $\xi \in \left(\sigma_X(\eta^X_k),\aleph_{n_k + (l-1)}\right) \cap \op{Card}^K$, then $\op{o}^K(\sigma_X(\eta^X_k)) \geq \xi$.
\end{fact}
We will need the above fact for the proof of Theorem \ref{thm_MainTheorem_three}:
\begin{proof}[Proof of Theorem \ref{thm_MainTheorem_three}]
We will do the proof for cofinalities $\aleph_1$ and $\aleph_2$, it is not hard to see that this case is representative. We just need to consider two sequences $\vec{S} := \<S^n_{f(n)} : n \geq 8\>$ and $\vec{T}:=\<S^n_{g(n)} : n \geq 4\>$ where
\begin{align*}
f(n) & = \begin{cases} 1 & n \mod 8 = 0,1,2,3 \\ 2 & n \mod 8 = 4,5,6,7 \end{cases} \\
g(n) & = \begin{cases} 1 & n \mod 4 = 0,1 \\ 2 & n \mod 4 = 2,3 \end{cases}
\end{align*}
Assume both $\vec{S}$ and $\vec{T}$ are mutually stationary. Using Theorem \ref{thm_MainTheorem_two} we get a sequence $\<\kappa_n: n <\omega\>$ and $\<\lambda_n: n< \omega\>$ s.t. for all $n < \omega$ there exists $\kappa < \kappa_n$ and $\kappa' < \lambda_n$ with $\op{o}^K(\kappa) \geq \kappa^{+3}_n$ and $\op{o}^K(\kappa') \geq \lambda^+_n$.
As to the identity of the $\kappa_n$'s they are the $\aleph_k$'s with either $k$ at least some number $n^*$ and $k \mod 8 = 0$ or $k \geq n^*$ and $k \mod 8 = 4$. Similarly, the $\lambda_n$'s are the $\aleph_k$'s with either $k$ at least some number $n^*$ and $k \mod 4 = 0$ or $k \geq n^*$ and $k \mod 4 = 2$. Of course, we can assume the two $n^*$'s to be the same.
Our job is now to simply check all of the 4 possible combinations and see that there must be some overlap on the $K$-sequence. By symmetry it suffices to examine just two of those cases.
Take some $k$ big enough with $k \mod 8 = 0$. Assume there is some $\kappa < \aleph_k$ with $\op{o}^K(\kappa) \geq \aleph_{k+3}$. The first case we look at is that there is $\kappa' < \aleph_{k + 2}$ with $\op{o}^K(\kappa') \geq \aleph_{k + 3}$. This then tells us that there must be some other $\kappa'' < \aleph_{k - 2}$ with $\op{o}^K(\kappa'') \geq \aleph_{k - 1}$. In our situation we have that $\kappa < \aleph_k$ is a regular cardinal in $K$ thus by Fact \ref{thefanciestoffactums} we have that $\op{o}^K(\kappa'') \geq \kappa$. If $\nu$ was the index of the order zero measure on $\kappa$ then $K \vert\vert \nu$ is a $0^\P$ type mouse.
The other case works similar. Assume now that $\kappa' < \aleph_k$ with $\op{o}^K(\kappa') \geq \aleph_{k+1}$ exists. Then we also have $\kappa'' < \aleph_{k+4}$ measurable in $K$. As before - but applying Fact \ref{thefanciestoffactums} at $\kappa$ instead - we actually have $\op{o}^K(\kappa) \geq \kappa''$ and thus $0^\P$.
As mentioned before, the remaining two cases are dealt with by a symmetric argument.
\end{proof}
Finally, the proof of theorem \ref{thm_MainTheorem_four}:
So let us fix $ 1 < m < \omega$ and a mutually stationary sequence $\vec{S}:= \<S_n: n > m\>$ s.t for all $n$, $S_n$ concentrates on a fixed cofinality $\mu_n$ s.t. $\liminf\limits_{m < n < \omega} \mu_n = \aleph_\omega$. It is easy to see that we can require all the $\mu_n$ to be uncountable.
We shall do the following proof in greater generality. The above hypothesis is almost certainly very strong, close to inconsistent even. We believe it should be possible to extract an inner model with a Woodin cardinal from the hypothesis. Considering that the consistency of the statement is unsure, it might not be a worthy pursuit to do so.
We assume for a contradiction:
\begin{itemize}
\item[(a)] $K$ is a core model satisfying weak covering at all but finitely many cardinals;
\item[(b)] if $E$ is a total extender on the $K'$-sequence where $K' \trianglelefteq K$, $\kappa$ is it's critical point and $\nu$ it's index, then $\nu$ is a successor cardinal in $\op{Ult}(K';E)$ and $\op{cof}((\kappa^+)^{K'}) = \op{cof}(\nu)$;
\item[(c)] there exists some $X \prec (H_{\omega};\in, K \cap H_\omega,\ldots)$ s.t. $\sup(X \cap \aleph_n) \in S_n$ for all $m < n <\omega$ and in the co-iteration of $K$ and $K_X$ which is not necessarily linear, $K_X$ does not move and $K$ drops along its main branch.
\end{itemize}
The above is satisfied if $0^\P$ does not exist as evidenced by the core model below $0^\P$(, except (b) which is not quite true, but we can make do by substituting $(\nu^+)^{\op{Ult}(K',E)}$ for $\nu$, the former does have the right cofinality as shown in the proof of Lemma \ref{lemma_nuplus}). We do not know if it is satisfied if there is no inner model with a Woodin cardinal.
So, let us write $\mathcal{T}_X$ for the iteration tree on $K$ and $b_X$ for its main branch from assumption (c). Let $\theta_X$ be the length of $\mathcal{T}_X$, $\<N^X_i,\kappa^X_i,\nu^X_i,m^X_i: i < \theta_X\>$ be the iteration's models, critical points, indices and degrees. As before we can show that $b_X$ has limit type. So there is some $n^*$ s.t whenever $\nu^X_i \geq \beta^X_{n^*}$ and $i \in b$ then $\op{cof}(\rho_{m^X_i}(N^X_i))$ is constant in $i$. Call this constant value $\lambda_X$. W.l.o.g. $\lambda_X < \mu_n$ for all $n > n^*$.
\begin{observation}\label{obs_succcof_two}
Let $n > n^*$. Let $\alpha \in [\beta^X_n,\beta^X_{n+1})$ be s.t. $K^X \models \exists \gamma :\alpha = \gamma^+$. Then $\op{cof}(\alpha) > \lambda_X$.
\end{observation}
\begin{proof}
If $\alpha = \beta^X_{n+1}$ then this is by choice of our sequence. If not, then by weak covering $\op{cof}(\sigma_X(\alpha)) = \aleph_n$. W.l.o.g $X$ is closed under some function witnessing this. But this easily gives $\op{cof}(\alpha) = \op{cof}(\beta^X_n) = \mu_n > \lambda_X$.
\end{proof}
We can now derive a contradiction finishing the proof of theorem \ref{thm_MainTheorem_four}:
Let $i + 1 \in b$ be s.t. $\nu^X_i \geq \beta^X_{n^*}$. On the one hand we have that $\nu^X_i$ is a successor cardinal of $K_X$. Thus by Observation \ref{obs_succcof_two} $\op{cof}(\nu^X_i) > \lambda_X$.
On the other hand by assumption (b) $\op{cof}(\nu^X_i) = \op{cof}(((\kappa^X_i)^+)^{M^X_i})$; furthermore, $((\kappa^X_i)^+)^{M^X_i} = ((\kappa^X_i)^+)^{M^X_j}$ because of agreement between models in iteration trees and, crucially, the fact that there occur no more drops on $b_X$ from this stage on. (Here $j$ is the $\mathcal{T}^X$-predecessor of $i + 1$). Clearly though, $((\kappa^X_i)^+)^{M^X_j}$ is a regular cardinal of $M^X_j$, and that model is sound above $\kappa^X_i$. So, Lemma \ref{lem_teclemma} applies and gives $\op{cof}(((\kappa^X_i)^+)^{M^X_i}) = \lambda_X$. Hence $\op{cof}(\nu^X_i) = \lambda_X$. Contradiction!
\section{Introduction}
Mutual stationarity was originally introduced in \cite{MR1846032} to study saturation properties of non-stationary ideals.
\begin{definition}
Let $\lambda$ be an ordinal, any ordinal. Let $\<\kappa_i : i < \lambda\>$ be an increasing sequence of regular cardinals, $\bar{\kappa} := \sup\limits_{i < \lambda} \kappa_i$. We say a sequence $\<S_i : i < \lambda\>$, where $S_i \subseteq \kappa_i$ is stationary, is mutually stationary iff the set
\[ \{ A \subset \bar{\kappa} \vert \forall i < \lambda: \kappa_i \in A \Rightarrow \sup(A \cap \kappa_i) \in S_i\} \]
is stationary, i.e. contains a substructure of every structure with countable signature on $\bar{\kappa}$.
\end{definition}
The most remarkable result from the above paper is the $\op{ZFC}$ fact that any sequence of stationary sets all of which concentrate on points of countable cofinalities is mutually stationary, no matter its length. It is also shown that an analog theorem for sets concentrating on cofinality $\omega_1$ can not be proven in $\op{ZFC}$. We do not currently know if it is even consistent, though a lower bound for it's consistency is known (see \cite{MR2267155},\cite{MR2803948}).
However, we are only going to discuss sequences that do not concentrate on a fixed cofinality. We shall also limit ourselves to stationary subsets of the $\aleph_n$'s, $n$ a natural number.
All mutually stationary sequences appearing in the paper will have limit length.
We will mention some prior results involving sets concentrating on countable cofinality to draw some parallels with the results from this paper. We start with this result:
\begin{theorem}[Magidor]
Let $\<\kappa_i : i < \omega\>$ be an increasing sequence of measurable cardinals. Then there exists a generic extension of the universe $V\left[G\right]$ in which $\kappa_i$ becomes $\aleph_{2{i+1}}$ and the sequence $\<S^2_0,S^3_1,S^4_0,S^5_1,S^6_0,\ldots\>$ is mutually stationary.
\end{theorem}
(Note: We will often use the following notation: $S^m_n := \{ \alpha < \aleph_m \vert \op{cof}(\alpha) = \aleph_n \}$.)
The theorem can be stated more generally, the real limitations being that all but finitely many sets in the sequence concentrate on one of two cofinalities, $\omega$ being one of them, and if a set in the sequence concentrates on countable cofinality then the next one does not. Here the points in the sequence concentrating on countable cofinality correspond to former measurable cardinals and their successors are not collapsed in the construction and correspond to points concentrating on the other cofinality.
If one wants to do away with this non-accumulation property of points concentrating on countable cofinality one uses supercompact cardinals instead \footnote{Successors of supercompacts might be collapsed in this construction, this being a classic use of indestructability.}. In that case for any given $f:\omega \rightarrow 2$ in the ground model there exists a generic extension in which the sequence $\vec{S}^f_2 := \<S^m_{f(m)}: 2 \leq m < \omega\>$ is mutually stationary (see \cite{MR2250537}).
This result can be improved using a competely different approach. Jensen has shown the consistency of a forcing axiom (relative to one supercompact) that implies the mutual stationarity of $\vec{S}^f_2$ for all $f:\omega \rightarrow 2$ simultaneously. (See \cite{Jensen_FA_CH}.)
The Magidor result, too, can be improved:
\begin{theorem}[Koepke]
Let $\kappa$ be a measurable cardinal. Then there exists a generic extension of the universe $V\left[G\right]$ in which $\kappa$ becomes $\aleph_\omega$ and the alternating sequence $\<S^2_0,S^3_1,S^4_0,S^5_1,S^6_0,\ldots\>$ is mutually stationary.
\end{theorem}
It is not hard to see that this is optimal. There is an interesting switch that happened here. In the Koepke result different limitations apply: all but finitely many sets in the sequence concentrate on one of two cofinalities, $\omega$ being one of them, and if a set in the sequence concentrates on the \emph{other} cofinality then the next one does not. Here the points in the sequence concentrating on the \emph{other} cofinality correspond to points in a Prikry sequence and their successors are not collapsed in the construction and correspond to points concentrating on \emph{countable} cofinality. (See \cite{MR2276663}.)
This leads us to ask the following question. Is it possible to force the mutual stationarity of the sequence $\<S^2_0,S^3_1,S^4_1,S^5_0,S^6_1,S^7_1,\ldots\>$ from finitely many measurable cardinals?
From now on, all sets in a mutually stationary sequence will concentrate on points of uncountable cofinality. The following result is an analog to Magidor's result above.
\begin{theorem}[Liu-Shelah]\label{thm_LiuShelah_1}
Let $1 \leq m < k$ be natural numbers. Let $A \subset \omega$ be infinite s.t.
\[ n \in A \Rightarrow n + 1 \notin A\]
for all $n < \omega$. Let $f:\omega \rightarrow \{n,k\}$ be defined by
\[ f(n) :=\begin{cases} m & n \in A \\ k & n \notin A.\end{cases} \]
Let $\<\kappa_i : i < \omega\>$ be an increasing sequence of cardinals of Mitchell order at least $\omega_m + 1$. Then there exists a generic extension in which $\<\kappa_i : i < \omega\>$ is the increasing enumeration of $\<\aleph_n : n > k, n \in A\>$ and the sequence $\vec{S}^f_{k +1} := \< S^n_{f(n)} : k < n < \omega\>$ is mutually stationary.
\end{theorem}
(There was a significantly weaker precursor result in \cite{KLiuOldPaper}, but it has been superceded by this one from \cite{MR1469093}.)
We do not know about a higher level analog to the Koepke result, but we think that it should exist.
The Liu-Shelah paper \cite{MR1469093} has another result, one which is nominally very powerful.
\begin{theorem}[Liu-Shelah]\label{PCF}
Assume $\max(\op{pcf}(\{\aleph_n: n <\omega\})) = \aleph_{\omega + n^*}$. Let $1 < m^* < \omega$. Let $I$ be the ideal of finite subsets of $\omega$. Let $\<A_i:i < n^*\>$ be a partition of $\omega$ such that $\prod\limits_{k \in A_i} \aleph_k \slash I$ has true cofinality $\aleph_{\omega + {i + 1}}$ for $i < n^*$. Let $\<n_i: i < n^*\> \subset \left[1,m^*\right]$ be arbitrary. Define a function $f:\omega \rightarrow \omega$ by
\[ f(n) = n_i :\Leftrightarrow n \in A_i. \]
Then the sequence $\vec{S}^f_{m^* + 1}$ is mutually stationary.
\end{theorem}
Note that the requirement here is the failure of $\op{SCH}$ at $\aleph_\omega$. So, we are still below $0^\P$. We are interested to know if this theorem can be used to generate mutually stationary sequences not already covered by Theorem \ref{thm_LiuShelah_1}. For that end we do need the ability to control for the partition $\<A_i:i < n^*\>$\footnote{Doing so might necessitate large cardinals beyond $0^\P$.}. Unfortunately, we do not know how to do that. (See also Question \ref{question_5}.)
We now state the main results of this paper. Theorem \ref{thm_MainTheorem} shows that the upper bounds obtained by Liu-Shelah in Theorem \ref{thm_LiuShelah_1} are close to optimal:
\begin{theorem}\label{thm_MainTheorem}
Let $1 < m$ be a natural number. Suppose $\langle S_n \ | \ n \ge m + 1 \rangle$ is a mutually stationary sequence such that:
\begin{enumerate}
\item for every $n \ge m + 1$, $S_n$ is stationary in $\omega_n$ and concentrates on a fixed cofinality $\mu_n$;
\item $\langle \mu_n \ |\ n \ge m + 1 \rangle$ is \textbf{not} eventually constant; and
\item $\omega_1 \le \mu^*:= \liminf_{n \to \infty} \mu_n < \aleph_\omega$.
\end{enumerate}
Then there is an inner model $W$ such that: for infinitely many $n \in \omega$:
\[
V \models \big\{ \alpha < \omega_n \ | \ o^W(\alpha) \ge \mu^* \big\} \text{ is stationary in } \omega_n
\]
\end{theorem}
The hypotheses of Theorem \ref{thm_MainTheorem} are consistent, by the Liu-Shelah Theorem \ref{thm_LiuShelah_1}. For example, mutual stationarity of the sequence
\begin{equation*}
\langle S^3_2, S^4_1, S^5_2, S^6_1, \dots, S^{2k-1}_2, S^{2k}_1, \dots \rangle
\end{equation*}
falls under the hypothesis of Theorem \ref{thm_MainTheorem} (with $\liminf_{n \to \infty} \mu_n = \omega_1$).
We do not know if the hypotheses of these following theorems is consistent. Theorem \ref{thm_MainTheorem_two} and Theorem \ref{thm_MainTheorem_three} have analogs in the countable case, mentioned in the introduction, which we do know to be consistent. Therefore we are confident that these hypotheses will be found to be consistent in the end. We are less confident about Theorem \ref{thm_MainTheorem_four}, but will include it anyway as it presents only a minimal time investment.
Furthermore, these hypotheses cover the most obvious variations of the hypothesis of our main theorem, Theorem \ref{thm_MainTheorem}. We feel the paper would be incomplete without addressing them.
\begin{theorem}\label{thm_MainTheorem_two}
Assume $0^\P$ does not exist. Fix natural numbers $l > 1, m > 1$. Suppose $\langle S_n \ | \ n \ge m + 1 \rangle$ is a sequence such that for every $n \ge m + 1$:
\begin{enumerate}
\item $S_n$ is stationary in $\omega_n$ and concentrates on a fixed uncountable cofinality $\mu_n$; and there exists a strictly increasing sequence $\langle n_k : k < \omega \rangle$ with
\item $n_{k+1} \geq n_k + l$ for all $k < \omega$
\item $\langle \mu_{n_k} \ |\ k < \omega \rangle$ is not eventually constant
\item $\mu_{n_k} = \mu_{n_k + i}$ for all $k < \omega$ and $i < l$
\item $\langle S_n \ | \ n \ge m + 1 \rangle$ is mutually stationary.
\end{enumerate}
Then in $K$ there is an infinite sequence $\<\kappa_n: n < \omega\> \subset \{\aleph_n : n < \omega\}$ s.t for all $n < \omega$ there is $\kappa < \kappa_n$ s.t $(\kappa^+)^K < \kappa_n$ and $\op{o}^K(\kappa) \geq (\kappa_n)^{+(l-1)}$.
\end{theorem}
\begin{theorem}\label{thm_MainTheorem_three}
Let $1 \leq n,k < m < \omega$ and assume that the sequence $\langle S^{n+m}_{f(n)} : n < \omega\> \rangle$ is mutually stationary for all $f:\omega \rightarrow \{n,k\}$. Then $0^\P$ exists.
\end{theorem}
By the results of Liu-Shelah mentioned in Theorem \ref{thm_LiuShelah_1}, our Theorem \ref{thm_MainTheorem} is almost an equiconsistency. However, if we alter the assumption of Theorem \ref{thm_MainTheorem} to require that $\liminf_{n \to \infty} \mu_n = \aleph_\omega$, the consistency strength jumps considerably, as shown by the following Theorem \ref{thm_MainTheorem_four}. In fact, the hypotheses of Theorem \ref{thm_MainTheorem_four} is an apparent strengthening of stating that $\aleph_\omega$ is a Jonsson cardinal, which is not known to be consistent at all.
\begin{theorem}\label{thm_MainTheorem_four}
Fix $1 \leq m < \omega$. Suppose $\langle S_n \ | \ n \ge m \rangle$ is a mutually stationary sequence such that for every $n \ge m$:
\begin{enumerate}
\item $S_n$ is stationary in $\omega_n$ and concentrates on a fixed cofinality $\mu_n$;
\item $\liminf_{n \to \infty} \mu_n = \aleph_\omega$.\footnote{Equivalently, any given cofinality appears only boundedly often in $\<\mu_n: n \ge m\>$.}
\end{enumerate}
Then $0^\P$ exists.
\end{theorem}
\section{Open Problems}\label{sec_OpenProblems}
\begin{question}
Is it possible to force, starting from a model with at most finitely many measurable cardinals, that the sequence $\<S^2_0,S^3_1,S^4_1,S^5_0,S^6_1,S^7_1,\ldots\>$ is mutually stationary?
\end{question}
\begin{question}
What is the upper bound for the existence of a mutually stationary sequence satisfying the hypothesis of Theorem \ref{thm_MainTheorem_two}?
\end{question}
\begin{question}
Is the hypothesis of Theorem \ref{thm_MainTheorem_three} consistent relative to large cardinals?
\end{question}
\begin{question}
Does ``$\aleph_\omega$ is Jonsson" imply that there exists in a - possibly trivial - forcing extension $V\left[G\right]$ a mutually stationary sequence satisfying the hypothesis of Theorem \ref{thm_MainTheorem_four} relative to $V\left[G\right]$?
\end{question}
\begin{question}\label{question_5}
Is it possible to generate mutually stationary sequences not coverded by Theorem \ref{thm_LiuShelah_1}, e.g. the sequence $\<S^4_1,S^5_1,S^6_2,S^7_2,S^8_1,S^9_1,S^10_2,S^11_2,\ldots\>$, using Theorem \ref{PCF}, i.e. is it possible to have $\op{cof}(\prod\limits_{n \in A_0} \aleph_n) = \aleph_{\omega + 1}$ and $\op{cof}(\prod\limits_{n \in A_1} \aleph_n) = \aleph_{\omega + 2}$ where $A_0 := \{ n < \omega\vert n \mod 4 = 0,1\}$ and $A_1 := \{ n < \omega \vert n \mod 4 = 2,3\}$ or vice versa?
\end{question}
\section{Preliminaries}\label{sec_Prelims}
\subsection{Inner model theory}
Unless otherwise stated, we follow the conventions of Zeman~\cite{ZemanBook}, assume that $0^\P$ does not exist, and let $K$ denote the core model (see Chapter 8 of \cite{ZemanBook}). Like \cite{ZemanBook}, we use Jensen indexing of extenders. We will heavily depend on the following lemma.
\begin{lemma}\label{lem_teclemma}
Let $M$ be a premouse. Let $n$ and $\kappa$ be such that $M$ is $n + 1$-sound above $\kappa$. Assume that $\lambda \in M$ is such that
\[ \kappa < \lambda \leq \rho^M_n \]
and $\op{cof}^M(\lambda) > \kappa$. Then $\op{cof}^V(\lambda) = \op{cof}^V(\rho^M_n)$.
\end{lemma}
\begin{proof}
We can assume that $n = 0$, otherwise replace $M$ by its $n$-th reduct. Define $f:\op{On} \cap M \rightarrow \lambda$ by
\[ \xi \mapsto \sup(\op{Hull}^{M || \xi}_1(\kappa \cup \{p^M_1\}) \cap \lambda). \]
By assumption this is well-defined and cofinal. It is also clearly increasing. Hence, we are done.
\end{proof}
We will need the following basic fact about normal fine-structural iterations.
\begin{fact}[See Lemma 4.2.2 of \cite{ZemanBook}]\label{fact_DegreesStabilize}
Suppose $\langle M_i \ | \ i \le \theta \rangle$ is a normal fine-structural iteration of a premouse $M = M_0$. Let $\kappa_i$ denote the critical point of the $i$-th stage. Assume that the ultimate projectum of $M_0$ is $\le \kappa_0$. Then for every $i < \theta$, the ultimate projectum of $M_i$ is $\le \kappa_i$. Let $\text{deg}(M_i,\kappa_i)$ denote the maximal $n \in \omega$ such that $\kappa_i < \omega \rho^M_n$. If $\theta$ is a limit ordinal, then $\langle \text{deg}(M_i,\kappa_i) \ | \ i < \theta \rangle$ is eventually constant.
\end{fact}
\subsection{Facts about mutual stationarity}\label{sec_Prelim_MutualStat}
The following lemma will be used to modify the members of sets witnessing mutual stationarity:
\begin{lemma}\label{lem_AddOrdsToBottom}
Suppose $\langle S_n \ | \ n \ge n_0 \rangle$ is a sequence such that $S_n$ is a stationary subset of $\omega_n$ for every $n \ge n_0$. Fix an algebra $\mathfrak{A} = (H_{\aleph_{\omega+1}}, \in, \dots)$ and assume that $X \prec \mathfrak{A}$ and $\text{sup}(X \cap \omega_n) \in S_n$ for every $n \ge n_0$. Fix a regular uncountable $\mu < \aleph_\omega$ and set
\begin{equation*}
X':= \text{Sk}^{\mathfrak{A}}\big( X \cup \mu \big)
\end{equation*}
Then for all $n$ such that $\omega_n > \mu$:
\begin{equation*}
\text{sup}(X' \cap \omega_n) = \text{sup}(X \cap \omega_n)
\end{equation*}
\end{lemma}
\begin{proof}
The $\ge$ direction is trivial. For the $\le$ direction, fix an $n$ such that $\mu < \omega_n$. Let $\eta$ be an element of $\omega_n \cap X'$. Then there is a function $f \in X$ and an ordinal $\xi < \mu$ such that $\eta = f(\xi)$. Let $h$ be the restriction of $f$ to those inputs from $\mu$ whose outputs are below $\omega_n$. Since $\mu$ is among the $\aleph_k$'s then $\mu \in X$, and so since $f \in X$ it follows that $h \in X$. Since $\omega_n$ is regular and $> \mu$ then $\text{sup}(\text{range}(h) ) \in X \cap \omega_n$. Thus $\eta = f(\xi) = h(\xi) < \text{sup}( X \cap \omega_n)$.
\end{proof}
\begin{corollary}\label{cor_ReplaceBottom}
Suppose $\vec{S} = \langle S_n \ | \ n \ge n_0 \rangle$ is mutually stationary, where $S_n \subset \omega_n$ for each $n \ge n_0$. Let $\mu < \aleph_\omega$ be fixed, and let $n_1$ be such that $\mu < \omega_{n_1}$. Then the mutual stationarity of $\langle S_n \ | \ n \ge n_1 \rangle$ is witnessed by models which contain $\mu$ as a subset.
\end{corollary}
The following lemma can be easily proved by induction on $n$:
\begin{lemma}\label{lem_Unif_Implies_Cover}
Assume $\mu < \aleph_\omega$ is regular, $\mu \subset X \prec H_{\aleph_{\omega+1}}$, and $\text{sup}(X \cap \omega_n)$ has cofinality $\ge \mu$ whenever $\omega_n \ge \mu$. Then for every such $n$, every $<\mu$-sized subset of $X \cap \omega_n$ is covered by some $<\mu$-sized set from $X$. In particular, $X \cap \aleph_\omega$ is a $<\mu$-closed set of ordinals.
\end{lemma}
\section{Proof of Theorem \ref{thm_MainTheorem}}
In this section we prove Theorem \ref{thm_MainTheorem}. Define
\begin{equation*}
\mu^*:= \liminf_{n \to \infty} \mu_n
\end{equation*}
Recall we are assuming that
\begin{equation}\label{eq_MuStarLessOmega}
\mu^* < \aleph_\omega
\end{equation}
\begin{remark}
The case where $\mu^* = \aleph_\omega$ is Theorem \ref{thm_MainTheorem_four}. However, unlike the assumptions of Theorem \ref{thm_MainTheorem}, the assumptions of Theorem \ref{thm_MainTheorem_four} are not known to be consistent.
\end{remark}
As described in Section \ref{sec_Prelims}, we work with the core model $K$ below $0^\P$.\footnote{If $0^\P$ exists then by iterating $0^\P$ one easily obtains an inner model as in the conclusion of Theorem \ref{thm_MainTheorem}.}
First we state a couple of theorems which are proved in \cite{CoxCoveringPaper}:
\begin{theorem}[\cite{CoxCoveringPaper}, Lemma 44]\label{thm_MitchellTheorem}
Let $K$ be the core model below 0-pistol and $\lambda$ an uncountable cardinal. Assume $S$ is a stationary collection of $X \prec H_{\lambda}$ such that
\begin{equation*}
\text{cof}(\omega) \cap \lambda \cap \text{lim}(X \cap \lambda) \subset X
\end{equation*}
For each $X \in S$ let $\sigma_X: H_X \to X \prec H_\lambda$ be the inverse of the Mostowski collapse of $X$, and let $K_X:= \sigma_X^{-1}[K\cap H_\lambda]$. Then for all but nonstationarily many $X \in S$, in the coiteration of $K$ with $K_X$:
\begin{itemize}
\item The $K$ side truncates to a mouse of size at most $|\text{crit}(\sigma_X)|$ by stage 1 of the coiteration;
\item the $K_X$ side of the coiteration is trivial.
\end{itemize}
\end{theorem}
\begin{notation}\label{notation_IterationStuff}
Let $S$ be as in the hypothesis of Theorem \ref{thm_MitchellTheorem}. For each $X \in S$ we let $\theta_X$ denote the length of the $K$ versus $K_X$ coiteration, and let $\langle N^X_i, \kappa^X_i, E^X_i \ | \ i < \theta_X \rangle$ denote the sequence of mice, critical points, and applied extenders on the $K$ side of the coiteration.\footnote{Recall from Theorem \ref{thm_MitchellTheorem} that the $K_X$ side of the coiteration is trivial.} For $i \le j < \theta_X$ let $\pi^X_{i,j}$ denote the (possibly partial) iteration map from $N^X_i \to N^X_j$.
\end{notation}
The following theorem was a generalization of a Covering Theorem of Mitchell:\footnote{E.g.\ it removed all cardinal arithmetic assumptions from the hypotheses.}
\begin{theorem}[Theorem 1 of Cox~\cite{CoxCoveringPaper}]\label{thm_CovThm}
Assume $0^\P$ does not exist, and let $K$ be the core model. Suppose $\gamma$ is an ordinal, $\gamma > \omega_2$, $\text{cf}(\gamma) < |\gamma|$, and $\gamma$ is regular in $K$. Then $\gamma$ is measurable in $K$. Moreover, if $\text{cf}(\gamma) > \omega$ then in $K$, $\gamma$ has Mitchell order at least $\text{cf}^V(\gamma)$.
\end{theorem}
We now commence with the proof of Theorem \ref{thm_MainTheorem}. Fix a large regular $\theta$ and a structure $\mathfrak{A} = (H_\theta, \in, \vec{S}, \dots)$ for the remainder of the proof. For each $X$ witnessing mutual stationarity of $\vec{S}$, let $\sigma_X: H_X \to X \prec \mathfrak{A}$ be the inverse of the collapsing map of $X$ and let $K_X$ denote $\sigma_X^{-1}[K \cap H_\theta]$.
Recall that we are assuming $\mu^*= \liminf_{n \to \infty} \mu_n < \aleph_\omega$. By Corollary \ref{cor_ReplaceBottom}, if we let $m_1$ be large enough so that $\omega_{m_1} > \mu^*$, then the mutual stationarity of $\langle S_n \ | \ n \ge m_1 \rangle$ is witnessed by sets containing $\mu^*$ as a subset; let $T$ denote this stationary set. Lemma \ref{lem_Unif_Implies_Cover}, together with the fact that $\mu_n \ge \mu^*$ for all $n \ge m_1$ and $\mu^* \subset X$ for all $X \in T$, yields:
\begin{observation}\label{obs_OmegaClosedOrds}
For every $X \in T$, $X \cap \aleph_\omega$ is closed under limits of cofinality less than $\mu^*$. In particular, since Theorem \ref{thm_MainTheorem} assumes that $\mu^* \ge \omega_1$, then $X \cap \aleph_\omega$ is an $\omega$-closed set of ordinals and thus Theorem \ref{thm_MitchellTheorem} applies.
\end{observation}
For $X \in T$ let $\beta^X_\omega:= \sigma_X^{-1}(\aleph_\omega)$. By Observation \ref{obs_OmegaClosedOrds} and Theorem \ref{thm_MitchellTheorem}, for every $X \in T$ the following facts hold for the coiteration of $K$ with $K_X ||\beta^X_\omega$:
\begin{equation}\label{eq_K_X_side_trivial}
\text{ the $K$ versus $K_X || \beta^X_\omega$ coiteration is trivial on the $K_X || \beta^X_\omega$ side}
\end{equation}
and
\begin{equation}\label{eq_Truncates}
K \text{ truncates to a mouse of size at most } |\text{crit}(\sigma_X)| \text{ by stage 1}
\end{equation}
For each $X \in T$ and $n > m_1$ let
\begin{equation*}
\beta^X_n:= \sigma_X^{-1}(\omega_n)
\end{equation*}
Since $\text{cf}(X \cap \omega_n) = \mu_n$ for all $n > m_1$, then $\text{cf}^V(\beta^X_n) = \mu_n$. So the assumptions of the theorem imply that for every $X \in T$:
\begin{equation}\label{eq_Vec_beta_not_Constant}
\langle \text{cf}^V(\beta^X_n) \ | \ n >m_1 \rangle \text{ is not eventually constant}
\end{equation}
Let $\theta_X$ denote the length of the coiteration of $K$ with $K_X || \beta^X_\omega$; equivalently, $\theta_X$ is the least stage of the $K$ versus $K_X$ coiteration such that all disagreements below $\beta^X_\omega$ have been resolved. Let $\langle N^X_i, \kappa^X_i, \nu^X_i \ | \ i < \theta_X \rangle$ denote the mice, critical point, and iteration index of the mouse on the $K$-side of the coiteration of $K$ with $K_X || \beta_\omega^X$. Note that by \eqref{eq_K_X_side_trivial} it follows that for all $i < \theta_X$:
\begin{equation}
\nu^X_i = \text{o}^{K_X}(\kappa^X_i)
\end{equation}
The following argument is due to Magidor:
\begin{lemma}[Magidor~\cite{MR939805}]\label{lem_Magidor}
For every $X \in T$:
\begin{equation*}
\{ \kappa^X_i \ | \ i < \theta^X \} \cap \beta^X_\omega \text{ is cofinal in } \beta^X_\omega
\end{equation*}
\end{lemma}
\begin{proof}
Assume not. By \eqref{eq_K_X_side_trivial} and universality of $K$, $M^X_{\theta_X}$ end extends $K_X || \beta^X_\omega$. Let $\eta_X$ be the strict supremum of $\{ \kappa_i^X \ | \ i < \theta_X \}$; by assumption, $\eta_X < \beta^X_\omega$. Now \eqref{eq_Truncates} implies that $M^X_{\theta_X}$ projects below $\eta_X$ and is sound above $\eta_X$.\footnote{This is a routine inductive proof; see e.g.\ the proof of Lemma 6.6.4 of Zeman~\cite{ZemanBook}.} Let $M$ be the maximal initial segment of $M^X_{\theta_X}$ such that $\beta^X_\omega$ is a cardinal in $M$. If $M = M^X_{\theta_X}$ then we have already shown that there is some $\eta < \beta^X_\omega$ such that $M$ projects below $\eta$ and is sound above $\eta$. If $M$ is a proper initial segment of $M^X_{\theta_X}$ then, since $\beta^X_\omega$ is definably collapsed over $M$, it follows that $M$ projects strictly below $\beta^X_\omega$ and, being a proper initial segment of a mouse, is (fully) sound. In either case there are $n^*,m^* < \omega$ such that
\[ \rho^M_{n^*+1} \leq \beta^X_k < \beta^X_\omega \le \rho^M_{n^*} \]
for all $k \geq m^*$. Fix any $k > m^*$. Since $\beta^X_k$ is regular in $K_X$, $\beta^X_\omega$ is a cardinal in $M$, and $M$ end-extends $K_X || \beta^X_\omega$, it follows by acceptability that $\beta^X_k$ is regular in $M$. But then by Lemma \ref{lem_teclemma} together with the soundness properties of $M$ established above, $\op{cof}(\beta^X_k) = \op{cof}(\rho^M_{n^*})$ for all but finitely many $k$. This contradicts \eqref{eq_Vec_beta_not_Constant}.
\end{proof}
Note that Lemma \ref{lem_Magidor} implies that:
\begin{equation}\label{eq_Theta_Limit_Ord}
\forall X \in T \ \ \theta_X \text{ is a limit ordinal}
\end{equation}
Lemma \ref{lem_Magidor}, together with the fact that there are only finitely many truncations in an iteration, yield that for every $X \in T$ there is an $n_X \in \omega$ such that, whenever $i < \theta_X$ and $\kappa^X_i \ge \beta^X_{n_X}$, then $i$ is not a truncation stage; i.e.\ all truncations of the $K$ versus $K_X ||\beta^X_\omega$ coiteration must occur before the critical points reach $\beta^X_{n_X}$. Using \eqref{eq_Theta_Limit_Ord} and Fact \ref{fact_DegreesStabilize}, it follows that for each $X \in T$ the sequence
\begin{equation*}
\langle \op{deg}(N^X_i, \kappa^X_i) \ | \ \kappa^X_i \ge \beta^X_{n_X} \ \text{and} \ i < \theta_X \rangle
\end{equation*}
is, eventually, a constant sequence of natural numbers.
So by increasing $n_X$ if necessary, we may also assume that $\op{deg}(N^X_i, \kappa^X_i)$ is constant with value $m_X$ for all $i$ such that $\kappa^X_i \ge \beta^X_{n_X}$. By countable completeness of the nonstationary ideal:
\begin{equation}\label{eq_m_star_n_star}
\exists m^*, n^* \in \omega \ \ \exists T' \subset T \text{ stationary } \forall X \in T' \ n_X = n^* \ \text{and} \ m_X = m^*
\end{equation}
Let $X \in T'$. Since (total) iteration maps are cofinal, we have that the cofinality of $\rho_{m^*}(N^X_i)$ is constant for all $i$ which satisfy:
\begin{equation}\label{eq_NoDropInterval}
\beta^X_{n^*} \le \kappa^X_i < \beta^X_\omega
\end{equation}
For each $X \in T'$ let $\lambda_X$ denote the constant cofinality of $\rho_{m^*}(N^X_i)$, for those $i$ satisfying (\ref{eq_NoDropInterval}).
For each $n \in \omega$ define:
\begin{equation}\label{eq_Theta_X_n}
\theta^X_n:= \text{ the least stage such that } \kappa^X_{\theta^X_n} \ge \beta^X_n
\end{equation}
Note that:
\begin{equation}
\forall n \in \omega \ \ \Big|K_X || \beta^X_n\Big| = \Big| X \cap \omega_n \Big| < \aleph_\omega
\end{equation}
Combined with \eqref{eq_Truncates} and Lemma 4.4.1 of Zeman~\cite{ZemanBook},
this implies that $|N^X_i| < \aleph_\omega$ for all $i \in (1, \theta_X)$. In particular:
\begin{equation}
\forall X \in T' \ \ \lambda_X < \aleph_\omega
\end{equation}
So $\lambda_X \in \{ \omega_k \ |\ k \in \omega \} \subset X$. Thus by pressing down there is some fixed infinite cardinal $\lambda^* < \aleph_\omega$ and a stationary $T'' \subset T'$ such that $\lambda_X = \lambda^*$ for all $X \in T''$. Since $\langle \mu_n \ | \ n > m_1 \rangle$ is not eventually constant:
\begin{equation}\label{eq_InfinitelyOften}
I:= \{ n \in \omega \ | \ \lambda^* \ne \mu_n \big( = \text{cf}^V(\beta_n) \big) \} \text{ is infinite}
\end{equation}
We now consider two cases. If, for some $X \in T''$ and $n \in I \cap (n^*, \omega)$, there is an iterate $N^X_i$ such that $\text{crit}(E^X_i) < \beta^X_n$ but the generators of $E^X_i$ are cofinal in $\beta^X_n$, then by iterating this extender we can obtain a model as in the conclusion of Theorem \ref{thm_MainTheorem}. \textbf{So from now on we assume there is no such extender}, i.e.\ assume:
\begin{eqnarray}\label{eq_OnlyOneGenerator}
\begin{split}
\forall X \in T'' \ \ \forall n \in I \cap (n^*, \omega) \ \ \forall i < \theta^X_n \\
\text{ the generators of } E^X_i \text{ are bounded below } \beta^X_n
\end{split}
\end{eqnarray}
\begin{lemma}\label{lem_CofInBeta_n}
For every $X \in T''$ and for all $n \in I \cap (n^*, \omega)$: the critical points of the coiteration are cofinal in $\beta_n^X$.
\end{lemma}
\begin{proof}
Fix $n > n^*$ s.t. $n \in I$; i.e.\ $\op{cof}(\beta^X_n) \neq \lambda^* = \lambda_X$. Now let us assume for a contradiction that there is an $i$ with $\kappa^X_i < \beta^X_n$ but $\kappa^X_{i+1} \geq \beta^X_n$. Note that $\beta^X_n$ is regular in $K_X$. Since $K_X || \beta_\omega^X$ doesn't move in the coiteration, $i > n^*$ (in particular $i$ isn't a truncation stage), and by acceptability, it follows that $\beta^X_n$ is also regular in $N^X_{i+1}$. Furthermore, assumption \eqref{eq_OnlyOneGenerator} implies that the generators of $E^X_i$ are bounded by some $\zeta < \beta^X_n$, which in turn implies that $N^X_{i+1}$ is sound above $\zeta + 1$. Also $\rho_{m^*}(N^X_{i+1}) \geq \kappa^X_{i+1} \geq \beta^X_n$ (recall $m^*$ was defined in \eqref{eq_m_star_n_star} as the uniform eventual value of $\op{deg}(N^X_j, \kappa^X_j)$). So we can apply Lemma \ref{lem_teclemma} to conclude that $\op{cof}(\beta^X_n) = \lambda_X$. But this contradicts our choice of $n$!
\end{proof}
In particular if $X \in T''$ and $n \in I \cap (n^*,\omega)$, then $\theta_n^X$ is a limit ordinal and $\text{cf}^V(\theta^X_n) = \text{cf}^V(\beta^X_n) = \mu_n$; here $\theta^X_n$ is as defined in \eqref{eq_Theta_X_n}.
\begin{lemma}\label{lem_thread}
Let $X \in T''$ and $n \in I \cap (n^*, \omega)$. Then the following set is closed and unbounded in $\theta^X_n$:
\begin{equation*}
C^X_n:= \{ j < \theta^X_n \ | \ \pi^X_{j, \theta^X_n}(\kappa^X_j) = \beta^X_n \}
\end{equation*}
\end{lemma}
\begin{proof}
First we show that $C^X_n$ is unbounded in $\theta^X_n$. Assume not, and let $i_0 < \theta^X_n$ be a bound on $C^X_n$. By Lemma \ref{lem_CofInBeta_n}, $\theta^X_n$ is a limit ordinal. So there is some $j^* \in (i_0, \theta^X_n)$ such that $\beta^X_n$ has a preimage in $N^X_{j^*}$, say $\bar{\beta}$. We claim that
\begin{equation}\label{eq_LessBarBeta}
\kappa^X_{j^*} < \bar{\beta}
\end{equation}
Suppose not. Our assumptions imply that these two ordinals are not equal, so it must be that $\kappa^X_{j^*} > \bar{\beta}$. But $\kappa^X_{j^*} < \beta^X_n$ (since $j^* < \theta^X_n$), so since $\pi^X_{j^*,\theta^X_n} \restriction \kappa^X_{j^*} = \text{id}$ this would imply that $\beta^X_n < \beta^X_n$, a contradiction.
Since $\beta^X_n$ is regular in $K_X || \beta^X_\omega$, $\theta^X_n$ is past all truncation points of the $K$ versus $K_X || \beta^X_\omega$ coiteration, and $K_X$ does not move in the coiteration, it follows that $\beta^X_n$ is regular in $N^X_{\theta^X_n}$. So by elementarity of the iteration map:
\begin{equation}\label{eq_BarBetaRegular}
\bar{\beta} \text{ is regular in } N^X_{j^*}
\end{equation}
So, our iteration embeddings are continuous at $\bar{\beta}$ and thus $\op{cof}(\bar{\beta}) = \op{cof}(\beta^X_n) \neq \lambda_X$, where the latter inequality is because $n \in I$.
On the other hand $N^X_{j^*}$ is sound above $\kappa^X_{j^*}$, $\bar{\beta}$ is regular in $N^X_{j^*}$ by \eqref{eq_BarBetaRegular}, and $\bar{\beta}$ is strictly above $\kappa^X_{j^*}$ by \eqref{eq_LessBarBeta}. So we can conclude by Lemma \ref{lem_teclemma} that $\op{cof}(\bar{\beta}) = \lambda_X$. This is a contradiction, completing the proof that $C^X_n$ is unbounded. That $C^X_n$ is closed below $\theta^X_n$ is a routine exercise, using the fact that the critical points of the iteration are increasing.
\end{proof}
Let $I'$ denote the tail end of $I$ beyond $n^*$, and also ensure that
\[
(\mu^*)^+ < \omega_{\text{min}(I')}
\]
For the rest of the proof, fix some $n \in I'$; by \eqref{eq_InfinitelyOften} there are infinitely many such $n$. Also fix some $X \in T''$. Observe that if $C^X_n$ is as in the statement of Lemma \ref{lem_thread}, then
\begin{equation*}
D^X_n:= \{ \alpha \ | \ \exists j \in C^X_n \ \ \alpha = \kappa^X_j \} \text{ is club in } \beta_n
\end{equation*}
Also observe that if $j \in C^X_n$ then since $j$ is past all truncations, $\kappa^X_j$ is a regular cardinal in the $j$-th iterate of $K_X$; but since $K_X$ doesn't move in the coiteration this just means $\kappa^X_j$ is regular in $K_X$. Thus
\begin{equation*}
\forall \alpha \in D^X_n \ \ K_X \models \ \alpha \text{ is regular}
\end{equation*}
and so by elementarity of $\sigma_X$ it follows that:
\begin{equation}\label{eq_WideTildeReg}
\forall \alpha \in \widetilde{D}^X_n:= \sigma_X[D^X_n] \ \ \ K \models \alpha \text{ is regular}
\end{equation}
By Observation \ref{obs_OmegaClosedOrds}, $\widetilde{D}^X_n$ is closed under limits of cofinality $<\mu^*$. Also $\widetilde{D}^X_n$ is cofinal in $\text{sup}\big( \sigma_X [ \beta_n] \big) = \text{sup}(X \cap \omega_n)$. Together with \eqref{eq_WideTildeReg} it follows that
\begin{align}\label{eq_AllLimitsOfReg}
\begin{split}
\forall \eta \in \text{lim} \big( \widetilde{D}^X_n \big) \cap \text{cof}(\ge \mu^*) \text{, all but nonstationarily} \\
\text{many members of } \eta \cap \text{cof}(<\mu^*) \text{ are regular in } K
\end{split}
\end{align}
The notation $\text{lim} \big( \widetilde{D}^X_n \big) \cap \text{cof}(\ge \mu^*)$ in \eqref{eq_AllLimitsOfReg} really means \textbf{all} limits of $\widetilde{D}^X_n$ of cofinality $\ge \mu^*$, not just those below $\text{sup}(X \cap \omega_n)$. In particular, it includes the ordinal $\text{sup}(X \cap \omega_n)$.\footnote{And $\text{sup}(X \cap \omega_n)$ might be the \emph{only} element of $\text{lim} \big( \widetilde{D}^X_n \big) \cap \text{cof}(\ge \mu^*)$, in the case that $\mu^* = \mu_n$.}
\begin{globalClaim}\label{clm_EtaLargeOrder}
If $\eta \in \text{lim} \big( \widetilde{D}^X_n \big) \cap \text{cof}(\ge \mu^*)$ \textbf{and} $\text{cf}(\eta) < \omega_{n-1} \le \eta$, then $o^K(\eta) \ge \text{cf}^V(\eta) \ge \mu^*$.
\end{globalClaim}
\begin{proof}
Fix such an $\eta$. The assumptions of the claim guarantee that $\omega < \text{cf}(\eta) < |\eta|$ and that $\eta > \omega_2$; so by Theorem \ref{thm_CovThm}, to prove that $o^K(\eta) \ge \text{cf}^V(\eta)$ it suffices to prove that $\eta$ is regular in $K$. Suppose for a contradiction that $\eta$ is singular in $K$. In $K$, fix some continuous $\vec{\eta} = \langle \eta_i \ | \ i < \text{cf}^K(\eta) \rangle$ which is cofinal in $\eta$ and such that $\eta_0 > \text{cf}^K(\eta)$. Then every member of
\[
E:= \{ \eta_i \ | \ i \text{ is a limit ordinal} \}
\]
is singular in $K$,\footnote{Because $\vec{\eta} \restriction i$ witnesses singularity of $\eta_i$.} and moreover $E$ is club in $\eta$. So in particular, almost every member of $\eta \cap \text{cof}(<\mu^*)$ is singular in $K$. This contradicts \eqref{eq_AllLimitsOfReg}.
\end{proof}
\begin{globalClaim}\label{clm_StatMany}
The set of $\eta$ which satisfy the assumptions of Claim \ref{clm_EtaLargeOrder} is stationary in $\omega_n$.
\end{globalClaim}
\begin{proof}
Note that $\mu_n \ge \mu^*$; we consider two cases, depending on whether this inequality is strict.
If $\mu_n = \mu^*$ then $\text{sup}(X \cap \omega_n)$ is a $\mu^*$-cofinal limit of $\widetilde{D}^X_n$.\footnote{Possibly the only such limit of $\widetilde{D}^X_n$; i.e.\ in the case $\mu_n = \mu^*$, then at most nonstationarily many members of $\widetilde{D}^X_n$ are $\mu^*$-cofinal.} Also, since $n \in I'$ then $\mu^* < \omega_{n-1}$, and so the cofinality of $\text{sup}(X \cap \omega_n)$ is strictly less than $\omega_{n-1}$. Finally, note that
\begin{equation*}
\bigcup_{X \in T''} \{ \text{sup}(X \cap \omega_n) \}
\end{equation*}
is stationary in $\omega_n$, because $T''$ is stationary.
If $\mu_n > \mu^*$, then $Q^X_n:= \text{lim}(\widetilde{D}^X_n) \cap \text{cof}(\mu^*) \cap [\omega_{n-1}, \text{sup}(X \cap \omega_n))$ is stationary (in fact $\mu^*$-club) in $\text{sup}(X \cap \omega_n)$ for all $X \in T''$. Also since $n \in I'$ then $\mu^* < \omega_{n-1}$. It follows that every $\eta \in Q^X_n$ satisfies the assumptions of Claim \ref{clm_EtaLargeOrder}. Finally, note that because each $Q^X_n$ is stationary in $\text{sup}(X \cap \omega_n)$ and $T''$ is stationary, it follows that
\begin{equation*}
\bigcup_{X \in T''} \ \ Q^X_n
\end{equation*}
is stationary in $\omega_n \cap \text{cof}(\mu^*)$, which completes the proof of the claim.
\end{proof}
Thus Claims \ref{clm_EtaLargeOrder} and \ref{clm_StatMany} imply that for any $n \in I'$, there are stationarily many $\eta < \omega_n$ such that $o^K(\eta) \ge \mu^*$. This completes the proof of Theorem \ref{thm_MainTheorem}.
\end{document} | b6716d9f8624251601da2353029508c8e49a4ae6 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Conclusion}
\label{sec:conclusion}
We have described \sys, a system for proving equivalence of SQL
rewrite rules. In support of \sys, we defined a formal language \sem,
following closely SQL's syntax. Our semantics extends that of SQL
from finite relations to infinite relations, and uses univalent types
from Homotopy Type Theory to represent and prove equality of cardinal
numbers (finite and infinite). We have demonstrated the power and
flexibility of \sys by proving the correctness of several powerful
optimization rules found in the database literature.
\section{Denotation of \sem}
\label{sec:denotation}
In this section we define the denotational semantics of \sem. We
first discuss the translation of \sem constructs into \outputLang.
Then, in~\secref{more-sql},
we describe how advanced features of SQL (such as integrity constraints
and indexes) can be expressed using
\sem and subsequently translated.
Figure~\ref{fig:denote-query} shows the translation of \sem to
\outputLang. The translation rules make use of contexts. A {\em context schema} $\Gamma$ is a schema (see the
definition of $\texttt{Schema}$ in Fig.~\ref{fig:data-model}); a {\em context}
$g$ is a tuple of type $\texttt{Tuple}\; \Gamma$ associated to that schema.
Intuitively, the context consists of the concatenation of all tuple
variables occurring in a surrounding scope.
For example, consider the \sem query with correlated subqueries
in~\figref{contexts} where path expressions are used to refer to
relations in predicates, as discussed in~\secref{data-model}. As in
standard SQL, evaluation proceeds from the outermost to the innermost
query, starting with query $q_1$. After the \texttt{FROM} clause in
$q_1$ is processed, the context consists of $R_1$'s schema
($\sigma_{R_1}$), which is then passed to the query $q_2$. In turn,
$q_2$ then processes its \texttt{FROM} clause, and appends the schema
of $R_2$ to the context ($\node{\sigma_{R_1}}{\sigma_{R_2}}$), and
this context is used to evaluate the path expression $q_2$'s predicate
(\texttt{right.$k$ = left.$k$}, i.e., \texttt{$R_1$.k = $R_2$.k}), and
similarly when $q_3$ evaluates its predicate.
In our system, contexts are implemented as tuples.
To make passing of contexts explicit, in the following,
each \sem construct takes in a context tuple (represented by {\tt Tuple $\Gamma$}),
and is translated to functions that take in
both a tuple ($t$) and a context tuple ($g$).
\begin{figure}
\begin{small}
\[
\begin{array}{ll}
\texttt{SELECT $*$ FROM $R_1$ WHERE} & q_1 \\
\quad {\color{blue}\texttt {-- predicate in $q_2$: $R_2$.b = $R_1$.a }} & \\
\quad \texttt {EXISTS SELECT $*$ FROM $R_2$ WHERE right.$p_2$=left.$p_1$ AND} & q_2\\
\qquad {\color{blue}\texttt {-- predicate in $q_3$: $R_3$.c=$R_2$.b }} & \\
\qquad \texttt{EXISTS SELECT $*$ FROM $R_3$} & \\
\hspace{0.7in} \texttt{WHERE right.$p_3$=left.right.$p_2$} & q_3
\end{array}
\]
\end{small}
\hrule
\vspace{0.1in}
\centering
\begin{tabular}{|c|l|}\hline
Query & Context schema \\ \hline
init & \texttt{$\Gamma_0$=empty} \\ \hline
$q_1$ & \texttt{$\Gamma_1$=node $\Gamma_0$ $\sigma_{R_1}$} \\ \hline
$q_2$ & \texttt{$\Gamma_2$=node $\Gamma_1$ $\sigma_{R_2}$} \\ \hline
$q_3$ & \texttt{$\Gamma_3$=node $\Gamma_2$ $\sigma_{R_3}$} \\ \hline
\end{tabular}
\figlabel{contexts}
\caption{Using Contexts in Evaluating Correlated Subqueries}
\end{figure}
\subsection{Denoting Basic HoTTSQL Constructs}
\begin{figure*}[t]
\centering
\[
\begin{array}{llll}
\multicolumn{3}{l}{ \framebox[1.1\width]{
$\denoteQuery{\Gamma}{q}{\sigma} : \texttt{Tuple} \; \Gamma \rightarrow \texttt{Tuple} \; \sigma \rightarrow \mathcal{U} $} } & \texttt{(* $Query$ *)} \\ \\
\denoteQuery{\Gamma}{table}{\sigma} & \triangleq &
\lambda \; \context \; \tuple. \; \denoteTable{table} \; t \\
\denoteQuery{\Gamma}{\SELECT{p}{q}}{\sigma} & \triangleq &
\multicolumn{2}{l}{
\lambda \; \context \; \tuple. \; \sum_{t':\texttt{Tuple} \; \sigma'}{
(\denoteProj{p}{\node{\Gamma}{\sigma'}}{\sigma} \; \mkPair{g}{t'} = t) \times
\denoteQuery{\Gamma}{q}{\sigma'}} \; g \; t'} \\
\denoteQuery{\Gamma}{\FROM{q_1, q_2}}{\node{\sigma_1}{\sigma_2}} & \triangleq &
\lambda \; \context \; \tuple. \;
\denoteQuery{\Gamma}{q_1}{\sigma_1} \; g \; \fst{t} \times
\denoteQuery{\Gamma}{q_2}{\sigma_2} \; g \; \snd{t} \\
\denoteQuery{\Gamma}{\FROM{q}}{\sigma} & \triangleq &
\lambda \; \context \; \tuple. \; \callQuery{\Gamma}{q}{\sigma} \\
\denoteQuery{\Gamma}{\WHERE{q}{b}}{\sigma} & \triangleq &
\lambda \; \context \; \tuple. \;
\callQuery{\Gamma}{q}{\sigma} \times
\denotePred{\node{\Gamma}{\sigma}}{b} \; \mkPair{g}{t} \\
\denoteQuery{\Gamma}{\UNIONALL{q_1}{q_2}}{\sigma} & \triangleq &
\lambda \; \context \; \tuple. \;
\callQuery{\Gamma}{q_1}{\sigma} +
\callQuery{\Gamma}{q_2}{\sigma} \\
\denoteQuery{\Gamma}{\EXCEPT{q_1}{q_2}}{\sigma} & \triangleq &
\lambda \; \context \; \tuple. \;
\callQuery{\Gamma}{q_1}{\sigma} \times
(\negate{(\callQuery{\Gamma}{q_2}{\sigma})}) \\
\denoteQuery{\Gamma}{\DISTINCT{q}}{\sigma} & \triangleq &
\lambda \; \context \; \tuple. \; \merely{\callQuery{\Gamma}{q}{\sigma}} \\ \\
\multicolumn{3}{l}{ \framebox[1.1\width]{$ \denotePred{\Gamma}{b}
: \texttt{Tuple} \; \Gamma \rightarrow \mathcal{U} $ } } & \texttt{(* $Predicate$ *)}\\
\\
\denotePred{\Gamma}{e_1 = e_2} & \triangleq &
\lambda \; \context. \;(\callExpr{\Gamma}{e_1}{\tau} = \callExpr{\Gamma}{e_2}{\tau}) \\
\denotePred{\Gamma}{\AND{b_1}{b_2}} & \triangleq &
\lambda \; \context. \; \callPred{\Gamma}{b_1} \times \callPred{\Gamma}{b_2} \\
\denotePred{\Gamma}{\OR{b_1}{b_2}} & \triangleq &
\lambda \; \context. \; \merely{\callPred{\Gamma}{b_1} + \callPred{\Gamma}{b_2}} \\
\denotePred{\Gamma}{\NOT{b}} & \triangleq &
\lambda \; \context. \; \negate{(\callPred{\Gamma}{b})} \\
\denotePred{\Gamma}{\EXISTS{q}} & \triangleq &
\lambda \; \context. \; \merely{\sum_{t:\texttt{Tuple} \; \sigma}{\callQuery{\Gamma}{q}{\sigma}}} \\
\denotePred{\Gamma}{\texttt{FALSE}} & \triangleq & \lambda \; \context. \; \textbf{0} \\
\denotePred{\Gamma}{\texttt{TRUE}} & \triangleq & \lambda \; \context. \; \textbf{1} \\
\denotePred{\Gamma}{\CastPred{p}{b}} & \triangleq &
\lambda \; \context. \; \denotePred{\Gamma'}{b} \; (\denoteProj{p}{\Gamma}{\Gamma'} \; g) \\
\\
\multicolumn{3}{l}{ \framebox[1.1\width]{$ \denoteExpr{\Gamma}{e}{\tau} :
\texttt{Tuple} \; \Gamma \rightarrow \denote{\tau} $} } & \texttt{(* $Expression$ *)} \\ \\
\denoteExpr{\Gamma}{\Var \; p}{\tau} & \triangleq &
\lambda \; \context. \; \denoteProj{p}{\Gamma}{\leaf \; \tau} \; g \\
\denoteExpr{\Gamma}{f(e_1, \ldots)}{\tau} & \triangleq &
\lambda \; \context. \; \denote{f}(\callExpr{\Gamma}{e_1}{\tau_1}, \ldots ) \\
\denoteExpr{\Gamma}{agg(q)}{\tau'} & \triangleq &
\lambda \; \context. \; \denote{agg} \; ( \denoteQuery{\Gamma}{q}{\leaf \; \tau} \; g ) \\
\denoteExpr{\Gamma}{\CastExpr{p}{e}}{\tau} & \triangleq &
\lambda \; \context. \; \denoteExpr{\Gamma'}{e}{\tau} \; (\denoteProj{c}{\Gamma}{\Gamma'} \; g) \\ \\
\multicolumn{3}{l}{ \framebox[1.1\width]{$
\denoteProj{p}{\Gamma}{\Gamma'} : \texttt{Tuple} \; \Gamma \rightarrow \texttt{Tuple} \; \Gamma'$ } } & \texttt{(* $Projection$ *)} \\ \\
\denoteProj{*}{\Gamma}{\Gamma} & \triangleq & \lambda \; \context. \; g \\
\denoteProj{\texttt{Left}}{\node{\Gamma_0}{\Gamma_1}}{\Gamma_0} & \triangleq & \lambda \; \context. \; \fst{g} \\
\denoteProj{\texttt{Right}}{\node{\Gamma_0}{\Gamma_1}}{\Gamma_1} & \triangleq & \lambda \; \context. \; \snd{g} \\
\denoteProj{\texttt{Empty}}{\Gamma}{\texttt{empty}} & \triangleq & \lambda \; \context. \; \texttt{unit} \\
\denoteProj{\Compose{p_1}{p_2}}{\Gamma}{\Gamma''} & \triangleq & \lambda \; \context. \;
\denoteProj{p_2}{\Gamma'}{\Gamma''}\;(\denoteProj{p_1}{\Gamma}{\Gamma'} \; g) \\
\denoteProj{\Duplicate{p_1}{p_2}}{\Gamma}{\node{\Gamma_0}{\Gamma_1}} & \triangleq &
\lambda \; \context. \; (\denoteProj{p_1}{\Gamma}{\Gamma_0} \; g, \; \denoteProj{p_2}{\Gamma}{\Gamma_1} \; g) \\
\denoteProj{\Evaluate{e}}{\Gamma}{\leaf \; \tau} & \triangleq & \lambda \; \context. \; \callExpr{\Gamma}{e}{\tau} \\
\end{array}
\]
\caption{Denotational Semantics of \inputLang}
\label{fig:denote-query}
\end{figure*}
\paragraph{Queries} A query $q$ is denoted to a function
from $q$'s context tuple (of type $\texttt{Tuple} \; \Gamma$) to a HoTT-Relation
(of type \( \texttt{Tuple} \; \sigma \rightarrow \mathcal{U} \)):
\[ \denoteQuery{\Gamma}{q}{\sigma}: Tuple \; \Gamma \rightarrow Tuple \; \sigma \rightarrow \mathcal{U} \]
The \texttt{FROM} clause is recursively denoted to a series of cross products of
HoTT-Relations. Each cross product is denoted using $\times$ as shown in
Section~\ref{sec:target-lang}. For example:
\[
\begin{array}{ll}
\denoteQuery{\Gamma}{\FROM{q_1 , \; q_2}}{\sigma} & \triangleq \\
\lambda \; \context \; \tuple. \; (\denoteQuery{\Gamma}{q_1}{\sigma} \; g \; \fst{t}) \times
(\denoteQuery{\Gamma}{q_2}{\sigma} \; g \; \snd{t})
\end{array}
\]
\noindent where $\fst{t}$ and $\snd{t}$ indexes into the context tuple $\Gamma$
to retrieve the schemas of $q_1$ and $q_2$ respectively.
Note the manipulation of the context tuple in the denotation of
\texttt{WHERE}: for each tuple $t$, we first evaluate $t$ against the
query before \texttt{WHERE}, using the context tuple $g$. After
that, we evaluate the predicate $b$ by first constructing a new
context tuple as discussed (namely, by concatenating $\Gamma$ and
$\sigma$, the schema of $q$), passing it the combined tuple
$(g,t)$. The combination is needed as $t$ has schema $\sigma$
while the predicate $b$ is evaluated under the schema
$\node{\Gamma}{\sigma}$, and the combination is easily accomplished
as $g$, the context tuple, has schema $\Gamma$.
\texttt{UNION ALL}, \texttt{EXCEPT}, and \texttt{DISTINCT} are denoted using +,
negation ($\negate{n}$) and merely ($\merely{n}$) on univalent types as shown in
Section~\ref{sec:target-lang}.
\paragraph{Predicates} A predicate $b$ is denoted to a function
from a tuple (of type $\texttt{Tuple} \; \Gamma$) to a univalent type (of type
$\mathcal{U}$):
\[ \denotePred{\Gamma}{b} : \texttt{Tuple} \; \Gamma \rightarrow \mathcal{U} \]
\noindent More specifically, the return type $\mathcal{U}$
must be a \emph{squash type}~\cite[Ch. 3.3]{hottBook}.
A squash type can only be a type of 1 element,
namely \textbf{1}, and a type of 0 element, namely \textbf{0}. \inputLang
program with the form \texttt{$q$ WHERE $b$} is denoted to the
Cartesian product between a univalent type and a mere proposition.
As an example, suppose a particular tuple $t$ has multiplicity 3 in
query $q$, i.e., $q\; t = \denote{R}{t} = {\bf 3}$, where ${\bf 3}$ is a univalent
type. Since predicates are denoted to propositions, applying the tuple to
the predicate returns either ${\bf 1}$ or ${\bf 0}$, and the overall result
of the query for tuple $t$ is then either ${\bf 3} \times {\bf 0} = {\bf 0}$,
or ${\bf 3} \times {\bf 1} = {\bf 1}$, i.e., a squash type.
\paragraph{Expressions and Projections} A value expression $e$ is
denoted to a function from a tuple (of type $\texttt{Tuple} \; \Gamma$) to its data
type, such as {\tt int} and {\tt bool} \ ($\denote{\tau}$):
\[ \denoteExpr{\Gamma}{e}{\tau} : \texttt{Tuple} \; \Gamma \rightarrow \denote{\tau} \]
\noindent A projection $p$ from $\Gamma$ to $\Gamma'$ is denoted to a function from a tuple of type $\texttt{Tuple} \; \Gamma$ to a tuple of type $\texttt{Tuple} \; \Gamma'$.
\[\denoteProj{p}{\Gamma}{\Gamma'} : \texttt{Tuple} \; \Gamma \rightarrow \texttt{Tuple} \; \Gamma'\]
Projections are recursively defined.
A projection can be composed by two projections using ``{\tt .}''. The composition of two projection, ``$\Compose{p_1}{p_2}$'', where $p_1$ is a projection from $\Gamma$ to $\Gamma'$ and $p_2$ is a projection from $\Gamma'$ to $\Gamma''$, is denoted to a function from a tuple of type $\texttt{Tuple} \; \Gamma$ to a tuple of type $\texttt{Tuple} \; \Gamma''$ as follows:
\[\lambda \; \context. \; \denoteProj{p_2}{\Gamma'}{\Gamma''}\;(\denoteProj{p_1}{\Gamma}{\Gamma'} \; g) \]
We apply the denotation of $p_1$, which is a function of type $\texttt{Tuple}\;\Gamma \rightarrow \texttt{Tuple} \; \Gamma'$, to the argument of composed projection $g$, then apply the denotation of $p_2$ to the result of application.
A projection can also be combined by two projections using ``{\tt ,}''. The combining of two projection, $\Duplicate{p_1}{p_2}$, is denoted to:
\[ \lambda \; \context. \; (\denoteProj{p_1}{\Gamma}{\Gamma_0} \; g, \; \denoteProj{p_2}{\Gamma}{\Gamma_1} \; g) \]
%
where we apply the denotation of $p_1$ and the denotation of $p_2$ to the argument of combined projection ($g$) separately, and combine their results using the constructor of a pair.
\subsection{Denoting Derived HoTTSQL Constructs}
\label{sec:more-sql}
\sem supports additional SQL features including group by, integrity constraints, and index.
All such features are commonly utilized in query optimization.
\paragraph{Grouping} Grouping is a widely-used relational operator that projects rows
with a common value into separate groups, and applies an aggregation function
(e.g., average) to each group.
In SQL, this is supported via the \texttt{GROUP BY} operator that takes in the attribute
names to form groups. \sem supports grouping by de-sugaring \texttt{GROUP BY} using
correlated subqueries that returns a single attribute relation,
and applying aggregation function to the resulting relation~\cite{BunemanLSTW94SIGMOD}.
Below is an example of such rewrite expressed using SQL:
\[
\begin{array}{ll}
\texttt{SELECT $k$, SUM($g$) FROM $R$ GROUP BY $k$} \vspace{0.05in} \\
\vspace{0.05in}\hspace{0.8in} \text{rewrites to} \Downarrow \\
\texttt{SELECT DISTINCT $k$, SUM( SELECT $g$ FROM $R$} \\
\qquad \qquad \qquad \qquad \qquad \quad \;\; \texttt{ WHERE $R.k = R_1.k$)} \\
\texttt{FROM $R$ AS $R_1$}
\end{array}
\]
\noindent We will illustrate using grouping in rewrite rules in~\secref{agg-rule}.
\paragraph{Integrity Constraints}
Integrity constraints are used in real-world database systems and
facilitate various semantics-based query optimizations~\cite{DarFJST96}.
\sem supports two
important integrity constraints: keys and functional dependency, through
syntactic rewrite.
A {\em key constraint} requires an attribute to have unique values among all
tuples in a relation. In \sem, a projection $k$ is a key to the relation $R$ if
the following holds:
\[
\begin{array}{l}
\texttt{key ($k$) ($R$)} := \\
\qquad \denoteQuery{\texttt{empty}}{\texttt{SELECT * FROM $R$}}{\sigma} = \\
\qquad \llbracket \texttt{empty} \vdash \texttt{SELECT Left.* FROM $R$, $R$} \\
\qquad \texttt{ WHERE (\Var{Right.Left.$k$}) = (\Var{Right.Right.$k$)}}: \sigma \rrbracket
\end{array}
\]
To see why this definition satisfies the key constraint, note that $k$
is a key in $R$ if and only if $R$ equals to its self-join
on $k$ after converting the result into a set using \texttt{DISTINCT}.
Intuitively, if $k$ is a key, then self-join of $R$ on $k$ will keep all the tuples of
$R$ with each tuple's multiplicity unchanged. Conversely, if $R$ satisfies the
self-join criteria, then attribute $k$ holds unique values in $R$ and is hence a key.
\paragraph{Functional Dependencies}
Keys are used in defining functional dependencies and indexes.
A \emph{functional dependency} constraint from attribute $a$ to $b$ requires that for any
two tuples $t_1$ and $t_2$ in $R$, $(t_1.a = t_2.a) \rightarrow (t_1.b = t_2.b)$
In \sem, two projections $a$ and $b$ forms a functional dependency in relation $R$ if
the following holds:
\[
\begin{array}{l}
\texttt{fd ($a\; b$) ($R$)} := \\
\qquad \texttt{key Left.* } \llbracket \texttt{empty} \vdash \texttt{DISTINCT SELECT a, b} \\
\qquad \qquad \qquad \qquad \quad \texttt{FROM R}:\node{(\leaf{\tau_a})}{(\leaf{\tau_b})} \rrbracket
\end{array}
\]
If $a$ and $b$ forms a functional dependency, then $a$ should be a key in
the relation the results from projecting $a$ and $b$ from $R$ followed by
de-duplication. The converse argument follows similarly.
\paragraph{Index}
An index on an attribute $a$ is a data structure
that speeds up the retrieval of tuples with a given value of $a$~\cite[Ch. 8]{dbSysBook}.
To reason about rewrite rules that use indexes, we follow the idea that index can
be treated as a logical relation rather than physical data structure from
Tsatalos et al~\cite{TsatalosSI94}. Since defining index as an relation
requires a unique identifier of each tuple (analogous to a pointer to each
tuple in the physical implementation of an index in database systems), we
define index as a \sem query that projects on the a key of the relation and the
index attribute. For example, if $k$ is a key of relation $R$, an index $I$ of
$R$ on attribute $a$ can be defined as:
\[
\texttt{index}(a, R) := \texttt{SELECT $k$, $a$ FROM $R$}
\]
\noindent In Section~\ref{sec:index-rule}, we show example rewrite rules that utilize
indexes that are commonly used in query optimizers.
\section{Discussion}
\label{sec:discussion}
{\bf Limitations} Our system does currently not support three SQL
features: NULL's with their associated three-valued-logic, outer
joins, and windows functions. However, all can be expressed in \sem,
at the cost of some added complexity, as we explain now.
When any argument to an expression is NULL, then the expression's
output is NULL; this feature can easily be supported by modifying the
external operators. When an argument of a comparison predicate is
NULL, then the resulting predicate has value \texttt{unknown}, and SQL
uses three valued logic to compute predicates: it defines
$0 = \texttt{false}, 1/2 = \texttt{unknown}, 1=\texttt{true}$, and the
logical operators $x \texttt{ and } y = \min(x,y)$,
$x \texttt{ or } y = \max(x,y)$, $\texttt{not}(x) = 1-x$; a
select-from-where query returns all tuples for which the
where-predicate evaluates to \texttt{true} (i.e. not \texttt{false} or
\texttt{unknown}). As a consequence, the law of excluded middle
fails, for example the query:
$$\SelectFromWhere{\texttt{*}}{R}{a=5 \texttt{ or } a \neq 5}$$
is not equivalent to $\SelectFrom{\texttt{*}}{R}$. This, too, could be
currently expressed \sem\ by encoding the predicates as external
functions that implement the 3-valued logic. However, by doing so one
hides from the rewrite rules the equality predicate, which plays a key
role in joins. In future versions, we plan to offer native support
for NULL's, to simplify the task of proving rewrite rules over
relations with NULLs.
Both outer joins and windows functions are directly expressible in
\sem. For example, a left outer join of two relations $R(a,b)$,
$S(b,c)$ can be expressed by first joining $R$ and $S$ on $b$, and
union-ing the result with
\[
\begin{array}{ll}
\SelectFrom{R.*,\texttt{NULL}}{S} \texttt{ EXCEPT } \\
\SelectFromWhere{R.*,\texttt{NULL}}{S}{R.b=S.b}
\end{array}
\]
A direct
implementation in \sem\ would basically have to follow the same
definition of left outer joins.
{\bf Finite v.s. Infinite Semantics} Recall that our semantics extends
the standard bag semantics of SQL in two ways: we allow a relation to
have infinitely many distinct elements, and we allow each element to
have an infinite multiplicity. To the best of our knowledge, our
system is the first that interprets SQL over infinite relations. This
has two consequences. First, our system cannot check the equivalence
of two SQL expressions that return the same results on all finite
relations, but differ on some infinite relations. It is well-known
that there exists First Order sentences, called {\em infinity axioms},
that do not admit any finite model, but admit infinite models. For
example \cite[pp.307]{DBLP:books/sp/BorgerGG1997} the sentence
$\varphi \equiv \forall x \exists y \forall z(\neg R(x,x) \wedge
R(x,y) \wedge (R(y,z) \rightarrow R(x,z)))$ is an infinity axiom. It
is possible to write a SQL query that checks $\varphi$, then returns
the empty set if $\varphi$ is false, or returns a set consisting of a
single value (say, 1) if $\varphi$ is true: call this query $Q_1$.
Call $Q_2$ the query \SelectFromWhere{\texttt{DISTINCT} 1}{R}{2=3}.
Then $Q_1=Q_2$ over all finite relations, but $Q_1 \neq Q_2$ not over
infinite relations. Thus, one possible disadvantage of our semantics
is that we cannot prove equivalence of queries that encode infinity
axioms. However, none of the optimization rules that we found in the
literature, and discussed in this paper, encode an infinity axiom.
Hence we argue that, for practical purposes, extending the semantics
to infinite relations is a small price to pay for the added simplicity
of the equivalence proofs. Second, by generalizing SQL queries to
both finite and infinite relations we make our system theoretically
complete: if two queries are equivalent then, by G\"odel's
completeness theorem, there exists a proof of their equivalence.
Finding the proof is undecidable (it is recursively enumerable, r.e.):
our system does not search for the proof, instead the user has to find
it, and our system will verify it. Contrast this with a system whose
semantics is based on finite relations: such a system cannot have a
complete proof system for SQL query equivalence. Indeed, if such a
complete proof system existed, then SQL query equivalence would be
r.e. (since we can enumerate all proofs and search for a proof of
$Q_1=Q_2$), and therefore equivalence would be decidable (since it is
also co-r.e., because we can enumerate all finite relations, searching
for an input s.t. $Q_1 \neq Q_2$). However, by Trakthenbrot's, query
equivalence is undecidable. Recall that Trakthenbrot's
theorem~\cite{Trakhtenbrot50DANUSSR,DBLP:books/sp/Libkin04} states
that the problem {\em given an FO sentence $\varphi$, check if
$\varphi$ has a finite model} is undecidable. We can reduce this
problem to query equivalence by defining $Q_1$ to be a query that
checks checks $\varphi$ and returns the empty set if $\varphi$ is
false, or returns some non-empty set if $\varphi$ is true, and
defining $Q_2$ to be the query that always returns the empty set (as
above), then checking $Q_1 \equiv Q_2$). Thus, by extending our
semantics to infinite relations we guarantee that, whenever two
queries are equivalent, there exists a proof of their equivalence.
\section{Introduction}
\seclabel{intro}
From purchasing plane tickets to browsing social networking websites,
we interact with database systems on a daily basis.
Every database system consists of a query optimizer that
takes in an input query and
determines the best program, also called a query plan, to execute
in order to retrieve the
desired data. Query optimizers typically consists of two components:
a query plan enumerator that generates query plans that are semantically
equivalent to the input query, and a plan selector that chooses
the optimal plan from the enumerated ones to execute
based on a cost model.
The key idea behind plan enumeration is to apply {\em rewrite rules}
that transform a given query plan into another one, hopefully one with
a lower cost. While numerous plan rewrite rules have been proposed
and implemented, unfortunately designing such rules remains a highly
challenging task. For one, rewrite rules need to be {\em semantically
preserving}, i.e., if a rule transforms query plan $Q$ into $Q'$,
then the results (i.e., the relation) returned from executing $Q$ must
be the same as those returned from $Q'$, and this has to hold for {\em
all} possible input database schemas and instances. Obviously,
establishing such proof for any non-trivial query rewrite rule is not
an easy task.
Coupled with that, the rich language constructs and subtle semantics of
SQL, the de facto programming language used to interact with relational
database systems,
only makes the task even more difficult.
As a result, while various rewrite rules
have been proposed and studied extensively
in the data management research community~\cite{MumickFPR90SIGMOD, Muralikrishna92VLDB, LevyMS94VLDB, SeshadriHPLRSSS96SIGMOD},
to the best of our knowledge only the trivial ones have been
formally proven to be semantically preserving. This has unfortunately
led to dire consequences as incorrect query results have been returned
from widely-used database systems due to unsound rewrite rules, and
such bugs can often go undetected for extended periods of time
\cite{GanskiW87SIGMOD, MySQLBug, PostgresBug}.
In this paper we describe a system to formally verify the equivalence
of two SQL expressions. We demonstrate its utility by proving correct
a large number of query rewrite rules that have been described in the
literature and are currently used in popular database systems. We also show
that, given counter examples, common mistakes made in query optimization fail to pass our
formal verification, as they should. Our system shares similar
high-level goals of building formally verified systems using theorem
provers and proof assistants as recent work has demonstrated
\cite{Leroy09JACM, SEL4, FSCQ}.
The biggest challenge in designing a formal verification system for
equivalence of SQL queries is choosing the right SQL semantics. Among
the various features of SQL, the language uses both set and bag
semantics and switches freely between them, making semantics
definition of the language a difficult task. Although SQL is an ANSI
standard~\cite{sql2011}, the ``formal'' semantics defined there is of
little use for formal verification: it is loosely described in English
and has resulted in conflicting interpretations \cite{Date89AW}.
Researchers have defined two quite different rigorous semantics
of SQL. The first comes from the formal methods
community~\cite{MalechaMSW10POPL, VeanesGHT09ICFEM, VeanesTH10LAPR16},
where SQL relations are interpreted as lists, and SQL queries as
functions over lists; two queries are equivalent if they return lists
that are equal up to element permutation (under bag semantics) or up
to element permutation and duplicate elimination (under set
semantics). The problem with this semantics is that even the simplest
equivalences require lengthy proofs in order to reason about
order-independence or duplicate elimination, and these proofs become
huge when applied to rewrites found in real-world optimizations.
The second semantics comes from the database theory community and uses
only set semantics \cite{TheAliceBook, BunemanLSTW94SIGMOD,
NegriPS91TODS}. This line of work has led to theoretical results
characterizing languages for which query equivalence is decidable (and
often fully characterizing the complexity of the equivalence problem),
and separating them from richer languages where equivalence is
undecidable~\cite{ChandraM77STOC,SagivY80JACM,DBLP:conf/icdt/Ullman97,DBLP:conf/icdt/GeckKNS16}.
For example, equivalence is decidable (and $\Pi^P_2$-complete for a
fixed database schema~\cite{DBLP:conf/icdt/Ullman97}, and
coNEXPTIME-complete in general~\cite{DBLP:conf/icdt/GeckKNS16}) for
conjunctive queries with safe negation, but undecidable for
conjunctive queries with unsafe negation. Unfortunately, this
approach is of limited use in practice, because most query
optimization rules use features that places them in the undecidable
language fragments.
This paper contributes a new semantics for SQL that is both simple
and allows simple proofs of query equivalence. We then demonstrates its
effectiveness by proving the correctness of various powerful query
optimization rules described in the literature.
Our semantics consists of two non-trivial generalizations of
$K$-relations. $K$-relations were introduced by Green et al. in the
database theory community~\cite{GreenKT07PODS}, and represent a
relation as mathematical function that takes as input a tuple and
returns its multiplicity in the relation, with 0 meaning that the
tuple does not exist in the relation. A $K$-relation is required to
have finite support, meaning that only a finite set of tuples have
multiplicity $>0$. $K$-relations greatly simplify reasoning about
SQL: under set semantics, a relation is simply a function that returns
0 or 1 (i.e., a Boolean value), while under bag semantics it returns a
natural number (i.e., a tuple's multiplicity).
Database operations such as join or union become
arithmetic operations on the corresponding multiplicities: join
becomes multiplication, union becomes addition. Determining the
equivalence of a rewrite rule that transforms a query $Q$ into another
query $Q'$ reduces to checking the equivalence of the functions they
denote. For example, proving that the join operation is associative
reduces to proving that multiplication is associative. As we will
show, reasoning about functions over cardinals is much easier than
writing inductive proofs on data structures such as lists.
However, $K$-relations as defined by~\cite{GreenKT07PODS} are
difficult to use in proof assistants, because one needs to prove for
every SQL expression under consideration that the $K$-relation it
returns has finite support: this is easy with pen-and-paper, but very
hard to encode for a proof assistant. Without a guarantee of finite
support, some operations are undefined, for example projection on an
attribute requires infinite summation. Our first generalization of
$K$-relations is to drop the finite support requirement, and meanwhile
allow the multiplicity of a tuple to be any cardinal number as opposed
to a (finite) natural number. Then the possibly infinite sum
corresponding to a projection is well defined. With this change, SQL
queries are interpreted over finite and infinite bags, where some
tuples may have infinite multiplicities. To the best of our
knowledge, ours is the first SQL semantics that interprets relations
as both finite and infinite; we discuss some implications in
Sec.~\ref{sec:discussion}.
Our second generalization of $K$-relations is to replace cardinal
numbers with univalent types. Homotopy Type Theory
(HoTT)~\cite{hottBook} has been introduced recently as a
generalization of classical type theory by adding membership and
equality proofs. A {\em univalent type} is a cardinal number (finite
of infinite) together with the ability to prove equality.
To summarize, we define a SQL semantics where a relation is
interpreted as a function mapping each tuple to a univalent type,
whose cardinality represents the multiplicity of the tuple in the
relation, and a SQL query is interpreted as a function from input
relations to an output relation. We call the SQL language with this
particular semantics \sem. Our language covers all major features of
SQL. In addition, since univalent types have been integrated into the
Coq proof assistant, we leverage that implementation to prove
equivalences of SQL expressions.
To demonstrate the effectiveness of \sem, we implemented a new system
called \sys (Database OPtimizations CERTified) for proving equivalence
of SQL rewrite rules. We have used \sys to prove many well-known
and commonly-used rewrite rules from the data management research
literature, many of which have never been formally proven correct
before:
aggregates~\cite{ChaudhuriD97SIGMOD,DBLP:conf/pods/KhamisNR16}, magic
sets rewriting~\cite{BancilhonMSU86PODS}, query rewriting using
indexes~\cite{TsatalosSI94}, and equivalence of conjunctive
queries~\cite{TheAliceBook}. All our proofs require at most a few
dozens lines of Coq code using \sys, as shown in Fig.~\ref{fig:rules}. All
definitions and proofs presented in this paper are open-source and
available online.\footnote{\url{http://dopcert.cs.washington.edu}}
In summary, this paper makes the following contributions:
\begin{itemize}
\item We present \sem, a (large fragment) of SQL whose semantics
generalizes $K$-relations to infinite relations and univalent types.
The goal of this semantics is to enable easy proofs for the
equivalence of query rewrite rules. (Sec.~\ref{sec:semantics}.)
\item We prove a wide variety of well-known and widely-used SQL
rewrite rules, where many of them have not be formally proven
before; each proof require at most a few dozens lines of Coq code
using \sys. (Sec.~\ref{sec:denotation}.)
\item We implement \sys, a new system written in Coq for checking the
equivalence of SQL rewrite rules. \sys comes with a number of
heuristic tactics for deciding the equivalence of arbitrary rewrite
rules, and a fully automated procedure for deciding rewrite rules
involving conjunctive queries, where conjunctive queries represent a
fragment of SQL where equivalence is
decidable. (Sec.~\ref{sec:rules}.)
\end{itemize}
The rest of this paper is organized as follows. In~\secref{overview},
we given an overview and motivation for a new semantics for SQL. We
then introduce \sem in~\secref{semantics}, its semantics in
\secref{denotation}, and demonstrate our results in~\secref{rules}.
Related work is presented in \secref{related}. We include some
discussion in \secref{discussion} and conclude in \secref{conclusion}.
\section{Overview}
\label{sec:overview}
\begin{figure}
\begin{small}
\textbf{Rewrite Rule}:
\begin{flalign*}
\quad\; & \SelectFromWhere{\texttt{*}}{(\UNIONALL{R}{S})}{b} \quad \equiv & \\
& \UNIONALL{(\SelectFromWhere{\texttt{*}}{R}{b})}{(\SelectFromWhere{\texttt{*}}{S}{b})} &
\end{flalign*}
\textbf{\sem Denotation}:
\begin{flalign*}
\Rightarrow & \lambdaFn{t}{(\denote{R} \; t + \denote{S} \; t) \times \denote{b} \; t}
\equiv
\lambdaFn{t}{\denote{R} \; t \times \denote{b} \; t + \denote{S} \; t \times \denote{b} \; t} &
\end{flalign*}
\textbf{\sem Proof}:
Apply distributivity of $\times$ over $+$.
\end{small}
\caption{Proving a rewrite rule using \sem.
Recall that \texttt{UNION ALL} means bag-union in SQL, which in \sem
is translated to addition of tuple multiplicities in the two input relations.
}
\label{fig:union-slct}
\end{figure}
\paragraph{SQL} The basic datatype in SQL is a {\em relation}, which has a
{\em schema} (a relation name $R$ plus attribute names $\sigma$), and
an {\em instance} (a bag of tuples). A SQL query maps one or more
input relations to a (nameless) output relation. For example, if a
relation with schema $R(a,b)$ has instance
$\set{(1,40),(2,40),(2,50)}$ then the SQL query
\begin{flalign*}
\texttt{Q1:} & \SelectFrom{a}{R}
\end{flalign*}
returns the bag $\set{1,2,2}$.
SQL freely mixes set and bag semantics, where a set is simply a bag
without duplicates and uses the \texttt{distinct} keyword to remove
duplicates. For example, the query:
\begin{flalign*}
\texttt{Q2:} & \SelectFrom{\texttt{DISTINCT } a}{R}
\end{flalign*}
returns the set $\set{1,2}$.
\paragraph{List Semantics} Previous approaches to mechanizing formal proofs
of SQL query equivalences represent bags as
list~\cite{MalechaMSW10POPL, VeanesGHT09ICFEM, VeanesTH10LAPR16}.
Every SQL query admits a natural interpretation over lists, using a
recursive definition~\cite{DBLP:journals/tcs/BunemanNTW95}. To prove
that two queries are equivalent, one uses their inductive definition
on lists, and proves that the two results are equal up to element
reordering and duplicate elimination (for set semantics).
The main challenges in this approach are coming up with the induction
hypothesis, and dealing with list equivalence under permutation and
duplicate elimination.
Inductive proofs quickly grow in complexity, even for simple query
equivalences. Consider the following query:
\begin{flalign*}
\texttt{Q3:} & \SelectFromWhere{\texttt{DISTINCT } x.a}{R \; \texttt{AS} \; x, R \; \texttt{AS} \; y}{x.a=y.a}
\end{flalign*}
\texttt{Q3} is equivalent to \texttt{Q2}, because it performs a
redundant self-join: the inductive proof of their equivalence is quite
complex, and has, to the best of our knowledge, not been done formally
before. A much simpler rewrite rule, the commutativity of selection,
requires 65 lines of Coq proof in prior work~\cite{MalechaMSW10POPL}, and only 10
lines of Coq proof in our semantics.
Powerful database query optimizations, such as magic sets rewrites
and conjunctive query equivalences, are based on generalizations of
redundant self-joins elimination like $\texttt{Q2}\equiv\texttt{Q3}$,
but significantly more complex (see Sec.~\ref{sec:rules}), and
inductive proofs become impractical.
This motivated us to consider a different semantics; we do not use
list semantics in this paper.
\paragraph{$K$-Relation SQL Semantics} An alternative approach introduced
in~\cite{GreenKT07PODS} is to represent relations as functions that
map every tuple to a value that indicates how many times it appears
in the relation. If the relation is a bag, then the function
returns a natural number, and if it is a set then it returns a value
in $\set{0,1}$.
More generally, a commutative semi-ring is a structure
${\bf K} = (K, +, \times, 0, 1)$ where both $(K,+,0)$ and
$(K,\times,1)$ are commutative monoids, and $\times$ distributes
over $+$. For a fixed set of attributes $\sigma$, denote
$\texttt{Tuple}(\sigma)$ the type of tuples with attributes $\sigma$. A
$K$-relation~\cite{GreenKT07PODS} is a function:
\begin{align*}
\denote{R}: & \texttt{Tuple} \; \sigma \rightarrow K
\end{align*}
with finite support, meaning that the set
$\setof{t}{\denote{R}\; t \neq 0}$ is finite. A bag is an
$\mathds{N}$-relation, and a set is a $\mathds{B}$-relation. All relational
operators are expressed in terms of the semi-ring operations, for
example:
\begin{align*}
& \denote{\UNIONALL{R}{S}} = \lambdaFn{t}{\denote{R} \; t+\denote{S} \; t} \\
& \denote{\SelectFrom{\texttt{*}}{R,S}} = \lambdaFn{(t_1,t_2)}{\denote{R} \; t_1 \times \denote{S} \; t_2} \\
& \denote{\SelectFromWhere{\texttt{*}}{R}{b}} = \lambdaFn{t}{\denote{R} \; t \times \denote{b} \; t}\\
& \denote{\SelectFrom{x.a}{R}} = \lambdaFn{t}{\sum_{t' \in \texttt{Tuple} \; \sigma}(t = \D{a} \; t') \times \denote{R} \; t'}\\
& \denote{\SelectFrom{\DISTINCT \texttt{*}}{R}} = \lambdaFn{t}{\merely{\denote{R} \; t}} \\
\end{align*}
where, for any predicate $b$: $\denote{b} \; t = 1$ if the
predicate holds on $t$, and $\denote{b} \; t = 0$ otherwise. The
function $\merely{\ }$ is defined as $\merely{x}=0$ when $x=0$, and
$\merely{x}= 1$ otherwise (see Subsec.~\ref{sec:target-lang}). The
projection $\D{a} \; t'$ returns the attribute $a$ of the tuple $t'$,
while equality $(x=y)$ is interpreted as 0 when $x\neq y$ and 1
otherwise.
To prove that two SQL queries are equal one has to prove that two
semi-ring expressions are equal. For example,
Fig.~\ref{fig:union-slct} shows how we can prove that selections
distribute over unions, by reducing it to the distributivity of
$\times$ over $+$, while Fig.~\ref{fig:magic-distinct} shows the proof
of the equivalence for $\texttt{Q2} \equiv \texttt{Q3}$.
\begin{figure}
\begin{small}
\textbf{Rewrite Rule}:
\begin{flalign*}
\quad\;
& \SelectFromWhere{\texttt{DISTINCT } x.a}{R \; \texttt{AS} \; x, R \; \texttt{AS} \; y}{x.a=y.a} \quad \equiv &\\
& \SelectFrom{\texttt{DISTINCT } a}{R}&
\end{flalign*}
\textbf{Equational \sem Proof}:
\begin{flalign*}
\Rightarrow & \lambdaFn{t}{\merely{\sum_{t_1, t_2} (t = \D{a}\; t_1) \times (\D{a} \; t_1= \D{a} \; t_2) \times \denote{R} \; t_1 \times \denote{R} \; t_2}} \equiv& \\
& \lambdaFn{t}{\merely{\sum_{t_1, t_2} (t = \D{a} \; t_1) \times (t = \D{a} \; t_2) \times \denote{R} \; t_1 \times \denote{R} \; t_2}} \equiv& \\
& \lambdaFn{t}{\merely{(\sum_{t_1} (t = \D{a} \; t_1) \times \denote{R} \; t_1) \times (\sum_{t_2} (t = \D{a} \; t_2) \times \denote{R} \; t_2)}} \equiv& \\
& \lambdaFn{t}{\merely{\sum_{t_1} (t = \D{a} \;t_1) \times \denote{R} \; t_1}}&
\end{flalign*}
We used the following semi-ring identities:
\begin{flalign*}
(a=b) \times (b=c) \equiv & (a=b) \times (a=c)\\
\sum_{t_1, t_2} E_1(t_1) \times E_2(t_2) \equiv & \sum_{t_1} E_1(t_1) \times \sum_{t_2} E_2(t_2)\\
\merely{n \times n} \equiv & \merely{n}
\end{flalign*}
\textbf{Deductive \sem Proof}:
\begin{flalign*}
\Rightarrow \forall \; t. & \exists_{t_0} (\denote{a} \; t_0 = t) \land \denote{R} \; t_0 \leftrightarrow & \\
& \exists_{t_1, t_2} (\denote{a} \; t_1 = t) \land \denote{R} \; t_1 \land \denote{R} \; t_2 \land (\denote{a} \; t_1 = \denote{a} \; t_2) &
\end{flalign*}
Then case split on $\leftrightarrow$.
Case $\rightarrow$: instantiate both $t_1$ and $t_2$ with $t_0$, then apply hypotheses.
Case $\leftarrow$: instantiate $t_0$ with $t_1$, then apply hypotheses.
\end{small}
\caption{The proof of equivalence $\texttt{Q2} \equiv \texttt{Q3}$.}
\label{fig:magic-distinct}
\end{figure}
Notice that the definition of projection requires that the relation
has finite support; otherwise, the summation is over an infinite set
and is undefined in $\mathds{N}$. This creates a major problem for our
equivalence proofs, since we need to prove, for each intermediate
result of a SQL query, that it returns a relation with finite support.
This adds significant complexity to the otherwise simple proofs of
equivalence.
\paragraph{\sem Semantics} To handle this challenge, our semantics
generalizes $K$-Relation in two ways: we no longer require relations to
have finite support, and we allow the multiplicity of a tuple to be an
arbitrary cardinality (possibly infinite). More precisely, in our
semantics a relation is interpreted as a function:
\begin{align*}
& \texttt{Tuple} \; \sigma \rightarrow \mathcal{U}
\end{align*}
where $\mathcal{U}$ is the class of homotopy types. We call such a
relation a \emph{HoTT-relation}. A homotopy type $n \in \mathcal{U}$ is an
ordinary type with the ability to prove membership and equality between
types.
Homotopy types form a commutative semi-ring and can well represent cardinals.
Cardinal number 0 is the empty homotopy type \textbf{0}, 1 is the unit type \textbf{1}, multiplication is the product type $\times$, addition is the sum type $+$, infinite summation is the dependent product type $\Sigma$, and truncation is the squash type $\merely{.}$.
Homotopy types generalize natural numbers and their semiring operations,
and is now well integrated with automated proof assistants like Coq \footnote{After adding the Univalence Axiom to Coq's underlying type theory.}. As we show in the rest of this paper, the equivalence proofs retain the simplicity of $\mathds{N}$-relations and can be
easily mechanized, but without the need to prove finite support.
In addition, homotopy type theory unifies squash type and proposition. Using the fact that propositions as types in homotopy type theory \cite[Ch 1.11]{hottBook}, in order to prove the equivalence of two squash types, $\merely{p}$ and $\merely{q}$, it is sufficient to just prove the bi-implication ($p \leftrightarrow q$), which is arguably easier in Coq. For example, transforming the equivalence proof of Figure~\ref{fig:magic-distinct} to bi-implication would not require a series of equational rewriting using semi-ring identities any more, which is complicated because it is under the variable bindings of $\Sigma$. The bi-implication can be proved in Coq by deduction easily.
The queries of the above rewrite rule fall in the well studied category of conjunctive queries,
for which equality is decidable (equality between arbitrary SQL queries is undecidable).
Using Coq's support for automating deductive reasoning (with \emph{Ltac}), we
have implemented a decision procedure for the equality of conjunctive
queries, the aforementioned rewrite rule can thus be proven in one line of Coq code.
\section{Appendix Title}
\bibliographystyle{abbrvnat}
\section{Related Work}
\label{sec:related}
\subsection{Query Rewriting}
Query rewriting based on equivalence rules is an essential part of modern query optimizers. Rewrite rules are either fired by a forward chaining rule engine in a Starburst optimizer framework \cite{HaasFLP89SIGMOD,PiraheshHH92SIGMOD}, or are used universally to represent the search space in Exodus \cite{GraefeD87SIGMOD} and its successors, including Volcano \cite{GraefeM93ICDE} and Cascades \cite{Graefe95aDEB}.
Using \sys, we formally prove a series of rewrite rules from the database literature. Those rules include basic algebraic rewrites such as selection push down \cite{Ullman89}, rewrite rules using indexes \cite{dbSysBook}, and unnesting aggregates with joins \cite{Muralikrishna92VLDB}. We are able to prove one of the most complicated rewrite rules that is also widely used in practice: magic set rewrites \cite{MumickFPR90SIGMOD, BancilhonMSU86PODS, SeshadriHPLRSSS96SIGMOD}. Magic set rewrites involve many SQL features such as correlated subqueries, aggregation and group by. To our best knowledge, its correctness has not been formally proven before.
\sys automates proving rewrite rules on decidable fragments of SQL. According to Codd's theorem \cite{Codd72}, relational algebra and relational calculus (formulas of first-order logic on database instances) are equivalent in expressive power. Thus, the equivalence between two SQL queries is in general undecidable \cite{Trakhtenbrot50DANUSSR}. Extensive research has been done to study the complexity of containment and equivalence of fragments of SQL queries under bag semantics and set semantics \cite{ChandraM77STOC, ChaudhuriV92PODS, IoannidisR95TODS, DBLP:conf/icdt/GeckKNS16, JayramKV06PODS, SagivY80JACM, Meyden92PODS}. We list the results in Figure~\ref{fig:complexity}.
\subsection{SQL Semantics}
SQL is the de-facto language for relational database systems. Although the SQL language is an ANSI/ISO standard \cite{sql2011}, it is loosely described in English and leads to conflicting interpretations \cite{Date89AW}. Previous related formalizations of various fragments of SQL include relational algebra \cite{TheAliceBook}, comprehension syntax \cite{BunemanLSTW94SIGMOD}, and recursive and non-recursive Datalog \cite{ChaudhuriV92PODS}. These formalisms are not suited for rigorous reasoning about the correctness of real world rewrite rules since they mostly focus exclusively on set semantics. In addition, in order to express rewrite rules in these formalism, non-trivial transformation from SQL are required.
Previous SQL formalizations in proof systems include \cite{MalechaMSW10POPL, BenzakenCD14ESOP, VeanesGHT09ICFEM, VeanesTH10LAPR16}. In \cite{VeanesGHT09ICFEM, VeanesTH10LAPR16}, SQL semantics are encoded in the Z3 SMT solver for test generation. In \cite{MalechaMSW10POPL}, an end to end verified prototype database system is implemented in Coq. In \cite{BenzakenCD14ESOP}, a relational data model and relational algebra are implemented in Coq. Compared with \cite{MalechaMSW10POPL, BenzakenCD14ESOP}, \sem covers all important SQL feature such as bags, aggregation, group by, indexes, and correlated subqueries. As a result, we are able to express a wide range of rewrite rules. Unlike \cite{MalechaMSW10POPL}, we did not build an end to end formally verified database system.
In \sem, SQL features like aggregation on group by and indexes are supported through syntactic rewrites. Rewriting aggregation on group by using aggregation on relations and correlated subqueries is based on \cite{BunemanLSTW94SIGMOD}. We use logical relation to represent indexes in \sem. This was firstly proposed by Tastalos et al \cite{TsatalosSI94}.
\subsection{Related Formal Semantics in Proof Systems}
In the past decades, a number of formal semantics in different application domains were developed using proof systems for software verification. The CompCert compiler \cite{Leroy09JACM} specifies the semantics of a subset of C in Coq and provides machine checkable proofs of the correctness of compilation. HALO denotes Haskell to first-order logic for static verification of contracts \cite{VytiniotisJCR13POPL}.
Bagpipe \cite{WeitzWTEKT2016:TR} developed formal semantics for the Border Gateway Protocol (BGP) to check the correctness of BGP configurations. SEL4 \cite{SEL4} formally specifies the functional correctness of an OS kernel in Isabelle/HOL and developed a verified OS kernel. FSCQ \cite{FSCQ} builds a crash safe file system using an encoding of crash Hoare logic in Coq. With formal semantics in proof systems, there are more verified system developed such as Verdi \cite{WilcoxWPTWEA15PLDI}, Verve \cite{YangH10PLDI}, Bedrock \cite{Chlipala13} and Ironclad \cite{HawblitzelHLNPZZ14OSDI}.
\section{\sys: A Verified System for Proving Rewrite Rules}
\label{sec:rules}
To demonstrate the effectiveness of \sem, we implement \sys, a system written in Coq for checking the equivalence of SQL rewrite rules. \sys consists of four parts, 1) the denotational semantics of \sem, 2) a library consisting of lemmas and tactics that can be used as building blocks for constructing proofs of arbitrary rewrite rules, 3) a fully automated decision procedure for the equivalence of rewrite rules consisting only of conjunctive queries, and 4) a number of proofs of existing rewrite rules from the database literature and real world systems.
\sys relies on the Homotopy Type Theory Coq library~\cite{HoTTCoq}. Its trusted code base contains 296 lines of specification of \sem. Its verified part contains 405 lines of library code (including the decision procedure for conjunctive queries), and 1094 lines of code that prove well known SQL rewrite rules.
In the following sections, we first show various rewrite rules and the lemmas they use from the \sys library, and then explain our automated decision procedure.
\subsection{Proving Rewrite Rules in \sys by Examples}
We proved 23 rewrite rules from both the database literature and real world
optimizers using \sem. Figure~\ref{fig:rules} shows the number of
rewrite rules that we proved in each category and the average lines of code
(LOC) required per proof.
\begin{figure}
\centering
\begin{tabular}{llll}
Category & No. of rules & Avg. LOC (proof only) \\ \hline
Basic & 8 & 11.1 \\
Aggregation & 1 & 50 \\
Subquery & 2 & 17 \\
Magic Set & 7 & 30.3 \\
Index & 3 & 64 \\
Conjunctive Query & 2 & 1 (automatic) \\
\textbf{Total} & 23 & 25.2 \\
\end{tabular}
\caption{Rewrite rules proved}
\label{fig:rules}
\end{figure}
The following sections show a sampling of interesting rewrite rules in these categories.
Sec~\ref{sec:basic-rewrite} shows how two basic rewrite rules are proved.
Sec~\ref{sec:agg-rule} shows how to prove a rewrite rule involving aggregation.
Sec~\ref{sec:magic} shows how to prove the magic set rewrite rules.
Sec~\ref{sec:index-rule} shows how to state a rewrite rule involving indexes.
\subsubsection{Basic Rewrite Rules}
\label{sec:basic-rewrite}
Basic rewrites are simple rewrite rules that are fundamental building blocks
of the rewriting system. These rewrites are also very effective in terms of
reducing query execution time. We demonstrate how to prove the correctness of
basic rewrite rules in \sys using two examples: selection push down and
commutativity of joins.
\paragraph{Selection Push Down} Selection push down moves a selection (filter)
directly after the scan of the input table to dramatically reduce the amount of
data in the execution pipeline as early as possible. It is known as one of most
powerful rules in database optimizers~\cite{dbSysBook}.
We formulate selection push down as the following rewrite rule in \inputLang:
\begin{small}
\[
\begin{array}{ll}
\denoteQuery{\Gamma}{\texttt{SELECT * FROM $R$ WHERE $p_1$ AND $p_2$ }}{\sigma} & \equiv \\
\llbracket \Gamma \vdash \texttt{SELECT * FROM} & \\
\multicolumn{2}{l}{ \qquad \texttt{(SELECT * FROM $R$ WHERE $p_1$) WHERE $p_2$ : $\sigma$} \rrbracket} \\
\end{array}
\]
\end{small}
This will be denoted to:
\begin{small}
\[
\begin{array}{ll}
\lambda \; \context \; \tuple. \; \D{p_1} \mkPair{g}{t} \times \llbracket p_2 \rrbracket \; \mkPair{g}{t} \times \D{R} \; g \; t & \equiv \\
\lambda \; \context \; \tuple. \; \D{p_2} \; \mkPair{g}{t} \times ( \D{p_1} \; \mkPair{g}{t} \times \D{R} \; g \; t) \\
\end{array}
\]
\end{small}
The proof proceeds by functional extensionality \footnote{Function
extensionality is implied by the Univalence Axiom.} and the associativity and
commutativity of $\times$.
\paragraph{Commutativity of Joins} Commutativity of joins allows an optimizer to rearrange
the order of joins in order to get the join order with best performance. This is one
of the most fundamental rewrite rules that almost every optimizer uses. We formulate the
commutativity of joins in \inputLang as follows:
\begin{small}
\[
\begin{array}{ll}
\denoteQuery{\Gamma}{\texttt{SELECT * FROM $R$, $S$}}{\Pair{\sigma_R}{\sigma_S}} \quad \equiv \\
\denoteQuery{\Gamma}{\texttt{SELECT Right.Right.*, Right.Left.* FROM $S$, $R$}}{\Pair{\sigma_R}{\sigma_S}}
\end{array}
\]
\end{small}
Note that the select clause flips the tuples from $S$ and $R$, such that the order of the tuples matches the original query.
This will be denoted to:
\begin{small}
\[
\begin{array}{ll}
\lambda \; \context \; \tuple. \; \D{R} \; g \; \fst{t} \times \D{S} \; g \; \snd{t} \quad \equiv \\
\lambda \; \context \; \tuple. \; \sum_{t_1} \D{S} \; g \; \fst{t_1} \times \D{R} \; g \; \snd{t_1} \times ((\snd{t_1},\fst{t_1}) = t)
\end{array}
\]
\end{small}
The proof uses Lemma~\ref{lem:sum_pair} provided by the \sys library.
\begin{lemma}
\label{lem:sum_pair}
Let $A$ inhabit $\mathcal{U}$, and have $P: A \rightarrow \mathcal{U}$ be a type family, then we have:
\[
\sum_{x:A \times B} P \; x = \sum_{x: B \times A} P \; (\snd{x}, \fst{x})
\]
\end{lemma}
Together with the fact that $t_1 = (\fst{t_1}, \snd{t_1})$, the rewrite rule's right hand side becomes:
\begin{small}
\[ \lambda \; \context \; \tuple. \; \sum_{t'} \D{S} \; g \; \snd{t'} \times \D{R} \; g \; \fst{t'} \times (t' = t ) \]
\end{small}
The proof then uses Lemma~\ref{lem:pair_eq} provided by the \sys library.
\begin{lemma}
\label{lem:pair_eq}
Let $A$ and $B$ inhabit $\mathcal{U}$, and have $P: A \times B \rightarrow \mathcal{U}$, then we have:
\[
P\; x = \sum_{x'} P \; x' \times (x' = x)
\]
\end{lemma}
After applying Lemma~\ref{lem:pair_eq}, the right hand side becomes the following,
and we can finish the proof by applying commutativity of $\times$:
\begin{small}
\[
\lambda \; \context \; \tuple. \; \D{S} \; g \; \snd{t} \times \D{R} \; g \; \fst{t}
\]
\end{small}
\subsubsection{Aggregation and Group By Rewrite Rules}
\label{sec:agg-rule}
Aggregation and Group By are widely used in analytic queries \cite{ChaudhuriD97SIGMOD}. The standard data analytic benchmark TPC-H \cite{tpch} has 16 queries with group by and 21 queries with aggregation out of a total of 22 queries. Following is an example rewrite rule for aggregate queries.
The query on the left-hand side groups the relation $R$ by the column $k$, sums
all values in the $b$ column for each resulting partition, and then removes all
results except the partition whose column $k$ is equal to the constant $l$.
This can be rewritten to the faster query that first removes all tuples from
$R$ whose column $k \neq l$, and then computes the sum.
\begin{small}
\[
\begin{array}{ll}
\llbracket \Gamma \vdash \texttt{SELECT * FROM (SELECT $k$, SUM($b$) FROM $R$ GROUP BY $k$)} & \\
\qquad \; \texttt{WHERE $(\Var{k}) = l$} : \sigma \rrbracket \quad \equiv \\
\denoteQuery{\Gamma}{\texttt{SELECT $k$, SUM($b$) FROM $R$ WHERE $(\Var{k})=l$ GROUP BY $k$ }}{\sigma}
\end{array}
\]
\end{small}
As shown in Sec.~\ref{sec:more-sql}, we use a correlated subquery and a unary
aggregate function (which takes a \inputLang query as its input) to represent
aggregation on group by SQL queries. After de-sugaring, the group by query
becomes \texttt{SELECT DISTINCT $\ldots$}. The rule will thus be denoted to:
\begin{small}
\[
\begin{array}{ll}
\lambda \; \context \; \tuple. \; (\fst{t} = \denote{l}) \times \| \sum_{t_1} \denote{R} \; t_1 \times
(\fst{t} = \denote{k} \; t_1) \times \\
\qquad \quad (\snd{t} = \denote{\texttt{SUM}} \; (\lambda \; t'. \; \sum_{t_2} (\denote{k} \; t_1 = \denote{k} \; t_2) \times \\
\qquad \qquad \quad \denote{R} \; t_2 \times (\denote{b} \; t_1 = t')) \| \\
\quad \equiv \quad \\
\lambda \; \context \; \tuple. \; \| \sum_{t_1} (\denote{k} \; t_1 = \denote{l}) \times \denote{R} \; t_1 \times
(\fst{t} = \denote{k} \; t_1) \times \\
\qquad \quad (\snd{t} = \denote{\texttt{SUM}} \; (\lambda \; t'. \; \sum_{t_2}
(\denote{k} \; t_1 = \denote{k} \; t_2) \times (\denote{k} \; t_2 = \denote{l})
\times \\
\qquad \qquad \quad \denote{R} \; t_2 \times (\denote{b} \; t_1 = t')) \|
\end{array}
\]
\end{small}
The proof proceeds by functional extensionality, after which both sides become squash types.
The proof then uses the fundamental lemma about squash types, where
for all squash types $A$ and $B$,
$(A \leftrightarrow B) \Rightarrow (A = B)$.
It thus suffices to prove by cases the bi-implication ($\leftrightarrow$) of both sides.
In both cases, instantiate $t_1$ with $t_1$ ($t_1$ is the witness of the $\Sigma$ hypothesis).
It follows that $t.1 = \denote{l} = \denote{k} \; t_1$, and thus that $\denote{k} \; t_2 = \denote{l}$ inside $\texttt{SUM}$.
\subsubsection{Magic Set Rewrite Rules}
\label{sec:magic}
Magic set rewrites are well known rewrite rules that were originally used in
the recursive query processing in deductive databases \cite{BancilhonMSU86PODS,
RohmerLK86NGC}. It was then used for rewriting complex decision support queries
and has been implemented in commercial systems such as IBM's DB2 database
\cite{SeshadriHPLRSSS96SIGMOD, MumickFPR90SIGMOD}. Below is an example of a
complex magic set rewrite from \cite{SeshadriHPLRSSS96SIGMOD}.
\noindent Original Query:
\begin{small}
\[
\begin{array}{ll}
\texttt{CREATE VIEW $DepAvgSal$ AS} \\
\qquad \texttt{(SELECT $E.did$, AVG($E.sal$) AS $avgsal$ FROM $Emp$ $E$} \\
\qquad \texttt{GROUP BY $E.did$);} \\
\texttt{SELECT $E.eid$, $E.sal$ FROM $Emp$ $E$, $Dept$ $D$, $DepAvgSal$ $V$ } \\
\texttt{WHERE $E.did = D.did$ AND $E.did = V.did$ AND $E.age < 30$} \\
\qquad \texttt{AND $D.budget > 100000$ AND $E.sal > V.avgsal$}\\
\end{array}
\]
\end{small}
\noindent Rewritten Query:
\begin{small}
\[
\begin{array}{ll}
\texttt{CREATE VIEW $PartialResult$ AS} \\
\qquad \texttt{(SELECT $E.eid$, $E.sal$, $E.did$ FROM $Emp$ $E$, $Dept$ $D$}\\
\qquad \; \texttt{WHERE $E.did = D.did$ AND $E.age < 30$ AND} \\
\qquad \qquad \texttt{$D.budget > 100000$);}\\
\texttt{CREATE VIEW $Filter$ AS} \\
\qquad \texttt{(SELECT DISTINCT $P.did$ FROM $PartialResult$ $P$);}\\
\texttt{CREATE VIEW $LimitedDepAvgSal$ AS} \\
\qquad \texttt{(SELECT $F.did$, AVG($E.sal$) AS $avgsal$} \\
\qquad \; \texttt{FROM $Filter$ $F$, $Emp$ $E$} \\
\qquad \; \texttt{WHERE $E.did = F.did$} \\
\qquad \; \texttt{GROUP BY $F.did$);} \\
\texttt{SELECT $P.eid$, $P.sal$} \\
\texttt{FROM $PartialResult$ $P$, $LimitedDepAvgSal$ $V$} \\
\texttt{WHERE $P.did = V.did$ AND $P.sal > V.avgsal$}\\
\end{array}
\]
\end{small}
This query aims to find each young employee in a big department ($D.budget > 100000$) whose salary is higher then the average salary in her department. Magic set rewrites use the fact that only the average salary of departments that are big and have young employees need to be computed. As described in \cite{SeshadriHPLRSSS96SIGMOD}, all magic set rewrites can be composed from just three basic rewrite rules on semijoins, namely introduction of $\theta$-semijoin, pushing $\theta$-semijoin through join, and pushing $\theta$-semijoin through aggregation.
Following, we show how to state all three rewrite rules using \sys, and show how to prove two.
We firstly define $\theta$-semijoin as a syntactic rewrite in \sem:
\begin{small}
\[
\begin{array}{ll}
\texttt{$A$ SEMIJOIN $B$ ON $\theta$} \triangleq \\
\texttt{SELECT * FROM $A$ WHERE EXISTS (SELECT * FROM $B$ WHERE $\theta$)}
\end{array}
\]
\end{small}
\paragraph{Introduction of $\theta$-semijoin}
This rules shows how to introduce semijoin from join and selection.
Using semijoin algebra notation, this rewrite can be expressed as follows:
\[
R_1 \bowtie_{\theta} R_2 \equiv R_1 \bowtie_{\theta} (R_2 \ltimes_{\theta} R_1)
\]
\noindent Using \inputLang, the rewrite can be expressed as follows:
\begin{small}
\[
\begin{array}{ll}
\denoteQuery{\Gamma}{\texttt{SELECT * FROM $R_2$, $R_1$ WHERE $\theta$}}{\sigma} & \quad \equiv \quad \\
\multicolumn{2}{l}{\denoteQuery{\Gamma}{\texttt{SELECT * FROM ($R_2$ SEMIJOIN $R_1$ ON $\theta$), $R_1$ WHERE $\theta$ }}{\sigma}}
\end{array}
\]
\end{small}
\noindent which is denoted to:
\begin{small}
\[
\begin{array}{ll}
\lambda \; \context \; \tuple. \; \denote{\theta} \; \mkPair{g}{t} \times \denote{R_2} \; g \; \fst{t} \times \denote{R_1} \; g \; \snd{t} & \quad \equiv \quad \\
\lambda \; \context \; \tuple. \; \denote{\theta} \; \mkPair{g}{t} \times \denote{R_2} \; g \; \fst{t} \times \denote{R_1} \; g \; \snd{t} \times{} \\
\qquad \quad \merely{\sum_{t_1}\denote{\theta} \; \mkPair{g}{\mkPair{\fst{t}}{t_1}} \times \denote{R_1} \; g \; t_1}
\end{array}
\]
\end{small}
\noindent The proof uses Lemma~\ref{lem:hprop_prod} provided by the \sys library.
\begin{lemma}
\label{lem:hprop_prod}
$\forall P, T: \mathcal{U}$, where $P$ is either $\textbf{0}$ or $\textbf{1}$, we have:
\[ (T \rightarrow P) \Rightarrow ((T \times P) = T) \]
\end{lemma}
\begin{proof}
Intuitively, this can be proven by cases on $T$.
If $T$ is inhabited, then $P$ holds by assumption, and $T \times \textbf{1} = T$.
If $T = 0$, then $\textbf{0} \times P = \textbf{0}$.
\end{proof}
Using this lemma, it remains to be shown that
$\denote{\theta} \; \mkPair{g}{t}$ and $\denote{R_2} \; g \; \fst{t}$ and $\denote{R_1} \; g \; \snd{t}$
imply
$\merely{\sum_{t_1}\denote{\theta} \; \mkPair{g}{\mkPair{\fst{t}}{t_1}} \times \denote{R_1} \; g \; t_1}$.
We show this by instantiating $t_1$ with $t.2$, and then by hypotheses.
\paragraph{Pushing $\theta$-semijoin through join}
The second rule in magic set rewrites is
the rule for pushing $\theta$-semijoin through join, represented in semijoin algebra as:
\[
(R_1 \bowtie_{\theta_1} R_2) \ltimes_{\theta_2} R_3 \equiv (R_1 \bowtie_{\theta_1}R_2') \ltimes_{\theta_2} R_3
\]
\noindent where $R_2' = E_2 \ltimes_{\theta_1 \land \theta_2} (R_1 \bowtie R_3)$.
This rule can be written in \inputLang as below:
\begin{small}
\[
\begin{array}{ll}
\lambda \; \context \; \tuple. \; \llbracket \Gamma \vdash \texttt{(SELECT * FROM $R_1$, $R_2$ WHERE $\theta_1$)} \\
\qquad \qquad \quad \texttt{SEMIJOIN $R_3$ ON $\theta_2$}:\Pair{\sigma_1}{\sigma_2} \rrbracket \quad \equiv \quad \\
\lambda \; \context \; \tuple. \; \llbracket \Gamma \vdash \texttt{(SELECT *} \\
\qquad \qquad \quad \texttt{ FROM $R_1$, ($R_2$ SEMIJOIN (FROM $R_1$, $R_3$) ON $\theta_1$ AND $\theta_2$)} \\
\qquad \qquad \quad \texttt{ WHERE $\theta_1$) SEMIJOIN $R_3$ ON $\theta_2$} :\Pair{\sigma_1}{\sigma_2} \rrbracket
\end{array}
\]
\end{small}
\noindent The rule is denoted to:
\begin{small}
\[
\begin{array}{ll}
\lambda \; \context \; \tuple. \; \merely{\sum_{t_1} \denote{\theta_2} \; \mkPair{g}{\mkPair{t}{t_1}} \times \denote{R_3} \; g \; t_1} \times \\
\qquad \quad \denote{\theta_1} \; \mkPair{g}{t} \times \denote{R_1} \; g \; \fst{t} \times \denote{R_2} \; g \; \snd{t}
\qquad \equiv \\
\lambda \; \context \; \tuple. \; \merely{\sum_{t_1} \denote{\theta_2} \; \mkPair{g}{\mkPair{t}{t_1}} \times \denote{R_3} \; g \; t_1} \times \\
\qquad \quad \denote{\theta_1} \; \mkPair{g}{t} \times \denote{R_1} \; g \; \fst{t} \times \denote{R_2} \; g \; \snd{t} \times{} \\
\qquad \quad \| \sum_{t_1} \denote{\theta_1} \; \mkPair{g}{\mkPair{\fst{t_1}}{\snd{t}}} \times
\denote{\theta_2} \; \mkPair{g}{\mkPair{\mkPair{\fst{t_1}}{\snd{t}}}{\snd{t_1}}}
\\
\qquad \quad \times \denote{R_1} \; g \; \fst{t_1} \times \denote{R_3} \; g \; \snd{t_1} \|
\end{array}
\]
\end{small}
We can prove this rule by using a similar approach to the one used to prove introduction of $\theta$-semijoin:
rewriting the right hand side using Lemma~\ref{lem:hprop_prod}.
and then instantiating $t_1$ with $(t.1, t_1)$ ($t_1$ is the witness of the $\Sigma$ hypothesis).
\paragraph{Pushing $\theta$-semijoin through aggregation}
The final rule is pushing $\theta$-semijoin through aggregation:
\newcommand{_{\bar{g}}\mathcal{F}_{\bar{f}}}{_{\bar{g}}\mathcal{F}_{\bar{f}}}
\[
_{\bar{g}}\mathcal{F}_{\bar{f}}(R_1) \ltimes_{c_1 = c_2} R_2 \equiv _{\bar{g}}\mathcal{F}_{\bar{f}}(R_1 \ltimes_{c_1 = c_2} R_2)
\]
\noindent where $_{\bar{g}}\mathcal{F}_{\bar{f}}$ is a grouping/aggregation operator ($_{\bar{g}}\mathcal{F}_{\bar{f}}$ was firstly defined in \cite{SeshadriHPLRSSS96SIGMOD}), and $\bar{g}$ denotes the group by attributes and $\bar{f}$ denotes the aggregation function. In this rule, one extra condition is that $c_1$ is from the attributes in $\bar{g}$ and $c_2$ is from the attributes of $R_2$. This rule can be written in \inputLang as below:
\begin{small}
\[
\begin{array}{ll}
\llbracket \Gamma \vdash \texttt{(SELECT $c_1$, COUNT($a$) FROM $R_1$ GROUP BY $c_1$)} \\
\qquad \quad \texttt{SEMIJOIN $R_2$ ON $c_1 = c_2$ } : \sigma \rrbracket \quad \equiv \\
\llbracket \Gamma \vdash \texttt{SELECT $c_1$, COUNT($a$)} \\
\qquad \quad \texttt{FROM ($R_1$ SEMIJOIN $R_2$ ON $c_1 = c_2$) GROUP BY $c_1$} : \sigma \rrbracket
\end{array}
\]
\end{small}
\noindent We omit the proof here for brevity.
\subsubsection{Index Rewrite Rules}
\label{sec:index-rule}
As introduced in Section~\ref{sec:more-sql}, we define an index as a \sem query that projects on the indexed attribute and the primary key of a relation. Assuming $k$ is the primary key of relation $R$, and $I$ is an index on column $a$:
\[ I := \texttt{SELECT $k$, $a$ FROM $R$} \]
\noindent We prove the following common
rewrite rule that converts a full table scan to a lookup on an index and a join:
\begin{small}
\[
\begin{array}{ll}
\denoteQuery{\Gamma}{\texttt{SELECT * FROM $R$ WHERE $a = l$ }}{\sigma} \quad \equiv \\
\llbracket \Gamma \vdash \texttt{SELECT * FROM $I$, $R$} \\
\qquad \quad \texttt{WHERE $a = l$ AND Right.Left.$k$ = Right.Right.$k$ } : \sigma \rrbracket
\end{array}
\]
\end{small}
\noindent We omit the proof here for brevity.
\subsection{Automated Decision Procedure for Conjunctive Queries}
\label{sec:automation}
\begin{figure*}
\centering
\begin{tabular}{|l|l|l|l|l|} \hline
& Containment (Set) & Containment (Bag) & Equivalence (Set) & Equivalence (Bag) \\ \hline
Conjunctive Queries & NP-Complete \cite{ChandraM77STOC} & Open & NP-Complete \cite{ChandraM77STOC} & Graph Isomorphism \cite{ChaudhuriV92PODS} \\ \hline
Union of Conjunctive Queries & NP-Complete \cite{SagivY80JACM} & Undecidable \cite{IoannidisR95TODS} & NP-Complete \cite{SagivY80JACM} & Open \\ \hline
Conjunctive Query with $\neq$, $\geq$, and $\leq$ &
$\Pi_2^p$-Complete \cite{Meyden92PODS} & Undecidable \cite{JayramKV06PODS} & $\Pi_2^p$-Complete \cite{Meyden92PODS} & Undecidable \cite{JayramKV06PODS} \\ \hline
First Order (SQL) Queries & Undecidable \cite{Trakhtenbrot50DANUSSR} & Undecidable & Undecidable & Undecidable \\ \hline
\end{tabular}
\caption{Complexities of Query Containment and Equivalence}
\label{fig:complexity}
\end{figure*}
\begin{figure}
\tikzset{
internode/.style = {shape=circle, draw, align=center}
}
\begin{tikzpicture}
\begin{scope}
\node [internode] (root) {}
child {node (n1) {$\sigma_{R_1}$}}
child {node (n2) {$\sigma_{R_2}$}};
\node [text width=6cm, anchor=east] at (6.5, 0) {Schema of $t_1$ (left)};
\end{scope}
\begin{scope}[xshift=4cm]
\node [internode] {}
child {node [internode] {}
child {node (n3) {$\sigma_{R_1}$}}
child {node (n4) {$\sigma_{R_1}$}}
}
child {node (n5) {$\sigma_{R_2}$}};
\node [text width=6cm, anchor=east] at (6.5, 0) {Schema of $t_1$ (right)};
\end{scope}
\draw[->, dashed, blue] (n3)--(n1);
\draw[->, dashed, blue] (n4)--(n1);
\draw[->, dashed, blue] (n5)--(n2);
\end{tikzpicture}
\begin{tikzpicture}
\begin{scope}
\node [internode] (root) {}
child {node (n1) {$\sigma_{R_1}$}}
child {node (n2) {$\sigma_{R_2}$}};
\node [text width=6cm, anchor=east] at (6.5, 0) {Schema of $t_1$ (left)};
\end{scope}
\begin{scope}[xshift=4cm]
\node [internode] {}
child {node [internode] {}
child {node (n3) {$\sigma_{R_1}$}}
child {node (n4) {$\sigma_{R_1}$}}
}
child {node (n5) {$\sigma_{R_2}$}};
\node [text width=6cm, anchor=east] at (6.5, 0) {Schema of $t_1$ (right)};
\end{scope}
\draw[->, dashed, red] (n1)--(n3);
\draw[->, dashed, red] (n2)--(n5);
\end{tikzpicture}
\caption{The mappings found to prove the conjunctive query example, blue lines show the mapping found to prove left $\rightarrow$ right, red lines show the mapping found to prove right $\rightarrow$ left. }
\label{fig:mappings}
\end{figure}
The equivalence of two SQL queries is in general undecidable. Figure~\ref{fig:complexity} shows the complexities of deciding containment and equivalence of subclasses of SQL. The most well-known subclass are conjunctive queries, which are of the form \texttt{DISTINCT SELECT $p$ FROM $q$ WHERE $b$}, where $p$ is a sequence of arbitrarily many attribute projections, $q$ is the cross product of arbitrarily many input relations, and $b$ is a conjunct consisting of arbitrarily many equality predicates between attribute projections.
We implement a decision procedure to automatically prove the equivalence of conjunctive queries in \inputLang. After denoting the \inputLang query to \outputLang,
the decision procedure automates steps similar to the proof in Section \ref{sec:agg-rule}.
First, after applying functional extensionality, both sides become squash types due to the \texttt{DISTINCT} clause. The procedure then applies the fundamental lemma about squash types $\forall A B, (A \leftrightarrow B) \Rightarrow (A = B)$. In both cases of the resulting
bi-implication, the procedure tries all possible instantiations of the $\Sigma$, which is due to the \texttt{SELECT} clause. This search for the correct instantiation is implemented using Ltac's built-in backtracking support. The procedure then rewrites all equalities and tries to discharge the proof by direct application of hypotheses.
The following is an example of two equivalent conjunctive SQL queries that we can solve using our decision procedure:
\begin{small}
\[
\begin{array}{ll}
\texttt{SELECT DISTINCT $x$.$c_1$ FROM $R_1$ AS $x$, $R_2$ AS $y$} \\
\texttt{WHERE $x$.$c_2$ = $y$.$c_3$} & \quad \equiv \\
\texttt{SELECT DISTINCT $x$.$c_1$ FROM $R_1$ AS $x$, $R_1$ AS $y$, $R_2$ AS $z$} \\
\texttt{WHERE $x$.$c_1$ = $y$.$c_1$ AND $x$.$c_2$ = $z$.$c_3$}
\end{array}
\]
\end{small}
The same queries can be expressed in \inputLang as follows, where the schema of $R_i$ is $\sigma_{R_i}$:
\begin{small}
\[
\begin{array}{ll}
\llbracket \Gamma \vdash \texttt{DISTINCT SELECT Right.Left.$c_1$ FROM $R_1$, $R_2$} \\
\qquad \quad \texttt{WHERE Right.Left.$c_2$ = Right.Right.$c_3$} :\sigma \rrbracket \quad \equiv \\
\llbracket \Gamma \vdash \texttt{DISTINCT SELECT Right.Left.Left.$c_1$} \\
\qquad \quad \texttt{FROM (FROM $R_1$, $R_1$), $R_2$} \\
\qquad \quad \texttt{WHERE Right.Left.Left.$c_1$ = Right.Left.Right.$c_1$ AND } \\
\qquad \qquad \qquad \texttt{Right.Left.Left.$c_2$ = Right.Right.$c_3$} : \sigma \rrbracket
\end{array}
\]
\end{small}
\noindent which is denoted as:
\begin{small}
\[
\begin{array}{ll}
\lambda \; \context \; \tuple. \; \| \sum_{t_1}
\denote{R_1} \; g \fst{t_1} \times \denote{R_2} \; g \; \snd{t_1} \times{} \\
\qquad \quad (\denote{c_2} \; \fst{t_1} = \denote{c_3} \; \snd{t_1} ) \times {} \\
\qquad \quad (\denote{c_1} \; \fst{t_1} = t) \times \| \qquad \equiv \\
\lambda \; \context \; \tuple. \; \| \sum_{t_1} \denote{R_1} \; g \; \fst{\fst{t_1}} \times \denote{R_1} \; g \; \snd{\fst{t_1}} \times \denote{R_2} \; g \; \snd{t_1} \times{} \\
\qquad \quad (\denote{c_1} \; \fst{\fst{t_1}} = \denote{c_1} \; \snd{\fst{t_1}}) \times (\denote{c_2} \; \fst{\fst{t_1}} = \denote{c_3} \; \snd{t_1}) \times{} \\
\qquad \quad (\denote{c_1} \fst{\fst{t_1}} = t) \|
\end{array}
\]
\end{small}
The decision procedure turns this goal into a bi-implication, which it proves by cases.
For the $\rightarrow$ case, the decision procedure destructs the available $\Sigma$ witness into
tuple $t_x$ from $R_1$ and $t_y$ from $R_2$ and tries all instantiations of $t_1$ using these tuples. The instantiation $t_1 = ((t_x, t_x), t_y)$ allows the procedure to complete the proof after rewriting all equalities.
For the $\leftarrow$ case, the available tuples are $t_x$ from $R_1$, $t_y$ from $R_1$, and $t_z$ from $R_2$.
The instantiation $t_1 = (t_x, t_z)$ allows the procedure to complete the proof after rewriting all equalities. This assignment is visualized in Figure~\ref{fig:mappings}.
\section{\sem and Its Semantics}
\label{sec:semantics}
In this section, we present \sem, a SQL-like language for expressing rewrite rules.
To simplify discussion, we first describe how relational
data structures are modeled
in Section~\ref{sec:data-model}. We then describe \sem, a language built
on top of our relational data structures that covers
all major features of SQL, in Section~\ref{sec:input-lang}. Next, we define
\outputLang, the formal expressions into which \sem is translated, in
Section~\ref{sec:target-lang}.
\subsection{Data Model}
\label{sec:data-model}
We first describe how schemas for relations and tuples are modeled
in \sem. Both of these foundational concepts from relational
theory~\cite{Codd70CACM} are what \sem uses to build upon.
\begin{figure}[t]
\[
\begin{array}{llll}
\tau \in \texttt{Type} & ::= & \texttt{int} ~|~ \texttt{bool} ~|~ \texttt{string} ~|~ \; \ldots \\
\denote{\texttt{int}} & ::= & \mathds{Z} \\
\denote{\texttt{bool}} & ::= & \mathds{B} \\
\denote{\texttt{string}} & ::= & String \\
\ldots & & \\ \\
\sigma \in \texttt{Schema} & ::= & \texttt{empty} \\
& \; \mid & \leaf{\tau} \\
& \; \mid & \node{\sigma_1}{\sigma_2} \\ \\
\texttt{Tuple} \; \texttt{empty} & ::= & \texttt{Unit} \\
\texttt{Tuple} \; (\node{\sigma_1}{\sigma_2}) & ::= & \texttt{Tuple} \; \sigma_1 \times \texttt{Tuple} \; \sigma_2 \\
\texttt{Tuple} \; (\leaf{\tau}) & ::= & \denote{\tau} \\
\end{array}
\]
\caption{Data Model of \sem}
\label{fig:data-model}
\end{figure}
\begin{figure}[t]
\begin{minipage}{.4\columnwidth}
\tikzset{
internode/.style = {shape=circle, draw, align=center}
}
\begin{tikzpicture}
\begin{scope}
\node [internode, anchor=north] (root) at (0, 0) {}
child {node {\texttt{string}}}
child {node [internode] {}
child {node {\texttt{int}}}
child {node {\texttt{bool}}}
};
\end{scope}
\end{tikzpicture}
\end{minipage}
\begin{minipage}{.6\columnwidth}
\begin{small}
\[
\begin{array}{lll}
\sigma : \texttt{Schema} & = & \node{(\leaf{\texttt{int}})}{} \\
& & \quad (\node{(\leaf{\texttt{int}})}{} \\
& & \quad\quad (\leaf{\texttt{bool}})) \\
\texttt{Tuple} \; \sigma & = & String \times (\mathds{Z} \times \mathds{B}) \\
t : \texttt{Tuple} \; \sigma & = & \mkPair{``Bob''}{\mkPair{52}{\texttt{true}}}
\end{array}
\]
\end{small}
\end{minipage}
\caption{An Example of \sem Schema and Tuple}
\label{ex:data-model}
\end{figure}
\paragraph{Schema and Tuple}
We briefly review the standard SQL definitions of a schema and a tuple.
Conceptually, a database schema is an unordered bag of $(n, \tau)$ pairs, where
$n$ is an attribute name, and $\tau$ is the type of the attribute.
For example, the schema of a table containing personal information might be:
\[ \{(\text{Name}, \texttt{string}), (\text{Age}, \texttt{int}), (\text{Married}, \texttt{bool})\} \]
A database tuple is a collection of values that conform to a
given schema. For example, the following is a tuple with the
aforementioned schema:
\[ \{\text{Name}:\text{``Bob''}; \; \text{Age}:52; \; \text{Married}: \texttt{true} \} \]
Attributes from tuples are accessed using record syntax. For instance
$t.\text{Name}$ returns ``Bob'' where $t$ refers to the tuple above.
As shown in Figure~\ref{fig:data-model},
we assume there exists a set of SQL types \texttt{Type}, which can be
denoted into types in Coq.
In \sem, we define schemas and tuples as follows. A schema is modeled
as a collection of types
organized in a binary tree, with each type corresponding to an attribute.
As shown in~\figref{data-model}, a schema can be constructed
from the $\texttt{empty}$ schema,
an individual type $\tau$, or recursively
from two schema nodes $s_1$ and $s_2$, corresponding to the branches
of the subtree.
As we will see, this organization is beneficial in
both writing \sem rewrite rules and also reasoning about the equivalence
of schemas.
The tuple type in \sem is defined as a dependent type on a schema. As shown in~\figref{data-model}, a tuple is an nested pair with the identical structure as its schema. Given a schema $s$,
if $s$ is the empty schema, then the (only) instance of $\texttt{Tuple} \; \texttt{empty}$ is {\tt Unit} (i.e., empty) tuple.
Otherwise, if $s$ is a leaf node in the schema tree with type $\tau$,
then a tuple is simply a value of the type $\D{\tau}$.
Finally, if $s$ is recursively
defined using two schemas $s_1$ and $s_2$, then the resulting tuple
is an instance of a product type $\texttt{Tuple}(s_1) \times \texttt{Tuple}(s_2)$.
As an illustration,
Figure~\ref{ex:data-model} shows a tuple $t$ and its schema $\sigma$, where $\sigma$ =
\texttt{(node (leaf \textrm{Int}) (node (leaf \textrm{Int}) (leaf
\textrm{Int})))} and
$t$ has the nested pair type $(String \times (\mathds{Z} \times \mathds{B}))$.
To access an element from a tuple, \sem uses path
expressions with selectors $\texttt{Left}$ and $\texttt{Right}$. For instance,
the path $\texttt{Left}.\texttt{Right}$ retrieves
the value 52 from the tuple $t$ in \figref{data-model}.
As will be shown in \secref{denotation}, path expressions will be denoted to standard
pair operations, i.e., $\texttt{Left}$ will be denoted to ``$\fst{}$'', which returns the first element from
a pair, and $\texttt{Left}$ will be denoted to ``$\snd{}$'', which returns the second. Such expressions can be
composed to retrieve nested pair types. For instance, $\texttt{Left}.\texttt{Right}$ will be denoted to ``$\snd{\fst{}}$'', thus it retrieves 52 from $t$ in \figref{data-model}.
\paragraph{Relation}
In \sem a relation is modeled as a function from tuples to homotopy
types called HoTT-relations, $\texttt{Tuple} \; \sigma \rightarrow \mathcal{U}$, as
already discussed in~\secref{overview}; we define homotopy types
shortly.
\paragraph{Discussion}
We briefly comment on our choice of data model. There are two
approaches to defining tuples in database theory \cite{TheAliceBook}:
the named approach and unnamed approach. We chose an unnamed
approach, because it avoids name collisions, and because proof
assistants like Coq can decide schema equivalence of unnamed schemas
based on structural equality. Previous work~\cite{MalechaMSW10POPL}
adopted an unnamed approach as well, for the same reason. Our choice
for representing tuples as trees, rather than as ordered lists, is
non-standard: we do this in order to allow our language to express
generic rewrite rules, without specifying a particular schema for the
input relations, see \secref{rewriteRules}.
Finally, we note that, in our model, a tuple is a dependent type,
which depends on
its schema. We use dependent types to ensure that a tuple, which is a nested pair, must have the same structure as its schema, which is a binary tree, by construction. This allows us to denote a path expression (composed by $\texttt{Left}$, $\texttt{Right}$) to a series of corresponding pair operations (composed by ``$\fst{}$'', ``$\snd{}$'') easily.
\subsection{\sem: A SQL-like Query Language}
\label{sec:input-lang}
We now describe \sem, our source language used to
express rewrite rules.
Figure~\ref{fig:sql-syntax} defines the syntax of \sem. We divide the
language constructs of \sem
into four categories: queries, predicates, expressions, and projections.
\paragraph{Queries} A query takes in relation(s) and outputs another relation.
The input to a query can be a base relation (called a $Table$ in~\figref{sql-syntax})
or other queries, including
projections, cross product, selections, bag-wise operations (\texttt{UNION ALL} and \texttt{EXCEPT}),
and finally conversion to sets (\texttt{DISTINCT}).
\paragraph{Predicates} Predicates are used as part of selections
(i.e., filtering of tuples) in queries. Given a tuple $t$, predicates return
a Boolean value to indicate whether $t$ should be retained in
the output relation.
\paragraph{Expressions} Expressions are used both in predicates and projections,
and they evaluate to values (e.g., of type $\mathds{Z}$, $\mathds{B}$, $String$, etc).
Expression includes conversions from projection to expression ($\Var{p}$), uninterpreted functions on expressions,
aggregators of a query, and casts of an expression ($\CastExpr{p}{e}$).
$\Var{p}$ converts a projection $p$ into an expression. For example, $\Var{a} = \Var{b}$ is an equality predicate on attribute $a$ and attribute $b$, where $\Var{a}$ and $\Var{b}$ are the expressions representing attribute $a$ and $b$. As shown next, we use a projection to represent an attribute.
Uninterpreted functions of expressions $f(e_1, \ldots, e_n)$ are used to represent arithmetic operations on expressions such as addition, multiplication, division, mode, and constants (which are nullary uninterpreted functions).
$\texttt{CASTEXPR}$\; is a special construct that is used to represent castings of meta-variables to express generic query rewrite rules. A comprehensive discussion of meta-variables can be found in \secref{rewriteRules}. A normal (non-generic) query would not need $\texttt{CASTEXPR}$.
\paragraph{Projections} When applied to a relation,
projections denote a subset of attributes to be returned.
A projection is defined to be
a tuple to tuple function. It can be the identity function (\texttt{*}), or
return the subtree denoted by any of the
path expressions as discussed in~\secref{data-model}, such as returning the
left (\texttt{Left}) or right (\texttt{Right}) subtree of the tuple.
Any empty tuple can also be produced using \texttt{empty}. Multiple
projections can be composed using ``{\tt .}''.
Two projections $p_1$ and $p_2$ can be applied to the input tuple separately with the results combined together using ``{\tt ,}''. $\Evaluate{p}$ is used to convert a projection to an expression.
Below are examples of \sem query using projections:
\vspace{2mm}
{\centering
\begin{tabular}{lll}
& SQL & \inputLang \\ \hline
$q_1$ & \texttt{SELECT $R.*$ FROM $R, S$} & \texttt{SELECT Left.* FROM $R, S$} \\
$q_2$ & \texttt{SELECT $S.*$ FROM $R, S$} & \texttt{SELECT Right.* FROM $R, S$} \\
$q_3$ & \texttt{SELECT $S.p$ FROM $R, S$} & \texttt{SELECT Right.$p$ FROM $R, S$} \\
$q_4$ & \texttt{SELECT $R.p_1$, $S.p_2$} & \texttt{SELECT Left.$p_1$, Right.$p_2$} \\
&\texttt{FROM $R, S$} & \texttt{FROM $R, S$} \\
$q_5$ & \texttt{SELECT $p_1+p_2$ FROM $R$} & \texttt{SELECT $\Evaluate add(\Var{p_1},\Var{p_2})$} \\
& & \texttt{FROM $R$} \\
\end{tabular}
}
In $q_1$, we compose the path expressions
\texttt{Left} and \texttt{*} to represent projecting all attributes of $R$ from
a tuple that is in the result of $R \bowtie S$ \footnote{Technically
$\bowtie$ denotes natural join in relational theory~\cite{Codd70CACM}. However, since
\sem schemas are unnamed, there are no shared
names between schemas, and thus natural joins are equivalent to cross products.
Hence, we use $\bowtie$ for cross product of relations to distinguish
it from Cartesian product of types ($\times$) to be discussed in~\secref{target-lang}.}. In $q_3$,
we compose \texttt{Right} and $k$ to project to a single attribute from $S$.
The variable $k$ is a projection to a singleton tuple, which is the way to represent attributes in our semantics.
In $q_4$, we are projecting one attribute $p_1$ from $R$, and another
attribute $p_2$ from $S$ using the projection combinator ``{\tt ,}''.
In $q_5$, to represent $p_1+p_2$, we first convert attributes ($p_1$,
$p_2$) to expressions ($\Var{p_1}$, $\Var{p_2}$), then
use an uninterpreted function $\texttt{add}$ to represent addition, and cast that
back to a projection using $\Evaluate$.
\subsection{Expressing Rewrite Rules}
\seclabel{rewriteRules}
\sem is a language for expressing query rewrite rules, and each such
rule needs to hold over all relations (i.e., both sides of each rule
need to return the same relation for all schemas and instances, as
discussed in~\secref{overview}), and likewise for predicates and
expressions. To facilitate that, \sem allows users to declare
meta-variables for queries, predicates, and expressions, and uses two
functions $\CastPred{}{}$ and $\CastExpr{}{}$, as we illustrate next.
First, consider meta-variables. Referring to~\figref{union-slct}, the
base tables $R$ and $S$ are meta-variables that can be quantified over
all possible relations, and $b$ is a meta-variable ranging over
predicates.
For another example, consider the rewrite rule in
Fig.~\ref{fig:magic-distinct}: we want to say that the rule holds for
any relation $R$ with a schema having an attribute $a$. We express
this in \sem by using a meta-variable $p$ instead of the attribute $a$:
\begin{flalign*}
\quad\;
& \texttt{DISTINCT} \; \SelectFrom{\texttt{Left}.p}{R, R} \\
& \texttt{WHERE \Var{\texttt{Left}.p} = \Var{\texttt{Right}.p}}\\
& \equiv \ \texttt{DISTINCT} \; \SelectFrom{p}{R}
\end{flalign*}
Since our data model is a tree, it is very convenient to use a
meta-variable $p$ to navigate to any leaf (corresponding to an
attribute), and it also easy to concatenate two schemas using the
$\node{}{}$ constructor; in contrast, a data model based on ordered
lists would make the combination of navigation and concatenation more
difficult.
Second, we explain the functions $\CastPred{}{}$ and $\CastExpr{}{}$
in \sem with another example. Consider the rewrite rule (to be
presented in \secref{basic-rewrite}) for pushing down selection
predicates, written informally as:
\begin{small}
\begin{flalign*}
\quad\;
& \SelectFrom{\texttt{*}}{R, \; (\SelectFromWhere{\texttt{*}}{S}{b})} \quad \equiv &\\
& \SelectFromWhere{\texttt{*}}{R, \; S}{b} &
\end{flalign*}
\end{small}
In this rule, $b$ is a meta-variable that ranges over all
possible Boolean predicates. In standard SQL, the two occurrences of
$b$ are simply identical expressions, but in \sem the second
occurrence is in a environment that consists of the schemas of both
$R$ and $S$.
In \sem, this is done rigorously using the $\CastPred{}$ construct:
\begin{small}
\begin{flalign*}
\quad\;
& \SelectFrom{\texttt{*}}{R, \; (\SelectFromWhere{\texttt{*}}{S}{b})} \quad \equiv &\\
& \SelectFromWhere{\texttt{*}}{R, \; S}{(\CastPred{\texttt{Right}}{b})} &
\end{flalign*}
\end{small}
The expression $\CastPred{\texttt{Right}}{b}$ is function composition: it
applies $\texttt{Right}$ first, to obtain the schema of $R$, then evaluates
the predicate $b$ on the result.
Requiring explicit casts is an important feature of \sem: doing so ensures that
rewrite rules are only applicable in situations where they are valid. In the example
above, the rule is only valid for all predicates $b$ where $b$ refers only to attributes
in $S$, and the cast operation makes that explicit. The {\tt CASTEXPR} construct works similarly
for expressions.
\begin{figure}[t]
\centering
\[
\begin{array}{llll}
q \in \texttt{Query} & ::= & \texttt{Table} \\
& \; \mid & \texttt{SELECT} \; p \; q \\
& \; \mid & \texttt{FROM} \; q_1, \ldots, q_n \\
& \; \mid & q \; \texttt{WHERE} \; p \\
& \; \mid & q_1 \; \texttt{UNION ALL} \; q_2 \\
& \; \mid & q_1 \; \texttt{EXCEPT} \; q_2 \\
& \; \mid & \texttt{DISTINCT} \; q \\
b \in \texttt{Predicate} & ::= & e_1 \; \texttt{=} \; e_2 \\
& \; \mid & \texttt{NOT} \; b
\mid b_1 \; \texttt{AND} \; b_2
\mid b_1 \; \texttt{OR} \; b_2
\mid \texttt{true} \mid \texttt{false} \\
& \; \mid & \texttt{CASTPRED} \; p \; b \\
& \; \mid & \texttt{EXISTS} \; q \\
e \in \texttt{Expression} & ::= & \Var \; p \\
& \; \mid & f(e_1, \ldots, e_n)
\mid agg(q) \\
& \; \mid & \texttt{CASTEXPR} \; p \; e \\
p \in \texttt{Projection} & ::= & \texttt{*} \mid \texttt{Left} \mid \texttt{Right} \mid \texttt{Empty} \\
& \; \mid & \Compose{p_1}{p_2} \\
& \; \mid & \Duplicate{p_1}{p_2} \\
& \; \mid & \Evaluate{e}
\end{array}
\]
\caption{Syntax of \inputLang}
\label{fig:sql-syntax}
\end{figure}
\subsection{UniNomials}
\label{sec:target-lang}
The interpretation of a \inputLang expression is a formal expression over \outputLang,
which is an algebra of univalent types.
\begin{definition}
\outputLang, the algebra of Univalent Types, consists of
$(\mathcal{U}, \textbf{0}, \textbf{1}, +, \times, \negate{\cdot}, \merely{\cdot},
\sum)$, where:
\begin{itemize}
\item $(\mathcal{U}, \textbf{0}, \textbf{1}, +, \times)$ forms a semi-ring, where
$\mathcal{U}$ is the universe of univalent types, $\textbf{0}, \textbf{1}$ are the
empty and singleton types, and $+, \times$ are binary operations
$\mathcal{U} \times \mathcal{U} \rightarrow \mathcal{U}$: $n_1 + n_2$,
is the direct sum, and $n_1 \times n_2$ is the
Cartesian product.
\item $\negate{\cdot}, \merely{\cdot}$ are derived unary operations
$\mathcal{U} \rightarrow \mathcal{U}$, where $(\negate{\textbf{0}}) = \textbf{1}$ and
$(\negate{n}) = \textbf{0}$ when $n\neq 0$, and
$\merely{n}=\negate{(\negate{n})}$.
\item $\sum : (A \rightarrow \mathcal{U}) \rightarrow \mathcal{U}$ is the infinitary
operation: $\sum f$ is the direct sum of the set of types
$\setof{f(a)}{a \in A}$.
\end{itemize}
\end{definition}
Following standard notation~\cite{hottBook}, we say that the homotopy type $n$ inhabits
some universe $\mathcal{U}$. The base cases of $n$ come from the
denotation of HoTT-Relation and equality of two tuples (In HoTT, propositions are squash types, which are {\textbf{0}} or {\textbf{1}} \cite[Ch 1.11]{hottBook}). The denotation of \sem will be shown Section~\ref{sec:denotation}. There are 5 type-theoretic operations on $\mathcal{U}$:
\paragraph{Cartesian product ($\times$)} Cartesian product of univalent types is
analogously the same concept as the Cartesian product of two sets. For
$A, B:\mathcal{U}$, the cardinality of $A \times B$ is the cardinality of $A$ multiplied by
the cardinality of $B$. For example, we denote the cross product of two
HoTT-Relations using the Cartesian product of univalent types:
\[ \denote{R_1 \bowtie R_2} \triangleq \lambda \; \tuple. \; (\denote{R_1} \; \fst{t}) \times (\denote{R_2} \; \snd{t}) \]
The result of $R_1 \bowtie R_2$ is a HoTT-Relation with type
$\texttt{Tuple} \; \sigma_{R_1 \bowtie R_2}$ $\rightarrow \mathcal{U}$.
For every tuple
$t \in R_1 \bowtie R_2$, its cardinality equals to the cardinality of $t$'s left
sub-tuple ($\fst{t}$) in $R_1$ ($\denote{R_1} \; \fst{t}$) multiplied by the
cardinality of $t$'s right sub-tuple ($\snd{t}$) in $R_2$ ($\denote{R_2} \;
\snd{t}$).
\paragraph{Disjoint union ($+$)} Disjoint union on univalent types
is analogously the same
concept as union on two disjoint sets. For $A,B: \mathcal{U}$, the cardinality of $A
+ B$ is the cardinality of $A$ adding the cardinality of $B$. For example,
\texttt{UNION ALL} denotes to $+$:
\[ \denote{\UNIONALL{R_1}{R_2}} \triangleq \lambda \; \tuple. \; (\denote{R_1} \; t) + (\denote{R_2} \; t) \]
In SQL, \texttt{UNION ALL} means bag semantic union of two relations. Thus a
tuple $t \in \UNIONALL{R_1}{R_2}$ has a cardinality of its
cardinality in $R_1$ ($\denote{R_1} \; t$) added by its cardinality in $R_2$
($\denote{R_2} \; t$).
We also denote logical \texttt{OR} of two predicates using $+$. $A+B$ is corresponded
type-theoretic operation of logical OR if $A$ and $B$ are squash types (recall that
squash types are {\textbf{0}} or \textbf{1} ~\cite[Ch 1.11]{hottBook}).
\paragraph{Squash ($\merely{n}$)} Squash is a type-theoretic operation that
truncates a univalent type to {\textbf{0}} or \textbf{1}. For $A:\mathcal{U}$, $\merely{A} =
\textbf{0}$ if $A$'s cardinality is zero and $\merely{A} = \textbf{1}$ otherwise. An
example of using squash types is in denoting \texttt{DISTINCT} (\texttt{DISTINCT}
means removing duplicated tuples in SQL, i.e., converting a bag to a set):
\[ \denote{\DISTINCT{R}} \triangleq \lambda \; \tuple. \; \merely{\denote{R} \; t} \]
For a tuple $t \in \DISTINCT{R}$, its cardinality equals to $1$ if its cardinality in $R$ is non-zero and equals to $0$ otherwise. This is exactly $\merely{\denote{R}\; t}$.
\paragraph{Negation ($\negate{n}$)} If $n$ is a squash type, $\negate{n}$ is the
negation of $n$. We have $\negate{\textbf{0}} = \textbf{1}$ and $\negate{\textbf{1}} = \textbf{0}$.
Negation is used to denote negating a predicate and to denote \texttt{EXCEPT} . For
example, \texttt{EXCEPT} is used to denote negation:
\[ \denote{R_1 \; \texttt{EXCEPT} \; R_2} \triangleq \lambda \; \tuple. \; (\denote{R_1} \; t) \times (\negate{\merely{\denote{R_2} \; t}}) \]
A tuple $t \in R_1 \; \texttt{EXCEPT} \; R_2$ retains its multiplicity
in $R_1$ if its multiplicity in $R_2$ is
not 0 (since if $\denote{R_2} \; t \neq \textbf{0}$, then ${\merely{\denote{R_2} \; t} \rightarrow \textbf{0}} = \textbf{1} $).
\paragraph{Summation ($\sum$)} Given $A:\mathcal{U}$ and $B:A \rightarrow \mathcal{U}$,
\;$\sum_{x:A} B(x)$ is a dependent pair type
$\sum$ is used to denote projection. For example:
\[ \denote{\texttt{SELECT $k$ FROM $R$}} \triangleq
\lambda \; \tuple. \; \sum_{t': \texttt{Tuple} \; \sigma_R}
\merely{\denote{k} \; t' = t} \times \denote{R} \; t' \]
For a tuple $t$ in the result of this projection query, its cardinality is the summation
of the cardinalities of all tuples of schema $\sigma_A$ that also has the
same value on column $k$ with $t$. Here $\merely{\denote{k} \; t' =
t}$ equals to \textbf{1}~if $t$ and $t'$ have same value on $k$,
otherwise it equals to \textbf{0}. Unlike $K$-Relations,
using univalent types allow us to support summation over an infinite domain
and evaluate expressions such as the projection described above.
In general, proving rewrite rules in \outputLang enables us to use
powerful automatic proving techniques such as associative-commutative term
rewriting in
semi-ring structures (recall that $\mathcal{U}$ is a semi-ring) similar to the
\texttt{ring} tactic~\cite{ringtac} and
Nelson-Oppen algorithm on congruence closure~\cite{NelsonO80JACM}. Both
of which mitigate our proof burden.
| 12a4b7741aa6396de72fd6809078014085d21ee3 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
\paragraph{\bf Hard lattice problems.} Lattices are discrete subgroups of $\mathbb{R}^d$. More concretely, given a basis $B = \{\vc{b}_1, \dots, \vc{b}_d\} \subset \mathbb{R}^d$, the lattice $\mathcal{L} = \mathcal{L}(B)$ generated by $B$ is defined as $\mathcal{L}(B) = \left\{\sum_{i=1}^d \lambda_i \vc{b}_i: \lambda_i \in \mathbb{Z}\right\}$. Given a basis of a lattice $\mathcal{L}$, the Shortest Vector Problem (SVP) asks to find a shortest non-zero vector in $\mathcal{L}$ under the Euclidean norm, i.e., a non-zero lattice vector $\vc{s}$ of norm $\|\vc{s}\| = \lambda_1(\mathcal{L}) := \min_{\vc{v} \in \mathcal{L} \setminus \{\vc{0}\}} \|\vc{v}\|$. Given a basis of a lattice and a target vector $\vc{t} \in \mathbb{R}^d$, the Closest Vector Problem (CVP) asks to find a vector $\vc{s} \in \mathcal{L}$ closest to $\vc{t}$ under the Euclidean distance, i.e.\ such that $\|\vc{s} - \vc{t}\| = \min_{\vc{v} \in \mathcal{L}} \|\vc{v} - \vc{t}\|$.
These two hard problems are fundamental in the study of lattice-based cryptography, as the security of these schemes is directly related to the hardness of SVP and CVP in high dimensions. Various other hard lattice problems, such as Learning With Errors (LWE) and the Shortest Integer Solution (SIS) problem are closely related to SVP and CVP, and many reductions between these and other hard lattice problems are known; see e.g.\ \cite[Figure 3.1]{laarhoven12kolkata} or \cite{stephens16} for an overview. These reductions show that being able to solve CVP efficiently implies that almost all other lattice problems can also be solved efficiently in the same dimension, which makes the study of the hardness of CVP even more important for choosing parameters in lattice-based cryptography.
\paragraph{\bf Algorithms for SVP and CVP.} Although SVP and CVP are both central in the study of lattice-based cryptography, algorithms for SVP have received somewhat more attention, including a benchmarking website to compare different algorithms~\cite{svp}. Various SVP methods have been studied which can solve CVP as well, such as enumeration (see e.g.\ \cite{kannan83, fincke85, gama10, micciancio15}), discrete Gaussian sampling~\cite{aggarwal15, aggarwal15b}, constructing the Voronoi cell of the lattice~\cite{agrell02, micciancio10}, and using a tower of sublattices~\cite{becker14}. On the other hand, for the asymptotically fastest method in high dimensions for SVP\footnote{To obtain provable guarantees, sieving algorithms are commonly modified to facilitate a somewhat artificial proof technique, which drastically increases the time complexity beyond e.g.\ the discrete Gaussian sampler and the Voronoi cell algorithm~\cite{ajtai01, nguyen08, pujol09, micciancio10b}. On the other hand, if some natural heuristic assumptions are made to enable analyzing the algorithm's behavior, then sieving clearly outperforms these methods. We focus on heuristic sieving in this paper.}, lattice sieving, it is not known how to solve CVP with similar costs as SVP.
After a series of theoretical works on constructing efficient heuristic sieving algorithms~\cite{nguyen08, micciancio10b, wang11, zhang13, laarhoven15crypto, laarhoven15latincrypt, becker15nns, becker16cp, becker16lsf} as well as practical papers studying how to speed up these algorithms even further~\cite{milde11, schneider11, schneider13, bos14, fitzpatrick14, ishiguro14, mariano14, mariano14b, mariano15, mariano16pdp, mariano16}, the best time complexity for solving SVP currently stands at $2^{0.292d + o(d)}$~\cite{becker16lsf, mariano16}. Although for various other methods the complexities for solving SVP and CVP are similar~\cite{gama10, micciancio10, aggarwal15b}, one can only guess whether the same holds for lattice sieving methods. To date, the best heuristic time complexity for solving CVP in high dimensions stands at $2^{0.377d + o(d)}$, due to Becker--Gama--Joux~\cite{becker14}.
\subsection{Contributions}
In this paper we revisit heuristic lattice sieving algorithms, as well as the recent trend to speed up these algorithms using nearest neighbor searching, and we investigate how these algorithms can be modified to solve CVP and its generalizations. We present two different approaches for solving CVP with sieving, each of which we argue has its own merits.
\paragraph{\bf Adaptive sieving.} In \textit{adaptive sieving}, we adapt the entire sieving algorithm to the problem instance, including the target vector. As the resulting algorithm is tailored specifically to the given CVP instance, this leads to the best asymptotic complexity for solving a single CVP instance out of our two proposed methods: $2^{0.292d + o(d)}$ time and space. This method is very similar to solving SVP with lattice sieving, and leads to equivalent asymptotics on the space and time complexities as for SVP. The corresponding space-time tradeoff is illustrated in Figure~\ref{fig:1}, and equals that of \cite{becker16lsf} for solving SVP.
\paragraph{\bf Non-adaptive sieving.} Our main contribution, \textit{non-adaptive sieving}, takes a different approach, focusing on cases where several CVP instances are to be solved on the same lattice. The goal here is to minimize the costs of computations depending on the target vector, and spend more time on preprocessing the lattice, so that the amortized time complexity per instance is smaller when solving many CVP instances on the same lattice. This is very closely related to the Closest Vector Problem with Preprocessing (CVPP), where the difference is that we allow for exponential-size preprocessed space. Using nearest neighbor techniques with a balanced space-time tradeoff, we show how to solve CVPP with $2^{0.636d + o(d)}$ space and preprocessing, in $2^{0.136d + o(d)}$ time. A continuous tradeoff between the two complexities can be obtained, where in the limit we can solve CVPP with $(1/\varepsilon)^{O(d)}$ space and preprocessing, in $2^{\varepsilon d + o(d)}$ time. This tradeoff is depicted in Figure~\ref{fig:1}.
A potential application of non-adaptive sieving is as a subroutine within enumeration methods. As described in e.g.\ \cite{gama10}, at any given level in the enumeration tree, one is attempting to solve a CVP instance in a lower-dimensional sublattice of $\mathcal{L}$, where the target vector is determined by the path chosen from the root to the current node in the tree. That means that if we can preprocess this sublattice such that the amortized time complexity of solving CVPP is small, then this could speed up processing the bottom part of the enumeration tree. This in turn might help speed up the lattice basis reduction algorithm BKZ~\cite{schnorr87, schnorr94, chen11}, which commonly uses enumeration as its SVP subroutine, and is key in assessing the security of lattice-based schemes. As the preprocessing needs to be performed once, CVPP algorithms with impractically large preprocessing costs may not be useful, but we show that with sieving the preprocessing costs can be quite small.
\begin{figure}[!t]
{\center
\includegraphics{sievingcvp-figure1}}
\caption{Heuristic complexities for solving the Closest Vector Problem (CVP), the Closest Vector Problem with Preprocessing (CVPP), Bounded Distance Decoding with Preprocessing ($\delta$-BDDP), and the Approximate Closest Vector Problem with Preprocessing ($\kappa$-CVPP). The red curve shows CVP complexities of Becker--Gama--Joux~\cite{becker14}. The left blue curve denotes CVP complexities of adaptive sieving. The right blue curve shows exact CVPP complexities using non-adaptive sieving. Purple curves denote relaxations of CVPP corresponding to different parameters $\delta$ (BDD radius) and $\kappa$ (approximation factor). Note that exact CVPP corresponds to $\delta$-BDDP with $\delta = 1$ and to $\kappa$-CVPP with $\kappa = 1$.\label{fig:1}}
\end{figure}
\paragraph{\bf Outline.} The remainder of the paper is organized as follows. In Section~\ref{sec:pre} we describe some preliminaries, such as sieving algorithms and a useful result on nearest neighbor searching. Section~\ref{sec:ad} describes adaptive sieving and its analysis for solving CVP without preprocessing. Section~\ref{sec:non} describes the preprocessing approach to solving CVP, with complexity analyses for exact CVP and some of its relaxations.
\section{Preliminaries}
\label{sec:pre}
\subsection{Lattice sieving for solving SVP}
Heuristic lattice sieving algorithms for solving the shortest vector problem all use the following basic property of lattices: if $\vc{v}, \vc{w} \in \mathcal{L}$, then their sum/difference $\vc{v} \pm \vc{w} \in \mathcal{L}$ is a lattice vector as well. Therefore, if we have a long list $L$ of lattice vectors stored in memory, we can consider combinations of these vectors to obtain new, shorter lattice vectors. To make sure the algorithm makes progress in finding shorter lattice vectors, $L$ needs to contain a lot of lattice vectors; for vectors $\vc{v}, \vc{w} \in \mathcal{L}$ of similar norm, the vector $\vc{v} - \vc{w}$ is shorter than $\vc{v}, \vc{w}$ iff the angle between $\vc{v}, \vc{w}$ is smaller than $\pi/3$, which for random vectors $\vc{v}, \vc{w}$ occurs with probability $(3/4)^{d/2 + o(d)}$. The expected space complexity of heuristic sieving algorithms follows directly from this observation: if we draw $(4/3)^{d/2 + o(d)}$ random vectors from the unit sphere, we expect a large number of pairs of vectors to have angle less than $\pi/3$, leading to many short difference vectors. This is exactly the heuristic assumption used in analyzing these sieving algorithms: when normalized, vectors in $L$ follow the same distribution as vectors sampled uniformly at random from the unit sphere.
\begin{heuristic} \label{heur:1}
When normalized, the list vectors $\vc{w} \in L$ behave as i.i.d.\ uniformly distributed random vectors from the unit sphere $\mathcal{S}^{d-1} := \{\vc{x} \in \mathbb{R}^d: \|\vc{x}\| = 1\}$.
\end{heuristic}
Therefore, if we start by sampling a list $L$ of $(4/3)^{d/2 + o(d)}$ long lattice vectors, and iteratively consider combinations of vectors in $L$ to find shorter vectors, we expect to keep making progress. Note that naively, combining pairs of vectors in a list of size $(4/3)^{d/2 + o(d)} \approx 2^{0.208d + o(d)}$ takes time $(4/3)^{d + o(d)} \approx 2^{0.415d + o(d)}$.
\paragraph{\bf The Nguyen-Vidick sieve.} The heuristic sieve algorithm of Nguyen and Vidick~\cite{nguyen08} starts by sampling a list $L$ of $(4/3)^{d/2 + o(d)}$ long lattice vectors, and uses a \textit{sieve} to map $L$, with maximum norm $R := \max_{\vc{v} \in L} \|\vc{v}\|$, to a new list $L'$, with maximum norm at most $\gamma R$ for $\gamma < 1$ close to $1$. By repeatedly applying this sieve, after $\operatorname{poly}(d)$ iterations we expect to find a long list of lattice vectors of norm at most $\gamma^{\operatorname{poly}(d)} R = O(\lambda_1(\mathcal{L}))$. The final list is then expected to contain a shortest vector of the lattice. Algorithm~\ref{alg:nv} in Appendix~\ref{app:alg} describes a sieve equivalent to Nguyen-Vidick's original sieve, to map $L$ to $L'$ in $|L|^2$ time.
\paragraph{\bf Micciancio and Voulgaris' GaussSieve.} Micciancio and Voulgaris used a slightly different approach in the GaussSieve~\cite{micciancio10b}. This algorithm reduces the memory usage by immediately \textit{reducing} all pairs of lattice vectors that are sampled. The algorithm uses a single list $L$, which is always kept in a state where for all $\vc{w}_1, \vc{w}_2 \in L$, $\|\vc{w}_1 \pm \vc{w}_2\| \geq \|\vc{w}_1\|, \|\vc{w}_2\|$, and each time a new vector $\vc{v} \in \mathcal{L}$ is sampled, its norm is reduced with vectors in $L$. After the norm can no longer be reduced, the vectors in $L$ are reduced with $\vc{v}$. Modified list vectors are added to a stack to be processed later (to maintain the pairwise reduction-property of $L$), and new vectors which are pairwise reduced with $L$ are added to $L$. Immediately reducing all pairs of vectors means that the algorithm uses less time and memory in practice, but at the same time Nguyen and Vidick's heuristic proof technique does not apply here. However, it is commonly believed that the same bounds $(4/3)^{d/2 + o(d)}$ and $(4/3)^{d + o(d)}$ on the space and time complexities hold for the GaussSieve. Pseudocode of the GaussSieve is given in Algorithm~\ref{alg:gauss} in Appendix~\ref{app:alg}.
\subsection{Nearest neighbor searching}
Given a data set $L \subset \mathbb{R}^d$, the nearest neighbor problem asks to preprocess $L$ such that, when given a query $\vc{t} \in \mathbb{R}^d$, one can quickly return a nearest neighbor $\vc{s} \in L$ with distance $\|\vc{s} - \vc{t}\| = \min_{\vc{w} \in L} \|\vc{w} - \vc{t}\|$. This problem is essentially identical to CVP, except that $L$ is a finite set of unstructured points, rather than the infinite set of all points in a lattice $\mathcal{L}$.
\paragraph{\bf Locality-Sensitive Hashing/Filtering (LSH/LSF).} A celebrated technique for finding nearest neighbors in high dimensions is Locality-Sensitive Hashing (LSH)~\cite{indyk98, wang14}, where the idea is to construct many random partitions of the space, and store the list $L$ in hash tables with buckets corresponding to regions. Preprocessing then consists of constructing these hash tables, while a query $\vc{t}$ is answered by doing a lookup in each of the hash tables, and searching for a nearest neighbor in these buckets. More details on LSH in combination with sieving can be found in e.g.\ \cite{laarhoven15crypto, laarhoven15latincrypt, becker15nns, becker16cp}.
Similar to LSH, Locality-Sensitive Filtering (LSF)~\cite{becker16lsf, laarhoven15nns} divides the space into regions, with the added relaxation that these regions do not have to form a partition; regions may overlap, and part of the space may not be covered by any region. This leads to improved results compared to LSH when $L$ has size exponential in $d$~\cite{becker16lsf, laarhoven15nns}. Below we restate one of the main results of~\cite{laarhoven15nns} for our applications. The specific problem considered here is: given a data set $L \subset \mathcal{S}^{d-1}$ sampled uniformly at random, and a random query $\vc{t} \in \mathcal{S}^{d-1}$, return a vector $\vc{w} \in L$ such that the angle between $\vc{w}$ and $\vc{t}$ is at most $\theta$. The following result further assumes that the list $L$ contains $n = (1 / \sin \theta)^{d + o(d)}$ vectors.
\begin{lemma} \label{lem:nns} \cite[Corollary 1]{laarhoven15nns}
Let $\theta \in (0, \frac{1}{2} \pi)$, and let $u \in [\cos \theta, 1/\cos \theta]$. Let $L \subset \mathcal{S}^{d-1}$ be a list of $n = (1 / \sin \theta)^{d + o(d)}$ vectors sampled uniformly at random from $\mathcal{S}^{d-1}$. Then, using spherical LSF with parameters $\alpha_{\mathrm{q}} = u \cos \theta$ and $\alpha_{\mathrm{u}} = \cos \theta$, one can preprocess $L$ in time $n^{1 + \rho_{\mathrm{u}} + o(1)}$, using $n^{1 + \rho_{\mathrm{u}} + o(1)}$ space, and with high probability answer a random query $\vc{t} \in \mathcal{S}^{d-1}$ correctly in time $n^{\rho_{\mathrm{q}} + o(1)}$, where:
\begin{align}
n^{\rho_{\mathrm{q}}} &= \left(\frac{\sin^2 \theta \, (u \cos \theta + 1)}{u \cos \theta - \cos 2 \theta}\right)^{d/2}, \quad n^{\rho_{\mathrm{u}}} = \left(\frac{\sin^2 \theta}{1 - \cot^2 \theta\left(u^2 - 2 u \cos \theta + 1\right)}\right)^{d/2}. \label{eq:main3}
\end{align}
\end{lemma}
Applying this result to sieving for solving SVP, where $n = \sin(\frac{\pi}{3})^{-d + o(d)}$ and we are looking for pairs of vectors at angle at most $\frac{\pi}{3}$ to perform reductions, this leads to a space and preprocessing complexity of $n^{0.292d + o(d)}$, and a query complexity of $2^{0.084d + o(d)}$. As the preprocessing in sieving is only performed once, and queries are performed $n \approx 2^{0.208d + o(d)}$ times, this leads to a reduction of the complexities of sieving (for SVP) from $2^{0.208d + o(d)}$ space and $2^{0.415d + o(d)}$ time, to $2^{0.292d + o(d)}$ space and time~\cite{becker16lsf}.
\section{Adaptive sieving for CVP}
\label{sec:ad}
We present two methods for solving CVP using sieving, the first of which we call \textit{adaptive sieving} -- we adapt the entire sieving algorithm to the particular CVP instance, to obtain the best overall time complexity for solving one instance. When solving several CVP instances, the costs roughly scale linearly with the number of instances.
\subsubsection{Using one list.} The main idea behind this method is to translate the SVP algorithm by the target vector $\vc{t}$; instead of generating a long list of lattice vectors reasonably close to $\vc{0}$, we generate a list of lattice vectors close to $\vc{t}$, and combine lattice vectors to find lattice vectors even closer vectors to $\vc{t}$. The final list then hopefully contains a closest vector to $\vc{t}$.
One quickly sees that this does not work, as the fundamental property of lattices does not hold for the lattice coset $\vc{t} + \mathcal{L}$: if $\vc{w}_1, \vc{w}_2 \in \vc{t} + \mathcal{L}$, then $\vc{w}_1 \pm \vc{w}_2 \notin \vc{t} + \mathcal{L}$. In other words, two lattice vectors close to $\vc{t}$ can only be combined to form lattice vectors close to $\vc{0}$ or $2 \vc{t}$. So if we start with a list of vectors close to $\vc{t}$, and combine vectors in this list as in the Nguyen-Vidick sieve, then after one iteration we will end up with a list $L'$ of lattice vectors close to $\vc{0}$.
\subsubsection{Using two lists.} To make the idea of translating the whole problem by $\vc{t}$ work for the Nguyen-Vidick sieve, we make the following modification: we keep track of two lists $L = L_{\vc{0}}$ and $L_{\vc{t}}$ of lattice vectors close to $\vc{0}$ and $\vc{t}$, and construct a sieve which maps two input lists $L_{\vc{0}}, L_{\vc{t}}$ to two output lists $L_{\vc{0}}', L_{\vc{t}}'$ of lattice vectors slightly closer to $\vc{0}$ and $\vc{t}$. Similar to the original Nguyen-Vidick sieve, we then apply this sieve several times to two initial lists $(L_{\vc{0}}, L_{\vc{t}})$ with a large radius $R$, to end up with two lists $L_{\vc{0}}$ and $L_{\vc{t}}$ of lattice vectors at distance at most approximately $\sqrt{4/3} \cdot \lambda_1(\mathcal{L})$ from $\vc{0}$ and $\vc{t}$\footnote{Observe that by the Gaussian heuristic, there are $(4/3)^{d/2 + o(d)}$ vectors in $\mathcal{L}$ within any ball of radius $\sqrt{4/3} \cdot \lambda_1(\mathcal{L})$. So the list size of the NV-sieve will surely decrease below $(4/3)^{d/2}$ when $R < \sqrt{4/3} \cdot \lambda_1(\mathcal{L})$.}. The argumentation that this algorithm works is almost identical to that for solving SVP, where we now make the following slightly different heuristic assumption.
\begin{heuristic} \label{heur:2}
When normalized, the list vectors $L_{\vc{0}}$ and $L_{\vc{t}}$ in the modified Nguyen-Vidick sieve both behave as i.i.d.\ uniformly distributed random vectors from the unit sphere.
\end{heuristic}
The resulting algorithm, based on the Nguyen-Vidick sieve, is presented in Algorithm~\ref{alg:nv-adaptive}.
\begin{algorithm}[!t]
\caption{The adaptive Nguyen-Vidick sieve for finding closest vectors}
\label{alg:nv-adaptive}
\begin{algorithmic}[1]
\Require Lists $L_{\vc{0}}, L_{\vc{t}} \subset \mathcal{L}$ containing $(4/3)^{d/2 + o(d)}$ vectors at distance $\leq R$ from $\vc{0}, \vc{t}$
\Ensure Lists $L_{\vc{0}}', L_{\vc{t}}' \subset \mathcal{L}$ contain $(4/3)^{d/2 + o(d)}$ vectors at distance $\leq \gamma R$ from $\vc{0}, \vc{t}$
\State Initialize empty lists $L_{\vc{0}}', L_{\vc{t}}'$
\For{\textbf{each} $(\vc{w}_1, \vc{w}_2) \in L_{\vc{0}} \times L_{\vc{0}}$}
\If{$\|\vc{w}_1 - \vc{w}_2\| \leq \gamma R$}
\State Add $\vc{w}_1 - \vc{w}_2$ to the list $L_{\vc{0}}'$
\EndIf
\EndFor
\For{\textbf{each} $(\vc{w}_1, \vc{w}_2) \in L_{\vc{t}} \times L_{\vc{0}}$}
\If{$\|(\vc{w}_1 - \vc{w}_2) - \vc{t}\| \leq \gamma R$}
\State Add $\vc{w}_1 - \vc{w}_2$ to the list $L_{\vc{t}}'$
\EndIf
\EndFor
\State \Return $(L_{\vc{0}}', L_{\vc{t}}')$
\end{algorithmic}
\end{algorithm}
\subsubsection{Main result.} As the (heuristic) correctness of this algorithm follows directly from the correctness of the original NV-sieve, and nearest neighbor techniques can be applied to this algorithm in similar fashion as well, we immediately obtain the following result. Note that space-time tradeoffs for SVP, such as the one illustrated in \cite[Figure 1]{becker16lsf}, similarly carry over to solving CVP, and the best tradeoff for SVP (and therefore CVP) is depicted in Figure~\ref{fig:1}.
\begin{theorem}
Assuming Heuristic~\ref{heur:2} holds, the adaptive Nguyen-Vidick sieve with spherical LSF solves CVP in time $\mathrm{T}$ and space $\mathrm{S}$, with
\begin{align}
\mathrm{S} = (4/3)^{d/2 + o(d)} \approx 2^{0.208 d + o(d)}, \quad \mathrm{T} = (3/2)^{d/2 + o(d)} \approx 2^{0.292 d + o(d)}.
\end{align}
\end{theorem}
An important open question is whether these techniques can also be applied to the faster GaussSieve algorithm to solve CVP. The GaussSieve seems to make even more use of the property that the sum/difference of two lattice vectors is also in the lattice, and operations in the GaussSieve in $\mathcal{L}$ cannot as easily be \textit{mimicked} for the coset $\vc{t} + \mathcal{L}$. Solving CVP with the GaussSieve with similar complexities is left as an open problem.
\section{Non-adaptive sieving for CVPP}
\label{sec:non}
Our second method for finding closest vectors with heuristic lattice sieving follows a slightly different approach. Instead of focusing only on the total time complexity for one problem instance, we split the algorithm into two phases:
\begin{itemize}
\item Phase 1: Preprocess the lattice $\mathcal{L}$, without knowledge of the target $\vc{t}$;
\item Phase 2: Process the query $\vc{t}$ and output a closest lattice vector $\vc{s} \in \mathcal{L}$ to $\vc{t}$.
\end{itemize}
Intuitively it may be more important to keep the costs of Phase 2 small, as the preprocessed data can potentially be reused later for other instances on the same lattice. This approach is essentially equivalent to the Closest Vector Problem with Preprocessing (CVPP): preprocess $\mathcal{L}$ such that when given a target vector $\vc{t}$ later, one can quickly return a closest vector $\vc{s} \in \mathcal{L}$ to $\vc{t}$. For CVPP however the preprocessed space is usually restricted to be of polynomial size, and the time used for preprocessing the lattice is often not taken into account. Here we will keep track of the preprocessing costs as well, and we do not restrict the output from the preprocessing phase to be of size $\operatorname{poly}(d)$.
\subsubsection{Algorithm description.} To minimize the costs of answering a query, and to do the preprocessing independently of the target vector, we first run a standard SVP sieve, resulting in a large list $L$ of almost all short lattice vectors. Then, after we are given the target vector $\vc{t}$, we use $L$ to reduce the target. Finally, once the resulting vector $\vc{t}' \in \vc{t} + \mathcal{L}$ can no longer be reduced with our list, we hope that this reduced vector $\vc{t}'$ is the shortest vector in the coset $\vc{t} + \mathcal{L}$, so that $\vc{0}$ is the closest lattice vector to $\vc{t}'$ and $\vc{s} = \vc{t} - \vc{t}'$ is the closest lattice vector to $\vc{t}$.
The first phase of this algorithm consists in running a sieve and storing the resulting list in memory (potentially in a nearest neighbor data structure for faster lookups). For this phase either the Nguyen-Vidick sieve or the GaussSieve can be used. The second phase is the same for either method, and is described in Algorithm~\ref{alg:nonadaptive} for the general case of an input list essentially consisting of the $\alpha^{d + o(d)}$ shortest vectors in the lattice. Note that a standard SVP sieve would produce a list of size $(4/3)^{d/2 + o(d)}$ corresponding to $\alpha = \sqrt{4/3}$.
\begin{algorithm}[!t]
\caption{Non-adaptive sieving (Phase 2) for finding closest vectors}
\label{alg:nonadaptive}
\begin{algorithmic}[1]
\Require A list $L \subset \mathcal{L}$ of $\alpha^{d/2 + o(d)}$ vectors of norm at most $\alpha \cdot \lambda_1(\mathcal{L})$, and $\vc{t} \in \mathbb{R}^d$
\Ensure The output vector $\vc{s}$ is the closest lattice vector to $\vc{t}$ (w.h.p.)
\State Initialize $\vc{t}' \leftarrow \vc{t}$
\For{\textbf{each} $\vc{w} \in L$}
\If{$\|\vc{t}' - \vc{w}\| \leq \|\vc{t}'\|$}
\State Replace $\vc{t}' \leftarrow \vc{t}' - \vc{w}$ and restart the \textbf{for}-loop
\EndIf
\EndFor
\State \Return $\vc{s} = \vc{t} - \vc{t}'$
\end{algorithmic}
\end{algorithm}
\subsubsection{List size.} We first study how large $L$ must be to guarantee that the algorithm succeeds. One might wonder why we do not fix $\alpha = \sqrt{4/3}$ immediately in Algorithm~\ref{alg:nonadaptive}. To see why this choice of $\alpha$ does not suffice, suppose we have a vector $\vc{t}' \in \vc{t} + \mathcal{L}$ which is no longer reducible with $L$. This implies that $\vc{t}'$ has norm approximately $\sqrt{4/3} \cdot \lambda_1(\mathcal{L})$, similar to what happens in the GaussSieve. Now, unfortunately the fact that $\vc{t}'$ cannot be reduced with $L$ anymore, does \textit{not} imply that the closest lattice point to $\vc{t}'$ is $\vc{0}$. In fact, it is more likely that there exists an $\vc{s} \in \vc{t} + \mathcal{L}$ of norm slightly more than $\sqrt{4/3} \cdot \lambda_1(\mathcal{L})$ which is closer to $\vc{t}'$, but which is not used for reductions.
By the Gaussian heuristic, we expect the distance from $\vc{t}$ and $\vc{t}'$ to the lattice to be $\lambda_1(\mathcal{L})$. So to guarantee that $\vc{0}$ is the closest lattice vector to the reduced vector $\vc{t}'$, we need $\vc{t}'$ to have norm at most $\lambda_1(\mathcal{L})$. To analyze and prove correctness of this algorithm, we will therefore prove that, under the assumption that the input is a list of the $\alpha^{d + o(d)}$ shortest lattice vectors of norm at most $\alpha \cdot \lambda_1(\mathcal{L})$ for a particular choice of $\alpha$, w.h.p.\ the algorithm reduces $\vc{t}$ to a vector $\vc{t}' \in \vc{t} + \mathcal{L}$ of norm at most $\lambda_1(\mathcal{L})$.
To study how to set $\alpha$, we start with the following elementary lemma regarding the probability of reduction between two uniformly random vectors with given norms.
\begin{lemma} \label{lem:1}
Let $v, w > 0$ and let $\vc{v} = v \cdot \vc{e}_v$ and $\vc{w} = w \cdot \vc{e}_w$. Then:
\begin{align}
\mathbb{P}_{\vc{e}_v, \vc{e}_w \sim \mathcal{S}^{d-1}}\Big(\|\vc{v} - \vc{w}\|^2 < \|\vc{v}\|^2\Big) \sim \left[1 - \left(\tfrac{w}{2v}\right)^2\right]^{d/2 + o(d)}.
\end{align}
\end{lemma}
\begin{proof}
Expanding $\|\vc{v} - \vc{w}\|^2 = v^2 + w^2 - 2 v w \ip{\vc{e}_v}{\vc{e}_w}$ and $\|\vc{v}\|^2 = v^2$, the condition $\|\vc{v} - \vc{w}\|^2 < \|\vc{v}\|^2$ equals $\frac{w}{2v} < \ip{\vc{e}_v}{\vc{e}_w}$. The result follows from \cite[Lemma 2.1]{becker16lsf}.
\end{proof}
Under Heuristic~\ref{heur:1}, we then obtain a relation between the choice of $\alpha$ for the input list and the expected norm of the reduced vector $\vc{t}'$ as follows.
\begin{lemma} \label{lem:2}
Let $L \subset \alpha \cdot \mathcal{S}^{d-1}$ be a list of $\alpha^{d + o(d)}$ uniformly random vectors of norm $\alpha > 1$, and let $\vc{v} \in \beta \cdot \mathcal{S}^{d-1}$ be sampled uniformly at random. Then, for high dimensions $d$, there exists a $\vc{w} \in L$ such that $\|\vc{v} - \vc{w}\| < \|\vc{v}\|$ if and only if
\begin{align}
\alpha^4 - 4 \beta^2 \alpha^2 + 4\beta^2
< 0. \label{eq:a}
\end{align}
\end{lemma}
\begin{proof}
By Lemma~\ref{lem:1} we can reduce $\vc{v}$ with $\vc{w} \in L$ with probability similar to $p = [1 - \frac{\alpha^2}{4\beta^2}]^{d/2 + o(d)}$. Since we have $n = \alpha^{d + o(d)}$ such vectors $\vc{w}$, the probability that none of them can reduce $\vc{v}$ is $(1 - p)^n$, which is $o(1)$ if $n \gg 1/p$ and $1 - o(1)$ if $n \ll 1/p$. Expanding $n \cdot p$, we obtain the given equation~\eqref{eq:a}, where $\alpha^4 - 4 \beta^2 \alpha^2 + 4 \beta^2 < 0$ implies $n \gg 1/p$.
\end{proof}
Note that in our applications, we do not just have a list of $\alpha^{d + o(d)}$ lattice vectors of norm $\alpha \cdot \lambda_1(\mathcal{L})$; for any $\alpha_0 \in [1, \alpha]$ we expect $L$ to contain $\alpha_0^{d + o(d)}$ lattice vectors of norm at most $\alpha_0 \cdot \lambda_1(\mathcal{L})$. To obtain a reduced vector $\vc{t}'$ of norm $\beta \cdot \lambda_1(\mathcal{L})$, we therefore obtain the condition that for \textit{some} value $\alpha_0 \in [1, \alpha]$, it must hold that $\alpha_0^4 - 4 \beta^2 \alpha_0^2 + 4\beta_0^2 < 0$.
From~\eqref{eq:a} it follows that $p(\alpha^2) = \alpha^4 - 4 \beta^2 \alpha^2 + 4\beta^2$ has two roots $r_1 < 2 < r_2$ for $\alpha^2$, which lie close to $2$ for $\beta \approx 1$. The condition that $p(\alpha_0^2) < 0$ for some $\alpha_0 \leq \alpha$ is equivalent to $\alpha > r_1$, which for $\beta = 1 + o(1)$ implies that $\alpha^2 \geq 2 + o(1)$. This means that asymptotically we must set $\alpha = \sqrt{2}$, and use $n = 2^{d/2 + o(d)}$ input vectors, to guarantee that w.h.p.\ the algorithm succeeds. A sketch of the situation is also given in Figure~\ref{fig:2a}.
\begin{figure}[!t]
\subfloat[For solving \textbf{exact CVP}, we must reduce the vector $\vc{t}$ to a vector $\vc{t}' \in \vc{t} + \mathcal{L}$ of norm at most $\lambda_1(\mathcal{L})$. The nearest lattice point to $\vc{t}'$ lies in a ball of radius approximately $\lambda_1(\mathcal{L})$ around $\vc{t}'$ (blue), and almost all the mass of this ball is contained in the (black) ball around $\vc{0}$ of radius $\sqrt{2} \cdot \lambda_1(\mathcal{L})$. So if $\vc{s} \in \mathcal{L} \setminus \{\vc{0}\}$ had lain closer to $\vc{t}'$ than $\vc{0}$, we would have reduced $\vc{t}'$ with $\vc{s}$, since $\vc{s} \in L$.\label{fig:2a}]{%
\includegraphics{sievingcvp-figure2a}}%
\hfill
\subfloat[For \textbf{variants of CVP}, a choice $\alpha$ for the list size implies a norm $\beta \cdot \lambda_1(\mathcal{L})$ of $\vc{t}'$. The nearest lattice vector $\vc{s}$ to $\vc{t}'$ lies within $\delta \cdot \lambda_1(\mathcal{L})$ of $\vc{t}'$ ($\delta = 1$ for approx-CVP), so with high probability $\vc{s}$ has norm approximately $(\sqrt{\beta^2 + \delta^2}) \cdot \lambda_1(\mathcal{L})$. For $\delta$-BDD, if $\sqrt{\beta^2 + \delta^2} \leq \alpha$ then we expect the nearest point $\vc{s}$ to be in the list $L$. For $\kappa$-CVP, if $\beta \leq \kappa$, then the lattice vector $\vc{t} - \vc{t}'$ has norm at most $\kappa \cdot \lambda_1(\mathcal{L})$.\label{fig:2b}]{
\includegraphics{sievingcvp-figure2b}}%
\caption{Comparison of the list size complexity analysis for CVP (left) and BDD/approximate CVP (right). The point $\vc{t}$ represents the target vector, and after a series of reductions with Algorithm~\ref{alg:nonadaptive}, we obtain $\vc{t}' \in \vc{t} + \mathcal{L}$. Blue balls around $\vc{t}'$ depict regions in which we expect the closest lattice point to $\vc{t}'$ to lie, where the blue shaded area indicates a negligible fraction of this ball~\cite[Lemma 2]{becker16lsf}.\label{fig:2}}
\end{figure}
\subsubsection{Modifying the first phase.} As we will need a larger list of size $2^{d/2 + o(d)}$ to make sure we can solve CVP exactly, we need to adjust Phase 1 of the algorithm as well. Recall that with standard sieving, we reduce vectors iff their angle is at most $\theta = \frac{\pi}{3}$, resulting in a list of size $(\sin \theta)^{-d + o(d)}$. As we now need the output list of the first phase to consist of $2^{d/2 + o(d)} = (\sin \theta')^{-d + o(d)}$ vectors for $\theta' = \frac{\pi}{4}$, we make the following adjustment: only reduce $\vc{v}$ and $\vc{w}$ if their common angle is less than $\frac{\pi}{4}$. For unit length vectors, this condition is equivalent to reducing $\vc{v}$ with $\vc{w}$ iff $\|\vc{v} - \vc{w}\|^2 \leq (2 - \sqrt{2}) \cdot \|\vc{v}\|^2$. This further accelerates nearest neighbor techniques due to the smaller angle $\theta$. Pseudocode for the modified first phase is given in Appendix~\ref{app:alg2}
\subsubsection{Main result.} With the algorithm in place, let us now analyze its complexity for solving CVP. The first phase of the algorithm generates a list of size $2^{d/2 + o(d)}$ by combining pairs of vectors, and naively this can be done in time $\mathrm{T}_1 = 2^{d + o(d)}$ and space $\mathrm{S} = 2^{d/2 + o(d)}$, with query complexity $\mathrm{T}_2 = 2^{d/2 + o(d)}$. Using nearest neighbor searching (Lemma~\ref{lem:nns}), the query and preprocessing complexities can be further reduced, leading to the following result.
\begin{theorem} \label{thm:2}
Let $u \in (\frac{1}{2} \sqrt{2}, \sqrt{2})$. Using non-adaptive sieving, we can solve CVP with preprocessing time $\mathrm{T}_1$, space complexity $\mathrm{S}$, and query time complexity $\mathrm{T}_2$ as follows:
\begin{align}
\mathrm{S} = \mathrm{T}_1 &= \left(\frac{1}{u (\sqrt{2} - u)}\right)^{d/2 + o(d)}, \qquad \mathrm{T}_2 = \left(\frac{\sqrt{2} + u}{2 u}\right)^{d/2 + o(d)}.
\end{align}
\end{theorem}
\begin{proof}
These complexities follow from Lemma~\ref{lem:nns} with $\theta = \frac{\pi}{4}$, noting that the first phase can be performed in time and space $\mathrm{T}_1 = \mathrm{S} = n^{1 + \rho_{\mathrm{u}}}$, and the second phase in time $\mathrm{T}_2 = n^{\rho_{\mathrm{q}}}$.
\end{proof}
To illustrate the time and space complexities of Theorem~\ref{thm:2}, we highlight three special cases $u$ as follows. The full tradeoff curve for $u \in (\frac{1}{2} \sqrt{2}, \sqrt{2})$ is depicted in Figure~\ref{fig:1}.
\begin{itemize}
\item Setting $u = \frac{1}{2} \sqrt{2}$, we obtain $\mathrm{S} = \mathrm{T}_1 = 2^{d/2 + o(d)}$ and $\mathrm{T}_2
\approx 2^{0.2925d + o(d)}$.
\item Setting $u = 1$, we obtain $\mathrm{S} = \mathrm{T}_1
\approx 2^{0.6358 d + o(d)}$ and $\mathrm{T}_2
\approx 2^{0.1358 d + o(d)}$.
\item Setting $u = \frac{1}{2}(\sqrt{2} + 1)$, we get $\mathrm{S} = \mathrm{T}_1 = 2^{d + o(d)}$ and $\mathrm{T}_2
\approx 2^{0.0594 d + o(d)}$.
\end{itemize}
The first result shows that the query complexity of non-adaptive sieving is never worse than for adaptive sieving; only the space and preprocessing complexities are worse. The second and third results show that CVP can be solved in significantly less time, even with preprocessing and space complexities bounded by $2^{d + o(d)}$.
\paragraph{\bf Minimizing the query complexity.} As $u \to \sqrt{2}$, the query complexity keeps decreasing while the memory and preprocessing costs increase. For arbitrary $\varepsilon > 0$, we can set $u = u_\varepsilon \approx \sqrt{2}$ as a function of $\varepsilon$, resulting in asymptotic complexities $\mathrm{S} = \mathrm{T}_1 = (1/\varepsilon)^{O(d)}$ and $\mathrm{T}_2 = 2^{\varepsilon d + o(d)}$. This shows that it is possible to obtain a slightly subexponential query complexity, at the cost of superexponential space, by taking $\varepsilon = o(1)$ as a function of $d$.
\begin{corollary} \label{thm:3}
For arbitrary $\varepsilon > 0$, using non-adaptive sieving we can solve CVPP with preprocessing time and space complexities $(1/\varepsilon)^{O(d)}$, in time $2^{\varepsilon d + o(d)}$. In particular, we can solve CVPP in $2^{o(d)}$ time, using $2^{\omega(d)}$ space and preprocessing.
\end{corollary}
Being able to solve CVPP in subexponential time with superexponential preprocessing and memory is neither trivial nor quite surprising. A naive approach to the problem, with this much memory, could for instance be to index the entire fundamental domain of $\mathcal{L}$ in a hash table. One could partition this domain into small regions, solve CVP for the centers of each of these regions, and store all the solutions in memory. Then, given a query, one looks up which region $\vc{t}$ is in, and returns the answer corresponding to that vector. With a sufficiently fine-grained partitioning of the fundamental domain, the answers given by the look-ups are accurate, and this algorithm probably also runs in subexponential time.
Although it may not be surprising that it is possible to solve CVPP in subexponential time with (super)exponential space, it is not clear what the complexities of other methods would be. Our method presents a clear tradeoff between the complexities, where the constants in the preprocessing exponent are quite small; for instance, we can solve CVPP in time $2^{0.06d + o(d)}$ with less than $2^{d + o(d)}$ memory, which is the same amount of memory/preprocessing of the best provable SVP and CVP algorithms~\cite{aggarwal15, aggarwal15b}. Indexing the fundamental domain may well require much more memory than this.
\subsection{Bounded Distance Decoding with Preprocessing}
We finally take a look at specific instances of CVP which are easier than the general problem, such as when the target $\vc{t}$ lies unusually close to the lattice. This problem naturally appears in practice, when a private key consists of a \textit{good basis} of a lattice with short basis vectors, and the public key is a \textit{bad basis} of the same lattice. An encryption of a message could then consist of the message being mapped to a lattice point $\vc{v} \in \mathcal{L}$, and a small error vector $\vc{e}$ being added to $\vc{v}$ ($\vc{t} = \vc{v} + \vc{e}$) to hide $\vc{v}$. If the noise $\vc{e}$ is small enough, then with a good basis one can decode $\vc{t}$ to the closest lattice vector $\vc{v}$, while someone with the bad basis cannot decode correctly. As decoding for arbitrary $\vc{t}$ (solving CVP) is known to be hard even with knowledge of a good basis~\cite{micciancio01e, feige02, regev04d, alekhnovich05}, $\vc{e}$ needs to be very short, and $\vc{t}$ must lie unusually close to the lattice.
So instead of assuming target vectors $\vc{t} \in \mathbb{R}^d$ are sampled at random, suppose that $\vc{t}$ lies at distance at most $\delta \cdot \lambda_1(\mathcal{L})$ from $\mathcal{L}$, for $\delta \in (0,1)$. For adaptive sieving, recall that the list size $(4/3)^{d/2 + o(d)}$ is the minimum initial list size one can hope to use to obtain a list of short lattice vectors; with fewer vectors, one would not be able to solve SVP.\footnote{The recent paper \cite{bai16} discusses how to use less memory in sieving, by using triple- or tuple-wise reductions, instead of the standard pairwise reductions. These techniques may also be applied to adaptive sieving to solve CVP with less memory, at the cost of an increase in the time complexity.} For non-adaptive sieving however, it may be possible to reduce the list size below $2^{d/2 + o(d)}$.
\subsubsection{List size.} Let us again assume that the preprocessed list $L$ contains almost all $\alpha^{d + o(d)}$ lattice vectors of norm at most $\alpha \cdot \lambda_1(\mathcal{L})$. The choice of $\alpha$ implies a maximum norm $\beta_{\alpha} \cdot \lambda_1(\mathcal{L})$ of the reduced vector $\vc{t}'$, as described in Lemma~\ref{lem:2}. The nearest lattice vector $\vc{s} \in \mathcal{L}$ to $\vc{t}'$ lies within radius $\delta \cdot \lambda_1(\mathcal{L})$ of $\vc{t}'$, and w.h.p.\ $\vc{s} - \vc{t}'$ is approximately orthogonal to $\vc{t}'$; see Figure~\ref{fig:2b}, where the shaded area is asymptotically negligible. Therefore w.h.p.\ $\vc{s}$ has norm at most $(\sqrt{\beta_{\alpha}^2 + \delta^2}) \cdot \lambda_1(\mathcal{L})$. Now if $\sqrt{\beta_{\alpha}^2 + \delta^2} \leq \alpha$, then we expect the nearest vector to be contained in $L$, so that ultimately $\vc{0}$ is nearest to $\vc{t}'$. Substituting $\alpha^4 - 4 \beta^2 \alpha^2 + 4 \beta^2 = 0$ and $\beta^2 + \delta^2 \leq \alpha^2$, and solving for $\alpha$, this leads to the following condition on $\alpha$.
\begin{align}
\alpha^2 \geq \tfrac{2}{3} (1 + \delta^2) + \tfrac{2}{3} \sqrt{(1 + \delta^2)^2 - 3 \delta^2} \, . \label{eq:a2}
\end{align}
Taking $\delta = 1$, corresponding to exact CVP, leads to the condition $\alpha \geq \sqrt{2}$ as expected, while in the limiting case of $\delta \to 0$ we obtain the condition $\alpha \geq \sqrt{4/3}$. This matches experimental observations using the GaussSieve, where after finding the shortest vector, newly sampled vectors often cause \textit{collisions} (i.e.\ being reduced to the $\vc{0}$-vector). In other words, Algorithm~\ref{alg:nonadaptive} often reduces target vectors $\vc{t}$ which essentially lie \textit{on} the lattice ($\delta \to 0$) to the $\vc{0}$-vector when the list has size $(4/3)^{d/2 + o(d)}$. This explains why collisions in the GaussSieve are common when the list size grows to size $(4/3)^{d/2 + o(d)}$.
\subsubsection{Main result.} To solve BDD with a target $\vc{t}$ at distance $\delta \cdot \lambda_1(\mathcal{L})$ from the lattice, we need the preprocessing to produce a list of almost all $\alpha^{d + o(d)}$ vectors of norm at most $\alpha \cdot \lambda_1(\mathcal{L})$, with $\alpha$ satisfying~\eqref{eq:a2}. Similar to the analysis for CVP, we can produce such a list by only doing reductions between two vectors if their angle is less than $\theta$, where now $\theta = \arcsin(1 / \alpha)$. Combining this with Lemma~\ref{lem:1}, we obtain the following result.
\begin{theorem} \label{thm:BDD}
Let $\alpha$ satisfy \eqref{eq:a2} and let $u \in (\sqrt{\frac{\alpha^2 - 1}{\alpha^2}}, \sqrt{\frac{\alpha^2}{\alpha^2 - 1}})$. Using non-adaptive sieving, we can heuristically solve BDD for targets $\vc{t}$ at distance $\delta \cdot \lambda_1(\mathcal{L})$ from the lattice, with preprocessing time $\mathrm{T}_1$, space complexity $\mathrm{S}$, and query time complexity $\mathrm{T}_2$ as follows:
\begin{align}
& \qquad \mathrm{S} = \left(\frac{1}{1 - (\alpha^2 - 1) (u^2 - \frac{2 u}{\alpha}
\sqrt{\alpha^2 - 1} + 1)}\right)^{d/2 + o(d)}, \\
\mathrm{T}_1 &= \max\left\{\mathrm{S}, \ (3/2)^{d/2 + o(d)}\right\}, \qquad \mathrm{T}_2 = \left(\frac{\alpha + u \sqrt{\alpha^2 - 1}}{2 \alpha - \alpha^3 + \alpha^2 u \sqrt{\alpha^2 - 1}}\right)^{d/2 + o(d)}.
\end{align}
\end{theorem}
\begin{proof}
These complexities directly follow from applying Lemma~\ref{lem:nns} with $\theta = \arcsin(1/\alpha)$, and again observing that Phase 1 can be performed in time $\mathrm{T}_1 = n^{1 + \rho_{\mathrm{u}}}$ and space $\mathrm{S} = n^{1 + \rho_{\mathrm{u}}}$, while Phase 2 takes time $\mathrm{T}_2 = n^{\rho_{\mathrm{q}}}$. Note that we cannot combine vectors whose angles are larger than $\frac{\pi}{3}$ in Phase 1, which leads to a lower bound on the preprocessing time complexity $\mathrm{T}_1$ based on the costs of solving SVP.
\end{proof}
Theorem~\ref{thm:BDD} is a generalization of Theorem~\ref{thm:2}, as the latter can be derived from the former by substituting $\delta = 1$ above. To illustrate the results, Figure~\ref{fig:1} considers two special cases:
\begin{itemize}
\item For $\delta = \frac{1}{2}$, we find $\alpha \approx 1.1976$, leading to $\mathrm{S} \approx 2^{0.2602d + o(d)}$ and $\mathrm{T}_2 = 2^{0.1908d + o(d)}$ when minimizing the space complexity.
\item For $\delta \to 0$, we have $\alpha \to \sqrt{4/3} \approx 1.1547$. The minimum space complexity is therefore $\mathrm{S} = (4/3)^{d/2 + o(d)}$, with query complexity $\mathrm{T}_2 = 2^{0.1610d + o(d)}$.
\end{itemize}
In the limit of $u \to \sqrt{\frac{\alpha^2}{\alpha^2 - 1}}$ we need superexponential space/preprocessing $\mathrm{S}, \mathrm{T}_1 \to 2^{\omega(d)}$ and a subexponential query time $\mathrm{T}_2 \to 2^{o(d)}$ for all $\delta > 0$.
\subsection{Approximate Closest Vector Problem with Preprocessing}
Given a lattice $\mathcal{L}$ and a target vector $\vc{t} \in \mathbb{R}^d$, approximate CVP with approximation factor $\kappa$ asks to find a vector $\vc{s} \in \mathcal{L}$ such that $\|\vc{s} - \vc{t}\|$ is at most a factor $\kappa$ larger than the real distance from $\vc{t}$ to $\mathcal{L}$. For random instances $\vc{t}$, by the Gaussian heuristic this means that a lattice vector counts as a solution iff it lies at distance at most $\kappa \cdot \lambda_1(\mathcal{L})$ from $\vc{t}$.
\subsubsection{List size.} Instead of reducing $\vc{t}$ to a vector $\vc{t}'$ of norm at most $\lambda_1(\mathcal{L})$ as is needed for solving exact CVP, we now want to make sure that the reduced vector $\vc{t}'$ has norm at most $\kappa \cdot \lambda_1(\mathcal{L})$. If this is the case, then the vector $\vc{t} - \vc{t}'$ is a lattice vector lying at distance at most $\kappa \cdot \lambda_1(\mathcal{L})$, which w.h.p.\ qualifies as a solution. This means that instead of substituting $\beta = 1$ in Lemma~\ref{lem:2}, we now substitute $\beta = \kappa$. This leads to the condition that $\alpha_0^4 - 4\kappa^2 \alpha_0^2 + 4 \beta^2 < 0$ for some $\alpha_0 \leq \alpha$. By a similar analysis $\alpha^2$ must therefore be larger than the smallest root $r_1 = 2\kappa (\kappa - \sqrt{\kappa^2 - 1})$ of this quadratic polynomial in $\alpha^2$. This immediately leads to the following condition on $\alpha$:
\begin{align}
\alpha^2 \geq 2 \kappa \left(\kappa - \sqrt{\kappa^2 - 1}\right). \label{eq:a3}
\end{align}
A sanity check shows that $\kappa = 1$, corresponding to exact CVP, indeed results in $\alpha \geq \sqrt{2}$, while in the limit of $\kappa \to \infty$ a value $\alpha \approx 1$ suffices to obtain a vector $\vc{t}'$ of norm at most $\kappa \cdot \lambda_1(\mathcal{L})$. In other words, to solve approximate CVP with very large (constant) approximation factors, a preprocessed list of size $(1 + \varepsilon)^{d + o(d)}$ suffices.
\subsubsection{Main result.} Similar to the analysis of CVPP, we now take $\theta = \arcsin(1/\alpha)$ as the angle with which to reduce vectors in Phase 1, so that the output of Phase 1 is a list of almost all $\alpha^{d + o(d)}$ shortest lattice vectors of norm at most $\alpha \cdot \lambda_1(\mathcal{L})$. Using a smaller angle $\theta$ for reductions again means that nearest neighbor searching can speed up the reductions in both Phase 1 and Phase 2 even further. The exact complexities follow from Lemma~\ref{lem:nns}.
\begin{theorem} \label{thm:aCVP}
Using non-adaptive sieving with spherical LSF, we can heuristically solve $\kappa$-CVP with similar complexities as in Theorem~\ref{thm:BDD}, where now $\alpha$ must satisfy \eqref{eq:a3}.
\end{theorem}
Note that only the dependence of $\alpha$ on $\kappa$ is different, compared to the dependence of $\alpha$ on $\delta$ for bounded distance decoding. The complexities for $\kappa$-CVP arguably decrease \textit{faster} than for $\delta$-BDD: for instance, for $\kappa \approx 1.0882$ we obtain the same complexities as for BDD with $\delta = \frac{1}{2}$, while $\kappa = \sqrt{4/3} \approx 1.1547$ leads to the same complexities as for BDD with $\delta \to 0$. Two further examples are illustrated in Figure~\ref{fig:1}:
\begin{itemize}
\item For $\kappa = 2$, we have $\alpha \approx 1.1976$, which for $u \approx 0.5503$ leads to $\mathrm{S} = \mathrm{T}_1 = 2^{0.2602 d + o(d)}$ and $\mathrm{T}_2 = 2^{0.1908 d + o(d)}$, and for $u = 1$ leads to $\mathrm{S} = \mathrm{T}_1 = 2^{0.3573 d + o(d)}$ and $\mathrm{T}_2 = 2^{0.0971 d + o(d)}$.
\item For $\kappa \to \infty$, we have $\alpha \to 1$, i.e.\ the required preprocessed list size approaches $2^{o(d)}$ as $\kappa$ grows. For sufficiently large $\kappa$, we can solve $\kappa$-CVP with a preprocessed list of size $2^{\varepsilon d + o(d)}$ in at most $2^{\varepsilon d + o(d)}$ time. The preprocessing time is given by $2^{0.2925 d + o(d)}$.
\end{itemize}
The latter result shows that for any superconstant approximation factor $\kappa = \omega(1)$, we can solve the corresponding approximate closest vector problem with preprocessing in subexponential time, with an exponential preprocessing time complexity $2^{0.292d + o(d)}$ for solving SVP and generating a list of short lattice vectors, and a subexponential space complexity required for Phase 2. In other words, even without superexponential preprocessing/memory we can solve CVPP with large approximation factors in subexponential time.
To compare this result with previous work, note that the lower bound on $\alpha$ from \eqref{eq:a3} tends to $1 + 1/(8 \kappa^2) + O(\kappa^{-4})$ as $\kappa$ grows. The query space and time complexities are further both proportional to $\alpha^{\Theta(d)}$. To obtain a polynomial query complexity and polynomial storage after the preprocessing phase, we can solve for $\kappa$, leading to the following result.
\begin{corollary} \label{cor:acvpp-poly}
With non-adaptive sieving we can heuristically solve approximate CVPP with approximation factor $\kappa$ in polynomial time with polynomial-sized advice iff $\kappa = \Omega(\sqrt{d / \log d})$.
\end{corollary}
\begin{proof}
The query time and space complexities are given by $\alpha^{\Theta(d)}$, where $\alpha = 1 + \Theta(1 / \kappa^2)$. To obtain polynomial complexities in $d$, we must have $\alpha^{\Theta(d)} = d^{O(1)}$, or equivalently:
\begin{align}
1 + \Theta\left(\frac{1}{\kappa^2}\right) = \alpha = d^{O(1/d)} = \exp \, O\left(\frac{\log d}{d}\right) = 1 + O\left(\frac{\log d}{d}\right).
\end{align}
Solving for $\kappa$ leads to the given relation between $\kappa$ and $d$.
\end{proof}
Apart from the heuristic assumptions, this approximation factor is equivalent to Aharonov and Regev~\cite{aharonov04}, who showed that the decision version of CVPP with approximation factor $\kappa = \Omega(\sqrt{d / \log d})$ can provably be solved in polynomial time. This further (heuristically) improves upon results of~\cite{lagarias90b, dadush14}, who are able to solve search-CVPP with polynomial time and space complexities for $\kappa = O(d^{3/2})$ and $\kappa = \Omega(d / \sqrt{\log d})$ respectively. Assuming the heuristic assumptions are valid, Corollary~\ref{cor:acvpp-poly} closes the gap between these previous results for decision-CVPP and search-CVPP with a rather simple algorithm: (1) preprocess the lattice by storing all $d^{O(1)}$ shortest vectors of the lattice in a list; and (2) apply Algorithm~\ref{alg:nonadaptive} to this list and the target vector to find an approximate closest vector. Note that nearest neighbor techniques only affect leading constants; even without nearest neighbor searching this would heuristically result in a polynomial time and space algorithm for $\kappa$-CVPP with $\kappa = \Omega(\sqrt{d / \log d})$. An interesting open problem would be to see if this result can be made provable for arbitrary lattices, without any heuristic assumptions.
\section*{Acknowledgments}
The author is indebted to L\'{e}o Ducas, whose initial ideas and suggestions on this topic motivated work on this paper. The author further thanks Vadim Lyubashevsky and Oded Regev for their comments on the relevance of a subexponential time CVPP algorithm requiring (super)exponential space. The author is supported by the SNSF ERC Transfer Grant CRETP2-166734 FELICITY.
\bibliographystyle{alpha}
| fd82d8f9ab6f29ef538142e55ebff69ba418ca10 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Exciton mediated many-body interactions give rise to a host of physical effects \cite{qiu2016screening,sahin2016computing,cudazzo2016exciton}
that determine the opto-electronic properties
of low dimensional transition metal dichalcogenides, MX$_2$
(M = Mo, W, Nb and X = S, Se) \cite{scharf2016probing,rama,ugeda2014giant,wang2012electronics,makatom,qiu2013,
splen,komsa2012effects,kormanyos2013monolayer},
with important consequences for fundamental and applied research.
The confinement of correlated charge carriers or excitons to a narrow region of space in low dimensional transition metal dichalcogenides (TMDCs) leads to
unique photoluminescence properties \cite{splendiani2010emerging,plechinger2012low,gao2016localized,ji2013epitaxial,zhu2016strongly,eda2013two,ghatak2011nature,bergh,molinaspin,mai2013many} that are otherwise absent in the bulk configurations \cite{saigal2016exciton}.
The availability of state-of-the art exfoliation
techniques \cite{novoselov2005two,varrla2015large,chen2015nanoimprint,fan2015fast} enable
fabrication of low dimensional transition metal dichalcogenides that is useful for applications
\cite{ou2014ion,li2016charge,perebeinos2015metal,beck2000,tsai2013few,bernardi2013extraordinary, he2012fabrication,wi2014enhancement,bertolazzi2013nonvolatile,ji2013epitaxial,yu2016evaluation,
park2016mos2,eda2013two,radisavljevic2011integrated,lembke2015single,pospischil2014solar,
zhang2014m,yoon2011good}. The excitonic processes that determine the
performance of TMDC-based electronic devices include
defect assisted scattering and trapping by surface states \cite{shi2013exciton}, decay via exciton-exciton
annihilation \cite{shin2014photoluminescence,ye2014exciton,konabe2014effect}, phonon assisted relaxation \cite{thilrelaxjap}, and capture by mid-gap defects through
Auger processes \cite{wang2015fast}. Excitonic processes that result in the formation of complex trions \cite{mak,berkelbach2013theory,
thiltrion} and electron-hole recombination with generation of hot carriers \cite{kozawa2014photocarrier}
are also of importance in device performances.
Dynamical processes incorporating exciton-phonon interactions underlie
the opto-electronic properties
of monolayer transition metal dichalcogenides \cite{jones2013optical}.
The strength of the interactions between coupled charge carriers and phonons
is deduced from experimental measurements of the dephasing times \cite{nie2014ultrafast},
exciton linewidths \cite{selig2016excitonic},
photoluminescence, \cite{mouri2013tunable} and other parameters such
as the exciton mobility and luminescence rise times.
The exciton formation time is determined by a complicated interplay of various
dynamical processes in the picosecond time scale \cite{siantidis2001dynamics}
and is linked to the efficient operation of optoelectronic devices.
To this end, a comprehensive understanding of how newly
generated electron-hole pairs relax energetically to
form excitons still remain unclear.
Recently decay times of $\approx$ 0.3 ps
of the transient absorption signal subsequent to
the interband excitation of the monolayer WSe$_2$, MoS$_2$, and MoSe$_2$
was recorded in time-resolved measurements \cite{ceballos2016exciton}.
The ultrafast decay times were
deduced as the exciton formation times from electron-hole pairs in monolayer systems.
Motivated by these considerations, we examine a mechanism by which excitons are formed from
an initial state of unbound electron-hole pairs to account
for the observed short exciton formation time \cite{ceballos2016exciton} in
common TMDCs (MoS$_2$, MoSe$_2$, WS$_2$, and WSe$_2$).
While the focus of this paper is on the theoretical aspects of excitonic
interactions, the overall aim is to seek an understanding of the critical
factors that limit the exciton formation time which is of relevance
to experimental investigations involving device applications.
The unequal charges of the basis atoms in polar crystals allow
a moving electron to polarize the electric field of its surrounding medium.
The polarization effects displaces the ions giving rise to lattice vibrations of
a optical phonon frequency in resonance with the polarization field, and enable
direct Fr\"ohlich coupling between phonons and charge carriers.
In this work we consider that the excitons are created via the two-dimensional
Fr\"ohlich interaction which provides a critical pathway by which
charge carriers undergo energy loss to optical phonons at elevated temperatures
in the monolayers MoS$_2$ and other transition-metal dichalcogenides \cite{kaasbjerg2014hot}.
The exciton is a neutral quasiparticle, and polarization effects due to the
longitudinal optical waves may appear to have less influence
than those associated with polarization effects of the individual electron or hole.
In reality the internal state of the exciton undergoes dipole type transitions and there occurs measurable effects due to Fr\"ohlich interactions in constrained systems.
The focus on LO phonons in the exciton formation process in this study is justified by the large strength of excitonic interactions with high frequency phonons that arise due to the strong confinement of the exciton wave-functions in the real space of monolayer systems. Moreover the exciton-phonon
interaction is long ranged due to the existence of polarization effects despite
large separations between charge carriers and the ions in the material
system. The phonon-limited mobility is largely dominated by polar optical scattering
via the Fr\"ohlich interaction at room temperatures \cite{kaasbjerg12}.
Exciton formation may take place via deformation potential coupling
to acoustic phonons \cite{oh2000excitonic,thilagam1993generation}, but is likely to occur
with less efficiency due to the high exciton binding energies
\cite{makatom,chei12,ugeda2014giant,hill2015observation,chei12,komsa2012effects,thiljap}
in monolayer dichalcogenides.
In conventional semiconductors such as the two band GaAs material system, excitons are formed
via the Fr\"ohlich interaction in the picosecond
time range \cite{siantidis2001dynamics,oh2000exciton}. While excitons in
GaAs materials are dominantly formed at the center of
the Brillouin zone center, the formation process occurs at the non-central points in the momentum space
of monolayer TMDCs \cite{jones2013optical}. This gives rise to
quantitative differences in the exciton
creation times between GaAs and TMDCs. For excitation energies higher than the band-gap of monolayer systems,
the electron-hole pair creates an exciton with
a non-zero wavevector associated with its center-of-mass
motion \cite{siantidis2001dynamics,oh2000exciton}. The exciton subsequently relaxes to the
zero wavevector state with emission of acoustic or LO phonons
before undergoing radiative recombination by emitting
a photon. To this end, the formation time of an exciton as a function of exciton wave
vector is useful in analyzing the luminescence rise times that can be
measured experimentally.
In this study we employ the exciton-LO phonon
interaction operator to estimate the exciton formation times in monolayer transition metal dichalcogenides.
The formation time of excitons is determined using
the interaction Hamiltonian which describes the
conversion of the photoexcited free electron-hole pair
to a final exciton state initiated by exciton-phonon Fr\"ohlich interactions,
and accompanied by absorption or emission
of phonons. The dependence of the
exciton formation time on several parameters
such as the temperatures of the crystal lattice, charge carriers and
excitons as well as the densities of charge carriers and excitons will be closely
examined in this study.
\section{Formation of Excitons in monolayer Molybdenum Disulfide \label{basic}}
\subsection{Exciton-LO phonon Hamiltonian}
We project the single monolayer of a hexagonally ordered plane of metal atoms sandwiched between two other hexagon planes of chalcogens onto a quasi two-dimensional space \cite{cho2008,mouri2013}. The motion of the exciton is generally confined to the parallel two-dimensional $XY$ layers of the atomic planes with restricted electron and hole motion in the $z$ direction perpendicular to the monolayer plane.
The monolayer MoS$_2$ has nine
phonon branches consisting of three acoustic and six optical branches.
The two lowest optical branches are weakly coupled to
the charge carriers are therefore not expected to play a significant role in the creation
of excitons. The next two phonon branches at the $\Sigma$ point positioned at energies 48 meV \cite{kaasbjerg12}
are linked to polar optical modes, which play a
critical role in the formation of exciton after photoexcitation of the material system.
The roles of the homopolar dispersionless mode at 50 meV which typically occurs in layered
structures as well as the sixth phonon mode with the highest energy will not be
considered here. Due to the large difference in momentum between valleys in
TMDCs, we assume that the exciton
formation occurs via an LO phonon-assisted intravalley process which preserves the valley
polarization in the monolayer system.
The Hamiltonian term associated with the interaction between excitons and
LO phonons is obtained by summing the electron-LO phonon and hole-LO phonon interaction
Hamiltonians as follows
\bea
H^{op}({\bf r_e,r_h}) &=&
{\sum_{\bf q}} {\mathcal C}
\left [ \exp (i {\bf q.r_e}) - \exp (i {\bf q.r_h}) b_{\vec{q},q_z}^{} + c.c \right ],\;
\label{hop} \\
{\mathcal C} &=& \frac{i e}{|\vec{q}|} \sqrt{\frac{\hbar \omega_{LO}}{2 \epsilon_o V} (\frac{1}{\kappa_{\infty}}-
\frac{1}{\kappa_{0}})}\; {\rm erfc[} |\vec{q}| \; \sigma/2]
\label{coup}
\eea
where ${\bf{r_e}}=\left(x_e,y_e,z_e\right) = \left(\vec{r_e},z_e\right)$
and ${\bf{r_h}} = \left( x_h,y_h,z_h \right) = \left( \vec{r_h},z_h \right)$
denote the respective space coordinates of the electron and hole, and
$\vec{r_e}$ (or $\vec{r_h}$) marked with an arrow
denotes the monolayer in-plane coordinates of the electron (or hole).
The phonon creation and annihilation operators are denoted by
$b^{\dagger}_{\vec{q},q_z}$ and $b_{\vec{q},q_z}$, respectively, where
${\bf{q}}=\left(\vec{q}, q_z\right)$ is composed
of the in-plane $\vec{q}$ and perpendicular $q_z$ components of the phonon wavevector.
The term $\omega_{LO}$ denotes the frequency of the LO phonon, $\epsilon_o$ is the permittivity of free space, $V$ is the volume of the crystal. The low-frequency and low-frequency relative dielectric constants are given by
$\kappa_{0}$ and $\kappa_{\infty}$, respectively. The inclusion of the
complementary error function $ {\rm erfc[q} \sigma/2]$ where $\sigma$ is the effective width of the electronic
Bloch states is based on the constrained interaction
of LO phonon with charge carriers in two-dimensional materials \cite{kaasbjerg12}.
For the monolayer MoS$_2$, the Fr\"ohlich coupling constant of 98 meV and an effective
width $\sigma$ = 4.41 \AA \; provide good fit to the interaction energies
evaluated from first principles in the long-wavelength limit \cite{kaasbjerg12}.
Due to dielectric screening,
the Fr\"ohlich interaction decreases with increase in the phonon momentum,
and larger coupling values $( \ge 330 )$ meV were obtained in the small momentum
limit in another study \cite{sohier2016two}. The Fr\"ohlich coupling constants obtained in
earlier works \cite{kaasbjerg12,sohier2016two} will be used in this study to compute the
formation times of excitons.
The field operator $\hat\Psi^\dagger_{_{e-h}}$
of a pair of electron and hole with a centre of mass that moves freely
is composed of electron and hole operators as follows
\be
\label{field}
\hat\Psi^\dagger_{_{e-h}}({\vec{R}},{\vec{r}},z_e,z_h)
= \frac{1}{A} \sum_{{\vec{K}, \vec{k}}}
e^{- i\vec{K} \cdot \vec{R}} e^{- i\vec{k} \cdot \vec{r}} \psi_e(z_e) \; \psi_h(z_h) \;
a_{v,{\alpha_h \vec{K} - \vec{k}}}^{\dagger}\; a_{c,{\alpha_e \vec{K} + \vec{k}}}^{\dagger}\;,
\ee
where $A$ is the two-dimensional quantization area in the monolayer plane,
and $a_{v,\vec{K}}^{\dagger}$ ($a_{c,\vec{K}}^{\dagger}$) are the respective hole
and electron creation operators with in-plane wavevector $\vec{K}$.
The center-of-mass wavevector $\vec{K} = \vec{k_e}+\vec{k_h}$ and the wavevector of the
relative motion $\vec{k} = \alpha_h \vec{k_e}- \alpha_e \vec{k_h}$ where
$\vec{k_e}$ ($\vec{k_h}$) is the electron (hole) in-plane wavevector,
with $\alpha_e ={m_e}/(m_e+m_h)$, $\alpha_h = {m_h}/(m_e+m_h)$
where $m_e$ ($m_h$ ) is the effective mass of the electron (hole).
In Eq.\ref{field}, the excitonic center of mass coordinate $\vec{R}$ and relative
coordinate $\vec{r}$ parallel to the monolayer plane are given by
\bea
\label{coord}
\vec{R} &=& \alpha_e \vec{r_e} + \alpha_h \vec{r_h}\;, \\
\nonumber
\vec{r} &=& \vec{r_e} -\vec{r_h}.
\eea
The electron and hole wave functions ($\psi_e(z_e)$, $\psi_h(z_h)$)
in the lowest-energy states are given by $\mathcal{N} \; \cos[\frac{\pi z_j}{L_w}]$ (j = e,h)
for $|z_j| \le \frac{L_w}{2}$, and 0 for $|z_j| > \frac{L_w}{2}$. The term $\mathcal{N}$ denotes the normalization constant and
$L_w$ is the average displacement of electrons and holes in the $z$ direction
perpendicular to the monolayer surface \cite{thilrelaxjap}.
\subsection{Exciton creation Hamiltonian}
The field operator $\hat\Psi^\dagger_{ex}$
of an exciton located at $\left(\vec{R},\vec{r},z_e,z_h \right)$
differs from the field operator $\hat\Psi^\dagger_{_{e-h}}$
of a free moving pair of electron and hole (see Eq.\ref{field}), and
is given by \cite{taka2,oh2000exciton,thilagam2006spin}
\be
\label{fieldex}
\hat\Psi^\dagger_{ex}({\vec{R}},{\vec{r}},z_e,z_h)
= \frac{1}{A} \sum_{\vec{K}}
e^{- i\vec{K} \cdot \vec{R}} \; \rho_{ex}^\star(\vec{r}) \; \psi_e(z_e) \; \psi_h(z_h) \;
B_{\vec{K}}^{\dagger},
\ee
where $B_{\vec{K}}^{\dagger}$ is the exciton creation operator with
center-of-mass wavevector $\vec{K}$ parallel to the monolayer plane.
The 1s two-dimensional exciton wavefunction
$\rho_{ex}(\vec{r})$ = $\sqrt{\frac{2 \beta^2}{\pi}} \exp(- \beta |\vec{r}|)$
where $\beta$ is a variational parameter.
Using Eqs. \ref{hop}, \ref{field} and \ref{fieldex}, the Hamiltonian associated with the
formation of an exciton from an initial state of free electron-hole
pair with absorption/emission of an LO phonon appear as
\bea
\label{int}
H^{F}_I &=& \frac{1}{\sqrt{A}}
\sum_{\vec{K},\vec{k},\vec{q},q_z} \; \lambda_o \; {\mathcal F_-}(\vec{k},\vec{q},q_z)\;
{\rm erfc[} \frac{|\vec{q}| \; \sigma}{2}]
B_{\vec{K}}^{\dagger}\; b_{\vec{q}, q_z} \;
a_{v,{\alpha_h (\vec{K}-\vec{q}) - \vec{k}}}\; a_{c,{\alpha_e (\vec{K}-\vec{q}) + \vec{k}}}\; \\
\nonumber
&+& \lambda_o \; {\mathcal F_+}(\vec{k},\vec{q},q_z)\; {\rm erfc[} \frac{|\vec{q}| \; \sigma}{2}] \;
B_{\vec{K}}^{\dagger}\; b_{\vec{q}, q_z}^{\dagger}
a_{v,{\alpha_h (\vec{K}+\vec{q}) - \vec{k}}}\; a_{c,{\alpha_e (\vec{K}+\vec{q}) + \vec{k}}}\;,
\\ \nonumber \\ \label{ffact}
{\mathcal F_{\mp}}(\vec{k},\vec{q},q_z) &=& {\mathcal F_e}(\pm q_z)\; {\mathcal G}(\vec{K} \pm \alpha_h \vec{q})
- {\mathcal F_h}(\pm q_z) \; {\mathcal G} (\vec{K} \mp \alpha_e \vec{q}),
\\ \label{ffact2}
{\mathcal G}(\vec{K} \pm \alpha_i \vec{q}) &=& \int \; d\vec{r} \rho_{ex}^\star(\vec{r})\;
e^{i(\vec{k} \pm \alpha_i \vec{q}) \cdot \vec{r}},
\\ \label{ffact3}
{\mathcal F_i}(q_z) &=& \int \; dz_i |\psi_i(z_i)|^2\; e^{i q_z z_j}, \; \; {\rm i = e,h}
\eea
where the coupling constant $\lambda_o = \sqrt{\frac{e^2 L_m \hbar \omega_{LO}}{2 \epsilon_o A} (\frac{1}{\kappa_{\infty}}- \frac{1}{\kappa_{0}})}$ and $L_m$ is the monolayer thickness. The form factor ${\mathcal G}$ is evaluated using the
explicit form of the two-dimensional exciton wavefunction $\rho_{ex}(\vec{r})$.
Likewise the second form factor ${\mathcal F}$ is computed using the
electron wavefunction $\psi_e(z_e)$ and hole wavefunction
$\psi_h(z_h)$.
\subsection{Exciton formation rate}
For transitions involving a single phonon with wavevector $\vec{q}$,
the formation rate of the exciton with wavevector $\vec{K}$
is computed by employing the Fermi golden rule
and the interaction operator in Eq.\ref{int}
as follows
\bea
\label{rate}
W(\vec{K} \pm \vec{q},q_z) &=& \frac{1}{A} \frac{2 \pi}{\hbar} \;
|\lambda_o|^2 \; |{\mathcal F_{\pm}}(\vec{k},\vec{q},q_z)|^2 \; {\rm erfc[} \frac{|\vec{q}| \; \sigma}{2}]^2 \;
f_h(\alpha_h (\vec{K} \pm \vec{q}) - \vec{k}) \; f_e(\alpha_e (\vec{K} \pm \vec{q}) + \vec{k})
\\
\nonumber &\times& (f_{ex}(\vec{K})+1) \;
(\overline n_{\bf q} + \frac{1}{2} \pm \frac{1}{2}) \; \delta(E_{ex} - E_{eh}^\pm \pm \hbar \omega_{LO}),
\eea
where the emission (absorption) of phonon is denoted by $+$ ($-$), and the exciton energy
$ E_{ex} = E_g + E_{_b}+ \frac{\hbar^2 |\vec{K}|^2}{2 (m_e+m_h)}$ where $E_{_b}$ is the
exciton binding energy. The energies of charge carrier is $E_{eh}^\pm = E_g +
\frac{\hbar^2 |\vec{K} \pm \vec{q}|^2}{2 (m_e + m_h)} + \frac{\hbar^2 |\vec{k}|^2}{2 \mu}$
where $\mu$, the reduced mass is obtained using $\frac{1}{\mu} = \frac{1}{m_e} + \frac{1}{m_h}$.
At low temperatures of the charge carriers the phonon bath can be considered
to be thermal equilibrium with negligible phonon-phonon scatterings and phonon decay
processes.The thermalized average occupation of phonons for low temperatures of the charge carriers is
given by
\be
\label{phon}
{\overline n_{q}} = [\exp \left(\frac{\hbar \omega_{LO}}{k_B T_l}\right)-1]^{-1},
\ee
where $T_l$ is the lattice
temperature and $\hbar \omega_{LO}$ is the energy of the LO phonon that is emitted
during the exciton generation process.
The relaxation of electrons and holes at high enough temperatures
($\approx 200 K$) generally displaces phonons beyond the equilibrium point when
phonon-phonon related processes become dominant.
The phonon Boltzmann equation \cite{kaasbjerg2014hot}
takes into account a common temperature that is achieved
as a result of equilibrium reached between electrons and phonons.
Hot-phonon effects is incorporated by replacing
the temperature $T_l$ in Eq.\ref{phon} by an effective lattice temperature
$T_{ph}$ \cite{kaasbjerg2014hot}.
The charge carriers are assumed to be in quasi-thermal equilibrium during the exciton formation
process. Consequently the occupation numbers ($f_h(K)), f_e(K)$) of hole and electron states in Eq.\ref{rate}
can be modeled using the Fermi-Dirac distribution
\bea
\label{fd}
f^i(K_i) &=&\left[ \exp \left( \frac{E(K_i) - \mu_i}{K_B T_i}\right ) + 1 \right]^{-1}, \quad \; { i = e, h}\\
\label{fd2}
\mu_{i} &=& K_B T_i \ln \left[\exp \left(\frac{\pi \hbar^2 n_i}{m_i k_b T_i} -1 \right) \right],
\eea
where the chemical potential $\mu_i$ is dependent on
the temperature $T_{i}$ and the two-dimensional density $n_i$ of the charge carriers.
When the mean inter-excitonic distance is higher
the exciton Bohr radius as considered to be the case in this study,
the exciton can be assumed to be an ideal boson \cite{thilpauli}
with a Bose-Einstein distribution \cite{ivanov1999bose}
\bea
\label{bes}
f^{ex}(K) &=&\left[ \exp \left( \frac{E(K) - \mu_{ex}}{K_B T_{ex}}\right ) - 1 \right]^{-1},\\
\label{bes2}
\mu_{ex} &=& K_B T_{ex} \ln \left[1-\exp \left(- \frac{2 \pi \hbar^2 n_{ex}}{g (m_e+m_h) k_b T_{ex}} \right) \right],
\eea
where $\mu_{ex}$ is the exciton chemical potential, $T_{ex}$ is the exciton temperature
and $n_{ex}$ is the exciton density. The degeneracy factor $g$ is obtained
as the product of the spin and valley degeneracy factors \cite{kaasbjerg12}.
\subsection{Numerical results of the Exciton formation Time}
The formation time of an exciton with
wavevector $\vec{K}$, $T(\vec{K})$ is obtained by summing the wavevectors $(\vec{k},\vec{q},q_z)$
over the rate obtained in Eq.\ref{rate}
\be
\label{formt}
\frac{1}{T(\vec{K})} = \sum_{\vec{k},\vec{q},q_z} \; W(\vec{K} \pm \vec{q},q_z)
\ee
To obtain quantitative estimates of the exciton formation time using Eq.\ref{formt}, we use the monolayer MoS$_2$ material parameters as $m_e$ = 0.51 $m_o$, $m_h$ = 0.58 $m_o$ \cite{jin2014intrinsic}
where $m_o$ is the free electron mass, and the coupling constant $\alpha_o$= 330 meV \cite{sohier2016two}.
We set the phonon energy $\hbar \omega_{LO}$ = 48 meV \cite{kaasbjerg12}, and the layer thickness $h$ = 3.13 \AA \; \cite{ding2011first} is used to determine
the upper limit of $\approx$ 6 \AA \; for $L_w$,
the average displacement of electrons and holes in the direction perpendicular to the plane of the
monolayer. We fix the effective lattice temperature
$T_{ph}$ = 15 K, but vary the electron and hole temperatures,
$T_e$ and $T_h$.
Fig. \ref{formK}a,b show the calculated values (using Eqs. \ref{ffact}- \ref{formt}) of the exciton formation
times as a function of exciton wavevector $\vec{K}$ with emission of an LO phonon
at different electron, hole and exciton temperatures
and densities, $n_e$ = $n_h$ = $n_{ex}$ = 1 $\times$ 10$^{11}$ cm$^{-2}$ and 5 $\times$ 10$^{11}$ cm$^{-2}$.
To obtain the results, we assume the temperatures to be the same for excitons and
unbound electron-hole pairs.
The results indicate that very fast
exciton formation times of less than one picosecond time occurs at
charge densities of 5 $\times$ 10$^{11}$ cm$^{-2}$ and carrier temperatures less than 300 K.
These ultrafast sub-picosecond exciton formation times are in agreement with recent experimental findings \cite{ceballos2016exciton} recorded at room temperatures in the monolayer MoS$_2$.
The exciton formation times are increased at the lower carrier densities of 1 $\times$ 10$^{11}$ cm$^{-2}$.
The wavevector of exciton states formed due to optical excitation
of the ground state of the crystal lies close
to zero due to selection rules. The results in Fig. \ref{formK}a,b
show that while excitons are dominantly created at $|\vec{K}|$ = 0 at low charge carrier temperatures ($\approx$ 50 K), exciton formation occurs most rapidly at non-zero exciton center-of-mass wavevectors
($|\vec{K}|_f \neq$) at higher temperatures ($T_e$ = $T_h$ $\ge $ 140 K) of the charge carriers.
At $T_e$ = $T_h$ $\approx $ 300 K, the shortest exciton formation time occurs at $|\vec{K}|_f$ =
0.04 $\times$ 10$^{10}$ m$^{-1}$ (about 5.6 meV).
The results in Fig. \ref{formK}a,b indicate that at exciton wavevectors greater than
0.06 $\times$ 10$^{10}$ m$^{-1}$, there is a notable increase in the exciton formation times linked to low electron-hole
plasma temperatures $T_e = T_h \le$ 80 K. At high carrier temperatures there is likely
conversion of newly formed composite bosons such as excitons into fermionic fragment species \cite{thilagam2013crossover}. The inclusion of considerations of
the quantum mechanical crossover of excitons into charge carriers at higher plasma temperatures
will add to greater accuracy when computing exciton formation times.
This currently lies beyond the scope of this study and will be considered in future
investigations where the role of the composite structure of excitons
in their formation rate will be examined.
\begin{figure}[htp]
\begin{center}
\subfigure{\label{Ka}\includegraphics[width=8.5cm]{WaveV}}\vspace{-1.1mm} \hspace{4.7mm}
\subfigure{\label{Kb}\includegraphics[width=8.5cm]{WaveV8}}\vspace{-1.1mm} \hspace{1.1mm}
\end{center}
\caption{(a)
The exciton formation time as a function of the exciton center-of-mass wavevector $|\vec{K}|$
in the monolayer MoS$_2$ at different temperatures, $T_e$ = $T_h$ = $T_{ex}$ (50 K, 80 K, 140 K, 200 K, 300 K).
We use $m_e$ = 0.51 $m_o$, $m_h$ = 0.58 $m_o$ \cite{jin2014intrinsic}
where $m_o$ is the free electron mass, the coupling constant $\alpha_o$ = 330 meV \cite{sohier2016two}
and $\hbar \omega_{LO}$ = 48 meV \cite{kaasbjerg12}.
The effective lattice temperature
$T_{ph}$ = 15 K, $L_w$ $\approx$ 6 \AA \;, the exciton binding energy,
$E_b$ = 300 meV \cite{thiljap} and densities, $n_e$ = $n_h$ = $n_{ex}$ = 1 $\times$ 10$^{11}$ cm$^{-2}$. \\
(b) The exciton formation time as a function of the center-of-mass wavevector $|\vec{K}|$
in the monolayer MoS$_2$ at different temperatures, $T_e$ = $T_h$ = $T_{ex}$ (50 K, 80 K, 140 K, 200 K, 300 K).
All other parameters used are the same as specified in (a) with the exception of
densities, $n_e$ = $n_h$ = $n_{ex}$ = 5 $\times$ 10$^{11}$ cm$^{-2}$.}
\label{formK}
\end{figure}
The effect of the variations within the electron-hole plasma temperatures or differences
between $T_e$ and $T_h$ on the exciton formation time is illustrated in Fig. \ref{diff}.
The formation times are computed at different exciton center-of-mass
wavevectors with the electron temperature fixed at $T_e$ = 250 K, and
exciton temperature $T_{ex}$ = 50 K. The charge densities $n_e$ = $n_h$ = $n_{ex}$ = 5 $\times$ 10$^{11}$ cm$^{-2}$
and all other parameters used are the same as specified in the caption for Fig. \ref{formK}.
At the larger wavevector $|\vec{K}|$ = 0.07 $\times$ 10$^{10}$ m$^{-1}$ ($\approx$ 17.1 meV)
the formation time is shortest when the hole temperature $T_h$ = $\approx $ 120.
With decrease in the center-of-mass wavevector $|\vec{K}|$, there is corresponding
decrease in the formation time when the hole temperature is allowed to decrease further from
the electron temperature. At the low exciton
wavevector $|\vec{K}|$ = 0.005 $\times$ 10$^{10}$ m$^{-1}$,
the shortest formation time occurs when the difference between $T_e$ and
$T_h$ reach the maximum possible value. These results demonstrate the interplay of competitive effects of the hole-phonon and the electron-phonon dynamics on a picosecond time scale which
results in a non-monotoinic temperature difference dependence $|T_h - T_e|$
of the exciton formation time.
\begin{figure}[htp]
\begin{center}
\subfigure{\label{figa}\includegraphics[width=9.9 cm]{holetemp}}\vspace{-1.1mm} \hspace{1.1mm}
\end{center}
\caption{The exciton formation time as a function of hole temperature
at different exciton center-of-mass wavevector $|\vec{K}|$ = 0.07 $\times$ 10$^{10}$ m$^{-1}$ (17.1 meV, red),
0.05 $\times$ 10$^{10}$ m$^{-1}$ (8.7 meV, blue), 0.02 $\times$ 10$^{10}$ m$^{-1}$ (1.4 meV, magenta),
0.005 $\times$ 10$^{10}$ m$^{-1}$ (0.08 meV, green). The electron temperature is fixed at $T_e$ = 250 K,
exciton temperature $T_{ex}$ = 50 K and the carrier densities $n_e$ = $n_h$ = 5 $\times$ 10$^{11}$ cm$^{-2}$.
and $n_{ex}$ = 5 $\times$ 10$^{11}$ cm$^{-2}$. All other parameters used are the same as specified in the caption for Fig. \ref{formK}.}
\label{diff}
\end{figure}
In Fig. \ref{concf}, the exciton formation time is plotted
as a function of the carrier density $n_e$ = $n_h$ at
different temperatures, $T_e$ = $T_h$ = $T_{ex}$ (300 K, 200 K, 100 K).
The exciton center-of-mass wavevector $|\vec{K}|$ = 0.03 $\times$ 10$^{10}$ m$^{-1}$.
All other parameters used are the same as specified in the caption for Fig. \ref{formK}.
Using the numerical values of the formation times,
we performed numerical fits using the following relation which involve the carrier concentrations \cite{oh2000exciton}
\be
\label{fitT}
T(n_i) = \frac{B}{n_i^p} \quad \; { i = e, h}
\ee
where $B$ and $p$ are fitting parameters. Using the results used
to obtain Fig. \ref{concf}, we get $B$ = 20.64 at $T_e$ = $T_h$ = 300 K,
$B$ = 10.35 at $T_e$ = $T_h$ = 200 K, $B$ = 3.56 at $T_e$ = $T_h$ = 100 K
and $B$ = 1.54 at $T_e$ = $T_h$ = 50 K. The constant $p \approx$ 2 irrespective
of the electron and hole temperatures. This implies an inverse
square-law dependence of the exciton formation time
on the electron/hole concentration. Consequently a square-law
dependence of the photoluminescence on excitation density is expected
to arise in the monolayer MoS$_2$ as well as other monolayer transition metal
dichalcogenides.
\begin{figure}[htp]
\begin{center}
\subfigure{\label{figa}\includegraphics[width=9.0cm]{conc}}\vspace{-1.1mm} \hspace{1.1mm}
\end{center}
\caption{The exciton formation time as a function of the carrier density $n_e$ = $n_h$ at
different temperatures, $T_e$ = $T_h$ = $T_{ex}$ (300 K, 200 K, 100 K).
The exciton center-of-mass wavevector $|\vec{K}|$ = 0.03 $\times$ 10$^{10}$ m$^{-1}$.
All other parameters used are the same as specified in the caption for Fig. \ref{formK}.}
\label{concf}
\end{figure}
\section{Exciton formation times for other exemplary monolayer transition metal
dichalcogenides}
The theoretical results obtained in this study for MoS$_2$ are expected to be applicable to
other low dimensional transition metal dichalcogenides. However subtle variations
in the exciton formation times are expected due to differences in the exciton-LO coupling
strengths and energies of the LO phonon in the monolayer materials.
The bare Fr\"ohlich interaction strengths obtained via ab initio techniques
give 334 mev (MoS$_2$), 500 meV (MoSe$_2$), 140 mev (WS$_2$) and 276 meV (WSe$_2$) \cite{sohier2016two},
hence the Molybdenum-based TMDCs possess higher exciton-phonon coupling
strengths than the Tungsten-based TMDCs.
A precise estimate of the exciton binding energy in the
monolayer TMDCs is not available, however a range of binding energies (100 to 800 meV) have been
reported for the monolayer systems \cite{hanbicki2015measurement,olsen2016simple,makatom,chei12,ugeda2014giant,
hill2015observation,choi2015linear,chei12,komsa2012effects,thiljap}. In order to compare
the exciton formation rates between Molybdenum-based TMDCs and
Tungsten-based TMDCs, we make use of the effective masses of electron and holes
at the $K$ energy valleys/peak position given in Ref. \cite{jin2014intrinsic}
and the Fr\"ohlich interaction strengths and
LO phonon energies given in Ref.\cite{sohier2016two}. To simplify the numerical
analysis, we fix the exciton binding
energies at $\approx$ 330 meV for all the TMDCs under investigation. This assumption is not expected
to affect the order of magnitude of the exciton formation times, and also to
not detract from the analysis of effects of Fr\"ohlich interaction strengths
on the formation times.
The results in Fig. \ref{comp} a, b show that the exciton formation times
of the selenide-based dichalcogenides
are smaller than the sulphide-based dichalcogenides at $T_e$ = $T_h$ = $T_{ex}$ = 100 K
and 300 K (with $|\vec{K}| \le$ 0.05 $\times$ 10$^{10}$ m$^{-1}$).
This is due to the comparatively higher Fr\"ohlich interaction strengths and lower LO phonon energies
of monolayer MoSe$_2$ and WSe$_2$. The results in Fig. \ref{comp}a,b also indicate that
excitons in the monolayer WS$_2$ are
dominantly created at non-zero center-of-mass wavevectors compared to the other three monolayer
dichalcogenide systems. This may be attributed to the comparatively lower effective masses
of hole and electron in the monolayer WS$_2$.
It is instructive to compare
the exciton formation times in Fig. \ref{comp}a,b with the
radiative lifetimes of zero in-plane momentum
excitons in suspended MoS2
monolayer of $\approx$ 0.18 - 0.30 ps at 5 K
\cite{wang2016radiative}. The lifetimes of excitons depend linearly on the
exciton temperature and increase to the picoseconds range at small temperatures and
is larger than 1 ns at the room temperature. This indicates that the exciton formation processes
are likely to dominate in the initial period when the TMDCs are optically excited
at high exciton temperatures. In the low temperature range (5 K - 20 K),
an interplay of competing effects of exciton generation and radiative decay
are expected to occur on the sub-picosecond time scale.
Environmental parameters such as impurity concentration, exciton density and density of
excess charge carriers that affect the stability of low dimensional trions will need
to be taken into account in order to accurately model the exciton generation process at the low temperature
regime.
The exciton formation scheme adopted in this study has been parameterized by physical quantities such
as the exciton density and charge carrier densities. It is not immediately clear whether
these parameters can be extracted directly using ab-initio quantum mechanical and
time-dependent density functional theory approaches.
Computations based on ab-initio techniques are generally numerically intensive and
time consuming which are the main challenges in modeling low dimensional material systems. It is expected
that improvements in first principles modeling of anisotropic systems may result
in more efficient and rewarding approaches to determining the density functions of excitons
and charge carriers in future investigations. The Auger process provides a non-radiative decay
channel for electron-hole pair recombination, hence this mechanism must be taken
into account for accurate predictions of exciton formation times in future studies.
\begin{figure}[htp]
\begin{center}
\subfigure{\label{Ka}\includegraphics[width=8.5cm]{comp-100}}\vspace{-1.1mm} \hspace{4.7mm}
\subfigure{\label{Kb}\includegraphics[width=8.5cm]{comp-300}}\vspace{-1.1mm} \hspace{1.1mm}
\end{center}
\caption{(a)
The exciton formation time as a function of the exciton center-of-mass wavevector $|\vec{K}|$
in common monolayer systems (MoS$_2$, MoSe$_2$, WS$_2$, WSe$_2$)
at temperatures, $T_e$ = $T_h$ = $T_{ex}$ = 100 K. The effective masses of electron and holes
at the $K$ energy valleys/peak are taken from Ref. \cite{jin2014intrinsic}
and the Fr\"ohlich interaction strengths and LO phonon energies are obtained from Ref.\cite{sohier2016two}.
The effective lattice temperature
$T_{ph}$ = 15 K, $L_w$ $\approx$ 6 \AA \;, and
densities, $n_e$ = $n_h$ = $n_{ex}$ = 5 $\times$ 10$^{11}$ cm$^{-2}$.\\
(b) The exciton formation time as a function of the exciton center-of-mass wavevector $|\vec{K}|$
in common monolayer systems at temperatures, $T_e$ = $T_h$ = $T_{ex}$ = 300 K.
All other parameters used are the same as specified in (a) above.
}
\label{comp}
\end{figure}
\section{Conclusion \label{conc}}
Transition metal chalcogenides have emerged as promising materials in which
excitons exist as stable quasi-particles with high binding energies and thus play
important roles in the optical processes of monolayer TMDCs.
The dynamics of excitons in monolayer transition metal
dichalcogenides has been extensively studied over the last
five years in terms of both theory and applications. However
the formation of excitons from free carriers has only been recently measured, and
in this work we develop a model within the framework of Fermi's Golden rule to calculate the
formation dynamics of excitons from free carriers.
This theoretical study is aimed at providing
a fundamental understanding of the exciton generation process
in optically excited monolayer transition metal dichalcogenides.
We focus on a mechanism by which excitons are generated via
the LO (longitudinal optical) phonon-assisted scattering process from free electron-hole pairs
in layered structures. The exciton formation time
is computed as a function of the exciton center-of-mass wavevector,
electron and hole temperatures and densities for known values of
the Fr\"ohlich coupling constant, LO phonon energy, lattice temperature and the exciton binding energy.
Our results show that the exciton is generated at
non-zero wavevectors at higher temperatures ($\ge $ 120 K) of charge carriers,
that is also dependent on the density of the electron and holes.
The inverse square-law dependence of the exciton formation time
on the density of charge carriers is also demonstrated by the
results of this study.
For monolayer MoS$_2$, we obtain exciton formation times on the picosecond time scale at
charge densities of 1 $\times$ 10$^{11}$ cm$^{-2}$ and carrier temperatures less than 100 K.
The exciton formation times decreases to the sub-picosecond time range at higher densities
(5 $\times$ 10$^{11}$ cm$^{-2}$) and electron-hole plasma temperatures ($\le$ 300 K).
These ultrafast formation times are in agreement with recent experimental results
($\approx$ 0.3 ps) for WSe$_2$, MoS$_2$, and MoSe$_2$ \cite{ceballos2016exciton}.
Due to the comparatively higher Fr\"ohlich interaction strengths and lower LO phonon energies
of monolayer MoSe$_2$ and WSe$_2$, the exciton formation times
of the selenide-based dichalcogenides
are smaller than the sulphide-based dichalcogenides at $T_e$ = $T_h$ = $T_{ex}$ = 100 K
and 300 K (with $|\vec{K}| \le$ 0.05 $\times$ 10$^{10}$ m$^{-1}$).
The results of this study is expected to be useful in understanding
the role of the exciton formation process in electroluminescence
studies \cite{sundaram2013electroluminescence,ye2014exciton} and exciton-mediated processes
in photovoltaic devices \cite{bernardi2013extraordinary,wi2014enhancement,tsuboi2015enhanced}.
| 9fc0f3c8016d0d8469363108d528984c8b99e653 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{A letter from Saratov} \label{story}
During his last trip to Moscow, the second author of this article came into possession of a remarkable mathematical letter. The custodian of the letter, a Russian businessman X who wished to remain anonymous,
presented the letter to Tabachnikov at the end of his lecture on configuration theorems in projective geometry \cite{ST} (like many successful contemporary Russian entrepreneurs, Mr. X has a degree in mathematics).
Mr. X explained that the letter had been written by his great-great-great grandfather, Konstantin Shestakov, shortly after he had been discharged from the Russian Army during the Napoleonic Wars. It appears that K. Shestakov had befriended the famous French mathematician, Jean-Victor Poncelet, then a prisoner of war in the Russian town of Saratov, and had been drawn into geometry by him.\footnote{According to the family history, Shestakov was well educated: he attended Saint Petersburg Engineering School.} The letter, dated fall of 1814 and apparently never mailed, was addressed to Shestakov's younger brother, Alexander, who had attended the Kazan Gymnasium with N. I. Lobachevsky.
We reproduce the first paragraph of the letter in the original Russian (in the modern orthography) and give its full translation into English.
\begin{quote}
{\small \it
\begin{otherlanguage}{russian}
Милый Саша!
Ты конечно слышал, что раны мои, полученные при битве под Смоленском, вынудили меня выйти в отставку, и что я постепенно восстанавливаю силы свои в Саратове. Это сонный город на Волге, где окромя роскошных обедов да балов у здешнего губернатора г. Панчулидзева и делать вовсе нечего. Служба моя состоит и в надзоре за здешней тюрьмой, однако заключённые ведут себя смирно и не нуждаются в постоянном присмотре. Пишу сообщить о своём житье-бытье, а также и с просьбой переслать математическую часть сего письма однокашнику твоему по гимназии Николаю Лобачевскому; как слыхал я, он теперь адъюнкт-профессор в Казани...
\end{otherlanguage}
}
\end{quote}
\begin{quote}
{\small \it
Dearest Sasha:
You doubtless have heard that the wounds sustained at a battle near Smolensk
forced me to withdraw from the army and that I have been recovering ever since in Saratov. Saratov is a dreary town on the Volga river which, aside from the several excellent dinner parties and balls thrown by
Governor Panchuleedzev, affords us almost nothing interesting to do. I have some hand in overseeing the prison, but the prisoners are peaceful and rarely need attention. I am writing this partly to tell you how I have been keeping myself, but also to ask you the favor of passing on the mathematical part of this letter to your former classmate Nikolai Lobachevsky who, as I understand it, is now an adjunct professor at the Kazan University.
I spent much of the winter holed up in my room, trying to avoid the oppressive regimen of drinking and other kinds of enforced merriment. As you know, I prefer a life of quiet contemplation. I had never quite become used to the cruel interruption of my studies brought about by the war. I confess that my bitterness over the situation had been driving me towards madness, but then everything changed.
I write to you now with great excitement, as a man given a divine gift.
My gift came to me in the form a prisoner, Lieutenant of the French Army Monsieur Poncelet. As luck would have it, Poncelet was assigned to me as an assistant of sorts, though the pedestrian kind of work required of him hardly touched on his enormous talents. It became clear almost immediately that he was nothing at all like our other prisoners. Quiet and unassuming, with a temperament much like my own, Poncelet preferred to keep his own company. The relatively undemanding work he did for me afforded him to pursue his true passion, which was geometry.
Poncelet explained to me that he was whiling away the time by reconstructing from first principles the theorems of geometry he had learned during his student days at \'Ecole Polytechnique in Paris. He said more than once that the great beauty of mathematics is that it can be created whole from very few principles. I knew something of geometry myself and I gradually became engaged in the project of this brilliant man. In the space of several months I went from casual observer to eager student to active participant.
You have to understand that Poncelet was no ordinary student! As time passed, I could see that he was not just trying to recreate the lessons of his masters, as I had first thought, but rather that he was striking out into the unknown. My excitement was almost as great as his when he explained to me his astonishing discovery about conic sections.
This is the part of the letter I request that you pass on to Lobachevsky.
I recall that your mathematics teacher, the venerable Grigorii Ivanovich, always praised Nikolai for his aptitude for mathematics. I trust that you will be able to follow this but, in any case, Lobachevsky certainly will, and I hope that it will interest him.
Suppose that $A$ and $B$ are elliptical conic sections with $A$ surrounding $B$. Imagine that one has a point $P_1$ on $A$, then draws a segment $P_1P_2$ which is tangent to $B$ and connects $P_1$ to $P_2$, also on $A$. Now repeat the construction, drawing the segment $P_2 P_3$ tangent to $B$ so that $P_3$ is on $A$. Imagine that this construction is repeated some number of times, say 100, so that $P_{101}$ is the same point as $P_1$. One has produced a kind of closed polygonal figure involving points $P_1, P_2$, etc., which ends up where it starts after 100 steps.
Poncelet's great discovery is that the same construction, starting with a
different point on $A$, will also repeat after 100 steps. That is, if one moves
$P_1$ to the new point $P'_1$ on $A$, and then produces points $P'_2, P'_3$, etc. then
it will again happen that $P'_{101} = P'_1$! You have to understand, finally, that
I have only used 100 as an example. Whether the figure repeats itself or not, and after how many steps, depends on the choice of $A$ and $B$. Poncelet encouraged me to think about this construction, as he did, as a kind of spinning polygon. He imagined moving the point $P_1$ continuously, so that the other points would move as well. Poncelet's great discovery is that this configuration of (say) 100 points and 100 segments spins around the conics $A$ and $B$ and remains intact.
Poncelet left the camp in June and I do not know what became of him, but I continued his work. I could not get the idea of the spinning configuration out of my mind and eventually I thought to ask some questions of my own. For simplicity I considered the case when $A$ and $B$ were both circles, with $A$ surrounding $B$ and $B$ being set in a position that was somewhat offset from the center of $A$. By adjusting the center and position of $B$ I could control the particulars of Poncelet's construction. After some amount of fussing around, I settled on a location and size of $B$ which caused the configuration to close up after 5 steps, making a star-like pattern.
I became interested in the question of how the center of gravity of the figure changed as it spun around. You might say that I approached this in an experimental way, drawing the figure on the page and then approximating the center of gravity by a kind of trick of drawing the points together. I wish that I could include diagrams of this laborious method, but perhaps they would mean nothing to you. Suffice it to say that my calculations seemed to show a promising result but were not accurate enough to convince me.
You probably remember my passion for precise reasoning! Eventually I joined many sheets together, making a kind of poster on which to do my calculations. I remember spending the bulk of a week scribbling on this poster, night after night, as gradually the beautiful answer revealed itself. It is yet another circle! As the great Poncelet figure spins around, its center of gravity traces out a circle as well! Understand that there is nothing special about the star configuration that I settled upon; I expect that the principle should be completely general. I am convinced that this is a first-rate discovery to rival Poncelet's own.
I regret that I have so far not hit upon a proof of my discovery, in the same way that Poncelet was able to find a proof for his. Alas, perhaps if my friend were still here we could find it together. Nonetheless, I have calculated things out to such a precision that I am confident in proclaiming this as a theorem.
I would be grateful if Lobachevsky could look into the matter.
Your brother Konstantin.
}
\end{quote}
\begin{figure}[ht]
\centering
\includegraphics[height=2in]{map}\qquad
\includegraphics[height=2in]{Panchuleedzev}
\caption{An old map of Saratov and its Governor A. D. Panchuleedzev (1758--1834).}
\label{Saratov}
\end{figure}
To put this letter into a historical context, Jean-Victor Poncelet participated in Napoleon's invasion of Russia and was wounded at the Battle of Krasnoi during Napoleon's retreat from Moscow in November of 1812. He
spent more than a year in Russian captivity in Saratov. There he developed projective geometry. Later, his {\it Saratov notebooks} became part of his {\it Trait\'e des propri\'et\'es projectives des figures} (1822)
and {\it Applications d'analyse et de g\'eom\'etrie} (1862). In particular, Poncelet discovered his celebrated porism during the Saratov exile.
N. I. Lobachevsky grew up in Kazan where he attended gymnasium\footnote {His teacher of mathematics was G. I. Kartashevsky.} and then the newly established Kazan University. The whole career of Lobachevsky was spent at this university where he served from 1814, as an adjunct professor, until 1846 (elected Chancellor in 1827).
\begin{figure}[ht]
\centering
\includegraphics[height=2in]{Poncelet2}\qquad\qquad
\includegraphics[height=2in]{Lobachevsky}
\caption{J.-V. Poncelet (1788--1867) and N. I. Lobachevsky (1792--1856).}
\label{mathematicians}
\end{figure}
The literature on the Poncelet theorem and its ramifications is vast; see, e.g., \cite{BB,BKOR,Do,DR1, DR2,Fl} and the references therein. Still, after having mulled over
Shestakov's letter at some length, we feel that his result appears to be a
new and surprising addition to the Poncelet theorem. What makes the result
especially surprising is that it makes a statement about the center of mass -- an
affine geometry construction -- in the context of projective geometry.
The one result we know in this direction is Weill's Theorem, which
deals with the centers of mass of Poncelet polygons in a special case.
One could view Shestakov's Theorem as a generalization of Weill's Theorem,
though Shestakov's result does not quite imply Weill's result.
In what follows, we will present and sketch modern proofs for
two versions of Shestakov's result.
For reasons we will explain, we think that he
most likely had the first statement in mind, but it seems reasonable to present
the second statement as well, which is a variant. At the end, and for the sake
of completeness, we will sketch a proof of Weill's Theorem.
\section{The theorem} \label{mainform}
Figure \ref{pentagon} features an instance of the Poncelet porism -- exactly
the kind Shestakov said that he had ``fussed around to arrange''.
The Poncelet theorem asserts that every point
of the outer conic is a vertex of such a polygon.
The interested reader can use\footnote{You can download the
program at http://www.math.brown.edu/$\sim$res/Java/PORISM.tar}
our program and see
the kind of animation envisioned by Poncelet.
\begin{figure}[ht]
\centering
\includegraphics[height=2.3in]{penta1.pdf}
\caption{A Poncelet pentagon.}
\label{pentagon}
\end{figure}
Let $P=(A_1\ldots A_n)$ be a Poncelet $n$-gon,
and let $(x_i,y_i)$ be the Cartesian coordinates
of the vertex $A_i$, $i=1,\ldots,n$.
There are really $3$ natural interpretations for the center of mass of
$P$:
\begin{itemize}
\item The center of mass of the vertices. We denote this by $CM_0(P)$.
\item The center of mass of the edges, when the edges are given a
uniform density. We denote this by $CM_1(P)$.
\item The center of mass of the $P$ when $P$ is considered as
a ``homogeneous lamina''.
We denote this by $CM_2(P)$. When $P$ is not
embedded, $CM_2(P)$ requires the interpretation
we give in Lemma \ref{CMcoord} below.
\end{itemize}
We provide formulas for the three kinds of centers below.
We will present version of Weill's Theorem for $CM_0$ and $CM_2$.
The result fails for $CM_1$ for reasons we will mention briefly below.
We think that Shestakov most likely had $CM_0$ in mind from his
description of ``drawing the points together''. This is the
sort of thing one would do when computing $CM_0(P)$.
The coordinates of these centroids are given in the next lemma. Let
$$
\ell_i=\sqrt{(x_{i+1}-x_i)^2 + (y_{i+1}-y_i)^2}
$$
be the length of $i$th side of the polygon, and let
$d_i = (x_i y_{i+1} - x_{i+1} y_i). $ Then $L(P)=\sum_i \ell_i$ is the perimeter of $P$, and
$A(P)=(1/2) \sum_i d_i$ is its area, counted with sign and multiplicity.
\begin{lemma}
\label{CMcoord}
One has:
\begin{itemize}
\item $CM_0(P) =\frac{1}{n} \sum_{i=1}^n (x_i, y_i),$
\item $CM_1(P) = \frac{1}{2L(P)} \sum_{i=1}^n \ell_i (x_i + x_{i+1},y_i + y_{i+1}),$
\item $CM_2(P)= \frac{1}{6 A(P)} \sum_{i=1}^n d_i (x_{i}+x_{i+1},y_{i}+y_{i+1}).$
\end{itemize}
\end{lemma}
\paragraph{Proof.} Consider the triangle $O A_i A_{i+1}$ where $O$ is the origin. The centroid of this triangle is at point $(A_i + A_{i+1})/3$ and it has the area $d_i/2$. Summing up over $i$ and dividing by the total area of $P$ yields the result for $CM_2(P)$. Likewise for $CM_0(P)$ and $CM_1(P)$.
$\Box$\bigskip
Here is (our interpretation of) Shestakov's Theorem.
\begin{theorem} \label{main}
Let $\gamma \subset \Gamma$ be a pair of nested ellipses that admit a 1-parameter family of Poncelet $n$-gons $P_t$. Then both loci $CM_0(P_t)$ and $CM_2(P_t)$ are ellipses
homothetic to $\Gamma$ (or single points).
\end{theorem}
As we mentioned above, Shestakov most likely had $CM_0(P)$ in mind.
Also, he has normalized so that the outer ellipse is a circle; in this
case the loci $CM_0(P_t)$ and $CM_2(P_t)$ are circles. For the proof
we will also normalize this way. Our proof relies on the fact that
$CM_0(P)$ and $CM_2(P)$ are rational expressions in the coordinates
of $P$. This is not true for $CM_1(P)$ and, as we mentioned above,
the result fails for $CM_1(P)$. The locus in this case is not
generally a conic section.
We wrote a computer program which tests Theorem \ref{main}.
Figure \ref{experiment} shows two pictures from this program. The left side
deals with $CM_0$ and the right side deals with $CM_2$.
In both cases, the loci $CM_0(P_t)$ and $CM_2(P_t)$ are black
circles. The two Poncelet polygons shown belong to the
same Poncelet family, but they are not the same polygon.
The two circles $CM_0(P_t)$ and $CM_2(P_t)$ are
different circles. We are not sure how they are
related to each other.
\begin{figure}[ht]
\centering
\includegraphics[height=2.3in]{penta2.pdf}\quad
\includegraphics[height=2.3in]{penta3.pdf}
\caption{$CM_0(P_t)$ and $CM_2(P_t)$.}
\label{experiment}
\end{figure}
\section{A proof} \label{mainpf}
Now we sketch a proof of Theorem \ref{main}. Let us start with the algebraic geometry proof of the Poncelet porism, see \cite{GH}, or \cite{Fl}, for a detailed account.
One starts with complexifying and projectivizing: we assume that $\gamma$ and $\Gamma$ are complex conics in general position in the complex projective plane ${\mathbb {CP}}^2$ obtained from the affine plane ${\mathbb C}^2$ by adding a line at infinity.
A complex conic is isomorphic to Riemann sphere, ${\mathbb {CP}}^1$.
Let $E$ be the set of flags $(x,L)$ consisting
of a point $x \in \Gamma$ and a line $L$ through
$x$ that is tangent to $\gamma$. The space
$E$ is naturally a Riemann surface, and
the projection $p:E \to \Gamma$ that takes
$(x,L)$ to point $x$ is a 2-fold branched covering with
four branch points. These branch points, invisible in Figure 5,
are the four intersection points of $\Gamma$ and $\gamma$.
One easily computes that the Euler characteristic of
$E$ is zero. Thus $E$ is a Riemann surface which,
topologically, is a torus.
\begin{figure}[ht]
\centering
\includegraphics[height=1.5in]{involution.pdf}
\caption{Two involutions of the flag space $E$.}
\label{involutions}
\end{figure}
One has two involutions on $E$
depicted in Figure \ref{involutions}.
$$
\sigma (x,L) = (x',L),\ \ \tau(x',L) = (x',L').
$$
The Poncelet map $T=\tau \circ \sigma$ is a
translation of $E$. If a translation of a torus
has a periodic point of order $n$, then $T^n$ is the
identity, and this proves the Poncelet theorem.
The two claims of Theorem \ref{main} are proved similarly,
so we let $CM$ mean either of the two centers of mass involved.
The group generated by $\sigma$ and $\tau$ is the dihedral group $D_n$. The coordinates $(x,y)$ of the center of mass $CM$ of the Poncelet polygons are $D_n$-invariant meromorphic functions on $E$. Indeed these coordinates are rational functions of the coordinates of the vertices of the polygons (Lemma \ref{CMcoord}), which, in turn, are rational functions of the rational parameter on the conic, which is a meromorphic function $p$ on $E$.
Without loss of generality, assume that $\Gamma$ is a circle. From the complex projective point of view, a circle is a conic that passes through the two circular points $(1:\pm i:0) \in {\mathbb {CP}}^2$ at infinity.
When does either coordinate $x$ or $y$ of $CM$ go to infinity? It happens exactly when one of the vertices of the Poncelet $n$-gons coincides with one of the two circular points.
This occurs twice for each vertex, with multiplicity one,\footnote{We assume that the conics are in general position.} and then both $x$ and $y$ go to infinity. Therefore the functions $x$ and $y$ have $4n$ simple poles on $E$, and these poles comprise two orbits of the group $D_n$, corresponding to the two circular points.
Let $A$ be a pole corresponding to point $(1:i:0)$, and let $z$ be a local holomorphic parameter on $E$ at $A$. Then, at $A$,
$$
x(z) = \frac{a_1}{z}+ a_2 + \ldots,\quad y(z) = \frac{ia_1}{z}+ a_3 + \ldots
$$
where $a_i \in {\mathbb C}$, and the dots denote terms of degree one or higher in $z$. Likewise, let $B$ be a pole corresponding to $(1:-i:0)$. We have, in a local parameter $w$ at $B$:
$$
x(w) = \frac{b_1}{w}+b_2 + \ldots,\quad y(w) = \frac{-ib_1}{w}+ b_3 + \ldots
$$
We want to find $u,v\in{\mathbb C}$ such that the function $F=(x-u)^2+(y-v)^2$ has no poles at $A$ and $B$. Expanding out $F$ at $A$ and $B$ yields:
$$
F=\frac{2a_1}{z}\left((a_2-u)+i(a_3-v)\right) + \ldots \ {\rm and}\ \ F=\frac{2b_1}{w}\left((b_2-u)-i(b_3-v)\right) + \ldots
$$
Equating both parentheses to zero
yields two linear equations on $u$ and $v$ which we can
easily solve.
Consider the function $F$ when
$u$ and $v$ are the solutions to these simultaneous equations.
The function $F$ is $D_n$-invariant and therefore has no poles at all.
Hence $F$ is holomorphic on $E$. But the only holomorphic
functions on a compact Riemann surface are constants. So,
$(X-u)^2+(Y-v)^2=r$, where $r$ is some constant. Since
$CM(P_t)$ contains a curve of points in ${\mathbb R}^2$, the
three constands $u,v,r$ must be real. In particular,
the complexified locus intersects ${\mathbb R}^2$ in the circle
of radius $r$ centered at $(u,v)$.
\begin{remark} \label{rmk}
{\rm
(i)
Here is a variant of
Theorem \ref{main}. Consider the
polygons $Q_t$ whose vertices are the tangency
points of the sides of the Poncelet polygons $P_t$
with the ellipse $\gamma$. Then the loci $CM_0(Q_t)$
and $CM_2(Q_t)$ are ellipses homothetic to $\gamma$. This is because
the polygon $Q_t$ is also Poncelet: its sides are
tangent to the conic which is polar dual to $\Gamma$
with respect to $\gamma$.\footnote{We thank A. Akopyan for this argument.}
\newline
(ii) There is a wealth of results about the loci of various triangle centers of Poncelet triangles. The interested reader is referred to \cite{Ro} for the incenters of 3-periodic billiard trajectories inside an ellipse, to
\cite{ZKM} for the orthocenters, Gergonne, Nagel, and Lemoine points in bicentric triangles, and to \cite{Sk} for the locus of
the isogonal conjugate points to a fixed point with respect to the Poncelet triangles inscribed in a circle.
}
\end{remark}
\section{Weill's Theorem} \label{Weill}
Here is Weill's Theorem, stated in the language of Theorem \ref{main}.
\begin{theorem}[Weill]
Let $\gamma \subset \Gamma$ be a pair of nested ellipses that admit a 1-parameter family of Poncelet $n$-gons $P_t$, and let $Q_t$ be the polygons fwhose vertices are the tangency points of the sides of $P_t$ with $\gamma$. Suppose that both ellipses are homothetic to each other. Then the locus $CM_0(Q_t)$ is a single point.
\end{theorem}
We present a proof adapted from \cite{MC}.
We start with Bertrand's proof of the Poncelet
theorem in the case when both conics are circles (see, e.g., \cite{Fl} or \cite{Sh}).)
There is a subtlety here that we want to emphasize.
For the proof of Poncelet's theorem, which belongs entirely to the domain
of projective geometry, one can always
normalize that both conics are circles. However, for Theorem \ref{main}
one cannot do this because the various centers of mass are not
projectively invariant.
Consider Figure \ref{Bertrand}. Call the map $A_i \mapsto A_{i+1}$ the Poncelet map. The proof consists of constructing a cyclic coordinate $t$ on the outer circle such that, in this coordinate, the Poncelet map is $t \mapsto t+c$, where the constant $c$ depends on the mutual position of the circles. The periodicity property of such a map depends only on $c$, and not on the starting point.
\begin{figure}[ht]
\centering
\includegraphics[height=3in]{circles2.pdf}
\caption{Invariant measure for the Poncelet map.}
\label{Bertrand}
\end{figure}
What follows is, essentially, the argument
from Theorem XXX, figure 102, in I. Newton's ``Principia" where Newton studies the gravitational attraction of spherical bodies.
Assume that points $A_i$ and $B_i$ are infinitesimally close (we keep them apart to avoid cluttering the figure). The triangles $A_iB_iX$ and $A_{i+1} B_{i+1}X$ are similar, and hence
$$
\frac{|A_i B_i| }{|A_iX|} = \frac{|A_{i+1} B_{i+1}|}{|B_{i+1}X|}.
$$
Denote by $F(A)$ the length of the tangent segment from a point $A$ on the outer circle $\Gamma$ to the inner circle $\gamma$. Then, in the limit $B_i \to A_i$,
$$
\frac{dA_i}{F(A_i)} = \frac{dA_{i+1}}{F(A_{i+1})},
$$
where $d A$ is the angular measure on the circle. It follows that the measure $dt(A) := dA/F(A)$ is invariant under the Poncelet map. Hence the map is a translation in the $t$-coordinate on $\Gamma$.
Now we are ready to prove Weill's Theorem. Denote the infinitesimal angle $A_iXB_i$ by $\varepsilon$. Up to infinitesimals, the angles made by the segments $A_i A_{i+1}$ and $B_i B_{i+1}$ with the circle are equal; let $\varphi$ denote this angle. Let the outer circle be unit, and the inner have radius $R$. Then, in the limit $B_i \to A_i$,
$$
\frac{|C_iD_i|}{|A_iA_{i+1}|} = \frac{\varepsilon R}{2\sin\varphi} = \frac{R}{2} \frac{|A_iB_i|}{|A_iX|} = \frac{R}{2}\, dt(A_i),
$$
where the second equality follows from the Sine Rule for triangle $A_iXB_i$.
Consider an infinitesimal motion of the Poncelet polygon $P=(\ldots A_i A_{i+1}\ldots)$ to $P'=(\ldots B_i B_{i+1} \ldots)$. The center of mass of the points $C_i$ is displaced by the infinitesimal vector
$$
\sum_i C_i D_i = \frac{R}{2}\, dt\, \sum_i A_i A_{i+1},
$$
and the latter sum is zero because the polygon $P$ is closed.
Therefore the velocity of the center of mass vanishes, and it remains stationary.
\medskip
\noindent
{\bf Acknowledgments}. We are deeply grateful to Mr. X for sharing with us the historical letter from his family archive. Many thanks to A. Akopyan, I. Dolgachev, D. Ivanov, J. Jeronimo, S. Lvovsky, S. Markelov for useful discussions, and to A. Zaslavsky for making available their harder-to-find article \cite{ZC}.
The authors were inspired by Martin Gardner's essay \cite{Ga}.
The first author was supported by NSF grant DMS-1204471.
The second author was supported by NSF grants DMS-1105442 and DMS-1510055.
| 1fbc93d87ad0e2012cd6e8bd32f81fa4244d7b84 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
\label{sec:intro}
\begin{figure}
\begin{center}
\begin{tabular}{c}
\includegraphics[width=10cm]{img/AGN_spectra.pdf}
\end{tabular}
\end{center}
\caption{\label{fig:sy1_2} A spectrum of a Seyfert 1 galaxy (blue) and a Seyfert 2 galaxy (red) taken with the VLT/X-SHOOTER instrument as part of the LLAMA survey\cite{davies2015}. Both AGN types show narrow permitted and forbidden emission lines, but only the type 1 AGN also has broad permitted lines and a strong blue continuum. The sub-classification of Seyfert 1 galaxies (1.2, 1.5, ...) is based on emission line ratios.}
\end{figure}
Optical spectra of active galactic nuclei (AGNs) can be classified into two distinct types depending on whether or not broad permitted emission lines are seen (Fig.~\ref{fig:sy1_2}). A Seyfert~1 (1.2, 1.5, 1.8, 1.9 or 1i) is an AGN where broad emission lines are detected (at variable strengths and wavelengths) while a Seyfert~2 galaxy does not show detectable broad emission lines. Since it has been observed that Seyfert~1 galaxies are more often seen face-on than Seyfert~2 galaxies, it has been speculated that obscuration by circum-nuclear dust may be responsible for the observed dichotomy\cite{keel1980}. Strong support has been given to the idea when Antonucci \& Miller showed in 1985\cite{antonucci1985} that broad lines are seen in the polarized light of the nearby prototypical Seyfert~2 galaxy NGC~1068. They interpreted these as scattered light from the intrinsic ``Seyfert~1'' nucleus of the source whose line of sight is blocked off by a geometrically and optically thick dusty disk, that was later\cite{krolik1986} dubbed ``torus''. This unification of Seyfert~1 and 2 galaxies by means of a dusty torus is a very popular idea and much research has been devoted to observe and model this dust distribution. We refer to Netzer 2015\cite{netzer2015} for a comprehensive recent review of AGN unification and will focus here on the contributions from infrared interferometry\footnote{More specifically, we will focus on long-baseline interferometry. We note that there are also two publications applying speckle interferometry on the nearby prototyptical AGN NGC~1068 in the H and K bands\cite{wittkowski1998,weigelt2004}.}.
The question of whether the ``unification'' model holds, has also implications for our understanding of the evolution of galaxies in general. A popular idea posits that mergers produce AGNs of high luminosity, i.e. quasars, which are thought to be first in an obscured and later in an unobscured phase\cite{sanders1988,hopkins2008}. Such an evolutionary scenario may not be applicable for lower luminosity, secularly evolving, systems but it is clear that the circum-nuclear gas and dust must play a key role in regulating the feeding of the AGN. A physical understanding of this region is therefore essential for grasping the effects AGN activity may have on the evolution of their host galaxies.
The ``torus'' picture and models based on that picture have been successful in explaining a number of observations including the detection of polarized broad lines\cite{ramosalmeida2016}, the collimation of ionization cones\cite{fischer2013}, its correspondence with the fraction of obscured sources\cite{maiolino1995} and the overall SED from the near- to far-infrared\cite{netzer2016}.
On the other hand, recent observations have also raised some questions. It is, for example, unclear whether ionization cones necessarily have to be collimated by the torus\cite{prieto2014}. Also, the detection of ``true'' type 2 objects, while rare, shows that some AGNs may not have a broad-line emitting region altogether\cite{shi2010} and modeling of the integrated ``torus'' SEDs (and later also of interferometric visibilities) have shown that the ``tori'' of type 2 Seyfert galaxies intrinsically have a higher covering factor and a larger number of dust-containing gas clumps\cite{ramosalmeida2011,elitzur2012,lopezgonzaga2016b}. These results already rule out the strictest (and astrophysically questionable) versions of ``unification'' requiring an identical ``torus'' in every AGN.
It it also challenging to explain the small, if any, anisotropy in the mid-IR emission from the ``torus'' at a given X-ray luminosity: Type 1 and type 2 AGNs essentially follow the same mid-IR--X-ray relation from low to high luminosities\cite{lutz2004,gandhi2009,asmus2011,asmus2015}. Apart from these observational complications, it is also unclear theoretically how a geometrically thick torus should exist dynamically in a disk configuration. If the dust were smoothly distributed and thermal random motion provided support against gravity, the large temperatures involved would immediately destroy the dust. The dust therefore has to be organized in clumps\cite{krolik1988}. What stabilizes these clumps against the surrounding medium or if they are stable at all is still an open question.
Much of the uncertainty about the geometry and dynamics of the ``torus'' comes from the fact that the circum-nuclear dust in AGNs is usually unresolved in single-dish high-resolution images\cite{asmus2014} -- a deficiency that infrared interferometry has partly solved in the last decade.
\section{Interferometric AGN observations}
\label{sec:obs}
\begin{table}[htbp]
\centering
\begin{tabular}{llll}
\hline
Reference & Interferometer & \# & summary of result \\
\hline
Swain et al. 2003\cite{swain2003} & KI & 1 & Marginally resolved emission in NGC~4151\\
Wittkowski et al. 2004\cite{wittkowski2004} & VINCI & 1 & \begin{tabular}{@{}l@{}}Low near-IR visibility for NGC~1068 argues for two-component \\ model\end{tabular}\\
Jaffe et al. 2004\cite{jaffe2004} & MIDI & 1 & Resolved two components of warm and hot dust in NGC~1068\\
Poncelet et al. 2006\cite{poncelet2006} & MIDI & 1 & Re-analysis of the Jaffe et al. 2004 MIDI data, find no hot dust\\
Meisenheimer et al. 2007\cite{meisenheimer2007} & MIDI & 1 & Nucleus of Centaurus~A: a dusty disk and synchrotron emission\\
Tristram et al. 2007\cite{tristram2007b} & MIDI & 1 & \begin{tabular}{@{}l@{}}Two-component structure of nuclear dust in Circinus, disk \\ component is warm and co-aligned with maser disk\end{tabular}\\
Beckert et al. 2008\cite{beckert2008} & MIDI & 1 & Nuclear dust in NGC~3783 consistent with clumpy torus model\\
Kishimoto et al. 2009a\cite{kishimoto2009} & MIDI \& KI & 4 & Evidence for a common radial structure in AGN tori\\
Raban et al. 2009\cite{raban2009} & MIDI & 1 & \begin{tabular}{@{}l@{}}Two-component structure of nuclear dust in NGC~1068, disk \\ component is hot and co-aligned with maser disk\end{tabular}\\
Tristram et al. 2009\cite{tristram2009} & MIDI & 8 & \begin{tabular}{@{}l@{}}Mid-IR sizes roughly scale with $\sqrt{L}$, no clear distinction \\ between type 1 and type 2 sources\end{tabular}\\
Burtscher et al. 2009\cite{burtscher2009} & MIDI & 1 & \begin{tabular}{@{}l@{}}The nuclear dust in the Seyfert 1 galaxy NGC~4151 has similar \\ properties as in Seyfert 2 galaxies\end{tabular}\\
Kishimoto et al. 2009b\cite{kishimoto2009b} & KI & 4 & \begin{tabular}{@{}l@{}}Interferometrically derived near-IR radii are slightly larger than\\ reverberation-based radii and therefore likely probing the\\ sublimation radius\end{tabular}\\
Pott et al. 2010\cite{pott2010} & KI & 1 & \begin{tabular}{@{}l@{}}No change in near-IR size of circum-nuclear dust in NGC~4151 \\ despite variable luminosity\end{tabular}\\
Burtscher et al. 2010\cite{burtscher2010} & MIDI & 1 & New mid-IR visibilities of Cen~A do not fit well to a dust disk\\
Kishimoto et al. 2011a\cite{kishimoto2011} & KI & 8 & Sublimation radius scales with $\sqrt{L}$\\
\begin{tabular}{@{}l@{}}Tristram \&\\ Schartmann 2011\end{tabular}\cite{tristram2011} & MIDI & 10 & \begin{tabular}{@{}l@{}}Differences in mid-IR sizes between type~1 and type~2 sources,\\ expected from models, are not seen observationally.\end{tabular}\\
Kishimoto et al. 2011b\cite{kishimoto2011b} & MIDI & 6 & Half-light radius in the mid-IR independent of luminosity\\
Weigelt et al. 2012\cite{weigelt2012} & AMBER & 1 & Marginally resolved near-IR emission in NGC~3783\\
H\"onig et al. 2012\cite{hoenig2012} & MIDI & 1 & \begin{tabular}{@{}l@{}}Majority of mid-IR emission originates from optically thin dust\\ in the polar region in NGC~424 and is part of the outflow\end{tabular}\\
H\"onig et al. 2013\cite{hoenig2013} & MIDI & 1 & Detection of dust in the polar region of NGC~3783\\
Burtscher et al. 2013\cite{burtscher2013} & MIDI & 23 & \begin{tabular}{@{}l@{}}MIDI AGN Large Programme results: half-light radius in the\\ mid-IR scales with luminosity, but with large scatter;\\ tori show a large diversity in intrinsic structure\end{tabular}\\
Kishimoto et al. 2013\cite{kishimoto2013} & KI & 7 & Evidence for a receding dust sublimation region in NGC~4151\\
Tristram et al. 2014\cite{tristram2014} & MIDI & 1 & \begin{tabular}{@{}l@{}}Updated model for the Circinus~galaxy including data from \\ shorter AT baselines; two-component structure confirmed with \\ larger structure in the polar direction; SED model predicts \\ sub-mm flux precisely as measured with ALMA;\\ no evidence for large amounts of cold gas\end{tabular}\\
Lopez-Gonzaga et al. 2014\cite{lopezgonzaga2014} & MIDI & 1 & \begin{tabular}{@{}l@{}}Updated model for NGC~1068 including data from shorter\\ AT baselines; two component structure confirmed with \\ larger structure in the polar direction\end{tabular}\\
Lopez-Gonzaga et al. 2016\cite{lopezgonzaga2016b} & MIDI & 23 & \begin{tabular}{@{}l@{}}Modeling shows that the observed $(u,v)$ coverages only allow\\ to detect elongations in 7/23 sources.\\ 5/7 are found to be significantly elongated, all in polar direction.\end{tabular}\\
\hline
\end{tabular}
\caption{Summary of all publications using long-baseline interferometry for studying AGNs (either data publication or new analysis). The third column gives the number of sources involved in the particular study.}
\label{tab:results}
\end{table}
Since the first interferometric observation of AGNs in the infrared\cite{swain2003} with the Keck Interferometer\cite{vasisht2003}, such observations have also been successfully performed with three interferometers at the VLTI: VINCI\cite{kervella2000,kervella2003}, MIDI\cite{leinert2003} and AMBER\cite{petrov2007}. These interferometers have in common that they are sensitive at near- or mid-infrared wavelengths, are assisted by AO systems and that they are fed by large, 8-10m class telescopes.
Observing AGNs by means of infrared interferometry is ``hard'' because AGNs are not very bright compared to more typical targets of stellar interferometers. For example, NGC~1068 and the Circinus~galaxy, the brightest AGNs in the southern sky, offer a flux density of about 10~Jy at 12~$\mu$m, comparable to the faint end of objects usually studied with this method\cite{leinert2004,boley2013}. All other targets are at least about ten times fainter. The situation is even more severe in the near-infrared where the atmosphere changes much more rapidly, requiring much shorter fringe exposures and fast AO performances. In addition, it turned out that the angular size of the structure in the near-infrared was found to be quite small at least for type 1 AGNs where AO works well with the optically bright nucleus. Thus, with the currently available baseline lengths, the near-infrared structure is only marginally resolved, except for NGC~1068 which is the only type 2 AGN successfully observed with near-infrared interferometry so far.
\begin{figure}
\begin{center}
\begin{tabular}{c}
\includegraphics[width=10cm]{img/MIDI_LP_flux_histogram.pdf}
\end{tabular}
\end{center}
\caption{\label{fig:midi_lp_flux_histogram} Histogram of the total 12 $\mu$m fluxes of AGNs successfully observed with MIDI. Compared to the two moderately bright sources, NGC~1068 and the Circinus~galaxy, all other sources are about a factor of ten fainter.}
\end{figure}
Most AGNs were therefore studied in the mid-infrared with MIDI at the VLTI, which was the only mid-IR interferometer so far that was sensitive enough to observe AGNs. There are, however, a number of technical challenges for studying such faint sources in the thermal infrared, i.e. in the atmospheric window at 8--13 $\mu$m. One of the difficulties in these observations was the fact that the standard visibility calibration is a very noisy quantity for faint mid-infrared sources since it involves comparing the correlated with the uncorrelated flux of the target. The latter is very difficult to measure due to the intense and fast varying background in the mid-infrared, aggravated by the fact that MIDI observes the sky after about 20 reflections on warm mirrors along the VLTI beam train. This can be circumvented by comparing directly the correlated fluxes, which are less sensitive to this noise since the noise is largely uncorrelated. But when comparing fluxes directly, it is necessary to know the stability of the atmospheric transfer function as well as the total flux of the target from a separate, but ideally close-in-time, observation. An additional challenge for the group-delay fringe tracking mechanism of MIDI is that coherence losses start to become a significant problem for a signal-to-noise-ratio (SNR) lower than about 10. This problem was mitigated by measuring the coherence loss on ``fake weak'' sources (i.e. bright sources mixed with noise) and deriving a correction function to apply to science data. Both of these calibration steps have been successfully employed for the faint AGNs observed as part of the MIDI AGN Large Programme\cite{burtscher2012,burtscher2013}, but see also an alternative approach for faint-object calibration with MIDI\cite{kishimoto2011b}.
\section{Results and Discussion}
\label{sec:results}
Until now, about 40 AGNs have been studied at near- and mid-infrared wavelengths with interferometry, leading to a total of 24 refereed papers. A short summary of all these publications is given in Tab.~\ref{tab:results} and here below we highlight a few of them and give a more general summary.
\subsection{Near-infrared: The dust sublimation radius}
Near-infrared interferometry, mostly at the Keck Interferometer, has shown that the majority of the nuclear near-infrared emission originates from thermal radiation of dust at or very close to the sublimation radius\cite{kishimoto2009b} and that this radius scales with $\sqrt{L}$, where $L$ is the luminosity of the AGN\cite{kishimoto2011}. These sizes are slightly larger than those derived from earlier observations of this region using dust reverberation mapping\cite{suganuma2006}, but the discrepancy is likely due to the well-known difference between flux- and response-weighted radii, respectively\cite{kishimoto2009b}. A similar effect is seen when comparing the H$\beta$ radii of the broad line region derived from reverberation mapping\cite{bentz2009b} and from comparing integrated spectroscopy with photoionization calculations (Schnorr-M\"uller et al., submitted).
\subsection{Mid-infrared: A disk and a wind}
\subsubsection{The best studied object: The Circinus galaxy}
The Circinus~galaxy is the ideal target for interferometric observations from Cerro Paranal. It is relatively bright in the mid-IR, highly resolved and, due to its southern declination, offers a nearly circularly symmetric $(u,v)$ coverage allowing to probe many different position angles at various spatial scales. Thanks to its highly resolved structure and its relatively large signal-to-noise ratio, the nuclear mid-infrared flux of the Circinus~galaxy has been dissected into three distinct components (see Fig.~\ref{fig:circinus})\cite{tristram2007,tristram2014}:
\begin{enumerate}
\item An elongated component of about 2~pc in size, containing the majority ($\sim 80\%$) of the flux. It seems to trace the southern edge of the near-side (north-west) ionization cone which also shows an enhanced flux in optical images\cite{wilson2000}. The observed wavelength-differential (or chromatic) phases observed for the Circinus~galaxy can be explained by introducing a strong gradient in the Silicate absorption feature of this component. This is consistent with the orientation of the host galaxy (i.e. the north-west is the near side with lower extinction).
\item A highly elongated component of about 1 pc full-width at half-maximum (FWHM). It is oriented roughly perpendicularly to the large component and in the same direction as the emission traced by 22~GHz water masers. We therefore identify it with the highly inclined and warped maser disk, but the proof of this co-location will only come from the next generation of interferometers.
\item An unresolved component containing $\sim 5\%$ of the total flux, potentially caused by radiative transfer effects in the dusty warped maser disk (Jud et al., submitted).
\end{enumerate}
\begin{figure}
\begin{center}
\begin{tabular}{c}
\includegraphics[width=10cm]{img/circinus_modelimage.pdf}
\end{tabular}
\end{center}
\caption{\label{fig:circinus} Model image for the VLTI/MIDI visibilities of the Circinus~galaxy from Tristram et al. 2014\cite{tristram2014}. The nuclear mid-IR flux is decomposed into three components: a 2-pc long, polar-elongated component with a large Silicate depth gradient (purple-green) contains most of the flux; a disk component (FWHM $\sim$ 1 pc) with a large axis ratio at the same position angle as the warped water maser disk (red--green-blue lines) and some unresolved flux (the bright spot at the blue end of the maser disk).}
\end{figure}
A detailed model image and flux decomposition has also been achieved for the only other bright AGN target for MIDI, the prototypical Seyfert~2 galaxy NGC~1068\cite{jaffe2004,raban2009,lopezgonzaga2014}, but it suffers from a very unfortunate $(u,v)$ coverage due to its declination near $0^{\circ}$. The usage of the VLTI Sub-Array (VISA), i.e. the smaller Auxiliary Telescopes (ATs), has allowed us to better characterize the extended component in this galaxy and determine its orientation in the polar direction as well. Attempts to use the re-locatable AT baselines to improve the $(u,v)$ coverage also for south-eastern baselines have failed, however, since correlated fluxes in this direction were too low for MIDI to rely on its internal group-delay tracking. Using PRIMA as a fringe tracker for MIDI\cite{mueller2010,pott2012} did not improve the situation since PRIMA was not sensitive enough to track fringes at this $(u,v)$ position either\footnote{The combination of PRIMA+MIDI worked well for stars, but not for AGNs whose typical nuclear $K-N$ colors are very red: 4--7 magnitudes (mag) for type 1 AGNs and 6--10 mag for type 2 AGNs\cite{burtscher2015}.}.
\subsubsection{Extending the sample}
Extending the sample of AGNs observed with MIDI proved difficult because the next brightest sources are about ten times fainter than the two sources mentioned above (see Fig.~\ref{fig:midi_lp_flux_histogram}). With a typical noise level in the correlated flux of about 0.05~Jy, these sources were observed at signal-to-noise ratios lower than 10 which caused artifacts in the group delay tracking (see above) that needed to be corrected.
Apart from the low fluxes, the sources of this extended sample were also much more compact than expected. For the inner edge of the dust distribution, $r_{\rm in}$, the scaling of the radius with luminosity $L$ is well established (see above and Tab.~\ref{tab:results}) and follows a $\sqrt{L}$ law over about four orders of magnitude in UV luminosity \cite{kishimoto2009b}. With the observed flux $F \propto L/D^2$ ($D$ being the distance to the source) and the observed size (angle) of the structure $\theta_{\rm in} \propto r_{\rm in}/D$, the angle the source subtends scales $\propto \sqrt{F}$. If all sources had the same intrinsic structure, we would expect the radius measured in the mid-infrared $\Theta_{\rm MIR}$ to scale similarly $\propto \sqrt{F_{\rm MIR}}$ where $F_{\rm MIR}$ is the observed flux in the mid-infrared. The fainter sources observed with MIDI do, however, not follow such a simple scaling relation. This issue has caused some controversy and led to papers claiming\cite{tristram2011} that the mid-IR size scales with $\sqrt{L}$, much flatter than this\cite{kishimoto2011b} and most recently and with the largest number of sources: that a relation exists but with a large intrinsic scatter\cite{burtscher2013}.
\begin{figure*}
\centering
\includegraphics[width=12cm]{img/LP_rescale.pdf}
\caption{\label{fig:LP_rescale} Visibilities of the relatively bright Circinus~galaxy compared with the sources of the MIDI AGN Large Programme\cite{burtscher2013}. The baselines are normalized to the mid-IR flux of the Circinus~galaxy under the expectation that the mid-IR size scales $\propto \sqrt{F}$. The color encodes the mid-IR luminosity of the sources, compared to the mid-IR luminosity of the Circinus~galaxy $L_{\rm circ}$. A typical error bar is shown below the legend. It is unclear what causes the large intrinsic diversity of dust structures.}
\end{figure*}
One way to visualize the observed diversity in {\em intrinsic} dust structure is to plot the observed visibilities\footnote{Note that we actually observe and model correlated fluxes directly for reasons of better calibratability (see above). The visibilities here are generated from the ratio of the correlated flux and the total flux, where the latter has been observed separately with a single-dish mid-IR instrument, mostly VLT/VISIR.} as a function baseline normalized to Circinus, see Fig.~\ref{fig:LP_rescale}. This normalization takes into account the expected scaling with flux, so that the normalized baseline is given by the observed baseline $\times \sqrt{F/F_{\rm Circinus}}$. As can be seen, the fainter sources (plotted as small circles) are less well resolved than expected from scaling the Circinus visibilities (large red squares) according to the sublimation radius of dust. Alternatively, one can also look only at the lowest visibility as a function of intrinsic resolution (i.e. resolution scaled to the sublimation radius)\cite{burtscher2013}, see Fig.~\ref{fig:fp}. This figure shows that the fraction of unresolved source (the ``point source fraction'') is largely independent of resolution. What drives the intrinsic diversity in the dusty environment of the AGN is not clear yet. The luminosity seems to have only a small effect (see Fig.~\ref{fig:LP_rescale}). For the inner radius, a possible scaling with the radio-to-optical luminosity has been explored and a tentative correlation has been found\cite{kishimoto2011} indicating that the structure of the circum-nuclear dust could be related to the accretion mechanism on sub-parsec scales.
\begin{figure}
\begin{center}
\begin{tabular}{c}
\includegraphics[width=10cm]{img/fp_rin_nocase1.pdf}
\end{tabular}
\end{center}
\caption{\label{fig:fp} Lowest visibilities (``point source fractions'') measured for the sample of 23 AGNs presented in Burtscher et al. 2013\cite{burtscher2013}. These visibilities are shown against the intrinsic resolution, i.e. the resolution in terms of the sublimation radius of dust. While the intrinsic resolution for the faint sources is comparable, they show a large diversity in unresolved flux. Type 1 AGNs are plotted in blue, type 2 AGNs in red.}
\end{figure}
The reason for the larger scatter in the (normalized) mid-infrared sizes compared to the near-infrared is that the sublimation temperature sets a physical inner boundary while a unique temperature-radius relation does not exist for the warm dust. This is consistent with clumpy torus models that predict a very shallow dependence of brightness temperature with radius (e.g. Fig.~10 in Tristram et al. 2007\cite{tristram2007b}).
\subsection{Dust in the polar region}
Perhaps the most spectacular result from AGN interferometry is the detection of dust in the polar region of several nearby AGNs\cite{hoenig2012,hoenig2013,tristram2014,lopezgonzaga2014}. Initially this was interpreted as the hot inner funnel of a clumpy toroidal distribution of dust (some torus models predict `X' shaped or polar emission from azimuthal, ``torus''-like, dust structures)\cite{tristram2014}. This interpretation suggested that radiative transfer effects played a large role in the apparent distribution of mid-infrared flux. A new interpretation of the polar emission as optically-thin emission from the sheath of the outflowing ionization cone\cite{hoenig2012,hoenig2013} was supported by a high-resolution SED compiled for the Circinus~galaxy. A high-resolution ALMA continuum measurement\cite{hagiwara2013} agreed extremely well with the flux predicted from modeling the warm dusty structure. This implied that the sub-mm flux is the Rayleigh-Jeans tail of the warm dusty structure and there is not any more cold dust that is hidden in the mid-IR.\cite{tristram2014}
A study assessing the frequency of polar dust emission examined the detectability of elongated structures given the very limited $(u,v)$ coverages of the observations in combination with observed noise levels and the level of unresolved flux \cite{lopezgonzaga2016a}. It was found that, given these observational limitations, only in seven out of 23 sources can a possible elongation (axis ratio and position angle) be constrained. And among these seven sources there are five that actually have elongated emission, all of them oriented much closer to the polar direction than orthogonal to it. This is consistent with observations on larger scales that typically find dust in the narrow-line region as well\cite{cameron1993,schweitzer2008,asmus2016}. However, single dish imaging also always finds a strong nuclear point source which only interferometry could resolve to see that most of the circum-nuclear parsec-scale dust emission originates from the outflow region, rather than from a ``torus''.
\section{Conclusions}
Infrared interferometry, and especially mid-infrared interferometry with MIDI at the VLTI, has been the key technology to dissect the otherwise unresolved nuclear dust structure in AGNs. It has become clear that the majority of the dust emission, and apparently also of the dust mass, is located in or near the outflow cone and possibly constitutes a dusty sheath around the ionizing cone, perhaps contributing to its collimation. This naturally solves the long-standing scale-height problem of the ``torus''. If the dust emission in the polar region is optically thin in the mid-IR as the good agreement of the sub-mm continuum with the extrapolated mid-IR model suggests, this would also solve the problem of essentially isotropic mid-IR emission as seen in the mid-IR -- X-ray correlation. The collimation of the ionizing radiation would then happen at the base of the dusty outflow (requiring only relatively little obscuration). The same structure could then also be responsible for the observed dichotomy in Seyfert spectra. Alternatively, the thin disk may provide the obscuration required to block our view onto the broad line region in Seyfert~1 galaxies. Its scale height as measured in the mid-IR and seen in the maser disks may not be sufficient, although the warp in the maser disk may help\cite{lawrence2010}. In the inner parts, this disk may have a puffed-up inner rim as suggested by the relatively large near-IR covering factors\cite{hoenig2013} and thus provide the opening angle required to explain the fraction of obscured sources.
It seems not exaggerated to blame mid-IR interferometry (and VLTI/MIDI in particular) for ``killing'' the postulated AGN torus by dissecting it into a thin disk and a polar-elongated outflow, none of which are represented in the classical AGN torus cartoon.
\section{Outlook}
The two main interferometric instruments used for observing AGNs, the Keck Interferometer in the near-infrared and VLTI/MIDI for the mid-infrared, are no longer operational, but a new generation of interferometers is arriving at the VLTI. GRAVITY\cite{eisenhauer2014} is already producing first science results and will be used to study AGNs in the near-infrared soon. With its superb sensitivity and fringe-tracking capabilities, it will reach of order 20 AGNs. The scientifically perhaps most thrilling prospect for AGN science is to use spectro-astrometry to measure the size and geometry of the broad line region\footnote{We note early attempts of using spectro-astrometry with VLTI/AMBER by R. Petrov et al.\cite{petrov2012}}. It will also image the inner rim of dust, measuring its size and orientation. In combination with infrared continuum reverberation mapping the measured angular size of the sublimation radius provides an independent distance estimate for nearby AGNs\cite{hoenig2014}.
The next generation mid-IR interferometer MATISSE\cite{lopez2014}, scheduled to arrive on Paranal in about two years, will offer four-telescope imaging interferometry in the L, M and N bands. With real images and higher resolution (shorter wavelengths), it will provide more reliable estimates for the geometry of the circumnuclear dust, possibly revealing what drives the dichotomy in its intrinsic properties.
In order to observe a significant number of AGNs, however, MATISSE will require a sensitive external fringe tracker, such as provided by the GRAVITY instrument. These authors therefore welcome very much ESO's efforts to make GRAVITY available as a fringe tracker for MATISSE (see the ``GRA4MAT'' project in these proceedings).
\section{Acknowledgements}
The authors would like to thank the VLTI teams at ESO for tremendous support over many years of observations. L.B. is supported by a DFG grant within the SPP 1573 ``Physics of the interstellar medium''
| ad23bcd4a53d3f50002d9a78ed9ecd501dd91b3e | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
The idea of optimally controlling the dynamical behavior of microsystems described by quantum mechanics has been an explicit dream for more than $50$ years \cite{Warren1993}. Among all the approaches to solving the resulting optimization problems, model-based optimal control relies on computer simulations of quantum dynamics and numerical optimization algorithms \cite{Schirmer2011}. The early attempt to seek the counterpart of the classical optimal control theory in quantum systems may be traced to Rabitz \textit{et al.} \cite{Peirce1988}, in which they explicated basic concepts of quantum optimal control theory which paved the way for later approaches. From then on, the studies have been enriched by numerous optimization algorithms with various applications.
Generally, a mathematical model of the underlying quantum system is required for designing optimal controls, and the strategies for solving this model-based optimization problem can be roughly divided into two classes. One approach is based on the necessary conditions derived from Pontryagin's Maximum Principle, i.e. "considering a pointer moving back and forth within the time interval" \cite{Schirmer2011}. For that, Zhu, Sundermann, and Ho \textit{et al.} proposed several quadratically convergent iterative algorithms \cite{Zhu1998a,*Zhu1998,Sundermann1999,Ho2010,Ho2011,Liao2011}; Sklarz, Reich, and Schirmer \textit{et al.} developed different kinds of Krotov algorithms \cite{Sklarz2002,Reich2012,Schirmer2011}. The other approach is to eliminate the "Quantum Constrains", i.e. eliminating the equality constrain of Schr\"odinger equation (SE) and apply classical optimization algorithms \cite{[][ pp. $215$-$288$]Boyd2004,Machnes2011}. The quantum gradient algorithm was first introduced by Shi \textit{et al.} based on difference method \cite{Shi1990}, and it was improved by Khaneja \textit{et al.} through a much more convenient way for calculating the gradient, saying the Gradient Ascent Pulse Engineering (GRAPE) method \cite{Khaneja2005}; besides, algorithms such as the Newton algorithm \cite{Kosloff1994,Palao2002,Palao2003}, quasi-Newton algorithm \cite{Borzi2002,Hohenester2007}, and various evolutionary algorithms \cite{Zahedinejad2014} were also proposed.
The efficiency of model-based optimizations is usually determined by two factors. One is how fast the optimization algorithm can converge to the desired solution, and the other is how fast the dynamical evolution under a given control protocol can be calculated. In order to reduce the number of iterations, Ho \textit{et al.} proposed the two-point boundary-value quantum control paradigm (TBQCP) \cite{Ho2010,Ho2011,Liao2011}, Eitan \textit{et al.} combined the Krotov and quasi-Newton methods \cite{Eitan2011}, and Watts \textit{et al.} constructed a easily computable mathematical expression for the target function \cite{Watts2015,Goerz2015}. More importantly, the simulation of quantum dynamical evolution need to be fast. Palao \textit{et al.} restricted the objective to only the states used directly as registers \cite{Palao2002,Palao2003}, Yip \textit{et al.} proposed the concept of "propagation toolkit" to simplify control functions \cite{Yip2003}, Arai \textit{et al.} introduced the reduced time-evolution operators \cite{Arai2015}.
In our previous work \cite{arXiv}, we proposed the Pulse Width Modulation (PWM) method for efficiently simulating time-dependent systems. It transforms an arbitrary bandwidth-limited Hamiltonian into a sequence of on and off "Hamiltonian Pulses" (called Hamiltonian in PWM form), and reduces the simulation time by a large extent. In this paper, we design a gradient algorithm based on PWM and study the resulting improvement on the optimization efficiency with respect to multi-qubit quantum systems. The rest of this paper is organised as follows. {Section~II} briefly reviews the PWM method, and {Sec.~III} presents the problem, derives the optimization algorithm for one-control and multi-control cases. We apply the algorithm to $3$-qubit and $6$-qubit systems, and compare computational efficiency with the GRAPE algorithm in {Sec.~IV}. We show how to smooth the optimized control function by example of transforming rectangular pulses into Gaussian pulse train in {Sec.~V}. Finally, we summarise all the results and draw the conclusions in {Sec.~VI}. In addition, {App.~A} shows different realizations of PWM transformation and compares PWM with $2$nd-order Split Operator Method (SPO), and {App.~B} presents the detailed Hamiltonian for the D-Norleucine we used for numerical simulation.
\section{Brief review of PWM}\label{PWM}
Pulse Width Modulation (PWM) is a well accepted concept in control technology in power electronics \cite{Holmes2003,[][ pp. $25$-$61$]Vasca2012}, which is based on the fact that any function $u(t)$ with finite frequency bandwidth can be well approximated at any precision by a sequence of rectangular pulses with unanimous amplitude $\varepsilon=\max_{t}{|u(t)|}$ but different pulse width $t_p$. To apply PWM to a quantum system, we define the Frequency Band $\Delta=[\omega_{min},\omega_{max}]$ as the interval from the minimum to the maximum frequency of all the time-dependent variables in the Hamiltonian, and the Frequency Scope $\Omega= M\omega_{min}$ as the frequency limit one should care about, where $M$ is called the Pulse Number within every time interval ${2\pi}/{\omega_{min}}$.
Consider an arbitrary time-dependent Hamiltonian in the semiclassical form
\begin{equation}\label{TD}
\hat H(t)=\hat H_0+\sum_{i=1}u_{i}(t)\hat H_{i},
\end{equation}
where $H_0$ and $H_i$'s are time-independent Hamiltonians, $u_{i}(t)$'s are time-dependent real functions with finite bandwidth. The procedure of applying PWM in quantum system is based on {Equal Integral Area Principle} (EAP) \cite{arXiv}; in short, we divide every time interval $2\pi/\omega_{min}$ into $M$ pieces with equal length $\tau$, in each we replace $u_i(t)$ by a rectangular function with amplitude $\varepsilon_i$ and width $t_{p,i}^{(k)}$ which contains equal integral area. In this way, the time-dependent Hamiltonian is transformed into the PWM form
\begin{equation}\label{TID}
\hat H(t)={\hat H}_0+ \sum_{i=1}s_{i}(t)\varepsilon_i{\hat H}_i,
\end{equation}
where $s_{i}(t)$ is a sign function which switches among $0,\pm1$; $\varepsilon_i\geq\max_{t}{|u_i(t)|}$ is a constant.
We further define anti-PWM transformation as transforming Eq.~\eqref{TID} into Eq.~\eqref{TD}, which replaces rectangular pulse in each time interval $\tau$ by a constant with the same integral area; moreover, the Piecewise Constant (PWC) result may be further smoothed through EAP, which may be desired in many cases \cite{Sundermann1999,Zahedinejad2015}. However, as we will see the anti-PWM transform cannot always work well when one sets a relatively large time interval $\tau$ for simplifying the optimization in complex systems, how to smooth PWM pulses through smooth pulses would be discussed in {Sec.~V}.
\section{Algorithm design}\label{DES}
Consider an arbitrary time-dependent Hamiltonian in the form of Eq.~\eqref{TD}, we name $u_i(t)$ as the $i$-th control signal, $\varepsilon_i$ as the bond on the $i$-th control field. Our aim is to tailor $u_i(t)$ to minimize the following defined objective $\rm IF$ at time $t=T$
\begin{equation}\label{IF}
{\rm IF}=1-\frac{1}{2^m}Re{\left\{tr\left[{W}^{\dagger}{U}(T,0)\right]\right\}},
\end{equation}
where ${W}$ is the target operation, $\rm IF$ is called the "Infidelity" in quantum information \cite{Vandersypen2005}.
Since optimizing the target Eq.~\eqref{IF} relays highly on the dynamics of the system, an efficient propagation scheme is crucial for quantum optimization. The broadly used approach is to break up the total evolution operator into small increments of duration $\tau$ in which the variation of the Hamiltonian operator is negligible, for example
\begin{equation}\label{OLD}
\hat U(t+\tau,t)=\exp{\Big[-i\tau(H_0+u(t+\frac{\tau}{2}){\hat H}_1)\Big]},
\end{equation}
where the sampling point $t+\frac{\tau}{2}$ is decided by the Exponential Midpoint Rule that usually guarantees a result with the highest precision \cite{Castro2004}. This approach is called the Piecewise Constant Scheme (PWC), which applies to nearly all the existing numerical quantum algorithms, e.g. the basic GRAPE algorithm \cite{Khaneja2005}.
However, PWM provides an alternative approach for solving SE, with the same simplicity but higher efficiency. Based on Eq.~\eqref{TID}, one can construct a corresponding time-independent SE in each interval $\tau$ and combine the results to calculate the propagator, for example
\begin{eqnarray*}
\hat U_M(t+\tau,t)&=&\exp{\Big[-it_f{\hat H}_0\Big]}
\exp{\Big[-it_{p}(H_0\pm\xi{\hat H}_1)\Big]}\\
&\times&\exp{\Big[-it_f{\hat H}_0\Big]},
\end{eqnarray*}
where $t_{p}=\varepsilon^{-1}\int_t^{t+\tau}dt'u(t')$, $t_{f}={\left(\tau-t_{p}\right)}/2$, subscript $M$ means it is calculated by PWM transformation.
Instead of using the amplitude of $u_i(t)$ as the decision variables of the system in GRAPE and all the other algorithms, the PWM approach utilizes the time intervals of rectangular pulses $t_{p,i}^{(k)}$ to be the decision variables. In the following, we derive the gradient formulas for combining PWM with the gradient descent algorithm.
\subsection{Gradient formula for one-control case}
Consider a Hamiltonian with only one control function
\begin{equation*}
{H}(t)={H}_0+u_x(t){H}_x~|u_x(t)|\leq\varepsilon_x,
\end{equation*}
we apply PWM transformation to the Hamiltonian and obtain the corresponding propagator of time $t=T$
\begin{widetext}
\begin{equation*}
U(T,0)=
e^{-iH_0t_f^{(M)}}e^{-i\left(H_0+g^{(M)}H_x\right)t_p^{(M)}}
\left(\prod_{k=1}^{M-1}e^{-iH_0\left(t_f^{(k)}+t_f^{(k+1)}\right)} e^{-i\left(H_0+g^{(k)}H_x\right)t_p^{(k)}}\right)e^{-iH_0t_f^{(1)}}.
\end{equation*}
\end{widetext}
where
\begin{equation*}
g^{(k)}=\left\{
\begin{array}{ll}
+\varepsilon_x, & \hbox{if the $k$-th pulse is positive} \\
-\varepsilon_x, & \hbox{otherwise}
\end{array}
\right.
\end{equation*}
is a hidden binary function which indicates whether the $k$-th pulse is positive or negative, $t_{p}^{(k)}$ is the time duration of the $k$-th pulses, and $t_{f}^{(k)}=\left(\tau-t_{p}^{(k)}\right)/2$. Note that the superscript turns out to be either $H_0$ or $H_0+g^{(k)}H_x$, in order to save computational resource we define
\begin{eqnarray}\label{ONEPRE}
H_0&=&D_0\Lambda_0D_0^{\dagger},\\\nonumber
H_0+g^{(k)}H_x&=&D_x\Lambda_xD_x^{\dagger},
\end{eqnarray}
\begin{eqnarray}\label{ONEV}
V_k=D_{x}\exp{\left[-it_p^{(k)}\Lambda_x\right]}D_{x}^{\dagger},
\end{eqnarray}
\begin{eqnarray}\label{ONEU}
&U_{1}=D_{0}\exp{\left[-it_f^{(1)}\Lambda_0\right]}D_0^{\dagger}\\\nonumber
&U_{k+1}=\left\{D_{0}\exp{\left[-i{(t_f^{(k)}+t_f^{(k+1)})}\Lambda_0\right]}
D_{0}^{\dagger}V_k\right\}U_{k},
\end{eqnarray}
where $k=1,\cdots,M-1$. Thus, we obtain the formula of the propagator and the corresponding gradient
\begin{eqnarray}\label{ONECON}
U(T,0)&=&D_0\exp{\left[-it_f^{(k)}\Lambda_0\right]}D_{0}^{\dagger}V_{k}U_{k}\\\nonumber
\frac{\partial U(T,0)}{\partial t_p^{(k)}} &=&\frac{-i}{2}U(T,0)\left(\left(V_{k}U_{k}\right)^{\dagger}H_x\left(V_{k}U_{k}\right)+U_k^{\dagger}H_xU_k\right)
\end{eqnarray}
\subsection{Gradient formula for multi-control case}
Consider a more general case where the Hamiltonian contains multiple control functions, e.g. the following $2$-control case
\begin{equation*}
{H}(t)={H}_0+u_x(t){H}_x+u_y(t){H}_y,~|u_{x,y}(t)|\leq\varepsilon_{x,y}.
\end{equation*}
We define:
if $t_{p,x}^{(k)}-t_{p,y}^{(k)}\geq0$:
\begin{eqnarray*}
&t_{a}^{(k)}=t_{p,x}^{(k)}, t_{b}^{(k)}=t_{p,y}^{(k)}, t_{d}^{(k)}=\frac{t_{p,x}^{(k)}-t_{p,y}^{(k)}}{2}, t_{f}^{(k)}=t_{f,x}^{(k)},\\ &H_{a}^{(k)}=g_{x}^{(k)}H_{x}, H_{b}^{(k)}=g_{y}^{(k)}H_{y};
\end{eqnarray*}
if $t_{p,x}^{(k)}-t_{p,y}^{(k)}\leq0$:
\begin{eqnarray*}
&t_{a}^{(k)}=t_{p,y}^{(k)}, t_{b}^{(k)}=t_{p,x}^{(k)}, t_{d}^{(k)}=\frac{t_{p,y}^{(k)}-t_{p,x}^{(k)}}{2}, t_{f}^{(k)}=t_{f,y}^{(k)},\\ &H_{a}^{(k)}=g_{y}^{(k)}H_{y}, H_{b}^{(k)}=g_{x}^{(k)}H_{x};
\end{eqnarray*}
where
\begin{equation*}
g_{i}^{(k)}=\left\{
\begin{array}{lll}
+\varepsilon_i, & \hbox{if the $k$-th pulse of $i$-th control is positive}\\
-\varepsilon_i, & \hbox{otherwise}
\end{array}
\right.
\end{equation*}
is a hidden binary function which indicates whether the $k$-th pulse of $i$-th control is positive or negative, $t_{p,i}^{(k)}$ for $i=x,y$ is the time duration of the $k$-th pulse of the $i$-th control, and $t_{f,i}^{(k)}=\left(\tau-t_{p,i}^{(k)}\right)/2$. In the similar way of one-control case, we obtain the corresponding propagator of time $t=T$
\begin{widetext}
\begin{eqnarray*}
U(T,0)=&&e^{-iH_0t_f^{(M)}}e^{-i\left(H_0+H_a\right)t_d^{(M)}}
e^{-i\left(H_0+H_a+H_b\right)t_b^{(M)}}
e^{-i\left(H_0+H_a\right)t_d^{(M)}}\nonumber\\
&&\times\Big(\prod_{k=1}^{M-1}e^{-iH_0\left(t_f^{(k)}+t_f^{(k+1)}\right)} e^{-i\left(H_0+H_a\right)t_d^{(k)}}
e^{-i\left(H_0+H_a+H_b\right)t_b^{(k)}}e^{-i\left(H_0+H_a\right)t_d^{(k)}}\Big)
e^{-iH_0t_f^{(1)}}.
\end{eqnarray*}
\end{widetext}
Similarly, we define
\begin{eqnarray}\label{TWOPRE}
H_0&=&D_0\Lambda_0D_0^{\dagger},\\\nonumber
H_0+H_a&=&D_a\Lambda_aD_a^{\dagger},\\\nonumber
H_0+H_a+H_b&=&D_b\Lambda_bD_b^{\dagger},
\end{eqnarray}
\begin{eqnarray}\label{TWOV}
V_{k,1}&=&D_{a}\exp{\left[-i{t_{d}^{(k)}}\Lambda_a\right]}D_{a}^{\dagger},\\\nonumber
V_{k,2}&=&D_{b}\exp{\left[-i{t_{b}^{(k)}}\Lambda_b\right]}D_{b}^{\dagger}V_{k,1},\\\nonumber
V_{k,3}&=&V_{k,1}V_{k,2};
\end{eqnarray}
\begin{eqnarray}\label{TWOU}
&U_{1}=D_{0}\exp{\left[-i{t_{f}^{(1)}}\Lambda_0\right]}D_{0}^{\dagger},\\\nonumber
&U_{k+1}=\left\{
D_0\exp{\left[-i\left(t_{f}^{(k)}+t_{f}^{(k+1)}\right)\Lambda_0\right]}D_{0}^{\dagger}
V_{k,3}\right\}U_{k},
\end{eqnarray}
where $k=1,\cdots,M-1$. Thus we obtain the formula of the propagator and the corresponding gradient
\begin{eqnarray}\label{TWOCON}
U(T,0)&=&D_0\exp{\left[-i{t_{f}^{(k)}}\Lambda_0\right]}D_{0}^{\dagger}V_{m,3}U_m\\\nonumber
\frac{\partial U(T,0)}{\partial t_{a}^{(k)}} &=&\frac{-i}{2}U(T,0)\left(\left(V_{k,3}U_{k}\right)^{\dagger}H_a\left(V_{k,3}U_{k}\right) +U_k^{\dagger}H_aU_k\right)\\\nonumber
\frac{\partial U(T,0)}{\partial t_{b}^{(k)}} &=&\frac{-i}{2}U(T,0)\Big(\left(V_{k,2}U_{k}\right)^{\dagger}H_b\left(V_{k,2}U_{k}\right)\\\nonumber
&&+\left(V_{k,1}U_{k}\right)^{\dagger}H_b\left(V_{k,1}U_{k}\right)\Big)
\end{eqnarray}
\subsection{Algorithm design}\label{EFF}
Without loss of generality, we present the whole procedure for the algorithm according to the one-control case:\\
(0) Diagonalize the Hamiltonian $H_0$ and $H_0+g^{(k)}H_x$ using Eq.~\eqref{ONEPRE}, and guess the initial pulse intervals $t_p^{(k)}$.
(1) Calculate the exponents $V_k$ and $U_k$ using Eq.~\eqref{ONEV} and \eqref{ONEU}, the propagator and the gradient using Eq.~\eqref{ONECON}.
(2) Evaluate ${\partial U(T,0)}/{\partial t_p^{(k)}}$ and update $t_p^{(k)}$ by gradient descent method (or other descent methods \cite{Boyd2004}).
(3) Evaluate the infidelity $\rm IF$ in Eq.~\eqref{IF} with these new $t_p^{(k)}$, repeat step (1)$\sim$(3) if the result is not satisfactory.
(4) Using anti-PWM transformation or smooth pulses to generate a smooth result.\\
Since Eq.~\eqref{ONEPRE} and \eqref{TWOPRE} allow one to diagonalize and calculate the exponents previously (offline), it transforms the large number of calculations of exponents with matrices on the shoulder Eq.~\eqref{OLD} into exponents with only scalars on the shoulder Eq.~\eqref{ONECON} and \eqref{TWOCON}, which may simplify the computational complexity and accelerate the simulation of quantum dynamics by a large extent; moreover, Eq.~\eqref{ONEPRE}-\eqref{ONEU} and Eq.~\eqref{TWOPRE}-\eqref{TWOU} can be recycled in Eq.~\eqref{ONECON} and \eqref{TWOCON} for calculating the time evolution and the corresponding gradient. Thus, we may expected that PWM should be very efficient for solving optimization problems, especially in complex situations.
\section{Optimization Examples}\label{NUM}
Consider the example of D-Norleucine molecule which contains six homonuclear carbon spins that can be encoded as qubits for Nuclear Magnetic Resonance (NMR) quantum computation. The system's Hamiltonian reads \cite{Zhang2015,Chen2015}
\begin{equation*}
{H}(t)={H}_0+u_x(t){H}_x+u_y(t){H}_y,~|u_{x,y}(t)|\leq\varepsilon
\end{equation*}
where the complete form of $H_0$ and $H_{x,y}$ are given in App.~\ref{HAM}, $u_{x,y}(t)$ are the control functions implemented by radio-frequency magnetic fields, $\varepsilon=2\rm MHz$ is the bound on the amplitude of control fields.
\subsection{3-qubit example with single control}\label{THREE}
\begin{figure}
\centering
\includegraphics[width=7.8cm]{F1}\\
\caption{(Color online) The infidelity $\rm IF$ of the PWM pulses with the iteration number in the optimization procedure from $10$ randomly chosen fields, for the gate control over $C_1$, $C_2$, and $C_3$ with accuracy requirement $\rm IF\leq1\times10^{-3}$ and $M=1\times10^5$.}\label{E1}
\end{figure}
\begin{figure}
\subfigure[]{\includegraphics[width=7.8cm]{F2}\label{E2}}
\subfigure[]{\includegraphics[width=7.8cm]{F3}\label{E3}}
\subfigure[]{\includegraphics[width=7.8cm]{F4}\label{E4}}
\caption{(Color online) The PWM optimized continuous field (a) in the time domain and (b) in the frequency domain, for the gate control over $C_1$, $C_2$, and $C_3$ with accuracy requirement $\rm IF\leq1\times10^{-3}$ and $M=1\times10^5$; (c) Infidelity $\rm IF$ of continuous control fields, which are transformed from those optimized PWM pulses with different pulse number $M$. The red line is fitted by the numerical samples (blue dots).}
\end{figure}
To illustrate how the PWM algorithm works, we choose the first three nuclears $C_1$, $C_2$, $C_3$ to seek an optimal solution. The target operation is to realize an $\frac{\pi}{2}$-rotation on $C_1$ around the $\langle\sigma_x\rangle$-axis at time $T=10 ms$ while remain the rest spins unchanged, saying
\begin{equation*}
{W}=\exp{\left(-i\frac{\pi}{4}\sigma_x\right)}\otimes I^2,
\end{equation*}
where $I$ is identity.
We start from $10$ randomly chosen initial fields of $u(0)$, set $M=1\times10^5$ to transform the Hamiltonian into the PWM form, and apply the gradient optimization algorithm to look for optimal solutions. Since there is no guarantee that the algorithm will converge to the global minimum \cite{Khaneja2005}, we add some random perturbations on the gradient to try to avoid potential trap. Figure \ref{E1} shows the corresponding results, where all the $10$ cases achieve the infidelity requirement $\rm IF\leq1\times10^{-3}$ within $210$ iterations. The blue line corresponds to a typical optimization procedure that achieves $\rm IF=0.84\times10^{-3}$ after $164$ iterations, where the infidelity could be extremely close to zero under the same condition when we allow more iterations, or when we apply a more sophisticated algorithm (such as the genetic algorithm). This indicates that the PWM algorithm is capable of finding highly accurate optimal solutions.
Since a continuous control field is more acceptable than a rectangular one in laboratory, we reverse the PWM transformation procedure to convert the optimized PWM control field back to the continuous field. Figure~\ref{E2} and \ref{E3} respectively show the corresponding continuous control field $u(t)$ in time domain and frequency domain, which slightly increases the infidelity but still guarantees a rather high control precision $\rm IF=0.95\times10^{-3}$. To test the performance of the continuous control fields transformed from PWM pulses with different parameter $M$, we vary $M$ from $50$ to $2.5\times10^{5}$ to optimize the same problem and anti-transformed the optimized PWM pulses into continuous fields. Figure~\ref{E4} reveals the relation between the performance of the continuous control field and the pulse number $M$ of the optimization process. We found that the anti-transformed fields could still provide a rather accurate result if the PWM time interval $\tau\leq1\mu s$ ($M\geq1\times10^{4}$) for this problem; however, it is very difficult to obtain a satisfactory continuous field when $\tau\geq1\mu s$ ($M\leq1\times10^{4}$). This can be explained by that EAP works under the condition of small time interval $\tau$, for it is equivalently using only the $1$st-order term of Magnus expansion \cite{arXiv}. Nevertheless, since in the latter case the time interval $\tau$ for preparing each pulse is relatively long, directly using on and off pulses may be applicable in contemporary laboratory conditions. On this regard, one may think the results for small pulse number $M$ is still acceptable, which further saves the optimization procedure by a large extent.
\subsection{6-qubit example with 2 controls}
\begin{figure}
\centering
\includegraphics[width=7.8cm]{F5}\\
\caption{(Color online) Infidelity ($\rm IF$) of the PWM pulses with the iteration number in the optimization procedure from $10$ randomly chosen fields, for the gate control over the whole $6$ spins with $M=1\times10^5$. The algorithm ceases at $\rm IF=3.89\times10^{-4}$ after $24$ iterations.}\label{F1}
\end{figure}
We consider a larger system with two control functions for manipulating all the six spins in the molecule, which is regarded to be very difficult for seeking optimal solutions \cite{Moore2011}. The target operation is to realize the CNOT gate on $C_1$ and $C_2$ at time $T=10s$ while remain the rest spins unchanged, saying
\begin{equation*}
{W}={U}_{CNOT}\otimes I^4,
\end{equation*}
where $I$ is identity. The time duration may touch the limit of the dephasing time in many NMR systems; nevertheless, it can be achieved in well-prepared samples and in a good experimental apparatus at reasonably high magnetic fields \cite{Vandersypen2005}. In this regard, we still focus on seeking optimal control solutions in closed systems.
We start from a randomly chosen initial fields of $u(0)$, set $M=1\times10^5$ to transform the Hamiltonian into the PWM form, and apply the gradient optimization algorithm to look for optimal solutions. Figure~\ref{F1} shows the optimization procedure which achieves the infidelity $\rm IF=3.89\times10^{-4}$ after $24$ iterations. We further changed other target operations and also met no any trap in the optimization process, which is consistent with the landscape of quantum system from the practical perspective, saying the traps of optimization are usually avoidable when the time duration is very long \cite{Zhdanov2015}. On the other hand, the PWM algorithm achieves a very high precision with only $24$ iterations. It reveals that PWM is highly efficient to handle high dimension, high accuracy, and long time interval problems which are regarded as rather difficult for other algorithms \cite{Moore2011}.
\subsection{Efficiency analysis}
\begin{figure}
\centering
\subfigure[]{\includegraphics[width=7.8cm]{F6}}
\subfigure[]{\includegraphics[width=7.8cm]{F7}}
\caption{(Color online) CPU time $t_c$ of the basic GRAPE algorithm (a) and the PWM algorithm (b) for calculating evolution $U(T,0)$ over the whole $6$ spins at different time duration $T\in[20,200]~\mu s$ and infidelity $\rm -lg(IF)\in[1,8]$.}\label{CPU}
\end{figure}
PWM speeds up quantum optimal control in accelerating the simulation of quantum dynamical evolution. To quantitatively compare the optimization efficiency of basic GRAPE algorithm and PWM based gradient algorithm, we suppose the iterations before achieving the target are of the same scale for the two algorithms, and compare the CPU time $t_c$ of the two method for calculating a desired propagator $U(T,0)$ \footnote{The CPU time $t_c$ is measured using the $tic$ and $toc$ commands in $Matlab~2014b$.}. For simplicity, we consider the $1$-control case, use the Hamiltonian of $6$-qubit example, and set the desired control function $u(t)=\sin{(\omega t)}$ for $\omega=50\rm KHz$.
Figure~\ref{CPU} displays the relation among the frequency scope $\Omega$ (which corresponds to pulse number $M$), time duration $t$, and the computational efficiency (CPU time $t_c$). Roughly speaking, PWM shows a distinct advantage in saving computational resources, at $T=20\mu s$, $\rm IF=1\times10^{-1}$ the PWM algorithm saves $29.6\%$ computational time for corresponding evolution; at $T=200\mu s$, $\rm IF=1\times10^{-8}$ the PWM algorithm saves $29.7\%$ computational time. Though compared with the low-dimensional results the computation advantage becomes more and more small with the increase of system dimension \cite{arXiv}, PWM still shows a nearly $30\%$ time saving in the $6$-qubit system. Since a quantum optimization problem usually requires solving SE many times, and solving that for high-dimensional problems needs extremely long computation time, the $30\%$ time saving in fact saves a large quantity of computational resources and makes it possible for optimizing complex systems.
\section{Smooth PWM Pulses}\label{GAU}
\begin{figure}
\subfigure[]{\includegraphics[width=8cm]{F8}\label{G1}}
\subfigure[]{\includegraphics[width=8cm]{F9}\label{G2}}
\caption{(Color online) (a), (b) PWM pulses in rectangular form (blue) and in Gaussian form (red) in the time domain and in the frequency domain, respectively.}
\end{figure}
Though we have demonstrated that PWM has the ability to efficiently find optimal controls in very complex systems, a smooth control is usually preferable in experimental realization such as superconducting control system \cite{Zahedinejad2015} or laser control system \cite{Assion1998,Sundermann1999}. Typically, this mission is implemented by introducing a smooth error function into the objective, and re-optimize the control problems \cite{Sundermann1999,Zahedinejad2015}; however, the way to smooth PWM pulses is more straight forward, and this approach may be also applied to smoothing results from other algorithms.
The first approach is using anti-PWM transformation, which replaces rectangular pulse by a continuous function with the same integral area in each time interval $\tau$. This approach also meets the conclusion in \cite{Zahedinejad2015}, where the author found "the target fidelity does not depend on the shape of the pulse" for the subspace-selective self-adaptive differential evolution algorithm (SuSSADE). Since the number of control parameters are equal, one may smooth the piecewise constant control field by using EAP and obtain the same continuous field, rather than sacrifice the efficiency by introducing the error function.
In fact, as we demonstrated in Sec.~{II} the anti-transformation does not always work when one sets a relatively large $\tau$. Here, we propose another approach for smoothing PWM pulses; that is, we transform the rectangular pulses into continuous pulses, e.g. Gaussian pulses. Detail discussion of this can be found in \cite{arXiv}; in short, the rectangular pulses can be well approximated by a sequence of Gaussian pulses $\exp{\Big[-{\pi t^2}/{t_{p}^2}\Big]}$ with the same value of $t_p$. Figure~\ref{G1} shows the relation between the rectangular pulses and the corresponding Gaussian pulses in time domain, Fig.~\ref{G2} reveals the properties of the twos in frequency domain. It demonstrates the Gaussian pulses can well approximate the PWM pulses within a large frequency band. However, though the Gaussian pulse works for a relatively larger $\tau$, it does not always work also. How to effectively smooth PWM pulses is still an open problem for further discussions.
\section{Conclusions}\label{CON}
Combining Pulse Width Modulation (PWM) and gradient algorithm, we applied PWM into quantum optimal control problems. Compared with the basic GRAPE algorithm, the our algorithm raised the efficiency by a large extant, which provides us a highly efficient way for solving various optimization problems. In fact, PWM can also be imbedded in other descendent algorithms, such as Newton algorithm, quasi-Newton algorithm, and various evolutionary algorithms. In addition, since we have demonstrated in \cite{arXiv} that the rectangular pulses generated by PWM are rather robust with noise and can also be implemented by other functions (e.g. the Gaussian pulse train), it indicates the result of PWM algorithm may be directly applied into experiments. Thus, the PWM algorithm may highly simplify the current experiment relying on subtle control of shaped pulses, and provide a very robust and reliable quantum optimal control design.
\section*{Acknowledgements}
This work is supported by the National Natural Science Foundation of China (Grants No. 61134008, No. 61374091, and No. 91221205). We thank Professor Herschel Rabitz from Princeton University for useful discussions.
| bc692e9f44094db92556e6c4a149ac1e8dce675a | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}\label{SEC:intro}
Let $p$ be a prime and let $G$ be a torsion-free abelian group.
The subject of this work is the group $Hom (G,\mathbb{Z}_p)$ of $p$-adic functionals on $G$.
This group is viewed as a $p$-adic dual of $G$, and is studied using the techniques of functional analysis.
This is a novel approach that adds clarity to a difficult area of study.
To emphasize this approach, the notation $G^{*p}$ is used instead of $Hom (G,\mathbb{Z}_p)$.
It is shown that an analogue of the Hahn--Banach Theorem (Proposition \ref{PROP:all-states-extend}) holds in this setting; broadly speaking, this means that $G^{*p}$ contains enough functionals to be interesting.
There is a natural homomorphism of $G$ into its $p$-adic double dual, $G^{*p*p}$, and one of the main results of this work is that the image of $G$ is dense in the relative topology that the double dual inherits as a subspace of $\mathbb{Z}_p^{G^{*p}}$ (Theorem \ref{THM:density}).
As a submodule of the free $\mathbb{Z}_p$-module $\mathbb{Z}_p^{G^{*p}}$, $G^{*p*p}$ is also a free $\mathbb{Z}_p$-module.
Thus, in particular, it is a pro-$p$ group, and the pro-$p$ topology agrees with the relative topology.
There is another pro-$p$ group, called the pro-$p$ completion of $G$, that shares this same property of having a homomorphism from $G$ into it, the image of which is dense.
Moreover, the pro-$p$ completion satisfies a universal property, which implies that there exists a continuous homomorphism from the pro-$p$ completion to the $p$-adic double dual.
The main result of Section \ref{SEC:completions} is Theorem \ref{THM:pro-p}, which states that this homomorphism has a continuous inverse, so the pro-$p$ completion and the $p$-adic double dual of $G$ are continuously isomorphic (in a way that preserves the image of $G$ in each of them).
The $\mathbb{Z}_p$-modules $G^{*p}$, as $p$ ranges over the set of all primes, give enough information to reconstruct the isomorphism class of the torsion-free abelian group $G$.
Section \ref{SEC:classification} describes how this is done in the finite-rank case; in particular, if $G$ has finite rank, then Theorem \ref{THM:classification-3} provides an isomorphism invariant for $G$.
(For a torsion-free abelian group $G$, let us use the term \emph{rank} to refer to the torsion-free rank of $G$, that is, the size of any maximal independent subset.)
This invariant is a matrix description of $G$ that is related to others due to Kurosch and Malcev from 1937 and 1938 respectively (see \cite{K:matrices} and \cite{M:matrices}, or \cite{K:summary} for a more modern description of the latter, with applications to developments since then).
\subsection{History}\label{SUBSEC:history}
This work contributes to the classification and description of torsion-free abelian groups, which is an area with a long history---an excellent overview of the theory is given in \cite{A:book}.
In 1937, Baer \cite{B:rank-one} classified the torsion-free abelian groups of rank one using an invariant called the \emph{type} of the group, which is an equivalence class of supernatural numbers.
At the same time, Kurosch \cite{K:matrices} and Malcev \cite{M:matrices} developed matrix descriptions of a finite-rank torsion-free abelian group.
Definitions \ref{DEF:group-determined} and \ref{DEF:factored} and Theorem \ref{THM:classification-3} from Section \ref{SEC:classification} provide a matrix description that is related to these.
The results here are most closely related to work of Warfield from 1968 \cite{W:homomorphisms}, in which, for a rank-$n$ group $M$ and a rank-$1$ group $G$, Warfield considers the dual $Hom (M,G)$ and the double dual $Hom($ $Hom ($ $M,G),G)$, and gives necessary and sufficient conditions for the natural homomorphism from $M$ to the double dual to be an isomorphism.
The functional analysis approach that is used here is modelled after the theory of functional analysis on partially ordered torsion-free abelian groups, an excellent reference for which is the book \cite{G:book}.
In particular, the $p$-adic Hahn--Banach Theorem that is proved here is modelled after Lemma 4.1 and Proposition 4.2 of \cite{G:book} (originally from \cite{GH:states}).
The matrix description of finite-rank torsion-free abelian groups that appears in Section \ref{SEC:classification} is one of the main results of this work (although by no means is it the only one).
Given that it is closely related to the matrix descriptions of Kurosch and Malcev, it would be natural to ask: What is new in this work that is not already present in the works of those authors?
In response to this question, I would point to the epigraph that precedes this introduction.
This epigraph is a quotation from Kurosch, in which he acknowledges that his matrices arise from a difficult sequence of calculations, and calls for a better theoretical understanding of the relationship between torsion-free abelian groups and $p$-adic integers.
The theory of $p$-adic functionals that is introduced here subsumes most of the difficult calculations, making them invisible, and so represents a clear answer to Kurosch's entreaty.
There is the added bonus that the matrices in the present work are simpler than Kurosch's or Malcev's in that they have fewer rows.
Furthermore Theorem \ref{THM:pro-p}, about the pro-$p$ completion of a torsion-free abelian group, is another consequence of this theory, and to my knowledge this result is entirely new.
\subsection{Notation}\label{SUBSEC:notation}
Let $p$ be a prime.
$\mathbb{Z}_p$ denotes the ring of $p$-adic integers and $\mathbb{Q}_p$ denotes its field of fractions.
The $p$-adic absolute value of an element $\alpha$ of $\mathbb{Q}_p$ or $\mathbb{Z}_p$ is denoted by $|\alpha |_p$.
If $n\in\mathbb{N}$ and $R$ is a ring, then $R^n$ denotes the module consisting of $n\times 1$ columns with entries in $R$, while $(R^n)^*$ denotes its dual consisting of $1\times n$ rows.
If $m,n\in\mathbb{N}$, then $M_{m\times n}(R)$ denotes the module of $m\times n$ matrices over $R$.
For $r\in R^n$, $r^t$ denotes its transpose in $(R^n)^*$.
For $A\in M_{m\times n}(R)$, $A^t$ denotes its transpose in $M_{n\times m}(R)$.
For $v = (v_1,\ldots, v_n)^t \in \mathbb{Z}_p^n$, the norm $\| v\|_p $ is defined by $\| v\|_p := \max_{1\leq i\leq n} \{ \| v_i\|_p\}$.
Let $G$ be a torsion-free abelian group and $S$ a subset of $G$.
Then $\langle S\rangle_*$ denotes the pure subgroup of $G$ generated by $S$; that is, $\langle S\rangle_* := \{ g\in G : ng \in \langle S\rangle$ for some $n\in\mathbb{N}\}$.
\subsection{Outline}\label{SUBSEC:outline}
Section \ref{SEC:examples} contains examples of torsion-free abelian groups with calculations of their $p$-adic functionals.
Section \ref{SEC:states} is the biggest section, and introduces the notation and terminology of $p$-adic functionals.
It includes a discussion of the $p$-adic topology on a torsion-free abelian group $G$ (Section \ref{SUBSEC:pseudometric}); the statement and proof of an analogue of the Hahn--Banach Theorem (Proposition \ref{PROP:all-states-extend} in Section \ref{SUBSEC:functionals}); results about the dual space of all $p$-adic functionals on $G$ (Section \ref{SUBSEC:state-spaces}); and results about double dual spaces, including the fact that the natural homomorphism from $G$ to its double dual has dense image (Theorem \ref{THM:density} in Section \ref{SUBSEC:double-duals}).
Section \ref{SEC:completions} contains a proof that the $p$-adic double dual of a torsion-free abelian group $G$ can be naturally identified with the pro-$p$ completion of $G$ (Theorem \ref{THM:pro-p}).
Section \ref{SEC:classification} uses the theory of $p$-adic functionals to classify finite-rank torsion-free abelian groups up to isomorphism; this is the content of Theorem \ref{THM:classification-3}.
The classifying invariant is a sequence of matrices that is related to Malcev's matrices, and is subject to the same equivalence relation as Malcev's matrices \cite{M:matrices}, although these matrices have fewer rows.
\section{Examples}\label{SEC:examples}
This section contains two examples to illustrate the theory.
The groups in both of these examples are inductive limits of stationary inductive sequences of finitely-generated free abelian groups---see \cite{D:stationary} for previous results about these groups, including a classification up to quasi-isomorphism.
There has already been work done on the real-valued functionals on \emph{ordered} abelian groups that are inductive limits of stationary inductive sequences of simplicial groups (see \cite{H:irrational} and \cite{M:stationary}).
The $p$-adic theory is similar to the theory in the ordered case: in particular, $p$-adic functionals come from rows that are combinations of certain left eigenvectors of a matrix representation of the homomorphism from the stationary inductive sequence \cite[Theorem 4.9]{M:finite-rank}.
\begin{example}\label{EX:favourite}
Consider the matrix $A = \twotwo{1}{1}{1}{4}$.
The characteristic polynomial of $A$ is $\chi_A(x) = x^2-5x+3$, which has roots $x = \frac{5\pm\sqrt{13}}{2}$.
These roots, being algebraic integers, have $3$-adic absolute value at most $1$, and, as their product is $\det A = 3$ and their sum is $\Tr A = 5$, exactly one of them must have $3$-adic absolute value less than $1$.
Let us use the symbol $\lambda$ to denote this eigenvalue.
Then eigenvectors of $A$ associated to $\lambda$ and its algebraic conjugate, $5-\lambda$, can be taken to be $v_1 := {{\lambda - 4}\choose{1}}$ and $v_2 := {{1-\lambda}\choose{1}}$.
Consider the torsion-free abelian group $G = \bigcup_{n=0}^\infty A^{-n}(\mathbb{Z}^2)$.
$\mathbb{Z}$ is dense in $\mathbb{Z}_3$, so, as $\mathbb{Z}^2\subseteq G$, $G$ contains elements that are arbitrarily close to $v_1$ and $v_2$ in the norm $\| \cdot \|_3$.
In particular, the sequence
\[
{{-1}\choose{1}}, {{-7}\choose{1}}, {{-7}\choose{1}}, {{-34}\choose{1}}, {{-115}\choose{1}}, {{128}\choose{1}}, \ldots
\]
has the property that its $n$th element is divisible by $3^n$ in $G$, and this sequence converges to the $\lambda$-eigenvector ${{\lambda - 4}\choose{1}}$ of $A$.
There is a metric on $G$ called the $3$-adic metric (Definition \ref{DEF:pseudometric}) that says that an element is small if it is divisible in $G$ by high powers of $3$.
Write $g\in G$ as $g = \alpha_1v_1+\alpha_2v_2$; then it can be proved that $g$ is divisible by high powers of $3$ if and only if $|\alpha_2|_3$ is small.
(Here is the idea of the proof: $g\in 3^nG$ if and only if $G^kg \in 3^n\mathbb{Z}^2$ for some $k\geq 0$.
But $G^kg = \lambda^k\alpha_1v_1 + (5-\lambda)^k\alpha_2v_2$, and $|\lambda^k|_3 = 3^{-k}$, which is already small in $p$-adic absolute value, so $G^kg$ will be small in the $3$-adic sense if and only if $|(5-\lambda)^k\alpha_2|_3 = |\alpha_2|_3$ is small.)
Any group homomorphism $f : G\to \mathbb{Z}_3$ can be represented as left multiplication by a row in $(\mathbb{Z}_p^2)^*$, and any such row has the form $\beta_1v_1^t + \beta_2v_2^t$, where $\beta_1,\beta_2\in \mathbb{Z}_3$.
As $A$ is symmetric, $v_1^t$ and $v_2^t$ are left eigenvectors associated to $\lambda$ and $5-\lambda$ respectively.
The fact that $f: G\to \mathbb{Z}_3$ is a group homomorphism implies that it must be continuous with respect to the $3$-adic metric on $G$, which happens if and only if $\beta_1 = 0$ (this is a consequence of the fact that $v_1^tv_2 = 0$, so the $v_1^t$ component of $f$ kills the component of $g$ that determines how $3$-divisible $g$ is).
Thus, up to multiplication by a scalar, there is a unique group homomorphism $f : G\to \mathbb{Z}_3$, which is given by left multiplication by $v_2^t$.
The image under $f$ of $G$ is
\begin{align*}
(1-\lambda, 1) \bigcup_{n=0}^\infty A^{-n}(\mathbb{Z}^2) & = (5-\lambda)^{-n}(1-\lambda,1)\mathbb{Z}^2 \\
& = \{ (5-\lambda)^{-n} [ (1-\lambda)a+b] : n\geq 0, a,b\in\mathbb{Z}\} \\
& = \Big\{ \Big( \frac{\lambda}{3}\Big)^n (c\lambda+d) : n\geq 0, c,d\in\mathbb{Z}\Big\}.
\end{align*}
$f$ is an isometry, so $G$ is in fact isomorphic to this image, which is a subgroup of $\mathbb{Z}_3$ because $\lambda/3\in\mathbb{Z}_3$.
\end{example}
The group in the next example admits two independent $3$-adic functionals.
\begin{example}\label{EX:rank-two-dual}
Consider the matrix $B = \left( \begin{array}{rrr} 0 & 0 & -3 \\ 1 & 0 & -1 \\ 0 & 1 & -1 \end{array}\right)$.
The characteristic polynomial of $B$ is $\chi_B(x) = x^3+x^2+x+3$; let us denote the roots of this by $\lambda, \mu,$ and $\nu$.
The product of these roots is $\det B = -3$, so at least one of them---say $\lambda$---has $3$-adic absolute value less than $1$.
But $\lambda\mu + \mu\nu + \nu\lambda = 1$, so $|\mu|_3= |\nu|_3 = 1$.
It is a consequence of the Newton Polygon Theorem \cite[Theorem 6.3.1]{C:local-fields} that $\lambda\in \mathbb{Q}_3$, and hence $\lambda\in \mathbb{Z}_3$, although the Theorem does not say anything $\mu$ and $\nu$.
$w_\lambda := (1,\lambda, \lambda^2)$, $w_\mu := (1,\mu,\mu^2)$, and $w_\nu := (1,\nu,\nu^2)$ are left eigenvectors of $B$ associated to the eigenvalues $\lambda, \mu,$ and $\nu$, respectively.
Let $v_\lambda,v_\mu,$ and $v_\nu$ denote corresponding right eigenvectors.
Consider the torsion-free abelian group $H = \bigcup_{n=0}^\infty B^{-n}(\mathbb{Z}^3)$.
Any element $g\in H$ can be written as a combination $g = \alpha_\lambda v_\lambda + \alpha_\mu v_\mu + \alpha_\nu v_\nu$, but in this case the coefficients will lie in the splitting field $K$ of $\chi_A(x)$ over $\mathbb{Q}_3$.
As was the case in Example \ref{EX:favourite}, $g$ is divisible by high powers of $3$ if and only if $\alpha_\mu$ and $\alpha_\nu$ are small in the extension of the $3$-adic absolute value to $K$.
Thus the only group homomorphisms from $H$ into $\mathbb{Z}_3$ are the combinations of $w_\mu$ and $w_\nu$.
Using the facts that $\lambda\mu\nu = -3$, $\lambda\mu + \mu\nu + \nu\lambda = 1$, and $\lambda + \mu + \nu = -1$, we can calculate
\begin{align*}
w_\mu + w_\nu & = (2,-1-\lambda,-1-\lambda^2)\in (\mathbb{Z}_3^3)^* \\
\text{and } \nu w_\mu + \mu w_\nu & = (-1-\lambda,2+2\lambda+2\lambda^2,2-\lambda-\lambda^2) \in (\mathbb{Z}_3^3)^*.
\end{align*}
Any homomorphism $f: H \to \mathbb{Z}_3$ can be written as a $\mathbb{Z}_3$-combination of these two rows.
(This follows from a rank argument: the group $Hom (H,\mathbb{Z}_3)$ is a free $\mathbb{Z}_3$-module, its rank cannot exceed $3$, it does not contain any non-zero multiple of $w_\lambda$, and we have found two independent rows that it does contain.)
\end{example}
\section{$p$-adic functionals}\label{SEC:states}
The goal of this section is to study the group $Hom (G,\mathbb{Z}_p)$, of homomorphisms from a torsion-free abelian group $G$ into $\mathbb{Z}_p$, from a functional analysis point of view.
To emphasize this point of view, these homomorphisms will be referred to as \emph{$p$-adic functionals} on $G$.
\begin{defn}\label{DEF:p-adic-functional}
Let $p$ be a prime and let $G$ be a torsion-free abelian group.
Then a \emph{$p$-adic functional} on $G$ is a group homomorphism $f : G\to \mathbb{Z}_p$.
\end{defn}
The starting point of this work is Proposition \ref{PROP:all-states-extend}, which is an analogue of the Hahn--Banach Theorem for the setting of torsion-free abelian groups.
Proposition \ref{PROP:all-states-extend} says that if $f$ is a $\mathbb{Z}_p$-valued group homomorphism on a subgroup $H$ of $G$ and $f$ is \emph{$t$-contractive} for some \emph{$p$-adic sublinear pseudometric} $t$ (Definition \ref{DEF:t-contractive}), then $f$ extends to a $p$-adic functional on $G$ that is also $t$-contractive.
This allows us to establish the existence of $p$-adic functionals on $G$ with desired properties by first finding functionals with those properties defined on some subgroup and then extending them to all of $G$.
A consequence of this theory is that the group of all $\mathbb{Z}_p$-valued homomorphisms on $G$ (which will be denoted by $G^{*p}$ to emphasize the functional analysis point of view) contains enough functionals to be interesting.
Section \ref{SUBSEC:state-spaces} deals with this group, both as a topological space and as a torsion-free abelian group.
The elements of $G$ are themselves $p$-adic functionals on $G^{*p}$, and there is a homomorphism from $G$ into the double dual group, which consists of $p$-adic functionals on the group of $p$-adic functionals on $G$.
It will be shown in Section \ref{SEC:completions} that this double dual group $G^{*p*p}$ coincides exactly with the pro-$p$ completion of $G$.
This result provides a means of describing the pro-$p$ completion of $G$, which is interesting in itself.
$p$-adic functionals are a $\mathbb{Z}_p$-submodule of the set of $\mathbb{Q}_p$-valued group homomorphisms on $G$ that are contractive with respect to a certain pseudometric; let us describe this pseudometric now.
\subsection{The $p$-adic pseudometric}\label{SUBSEC:pseudometric}
If $g$ is an element of a torsion-free abelian group $G$ and $n\in\mathbb{N}$, let us say that $g$ is \emph{divisible by $n$} or \emph{$n$-divisible in $G$} if there exists $h\in G$ such that $nh = g$.
Let us say that $G$ is $n$-divisible if each element of $G$ is $n$-divisible.
\begin{defn}\label{DEF:pseudometric}
Let $G$ be a torsion-free abelian group and $p$ be a prime.
Define the \emph{$p$-adic pseudometric} $d_p$ on $G$ by
\[
d_p(g,h) = p^{-k},
\]
where $k$ is the non-negative integer that is maximal with the property that $g-h$ is divisible by $p^k$.
If $g-h$ is divisible by unbounded powers of $p$, then $d_p(g,h) = 0$.
\end{defn}
The reader is invited to verify that $d_p$ satisfies the definition of a pseudometric; indeed, it satisfies the strong triangle inequality: $d_p(g,h) \leq \max \{ d_p(g,z),d_p(z,h)\}$, where $g,h,$ and $z\in G$.
$d_p$ also satisfies an absolute scalability condition: if $g,h\in G$ and $n\in\mathbb{Z}$, then $d_p(ng,nh) = |n|_pd_p(g,h)$.
$d_p$ is a metric if and only if $0 \in G$ is the only element that is divisible by arbitrarily high powers of $p$; in this case, let us refer to it as the \emph{$p$-adic metric}.
$d_p$ is bounded: $d_p(g,h)\leq 1$ for all $g,h\in G$.
The topology arising from the $p$-adic pseudometric is called the $p$-adic topology in \cite{K:abelian-groups}.
Let us use the symbol $G_p$ to denote the subgroup of $G$ containing all elements that are divisible by arbitrarily high powers of $p$.
It is not difficult to verify that $G_p$ is a pure subgroup.
Given a subgroup $H$ of $G$, let $\overline{H}^p$ denote the closure of $H$ in the topology arising from $d_p$.
Note that $\overline{\{ 0\}}^p = G_p$, so $\overline{H}^p$ contains $G_p$ for all subgroups $H$.
The following lemma will be useful later, when proving Lemma \ref{LEM:quotient}, which is used to prove Theorem \ref{THM:pro-p}, one of the main theorems.
It says that, if $H$ is a subgroup of $G$ such that any element of $G/H$ has order dividing $p^m$, then any element $g\in G$ that is not in $H$ must lie at a distance of at least $p^{-m}$ from $H$.
In particular, this implies that if $[G:H] = p^m$, then $H = \overline{H}^p$.
\begin{lemma}\label{LEM:subgroup-separation}
Let $p$ be a prime, let $G$ be a torsion-free abelian group with $p$-adic pseudometric $d_p$, let $H$ be a subgroup of $G$, and let $m\in\mathbb{N}$.
Then the following are equivalent.
\begin{enumerate}
\item The order of any element $g+H\in G/H$ divides $p^m$.
\item If $g, g'\in G$ and $d_p(g,g') \leq p^{-m}$, then $g+H = g'+H$.
\end{enumerate}
\end{lemma}
\begin{proof}
To prove that (1) implies (2), suppose that $g,g'\in G$ are such that $d_p(g,g') \leq p^{-m}$.
This means there exists $h\in G$ such that $p^mh = g-g'$.
But by (1), $p^mh\in H$, so
\[
g + H = g - g' + g' + H = p^{m}h + g' + H = g' + H.
\]
To see that (2) implies (1), pick $g\in G$.
Then
\[
d_p(p^mg,0) = p^{-m}d_p(g,0) \leq p^{-m},
\]
so, by (2), $p^mg + H = 0 + H$, that is, $p^mg\in H$.
\end{proof}
The following lemma establishes that the closure of a pure subgroup is pure.
\begin{lemma}\label{LEM:pure-closure}
Let $p$ be a prime, let $G$ be a torsion-free abelian group, and let $H$ be a pure subgroup of $G$.
Then $\overline{H}^p$ is also a pure subgroup.
\end{lemma}
\begin{proof}
It is easy to verify that $\overline{H}^p$ is a subgroup: if $g,g'\in \overline{H}^p$, then given $\epsilon > 0$ we can find $h,h'\in H$ such that $d_p(g,h), d_p(g',h') < \epsilon$, in which case $d_p(g+g',h+h') \leq \max \{ d_p(g,h),d_p(g',h')\} < \epsilon$.
To see that $\overline{H}^p$ is pure, suppose $ng\in \overline{H}^p$ for some $n\in \mathbb{Z}\backslash \{ 0\}$ and $g\in G$.
This is equivalent to saying that $d_p(ng+H,0) = 0$ in the quotient group $G/H$, which is torsion-free as $H$ is pure.
Thus $ng+H\in (G/H)_p$, which is pure in $G/H$, so $g+H\in (G/H)_p$, so $g\in \overline{H}^p$ as well.
\end{proof}
Closed pure subgroups will play a role in what follows, so it will be useful to give a name to such subgroups.
\begin{defn}\label{DEF:ideal-simple}
Let $p$ be a prime and let $G$ be a torsion-free abelian group.
Let us use the term \emph{$p$-ideal} to refer to any pure subgroup of $G$ that is closed in the topology arising from $d_p$.
Let us say that $G$ is \emph{$p$-simple} if $G_p$ is the only proper $p$-ideal of $G$.
\end{defn}
The group $G$ in Example \ref{EX:favourite} is $3$-simple: it has many proper non-trivial pure subgroups, each of which is isomorphic to $\mathbb{Z}$, but the closure of any of them in the metric $d_3$ is all of $G$.
Given a family $(J_\lambda)_{\lambda\in\Lambda}$ of $p$-ideals of $G$, the intersection $\bigcap_{\lambda\in\Lambda}J_\lambda$ is again a $p$-ideal.
The sum of the $J_\lambda$s is not necessarily pure or closed, but $\overline{\langle \bigcup_{\lambda\in\Lambda}J_\lambda \rangle_*}^p$ is a $p$-ideal, and is clearly the smallest one that contains each $J_\lambda$.
Thus the set of $p$-ideals of $G$ forms a complete lattice where the partial order is given by inclusion.
The minimal $p$-ideals---namely, the elements of this lattice that have nothing except $G_p$ below them---are of particular interest.
\begin{defn}\label{DEF:minimal-ideal}
Let $p$ be a prime and let $G$ be a torsion-free abelian group.
A $p$-ideal $J$ of $G$ is called \emph{minimal} if it properly contains $G_p$, but does not properly contain any other $p$-ideal of $G$.
\end{defn}
Lemma \ref{LEM:singly-generated-ideal}, below, establishes that a singly-generated $p$-ideal is minimal.
The proof of this fact uses the following lemma, which will also be used later to show that a $p$-ideal can be separated from an element not in its closure using a $p$-adic functional (Corollary \ref{COR:separating-state}).
\begin{lemma}\label{LEM:closure}
Let $p$ be a prime, let $G$ be a torsion-free abelian group, let $g\in G$, and let $H$ be a pure subgroup of $G$.
If the distances $\{ d_p(g,h) : h\in H\}$ are bounded below by $\epsilon > 0$, then the values $\{\frac{1}{|n|_p}d_p(ng,h) : h\in H, n\in \mathbb{Z}\backslash \{ 0\}\}$ are also bounded below by $\epsilon$.
\end{lemma}
\begin{proof}
Suppose otherwise, and let $n\in \mathbb{N}$ be minimal with the property that $\frac{1}{|n|_p}d_p(ng,h) < \epsilon$ for some $h\in H$.
Let us show first that $n$ is not divisible by $p$.
Suppose for a contradiction that $p$ that divides $n$---say $n = pn'$ for $n'\in \mathbb{N}$.
The assumption that $d_p(ng,h) < \epsilon|n|_p < 1$ implies that $ng-h$ is $p$-divisible.
Combined with the obvious fact that $ng$ is $p$-divisible, this implies that $h$ is $p$-divisible---say $h = ph'$ for some $h'\in H$.
But then $\frac{1}{|n'|_p}d_p(n'g,h') = \frac{1}{|n|_p}d_p(ng,h) < \epsilon$, contradicting the minimality of $n$.
Thus $n$ is not divisible by $p$.
Consider the element $g+H\in G/H$.
As $H$ is pure, $G/H$ is torsion-free.
$d_p(ng,h) < \epsilon$ in $G$, so $d_p(ng + H, 0) < \epsilon$ in $G/H$.
But $p$ does not divide $n$, so $d_p(g+H,0) = |n|_p d_p(g+H,0) = $ $d_p(ng+H,0) < \epsilon$.
But the statement that $d_p(g+H,0)< \epsilon$ in $G/H$ implies that $d_p(g,h) < \epsilon$ in $G$ for some $h\in H$, which is a contradiction.
\end{proof}
\begin{lemma}\label{LEM:singly-generated-ideal}
Let $p$ be a prime, let $G$ be a torsion-free abelian group, and suppose that $g\in G\backslash G_p$.
Then $\overline{\langle g\rangle_*}^p$ is a minimal $p$-ideal.
\end{lemma}
\begin{proof}
$\overline{\langle g\rangle_*}^p$ is certainly a $p$-ideal; to see that it is minimal, let us suppose that it properly contains another $p$-ideal $H$ and show that $H = G_p$.
Choose $h\in H$ and consider the $p$-ideal $\overline{\langle h\rangle_*}^p$.
By assumption, $g$ is not in this $p$-ideal, so there exists $\epsilon > 0$ such that $d_p(g,h')>\epsilon$ for all $h'\in \langle h\rangle_*$.
By Lemma \ref{LEM:closure}, the values $\{ \frac{1}{|n|_p}d_p(ng,h') : n\in \mathbb{Z}\backslash \{ 0\}, h'\in \langle h\rangle_*\}$ are also bounded below by $\epsilon$.
But the hypothesis that $h\in \overline{\langle g\rangle_*}^p$ means that, for all $k\in \mathbb{N}$, there exists $g_k\in \langle g\rangle_*$ such that $d_p(g_k,h)<p^{-k}$.
There exist relatively prime $n_k\in\mathbb{Z}$ and $m_k\in\mathbb{Z}\backslash \{ 0\}$ such that $m_kg_k = n_kg$.
Thus $d_p(n_kg,m_kh) < p^{-k}$.
Combined with the fact that $\frac{1}{|n_k|_p}d_p(n_kg,m_kh) > \epsilon$, this implies that $\frac{1}{|n_k|_p}p^{-k}> \epsilon$, that is, $|n_k|_p < p^{-k}/\epsilon$.
Thus, given $\delta > 0$, the value $|n_k|_p$ can be made less than $\delta$ by choosing large enough $k$.
So in particular, for large enough $k$, $p\mid n_k$, and hence $p\nmid m_k$.
But then
\begin{align*}
d_p(h,0) & = |m_k|_pd_p(h,0) \\
& = d_p(m_kh,0) \\
& \leq \max \{ d_p(n_kg,m_kh),d_p(n_kg,0)\} \\
& \leq \max \{ p^{-k},\delta\}.
\end{align*}
Thus $d_p(h,0) = 0$, so $h\in G_p$.
\end{proof}
An immediate consequence of Lemma \ref{LEM:singly-generated-ideal} is the following.
\begin{cor}\label{COR:minimal-ideals}
Let $p$ be a prime, let $G$ be a torsion-free abelian group, and let $H$ be a $p$-ideal of $G$.
Then $H$ is minimal if and only if it is singly generated as a $p$-ideal and not equal to $G_p$.
\end{cor}
\subsection{A Hahn--Banach Theorem for $p$-adic functionals}\label{SUBSEC:functionals}
Let $p$ be a prime, let $G$ be a torsion-free abelian group with $p$-adic pseudometric $d_p$, and let $f:G\to \mathbb{Z}_p$ be a group homomorphism (i.e., a $p$-adic functional).
Then $f$ is automatically contractive (and hence continuous) with respect to the $p$-adic pseudometric $d_p$ on $G$, meaning that $|f(g_1)-f(g_2)|_p\leq d_p(g_1,g_2)$ for all $g_1,g_2\in G$.
To see this, it is enough to observe that, if $g\in G$ is $p^n$-divisible---say $g = p^nh$ for some $h\in G$---then $|f(g)|_p = |p^nf(h)|_p = p^{-n}|f(h)|_p\leq p^{-n}$, as $|f(h)|_p\leq 1$.
If instead $f$ is only defined on a subgroup $H$ of $G$, then the above argument shows that, in order for there to exist an extension of $f$ to all of $G$, it is necessary that $f$ be contractive with respect to $d_p$.
It turns out that this condition is also sufficient (Proposition \ref{PROP:all-states-extend}).
But we can say more: if $f: H\to \mathbb{Z}_p$ is contractive with respect to some \emph{$p$-adic sublinear pseudometric} $t$ (Definition \ref{DEF:t-contractive}), then $f$ extends to a $p$-adic functional on $G$ that is also contractive with respect to $t$.
\begin{defn}\label{DEF:t-contractive}
Let $p$ be a prime and let $G$ be a torsion-free abelian group with $p$-adic pseudometric $d_p$.
Let $t : G\times G \to \mathbb{R}_{\geq 0}$ be a pseudometric.
Let us say that $t$ is a \emph{$p$-adic sublinear pseudometric} on $G$ if it satisfies the following additional conditions:
\begin{enumerate}
\item $t(g_1,g_2) = t(g_1+h,g_2+h)$ for all $g_1,g_2,h\in G$ ($t$ is translation invariant);
\item $|n|_pt(g_1,g_2) = t(ng_1,ng_2)$ for all $g_1,g_2\in G$, $n\in \mathbb{Z}$ ($t$ is absolutely scalable);
\item $t(g_1,g_2) \leq d_p(g_1,g_2)$ for all $g_1,g_2\in G$ ($t$ is dominated by $d_p$); and
\item $t(g_1,g_2) \leq \max\{ t(g_1,g_3),t(g_2,g_3)\}$ for all $g_1,g_2,g_3\in G$ ($t$ satisfies the strong triangle inequality).
\end{enumerate}
Let $H$ be a subgroup of $G$, and suppose that $f : H\to \mathbb{Z}_p$ is a group homomorphism that is contractive with respect to the $p$-adic sublinear pseudometric $t$, meaning that $|f(h_1)-f(h_2)|_p \leq t(h_1,h_2)$ for all $h_1,h_2\in H$.
Let us refer to such a homomorphism $f$ as a \emph{$t$-contractive functional} on $H$.
\end{defn}
Note in particular that $d_p$ itself is a $p$-adic sublinear pseudometric, and, if $t$ is another $p$-adic sublinear pseudometric, then any $t$-contractive functional is also a $d_p$-contractive functional.
Most of the results in this section are statements about the existence of $p$-adic functionals; these functionals are extensions of $t$-contractive functionals for some $p$-adic sublinear pseudometric $t$.
For most of the proofs, $t$ can be taken to be $d_p$; however, Proposition \ref{PROP:separate-from-subgroup} establishes the existence of $p$-adic functionals with very specific properties, and for this it will be necessary to use a more restrictive $t$ to get sharper information.
Note that, if $H$ is a subgroup of $G$ and $d_p$ is the $p$-adic pseudometric on $G$, then the $p$-adic pseudometric on $H$ is bounded below by the restriction of $d_p$ to $H$, and this bound may be strict.
Thus to say that $f$ is a $d_p$-contractive functional on $H$ is stronger than saying that $f$ is a $p$-adic functional on $H$.
Of course, if $H$ is a pure subgroup of $G$, then the $p$-adic pseudometric on $H$ coincides with the restriction to $H$ of that on $G$.
It is clear that the trivial homomorphism is a $p$-adic functional, and, if $G$ is $p$-divisible, then this is the only one.
It is natural to ask whether there exist any non-trivial $p$-adic functionals on $G$ if $G$ is not $p$-divisible.
Corollary \ref{COR:states-exist} shows that, if $G$ is not $p$-divisible, then non-zero $p$-adic functionals on $G$ do exist.
The following is the key lemma that makes everything else work.
\begin{lemma}\label{LEM:extend}
Let $p$ be a prime, let $G$ be a torsion-free abelian group, let $t$ be a $p$-adic sublinear pseudometric on $G$, let $H$ be a subgroup of $G$, and let $g\in G$.
Suppose there exists a $t$-contractive functional $f: H\to \mathbb{Z}_p$, and define, for each $h\in H$ and $n\in \mathbb{Z}\backslash \{ 0\}$, the closed subset $B(h,n) := \{ \alpha \in\mathbb{Q}_p : |n|_p|\alpha-\frac{1}{n}f(h)|_p \leq t(ng,h)\}$.
Then
\begin{enumerate}
\item $\bigcap_{h\in H,n\in\mathbb{Z}\backslash \{ 0\}} B(h,n)$ is non-empty.
\item If $f$ extends to a $t$-contractive functional $f'$ on $H + \mathbb{Z} g$, then $f'(g)\in \bigcap_{h\in H,n\in\mathbb{Z}\backslash \{ 0\}} B(h,n)$.
\item If $\beta \in \bigcap_{h\in H,n\in\mathbb{Z}\backslash \{ 0\}}B(h,n)$, then $f$ extends to a $t$-contractive functional $f'$ on $H + \mathbb{Z} g$ such that $f'(g) = \beta$.
\end{enumerate}
\end{lemma}
\begin{proof}
To prove statement (1), note that $B(0,1)$ consists of all $\alpha\in \mathbb{Q}_p$ such that $|\alpha|_p\leq t(g,0) \leq d_p(g,0) \leq 1$, so $B(0,1)\subseteq \mathbb{Z}_p$.
Then rewrite the intersection:
\[
\bigcap_{h\in H,n\in\mathbb{Z}\backslash\{ 0\}}B(h,n) = \bigcap_{h\in H,n\in\mathbb{Z}\backslash \{ 0\}} (B(0,1)\cap B(h,n)),
\]
the latter of which is an intersection of closed subsets of $\mathbb{Z}_p$.
$\mathbb{Z}_p$ is compact, so the collection of all closed subsets of the form $B(0,1)\cap B(h,n)$ has non-empty intersection if and only if it has the finite intersection property.
But $\bigcap_{i=1}^k(B(0,1)\cap B(h_i,n_i)) = B(0,1)\cap B(h_1,n_1)\cap \cdots \cap B(h_k,n_k)$, so it suffices to check that the collection of all subsets of the form $B(h,n)$ has the finite intersection property.
Because $|\cdot |_p$ satisfies the strong triangle inequality, if two closed disks in $\mathbb{Q}_p$ have non-empty intersection, then one is contained in the other.
Thus if $B(h_1,n_1)\cap B(h_2,n_2)$ is non-empty, it equals either $B(h_1,n_1)$ or $B(h_2,n_2)$.
So to show that a finite intersection of closed disks is non-empty, it suffices to show that all pairwise intersections are non-empty.
So let us verify to that any pair of subsets $B(h_1,n_1)$ and $B(h_2,n_2)$ with $h_1,h_2\in H$, $n_1,n_2\in \mathbb{Z}\backslash \{ 0\}$ has non-empty intersection.
By the strong triangle inequality,
\[
t(n_2h_1,n_1h_2) \leq \max \{ t(n_1n_2g,n_2h_1),t(n_1n_2g,n_1h_2)\}.
\]
Suppose without loss of generality that $t(n_1n_2g,n_2h_1)$ is the maximum of these two distances.
Then
\begin{align*}
|n_1n_2|_p |\frac{1}{n_1}f(h_1)-\frac{1}{n_2}f(h_2)|_p
& = |n_2f(h_1)-n_1f(h_2)|_p \\
& = |f(n_2h_1)-f(n_1h_2)|_p \\
& \leq t(n_2h_1,n_1h_2) \\
& \leq t(n_1n_2g,n_2h_1) = |n_2|_pt(n_1g,h_1),
\end{align*}
where the last equality comes from the absolute scalability of $t$.
Thus $|n_1|_p|\frac{1}{n_1}f(h_1)-\frac{1}{n_2}f(h_2)|_p $ $\leq t(n_1g,h_1)$, so $\frac{1}{n_2}f(h_2)\in B(h_1,n_1)$.
It is clearly in $B(h_2,n_2)$, so the intersection is non-empty.
Now let us prove statement (2).
For all $h\in H$ and $n\in\mathbb{Z}\backslash\{ 0\}$, $|n|_p|f'(g)-\frac{1}{n}f(h)|_p $ $ = |f'(ng) - f'(h)|_p$, which is bounded above by $t(ng,h)$ because $f'$ is a $t$-contraction.
Thus $f'(g)\in B(h,n)$.
Now let us prove statement (3).
Define $f': H + \mathbb{Z} g \to \mathbb{Q}_p$ as follows.
Given $g'\in H + \mathbb{Z} g$, write $g' = h + ng$ for some $h\in H$ and $n\in \mathbb{Z}$.
Then let $f'(g') = f(h) + n\beta\in \mathbb{Q}_p$.
$f'$ is well-defined.
To see this, suppose $g' = h_1+n_1g$ and $g' = h_2 + n_2g$; then $(n_1-n_2)g' - (h_2-h_1) = 0$.
If $n_1 = n_2$, then $h_1 = h_2$, in which case $f(h_1)+n_1\beta = f(h_2)+n_2\beta$.
If $n_2\neq n_2$, then $d_p((n_1-n_2)g,h_2-h_1) = 0$, so $B(h_2-h_1,n_1-n_2)$ is the singleton $\{ \frac{1}{n_1-n_2}f(h_2-h_1)\}$, in which case $\beta = \frac{1}{n_1-n_2}f(h_2-h_1)$ and $f(h_1) + n_1\beta = f(h_2)+n_2\beta$.
It is easy to verify that $f'$ is a group homomorphism.
Given that $f'$ is a group homomorphism and $t$ is translation invariant, to see that $f'$ is a $t$-contractive functional it is enough to check that $|f(g')|_p\leq t(g',0)$ for all $g'\in H+\mathbb{Z} g$.
If $g'\in H$, this follows immediately from the fact that $f$ is a $t$-contractive functional, so we may suppose that $g'\notin H$.
This means that $g' = h + ng$ for some $h\in H$ and $n\in \mathbb{Z}\backslash \{ 0\}$.
Then
\begin{align*}
|f'(g')|_p & = |f(h)+n\beta|_p \\
& = |n|_p|\beta - \frac{1}{n}f(-h)|_p \\
& \leq t(ng,-h) = t(h+ng,0),
\end{align*}
where the inequality on the last line comes from the fact that $\beta\in B(-h,n)$.
\end{proof}
Lemma \ref{LEM:extend} allows us to extend a $t$-contractive functional on a subgroup of $G$ to one on a larger subgroup, but to show that non-trivial $p$-adic functionals exist, we need to start with a non-trivial $t$-contractive functional on some subgroup of $G$.
The natural place to begin is with a cyclic subgroup.
The following corollary, which deals with functionals on cyclic subgroups of $G$, follows from Lemma \ref{LEM:extend} by taking $H = \{ 0\}$ and $t = d_p$.
\begin{cor}\label{COR:rank-one-seed}
Let $p$ be a prime, let $G$ be a torsion-free abelian group with $p$-adic pseudometric $d_p$, and let $g\in G$.
Then, given any $\beta\in\mathbb{Z}_p$ for which $|\beta|_p \leq d_p(g,0)$, there exists a $d_p$-contractive functional $f:\langle g\rangle\to \mathbb{Z}_p$ such that $f(g) = \beta$.
\end{cor}
The next proposition is an analogue of the Hahn--Banach Theorem for the setting of torsion-free abelian groups.
\begin{prop}\label{PROP:all-states-extend}
Let $p$ be a prime and let $G$ be a torsion-free abelian group with $p$-adic sublinear pseudometric $t$.
Suppose that $H$ is a subgroup of $G$ and $f$ is a $t$-contractive functional on $H$.
Then there exists a $t$-contractive $p$-adic functional $f'$ on $G$ that extends $f$.
\end{prop}
\begin{proof}
Consider the set $\mathcal{K}$ of pairs $(f',H')$, where $H'$ is a subgroup of $G$ that contains $H$ and $f'$ is a $t$-contractive functional on $H'$ that extends $f$.
$\mathcal{K}$ is non-empty because it contains $(f,H)$.
Give $\mathcal{K}$ a partial ordering by saying that $(f_1',H_1')\leq (f_2',H_2')$ if $H_1'\subseteq H_2'$ and $f_2'$ extends $f_1'$.
Any chain $((f_\lambda',H_\lambda'))_{\lambda\in\Lambda}$ in $\mathcal{K}$ is bounded above---an upper bound is $(F',\bigcup_{\lambda\in\Lambda}H_\lambda')$, where $F'$ is defined by $F'(h) = f_\lambda'(h)$ for any $\lambda$ for which $h\in H_\lambda'$.
Thus by Zorn's Lemma there exists a maximal pair $(f_{\max},H_{\max})$.
Suppose for a contradiction that $H_{\max}\neq G$.
Then there exists $g'\in G\backslash H_{\max}$.
By parts (1) and (3) of Lemma \ref{LEM:extend}, $f_{\max}$ extends to a $t$-contractive functional $f_{\max}'$ on the subgroup $H+\mathbb{Z} g'$.
$H+\mathbb{Z} g'$ contains $H_{\max}$, and hence contains $H$, and $f_{\max}'$ extends $f_{\max}$, and hence extends $f$.
Thus the pair $(f_{\max}',H+\mathbb{Z} g')$ is in $\mathcal{K}$, and is strictly greater than $(f_{\max},H_{\max})$, contradicting the maximality of that pair.
Hence $H_{\max} = G$ after all.
\end{proof}
Proposition \ref{PROP:all-states-extend} can be combined with Corollary \ref{COR:rank-one-seed} and Lemma \ref{LEM:closure} to produce the following corollaries, which say that any $g\in G$ realizes its maximum possible value on some functional, and $p$-adic functionals can separate a pure subgroup of $G$ from an element $g$ not in its closure.
\begin{cor}\label{COR:state-hits-distance}
Let $p$ be a prime, let $G$ be a torsion-free abelian group with $p$-adic pseudometric $d_p$, and let $g\in G$.
Then there exists a $p$-adic functional $f$ on $G$ such that $|f(g)|_p = d_p(g,0)$.
\end{cor}
\begin{proof}
By Corollary \ref{COR:rank-one-seed}, there exists a $d_p$-contractive functional $f$ on $\langle g\rangle$ such that $|f(g)|_p = d_p(g,0)$.
By Proposition \ref{PROP:all-states-extend}, $f$ extends to a $p$-adic functional on $G$.
\end{proof}
\begin{cor}\label{COR:states-exist}
Let $p$ be a prime and let $G$ be a torsion-free abelian group that is not $p$-divisible.
Then there exists a non-zero $p$-adic functional on $G$.
\end{cor}
\begin{cor}\label{COR:separating-state}
Let $p$ be a prime, let $G$ be a torsion-free abelian group with pure subgroup $H$, and suppose that $g$ is an element of $G$ that is not in $\overline{H}^p$.
Then there exists a $p$-adic functional $f$ on $G$ such that $f(h) = 0$ for all $h\in H$, but $f(g)\neq 0$.
\end{cor}
\begin{proof}
Let $d_p$ denote the $p$-adic pseudometric on $G$.
The zero homomorphism $f' : H\to \mathbb{Z}_p$ is a $d_p$-contractive functional.
To say that $g$ is not in $\overline{H}^p$ means that there is some $\epsilon > 0$ such that $d_p(g,h) \geq \epsilon$ for all $h\in H$.
By Lemma \ref{LEM:closure}, this means that the values $\{ \frac{1}{|n|_p}d_p(ng,h) : h\in H, n\in \mathbb{Z}\backslash \{ 0\}\}$ are bounded below by $\epsilon$.
But then, by Lemma \ref{LEM:extend} part (3), for any $\beta$ in the ideal $\{ \alpha\in\mathbb{Z}_p : |\alpha|_p\leq \epsilon\}$ of $\mathbb{Z}_p$, $f'$ extends to a $d_p$-contractive functional $f''$ on $H+\mathbb{Z} g$ such that $f''(g) = \beta$.
In particular, it is possible to find a non-zero $\beta$ with this property.
Then by Proposition \ref{PROP:all-states-extend}, $f''$ extends to the required $p$-adic functional $f$ on $G$.
\end{proof}
The following proposition will be useful in Section \ref{SEC:completions}, when we consider finite-index subgroups of a torsion-free abelian group $G$.
\begin{prop}\label{PROP:separate-from-subgroup}
Let $p$ be a prime, let $G$ be a torsion-free abelian group, and let $H$ be a subgroup of $G$.
Suppose $m\in \mathbb{N}$ is such that the order of any element $g+H\in G/H$ divides $p^m$.
Then for any $g\in G\backslash H$ there exists a $p$-adic functional $f$ on $G$ with the property that $|f(g)|_p >p^{-m} \geq |f(h)|_p$ for all $h\in H$.
\end{prop}
\begin{proof}
Let $d_p$ denote the $p$-adic pseudometric on $G$, and let $d_p^H$ denote the $p$-adic pseudometric on $H$ (which might be different from the restriction of $d_p$ to $H$).
For $g\in G$, define
\begin{align*}
v : G & \to \mathbb{N}\cup \{ 0\} \\
g & \mapsto \min \{ k\in \mathbb{N}\cup\{ 0\} : p^kg\in H\}.
\end{align*}
Then define a function $t : G\times G \to \mathbb{R}_{\geq 0}$ as follows:
\begin{align*}
t(g_1,g_2) & = \left\{ \begin{array}{rl} p^{-m}d_p^H(g_1,g_2) & \text{ if } g_1-g_2\in H \\
p^{-m}p^{v(g_1-g_2)} & \text{ otherwise.} \end{array}\right.
\end{align*}
Let us show that $t$ is a $p$-adic sublinear pseudometric.
Choose elements $g_1,g_2$, and $g_3\in G$.
The proof that $t$ satisfies the strong triangle inequality can be broken into cases, depending upon which of the differences $g_1-g_2$, $g_1-g_3$, $g_3-g_2$ lie in $H$.
If any two of these are in $H$, then so is the third, so the possibilities are (1) all are in $H$; (2) none is in $H$; (3) $g_1-g_2\in H$, $g_1-g_3,g_3-g_2\notin H$; and (4) without loss of generality, $g_1-g_2,g_3-g_2\notin H$, $g_1-g_3\in H$.
{\bf Case (1):} $g_1-g_2,g_1-g_3,g_3-g_2\in H$.
Then
\begin{align*}
t(g_1,g_2) & = p^{-m}d_p^H(g_1,g_2) \\
& \leq p^{-m}\max \{ d_p^H(g_1,g_3),d_p^H(g_3,g_2)\} \\
& = \max\{ t(g_1,g_3),t(g_3,g_2)\}.
\end{align*}
{\bf Case (2):} $g_1-g_2,g_1-g_3,g_3-g_2\notin H$.
Let $v_0 = \max\{ v(g_1-g_3),v(g_3-g_2)\}$.
Then $p^{v_0}(g_1-g_3), p^{v_0}(g_3-g_2)\in H$, and
\[
p^{v_0} (g_1-g_2) = p^{v_0} (g_1-g_3) + p^{v_0} (g_3-g_2) \in H,
\]
so $v(g_1-g_2)\leq v_0 = \max\{ v(g_1-g_3),v(g_3-g_2)\}$, which means
\begin{align*}
t(g_1,g_2) = p^{-m}p^{v(g_1-g_2)} & \leq \max\{p^{-m}p^{v(g_1-g_3)},p^{-m}p^{v(g_3-g_2)}\} \\
& = \max\{ t(g_1,g_3),t(g_3,g_2)\}.
\end{align*}
{\bf Case (3):} $g_1-g_2\in H$, $g_1-g_3,g_3-g_2\notin H$.
In this case $t(g_1,g_3),$ $t(g_3,g_2) \geq p^{-m}$, so
\[
t(g_1,g_2) = p^{-m}d_p^H(g_1,g_2) \leq p^{-m} \leq \max\{ t(g_1,g_3),t(g_3,g_2)\}.
\]
{\bf Case (4):} $g_1-g_2,g_3-g_2\notin H$, $g_1-g_3\in H$.
In this case $p^k (g_1-g_2) \in H$ if and only if $p^k(g_3-g_2)\in H$, so
\[
t(g_1,g_2) = p^{-m}p^{v(g_1-g_2)} = p^{-m}p^{v(g_3-g_2)} = t(g_3,g_2).
\]
Thus $t$ satisfies the strong triangle inequality.
Combined with the obvious facts that $t$ is symmetric and $t(g,g) = 0$ for all $g\in G$, this implies that $t$ is a pseudometric.
$t$ is clearly translation invariant, as the formulas $v(g_1-g_2)$ and $d_p^H(g_1,g_2)$ involve taking the difference of the group elements $g_1$ and $g_2$.
To see that $t$ is absolutely scalable, pick $n\in \mathbb{Z}$ and $g_1,g_2\in G$.
If $(g_1-g_2)\in H$, then $n(g_1-g_2)\in H$, and
\begin{align*}
t(ng_1,ng_2) & = p^{-m}d_p^H(ng_1,ng_2) \\
& = |n|_pp^{-m}d_p^H(g_1,g_2) \\
& = |n|_pt(g_1,g_2).
\end{align*}
If $n(g_1-g_2)\notin H$, then $g_1-g_2\notin H$, and
\begin{align*}
t(ng_1,ng_2) & = p^{-m}p^{v(n(g_1-g_2))} \\
& = p^{-m}p^{v(g_1-g_2)-v_p(n)} \\
& = p^{-m}p^{v(g_1-g_2)}|n|_p \\
& = |n|_pt(g_1,g_2),
\end{align*}
where $v_p$ is the $p$-adic valuation on $\mathbb{Q}$.
If $n(g_1-g_2)\in H$ but $g_1-g_2\notin H$, then $p^{v(g_1-g_2)} \mid n$; say $n = p^{v(g_1-g_2)}n'$, and
\begin{align*}
t(ng_1,ng_2) & = p^{-m}d_p^H(ng_1,ng_2) \\
& = p^{-m}|n'|_pd_p^H(p^{v(g_1-g_2)}g_1,p^{v(g_1-g_2)}g_2) \\
& = p^{-m}p^{-v_p(n')}\cdot 1 \\
& = p^{-m}p^{v(g_1-g_2)}p^{-v(g_1-g_2)-v_p(n')} \\
& = t(g_1,g_2)p^{-v_p(n)} \\
& = |n|_pt(g_1,g_2).
\end{align*}
Now let us check that $t$ is dominated by $d_p$.
Suppose that $g_1,g_2\in G$ and $g_1-g_2\in H$.
Suppose $g_1-g_2$ is $p^n$-divisible in $G$ for some $n\geq 0$.
If $n < m$, then $t(g_1,g_2) \leq p^{-m} < p^{-n}$.
If $n\geq m$, then there exists $g\in G$ such that $g_1-g_2 = p^ng = p^{n-m}p^mg \in p^{n-m}H$, so $g_1-g_2$ is $p^{n-m}$-divisible in $H$, and $t(g_1,g_2) = p^{-m}d_p^H(g_1,g_2) \leq p^{-m}p^{-(n-m)} = p^{-n}$.
As $n$ was arbitrary with the property that $g_1-g_2$ was $p^n$-divisible in $G$, it follows that $t(g_1,g_2)\leq d_p(g_1,g_2)$ if $g_1-g_2\in H$.
If $g_1-g_2\notin H$ and $p^n$ divides $g_1-g_2$, then $n < m$ by the hypothesis on $m$, and $p^{m-n}(g_1-g_2)\in H$.
Thus $v(g_1-g_2)\leq m-n$, so $t(g_1,g_2) = p^{-m}p^{v(g_1-g_2)} \leq p^{-n}$.
Thus $t(g_1,g_2)\leq d_p(g_1,g_2)$ if $g_1-g_2\notin H$, so $t$ is dominated by $d_p$.
This completes the proof that $t$ is a $p$-adic sublinear pseudometric.
Now for any $g\notin H$, the definition of $t$ implies that $p^{-m} < t(g,0)$.
Choose a group homomorphism $f' : \langle g\rangle \to \mathbb{Z}_p$ with the property that $|f'(g)|_p = t(g,0)$.
Then $f'$ is $t$-contractive, so by Proposition \ref{PROP:all-states-extend}, it extends to a $t$-contractive $p$-adic functional $f$ on $G$.
As $f$ is $t$-contractive, $|f(h)|_p\leq t(h,0) \leq p^{-m} < t(g,0) = |f(g)|_p$ for all $h\in H$.
\end{proof}
\subsection{Spaces of functionals}\label{SUBSEC:state-spaces}
\begin{defn}\label{DEF:state-space}
Let $p$ be a prime and let $G$ be a torsion-free abelian group.
Let us use the symbol $G^{*p}$ to denote $Hom(G,\mathbb{Z}_p)$, the set of $p$-adic functionals on $G$.
\end{defn}
$G^{*p}$ is a torsion-free abelian group, and as it is a subgroup of $\mathbb{Z}_p^G$, it contains no non-zero elements that are divisible by arbitrarily high powers of $p$, so the $p$-adic pseudometric on $G^{*p}$ is in fact a metric.
In what follows, it will be useful to view $G^{*p}$ as a topological space.
It turns out that the most useful topology is not the metric topology arising from the $p$-adic metric on $G^{*p}$, but rather the topology that $G^{*p}$ inherits from $\mathbb{Z}_p^{G}$.
\begin{defn}\label{DEF:relative-top}
Let $p$ be a prime and let $G$ be a torsion-free abelian group.
Let us refer to the topology that $G^{*p}$ inherits as a subspace of the product space $\mathbb{Z}_p^G$ as the \emph{relative topology} on $G^{*p}$.
\end{defn}
$\mathbb{Z}_p^G$ is a product of compact spaces, and hence is compact by Tychonoff's Theorem.
It is easy to verify that $G^{*p}$ is closed in the relative topology, and hence is compact.
$\mathbb{Z}_p$ is a principal ideal domain, and $G^{*p}$ is a submodule of the $\mathbb{Z}_p$-module $\mathbb{Z}_p^G$, which is free, so $G^{*p}$ is free as well \cite[Theorem 9.8]{R:algebra}.
Thus is makes sense to speak of a $\mathbb{Z}_p$-basis and the $\mathbb{Z}_p$-rank of $G^{*p}$.
The metric topology on $G^{*p}$ is finer than the relative topology in general, and if the $\mathbb{Z}_p$-rank of $G^{*p}$ is finite, then the two topologies agree, but this need not be true if $G^{*p}$ has infinite rank, as the following example illustrates.
\begin{example}\label{EX:infinite-rank-independence}
Let $p$ be a prime, let $G = \bigoplus_{n\in\mathbb{N}}\mathbb{Z}$ and, for $n\in\mathbb{N}$, let $e_n$ denote the $n$th standard basis element of $G$.
For $m\in\mathbb{N}$, let $f_m$ denote the $p$-adic functional defined by $f_m(e_n) = \delta_{m,n}$, and let $f_0$ denote the $p$-adic functional defined by $f_0(e_n) = 1$ for all $n\in\mathbb{N}$.
Then the sequence $f_1,f_1+f_2,f_1+f_2+f_3,\ldots $ converges to $f_0$ in the relative topology, but not in the metric topology.
\end{example}
Note that any $p$-adic functional on $G$ is entirely determined by its values on a maximal independent subset of $G$, which means that, if $G$ has finite rank, then the $\mathbb{Z}_p$-rank of $G^{*p}$ cannot exceed the rank of $G$---more precisely, it cannot exceed the rank of $G$ minus the rank of $G_p$.
Proposition \ref{PROP:simple}, below, says that $G$ is $p$-simple but not $p$-divisible if and only if the $\mathbb{Z}_p$-rank of $G^{*p}$ is $1$.
The proof of this fact uses the following easy lemma.
\begin{lemma}\label{LEM:kernel-ideal}
Let $p$ be a prime, let $G$ be a torsion-free abelian group, and let $f$ be a $p$-adic functional on $G$.
Then $\ker (f)$ is a $p$-ideal of $G$.
\end{lemma}
\begin{proof}
$f$ is a group homomorphism into a torsion-free abelian group, so its kernel is a pure subgroup of $G$.
$f$ is continuous with respect to the topology arising from the $p$-adic pseudometric on $G$, so its kernel is closed, as it is the pre-image of the closed subset $\{ 0\}$ of the metric space $\mathbb{Z}_p$.
\end{proof}
\begin{prop}\label{PROP:simple}
Let $p$ be a prime and let $G$ be a torsion-free abelian group.
Then $G$ is $p$-simple but not $p$-divisible if and only if, up to multiplication by a non-zero constant in $\mathbb{Z}_p$, there is a unique non-zero $p$-adic functional on $G$.
\end{prop}
\begin{proof}
For the ``only if'' direction, suppose that $G$ is $p$-simple but not $p$-divisible, then pick $f_1, f_2 \in G^{*p}$, and suppose that $f_1$ is non-zero.
Then there exists $g\in G$ such that $f_1(g)\neq 0$, which means in particular that $g\notin G_p$.
Suppose without loss of generality that $|f_1(g)|_p\geq |f_2(g)|_p$.
Then $f_0 := f_2 - \frac{f_2(g)}{f_1(g)}f_1\in G^{*p}$, and $g\in \ker(f_0)$.
By Lemma \ref{LEM:kernel-ideal}, $\ker(f_0)$ is a $p$-ideal of $G$.
$\ker(f_0)$ contains $g$, so is bigger than $G_p$; thus, as $G$ is $p$-simple, $\ker(f_0) = G$.
Now let us prove the ``if'' direction.
Note that, if $G$ is $p$-divisible, then it has no non-zero $p$-adic functionals.
Thus suppose that $G$ is not $p$-divisible and not $p$-simple, and let us construct two $\mathbb{Z}_p$-independent $p$-adic functionals on $G$.
Let $J$ be a proper $p$-ideal of $G$ that is strictly larger than $G_p$.
As $J$ is proper, there exists $g\in G\backslash J$; by Corollary \ref{COR:separating-state}, there exists a $p$-adic functional $f_1$ with the property that $f_1(h) = 0$ for all $h\in J$ and $f_1(g)\neq 0$.
As $J$ is larger than $G_p$, it is not $p$-divisible, so by Corollary \ref{COR:states-exist}, there exists a non-zero $p$-adic functional on $J$, which extends to a $p$-adic functional $f_2$ on $G$ by Proposition \ref{PROP:all-states-extend}.
$f_1$ and $f_2$ are both non-zero, and neither is a multiple of the other because $f_1$ is $0$ on $J$ while $f_2$ is not.
\end{proof}
The next lemma characterizes the $p$-adic metric on $G^{*p}$ in terms of the absolute values of the images of elements of $G$.
\begin{lemma}\label{LEM:state-space-metric}
Let $p$ be a prime, let $G$ be a torsion-free abelian group with $p$-adic pseudometric $d_p$, and let $f$ be a $p$-adic functional on $G$.
Then
\[
d_p^{*p}(f,0) = \max \Big( \Big\{ \frac{|f(g)|_p}{d_p(g,0)} : g\in G, d_p(g,0)\neq 0\Big\}\cup \{ 0\}\Big),
\]
where $d_p^{*p}$ denotes the $p$-adic metric on $G^{*p}$.
\end{lemma}
\begin{proof}
Suppose that $f = p^k f'$ for some $f'\in G^{*p}$.
Then, for any $g\in G$,
\begin{align*}
|f(g)|_p & = |p^kf'(g)|_p = p^{-k}|f'(g)|_p \leq p^{-k}d_p(g,0).
\end{align*}
Thus $\frac{|f(g)|_p}{d_p(g,0)} \leq d_p^{*p}(f,0)$ for all $g\in G$ with $d_p(g,0)\neq 0$.
Now suppose that $|f(g)|_p\leq p^{-k} d_p(g,0)$ for all $g\in G$.
Then define a function $f': G\to \mathbb{Q}_p$ by $f'(g) = \frac{1}{p^k}f(g)$ for $g\in G$.
This is clearly a group homomorphism into $\mathbb{Q}_p$, and in fact it takes values in $\mathbb{Z}_p$ because $|f'(g)|_p = |\frac{1}{p^k}f(g)|_p = p^k|f(g)|_p \leq d_p(g,0)$, so in this case, $d_p^{*p}(f,0)\leq p^{-k}$.
Thus $d_p^{*p}(f,0)\leq \max \{ \frac{|f(g)|_p}{d_p(g,0)} : g\in G, d_p(g,0)\neq 0\}$.
\end{proof}
The following corollary is an immediate consequence of Lemma \ref{LEM:state-space-metric}.
\begin{cor}\label{COR:state-max-attained}
Let $p$ be a prime, let $G$ be a torsion-free abelian group with $p$-adic pseudometric $d_p$, and suppose that $f$ is a non-zero $p$-adic functional on $G$.
Then there exists $g\in G$ with $d_p(g,0) = 1$ such that $d_p^{*p}(f,0) = |f(g)|_p$, where $d_p^{*p}$ denotes the $p$-adic metric on $G^{*p}$.
\end{cor}
The goal of the next sequence of results is to characterize the subsets of $G^{*p}$ that are sufficiently rich to contain, for each $g\in G$, a $p$-adic functional that achieves the maximum possible absolute value on $g$.
This is the content of Proposition \ref{PROP:spanning-states}, which says that this is occurs for subsets of $G^{*p}$ that have $\mathbb{Z}_p$-spans that are dense in $G^{*p}$ in the relative topology.
The proof of this result uses Lemma \ref{LEM:independence}, which in turn uses the following technical lemma.
\begin{lemma}\label{LEM:independence-rank}
Let $p$ be a prime, let $G$ be a torsion-free abelian group, and let $F\subset G^{*p}$ and $X\subset G$ be finite subsets of $G^{*p}$ and $G$ respectively, with the property that $\{ f|_X : f\in F\} \subseteq \mathbb{Z}_p^X$ is $\mathbb{Z}_p$-independent.
Then there exists $X'\subseteq X$ such that $X'$ and $F$ contain the same number of elements, and $\{ f|_{X'} : f\in F\}\subseteq \mathbb{Z}_p^{X'}$ is $\mathbb{Z}_p$-independent.
\end{lemma}
\begin{proof}
Let $F = \{ f_1,\ldots,f_k\}$ and $X = \{ g_1, \ldots, g_l\}$.
Certainly $l\geq k$, as $\{ f_i|_X : 1\leq i\leq k\}$ is independent.
Consider the matrix $B\in M_{k\times l}(\mathbb{Z}_p)$, the $(i,j)$th entry of which is $f_i(g_j)$.
View $B$ as a matrix over $\mathbb{Q}_p$; then if $l>k$ the Rank-Nullity Theorem says that there exists a non-zero vector $v = (v_1,\ldots, v_l)^t \in\mathbb{Q}_p^l$ such that $Bv = 0$.
Suppose that the entry $v_l$ has maximal absolute value among all entries of $v$; then, by replacing $v$ with $v/v_l$, we may suppose that $v$ in fact has entries in $\mathbb{Z}_p$, and $v_l = 1$.
Then $\{ f|_{X\backslash \{ g_l\}} : f\in F\}$ is $\mathbb{Z}_p$-independent.
To see this, suppose that $\alpha_1f_1+\cdots + \alpha_kf_k$ is $0$ on $X\backslash \{ g_l\}$, where $\alpha_i\in\mathbb{Z}_p$ for $1\leq i\leq k$.
Let us pick $\epsilon>0$ and show that $|\alpha_1f_1(g_l)+\cdots +\alpha_kf_k(g_l)|_p<\epsilon$.
$\mathbb{Z}$ is dense in $\mathbb{Z}_p$, so we can pick an element $w = (w_1,\ldots,w_l)^t\in\mathbb{Z}^l$ such that $|v_j-w_j|_p<\epsilon$ for all $1\leq j\leq l$.
We may insist further that $w_l=1$ because $v_l=1$.
Then $|w_1f_i(g_l)+\cdots + w_{l-1}f_i(g_{l-1})+f_i(g_l)|_p<\epsilon$.
Thus it is true that, for each $i\leq k$,
\begin{align*}
f_i(g_l) & = \big( f_i(g_l) + \sum_{j=1}^{l-1}w_jf_i(g_j)\big) - \sum_{j=1}^{l-1}w_jf_i(g_j),\\
\intertext{which implies}
\sum_{i=1}^k \alpha_if_i(g_l) & = \sum_{i=1}^k\alpha_i \Big[ \big( f_i(g_l) + \sum_{j=1}^{l-1}w_jf_i(g_j)\big) - \sum_{j=1}^{l-1}w_jf_i(g_j) \Big] \\
& = \sum_{i=1}^k\alpha_i \big( f_i(g_l) + \sum_{j=1}^{l-1}w_jf_i(g_j)\big) - \sum_{j=1}^{l-1}w_j \sum_{i=1}^k \alpha_i f_i(g_j) \\
& = \sum_{i=1}^k\alpha_i \big( f_i(g_l) + \sum_{j=1}^{l-1}w_jf_i(g_j)\big) - \sum_{j=1}^{l-1}w_j \cdot 0.
\end{align*}
Each term in the first sum has absolute value less than $|\alpha_i|\epsilon \leq \epsilon$, so by the strong triangle inequality, the entire sum has absolute value less than $\epsilon$.
Thus $\sum_{i=1}^k\alpha_if_i(g_l) = 0$, implying $\{ f|_X : f\in F\}$ is $\mathbb{Z}_p$-dependent, which is a contradiction.
Thus $\{ f_{X\backslash \{ g_l\}} : f\in F\}$ is $\mathbb{Z}_p$-independent.
We can apply the same argument to $X\backslash \{ g_l\}$ to show that either $X\backslash \{ g_l\}$ has the same number of elements as $F$, or else there exists a smaller set, the restriction of $F$ to which is $\mathbb{Z}_p$-independent.
Proceeding in this fashion eventually produces a set with the same size as $F$.
\end{proof}
The next lemma says that a functional $f\in G^{*p}$ can be separated from a subset $F\subseteq G^{*p}$ by an element of $G$ if and only if it is not in the relative-topology closure of the span of $F$.
\begin{lemma}\label{LEM:independence}
Let $p$ be a prime, let $G$ be a torsion-free abelian group, let $F$ be a non-empty subset of $G^{*p}$, and let $f\in G^{*p}$.
The following are equivalent:
\begin{enumerate}
\item There exists $g\in G$ such that $|f(g)|_p>|f'(g)|_p$ for all $f'\in F$.
\item $f$ is not in the relative-topology closure of the $\mathbb{Z}_p$-span of $F$.
\end{enumerate}
\end{lemma}
\begin{proof}
To prove that (1) implies (2), suppose for a contradiction that there exists $g\in G$ satisfying condition (1), but $f$ lies in the relative-topology closure of the $\mathbb{Z}_p$-span of $F$.
Let $M = \sup_{f'\in F} \{ |f'(g)|_p\}$.
Then, as $\mathbb{Z}_p$ is a discrete valuation ring, this supremum is achieved on $F$; that is, $M = \max_{f'\in F} \{|f'(g)|_p\} < $ $|f(g)|_p$.
Similarly, the fact that $\mathbb{Z}_p$ is a discrete valuation ring implies that the set $T_{|f(g)|_p} := \{ \beta\in\mathbb{Z}_p : |\beta |_p = |f(g)|_p\}$ is open in $\mathbb{Z}_p$.
Consider the relative-topology open set $\prod_{g'\in G}U_{g'}$, where $U_{g'} = \mathbb{Z}_p$ if $g'\neq g$ and $U_g = T_{|f(g)|_p}$.
$f$ is in this open set, so, by assumption, there exists some $\mathbb{Z}_p$ combination $\alpha_1f_1 + \cdots + \alpha_nf_n$ in this set, where $\alpha_i\in\mathbb{Z}_p$ and $f_1,\ldots , f_n\in F$.
But then, by the strong triangle inequality, $|f(g)|_p = $ $|\alpha_1f_1(g) + \cdots + \alpha_nf_n(g)|_p \leq $ $\max_{1\leq i \leq n} \{ |\alpha_i|_p|f_i(g)|_p\} \leq M < |f(g)|_p$, which is a contradiction.
Thus $f$ is not in the relative-topology closure of the $\mathbb{Z}_p$-span of $F$ after all.
Now let us prove that (2) implies (1).
The statement that $f$ is not in the relative-topology closure of the $\mathbb{Z}_p$-span of $F$ means that there exists a finite subset $X\subset G$ such that $f|_X$ is not in the $\mathbb{Z}_p$-span of $\{ f'|_X : f'\in F\}$.
Let $f_1 = f|_X$, and let $\{ f_2,\ldots, f_k\}$ be a basis for the $\mathbb{Z}_p$-module spanned by $\{ f'|_X : f'\in F\}$, which is a submodule of the free $\mathbb{Z}_p$-module $\mathbb{Z}_p^X$, and hence is free itself.
Then $\{ f_1,\ldots,f_k\}$ is independent over $\mathbb{Z}_p$.
By Lemma \ref{LEM:independence-rank}, we may, in fact, suppose that $X$ contains exactly $k$ elements, say $X = \{ g_1,\ldots, g_k\}$.
Let $A$ denote the $k\times k$ matrix, the $(i,j)$th entry of which is $f_i(g_j)\in\mathbb{Z}_p$.
$\{ f_1|_X,\ldots , f_k|_X\}$ is $\mathbb{Z}_p$-independent, so $A$ has non-zero determinant.
Let $\adj(A)$ denote the adjugate of $A$, that is, the transpose of the matrix of cofactors of $A$.
Then the entries of $\adj(A)$ are in $\mathbb{Z}_p$, and $A\adj(A) = \det(A)I$, where $I$ is the $k\times k$ identity matrix.
Let $e_1$ denote the first standard basis element of $\mathbb{Z}_p^k$.
Let $v = (v_1,\ldots,v_k)^t =$ $\adj(A)e_1\in\mathbb{Z}_p^k$, so that $Av = \det(A)e_1$.
$\mathbb{Z}$ is dense in $\mathbb{Z}_p$, so there exists $w = (w_1,\ldots,w_k)^t $ $\in\mathbb{Z}^k$ such that $|v_i-w_i|_p < \det(A)/2$ for each index $1\leq i \leq k$.
Then, by the strong triangle inequality, each entry of $Aw$ differs from the corresponding entry of $Av$ by something with absolute value less than $\max_{1\leq i,j\leq k} \{ |f_i(g_j)|_p\} \det(A)/2 \leq \det(A)/2$.
In particular, the first entry of $Aw$ has absolute value greater than $\det(A)/2$, and all other entries have absolute value less than this.
Thus $|f_1(w_1g_1+\cdots + w_kg_k)|_p > $ $\det(A)/2 > $ $|f_i(w_1g_1+\cdots + w_kg_k)|_p$ for all $2\leq i \leq k$.
$\{ f_2,\ldots, f_k\}$ is a basis for the $\mathbb{Z}_p$-module spanned by $\{ f'|_X : f'\in F\}$, so, given any $f'\in F$, there exist $\beta_2,\ldots,\beta_k\in\mathbb{Z}_p$ such that $f'(w_1g_1+\cdots + w_kg_k) = $ $\sum_{j=2}^k \beta_j f_j(w_1g_1+\cdots + w_kg_k)$.
But by the strong triangle inequality, the absolute value of this sum is at most $\max_{2\leq j \leq k} \{ |\beta_j|_p|f_j(w_1g_1+\cdots + w_kg_k)|_p\} \leq $ $\det(A)/2 < $ $|f(w_1g_1+\cdots + w_kg_k)|_p$.
Thus statement (1) is true for the element $w_1g_1+\cdots w_kg_k\in G$.
\end{proof}
The next result describes the subsets of $G^{*p}$, the relative-topology closures of the the spans of which are all of $G^{*p}$.
\begin{prop}\label{PROP:spanning-states}
Let $p$ be a prime, let $G$ be a torsion-free abelian group with $p$-adic pseudometric $d_p$, and let $F$ be a subset of $G^{*p}$.
The following are equivalent:
\begin{enumerate}
\item For each $g\in G$, there exists $f\in F$ such that $|f(g)|_p = d_p(g,0)$.
\item $G^{*p}$ is the relative-topology closure of the $\mathbb{Z}_p$-span of $F$.
\end{enumerate}
\end{prop}
\begin{proof}
To prove that (2) implies (1), suppose that $G^{*p}$ is the relative-topology closure of the $Z_p$-span of $F$, and pick $g\in G$.
By Corollary \ref{COR:state-hits-distance}, there exists a $p$-adic functional $f$ on $G$ such that $|f(g)|_p = d_p(g,0)$.
We can write $f(g)$ in terms of $\{ f'(g) : f'\in F\}$:
\begin{align*}
f(g) & = \alpha_1f_1(g) + \cdots + \alpha_kf_k(g),
\end{align*}
where $\alpha_i\in\mathbb{Z}_p$ and $(f_i)_{i=1}^k$ are elements of $F$.
For each $i\leq k$, $|f_i(g)|_p \leq d_p(g,0)$.
If the stronger statement $|f_i(g)|_p < d_p(g,0)$ were true for all $i\leq k$, then, by the strong triangle inequality, it would follow that
\begin{align*}
|f(g)|_p & \leq \max_{i\leq k} \{ |\alpha_i|_p|f_i(g)|_p\} < d_p(g,0),
\end{align*}
which is a contradiction.
Thus $|f_i(g)|_p = d_p(g,0)$ for at least one $f_i\in F$.
To prove that (1) implies (2), suppose that statement (1) is true, but the relative-topology closure of the $\mathbb{Z}_p$-span of $F$ is properly contained in $G^{*p}$.
Then pick a $p$-adic functional $f$ on $G$ that is not in the relative-topology closure of the $\mathbb{Z}_p$-span of $F$.
By Lemma \ref{LEM:independence}, there exists $g\in G$ such that $|f(g)|_p > $ $|f'(g)|_p$ for all $f'\in F$.
But this is impossible because, by statement (1), there exists $f'\in F$ such that $|f'(g)|_p = d_p(g,0)$, and $|f(g)|_p$ cannot exceed this value because $f$ is a $p$-adic functional on $G$.
\end{proof}
\subsection{Double duals}\label{SUBSEC:double-duals}
If $p$ is a prime and $G$ is a torsion-free abelian group, then the dual space $G^{*p}$ of $p$-adic functionals is also a torsion-free abelian group, and so has its own dual space $G^{*p*p}$.
There is a natural homomorphism $\Phi^p$ from $G$ to $G^{*p*p}$ (Definition \ref{DEF:phi-p}, below), the image of which is dense in $G^{*p*p}$ (Theorem \ref{THM:density}, below).
\begin{defn}\label{DEF:g-hat}
Let $p$ be a prime, let $G$ be a torsion-free abelian group, and let $g\in G$.
Define a function $\hat{g}^p : G^{*p}\to \mathbb{Z}_p$ by $\hat{g}^p(f) = f(g)$ for $f\in G^{*p}$.
\end{defn}
$\hat{g}^p$ is clearly a group homomorphism into $\mathbb{Z}_p$, so is an element of $G^{*p*p}$.
\begin{defn}\label{DEF:phi-p}
Let $p$ be a prime and let $G$ be a torsion-free abelian group.
Define a function $\Phi^p$ by
\begin{align*}
\Phi^p : G & \to G^{*p*p} \\
g & \mapsto \hat{g}^p.
\end{align*}
\end{defn}
Note that $\Phi^p$ is a group homomorphism: $\widehat{g+h}^p(f) = f(g+h) = f(g) + f(h) = \hat{g}^p(f) + \hat{h}^p(f)$.
Its kernel certainly contains $G_p$, and, by Corollary \ref{COR:state-hits-distance}, $G_p$ is the entire kernel.
Let $\hat{G}^p$ denote the image of $\Phi^p$: $\hat{G}^p := \{ \hat{g}^p : g\in G\}\subset G^{*p*p}$.
\begin{remark}\label{REM:not-tensor-product}
By tensoring with $\mathbb{Z}_p$ over $\mathbb{Z}$, we can embed any torsion-free abelian group $G$ into a free $\mathbb{Z}_p$-module.
Such embeddings have been studied in \cite{M:tensor-product} and \cite{K:matrices}; they are similar to the homomorphism $\Phi^p : G \to G^{*p*p}$, but are not generally the same.
If $G_p$ is not trivial, then $\Phi^p$ is not an embedding, and even when it is an embedding, it is not necessarily equivalent to the tensor product embedding.
Example \ref{EX:favourite} illustrates this fact: in that example, $G$ is a rank-$2$ group, so its tensor product with $\mathbb{Z}_3$ has $\mathbb{Z}_3$-rank equal to $2$, while its $p$-adic dual group $G^{*p}$ is generated over $\mathbb{Z}_3$ by a single row, and so $G^{*p*p}$ has rank equal to $1$.
\end{remark}
The main result of this section is the following.
\begin{thm}\label{THM:density}
Let $p$ be a prime and let $G$ be a torsion-free abelian group with $p$-adic pseudometric $d_p$.
Then $\Phi^p : G\to\hat{G}^p\subseteq G^{*p*p}$ is a pseudo-isometry, and $d_p(g,0) = d_p^{*p*p}(\hat{g}^p,0) = \max_{f\in G^{*p}}\{ |\hat{g}^p(f)|_p\}$, where $d_p^{*p*p}$ denotes the $p$-adic metric on $G^{*p*p}$.
Furthermore $\hat{G}^p$ is dense in $G^{*p*p}$ in the relative topology.
\end{thm}
\begin{proof}
Pick $g\in G$, and let us show that $d_p^{*p*p}(\hat{g}^p,0) = d_p(g,0) = \max_{f\in G^{*p}} \{ |\hat{g}^p(f)|_p\}$.
Certainly for any $f\in G^{*p}$, it is true that $|f(g)|_p\leq d_p(g,0)$, so it is enough to show that this value is attained for some $p$-adic functional $f$.
By Corollary \ref{COR:state-max-attained} applied to $\hat{g}^p$, there exists $f\in G^{*p}$ with $d_p^{*p}(f,0) = 1$ such that
\[
d_p^{*p*p}(\hat{g}^p,0) = |\hat{g}^p(f)|_p = |f(g)|_p \leq d_p(g,0) = d_p(g,0)d_p^{*p}(f,0) \leq |f(g)|_p,
\]
where the last inequality follows from Lemma \ref{LEM:state-space-metric}.
Thus $d_p^{*p*p}(\hat{g}^p,0) = d_p(g,0) = |\hat{g}^p(f)|_p = |f(g)|_p$.
Now let us show that $\hat{G}^p$ is dense in $G^{*p*p}$ in the relative topology.
Pick $f\in G^{*p}$.
By Corollary \ref{COR:state-max-attained}, there exists $g\in G$ with $d_p^{*p}(f,0) = |f(g)|_p = |\hat{g}^p(f)|_p$; by Proposition \ref{PROP:spanning-states}, this implies that the $\mathbb{Z}_p$-span of $\hat{G}^p$ is dense in $G^{*p*p}$ in the relative topology.
So given $\phi\in G^{*p*p}$, a finite subset $F = \{ f_1,\ldots,f_k\}\subseteq G^{*p}$, and $\epsilon > 0$, there exist $g_1,\ldots, g_l\in G$ and $\alpha_1,\ldots, \alpha_l\in \mathbb{Z}_p$ such that
\begin{align*}
|(\sum\alpha_j\hat{g}^p_j)(f_i) - \phi(f_i)|_p < \epsilon
\end{align*}
for each $1\leq i\leq k$.
For $1\leq j\leq l$ choose $a_j\in \mathbb{Z}$ such that $|a_j-\alpha_j|_p < \epsilon$.
Then $\sum a_j\hat{g}^p_j\in \hat{G}^p$ because $\hat{G}^p$ is a group, and
\begin{align*}
& |(\sum a_j\hat{g}^p_j )(f_i) - \phi(f_i)|_p \\
\leq & \max\{ |(\sum \alpha_j\hat{g}^p_j)(f_i)-\phi (f_i)|_p, |(\sum a_j\hat{g}^p_j)(f_i) - (\sum \alpha_j\hat{g}^p_j )(f_i)|_p\} \\
\leq & \max \{ \epsilon, |(a_1-\alpha_1)\hat{g}^p_1(f_i)|_p,\ldots, |(a_l-\alpha_l)\hat{g}^p_l(f_i)|_p\} \\
\leq & \max \{ \epsilon, \epsilon, \ldots, \epsilon\} = \epsilon.
\end{align*}
Thus $\hat{G}^p$ is dense in $G^{*p*p}$ in the relative topology.
\end{proof}
\section{Pro-$p$ completions}\label{SEC:completions}
Readers who are familiar with profinite groups might have noticed that the double dual $G^{*p*p}$ of a torsion-free abelian group $G$ shares important features with the pro-$p$ completion of $G$.
This is not a coincidence---the two objects are, in fact, the same.
\begin{defn}\label{DEF:pro-p}
A topological group $G$ is called a \emph{pro-$p$ group} if it is isomorphic as a topological group to the inverse limit of an inverse system of finite $p$-groups.
\end{defn}
The topology in Definition \ref{DEF:pro-p} may require some explanation.
Any inverse limit of an inverse system of groups has a standard realization as a subgroup of the direct product:
\[
\varprojlim (G_\lambda)_{\lambda\in\Lambda} = \{ (x_\lambda)_{\lambda\in\Lambda} \in \prod_{\lambda\in\Lambda} G_\lambda : \phi_{\mu\lambda}(x_\lambda) = x_\mu \text{ for all } \lambda\leq \mu\},
\]
where, for $\lambda\leq \mu$, $\phi_{\mu\lambda} : G_\lambda\to G_\mu$ is the corresponding map in the inverse system.
In the pro-$p$ case, the groups $G_\lambda$ in the direct product are finite $p$-groups, and hence are topological groups with the discrete topology.
The topology on the inverse limit is the relative topology that it inherits from the direct product.
Given an arbitrary group $G$, consider the set of all normal subgroups $N$ of $G$ for which $G/N$ is a finite $p$-group.
This is a partially ordered set under inclusion, which means that there is an inverse system of natural homomorphisms $\phi_{\mu\lambda} : G/N_\lambda \to G/N_\mu$ between the quotients, where $N_\mu\leq N_\lambda$.
The \emph{pro-$p$ completion} $\mathcal{K}_p(G)$ of $G$ is the inverse limit of this inverse system.
There is a natural embedding $\Psi^p : G\to \mathcal{K}_p(G)$, which sends $g\in G$ to $(gN_\lambda)_{\lambda\in\Lambda}$, and the image of $G$ under this embedding is a dense subgroup.
$\mathcal{K}_p(G)$ is clearly a pro-$p$ group.
Moreover it satisfies the following universal property with respect to $G$: if there exists a pro-$p$ group $H$ and a group homomorphism $\theta : G\to H$, then there exists a continuous homomorphism $\eta : \mathcal{K}_p(G)\to H$ such that $\theta = \eta\circ\Psi^p$.
If $G$ is a torsion-free abelian group, then $G^{*p*p}$ is a free $\mathbb{Z}_p$-module, and hence is a pro-$p$ group (and the pro-$p$ topology agrees with the relative topology on $G^{*p*p}$ \cite[Proposition 2.2.1, parts (a) and (b)]{RZ:profinite}).
Thus the universal property guarantees the existence of a continuous homomorphism $\eta : \mathcal{K}_p(G)\to G^{*p*p}$ such that $\eta(\Psi^p(g)) = \hat{g}^p$ for all $g\in G$.
Then in order to show that $G^{*p*p}$ is the pro-$p$ completion, it will be necessary to find a continuous homomorphism in the other direction.
Let us do so by finding a compatible system of homomorphisms from $G^{*p*p}$ into all of the finite $p$-group quotients of $G$, and then use the universal property of the inverse limit to construct the required homomorphism.
\begin{lemma}\label{LEM:quotient}
Let $p$ be a prime, let $G$ be a torsion-free abelian group, and let $H$ be a subgroup of $G$ of index $p^k$ for some $k\in\mathbb{N}$.
Then there exists a unique continuous homomorphism $\theta_H : G^{*p*p}\to G/H$ such that $\theta_H(\hat{g}^p) = g+H$ for all $g\in G$.
\end{lemma}
\begin{proof}
There exists $m\in\mathbb{N}\cup\{ 0\}$ such that the maximum order of any element $g+H\in G/H$ is $p^m$.
By Lemma \ref{LEM:subgroup-separation}, if $g,g'\in G$ with $d_p(g,g')\leq p^{-m}$, then $g$ and $g'$ are in the same $H$-coset.
Let $\{g_1,\ldots, g_{p^k}\}$ be a complete set of coset representatives of $G/H$, with $g_1\in H$, and let $F$ be a finite set of $p$-adic functionals such that, for each $2\leq i\leq p^k$, there exists $f\in F$ such that $|f(g_i)|_p > p^{-m} \geq |f(h)|_p$ for all $h\in H$.
(Such a set $F$ exists by Proposition \ref{PROP:separate-from-subgroup}.)
For $\phi\in G^{*p*p}$, consider the open neighbourhood $U_\phi$ of $\phi$ defined by $U_\phi = \prod_{f\in G^{*p}} U_f$, where $U_f = \{ \beta\in \mathbb{Z}_p : |\beta-\phi(f)|_p \leq p^{-m}\}$ if $f \in F$, and for $f\notin F$, $U_f = \mathbb{Z}_p$.
Then $U_\phi$ is an open subset of $\mathbb{Z}_p^{G^{*p}}$, so its intersection $U_\phi'$ with $G^{*p*p}$ is open in the relative topology on $G^{*p*p}$.
$\hat{G}^p$ is dense in $G^{*p*p}$ in the relative topology, so there exists some $g\in G$ such that $\hat{g}^p\in U_{\phi}'$.
Set $\theta_H(\phi) = g+H$, and let us show that this is well-defined.
Let us show that, if we had chosen a different element $g'\in G$ with $\hat{g}'^p\in U_\phi'$, then $g+H = g'+H$.
To see this, note that $\hat{g}'^p\in U_\phi'$ implies that $|f(g-g')|_p\leq p^{-m}$ for all $f\in F$.
Suppose that $g_j$ is the chosen representative of the coset $g-g' + H$; this means that $g_j - g + g'\in H$.
If $j = 1$, then $g_1\in H$, so $g-g'\in H$ as required.
Otherwise, pick $f\in F$ such that $|f(g_j)|_p > p^{-m} \geq |f(h)|_p$ for all $h\in H$, and note that, by the strong triangle inequality,
\[
p^{-m} < |f(g_j)|_p \leq \max \{ |f(g-g')|_p,|f(g_j-g+g')|_p\}.
\]
But $g_j-g+g'\in H$, so $|f(g_j-g+g')|_p \leq p^{-m}$, while $|f(g-g')|_p \leq \max\{ |f(g)-\phi(f)|_p,|f(g')-\phi(f)|_p\}$, which is less than or equal to $p^{-m}$ because $\hat{g}$ and $\hat{g}'$ are both in $U_\phi$.
This is a contradiction, so $g-g'\in H$ after all, and $\theta_H$ is well defined.
To see that $\theta_H$ is a group homomorphism, choose $\phi_1,\phi_2\in G^{*p*p}$ and observe that, if $g_1,g_2\in G$ satisfy $\hat{g}_1^p\in U_{\phi_1}$ and $\hat{g}_2^p\in U_{\phi_2}$, then $\widehat{g_1+g_2}^p\in U_{\phi_1+\phi_2}$ by the strong triangle inequality.
To see that $\theta_H$ is continuous, pick $g+H\in G/H$ and consider an element in its pre-image set: $\phi\in \theta_H^{-1}(\{ g+H\})$.
Then by the argument above, the open neighbourhood $U_\phi'\subseteq \theta_H^{-1}(g+H)$, so the pre-image of $g+H$ is open and $\theta_H$ is continuous.
To see that $\theta_H$ is unique with these properties, suppose that $\theta_H': G^{*p*p}\to G/H$ is another continuous group homomorphism such that $\theta_H'(\hat{g}^p) = g+H$ for all $g\in G$.
Then $\theta_H-\theta_H' : G^{*p*p}\to G/H$ is a continuous group homomorphism.
To see that it is a group homomorphism is easy.
To see that it is continuous, choose $g+H\in G/H$ and suppose $\phi\in (\theta_H-\theta_H')^{-1}(g+H)$.
Say $\theta_H(\phi) = g_1+H$ and $\theta_H'(\phi) = g_2+H$; then $g_1+g_2 +H = g+H$.
As $\theta_H$ and $\theta_H'$ are continuous, there exists open neighbourhoods $U_1,U_2$ of $\phi$ such that $U_1\subseteq \theta_H^{-1}(g_1+H)$ and $U_2\subseteq \theta_{H}'^{-1}(g_2+H)$; the open neighbourhood $U_1\cap U_2$ of $\phi$ then has the property that $U_1\cap U_2\subseteq (\theta_H-\theta_H')^{-1}(g+H)$.
Thus $\theta_H-\theta_H'$ is continuous.
The kernel of $\theta_H-\theta_H'$ is then a closed subset of $G^{*p*p}$, and, as it contains the dense subset $\hat{G}^p$, it must be everything.
\end{proof}
\begin{thm}\label{THM:pro-p}
Let $p$ be a prime and let $G$ be a torsion-free abelian group.
Then there is an isomorphism $\eta : \mathcal{K}_p(G) \to G^{*p*p}$ with the properties that $\eta(\Psi^p(g)) = \hat{g}^p$ for all $g\in G$ and $\eta$ is a homeomorphism.
\end{thm}
\begin{proof}
The existence of a continuous group homomorphism $\eta$ such that $\eta(\Psi^p(g)) = \hat{g}^p$ is guaranteed by the universal property of the pro-$p$ completion.
To see that $\eta$ is an isomorphism, let us construct an inverse homomorphism.
Consider the set of subgroups $H$ of $G$ such that $[G:H] = p^k$ for some $k\in\mathbb{N}\cup\{ 0\}$.
By Lemma \ref{LEM:quotient}, for each of these subgroups, there exists a continuous homomorphism $\theta_H:G^{*p*p}\to G/H$; moreover these homomorphisms are easily seen to be compatible with the natural homomorphisms between the finite $p$-group quotients of $G$.
Therefore by the universal property of the inverse limit, there is a continuous group homomorphism $\rho : G^{*p*p}\to \mathcal{K}_p(G)$.
Moreover, for any $g\in G$, the element $\hat{g}^p$ has image $g+H$ under $\theta_H$, which implies that $\rho(\hat{g}^p) = \Psi^p(g)$.
Let us show that $\rho$ is an inverse for $\eta$.
Note that $\rho\circ \eta$ is continuous, and, for $g\in G$, it is true that $\rho(\eta(\Psi^p(g))) = \rho(\hat{g}^p) = \Psi^p(g)$.
Thus $\rho\circ\eta$ is the identity on the dense subgroup $\{ \Psi^p(g) : g\in G\}$, so $\rho\circ\eta - id_{\mathcal{K}_p(G)}$ is a continuous group homomorphism that contains a dense subgroup in its kernel, and hence is trivial.
A similar argument shows that $\eta\circ\rho = id_{G^{*p*p}}$.
\end{proof}
\section{Classification}\label{SEC:classification}
The theory of $p$-adic functionals can be used to classify finite-rank torsion-free abelian groups.
Specifically, the collection of homomorphisms $\Phi^p : G\to G^{*p*p}$ for a torsion-free abelian group $G$, as $p$ ranges over the set of all primes, gives enough information to recover the isomorphism class of $G$.
Let $P$ denote the set consisting of all integer primes and let $G$ be a torsion-free abelian group.
Pick a maximal independent subset $\mathcal{B}$ of $G$.
For an element $(\alpha_\lambda)_{\lambda\in\Lambda} \in \mathbb{Z}_p^{\Lambda}$, define the sup-norm $\| (\alpha_\lambda)_{\lambda\in\Lambda} \|_p = \sup_{\lambda\in\Lambda} \{ |\alpha_\lambda|_p\}$.
It is easy to verify that $\|\cdot \|_p$ is indeed a norm, and that it satisfies the strong triangle inequality.
$G^{*p*p}$ is isomorphic to $\mathbb{Z}_p^{\Lambda_p}$ for some index set $\Lambda_p$; if we identify $G^{*p*p}$ with this latter group, then the homomorphism $\Phi^p$ sends elements of $G$ to $\Lambda_p$-tuples in $\mathbb{Z}_p^{\Lambda_p}$.
Moreover, by Theorem \ref{THM:density}, $\Phi^p$ is a pseudo-isometry, meaning that $d_p(g,0) = \| \Phi^p(g)\|_p$.
This is a powerful fact: it implies that $G$ is entirely determined by the values $(\Phi^p(g))_{p\in P}$ as $g$ ranges over the elements of $\mathcal{B}$.
Indeed, $G$ is isomorphic to a subgroup of $\langle\mathcal{B}\rangle\otimes \mathbb{Q}$, so in order to specify $G$ it is enough to be able to say, for each element $g = a_1g_1+\cdots +a_kg_k\in \langle\mathcal{B}\rangle\subseteq G$ and $m\in\mathbb{N}$, whether or not $g$ is in $mG$.
But $g\in mG$ if and only if $|m|_p \geq d_p(g,0) = \| \Phi^p(g)\|_p = \| a_1\Phi^p(g_1) + \cdots + a_k\Phi^p(g_k)\|_p$ for all $p\in P$.
This is summarized in the following proposition.
\begin{prop}\label{PROP:classification-1}
Let $G$ be a torsion-free abelian group, let $P$ denote the set of integer primes, and, for each $p\in P$, identify $G^{*p*p}$ with the group $\mathbb{Z}_p^{\Lambda_p}$ for some index set $\Lambda_p$.
If $\mathcal{B}$ is any maximal independent subset of $G$, then the values $(\Phi^p(g))_{p\in P, g\in\mathcal{B}}$ completely determine $G$.
\end{prop}
In the case that $G$ has finite rank, the data that specify $G$ can be stated more explicitly in terms of the following definition.
\begin{defn}\label{DEF:group-determined}
Let $n\in \mathbb{N}$, let $P$ be the set of all primes and let $(A_p)_{p\in P}$ be a sequence of matrices with $A_p\in M_{n_p\times n}(\mathbb{Z}_p)$.
Then the \emph{group determined by $(A_p)_{p\in P}$} is the torsion-free abelian group
\[
\mathcal{G}((A_p)_{p\in P}) := \{ v\in \mathbb{Q}^n : A_pv\in \mathbb{Z}_p^{n_p} \text{ for all } p\in P\}.
\]
\end{defn}
\begin{prop}\label{PROP:classification-2}
Let $P$ denote the set of integer primes and let $G$ be a torsion-free abelian group with finite rank, $n$.
Choose
\begin{enumerate}
\item a maximal independent set $\{ g_1,\ldots,g_n\}\subset G$; and
\item for each $p\in P$, a $\mathbb{Z}_p$-basis $\{ f_1^{(p)},\ldots, f_{n_p}^{(p)}\}$ for $G^{*p}$.
\end{enumerate}
For each $p\in P$, let $A_p = (f_i^{(p)}(g_j)) \in M_{n_p\times n}(\mathbb{Z}_p)$.
Then $G$ is isomorphic to $\mathcal{G}((A_p)_{p\in P})$.
\end{prop}
Thus the finite-rank torsion-free abelian group $G$ is completely described by a sequence, indexed by the prime numbers, of matrices $A_p$ with entries in $\mathbb{Z}_p$.
The rows of the $p$th matrix correspond to elements of a basis for $G^{*p}$.
This fact places a restriction on the structure of $A_p$; this restriction is expressed in terms of the following definition.
\begin{defn}\label{DEF:column-span}
Let $p$ be a prime, let $m,n\in\mathbb{N}$, and let $A\in M_{m\times n}(\mathbb{Z}_p)$.
Then define the \emph{column span} of $A$ to be
\[
C(A) := \spa_\mathbb{Z} \{ v/\| v\|_p : v \text{ is a non-zero column of } A\},
\]
where, for a vector $v\in\mathbb{Z}_p^{n}$, $\| v\|_p$ is the maximum of the $p$-adic absolute value of its entries.
\end{defn}
Then, by Theorem \ref{THM:density}, the matrices $A_p$ in Proposition \ref{PROP:classification-2} have the property that the column span $C(A_p)$ of $A_p$ is dense in $\mathbb{Z}_p^{n_p}$.
This implies in particular that $n_p\leq n$ for all primes $p$.
\begin{example}\label{EX:favourite-matrix-sequence}
Let $G$ be the group from Example \ref{EX:favourite}.
Then for $p\neq 3$, by choosing the most obvious basis for $G^{*p}$, we can arrange for the matrix $A_p$ to be $\twotwo{1}{0}{0}{1}$.
The $3$-adic functional described in that example gives rise to the matrix $A_3 = (1-\lambda,1)$, where $\lambda$ is the root of $x^2-5x+3=0$ for which $|\lambda|_3<1$.
\end{example}
\begin{example}\label{EX:two-three-five}
Let $G$ be the subgroup of $\mathbb{Q}^2$ generated by $\mathbb{Z}[1/2]\oplus \mathbb{Z}[1/3]$ and $\mathbb{Z}[1/5]{{1}\choose {1}}$.
Then the obvious choices of bases for $G^{*2}$, $G^{*3}$, and $G^{*5}$ yield matrices $A_2 = (0,1)$, $A_3 = (1,0)$, and $A_5 = (1,-1)$.
For $p > 5$, by choosing the most obvious basis for $G^{*p}$, we can arrange for the matrix $A_p$ to be $\twotwo{1}{0}{0}{1}$.
\end{example}
\begin{defn}\label{DEF:factored}
Let $P$ denote the set of primes and let $n\in\mathbb{N}$.
Suppose that $(m_p)_{p\in P}$ is a sequence of non-negative integers and $(A_p)_{p\in P}$ is a sequence of matrices satisfying
\begin{enumerate}
\item $A_p\in M_{m_p\times n}(\mathbb{Z}_p)$;
\item either $m_p = 1$ and $A_p$ is the zero row in $M_{1\times n}(\mathbb{Z}_p)$, or else $C(A_p)$ is dense in $\mathbb{Z}_p^{m_p}$ (see Definition \ref{DEF:column-span}); and
\end{enumerate}
Then let us refer to the matrices $(A_p)_{p\in P}$ as a \emph{factored form}.
If, for some rank-$n$ torsion-free abelian group $G$, it is true that $G\cong \mathcal{G}((A_p)_{p\in P})$, then let us refer to the matrices $(A_p)_{p\in P}$ as a \emph{factored form of $G$}.
\end{defn}
\begin{remark}
The matrices in the factored form of $G$ are related to Malcev's matrices \cite{M:matrices} (Malcev's matrices were not originally expressed as matrices with entries in $\mathbb{Z}_p$, but in \cite{K:summary} Kostromina demonstrates that they can be equivalently described that way).
Malcev's matrices $(M_p)_{p\in P}$ are square matrices, and they satisfy the condition that the first $n-n_p$ columns of $A_pM_p^t$ are zero, where $A_p$ comes from a factored form derived from an appropriate maximal independent subset of $G$.
The factored form is also related to the matrices of Kurosch \cite{K:matrices}.
Following the description in \cite[Chapter XIII, Section 93]{F:book}, the matrices of Kurosch come from the tensor product $G\otimes \mathbb{Z}_p$.
This decomposes (not necessarily uniquely) as a direct sum of a reduced part and its non-reduced (i.e. divisible) subgroup.
The divisible part is a $\mathbb{Q}_p$-vector space and the reduced part is a free $\mathbb{Z}_p$-module.
Let $\mathcal{B}$ be a maximal independent subset of $G$.
The coefficients in the $i$th row of the $p$th Kurosch matrix express the $i$th element of $\mathcal{B}$ as a combination of basis elements for the divisible and reduced parts of $G\otimes \mathbb{Z}_p$.
One main result of the present work is that the information from the divisible subgroup of $G\otimes \mathbb{Z}_p$ is extraneous; that is, the last $n_p$ columns of the Kurosch matrices are enough to determine $G$ completely.
\end{remark}
It is possible to characterize the homomorphisms between factored forms.
\begin{prop}\label{PROP:factored-homomorphisms}
Let $(A_p)_{p\in P}$ and $(B_p)_{p\in P}$ be two factored forms, with each matrix $A_p\in M_{m_p\times m}(\mathbb{Z}_p)$ and each matrix $B_p\in M_{n_p\times n}(\mathbb{Z}_p)$.
Then a matrix $V\in M_{n\times m}(\mathbb{Q})$ represents a group homomorphism from $\mathcal{G}((A_p)_{p\in P})$ to $\mathcal{G}((B_p)_{p\in P})$ if and only if, for all $p\in P$, every row of $B_pV$ lies in the $\mathbb{Z}_p$-module spanned by the rows of $A_p$.
\end{prop}
\begin{proof}
The ``only if'' part of the statement is clear, because left multiplication by any row $R$ of $B_p$ is a $p$-adic functional on $\mathcal{G}((B_p)_{p\in P})$, so the composition $g \mapsto Vg \mapsto RVg$ is a $p$-adic functional on $\mathcal{G}((A_p)_{p\in P})$, and hence must correspond to a $\mathbb{Z}_p$ combination of the rows of $A_p$.
For the ``if''' part of the statement, suppose that, for all primes $p$, every row of $B_pV$ lies in the $\mathbb{Z}_p$-module spanned by the rows of $A_p$.
Pick $g\in \mathcal{G}((A_p)_{p\in P})$.
To show that left multiplication by $V$ is a homomorphism, it is enough to show that, if $g\in p^k\mathcal{G}((A_p)_{p\in P})$ for some prime $p$ and $k\geq 0$, then $Vg \in p^k\mathcal{G}((B_p)_{p\in P})$.
So suppose $g\in p^k\mathcal{G}((A_p)_{p\in P})$.
By Definition \ref{DEF:group-determined}, this means that $A_pg\in p^k\mathbb{Z}_p^{m_p}$.
Thus if $R'$ is a row of $A_p$, then $R'g \in p^k\mathbb{Z}_p$.
As any row $R$ of $B_pV$ is in the $\mathbb{Z}_p$-span of the rows of $A_p$, this means that $Rg\in p^k\mathbb{Z}_p$ as well, so $B_pVg\in p^k\mathbb{Z}_p^{n_p}$, and by Definition \ref{DEF:group-determined} $Vg\in p^k\mathcal{G}((B_p)_{p\in P})$.
\end{proof}
We are now in a position to state Theorem \ref{THM:classification-3}, the main classification result.
It says that every finite-rank torsion-free abelian group has a factored form, and gives a characterization of isomorphisms between factored forms.
The first statement is a consequence of Proposition \ref{PROP:classification-2} and the discussion following it, while the second statement is a consequence of Proposition \ref{PROP:factored-homomorphisms}.
\begin{thm}\label{THM:classification-3}
Any finite-rank torsion-free abelian group $G$ has a factored form.
Moreover, if $(m_p,A_p)_{p\in P}$ and $(n_p,B_p)_{p\in P}$ are two factored forms, then $\mathcal{G}((A_p)_{p\in P})$ is isomorphic to $\mathcal{G}((B_p)_{p\in P})$ if and only if the following conditions hold.
\begin{enumerate}
\item $m_p = n_p$ for all $p\in P$;
\item there exists $V\in GL(n,\mathbb{Q})$ such that, for all $p\in P$, $B_p = U_pAV$ for some $U_p\in GL(n_p,\mathbb{Z}_p)$.
\end{enumerate}
\end{thm}
The matrix $V$ in Theorem \ref{THM:classification-3} should be thought of as a change of basis matrix that represents a new maximal independent subset in terms of the old one.
The matrices $U_p$ should be thought of as change of basis matrices that represent a new basis of the $p$-adic dual group in terms of the old basis.
\begin{remark}\label{REM:isomorphism-check}
In order to check that two factored forms represent isomorphic groups, it is necessary to check infinitely many matrix equations.
Specifically, condition (2) in the definition of a factored form (Definition \ref{DEF:factored}) implies that, if $B_p$ is not the zero row, then $B_pB_p^t$ is invertible over $\mathbb{Q}_p$.
Thus the equation in Theorem \ref{THM:classification-3} can be rearranged to obtain
\begin{align}\label{EQ:system}
A_pVB_p^t(B_pB_p^t)^{-1} & = U_p^{-1}.
\end{align}
So two factored forms represent isomorphic groups if and only if there exists a matrix $V\in GL(n,\mathbb{Q})$ such that, for each prime $p$, the matrix on the left hand side of Equation \ref{EQ:system} is an element of $GL(n_p,\mathbb{Z}_p)$.
\end{remark}
\bibliographystyle{abbrv}
| f7d4032a0da21657f931d5747c87abe2b41d619a | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}\label{sec1}
Let $\mathbb{Z}$ be a finite dimensional vector space, and $\mathbb{R}^{m\times n}$ be
the vector space of all $m\!\times\!n$ real matrices equipped with the trace inner product
$\langle \cdot,\cdot\rangle$ and its induced norm $\|\cdot\|_F$. Denote by $\|X\|_*$
the nuclear norm of a matrix $X\in\mathbb{R}^{m\times n}$, i.e., the sum of all singular
values of $X$, and by $\|X\|$ the spectral norm of $X$. Consider the following optimization problem
\begin{equation}\label{prob}
\min_{z\in\mathbb{Z}}\Big\{f(z)\!:\ g(z)\in K,\ (G(z),H(z))\in{\rm gph}\,\partial\|\cdot\|_*\Big\},
\end{equation}
where $f\!:\mathbb{Z}\to\mathbb{R}$ is a locally Lipschitz function,
$g\!:\mathbb{Z}\to\mathbb{R}^l\times\mathbb{R}^{m\times n}$ and
$G,H\!:\mathbb{Z}\to\mathbb{R}^{m\times n}$ are continuously differentiable mappings,
$K$ is a simple closed convex set of $\mathbb{R}^l\times\mathbb{R}^{m\times n}$,
and ${\rm gph}\,\partial\|\cdot\|_*$ denotes the graph of the subdifferential mapping
of the nuclear norm.
\medskip
By the proof of Lemma \ref{graph-subdiff} below, we know that $(G(z),H(z))\in{\rm gph}\,\partial\|\cdot\|_*$
if and only if $H(z)\!\in\mathop{\arg\max}_{Y\in\mathbb{B}}\langle G(z),Y\rangle$,
where $\mathbb{B}\!:=\!\big\{Z\in\mathbb{R}^{m\times n}\ |\ \|Z\|\le 1\big\}$ in this paper.
This shows that the constraint $(G(z),H(z))\!\in{\rm gph}\,\partial\|\cdot\|_*$
represents a kind of optimality conditions. Therefore, problem \eqref{prob}
is a mathematical program with a matrix equilibrium constraint
$(G(z),H(z))\!\in{\rm gph}\,\partial\|\cdot\|_*$, which extends the optimization
problems with polyhedral variational inequality constraints \cite{Ye99,Ye00},
second-order cone complementarity constraints \cite{Ye15,ZZW11},
or positive semidefinite (PSD) complementarity constraints \cite{DingYS14,WZZhang14}
to those with general matrix equilibrium constraints.
\medskip
Our interest in \eqref{prob} comes from the fact that it covers an equivalent
reformulation of low-rank optimization problems. Indeed, for the following rank minimization problem
\begin{equation}\label{rank-min}
\min_{X\in\mathbb{R}^{m\times n}}\!\Big\{{\rm rank}(X)\!:\ \|\mathcal{A}(X)-b\|\le\delta,\,X\in\Omega\Big\},
\end{equation}
from \cite[Section 3.1]{Bi14} and Lemma \ref{graph-subdiff} in Section \ref{sec2}
we know that it can be reformulated as
\begin{equation}\label{equiv-rank}
\min_{X,Y\in\mathbb{R}^{m\times n}}\!\Big\{\|Y\|_*\!:\ \|\mathcal{A}(X)-b\|\le\delta,\,
X\in\Omega,\,(X,Y)\in{\rm gph}\,\partial\|\cdot\|_*\Big\},
\end{equation}
where $\mathcal{A}\!: \mathbb{R}^{m\times n}\to\mathbb{R}^{N}$ is a sampling operator,
$b\in\mathbb{R}^{N}$ is a noisy observation vector, $\delta>0$ is a constant related to
the noise level, and $\Omega\subseteq\mathbb{R}^{m\times n}$ is a closed convex set.
Clearly, problem \eqref{equiv-rank} is a special case of \eqref{prob}
with $K=\{x\in\mathbb{R}^N\!:\|x\|\le\delta\}\times\Omega$, and
$g(z)=(\mathcal{A}(X)-b; X)$ and $(G(z),H(z))=(X,Y)$ for
$z=(X,Y)\in\mathbb{R}^{m\times n}\times\mathbb{R}^{m\times n}$.
\medskip
As it is well known, low-rank optimization problems have wide applications in
many fields such as statistics \cite{Negahban11}, system identification
and control \cite{Fazel02,FPST13}, signal and image processing \cite{ChenChi14},
machine learning \cite{Srebro04}, finance \cite{Pietersz04}, quantum tomography \cite{Gross11},
and so on. This motivates us to develop the optimality conditions and stability
results for problem \eqref{prob}, especially the equivalent reformulation
\eqref{equiv-rank} of the rank minimization problem \eqref{rank-min}. To achieve this goal,
an essential step is to provide the characterization for the regular and limiting normal cones
to the graph of the subdifferential mapping of the nuclear norm. In this work,
we shall resolve this critical problem and as a byproduct establish the (regular) coderivative of the projection operator onto
the unit spectral norm ball.
\medskip
Throughout this paper, we stipulate $m\le n$. Let $\mathbb{S}^m$ be the space of all $m\times m$
real symmetric matrices and $\mathbb{S}_{+}^m$ be the cone of all PSD matrices from $\mathbb{S}^m$.
Let $\mathbb{O}^{n_1\times n_2}$ be the set of all $n_1\times n_2$ real matrices with orthonormal columns,
and $\mathbb{O}^{n_1}$ be the set of all $n_1\times n_1$ real orthogonal matrices.
For $Z\in\!\mathbb{R}^{m\times n}$, $Z_{\alpha\beta}$ denotes the submatrix consisting of
those $Z_{ij}$ with $(i,j)\in\alpha\times\beta$. Let $e$ and $E$ be the vector and
the matrix of all ones whose dimensions are known from the context,
and for a vector $z$, ${\rm Diag}(z)$ denotes a diagonal matrix
which may be square or rectangular. For a given set $S$, $\delta_S(\cdot)$ means
the indicator function over $S$; $\mathcal{T}_S(x)$ and $\mathcal{T}_S^{i}(x)$ denote
the tangent cone and the inner tangent cone to $S$ at $x$, respectively; and $\mathcal{N}^{\pi}_S(x)$,
$\widehat{\mathcal{N}}_S(x)$ and $\mathcal{N}_S(x)$ denote the proximal normal
cone, the regular normal cone and the limiting normal cone to $S$ at $x$, respectively
(see \cite{RW98,Mordu06}).
\section{Preliminaries}\label{sec2}
This section includes three technical lemmas used for the subsequent analysis.
The first one gives some characterizations for the graph of
the subdifferential mapping $\partial\|\cdot\|_*$.
\begin{lemma}\label{graph-subdiff}
The graph of the subdifferential mapping $\partial\|\cdot\|_*$ has the following forms
\begin{align*}
{\rm gph}\,\partial\|\cdot\|_*\!={\rm gph}\,\mathcal{N}_{\mathbb{B}}^{-1}
&\!=\!\big\{(X,Y)\in\mathbb{R}^{m\times n}\times \mathbb{R}^{m\times n}\ |\
\|X\|_*-\langle X,Y\rangle=0,\,\|Y\|\le 1\big\}\\
&\!=\big\{(X,Y)\in\mathbb{R}^{m\times n}\times \mathbb{R}^{m\times n}\ |\
\Pi_{\mathbb{B}}(X+Y)=Y\big\}.
\end{align*}
\end{lemma}
\begin{proof}
Notice that $(X,Y)\in{\rm gph}\,\partial\|\cdot\|_*$ if and only if
$Y\in\partial\|X\|_*=\partial\delta_{\mathbb{B}}^*(X)$, where $\delta_{\mathbb{B}}^*$
is the conjugate function of $\delta_{\mathbb{B}}$. By \cite[Theorem 23.5]{Roc70},
$\partial\delta_{\mathbb{B}}^*(X)=(\partial\delta_{\mathbb{B}})^{-1}(X)=\mathcal{N}_{\mathbb{B}}^{-1}(X)$.
This shows that $(X,Y)\in{\rm gph}\,\partial\|\cdot\|_*$ iff
$(X,Y)\in{\rm gph}\,\mathcal{N}_{\mathbb{B}}^{-1}$, and the first equality follows.
Since $(X,Y)\in{\rm gph}\,\mathcal{N}_{\mathbb{B}}^{-1}$ if and only if
$X\in\mathcal{N}_{\mathbb{B}}(Y)$, which is equivalent to saying that
$Y\in\mathop{\arg\max}_{Z\in\mathbb{B}}\langle X,Z\rangle$ or equivalently
$\langle X,Y\rangle=\|X\|_*$ and $\|Y\|\le 1$. The second equality follows.
For the third equality, by \cite[Proposition 6.17]{RW98} $\Pi_{\mathbb{B}}(X+Y)=Y$
iff $X\in\mathcal{N}_{\mathbb{B}}(Y)$, which means that
${\rm gph}\,\partial\|\cdot\|_*=\big\{(X,Y)\in\mathbb{R}^{m\times n}\times \mathbb{R}^{m\times n}
\ |\ \Pi_{\mathbb{B}}(X+Y)=Y\big\}$.
\end{proof}
Lemma \ref{graph-subdiff} shows that the graph of $\partial\|\cdot\|_*$ has an intimate
link with $\Pi_{\mathbb{B}}(\cdot)$. Motivated by this, we here establish an important
property for the projection operator $\Pi_{\mathbb{B}}(\cdot)$.
\begin{lemma}\label{Calmly-Bdiff}
The projection map $\Pi_{\mathbb{B}}$ is calmly B-differentiable
at any given $X\in\mathbb{R}^{m\times n}$, i.e.,
\(
\Pi_{\mathbb{B}}(X+H)-\Pi_{\mathbb{B}}(X)-\Pi_{\mathbb{B}}'(X;H)=O(\|H\|^2)
\)
for any $\mathbb{R}^{m\times n}\ni H \to 0$.
\end{lemma}
\begin{proof}
Define $\psi(x)\!:={\rm mid}(-e,x,e)$ for $x\in\mathbb{R}^m$. The mapping
$\psi$ is Lipschitz continuous everywhere. For any given $x\in\mathbb{R}^m$
and any $h\in\mathbb{R}^m$, a simple calculation yields that
\begin{equation}\label{gdir-derivative}
\big[\psi'(x;h)\big]_i
=\left\{\!\begin{array}{cl}
0 & {\rm if}\ |x_i|>1,\\
{\rm sign}(x_i)\min(0,{\rm sign}(x_i)h_i) & {\rm if}\ |x_i|=1,\\
h_i & {\rm if}\ |x_i|<1
\end{array}\right.\ \ {\rm for}\ i=1,2,\ldots,m.
\end{equation}
It is easy to check that $\psi(x+h)-\psi(x)-\psi'(x;h)=0$. Hence,
$\psi$ is calmly B-differentiable in $\mathbb{R}^m$. Since $\psi$ is symmetric,
i.e., $\psi(x)=Q^{\mathbb{T}}\psi(Qx)$ for any signed permutation matrix
$Q\in\mathbb{R}^{m\times m}$ and $x\in\mathbb{R}^m$, the desired result
follows by invoking \cite[Theorem 5.5]{DingSST14}.
\end{proof}
Next we give the expression of the directional derivative of $\Pi_{\mathbb{B}}$
by \cite[Theorem 3.4]{DingSST14}.
\begin{lemma}\label{dir-derivative}
Let $\overline{Z}\in\mathbb{R}^{m\times n}$ have the SVD of the form
$\overline{Z}=\overline{U}\big[{\rm Diag}(\sigma(\overline{Z}))\ \ 0\big]\overline{V}^{\mathbb{T}}$, where
$\overline{U}\in\mathbb{O}^m$ and $\overline{V}=[\overline{V}_1\ \ \overline{V}_2]\in\mathbb{O}^{n}$
with $\overline{V}_1\in\mathbb{R}^{n\times m}$. Define the index sets
\begin{subnumcases}{}
\label{ab-index}
\alpha:=\big\{i\in\{1,\ldots,m\}\ |\ \sigma_i(\overline{Z})>1\big\},\ \beta:=\big\{i\in\{1,\ldots,m\}\ |\ \sigma_i(\overline{Z})=1\big\},\\
\label{gc-index}
\gamma:=\big\{i\in\{1,\ldots,m\}\ |\ \sigma_i(\overline{Z})<1\big\},\ c:=\big\{m\!+\!1,m\!+\!2,\ldots,n\big\}.
\end{subnumcases}
Let $\Omega_1,\Omega_2\in\mathbb{S}^{m}$ and $\Omega_3\in\mathbb{R}^{m\times (n-m)}$ be
the matrices associated with $\sigma(\overline{Z})$, defined by
\begin{align}\label{Omega1}
\big(\Omega_1\big)_{ij}
&:=\left\{\begin{array}{cl}
\frac{\min(1,\sigma_i(\overline{Z}))-\min(1,\sigma_j(\overline{Z}))}
{\sigma_i(\overline{Z})-\sigma_j(\overline{Z})}& {\rm if}\ \sigma_i(\overline{Z})\ne\sigma_j(\overline{Z}),\\
0 & {\rm otherwise}\\
\end{array}\right.\ \ i,j\in\{1,2,\ldots,m\},\\
\label{Omega2}
\big(\Omega_2\big)_{ij}
&:=\left\{\begin{array}{cl}
\frac{\min(1,\sigma_i(\overline{Z}))+\min(1,\sigma_j(\overline{Z}))}
{\sigma_i(\overline{Z})+\sigma_j(\overline{Z})}& {\rm if}\ \sigma_i(\overline{Z})+\!\sigma_j(\overline{Z})\ne 0,\\
0 & {\rm otherwise}\\
\end{array}\right.\ \ i,j\in\{1,2,\ldots,m\},\\
\label{Omega3}
\big(\Omega_3\big)_{ij}
&:=\left\{\begin{array}{cl}
\frac{\min(1,\sigma_i(\overline{Z}))}{\sigma_i(\overline{Z})}
& {\rm if}\ \sigma_i(\overline{Z})\ne 0,\\
0 & {\rm otherwise}\\
\end{array}\right.\ \ i\in\{1,\ldots,m\},j\in\{1,\ldots,n\!-\!m\}.
\end{align}
Then, for any $H\in\mathbb{R}^{m\times n}$, with $\widetilde{H}_1=\overline{U}^{\mathbb{T}}H\overline{V}_1$
and $\widetilde{H}=[\overline{U}^{\mathbb{T}}H\overline{V}_1\ \ \overline{U}^{\mathbb{T}}H\overline{V}_2]$
it holds that
\begin{align*}
\Pi_{\mathbb{B}}'(\overline{Z};H)
=\!\overline{U}\left[\begin{matrix}
(\Omega_2)_{\alpha\alpha}\circ \big(\mathcal{X}(\widetilde{H}_1)\big)_{\alpha\alpha}
&(\Omega_2)_{\alpha\beta}\!\circ\big(\mathcal{X}(\widetilde{H}_1)\big)_{\alpha\beta}& \widehat{H}_{\alpha\gamma}
&(\Omega_3)_{\alpha c}\!\circ\!\widetilde{H}_{\alpha c}\\
(\Omega_2)_{\beta\alpha }\!\circ\! \big(\mathcal{X}(\widetilde{H}_1)\big)_{\beta\alpha }&\widetilde{H}_{\beta\beta} \!-\!\Pi_{\mathbb{S}_{+}^{|\beta|}}(\mathcal{S}(\widetilde{H}_{\beta\beta}))&
\widetilde{H}_{\beta\gamma}&\widetilde{H}_{\beta c}\\
\widehat{H}_{\gamma\alpha}&\widetilde{H}_{\gamma\beta}&\widetilde{H}_{\gamma\gamma}
&\widetilde{H}_{\gamma c}\\
\end{matrix}\right]\overline{V}^{\mathbb{T}},
\end{align*}
where $\widehat{H}_{ij}=(\Omega_1)_{ij}\big(\mathcal{S}(\widetilde{H}_1)\big)_{ij}+(\Omega_2)_{ij}\big(\mathcal{X}(\widetilde{H}_1)\big)_{ij}$
for $(i,j)\in\alpha\times\gamma$ or $(i,j)\in\gamma\times\alpha$,
and $\mathcal{S}\!:\mathbb{R}^{m\times m}\to\mathbb{S}^m$ and
$\mathcal{X}\!:\mathbb{R}^{m\times m}\to\mathbb{R}^{m\times m}$ are two linear mappings
defined by
\begin{equation}\label{ST-oper}
\mathcal{S}(Z):=(Z\!+\!Z^{\mathbb{T}})/2\ \ {\rm and}\ \ \mathcal{X}(Z):=(Z\!-\!Z^{\mathbb{T}})/2
\quad\ \forall Z\in\mathbb{R}^{m\times m}.
\end{equation}
\end{lemma}
\section{Regular and limiting normal cones to ${\rm gph}\,\partial\|\cdot\|_*$}
In this section we shall derive the exact formula for the regular and limiting
normal cones to ${\rm gph}\,\partial\|\cdot\|_*$. First, we focus on the formula
of the regular normal cone to ${\rm gph}\,\partial\|\cdot\|_*$.
\subsection{Regular normal cone}
For the set ${\rm gph}\,\partial\|\cdot\|_*$, we shall verify that its regular normal cone
coincides with its proximal normal cone just as \cite{WZZhang14} did
for ${\rm gph}\,\mathcal{N}_{\mathbb{S}_{+}^m}(\cdot)$. This requires the following
two lemmas. Among others, Lemma \ref{tangent-graphf} characterizes the tangent cone
to ${\rm gph}\,\partial\|\cdot\|_*$, while Lemma \ref{prox-normalcone} provides
the characterization for the proximal normal cone to ${\rm gph}\,\partial\|\cdot\|_*$.
\begin{lemma}\label{tangent-graphf}
For any given $(X,Y)\!\in{\rm gph}\,\partial\|\cdot\|_*$, the following equalities hold:
\[
\mathcal{T}_{{\rm gph}\,\partial\|\cdot\|_*}^{i}(X,Y)\!=\!\mathcal{T}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)
=\!\big\{(G,H)\in\mathbb{R}^{m\times n}\times\mathbb{R}^{m\times n}\ |\
\Pi_{\mathbb{B}}'(X\!+\!Y,G\!+\!H)=\!H\big\}.
\]
\end{lemma}
\begin{proof}
Let $(G,H)$ be an arbitrary point from $\mathcal{T}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)$.
By the definition of tangent cone, there exist $t_k\downarrow 0$ and $(G^k,H^k)\to(G,H)$ such that
$(X,Y)+t_k(G^k,H^k)\in {\rm gph}\,\partial\|\cdot\|_*$ for each $k$. By Lemma \ref{graph-subdiff},
$\Pi_{\mathbb{B}}(X+Y+t_k(G^k+H^k))=Y+t_kH^k$. Notice that $Y=\Pi_{\mathbb{B}}(X+Y)$
by virtue of $(X,Y)\in{\rm gph}\,\partial\|\cdot\|_*$ and Lemma \ref{graph-subdiff}.
Then we have
\[
\Pi_{\mathbb{B}}'(X\!+\!Y,G\!+\!H)=
\lim_{k\to\infty}\frac{1}{t_k}\big(\Pi_{\mathbb{B}}(X\!+\!Y+t_k(G^k\!+\!H^k))-\Pi_{\mathbb{B}}(X\!+\!Y)\big)
=H.
\]
This, by the arbitrariness of $(G,H)$ in $\mathcal{T}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)$,
implies the following inclusion:
\[
\mathcal{T}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)\subseteq
\big\{(G,H)\in\mathbb{R}^{m\times n}\times\mathbb{R}^{m\times n}\ |\
\Pi_{\mathbb{B}}'(X\!+\!Y,G\!+\!H)=H\big\}.
\]
Since $\mathcal{T}_{{\rm gph}\,\partial\|\cdot\|_*}^{i}(X,Y)\!\subseteq\!\mathcal{T}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)$,
the rest only needs to establish the inclusion
\begin{equation}\label{converse-inclusion}
\big\{(G,H)\in\mathbb{R}^{m\times n}\times\mathbb{R}^{m\times n}\ |\ \Pi_{\mathbb{B}}'(X+Y,G+H)=H\big\}
\subseteq\mathcal{T}_{{\rm gph}\,\partial\|\cdot\|_*}^{i}(X,Y).
\end{equation}
To this end, let $(G,H)\in\mathbb{R}^{m\times n}\times\mathbb{R}^{m\times n}$
with $\Pi_{\mathbb{B}}'(X\!+\!Y,G\!+\!H)=H$. For any $t>0$, write $Z_t:=X\!+\!Y+t(G\!+\!H)$.
By the definition of $\Pi_{\mathbb{B}}(\cdot)$, we have
\(
Z_t-\Pi_{\mathbb{B}}(Z_t)\in\mathcal{N}_{\mathbb{B}}(\Pi_{\mathbb{B}}(Z_t)).
\)
In addition, from $\Pi_{\mathbb{B}}'(X+Y,G+H)=H$ and the definition of the directional derivative,
\[
\Pi_{\mathbb{B}}(Z_t)=\Pi_{\mathbb{B}}(X+Y)+tH+o(t)=Y+tH+o(t).
\]
This shows that $X+tG+o(t)\in\mathcal{N}_{\mathbb{B}}(\Pi_{\mathbb{B}}(Z_t))$,
and then $\big(X+tG+o(t),\Pi_{\mathbb{B}}(Z_t)\big)\in{\rm gph}\,\partial\|\cdot\|_*$ by Lemma \ref{graph-subdiff}.
Along with the last equality,
\(
{\rm dist}\big((X+tG,Y+tH),{\rm gph}\,\partial\|\cdot\|_*\big)
=o(t).
\)
This means that $(G,H)\in\mathcal{T}_{{\rm gph}\,\partial\|\cdot\|_*}^{i}(X,Y)$.
So, the inclusion in \eqref{converse-inclusion} follows.
\end{proof}
\begin{lemma}\label{prox-normalcone}
For any given $(X,Y)\!\in\!{\rm gph}\,\partial\|\cdot\|_*$, we have
$(X^*,Y^*)\!\in\!\mathcal{N}_{{\rm gph}\,\partial\|\cdot\|_*}^{\pi}\!(X,Y)$ iff
\begin{equation}\label{aim-ineq}
\langle X^*, W-\Pi'_{\mathbb{B}}(X+Y;W)\rangle+\langle Y^*,\Pi'_{\mathbb{B}}(X+Y;W)\rangle\le 0
\quad\ \forall W\in\mathbb{R}^{m\times n}.
\end{equation}
\end{lemma}
\begin{proof}
Let $(X^*,Y^*)\in\mathcal{N}_{{\rm gph}\,\partial\|\cdot\|_*}^{\pi}(X,Y)$.
We prove that inequality \eqref{aim-ineq} holds. For this purpose,
let $W$ be an arbitrary point from $\mathbb{R}^{m\times n}$. For any $t>0$, we write
\begin{equation}\label{temp-ineq0}
Y'_t:=\Pi_{\mathbb{B}}(X+Y+tW)\ \ {\rm and}\ \ X'_t:=X+Y+tW-\Pi_{\mathbb{B}}(X+Y+tW).
\end{equation}
Clearly, $X_t'\in\mathcal{N}_{\mathbb{B}}(Y_t')$. By Lemma \ref{graph-subdiff},
$Y'_t\in\partial\|X'_t\|_*$.
Since $(X^*,Y^*)\in\mathcal{N}_{{\rm gph}\,\partial\|\cdot\|_*}^{\pi}(X,Y)$,
by Part E of \cite[Chapter 6]{RW98} there exists $\varepsilon>0$ such that
for any $(X',Y')\in {\rm gph}\,\partial\|\cdot\|_*$,
\begin{equation*}
\langle (X^*, Y^*),(X',Y')-(X,Y)\rangle\leq \varepsilon\|(X',Y')-(X,Y)\|_F^2.
\end{equation*}
Take $(X',Y')=(X'_t,Y_t')$. From this inequality, it follows that
\begin{equation}\label{temp-ineq}
\langle (X^*, Y^*),(X_t'-X,Y_t'-Y)\rangle\leq \varepsilon\|(X_t',Y_t')-(X,Y)\|_F^2.
\end{equation}
Note that $Y=\Pi_{\mathbb{B}}(X+Y)$ since $(X,Y)\in{\rm gph}\,\partial\|\cdot\|_*$.
From \eqref{temp-ineq0} and \eqref{temp-ineq}, we have that
\begin{align*}
&\langle X^*, W-\Pi'_{\mathbb{B}}(X+Y;W)\rangle+\langle Y^*,\Pi'_{\mathbb{B}}(X+Y;W)\rangle \\
&\le \varepsilon\lim_{t\downarrow 0}\frac{1}{t}(\| X'_t-X\|_F^2+\|Y'_t-Y\|_F^2)\nonumber \\
&\leq \varepsilon \lim_{t\downarrow 0}\frac{1}{t}\big(3\|Y-\!\Pi_{\mathbb{B}}(X+Y+tW)\|_F^2+2t^2\|W\|_F^2\big)
\le\varepsilon\lim_{t\downarrow 0}\frac{1}{t}\big(5t^2\|W\|_F^2\big)=0,
\end{align*}
where the last inequality is using $Y=\Pi_{\mathbb{B}}(X+Y)$ and the global Lipschitz continuity
with modulus $1$ of the projection operator $\Pi_{\mathbb{B}}(\cdot)$.
This shows that \eqref{aim-ineq} holds. Conversely, suppose that \eqref{aim-ineq} holds.
We shall prove $(X^*,Y^*)\in\mathcal{N}_{{\rm gph}\,\partial\|\cdot\|_*}^{\pi}(X,Y)$.
By Lemma \ref{Calmly-Bdiff}, there exist $\delta>0$ and a constant $M>0$ such that
for any $Z'$ with $\|Z'-(X+Y)\|_F\leq \delta$,
\begin{align*}
\langle X^*,\Pi_{\mathbb{B}}(Z')-\Pi_{\mathbb{B}}(X+Y)\rangle
\le \langle Y^*,\Pi_{\mathbb{B}}'(X+Y;Z'-X-Y)\rangle+M\|Z'-(X+Y)\|_F^2,\\
\langle Y^*,\Pi_{\mathbb{B}}(Z')-\Pi_{\mathbb{B}}(X+Y)\rangle
\le \langle Y^*,\Pi_{\mathbb{B}}'(X+Y;Z'-X-Y)\rangle+M\|Z'-(X+Y)\|_F^2.
\end{align*}
Thus, for any $(X',Y')\in{\rm gph}\,\partial\|\cdot\|_*$ with $\|(X',Y')-(X,Y)\|_F\le \delta/2$, we have that
\begin{align}\label{equa1-XYStar}
&\langle X^*,\Pi_{\mathbb{B}}(X'\!+\!Y')-\Pi_{\mathbb{B}}(X\!+\!Y)\rangle
-\langle X^*,\Pi_{\mathbb{B}}'(X\!+\!Y;\Delta X\!+\!\Delta Y)\rangle\le M\|\Delta X\!+\!\Delta Y\|_F^2,\\
&\langle Y^*,\Pi_{\mathbb{B}}(X'\!+\!Y')-\Pi_{\mathbb{B}}(X\!+\!Y)\rangle
-\langle Y^*,\Pi_{\mathbb{B}}'(X\!+\!Y;\Delta X\!+\!\Delta Y)\rangle
\le M\|\Delta X \!+\!\Delta Y\|_F^2,
\label{equa2-XYStar}
\end{align}
where $\Delta X\!=\!X'-X$ and $\Delta Y=Y'-Y$. Along with $\Pi_{\mathbb{B}}(X'+Y')=Y'$
and $\Pi_{\mathbb{B}}(X+Y)=Y$,
\begin{align*}
\langle (X^*, Y^*),(X',Y')-(X,Y)\rangle
&=\langle X^*,\Delta X\!+\!\Delta Y-\Pi_{\mathbb{B}}(X'+Y')+\Pi_{\mathbb{B}}(X+Y)\rangle\\
&\quad +\langle Y^*,\Pi_{\mathbb{B}}(X'+Y')-\Pi_{\mathbb{B}}(X+Y)\rangle\\
&\le\langle X^*, \Delta X\!+\!\Delta Y-\Pi'_{\mathbb{B}}(X+Y;\Delta X\!+\!\Delta Y)\rangle\nonumber \\
&\qquad+ \langle Y^*, \Pi'_{\mathbb{B}}(X+Y; \Delta X\!+\!\Delta Y)\rangle + 2M\|\Delta X\!+\!\Delta Y\|_F^2\nonumber\\
&\leq 4M\|\Delta X\!+\!\Delta Y\|_F^2\nonumber,
\end{align*}
where the first inequality is using \eqref{equa1-XYStar} and \eqref{equa2-XYStar},
and the last one is by virtue of \eqref{aim-ineq} with $W=\Delta X\!+\!\Delta Y$.
Take $\varepsilon=\max\{4M, 2\|(X^*,Y^*)\|_F/\delta\}$. For any $(X',Y')\in {\rm gph}\,\partial\|\cdot\|_*$,
it holds that
\(
\langle (X^*, Y^*),(X',Y')-(X,Y)\rangle\leq \varepsilon\|(X',Y')-(X,Y)\|_F^2.
\)
This, by Part E of \cite[Chapter 6]{RW98}, shows that
$(X^*,Y^*)\in\mathcal{N}_{{\rm gph}\,\partial\|\cdot\|_*}^{\pi}(X,Y)$.
Thus, we finish the proof.
\end{proof}
Now we are in a position to establish the coincidence between the regular normal cone
to ${\rm gph}\,\partial\|\cdot\|_*$ and the proximal normal cone to ${\rm gph}\,\partial\|\cdot\|_*$.
\begin{proposition}\label{prop-relation}
For any given $(X,Y)\!\in{\rm gph}\,\partial\|\cdot\|_*$,
\(
\mathcal{\widehat{N}}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)\!=\!\mathcal{N}^{\pi}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y).
\)
\end{proposition}
\vspace{-0.3cm}
\begin{proof}
Take an arbitrary point $(X^*,Y^*)\in\widehat{\mathcal{N}}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)$.
By \cite[Proposition 6.5]{RW98}, $\langle (X^*,Y^*),(G,H)\rangle\le 0$
for any $(G,H)\in \mathcal{T}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)$.
From Lemma \ref{tangent-graphf}, clearly,
\(
\big(W-\Pi_{\mathbb{B}}'(X+Y,W),\Pi_{\mathbb{B}}'(X+Y,W)\big)\in \mathcal{T}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)
\)
for any $W\in\mathbb{R}^{m\times n}$, and then
\(
\langle X^*, W-\Pi_{\mathbb{B}}'(X+Y,W)\rangle+\langle Y^*, \Pi_{\mathbb{B}}'(X+Y,W)\rangle\leq 0.
\)
This, by Lemma \ref{prox-normalcone}, implies that $(X^*,Y^*)\in \mathcal{N}^{\pi}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)$,
and then $\widehat{\mathcal{N}}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)\subseteq\mathcal{N}^{\pi}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)$
follows. Next take an arbitrary point $(X^*,Y^*)\in\mathcal{N}^{\pi}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)$.
For any $(G,H)\in\mathcal{T}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)$,
by Lemma \ref{tangent-graphf} it follows that $\Pi_{\mathbb{B}}'(X+Y,G+H)=H$.
Using Lemma \ref{prox-normalcone} with $W=G+H$ and noting that $\Pi_{\mathbb{B}}'(X+Y,W)=H$
yields that
\(
\langle (X^*,Y^*),(G,H)\rangle\le 0,
\)
i.e., $(X^*,Y^*)\in \mathcal{\widehat{N}}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)$.
So, $\mathcal{N}^{\pi}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)\subseteq
\widehat{\mathcal{N}}_{{\rm gph}\,\partial\|\cdot\|_*}(X,Y)$.
The proof is completed.
\end{proof}
\medskip
Proposition \ref{prop-relation} shows that, to characterize the regular normal cone to
${\rm gph}\,\partial\|\cdot\|_*$, one only needs to characterize its proximal normal cone.
Next we shall employ Lemma \ref{prox-normalcone} and Lemma \ref{dir-derivative} to
derive the expression of the proximal normal cone to ${\rm gph}\,\partial\|\cdot\|_*$.
\begin{theorem}\label{proxnormal-theorem}
For any given $(X,Y)\in{\rm gph}\,\partial\|\cdot\|_*$, let $\overline{Z}=X+Y$ have the SVD as
given in Lemma \ref{dir-derivative}. With $\Omega_1$ and $\Omega_2$ in \eqref{Omega1}-\eqref{Omega2},
we define the following matrices
\begin{subequations}
\begin{equation*}\label{Theta}
\Theta_1:=\left[\begin{matrix}
0_{\alpha\alpha} & 0_{\alpha \beta} &(\Omega_1)_{\alpha\gamma}\\
0_{\beta\alpha}& 0_{\beta\beta}&E_{\beta\gamma}\\
(\Omega_1)_{\gamma\alpha}& E_{\gamma\beta}&E_{\gamma\gamma}\\
\end{matrix}\right],\
\Theta_2:=\left[\begin{matrix}
E_{\alpha\alpha} & E_{\alpha \beta} &E_{\alpha\gamma}\!-\!(\Omega_1)_{\alpha\gamma}\\
E_{\beta\alpha}& 0_{\beta\beta}&0_{\beta\gamma}\\
E_{\gamma\alpha}\!-\!(\Omega_1)_{\gamma\alpha}& 0_{\gamma\beta}&0_{\gamma\gamma}\\
\end{matrix}\right],\\
\end{equation*}
\begin{equation*}\label{Sigma}
\Sigma_1\!:=\!\left[\begin{matrix}
(\Omega_2)_{\alpha\alpha} & (\Omega_2)_{\alpha\beta} & (\Omega_2)_{\alpha\gamma}\\
(\Omega_2)_{\beta\alpha}& 0_{\beta\beta}&E_{\beta\gamma}\\
(\Omega_2)_{\gamma\alpha}& E_{\gamma \beta}&E_{\gamma\gamma}\\
\end{matrix}\right],
\Sigma_2\!:=\!\left[\begin{matrix}
E_{\alpha\alpha}\!-\!(\Omega_2)_{\alpha\alpha} & E_{\alpha \beta}\!-\!(\Omega_2)_{\alpha\beta}
&E_{\alpha\gamma}\!-\!(\Omega_2)_{\alpha\gamma}\\
E_{\beta\alpha}\!-\!(\Omega_2)_{\beta\alpha }& 0_{\beta \beta}&0_{\beta\gamma}\\
E_{\gamma\alpha}\!-\!(\Omega_2)_{\gamma\alpha}& 0_{\gamma\beta}&0_{\gamma\gamma}\\
\end{matrix}\!\right].
\end{equation*}
\end{subequations}
Then $(X^*,Y^*)\in \mathcal{N}_{{\rm gph}\,\partial\|\cdot\|_*}^{\pi}(X,Y)$
if and only if $(X^*,Y^*)$ satisfies the following conditions
\vspace{-0.3cm}
\begin{subequations}
\begin{align}\label{result-equa1}
\Theta_{1}\circ \mathcal{S}(\widetilde{Y}_1^*)+ \Theta_{2}\circ \mathcal{S}(\widetilde{X}_1^*)
+\Sigma_{1}\circ \mathcal{X}(\widetilde{Y}_1^*)+\Sigma_{2}\circ \mathcal{X}(\widetilde{X}_1^*)=0,\\
\label{result-equa2}
\widetilde{X}_{\alpha c}^*\circ (E_{\alpha c}-(\Omega_3)_{\alpha c})+\widetilde{Y}_{\alpha c}^*\circ (\Omega_3)_{\alpha c}=0,\qquad\qquad\\
\widetilde{Y}_{\beta c}^*=0,\, \widetilde{Y}_{\gamma c}^*=0,\,\widetilde{X}_{\beta\beta}^*\preceq 0,\,
\widetilde{Y}_{\beta\beta}^*\succeq 0,\qquad\qquad\quad
\label{result-equa3}
\end{align}
\end{subequations}
where $\widetilde{X}_1^*=\overline{U}^{\mathbb{T}}X^*\overline{V}_1,$
$\widetilde{Y}_1^*=\overline{U}^{\mathbb{T}}Y^*\overline{V}_1$,
$\widetilde{X}^*=\overline{U}^{\mathbb{T}}X^*\overline{V}$ and
$\widetilde{Y}^*=\overline{U}^{\mathbb{T}}Y^*\overline{V}$.
\end{theorem}
\begin{proof}
By Lemma \ref{prox-normalcone} and Lemma \ref{dir-derivative},
$(X^*,Y^*)\in\!\mathcal{N}_{{\rm gph}\,\partial\|\cdot\|_*}^{\pi}(X,Y)$
iff for any $H\in\mathbb{R}^{m\times n}$,
\begin{align}\label{main-ineq1}
&\left\langle \widetilde{X}^*,\widetilde{H}-\left[\begin{matrix}
(\Omega_2)_{\alpha\alpha}\circ \big(\mathcal{X}(\widetilde{H}_1)\big)_{\alpha\alpha}
&(\Omega_2)_{\alpha\beta}\!\circ\big(\mathcal{X}(\widetilde{H}_1)\big)_{\alpha\beta}
& \widehat{H}_{\alpha\gamma}&(\Omega_3)_{\alpha c}\!\circ\!\widetilde{H}_{\alpha c}\\
(\Omega_2)_{\beta\alpha }\!\circ\! \big(\mathcal{X}(\widetilde{H}_1)\big)_{\beta\alpha}&\widetilde{H}_{\beta\beta} \!-\!\Pi_{\mathbb{S}_{+}^{|\beta|}}(\mathcal{S}(\widetilde{H}_{\beta\beta}))&\widetilde{H}_{\beta\gamma}&\widetilde{H}_{\beta c}\\
\widehat{H}_{\gamma\alpha}&\widetilde{H}_{\gamma\beta}&\widetilde{H}_{\gamma\gamma}&\widetilde{H}_{\gamma c}\\
\end{matrix}\right]\right\rangle\nonumber\\
&+\left\langle \widetilde{Y}^*,\left[\begin{matrix}
(\Omega_2)_{\alpha\alpha}\circ \big(\mathcal{X}(\widetilde{H}_1)\big)_{\alpha\alpha}
&(\Omega_2)_{\alpha\beta}\!\circ\big(\mathcal{X}(\widetilde{H}_1)\big)_{\alpha\beta}& \widehat{H}_{\alpha\gamma}
&(\Omega_3)_{\alpha c}\!\circ\!\widetilde{H}_{\alpha c}\\
(\Omega_2)_{\beta\alpha }\!\circ\! \big(\mathcal{X}(\widetilde{H}_1)\big)_{\beta\alpha }&\widetilde{H}_{\beta\beta} \!-\!\Pi_{\mathbb{S}_{+}^{|\beta|}}(\mathcal{S}(\widetilde{H}_{\beta\beta}))&
\widetilde{H}_{\beta\gamma}&\widetilde{H}_{\beta c}\\
\widehat{H}_{\gamma\alpha}&\widetilde{H}_{\gamma \beta}&\widetilde{H}_{\gamma\gamma}&\widetilde{H}_{\gamma c}\\
\end{matrix}\right]\right\rangle\le 0
\end{align}
where $\widetilde{H}_1=\overline{U}^{\mathbb{T}}H\overline{V}_1$ and
$\widetilde{H}=\overline{U}^{\mathbb{T}}H\overline{V}$.
Take $H=\overline{U}_{\!\beta}M_{\beta\gamma}\overline{V}_{\!\gamma}^{\mathbb{T}}$
for any $M_{\beta\gamma}\in\mathbb{R}^{|\beta|\times\gamma}$. By the expressions
of $\widetilde{H}_1$ and $\widetilde{H}$ and equation \eqref{main-ineq1},
it is easy to obtain that $\widetilde{Y}_{\beta\gamma}^*=0$. Using the similar arguments, we can achieve that
$\widetilde{Y}_{\gamma\beta}^*=0,\widetilde{Y}_{\gamma\gamma}^*=0,\widetilde{Y}_{\beta c}^*=0,\widetilde{Y}_{\gamma c}^*=0$ and
\(
\widetilde{X}_{\alpha c}^*\circ (E_{\alpha c}\!-\!(\Omega_3)_{\alpha c})+\widetilde{Y}_{\alpha c}^*\circ (\Omega_3)_{\alpha c}=0.
\)
Taking $H=\overline{U}_{\!\beta}M_{\beta\beta}\overline{V}_{\!\beta}^{\mathbb{T}}$ for
any $M_{\beta\beta}\in\mathbb{S}_{-}^{|\beta|}$, from \eqref{main-ineq1} we have
$\widetilde{Y}_{\beta\beta}^*\succeq 0$; and by taking $H=\overline{U}_{\!\beta}M_{\beta\beta}\overline{V}_{\!\beta}^{\mathbb{T}}$
for any $M_{\beta\beta}\in\mathbb{S}_{+}^{|\beta|}$, we obtain that
\begin{align*}
0\ge \langle \widetilde{X}_{\beta\beta}^*-\widetilde{Y}_{\beta\beta}^*,
\Pi_{\mathbb{S}_{+}^{|\beta|}}(\mathcal{S}(\widetilde{H}_{\beta\beta}))\rangle
+\langle \widetilde{Y}_{\beta\beta}^*,\widetilde{H}_{\beta\beta}\rangle
=\langle \widetilde{X}_{\beta\beta}^*,M_{\beta\beta}\rangle,
\end{align*}
which implies that $\widetilde{X}_{\beta\beta}^*\preceq 0$. In addition,
taking $H=\overline{U}_{\!\alpha}M_{\alpha\alpha}\overline{V}_{\!\alpha}^{\mathbb{T}}$ for
any $M_{\alpha\alpha}\in\mathbb{R}^{|\alpha|\times|\alpha|}$ and observing that
$\langle Z,\mathcal{X}(M_{\alpha\alpha})\rangle=\langle\mathcal{X}(Z),M_{\alpha\alpha}\rangle$ for
any $Z\in\mathbb{R}^{|\alpha|\times|\alpha|}$, from \eqref{main-ineq1} we have
\begin{align*}
0\ge \langle\widetilde{X}_{\alpha\alpha}^*,M_{\alpha\alpha}\rangle
+\langle (\widetilde{Y}_{\alpha\alpha}^*\!-\!\widetilde{X}_{\alpha\alpha}^*)\circ(\Omega_2)_{\alpha\alpha},\mathcal{X}(M_{\alpha\alpha})\rangle\\
=\langle\widetilde{X}_{\alpha\alpha}^*,M_{\alpha\alpha}\rangle
+\langle \mathcal{X}\big(\widetilde{Y}_{\alpha\alpha}^*\!-\!\widetilde{X}_{\alpha\alpha}^*\big)\circ(\Omega_2)_{\alpha\alpha},M_{\alpha\alpha}\rangle,
\end{align*}
which implies that
\(
\widetilde{X}_{\alpha\alpha}^*+\mathcal{X}\big(\widetilde{Y}_{\alpha\alpha}^*\!-\!\widetilde{X}_{\alpha\alpha}^*\big)\circ(\Omega_2)_{\alpha\alpha}=0.
\)
Similarly, taking $H=\overline{U}_{\!\alpha}M_{\alpha\beta}\overline{V}_{\!\beta}^{\mathbb{T}}$ for
any $M_{\alpha\beta}\in\mathbb{R}^{|\alpha|\times|\beta|}$, from equation \eqref{main-ineq1} we obtain that
\begin{align*}
0&\ge \langle \widetilde{X}_{\alpha\beta}^*,M_{\alpha\beta}\rangle
+\langle \big(\mathcal{X}(\widetilde{Y}_1^*\!-\!\widetilde{X}_1^*)\big)_{\alpha\beta}\circ (\Omega_2)_{\alpha\beta},M_{\alpha\beta}\rangle,
\end{align*}
which shows that $\widetilde{X}_{\alpha\beta}^*+\big(\mathcal{X}(\widetilde{Y}_1^*\!-\!\widetilde{X}_1^*)\big)_{\alpha\beta}
\circ (\Omega_2)_{\alpha\beta}=0$. Using the similar way, we have
\begin{subequations}
\begin{align}
\label{equa1-ab}
\widetilde{X}_{\beta\alpha}^*+\big(\mathcal{X}(\widetilde{Y}_1^*\!-\!\widetilde{X}_1^*)\big)_{\beta\alpha}
\circ (\Omega_2)_{\beta\alpha}=0,\qquad\qquad\\
\label{equa1-ag}
\widetilde{X}_{\alpha\gamma}^*+\big(\mathcal{S}(\widetilde{Y}_1^*\!-\!\widetilde{X}_1^*)\big)_{\alpha\gamma}
\circ(\Omega_1)_{\alpha\gamma}+\big(\mathcal{X}(\widetilde{Y}_1^*\!-\!\widetilde{X}_1^*)\big)_{\alpha\gamma}
\circ(\Omega_2)_{\alpha\gamma}=0,\\
\label{equa2-ag}
\widetilde{X}_{\gamma\alpha}^*+\big(\mathcal{S}(\widetilde{Y}_1^*\!-\!\widetilde{X}_1^*)\big)_{\gamma\alpha}
\circ(\Omega_1)_{\gamma\alpha}+\big(\mathcal{X}(\widetilde{Y}_1^*\!-\!\widetilde{X}_1^*)\big)_{\gamma\alpha}
\circ(\Omega_2)_{\gamma\alpha}=0.
\end{align}
\end{subequations}
To sum up, the fact that inequality \eqref{main-ineq1} holds for any $H\in\mathbb{R}^{m\times n}$ implies that
\begin{align}\label{main-equa2}
\widetilde{X}_{\alpha\alpha}^*+\mathcal{X}(\widetilde{Y}_{\alpha\alpha}^*\!-\!\widetilde{X}_{\alpha\alpha}^*)\circ(\Omega_2)_{\alpha\alpha}=0,\ \ \widetilde{X}_{\beta\beta}^*\preceq 0,\ \widetilde{Y}_{\beta\beta}^*\succeq 0,\qquad\nonumber\\
\widetilde{X}_{\alpha\beta}^*+\big(\mathcal{X}(\widetilde{Y}_1^*\!-\!\widetilde{X}_1^*)\big)_{\alpha\beta}\circ (\Omega_2)_{\alpha\beta}=0,\,
\widetilde{X}_{\beta\alpha}^*+\big(\mathcal{X}(\widetilde{Y}_1^*\!-\!\widetilde{X}_1^*)\big)_{\beta\alpha}\circ (\Omega_2)_{\alpha\beta}=0,\nonumber\\
\widetilde{X}_{\alpha\gamma}^*+\big(\mathcal{S}(\widetilde{Y}_1^*\!-\!\widetilde{X}_1^*)\big)_{\alpha\gamma}
\circ(\Omega_1)_{\alpha\gamma}+\big(\mathcal{X}(\widetilde{Y}_1^*\!-\!\widetilde{X}_1^*)\big)_{\alpha\gamma}
\circ(\Omega_2)_{\alpha\gamma}=0,\qquad\\
\widetilde{X}_{\gamma\alpha}^*+\big(\mathcal{S}(\widetilde{Y}_1^*\!-\!\widetilde{X}_1^*)\big)_{\gamma\alpha}
\circ(\Omega_1)_{\gamma\alpha}+\big(\mathcal{X}(\widetilde{Y}_1^*\!-\!\widetilde{X}_1^*)\big)_{\gamma\alpha}
\circ(\Omega_2)_{\gamma\alpha}=0,\,\widetilde{Y}_{\beta\gamma}^*=0,\qquad\nonumber\\
\widetilde{Y}_{\gamma\beta}^*=0,\widetilde{Y}_{\gamma\gamma}^*=0,\widetilde{Y}_{\beta c}^*=0,\widetilde{Y}_{\gamma c}^*=0,\,
\widetilde{X}_{\alpha c}^*\circ (E_{\alpha c}-(\Omega_3)_{\alpha c})+\widetilde{Y}_{\alpha c}^*\circ (\Omega_3)_{\alpha c}=0.\nonumber
\end{align}
By the definitions of $\Theta_1,\Theta_2$ and $\Sigma_1,\Sigma_2$, equation
\eqref{main-equa2} can be compactly written as \eqref{result-equa1}-\eqref{result-equa3}.
Conversely, it is easy to check that if $(\widetilde{X}^*,\widetilde{Y}^*)$ satisfies
\eqref{main-equa2} or its compact form \eqref{result-equa1}-\eqref{result-equa3},
then \eqref{main-ineq1} holds for any $H\in\mathbb{R}^{m\times n}$, i.e.,
$(X^*,Y^*)\in\!\mathcal{N}_{{\rm gph}\,\partial\|\cdot\|_*}^{\pi}(X,Y)$.
\end{proof}
\begin{remark}\label{remark-pnormal}
For any given $(X,Y)\in{\rm gph}\,\partial\|\cdot\|_*$, let $\overline{Z}=X+Y$.
By Theorem \ref{proxnormal-theorem}, if $\|\overline{Z}\|<1$,
then $(X^*,Y^*)\in\mathcal{N}_{{\rm gph}\,\partial\|\cdot\|_*}^{\pi}(X,Y)$ if and only if
$Y^*=0$; if $\|\overline{Z}\|=1$, then $(X^*,Y^*)\in\mathcal{N}_{{\rm gph}\,\partial\|\cdot\|_*}^{\pi}(X,Y)$
if and only if $\widetilde{X}^*$ and $\widetilde{Y}^*$ take the following form
\[
\widetilde{X}^*\!=\!\left[\begin{matrix}
\widetilde{X}_{\beta\beta}^*& \widetilde{X}_{\beta\gamma}^*&\widetilde{X}_{\beta c}^*\\
\widetilde{X}_{\gamma\beta}^*& \widetilde{X}_{\gamma\gamma}^*&\widetilde{X}_{\gamma c}^*\\
\end{matrix}\right]\ \ {\rm and}\ \
\widetilde{Y}^*\!=\!\left[\begin{matrix}
\widetilde{Y}_{\beta\beta}^*& 0_{\beta\gamma}&0_{\beta c}\\
0_{\gamma\beta}& 0_{\gamma\gamma}& 0_{\gamma c}\\
\end{matrix}\right]\ {\rm with}\ \widetilde{X}_{\beta\beta}^*\preceq 0,\,
\widetilde{Y}_{\beta\beta}^*\succeq 0.
\]
\end{remark}
\subsection{Limiting normal cone}\label{subsec3.2}
Let $\beta$ be a nonempty index set and denote the set of all partitions of $\beta$
by $\mathscr{P}(\beta)$. Write
\(
\mathbb{R}_{>}^{|\beta|}:=\big\{z\in\mathbb{R}^{|\beta|}\!:\ z_1\ge\cdots\ge z_{|\beta|}>0\big\}.
\)
For any $z\in\mathbb{R}_{>}^{|\beta|}$, let $D(z)\in\mathbb{S}^{|\beta|}$ denote
the generalized first divided difference matrix of $h(t)=\min(1,t)$ at $z$,
which is defined as
\begin{equation}\label{Ddiff-matrix}
(D(z))_{ij}:=\left\{\begin{array}{cl}
\!\frac{\min(1,z_i)-\min(1,z_j)}{z_i-z_j}\in[0,1]&{\rm if}\ z_i\ne z_j,\\
0 &{\rm if}\ z_i=z_j\ge 1,\\
1 & {\rm otherwise}.
\end{array}\right.
\end{equation}
Write
\(
\mathcal{U}_{|\beta|}
:=\big\{\overline{\Omega}\in\mathbb{S}^{|\beta|}\!:\ \overline{\Omega}=\lim_{k\to\infty}D(z^k),\,
z^k\to e_{|\beta|},\, z^k\in\mathbb{R}_{>}^{|\beta|}\big\}.
\)
For each $\Xi_1\in\mathcal{U}_{|\beta|}$, by equation \eqref{Ddiff-matrix} there exists a partition $(\beta_{+},\beta_{0},\beta_{-})\in\mathscr{P}(\beta)$ such that
\begin{equation}\label{Xi1-matrix}
\Xi_1=\left[\begin{matrix}
0_{\beta_{+}\beta_{+}}& 0_{\beta_{+}\beta_{0}}&(\Xi_1)_{\beta_{+}\beta_{-}}\\
0_{\beta_{0}\beta_{+}}& 0_{\beta_{0}\beta_{0}}& E_{\beta_{0}\beta_{-}}\\
(\Xi_1)_{\beta_{+}\beta_{-}}^{\mathbb{T}}&E_{\beta_{-}\beta_{0}}&E_{\beta_{-}\beta_{-}}
\end{matrix}\right],
\end{equation}
where each entry of $(\Xi_1)_{\beta_{+}\beta_{-}}$ belongs to $[0,1]$.
Let $\Xi_2$ be the matrix associated to $\Xi_1$:
\begin{equation}\label{Xi2-matrix}
\Xi_2=\left[\begin{matrix}
E_{\beta_{+}\beta_{+}}& E_{\beta_{+}\beta_{0}}& E_{\beta_{+}\beta_{-}}\!-\!(\Xi_1)_{\beta_{+}\beta_{-}}\\
E_{\beta_{0}\beta_{+}}& 0_{\beta_{0}\beta_{0}}& 0_{\beta_{0}\beta_{-}}\\
E_{\beta_{-}\beta_{+}}\!-\!(\Xi_1)_{\beta_{+}\beta_{-}}^{\mathbb{T}}&0_{\beta_{-}\beta_{0}}&0_{\beta_{-}\beta_{-}}
\end{matrix}\right].
\end{equation}
With the above notations, we shall provide the exact formula for the limiting normal cone to
${\rm gph}\,\partial \|\cdot\|_*$ in the following theorem, whose proof is included in Appendix.
\begin{theorem}\label{normal-theorem}
For any given $(X,Y)\in{\rm gph}\,\partial \|\cdot\|_*$, let $\overline{Z}=X+Y$ have
the SVD as in Lemma \ref{dir-derivative} with $\alpha,\beta,\gamma$ and $c$ defined by
\eqref{ab-index}-\eqref{gc-index}. Then, $(G,H)\in\mathcal{N}_{{\rm gph}\,\partial \|\cdot\|_*}(X,Y)$
if and only if $(\widetilde{G},\widetilde{H})$ with $\widetilde{G}=\overline{U}^{\mathbb{T}}G\overline{V}$
and $\widetilde{H}=\overline{U}^{\mathbb{T}}\!H\overline{V}$ satisfies the following conditions
\begin{subequations}
\begin{align}\label{equa1-normal}
\Theta_1\circ\mathcal{S}(\widetilde{H}_1)+\Theta_2\circ \mathcal{S}(\widetilde{G}_1)
+\Sigma_1\circ \mathcal{X}(\widetilde{H}_1)+\Sigma_2\circ \mathcal{X}(\widetilde{G}_1)=0,\qquad\qquad\\
\label{equa2-normal}
\widetilde{G}_{\alpha c}+(\Omega_3)_{\alpha c}\circ
(\widetilde{H}_{\alpha c}-\widetilde{G}_{\alpha c})=0,\,\widetilde{H}_{\beta c}=0,\,
\widetilde{H}_{\gamma c}=0,\qquad\qquad\quad\\
\label{equa3-normal}
(\widetilde{G}_{\beta\beta},\widetilde{H}_{\beta\beta})\in\!
\bigcup_{Q\in\mathbb{O}^{|\beta|}\atop \Xi_1\in\mathcal{U}_{|\beta|}}
\!\left\{(M,N)\ \bigg|\!\left.\begin{array}{ll}
\Xi_1\circ\widehat{N}+\Xi_2\circ \mathcal{S}(\widehat{M})+\Xi_2\circ \mathcal{X}(\widehat{N})=0\\
\quad {\rm with}\ \widehat{N}=Q^{\mathbb{T}}NQ,\,\widehat{M}=Q^{\mathbb{T}}MQ,\\
\quad Q_{\beta_0}^{\mathbb{T}}MQ_{\beta_0}\preceq 0,\
Q_{\beta_0}^{\mathbb{T}}NQ_{\beta_0}\succeq 0
\end{array}\right.\!\right\},
\end{align}
\end{subequations}
where $\widetilde{G}_1=\overline{U}^{\mathbb{T}}G\overline{V}_1,\widetilde{H}_1\!=\overline{U}^{\mathbb{T}}H\overline{V}_1$,
and $\Theta_1,\Theta_2,\Sigma_1,\Sigma_2$ are the same as those before.
\end{theorem}
To close this paper, we point out that Theorem \ref{normal-theorem} also provides
the characterization for the coderivative of $\Pi_\mathbb{B}$. Indeed,
by Lemma \ref{graph-subdiff}, ${\rm gph}\,\Pi_{\mathbb{B}}=\mathcal{L}^{-1}({\rm gph}\,\partial\|\cdot\|_*)$ with
\[
\mathcal{L}(X,Y):=(X\!-\!Y,Y)\ \ {\rm for}\ \ (X,Y)\in\mathbb{R}^{m\times n}\times\mathbb{R}^{m\times n}.
\]
Since the linear map $\mathcal{L}\!:\mathbb{R}^{m\times n}\times\mathbb{R}^{m\times n}\to
\mathbb{R}^{m\times n}\times\mathbb{R}^{m\times n}$ is onto, from \cite[Exercise 6.7]{RW98}
\[
\mathcal{N}_{{\rm gph}\,\Pi_{\mathbb{B}}}(X,Y)\!=\!\mathcal{L}^*(\mathcal{N}_{{\rm gph}\,\partial\|\cdot\|_*}(\mathcal{L}(X,Y))),
\]
where $\mathcal{L}^*$ is the adjoint of $\mathcal{L}$.
By the definition of coderivative (see \cite[Definition 8.33]{RW98}),
\begin{equation}
W\in D^*\Pi_\mathbb{B}(X,Y)(S)\Longleftrightarrow (W,W\!-\!S)\in\mathcal{N}_{{\rm gph}\,\partial\|\cdot\|_*}(X\!-\!Y,Y).
\end{equation}
Similarly, Theorem \ref{proxnormal-theorem} also provides
the characterization for the regular coderivative:
\begin{equation}
W\in \widehat{D}^*\Pi_\mathbb{B}(X,Y)(S)\Longleftrightarrow (W,W\!-\!S)\in\widehat{\mathcal{N}}_{{\rm gph}\,\partial\|\cdot\|_*}(X\!-\!Y,Y).
\end{equation}
In addition, with the help of Theorem \ref{normal-theorem} and the Mordukhovich
criterion \cite[Proposition 3.5]{Mordu94} on the Aubin property of a multifunction,
one may easily obtain the practical conditions for the Aubin property of $\partial\|\cdot\|_*$.
In our future work, we shall use Theorem \ref{proxnormal-theorem}
and \ref{normal-theorem} to derive the optimality conditions of
the rank minimization problem \eqref{rank-min}.
| a95f7c897720a98f9f4e848060c6c628af81ae76 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Nowadays, millions of videos are being uploaded to the Internet every day. These videos capture all aspects of multimedia content about their uploader's daily life. These explosively growing user generated content videos online are becoming an crucial source of video data. Automatically categorizing videos into concepts, such as people actions, objects, etc., has become an important research topic. Recently many work have been proposed to tackle with building concept detectors both in image domain and video domain \cite{deng2009imagenet,liang2015towards,tang2012shifting,karpathy2014large,jiang2015exploiting}. However, the need for manual labels by human annotators has become one of the major important limitations for large-scale concept learning. It is even more so in video domain, since training concept detectors on videos is more challenging than on still images.
Many image datasets such as ImageNet~\cite{deng2009imagenet}, CIFAR~\cite{krizhevsky2009learning}, PASCAL VOC\cite{everingham2010pascal}, MS COCO~\cite{lin2014microsoft} and Caltech~\cite{fei2006one} have been collected and manually labeled. In video domain, some largest datasets such as UCF-101\cite{soomro2012ucf101}, MCG-WEBV~\cite{cao2009mcg}, TRECVID MED~\cite{over2014trecvid} and FCVID~\cite{jiang2015exploiting} are popular benchmark datasets for video classification.
Collecting such datasets requires a large amount of human effort that can take thousands of man hours. In addition, manually labeling video requires playing back the video, which is more time consuming and expensive than labeling still images. As a result, the largest labeled video collection, FCVID~\cite{jiang2015exploiting}, only contains about 0.09 million labels with 239 concept classes, much less than the 14 million labels with over 20,000 classes in the image collection ImageNet~\cite{deng2009imagenet}.
Many state-of-the-art models in visual classification are based on the neural networks~\cite{jiang2015exploiting,karpathy2014large,varadarajan2015efficient}. As the architecture gets deeper, the neural network would need more data to train in order to get better performance. However, more data needs more human supervision which are more expensive to acquire in the video domain.
Videos are available on the web and contain rich contextual information with a weak annotation about their content, such as their titles, descriptions and surrounding text. These webly-labeled data are orders of magnitude larger than that of any manually-labeled collections. Moreover, automatically extracted features from multiple modalities such as existing still image classification models, automatic speech recognition and optical character recognition tools can be useful additional information for the content of the video.
Figure \ref{cur-example} shows an example of webly-labeled video for walking with a dog. As we see, the textual metadata we get from the web videos contain useful but very noisy information. The multi-modal prior information we get is correlated across modalities, as the image classification results and speech transcript show high probability of dog appearance, while the textual metadata indicates the same content. Some of the videos (about 20\% in the FCVID dataset) have very little textual metadata and we can only obtain web labels via other modalities.
To address the problem of learning detectors from the big web data, in this paper, we utilize multi-modal information to harness prior knowledge from the web video without any manual efforts.
Existing methods on learning from noisy webly-labeled data has mainly focused on the image domain \cite{fergus2005learning,li2010optimol,bergamo2010exploiting,chen2015webly}. Existing studies demonstrated promising results in this direction. However, these methods are primarily based on some heuristic methods. It is not clear what objective is being optimized and where or even whether the learning process will converge. Moreover, these methods only utilize a single text modality in the image domain. It is unclear how to exploit the multi-modal prior knowledge for concept learning from the rich context of Internet video.
To utilize the large amount of webly-labeled video data for concept learning, we propose a learning framework called \textbf{WEbly-Labeled Learning (WELL)}. It is established on the theories called \textit{curriculum learning}~\cite{bengio2009curriculum} and \textit{self-paced learning}~\cite{kumar2010self}. The learning framework is motivated by human learning, where people generally start learning easier aspects of a concept, and then gradually take more complex examples into the learning process\cite{bengio2009curriculum,kumar2010self,jiang2015self}.
Following this idea, WELL learns a concept detector iteratively from first using a few samples with more confident labels (more related to the concept), then gradually incorporate more video samples with noisier labels. The algorithm combines the prior knowledge, called learning curriculum, extracted from the webly-labeled data with the dynamic information learned from the statistical model (self-paced) to determine which video samples to learn in the next iteration. This idea of easy-to-hard learning paradigm has been adopted for learning in noisy web data~\cite{jiang2014easy,chen2015webly,kumar2010self} and has been proved to be efficient to deal with noise and outliers.
Our proposed method generalizes such learning paradigm using a clear objective function. It is proved to be convex and is a also general framework that can incorporate state-of-the-art deep learning methods to learn robust detectors from noisy data. Our framework fundamentally changes self-paced learning and allows learning for video concept detectors at unlimited scale.
Figure~\ref{well-cnn} shows the architecture of the proposed method. We extract keyframe-level convolutional neural network features and feed them into WELL layer with average pooling and iterative learning process. We have also tried using other video features such as motion features and audio MFCC features.
Our contributions are threefold. First, we address the problem of learning robust video concept detectors from noisy web data through a general framework with solid theoretical justifications. We show that WELL not only outperforms state-of-the-art learning methods on noisy labels, but also, notably, achieves comparable results with state-of-the-art models trained using manual annotation on one of the largest video dataset. Second, we provide detailed comparison of different approaches to exploit multi-modal curriculum from noisy labels and verify that our method is robust against certain level of noisiness in the video data. Finally, the efficacy and the scalability have been empirically demonstrated on two public benchmarks, including by far the largest manually-labeled video set called FCVID~\cite{jiang2015exploiting} and the largest multimedia dataset called YFCC100M~\cite{thomee2015yfcc100m}. The promising results suggest that detectors trained on sufficient webly-labeled videos may outperform detectors trained on any existing manually-labeled datasets.
\begin{figure*}
\centering
\includegraphics[width=0.9\textwidth,height=1.5in]{well-cnn}
\vspace{-6mm}
\caption{\textbf{Learning architecture of WEbly-Labeled Learning (WELL)}}
\label{well-cnn}
\end{figure*}
\section{Related Work}
\textbf{Curriculum and Self-paced Learning}: Recently a learning paradigm called~\textit{curriculum learning} (CL) was proposed by Bengio et al., in which a model is learned by gradually incorporating from easy to complex samples in training so as to increase the entropy of training samples~\cite{bengio2009curriculum}. A curriculum determines a sequence of training samples and is often derived by predetermined heuristics in particular problems.
For example, Chen et al. designed a curriculum where images with clean backgrounds are learned before the images with noisy backgrounds~\cite{chen2015webly} , i.e. their method first builds a feature representation by a Convolutional Neural Network (CNN) on images with clean background and then they fine tune the models on images with noisy background.
In~\cite{spitkovsky2009baby}, the authors approached grammar induction, where the curriculum is derived in terms of the length of a sentence. Because the number of possible solutions grows exponentially with the length of the sentence, and short sentences are easier and thus should be learn earlier.
The heuristic knowledge in a problem often proves to be useful. However, the curriculum design may lead to inconsistency between the fixed curriculum and the dynamically learned models. That is, the curriculum is predetermined a prior and cannot be adjusted accordingly, taking into account the feedback about the learner. To alleviate the issue of CL, Kumar et al. designed a learning paradigm, called \emph{self-paced learning} (SPL)~\cite{kumar2010self}. SPL embeds curriculum design as a regularizer into the learning objective. Compared with CL, SPL exhibits two advantages: first, it jointly optimizes the learning objective with the curriculum, and thus the curriculum and the learned model are consistent under the same optimization problem; second, the learning is controlled by a regularizer which is independent of the loss function in specific problems. This theory has been successfully applied to various applications, such as matrix factorization~\cite{zhao2015self}, action/event detection~\cite{jiang2014self}, domain adaption~\cite{tang2012shifting}, tracking~\cite{supancic2013self} and segmentation~\cite{kumar2011learning}, reranking~\cite{jiang2014easy}, etc.
\textbf{Learning Detectors in Web Data}: Many recent studies have been proposed to utilize the large amount of noisy data from the Internet. For example, \cite{mitchell2015never} proposed a Never-Ending Language Learning (NELL) paradigm and built adaptive learners that makes use of the web data by learning different types of knowledge and beliefs continuously. Such learning process is mostly self-supervised, and previously learned knowledge enables learning further types of knowledge.
In the image domain, existing methods try to tackle the problem of constructing qualified training sets based on the search results of text or image search engines~\cite{fergus2005learning,li2007optimol,chen2013neil,li2014exploiting,divvala2014learning,liang2015towards}.
For example, \cite{fergus2005learning} extended the probabilistic Latent Semantic Analysis in visual domain and learned object categories using results from image search engines. \cite{li2007optimol} proposed an incremental learning paradigm that initialized from a few seed images and repeatedly trained models to refine the collected image dataset from the Internet.
NEIL~\cite{chen2013neil} followed the idea of NELL and learned from web images to form a large collection of concept detectors iteratively via a semi-supervised fashion. By combining the classifiers and the inter-concept relationships it learned, NEIL can be used for scene classification and object detection task.
\cite{li2014exploiting} tried to learn robust classifiers by considering the noisy textual information accompanied with web images. However, the down side is that the portion of the true positive samples has to be determined via prior knowledge, where in fact it is not accurate to assume the same number of true positive samples for any targeted concept.
\cite{divvala2014learning} introduced a webly-supervised visual concept learning method that automatically learns large amount of models for a wide range of variations within visual concepts. They discovered concept variances through vocabulary of online books, and then downloaded images based on text-search from the web to train object detection and localization models.
\cite{liang2015towards} presented a weakly-supervised method called Baby Learning for object detection from a few training images and videos. They first embed the prior knowledge into a pre-trained CNN. When given very few samples for a new concept, a simple detector is constructed to discover much more training instances from the online weakly labeled videos. As more training samples are selected, the concept detector keeps refining until a mature detector is formed.
Another recent work in image domain \cite{chen2015webly} proposed a webly supervised learning of Convolutional Neural Network. They utilized easy images from search engine like Google to bootstrap a first-stage network and then used noisier images from photo-sharing websites like Flickr to train an enhanced model.
In video domain, only few studies \cite{duan2012visual,han2015fast,varadarajan2015efficient} have been proposed for noisy data learning since training robust video concept detectors is more challenging than the problem in the image domain.
\cite{duan2012visual} tackled visual event detection problem by using SVM based domain adaptation method in web video data.
\cite{han2015fast} described a fast automatic video retrieval method using web images. Given a targeted concept, compact representations of web images obtained from search engines like Google, Flickr are calculated and matched to compact features of videos. Such method can be utilized without any pre-defined concepts.
\cite{varadarajan2015efficient} discussed a method that exploits the YouTube API to train large scale video concept detectors on YouTube. The method utilized a calibration process and hard negative mining to train a second order mixture of experts model in order to discover correlations within the labels.
Most of the existing methods are heuristic approaches as it is unclear what objective is being optimizing on the noisy data. Moreover, results obtained from the web search results is just one approach to acquire prior knowledge or curriculum. To the best of our knowledge, there have been no systematical studies on exploiting the multi-modal prior knowledge in video concept learning on noisy data. Since search engine algorithm is changing rapidly, it is unclear that how noisy the web labels are and how the level of noisiness in the data will affect performance. In this paper, we proposed a theoretically justified method with clear framework for curriculum constructing and model learning. We also empirically demonstrate its superior performance over representative existing methods and systemically verify that WELL is robust against the level of noisiness of the video data.
\vspace{-1mm}
\section{WEbly-Labeled Learning (WELL)}
\subsection{Problem Description}
In this paper, following~\cite{varadarajan2015efficient}, we consider a concept detector as a classifier and our goal is to train concept detectors from webly-labeled video data without any manual labeling effort. Given a noisy web video training set and a target concept set, we do not assume any distribution of the noise. Formally, we represent the training set as $\mathcal{D} = \{(\mathbf{x}_i,\mathbf{z}_i,\mathbf{\tilde{y}}_i)\}_{i=1}^n$
where $\mathbf{x}_i \in \mathbb{R}^m$ denotes the feature for the $i^{th}$ observed sample, and $\mathbf{z}_i$ represents its noisy web label, which generally means the prior knowledge we can get from the web without additional human effort that not only includes textual information provided by the uploaders in the video metadata but also includes prior knowledge from other modalities using existing tools like pre-trained Convolutional Neural Network image detector \cite{chatfield2014return}, Automatic Speech Recognition \cite{povey2011kaldi} and Optical character recognition \cite{smith2007overview}. The $\mathbf{\tilde{y}}_i \subset \mathcal{Y}$ is the inferred concept label set for the $i^{th}$ observed sample based on its noisy web label, and $\mathcal{Y}$ denotes the full set of target concepts. In our experiment, to simplify the problem, we apply our method on binary classification and infer binary labels $\tilde{y}_i$ from the noisy web labels.
The noisy web labels can be used to automatically infer concept labels by matching the concept name to the video textual metadata. For example, a video may be inferred to the concept label ``cat'' as its textual title contains cat. \cite{varadarajan2015efficient} utilizes the YouTube topic API, which is derived from the textual metadata, to automatically get concept labels for videos. The web labels are quite noisy as the webly-labeled concepts may not present in the video content whereas the concepts not in the web label may well appear.
\subsection{Model and Algorithm}
\subsubsection{Objective Function}
To leverage the noisy web labels in a principled way, we propose WEbly-Labeled Learning (WELL). Formally, given a training set $\mathcal{D}$ as described before, Let $L(\tilde{y}_i,g(\mathbf{x}_i,\mathbf{w}))$, or $\ell_i$ for short, denote the loss function which calculates the cost between the inferred label $\tilde{y}_i$ and the estimated label $g(\mathbf{x}_i,\mathbf{w})$. Here $\mathbf{w}$ represents the model parameter inside the decision function $g$.
For example, in our paper, $\mathbf{w}$ represents the weight parameters in the Convolutional Neural Network (CNN) and the Support Vector Machine (SVM).
Our objective function is to jointly learn the model parameter $\mathbf{w}$ and the latent weight variable $\mathbf{v}= [v_1,\cdots,v_n]^T$ by:
\vspace{-2mm}
\begin{equation}
\label{eq:spcl_obj}
\begin{split}
\!\min_{\mathbf{w},\mathbf{v}\in \lbrack 0,\!1]^{n}}\!\!\mathbb{E}(\mathbf{w},\!\mathbf{v}\!;\lambda,\!\Psi\!)
\!=\! \sum_{i=1}^n v_i L(\tilde{y}_i,\!g(\mathbf{x}_i,\!\mathbf{w})) \!+\! f(\mathbf{v};\! \lambda), \\ \text{ subject to } \mathbf{v} \in \Psi
\end{split}
\end{equation}
where $\mathbf{v=[}v_{1},v_{2},\cdots ,v_{n}\mathbf{]}^{T}$ denote the
latent weight variables reflecting the inferred labels' confidence. The weights determine a learning sequence of samples, where samples with greater weights tend to be learned earlier.
Our goal is to assign greater weights to the samples with more confident labels whereas smaller or zero weights to the samples with noisy labels. To this end, we employ the self-paced regularizer $f$, which controls the learning process of the model.
We consider the linear regularizer Eq.~\eqref{eq:linear_scheme} proposed in~\cite{jiang2015self}:
\begin{equation}
\label{eq:linear_scheme}
f(\mathbf{v};\lambda) = \frac{1}{2} \lambda \sum_{i=1}^n ( v_i^2 - 2v_i ).
\vspace{-1mm}
\end{equation}
Generally, a self-paced regularizer determines the scheme for penalizing the latent weight variables. Physically it resembles the learning schemes human used in understanding new concepts. The linear scheme corresponds to a prudent strategy, which linearly penalizes the samples that are different to what the model has already learned (see Eq.~\eqref{eq:linear_closedform}).
The hyper-parameter $\lambda$ $(\lambda > 0)$ is called ``model age'', which controls the pace at which the model learns new samples. When $\lambda$ is small only samples of with small loss will be considered. As $\lambda$ grows, more samples with larger loss will be gradually appended to train a ``mature'' mode.
$\Psi$ in Eq.~\eqref{eq:spcl_obj} is a curriculum region derived from noisy web labels $\mathbf{z}$ that incorporates the prior knowledge extracted from the webly-labeled data as a convex feasible region for the weight variables. The shape of the region weakly implies a prior learning sequence of samples, where the expected values for favored samples are larger. The curriculum region can be derived in a variety of ways that make use of different modalities. We will discuss this topic in details in following section. A straightforward approach is by counting the term frequency in the video's textual metadata. That is, for example, the chance of a video containing the concept ``cat'' become higher when it has more word ``cat'' in its title, description or tags.
Eq.~\eqref{eq:spcl_obj} represents a concise and general optimization model ~\cite{jiang2015self}. It combines the prior knowledge extracted from the noisy webly-labeled data (as the curriculum region) and the information dynamically learned during the training (via the self-paced regularizer). Intuitively, the prior knowledge serves as an instructor providing a guidance on learning the latent weights, but it leaves certain freedom for the model (the student) to adjust the actual weights according to its learning pace. Experimental results in Section~\ref{sec:experiments} demonstrate the learning paradigm can better overcome the noisy labels than heuristic approaches. Figure \ref{well-cnn} shows the learning process of our method.
Following~\cite{kumar2010self,jiang2015self}, we employ the alternative convex search algorithm to solve Eq.~\eqref{eq:spcl_obj}. Algorithm~\ref{alg:overall} takes the input of a curriculum region, an instantiated self-paced regularizer and a step size parameter; it outputs an optimal model parameter $\mathbf{w}$. First of all, it initializes the latent weight variables in the feasible region. Then it alternates between two steps until it finally converges: Step 3 learns the optimal model parameter with the fixed and most recent $\mathbf{v}^*$; Step 5 learns the optimal weight variables with the fixed $\mathbf{w}^*$. In the beginning, the model ``age'' is gradually increased so that more noisy samples will be gradually incorporated in the training. Step 3 can be conveniently implemented by existing off-the-shelf supervised learning methods such as the back propagation. Gradient-based methods can be used to solve the convex optimization problem in Step 4. According to~\cite{gorski2007biconvex}, the alternative search in Algorithm~\ref{alg:overall} converges as the objective function is monotonically decreasing and is bounded from below.
\setlength{\textfloatsep}{1pt}
\vspace{-3mm}
\IncMargin{1em}
\begin{algorithm}
\SetKwData{Left}{left}\SetKwData{This}{this}\SetKwData{Up}{up}
\SetKwInOut{Input}{input}\SetKwInOut{Output}{output}
\LinesNumbered
\Input{Input dataset $\mathcal{D}$, curriculum region $\Psi$, self-paced function $f$ and a step size $\mu$}
\Output{Model parameter $\mathbf{w}$}
\BlankLine
Initialize $\mathbf{v}^*$, $\lambda$ in the curriculum region\;
\While{not converged} {
Update $\mathbf{w}^* = \arg\min_{\mathbf{w}} \mathbb{E}(\mathbf{w},\mathbf{v}^*;\lambda, \Psi)$\;
Update $\mathbf{v}^* = \arg\min_{\mathbf{v}} \mathbb{E}(\mathbf{w}^*,\mathbf{v}; \lambda, \Psi)$\;
\lIf{$\lambda$ is small}{increase $\lambda$ by the step size $\mu$}
}
\Return $\mathbf{w}^*$
\caption{\label{alg:overall} WEbly-Labeled Learning (WELL).}
\end{algorithm}
\DecMargin{1em}
\vspace{-3mm}
At an early age when $\lambda$ is small, Step 4 in Algorithm~\ref{alg:overall} has an evident suppressing effect over noisy samples that have greater loss to the already learned model. For example, with a fixed $\mathbf{w}$, the unconstrained close-formed solution for the regularizer in Eq.~\eqref{eq:linear_scheme} equals
\begin{equation}
\label{eq:linear_closedform}
v_i^* =\begin{cases}
-\frac{1}{\lambda} \ell_i +1 & \ell_{i} < \lambda\\
0 & \ell_{i} \ge \lambda
\end{cases},
\end{equation}
where $v_i$ represents the $i$th element in the optimal solution $\mathbf{v}^* = [v_1^*, \cdots, v_n^*]^T$. Eq.~\eqref{eq:linear_closedform} called linear regularizer indicates the latent weight is proportional to the negative sample loss, and the sample whose loss is greater or equals to $\lambda$ will have zero weights and thus will not affect the training of the next model. As the model age grows, the hyper-parameter $\lambda$ increases, and more noisy samples will be used into training. The prior knowledge embedded in the curriculum region $\Psi$ is useful as it suggests a learning sequence of samples for the ``immature'' model. \cite{meng2015objective} theoretically proves that the iterative learning process is identical to optimizing a robust loss function on the noisy data.
If we keep increasing $\lambda$, the model will ultimately use every sample in the noisy data, which is undesirable as the labels of some noisy samples are bound to be incorrect. To this end, we stop increasing the age $\lambda$ after about a certain number of iterations (early stopping). The exact stopping iteration for each detector is automatically tuned in terms of its performance on a small validation set.
\begin{figure*}
\centering
\includegraphics[width=0.8\textwidth,height=0.3\textheight]{curriculum_example}
\caption{Curriculum Extraction Example. We automatically extract information using meaningful prior knowledge from several modalities and fuse them to get curriculum for WELL. Our method makes use of text, speech, visual cues while common methods like search engine only extract from textual information.}
\label{curriculum_example}
\end{figure*}
\subsubsection{Model Details}\label{sec:partial_curriculum}
In this section we discuss further details of the curriculum region $\Psi$ and the self-paced regularizer $f(\mathbf{v};\lambda)$.
$\Psi$ is a feasible region that embeds the prior knowledge extracted from the webly-labeled data. It physically corresponds to a convex search region for the latent weight variable. Given a set of training samples $\mathbf{X}=\{\mathbf{x}_{i}\}_{i=1}^{n}$,
we utilize the partial-order curriculum which generalizes the total-order curriculum by incorporating the incomplete prior over groups of samples. Samples in the confident groups should be learned earlier than samples in the less confident groups. It imposes no prior over the samples within the same group nor the samples not in any group. Formally, we define a partial order relation $\preceq$ such that $x_i \preceq x_j$ indicates that the sample $x_i$ should be learned no later than $x_j$ ($i,j \in [1,n]$). Similarly given two sample subsets $\mathbf{X}_{a} \preceq \mathbf{X}_{b}$ denotes the samples in $\mathbf{X}_{a}$ should be learned no later than the samples in $\mathbf{X}_{b}$.
In our problem, we extract the partial-order curriculum in the webly-labeled data in the following ways: we only distinguish the training order for groups of samples.
Information from different modalities of the web labels can be used for curriculum design. A straightforward way is to directly utilize the textual descriptions of the videos generated by the uploaders. We compare common ways to extract curriculum from web data for concept learning to the proposed novel method that utilize state-of-the-art topic modeling techniques in natural language processing.
In the following methods (Exact \& Stem Matching, Word Embedding and Latent Topic with Word Embedding), we first extract bag-of-words features from different modalities and then match them using specific matching methods to the concept words. Each video will then come with a matching score to each concept. In our experiment, we divide the data into two partial-order curriculum groups, where the videos with matching scores larger than zero will be in one group while others will be in the other group.
\textbf{Exact \& Stem Matching} We build curriculum directly using exact word matching or stemmed word matching between the textual metadata of the noisy videos to the targeted concept names.
\textbf{YouTubeTopicAPI} We directly utilize the YouTube topic API to search for videos that are related to the concept words. The topic API utilizes textual information of the uploaded videos to obtain related topics of the videos from Freebase.
\textbf{SearchEngine} We build curriculum using the search result from a text-based search engine. It is similar to related web-search based methods.
\textbf{Word Embedding} We use word embedding \cite{mikolov2013distributed} to match words in metadata to targeted concept words in order to deal with synonyms and related concepts. The word embedding is trained using Google News data.
\textbf{Latent Topic} We build curriculum based on the latent topic we learned from the noisy label. We incorporate Latent Dirichlet Allocation (LDA)~\cite{blei2003latent} to determine how each noisy labeled video is related to each target concept. The basic idea is that each web video consists of mixtures of topics (concepts), and each topic is characterized by a distribution of words.
Formally, given all the noisy information extracted from a web video and collected them as a document $\mathbf{d_i}$, which combines into a corpus $\mathbf{d}$, we have a target set of $\mathbf{k}$ topics, then the key inferential problem that we are going to solve is that of computing the posterior distribution of the latent topics given a corpus (how likely the videos are related to each target concept given the noisy information):
\begin{equation}
\label{eq:lda}
p(\theta,\mathbf{t}|\mathbf{d},\alpha,\beta) = \frac{p(\theta,\mathbf{t},\mathbf{d}|\alpha,\beta)}{p(\mathbf{d}|\alpha,\beta)}
\end{equation}
where $\theta$ is the topic distribution variable for the corpus, $\theta$ $\sim$ Dir($\alpha$), in which Dir($\alpha$) represents a uniform Dirichlet distribution with scaling parameter $\alpha$. The $\mathbf{t}$ is the topic assignment variable that indicates which topic each word belong to in the document.
$\beta$ is the Dirichlet prior on the per-topic word distribution, in which we impose asymmetric priors over the word distribution so that each learned topic will be seeded with particular words in our target concept. For example, a topic will be seeded with words "walk, dog" for the target concept "WalkingWithDog".
The parameter estimation in Eq~\eqref{eq:lda} can be done via Bayes methods.
However, Eq.~\eqref{eq:lda} is intractable to compute since $p(\mathbf{d}|\alpha,\beta)$ is intractable due to the coupling between $\theta$ and $\beta$ \cite{blei2003latent}. To solve this problem, approximate inference algorithms are introduced and we use the online variational inference algorithm from \cite{hoffman2010online}. The true posterior is approximated by a simpler distribution:
\begin{equation}
\label{eq:lda-vi}
q(\theta,\mathbf{t}|\gamma,\phi) = q(\theta|\gamma)\prod_{i=1}^{N}{q(t_i|\phi_i)}
\end{equation}
where the Dirichlet parameter $\gamma$ and the multinomial parameters ($\phi_1,...,\phi_N$) are the free variational parameters. Thus the maximization problem is equivalent to minimizing the Kullback-Leibler(KL) divergence between $q(\theta,\mathbf{t}|\gamma,\phi)$ and the posterior $p(\theta,\mathbf{t}|\mathbf{d},\alpha,\beta)$ \cite{blei2003latent}. The optimization problem can then be solve using Expectation-Maximization algorithm \cite{hoffman2010online}. The estimated parameters $\theta,t$ in Eq.~\eqref{eq:lda} is then used for constructing curriculum region in Eq.~\eqref{eq:spcl_obj}.
\textbf{Latent Topic with Word Embedding (LT+WE)} We first learn latent topics using LDA to replace the concept words with a topic word distribution and then match the latent topic words to the web label's bag-of-words features by using the word embeddings. We compare this method to the others using only textual information from the web videos.
We also use this method to get curriculum from other modalities such as Automatic Speech Recognition (ASR) \cite{povey2011kaldi}, Optical Character Recognition (OCR) \cite{smith2007overview} and basic image detector pre-trained on still images~\cite{ILSVRC15} (in this paper we use VGG net~\cite{simonyan2014very}, extract keyframe-level image classification results and average them to get video-level results.).
We extract bag-of-word features from them and combine them with linear weights.
Detailed fusion experiments can be found in Section \ref{sec:experiments}. We empirically set OCR's weight to be small as the results are much noisier than other features.
Figure~\ref{curriculum_example} shows an example of the noisy web video data and how the curriculum is extracted with different methods. Our method can utilize information from different modalities while common methods like search engine only consider textual information.
We compare the performance of different ways of curriculum design by training detectors directly in Section 4.
The labels in webly-labeled data are much noisier than manually-labeled data, and as a result, we found that the learning is prone to overfitting the noisy labels. To address this issue, inspired by the dropout technique in deep learning~\cite{srivastava2014dropout}, we use a dropout strategy for webly-labeled learning~\cite{liang2016learning}. It is implemented in the self-paced regularizer discussed in Section 3. With the dropout, the regularizers become:
\begin{equation}
\label{eq:dropout}
\vspace{-2mm}
\begin{split}
r_i(p) \sim \text{Bernoulli}(p) + \epsilon, (0 < \epsilon \ll 1)\\
f(\mathbf{v};\lambda, p) = \frac{1}{2} \lambda \sum_{i=1}^n (\frac{1}{r_i} v_i^2 - 2v_i),
\end{split}
\vspace{-2mm}
\end{equation}
where $\mathbf{r}$ is a column vector of independent Bernoulli random variables with the probability $p$ of being 1. Each of the element equals the addition of $r_i$ and a small positive constant $\epsilon$. Denote $\mathbb{E}_{\mathbf{w}} = \sum_{i=1}^n v_i \ell_i + f(\mathbf{v};\lambda)$ as the objective with the fixed model parameters $\mathbf{w}$ without any constraint, and the optimal solution $\mathbf{v}^* = [v_1^*, \cdots, v_n^*]^T= \argmin_{\mathbf{v}\in[0,1]^n} \mathbb{E}_{\mathbf{w}}$. We have:
\begin{equation}
\vspace{-2mm}
\label{eq:linear_dropout}
\begin{split}
\mathbb{E}_{\mathbf{w}} = \sum_{i=1}^n \ell_i v_i + \lambda ( \frac{1}{2r_i} v_i^2 - v_i );\\
\frac{\partial \mathbb{E}_{\mathbf{w}}}{\partial v_i} = \ell + \lambda v_i/r_i -\lambda = 0;\\
\Rightarrow v_i^* =\begin{cases}
r_i(-\frac{1}{\lambda} \ell_i +1) & \ell_{i} < \lambda\\
0 & \ell_{i} \ge \lambda
\end{cases}.
\end{split}
\vspace{-2mm}
\end{equation}
The dropout effect can be demonstrated in the closed-form solutions in Eq.~\eqref{eq:linear_dropout}: with the probability $1-p$, $v_i^*$ in both the equations approaches 0; with the probability $p$, $v_i^*$ approaches the solution of the plain regularizer discussed in Eq.~\eqref{eq:linear_scheme}. Recall the self-paced regularizer defines a scheme for learning samples. Eq.~\eqref{eq:linear_dropout} represent the new dropout learning scheme.
When the base learner is neural networks, the proposed dropout can be used combined with the classical dropout in~\cite{srivastava2014dropout}. The term dropout in this paper refers to dropping out samples in the iterative learning. By dropping out a sample, we drop out its update to the model parameter, which resembles the classical dropout used in neural networks. It operates on a more coarse-level which is useful for noisy data. When samples with incorrect noisy labels update a model, it will encourage the model to select more noisy labels. The dropout strategy prevents overfitting to noisy labels. It provides a way of combining many different sample subsets in different iterations in order to help avoid bad local minima. Experimental results substantiate this argument. In practice, we recommend setting two Bernoulli parameters for positive and negative samples on imbalanced data.
\section{Experiments}\label{sec:experiments}
In this section, we evaluate our method WELL for learning video detectors on noisy labeled data. The experiments are conducted on two major public benchmarks: FCVID and YFCC100M, where FCVID is by far one of the biggest manually annotated video dataset \cite{jiang2015exploiting}, and the YFCC100M dataset is the largest multimedia benchmark \cite{thomee2015yfcc100m}.
\subsection{Experimental Setup}
\textbf{Datasets, Features and Evaluation Metrics}
Fudan-columbia Video Dataset (FCVID) contains 91,223 YouTube videos (4,232 hours) from 239 categories. It covers a wide range of concepts like activities, objects, scenes, sports, DIY, etc. Detailed descriptions of the benchmark can be found in~\cite{jiang2015exploiting}. Each video is manually labeled to one or more categories. In our experiments, we do not use the manual labels in training, but instead we automatically generate the web labels according to the concept name appearance in the video metadata. The manual labels are used only in testing to evaluate our and the baseline methods. Following~\cite{jiang2015exploiting}, the standard train/test split is used.
The second set is YFCC100M \cite{thomee2015yfcc100m} which contains about 800,000 videos on Yahoo! Flickr with metadata such as the title, tags, the uploader, etc. There are no manual labels on this set and we automatically generate the curriculum from the metadata in a similar way. Since there are no annotations, we train the concept detectors on the most 101 frequent latent topics found in the metadata.
There are totally 47,397 webly labeled videos on the 101 concepts for training.
On FCVID, as the manual labels are available, the performance is evaluated in terms of the precision of the top 5 and 10 ranked videos (P@5 and P@10) and mean Average Precision (mAP) of 239 concepts. On YFCC100M, since there are no manual labels, for evaluation, we apply the detectors to a third public video collection called TRECVID MED which includes 32,000 Internet videos~\cite{over2014trecvid}. We apply the detectors trained on YFCC100M to the TRECVID videos and manually annotate the top 10 detected videos returned by each method for 101 concepts.
\textbf{Implementation Details}
We build our method on top of a pre-trained convolutional neural network as the low-level features (VGG network \cite{simonyan2014very}). We extract the key-frame level features and create a video feature by the average pooling. The same features are used across different methods on each dataset. The concept detectors are trained based on a hinge loss cost function. Algorithm~\ref{alg:overall} is used to train the concept models iteratively and the $\lambda$ stops increasing after 100 iterations. We automatically generate curriculum labels based on the video metadata, ASR, OCR and VGG net 1,000 classification results using latent topic modeling with word embedding matching as shown in Section 3, and derive a partial-order curriculum using the method discussed in Section 3.2.2.
\textbf{Baselines}
The proposed method is compared against the following five baseline methods which cover both the classical and the recent representative learning algorithms on webly-labeled data.
\textit{BatchTrain} trains a single SVM model using all samples in the multi-modal curriculum built as described in section 3.2.2 LT+WE.
\textit{Self-Paced Learning (SPL)} is a classical method where the curriculum is generated by the learner itself~\cite{kumar2010self}.
\textit{BabyLearning} is a recent method that simulates baby learning by starting with few training samples and fine-tuning using more weakly labeled videos crawled from the search engine \cite{liang2015towards}.
\textit{GoogleHNM} is a hard negative mining method proposed by Google \cite{varadarajan2015efficient}. It utilizes hard negative mining to train a second order mixture of experts model according to the video's YouTube topics. \textit{FastImage} \cite{han2015fast} is a video retrieval method that utilizes web images from search engine to match to the video with re-ranking. \textit{WELL} is the proposed method. The hyper-parameters of all methods including the baseline methods are tuned on the same validation set. On FCVID, the set is a standard development set with manual labels randomly selected from 10\% of the training set (No training was done using ground truth labels) whereas on YFCC100M it is also a 10\% proportion of noisy training set.
\subsection{Experiments on FCVID}
\textbf{Curriculum Comparison} As disscussed in Section 3.2.2, we compare different ways to build curriculum for noisy label learning.
Here we also compare their effectiveness by training concept detectors directly using the curriculum labels. The batch train model is used for all generated cirriculumn labels.
In Table \ref{exp-cur} we show the batch trained models' precision at 5, 10 and mean average precision on the test set of FCVID.
For LT+WE (Multi-modal), we extract curriculum from different modalities as shown in Section 3.2.2, and combine them using linear weights. The weights are hyper-parameters that are tuned on the validation set, and the optimal weights for textual metadata, ASR, image classification and OCR results are 1.0, 0.5, 0.5 and 0.05, respectively.
Results show that the curriculum generated by combining latent topic modeling and word embedding using multi-modal prior knowledge is the most accurate, which indicates our claim of exploiting multi-modal information is beneficial, and we use this method in WELL for the rest of the experiments.
\begin{table}[]
\centering
\caption{Curriculum BatchTrain Comparison}
\label{exp-cur}
\begin{tabular}{|l||c|c|c|}
\hline
Method & P@5 & P@10 & mAP \\ \hline \hline
ExactMatching & 0.730 & 0.713 & 0.419 \\
StemMatching & 0.782 & 0.763 & 0.469 \\
YouTubeTopicAPI & 0.587 & 0.563 & 0.315 \\
SearchEngine & 0.723 & 0.713 & 0.413 \\
WordEmbedding & 0.790 & 0.774 & 0.462 \\
LatentTopic & 0.731 & 0.716 & 0.409 \\
LT+WE & 0.804 & 0.795 & 0.473 \\
\textbf{LT+WE(Multi-modal)} & \textbf{0.838 } & \textbf{ 0.820 } & \textbf{ 0.486 } \\ \hline
\end{tabular}
\end{table}
\textbf{Baseline Comparison} Table~\ref{exps-basline} compares the precision and mAP of different methods where the best results are highlighted. As we see, the proposed WELL significantly outperforms all baseline methods, with statistically significant difference at $p$-level of 0.05. Comparing WELL with SPL, the effect of curriculum learning and dropout makes a significant difference in terms of performance, which suggests the importance of prior knowledge and preventing over-fitting in webly learning. The promising experimental results substantiate the efficacy of the proposed method.
\begin{table}[ht]
\centering
\footnotesize
\caption{Baseline comparison on FCVID}
\label{exps-basline}
\begin{tabular}{|l||c|c|c|c|c|c|}
\hline
Method & P@5 & P@10 & mAP \\ \hline \hline
BatchTrain & 0.838 & 0.820 & 0.486 \\
FastImage~\cite{han2015fast} &- &- & 0.284\\
SPL~\cite{kumar2011learning} & 0.793 & 0.754 & 0.414 \\
GoogleHNM~\cite{varadarajan2015efficient} & 0.781 & 0.757 & 0.472 \\
BabyLearning~\cite{liang2015towards} & 0.834 & 0.817 & 0.496 \\
\textbf{WELL} &\textbf{0.918}&\textbf{0.906} & \textbf{0.615} \\ \hline
\end{tabular}
\end{table}
\textbf{Robustness to Noise Comparison}
In this comparison we manually control the noisiness of the curriculum in order to systematically verify how our methods would perform with respect to the noisiness within the web data. The experimental results indicate the robustness of our method towards noisy labels. To this end, we randomly select video samples with ground truth labels for each concept, so that the precision of the curriculum labels are set at 20\%, 40\%, 60\%, 80\% and we fix the recall of all the labels. We then train WELL using such curriculum and test them on the FCVID testing set. We also compare WELL to three other methods with the same curriculum, among them \textit{GoogleHNM} is a recent method to train video concept detector with large-scale data. We exclude \textit{BabyLearning}, which relies on the returned results by the search engine, since in this experiment the curriculum is fixed . As shown in Table \ref{exp-noise}, as the noisiness of the curriculum grows (the precision drops), WELL maintains its performance while other methods drop significantly. Specifically, when the precision of the curriculum drops from 40\% to 20\%, other methods' mAP averagely drops 46.5\% while WELL's mAP only drops 19.1\% relatively. It shows that WELL is robust against different level of noise, which shows great potential in larger scale webly-labeled learning as the dataset gets bigger, the noisier it may become.
\begin{figure}[!ht]
\centering
\includegraphics[width=1.0\linewidth,height=55mm]{well-noise}
\caption{WELL performance with curriculum of different level of noisiness. p=k\% means the curriculum precision. The higher is k, the less noise is in the curriculum labels.}
\label{well-noise}
\end{figure}
\begin{table}[ht]
\centering
\footnotesize
\caption{WELL performance with curriculum of different level of noisiness. $p$ represents the precision of the curriculum. }
\vspace{1mm}
\label{exp-noise}
\begin{tabular}{|l||c|c|c|c|}
\hline
& p=20\% & p=40\% & p=60\% & p=80\% \\ \hline \hline
BatchTrain & 0.232 & 0.463 & 0.538 & 0.592 \\
SPL & 0.184 & 0.396 & 0.515 & 0.586 \\
GoogleHNM & 0.304 & 0.477 & 0.552 & 0.602 \\
\textbf{WELL} & \textbf{0.496} & \textbf{0.613} & \textbf{0.646} & \textbf{0.673} \\ \hline
\end{tabular}
\end{table}
\begin{table}[ht]
\centering
\footnotesize
\caption{WELL performance with curriculum of different level of noisiness. $p$ represents the precision of the curriculum. }
\vspace{1mm}
\label{exp-noise}
\begin{tabular}{|l||c|c|c|c|}
\hline
& p=80\% & p=60\% & p=40\% & p=20\% \\ \hline \hline
BatchTrain & 0.592& 0.538& 0.463& 0.232 \\
SPL & 0.586& 0.515& 0.396& 0.184 \\
GoogleHNM & 0.602& 0.552& 0.477& 0.304 \\
\textbf{WELL} & \textbf{0.673}& \textbf{0.646}& \textbf{0.613}& \textbf{0.496} \\ \hline
\end{tabular}
\end{table}
\textbf{Ground-truth Training Comparison}
In this part, we also compare our method with the state-of-the-art method trained using ground truth labels on FCVID (rDNN)~\cite{jiang2015exploiting}. We compare WELL trained using the static CNN features, the standard features provided by the authors~\cite{jiang2015exploiting}, and we also compare WELL using the late (average) fusion with CNN, motion and audio MFCC features (WELL-MM) to the method that achieves the best result on FCVID trained using the same multi-modal features. WELL-MM uses CNN, Motion and MFCC features, which is the same set of features as rDNN-F~\cite{jiang2015exploiting}. Noted that the state-of-the-art method uses the ground truth labels to train models, which includes 42,223 videos with manual labels, while our proposed method uses none of the human annotation into training but still be able to outperform one of the state-of-the-art results.
\begin{table}[ht]
\centering
\footnotesize
\caption{Ground-truth Training Comparison on FCVID. The methods with * are trained using human annotated labels. WELL-MM uses CNN, Motion and MFCC features, which is the same set of features as rDNN-F.}
\vspace{1mm}
\label{exps-sa}
\begin{tabular}{|l||c|c|c|c|c|c|}
\hline
Method & P@5 & P@10 & mAP \\ \hline \hline
WELL &\textbf{0.918}&\textbf{0.906} & \textbf{0.615} \\
Static CNN\cite{jiang2015exploiting}* &-&- & 0.638 \\
WELL-MM &\textbf{0.930}&\textbf{0.918} & \textbf{0.697} \\
rDNN-F\cite{jiang2015exploiting}* &-&- & 0.754 \\ \hline
\end{tabular}
\end{table}
\textbf{Noisy Dataset Size Comparison} To investigate the potential of concept learning on webly-labeled video data, we apply the methods on different sizes of subsets of the data. Specifically, we randomly split the FCVID training set into several subsets of 200, 500, 1,000, and 2,000 hours of videos, and train the models on each subset without using manual annotations. The models are then tested on the same test set. Table~\ref{exps-small} lists the average results of each type of subsets. As we see, the accuracy of WELL on webly-labeled data increases along with the growth of the size of noisy data while other webly learning methods' performance tend to be saturated.
Comparing to the methods trained using ground truth, In Table~\ref{exps-small}, WELL-MM trained using the whole dataset (2000h) outperforms rDNN-F(trained using manual labels) trained using around 1200h of data. And since the incremental performance increase of WELL-MM is close to linear, we conclude that with sufficient webly-labeled videos WELL-MM will be able to outperform the rDNN-F trained using 2000h of data, which is currently the largest manual labeled dataset.
\vspace{-3mm}
\begin{figure}[!ht]
\centering
\includegraphics[width=1.05\linewidth,height=70mm]{well-dataset}
\vspace{-7mm}
\caption{MAP comparison of models trained using web labels and ground-truth labels on different subsets of FCVID. The methods with * are trained using human annotated labels.}
\label{well-dataset}
\end{figure}
\begin{table}[ht]
\centering
\footnotesize
\caption{MAP comparison of models trained using web labels and ground-truth labels on different subsets of FCVID. The methods with * are trained using human annotated labels. Noted some of the numbers from \cite{jiang2015exploiting} are approximated from graphs.}
\vspace{1mm}
\label{exps-small}
\setlength{\tabcolsep}{4pt}
\renewcommand{\arraystretch}{1.1}
\begin{tabular}{|l||c|c|c|c|}
\hline
Dataset Size & 200h & 500h & 1000h & 2000h \\ \hline
BatchTrain & 0.364 & 0.422 & 0.452 & 0.486 \\
SPL~\cite{kumar2011learning} & 0.327 &0.379 &0.403 &0.414 \\
GoogleHNM~\cite{varadarajan2015efficient} & 0.361 & 0.421 &0.451 &0.472 \\
BabyLearning~\cite{liang2015towards} &0.390 & 0.447 & 0.481 & 0.496 \\
\textbf{WELL-MM} & \textbf{0.541} & \textbf{0.616} & \textbf{0.632} & \textbf{0.697} \\ \hline
Static CNN\cite{jiang2015exploiting}* & 0.485 & 0.561 & 0.604 & 0.638 \\
rDNN-F\cite{jiang2015exploiting}* & 0.550 & 0.620 & 0.650 & 0.754\\ \hline
\end{tabular}
\end{table}
\vspace{-3mm}
\subsection{Experiments on YFCC100M}
In the experiments on YFCC100M, we train 101 concept detectors on YFCC100M and test them on the TRECVID MED dataset which includes 32,000 Internet videos. Since there are no manual labels, to evaluate the performance, we manually annotate the top 10 videos in the test set and report their precisions in Table~\ref{exps-yfcc}. The MED evaluation is done by four annotators and the final results are averaged from all annotations. The Fleiss' Kappa value for these four annotators is 0.64.
A similar pattern can be observed where the comparisons substantiate the rationality of the proposed webly learning framework. Besides, the promising results on the largest multimedia set YFCC100M verify the scalability of the proposed method.
\vspace{-3mm}
\begin{table}[ht]
\centering
\footnotesize
\caption{Baseline comparison on YFCC100M}
\label{exps-yfcc}
\begin{tabular}{|l||c|c|c|c|c|c|}
\hline
Method & P@3 & P@5 & P@10\\ \hline \hline
BatchTrain & 0.535 & 0.513 & 0.487 \\
SPL~\cite{kumar2011learning} &0.485 & 0.463 & 0.454\\
GoogleHNM~\cite{varadarajan2015efficient} &0.541 & 0.525 & 0.500 \\
BabyLearning~\cite{liang2015towards} &0.548 & 0.519 & 0.466 \\
\textbf{WELL} &\textbf{0.667} & \textbf{0.663} & \textbf{0.649}\\ \hline
\end{tabular}
\end{table}
\begin{figure*}
\centering
\includegraphics[width=1.0\textwidth]{pick-all}
\vspace{-7mm}
\caption{WELL's example picks for different iterations}
\vspace{-5mm}
\label{well-pick-all}
\end{figure*}
\vspace{-4mm}
\subsection{Time Complexity Comparison}
The computation complexity of WELL is comparable to existing methods. For a single class, the complexity for our model and baseline models is $O(r \times n \times m)$, where $r$ is the number of iterations to converge, $n$ is the number of training samples and $m$ is the feature dimension. Theoretically, the complexity is comparable to the baseline models that have different $r$. In practice, on a 40 core-CPU machine, WELL and SPL takes 7 hours to converge (100 iterations) on FCVID with 239 concepts, whereas GoogleHNM and BabyLearning take around 5 hours.
In Table \ref{time}, we show the theoretical and actual run time for all methods.
\begin{table}[ht]
\centering
\caption{Runtime comparison across different methods. We report the time complexity on different method as well as their actual run time on FCVID (in hours).}
\label{time}
\begin{tabular}{|l||c|c|c|}
\hline
Method & Complexity & FCVID(h) \\ \hline \hline
BatchTrain & $O(n \times m)$ & 2.0 \\
SPL & $O(r \times n \times m)$ & 7.0 \\
GoogleHNM & $O(r \times n \times m)$ & 5.0 \\
BabyLearning & $O(r \times n \times m)$ & 5.0 \\
WELL & $O(r \times n \times m)$ & 7.0\\ \hline
\end{tabular}
\end{table}
\vspace{-3mm}
\subsection{Qualitative Analysis}
In this section we show training examples of WELL. In Figure~\ref{well-pick-all}, we demonstrate the positive samples that WELL select at different stage of training the concept "baseball", "forest", "birthday" and "cow".
For the concept "baseball", at early stage (1/93, 25/93), WELL selects easier and clearer samples such as those camera directly pointing at the playground, while at later stage (75/93, 93/93) WELL starts to train with harder samples with different lighting conditions and untypical samples for the concept.
For the concept "birthday", as we see, at later stage of the training, complex samples for birthday event like a video with two girl singing birthday song (75/84) and a video of celebrating birthday during hiking (84/84) are included in the training.
For the concept "forest", at the final iteration (95/95), a video of a man playing nunchaku is included, as the video title contains "Air Forester Nunchaku Freestyle" and it is included in the curriculum, which is reasonable as the curriculum is noisy and may contain false positive. Since WELL is able to leave outliers in later stage of the training, the affection of the false positives in the curriculum can be alleviated by early stopping. In our experiments, we stop increasing $\lambda$ after 100 iterations so in the final model, only a subset of samples are used.
\vspace{-2mm}
\section{Conclusions}
In this paper, we proposed a novel method called WELL for webly labeled video data learning. WELL extracts multi-modal informative knowledge from noisy weakly labeled video data from the web through a general framework with solid theoretical justifications.
WELL achieves the best performance only using webly-labeled data on two major video datasets. The comprehensive experimental results demonstrate that WELL outperforms state-of-the-art studies by a statically significant margin on learning concepts from noisy web video data. In addition, the results also verify that WELL is robust to the level of noisiness in the video data. The result suggests that with more webly-labeled data, which is not hard to obtain, WELL can potentially outperform models trained on any existing manually-labeled data.
\vspace{-3mm}
\bibliographystyle{abbrv}
\linespread{0.9}
\scriptsize
\section{Introduction}
Nowadays, millions of videos are being uploaded to the Internet every day. These videos capture all aspects of multimedia content about their uploader's daily life. These explosively growing user generated content videos online are becoming an crucial source of video data. Automatically categorizing videos into concepts, such as people actions, objects, etc., has become an important research topic. Recently many work have been proposed to tackle with building concept detectors both in image domain and video domain \cite{deng2009imagenet,liang2015towards,tang2012shifting,karpathy2014large,jiang2015exploiting}. However, the need for manual labels by human annotators has become one of the major important limitations for large-scale concept learning. It is even more so in video domain, since training concept detectors on videos is more challenging than on still images.
Many image datasets such as ImageNet~\cite{deng2009imagenet}, CIFAR~\cite{krizhevsky2009learning}, PASCAL VOC\cite{everingham2010pascal}, MS COCO~\cite{lin2014microsoft} and Caltech~\cite{fei2006one} have been collected and manually labeled. In video domain, some largest datasets such as UCF-101\cite{soomro2012ucf101}, MCG-WEBV~\cite{cao2009mcg}, TRECVID MED~\cite{over2014trecvid} and FCVID~\cite{jiang2015exploiting} are popular benchmark datasets for video classification.
Collecting such datasets requires a large amount of human effort that can take thousands of man hours. In addition, manually labeling video requires playing back the video, which is more time consuming and expensive than labeling still images. As a result, the largest labeled video collection, FCVID~\cite{jiang2015exploiting}, only contains about 0.09 million labels with 239 concept classes, much less than the 14 million labels with over 20,000 classes in the image collection ImageNet~\cite{deng2009imagenet}.
Many state-of-the-art models in visual classification are based on the neural networks~\cite{jiang2015exploiting,karpathy2014large,varadarajan2015efficient}. As the architecture gets deeper, the neural network would need more data to train in order to get better performance. However, more data needs more human supervision which are more expensive to acquire in the video domain.
Videos are available on the web and contain rich contextual information with a weak annotation about their content, such as their titles, descriptions and surrounding text. These webly-labeled data are orders of magnitude larger than that of any manually-labeled collections. Moreover, automatically extracted features from multiple modalities such as existing still image classification models, automatic speech recognition and optical character recognition tools can be useful additional information for the content of the video.
Figure \ref{cur-example} shows an example of webly-labeled video for walking with a dog. As we see, the textual metadata we get from the web videos contain useful but very noisy information. The multi-modal prior information we get is correlated across modalities, as the image classification results and speech transcript show high probability of dog appearance, while the textual metadata indicates the same content. Some of the videos (about 20\% in the FCVID dataset) have very little textual metadata and we can only obtain web labels via other modalities.
To address the problem of learning detectors from the big web data, in this paper, we utilize multi-modal information to harness prior knowledge from the web video without any manual efforts.
Existing methods on learning from noisy webly-labeled data has mainly focused on the image domain \cite{fergus2005learning,li2010optimol,bergamo2010exploiting,chen2015webly}. Existing studies demonstrated promising results in this direction. However, these methods are primarily based on some heuristic methods. It is not clear what objective is being optimized and where or even whether the learning process will converge. Moreover, these methods only utilize a single text modality in the image domain. It is unclear how to exploit the multi-modal prior knowledge for concept learning from the rich context of Internet video.
To utilize the large amount of webly-labeled video data for concept learning, we propose a learning framework called \textbf{WEbly-Labeled Learning (WELL)}. It is established on the theories called \textit{curriculum learning}~\cite{bengio2009curriculum} and \textit{self-paced learning}~\cite{kumar2010self}. The learning framework is motivated by human learning, where people generally start learning easier aspects of a concept, and then gradually take more complex examples into the learning process\cite{bengio2009curriculum,kumar2010self,jiang2015self}.
Following this idea, WELL learns a concept detector iteratively from first using a few samples with more confident labels (more related to the concept), then gradually incorporate more video samples with noisier labels. The algorithm combines the prior knowledge, called learning curriculum, extracted from the webly-labeled data with the dynamic information learned from the statistical model (self-paced) to determine which video samples to learn in the next iteration. This idea of easy-to-hard learning paradigm has been adopted for learning in noisy web data~\cite{jiang2014easy,chen2015webly,kumar2010self} and has been proved to be efficient to deal with noise and outliers.
Our proposed method generalizes such learning paradigm using a clear objective function. It is proved to be convex and is a also general framework that can incorporate state-of-the-art deep learning methods to learn robust detectors from noisy data. Our framework fundamentally changes self-paced learning and allows learning for video concept detectors at unlimited scale.
Figure~\ref{well-cnn} shows the architecture of the proposed method. We extract keyframe-level convolutional neural network features and feed them into WELL layer with average pooling and iterative learning process. We have also tried using other video features such as motion features and audio MFCC features.
Our contributions are threefold. First, we address the problem of learning robust video concept detectors from noisy web data through a general framework with solid theoretical justifications. We show that WELL not only outperforms state-of-the-art learning methods on noisy labels, but also, notably, achieves comparable results with state-of-the-art models trained using manual annotation on one of the largest video dataset. Second, we provide detailed comparison of different approaches to exploit multi-modal curriculum from noisy labels and verify that our method is robust against certain level of noisiness in the video data. Finally, the efficacy and the scalability have been empirically demonstrated on two public benchmarks, including by far the largest manually-labeled video set called FCVID~\cite{jiang2015exploiting} and the largest multimedia dataset called YFCC100M~\cite{thomee2015yfcc100m}. The promising results suggest that detectors trained on sufficient webly-labeled videos may outperform detectors trained on any existing manually-labeled datasets.
\begin{figure*}
\centering
\includegraphics[width=0.9\textwidth,height=1.5in]{well-cnn}
\vspace{-6mm}
\caption{\textbf{Learning architecture of WEbly-Labeled Learning (WELL)}}
\label{well-cnn}
\end{figure*}
\section{Related Work}
\textbf{Curriculum and Self-paced Learning}: Recently a learning paradigm called~\textit{curriculum learning} (CL) was proposed by Bengio et al., in which a model is learned by gradually incorporating from easy to complex samples in training so as to increase the entropy of training samples~\cite{bengio2009curriculum}. A curriculum determines a sequence of training samples and is often derived by predetermined heuristics in particular problems.
For example, Chen et al. designed a curriculum where images with clean backgrounds are learned before the images with noisy backgrounds~\cite{chen2015webly} , i.e. their method first builds a feature representation by a Convolutional Neural Network (CNN) on images with clean background and then they fine tune the models on images with noisy background.
In~\cite{spitkovsky2009baby}, the authors approached grammar induction, where the curriculum is derived in terms of the length of a sentence. Because the number of possible solutions grows exponentially with the length of the sentence, and short sentences are easier and thus should be learn earlier.
The heuristic knowledge in a problem often proves to be useful. However, the curriculum design may lead to inconsistency between the fixed curriculum and the dynamically learned models. That is, the curriculum is predetermined a prior and cannot be adjusted accordingly, taking into account the feedback about the learner. To alleviate the issue of CL, Kumar et al. designed a learning paradigm, called \emph{self-paced learning} (SPL)~\cite{kumar2010self}. SPL embeds curriculum design as a regularizer into the learning objective. Compared with CL, SPL exhibits two advantages: first, it jointly optimizes the learning objective with the curriculum, and thus the curriculum and the learned model are consistent under the same optimization problem; second, the learning is controlled by a regularizer which is independent of the loss function in specific problems. This theory has been successfully applied to various applications, such as matrix factorization~\cite{zhao2015self}, action/event detection~\cite{jiang2014self}, domain adaption~\cite{tang2012shifting}, tracking~\cite{supancic2013self} and segmentation~\cite{kumar2011learning}, reranking~\cite{jiang2014easy}, etc.
\textbf{Learning Detectors in Web Data}: Many recent studies have been proposed to utilize the large amount of noisy data from the Internet. For example, \cite{mitchell2015never} proposed a Never-Ending Language Learning (NELL) paradigm and built adaptive learners that makes use of the web data by learning different types of knowledge and beliefs continuously. Such learning process is mostly self-supervised, and previously learned knowledge enables learning further types of knowledge.
In the image domain, existing methods try to tackle the problem of constructing qualified training sets based on the search results of text or image search engines~\cite{fergus2005learning,li2007optimol,chen2013neil,li2014exploiting,divvala2014learning,liang2015towards}.
For example, \cite{fergus2005learning} extended the probabilistic Latent Semantic Analysis in visual domain and learned object categories using results from image search engines. \cite{li2007optimol} proposed an incremental learning paradigm that initialized from a few seed images and repeatedly trained models to refine the collected image dataset from the Internet.
NEIL~\cite{chen2013neil} followed the idea of NELL and learned from web images to form a large collection of concept detectors iteratively via a semi-supervised fashion. By combining the classifiers and the inter-concept relationships it learned, NEIL can be used for scene classification and object detection task.
\cite{li2014exploiting} tried to learn robust classifiers by considering the noisy textual information accompanied with web images. However, the down side is that the portion of the true positive samples has to be determined via prior knowledge, where in fact it is not accurate to assume the same number of true positive samples for any targeted concept.
\cite{divvala2014learning} introduced a webly-supervised visual concept learning method that automatically learns large amount of models for a wide range of variations within visual concepts. They discovered concept variances through vocabulary of online books, and then downloaded images based on text-search from the web to train object detection and localization models.
\cite{liang2015towards} presented a weakly-supervised method called Baby Learning for object detection from a few training images and videos. They first embed the prior knowledge into a pre-trained CNN. When given very few samples for a new concept, a simple detector is constructed to discover much more training instances from the online weakly labeled videos. As more training samples are selected, the concept detector keeps refining until a mature detector is formed.
Another recent work in image domain \cite{chen2015webly} proposed a webly supervised learning of Convolutional Neural Network. They utilized easy images from search engine like Google to bootstrap a first-stage network and then used noisier images from photo-sharing websites like Flickr to train an enhanced model.
In video domain, only few studies \cite{duan2012visual,han2015fast,varadarajan2015efficient} have been proposed for noisy data learning since training robust video concept detectors is more challenging than the problem in the image domain.
\cite{duan2012visual} tackled visual event detection problem by using SVM based domain adaptation method in web video data.
\cite{han2015fast} described a fast automatic video retrieval method using web images. Given a targeted concept, compact representations of web images obtained from search engines like Google, Flickr are calculated and matched to compact features of videos. Such method can be utilized without any pre-defined concepts.
\cite{varadarajan2015efficient} discussed a method that exploits the YouTube API to train large scale video concept detectors on YouTube. The method utilized a calibration process and hard negative mining to train a second order mixture of experts model in order to discover correlations within the labels.
Most of the existing methods are heuristic approaches as it is unclear what objective is being optimizing on the noisy data. Moreover, results obtained from the web search results is just one approach to acquire prior knowledge or curriculum. To the best of our knowledge, there have been no systematical studies on exploiting the multi-modal prior knowledge in video concept learning on noisy data. Since search engine algorithm is changing rapidly, it is unclear that how noisy the web labels are and how the level of noisiness in the data will affect performance. In this paper, we proposed a theoretically justified method with clear framework for curriculum constructing and model learning. We also empirically demonstrate its superior performance over representative existing methods and systemically verify that WELL is robust against the level of noisiness of the video data.
\vspace{-1mm}
\section{WEbly-Labeled Learning (WELL)}
\subsection{Problem Description}
In this paper, following~\cite{varadarajan2015efficient}, we consider a concept detector as a classifier and our goal is to train concept detectors from webly-labeled video data without any manual labeling effort. Given a noisy web video training set and a target concept set, we do not assume any distribution of the noise. Formally, we represent the training set as $\mathcal{D} = \{(\mathbf{x}_i,\mathbf{z}_i,\mathbf{\tilde{y}}_i)\}_{i=1}^n$
where $\mathbf{x}_i \in \mathbb{R}^m$ denotes the feature for the $i^{th}$ observed sample, and $\mathbf{z}_i$ represents its noisy web label, which generally means the prior knowledge we can get from the web without additional human effort that not only includes textual information provided by the uploaders in the video metadata but also includes prior knowledge from other modalities using existing tools like pre-trained Convolutional Neural Network image detector \cite{chatfield2014return}, Automatic Speech Recognition \cite{povey2011kaldi} and Optical character recognition \cite{smith2007overview}. The $\mathbf{\tilde{y}}_i \subset \mathcal{Y}$ is the inferred concept label set for the $i^{th}$ observed sample based on its noisy web label, and $\mathcal{Y}$ denotes the full set of target concepts. In our experiment, to simplify the problem, we apply our method on binary classification and infer binary labels $\tilde{y}_i$ from the noisy web labels.
The noisy web labels can be used to automatically infer concept labels by matching the concept name to the video textual metadata. For example, a video may be inferred to the concept label ``cat'' as its textual title contains cat. \cite{varadarajan2015efficient} utilizes the YouTube topic API, which is derived from the textual metadata, to automatically get concept labels for videos. The web labels are quite noisy as the webly-labeled concepts may not present in the video content whereas the concepts not in the web label may well appear.
\subsection{Model and Algorithm}
\subsubsection{Objective Function}
To leverage the noisy web labels in a principled way, we propose WEbly-Labeled Learning (WELL). Formally, given a training set $\mathcal{D}$ as described before, Let $L(\tilde{y}_i,g(\mathbf{x}_i,\mathbf{w}))$, or $\ell_i$ for short, denote the loss function which calculates the cost between the inferred label $\tilde{y}_i$ and the estimated label $g(\mathbf{x}_i,\mathbf{w})$. Here $\mathbf{w}$ represents the model parameter inside the decision function $g$.
For example, in our paper, $\mathbf{w}$ represents the weight parameters in the Convolutional Neural Network (CNN) and the Support Vector Machine (SVM).
Our objective function is to jointly learn the model parameter $\mathbf{w}$ and the latent weight variable $\mathbf{v}= [v_1,\cdots,v_n]^T$ by:
\vspace{-2mm}
\begin{equation}
\label{eq:spcl_obj}
\begin{split}
\!\min_{\mathbf{w},\mathbf{v}\in \lbrack 0,\!1]^{n}}\!\!\mathbb{E}(\mathbf{w},\!\mathbf{v}\!;\lambda,\!\Psi\!)
\!=\! \sum_{i=1}^n v_i L(\tilde{y}_i,\!g(\mathbf{x}_i,\!\mathbf{w})) \!+\! f(\mathbf{v};\! \lambda), \\ \text{ subject to } \mathbf{v} \in \Psi
\end{split}
\end{equation}
where $\mathbf{v=[}v_{1},v_{2},\cdots ,v_{n}\mathbf{]}^{T}$ denote the
latent weight variables reflecting the inferred labels' confidence. The weights determine a learning sequence of samples, where samples with greater weights tend to be learned earlier.
Our goal is to assign greater weights to the samples with more confident labels whereas smaller or zero weights to the samples with noisy labels. To this end, we employ the self-paced regularizer $f$, which controls the learning process of the model.
We consider the linear regularizer Eq.~\eqref{eq:linear_scheme} proposed in~\cite{jiang2015self}:
\begin{equation}
\label{eq:linear_scheme}
f(\mathbf{v};\lambda) = \frac{1}{2} \lambda \sum_{i=1}^n ( v_i^2 - 2v_i ).
\vspace{-1mm}
\end{equation}
Generally, a self-paced regularizer determines the scheme for penalizing the latent weight variables. Physically it resembles the learning schemes human used in understanding new concepts. The linear scheme corresponds to a prudent strategy, which linearly penalizes the samples that are different to what the model has already learned (see Eq.~\eqref{eq:linear_closedform}).
The hyper-parameter $\lambda$ $(\lambda > 0)$ is called ``model age'', which controls the pace at which the model learns new samples. When $\lambda$ is small only samples of with small loss will be considered. As $\lambda$ grows, more samples with larger loss will be gradually appended to train a ``mature'' mode.
$\Psi$ in Eq.~\eqref{eq:spcl_obj} is a curriculum region derived from noisy web labels $\mathbf{z}$ that incorporates the prior knowledge extracted from the webly-labeled data as a convex feasible region for the weight variables. The shape of the region weakly implies a prior learning sequence of samples, where the expected values for favored samples are larger. The curriculum region can be derived in a variety of ways that make use of different modalities. We will discuss this topic in details in following section. A straightforward approach is by counting the term frequency in the video's textual metadata. That is, for example, the chance of a video containing the concept ``cat'' become higher when it has more word ``cat'' in its title, description or tags.
Eq.~\eqref{eq:spcl_obj} represents a concise and general optimization model ~\cite{jiang2015self}. It combines the prior knowledge extracted from the noisy webly-labeled data (as the curriculum region) and the information dynamically learned during the training (via the self-paced regularizer). Intuitively, the prior knowledge serves as an instructor providing a guidance on learning the latent weights, but it leaves certain freedom for the model (the student) to adjust the actual weights according to its learning pace. Experimental results in Section~\ref{sec:experiments} demonstrate the learning paradigm can better overcome the noisy labels than heuristic approaches. Figure \ref{well-cnn} shows the learning process of our method.
Following~\cite{kumar2010self,jiang2015self}, we employ the alternative convex search algorithm to solve Eq.~\eqref{eq:spcl_obj}. Algorithm~\ref{alg:overall} takes the input of a curriculum region, an instantiated self-paced regularizer and a step size parameter; it outputs an optimal model parameter $\mathbf{w}$. First of all, it initializes the latent weight variables in the feasible region. Then it alternates between two steps until it finally converges: Step 3 learns the optimal model parameter with the fixed and most recent $\mathbf{v}^*$; Step 5 learns the optimal weight variables with the fixed $\mathbf{w}^*$. In the beginning, the model ``age'' is gradually increased so that more noisy samples will be gradually incorporated in the training. Step 3 can be conveniently implemented by existing off-the-shelf supervised learning methods such as the back propagation. Gradient-based methods can be used to solve the convex optimization problem in Step 4. According to~\cite{gorski2007biconvex}, the alternative search in Algorithm~\ref{alg:overall} converges as the objective function is monotonically decreasing and is bounded from below.
\setlength{\textfloatsep}{1pt}
\vspace{-3mm}
\IncMargin{1em}
\begin{algorithm}
\SetKwData{Left}{left}\SetKwData{This}{this}\SetKwData{Up}{up}
\SetKwInOut{Input}{input}\SetKwInOut{Output}{output}
\LinesNumbered
\Input{Input dataset $\mathcal{D}$, curriculum region $\Psi$, self-paced function $f$ and a step size $\mu$}
\Output{Model parameter $\mathbf{w}$}
\BlankLine
Initialize $\mathbf{v}^*$, $\lambda$ in the curriculum region\;
\While{not converged} {
Update $\mathbf{w}^* = \arg\min_{\mathbf{w}} \mathbb{E}(\mathbf{w},\mathbf{v}^*;\lambda, \Psi)$\;
Update $\mathbf{v}^* = \arg\min_{\mathbf{v}} \mathbb{E}(\mathbf{w}^*,\mathbf{v}; \lambda, \Psi)$\;
\lIf{$\lambda$ is small}{increase $\lambda$ by the step size $\mu$}
}
\Return $\mathbf{w}^*$
\caption{\label{alg:overall} WEbly-Labeled Learning (WELL).}
\end{algorithm}
\DecMargin{1em}
\vspace{-3mm}
At an early age when $\lambda$ is small, Step 4 in Algorithm~\ref{alg:overall} has an evident suppressing effect over noisy samples that have greater loss to the already learned model. For example, with a fixed $\mathbf{w}$, the unconstrained close-formed solution for the regularizer in Eq.~\eqref{eq:linear_scheme} equals
\begin{equation}
\label{eq:linear_closedform}
v_i^* =\begin{cases}
-\frac{1}{\lambda} \ell_i +1 & \ell_{i} < \lambda\\
0 & \ell_{i} \ge \lambda
\end{cases},
\end{equation}
where $v_i$ represents the $i$th element in the optimal solution $\mathbf{v}^* = [v_1^*, \cdots, v_n^*]^T$. Eq.~\eqref{eq:linear_closedform} called linear regularizer indicates the latent weight is proportional to the negative sample loss, and the sample whose loss is greater or equals to $\lambda$ will have zero weights and thus will not affect the training of the next model. As the model age grows, the hyper-parameter $\lambda$ increases, and more noisy samples will be used into training. The prior knowledge embedded in the curriculum region $\Psi$ is useful as it suggests a learning sequence of samples for the ``immature'' model. \cite{meng2015objective} theoretically proves that the iterative learning process is identical to optimizing a robust loss function on the noisy data.
If we keep increasing $\lambda$, the model will ultimately use every sample in the noisy data, which is undesirable as the labels of some noisy samples are bound to be incorrect. To this end, we stop increasing the age $\lambda$ after about a certain number of iterations (early stopping). The exact stopping iteration for each detector is automatically tuned in terms of its performance on a small validation set.
\begin{figure*}
\centering
\includegraphics[width=0.8\textwidth,height=0.3\textheight]{curriculum_example}
\caption{Curriculum Extraction Example. We automatically extract information using meaningful prior knowledge from several modalities and fuse them to get curriculum for WELL. Our method makes use of text, speech, visual cues while common methods like search engine only extract from textual information.}
\label{curriculum_example}
\end{figure*}
\subsubsection{Model Details}\label{sec:partial_curriculum}
In this section we discuss further details of the curriculum region $\Psi$ and the self-paced regularizer $f(\mathbf{v};\lambda)$.
$\Psi$ is a feasible region that embeds the prior knowledge extracted from the webly-labeled data. It physically corresponds to a convex search region for the latent weight variable. Given a set of training samples $\mathbf{X}=\{\mathbf{x}_{i}\}_{i=1}^{n}$,
we utilize the partial-order curriculum which generalizes the total-order curriculum by incorporating the incomplete prior over groups of samples. Samples in the confident groups should be learned earlier than samples in the less confident groups. It imposes no prior over the samples within the same group nor the samples not in any group. Formally, we define a partial order relation $\preceq$ such that $x_i \preceq x_j$ indicates that the sample $x_i$ should be learned no later than $x_j$ ($i,j \in [1,n]$). Similarly given two sample subsets $\mathbf{X}_{a} \preceq \mathbf{X}_{b}$ denotes the samples in $\mathbf{X}_{a}$ should be learned no later than the samples in $\mathbf{X}_{b}$.
In our problem, we extract the partial-order curriculum in the webly-labeled data in the following ways: we only distinguish the training order for groups of samples.
Information from different modalities of the web labels can be used for curriculum design. A straightforward way is to directly utilize the textual descriptions of the videos generated by the uploaders. We compare common ways to extract curriculum from web data for concept learning to the proposed novel method that utilize state-of-the-art topic modeling techniques in natural language processing.
In the following methods (Exact \& Stem Matching, Word Embedding and Latent Topic with Word Embedding), we first extract bag-of-words features from different modalities and then match them using specific matching methods to the concept words. Each video will then come with a matching score to each concept. In our experiment, we divide the data into two partial-order curriculum groups, where the videos with matching scores larger than zero will be in one group while others will be in the other group.
\textbf{Exact \& Stem Matching} We build curriculum directly using exact word matching or stemmed word matching between the textual metadata of the noisy videos to the targeted concept names.
\textbf{YouTubeTopicAPI} We directly utilize the YouTube topic API to search for videos that are related to the concept words. The topic API utilizes textual information of the uploaded videos to obtain related topics of the videos from Freebase.
\textbf{SearchEngine} We build curriculum using the search result from a text-based search engine. It is similar to related web-search based methods.
\textbf{Word Embedding} We use word embedding \cite{mikolov2013distributed} to match words in metadata to targeted concept words in order to deal with synonyms and related concepts. The word embedding is trained using Google News data.
\textbf{Latent Topic} We build curriculum based on the latent topic we learned from the noisy label. We incorporate Latent Dirichlet Allocation (LDA)~\cite{blei2003latent} to determine how each noisy labeled video is related to each target concept. The basic idea is that each web video consists of mixtures of topics (concepts), and each topic is characterized by a distribution of words.
Formally, given all the noisy information extracted from a web video and collected them as a document $\mathbf{d_i}$, which combines into a corpus $\mathbf{d}$, we have a target set of $\mathbf{k}$ topics, then the key inferential problem that we are going to solve is that of computing the posterior distribution of the latent topics given a corpus (how likely the videos are related to each target concept given the noisy information):
\begin{equation}
\label{eq:lda}
p(\theta,\mathbf{t}|\mathbf{d},\alpha,\beta) = \frac{p(\theta,\mathbf{t},\mathbf{d}|\alpha,\beta)}{p(\mathbf{d}|\alpha,\beta)}
\end{equation}
where $\theta$ is the topic distribution variable for the corpus, $\theta$ $\sim$ Dir($\alpha$), in which Dir($\alpha$) represents a uniform Dirichlet distribution with scaling parameter $\alpha$. The $\mathbf{t}$ is the topic assignment variable that indicates which topic each word belong to in the document.
$\beta$ is the Dirichlet prior on the per-topic word distribution, in which we impose asymmetric priors over the word distribution so that each learned topic will be seeded with particular words in our target concept. For example, a topic will be seeded with words "walk, dog" for the target concept "WalkingWithDog".
The parameter estimation in Eq~\eqref{eq:lda} can be done via Bayes methods.
However, Eq.~\eqref{eq:lda} is intractable to compute since $p(\mathbf{d}|\alpha,\beta)$ is intractable due to the coupling between $\theta$ and $\beta$ \cite{blei2003latent}. To solve this problem, approximate inference algorithms are introduced and we use the online variational inference algorithm from \cite{hoffman2010online}. The true posterior is approximated by a simpler distribution:
\begin{equation}
\label{eq:lda-vi}
q(\theta,\mathbf{t}|\gamma,\phi) = q(\theta|\gamma)\prod_{i=1}^{N}{q(t_i|\phi_i)}
\end{equation}
where the Dirichlet parameter $\gamma$ and the multinomial parameters ($\phi_1,...,\phi_N$) are the free variational parameters. Thus the maximization problem is equivalent to minimizing the Kullback-Leibler(KL) divergence between $q(\theta,\mathbf{t}|\gamma,\phi)$ and the posterior $p(\theta,\mathbf{t}|\mathbf{d},\alpha,\beta)$ \cite{blei2003latent}. The optimization problem can then be solve using Expectation-Maximization algorithm \cite{hoffman2010online}. The estimated parameters $\theta,t$ in Eq.~\eqref{eq:lda} is then used for constructing curriculum region in Eq.~\eqref{eq:spcl_obj}.
\textbf{Latent Topic with Word Embedding (LT+WE)} We first learn latent topics using LDA to replace the concept words with a topic word distribution and then match the latent topic words to the web label's bag-of-words features by using the word embeddings. We compare this method to the others using only textual information from the web videos.
We also use this method to get curriculum from other modalities such as Automatic Speech Recognition (ASR) \cite{povey2011kaldi}, Optical Character Recognition (OCR) \cite{smith2007overview} and basic image detector pre-trained on still images~\cite{ILSVRC15} (in this paper we use VGG net~\cite{simonyan2014very}, extract keyframe-level image classification results and average them to get video-level results.).
We extract bag-of-word features from them and combine them with linear weights.
Detailed fusion experiments can be found in Section \ref{sec:experiments}. We empirically set OCR's weight to be small as the results are much noisier than other features.
Figure~\ref{curriculum_example} shows an example of the noisy web video data and how the curriculum is extracted with different methods. Our method can utilize information from different modalities while common methods like search engine only consider textual information.
We compare the performance of different ways of curriculum design by training detectors directly in Section 4.
The labels in webly-labeled data are much noisier than manually-labeled data, and as a result, we found that the learning is prone to overfitting the noisy labels. To address this issue, inspired by the dropout technique in deep learning~\cite{srivastava2014dropout}, we use a dropout strategy for webly-labeled learning~\cite{liang2016learning}. It is implemented in the self-paced regularizer discussed in Section 3. With the dropout, the regularizers become:
\begin{equation}
\label{eq:dropout}
\vspace{-2mm}
\begin{split}
r_i(p) \sim \text{Bernoulli}(p) + \epsilon, (0 < \epsilon \ll 1)\\
f(\mathbf{v};\lambda, p) = \frac{1}{2} \lambda \sum_{i=1}^n (\frac{1}{r_i} v_i^2 - 2v_i),
\end{split}
\vspace{-2mm}
\end{equation}
where $\mathbf{r}$ is a column vector of independent Bernoulli random variables with the probability $p$ of being 1. Each of the element equals the addition of $r_i$ and a small positive constant $\epsilon$. Denote $\mathbb{E}_{\mathbf{w}} = \sum_{i=1}^n v_i \ell_i + f(\mathbf{v};\lambda)$ as the objective with the fixed model parameters $\mathbf{w}$ without any constraint, and the optimal solution $\mathbf{v}^* = [v_1^*, \cdots, v_n^*]^T= \argmin_{\mathbf{v}\in[0,1]^n} \mathbb{E}_{\mathbf{w}}$. We have:
\begin{equation}
\vspace{-2mm}
\label{eq:linear_dropout}
\begin{split}
\mathbb{E}_{\mathbf{w}} = \sum_{i=1}^n \ell_i v_i + \lambda ( \frac{1}{2r_i} v_i^2 - v_i );\\
\frac{\partial \mathbb{E}_{\mathbf{w}}}{\partial v_i} = \ell + \lambda v_i/r_i -\lambda = 0;\\
\Rightarrow v_i^* =\begin{cases}
r_i(-\frac{1}{\lambda} \ell_i +1) & \ell_{i} < \lambda\\
0 & \ell_{i} \ge \lambda
\end{cases}.
\end{split}
\vspace{-2mm}
\end{equation}
The dropout effect can be demonstrated in the closed-form solutions in Eq.~\eqref{eq:linear_dropout}: with the probability $1-p$, $v_i^*$ in both the equations approaches 0; with the probability $p$, $v_i^*$ approaches the solution of the plain regularizer discussed in Eq.~\eqref{eq:linear_scheme}. Recall the self-paced regularizer defines a scheme for learning samples. Eq.~\eqref{eq:linear_dropout} represent the new dropout learning scheme.
When the base learner is neural networks, the proposed dropout can be used combined with the classical dropout in~\cite{srivastava2014dropout}. The term dropout in this paper refers to dropping out samples in the iterative learning. By dropping out a sample, we drop out its update to the model parameter, which resembles the classical dropout used in neural networks. It operates on a more coarse-level which is useful for noisy data. When samples with incorrect noisy labels update a model, it will encourage the model to select more noisy labels. The dropout strategy prevents overfitting to noisy labels. It provides a way of combining many different sample subsets in different iterations in order to help avoid bad local minima. Experimental results substantiate this argument. In practice, we recommend setting two Bernoulli parameters for positive and negative samples on imbalanced data.
\section{Experiments}\label{sec:experiments}
In this section, we evaluate our method WELL for learning video detectors on noisy labeled data. The experiments are conducted on two major public benchmarks: FCVID and YFCC100M, where FCVID is by far one of the biggest manually annotated video dataset \cite{jiang2015exploiting}, and the YFCC100M dataset is the largest multimedia benchmark \cite{thomee2015yfcc100m}.
\subsection{Experimental Setup}
\textbf{Datasets, Features and Evaluation Metrics}
Fudan-columbia Video Dataset (FCVID) contains 91,223 YouTube videos (4,232 hours) from 239 categories. It covers a wide range of concepts like activities, objects, scenes, sports, DIY, etc. Detailed descriptions of the benchmark can be found in~\cite{jiang2015exploiting}. Each video is manually labeled to one or more categories. In our experiments, we do not use the manual labels in training, but instead we automatically generate the web labels according to the concept name appearance in the video metadata. The manual labels are used only in testing to evaluate our and the baseline methods. Following~\cite{jiang2015exploiting}, the standard train/test split is used.
The second set is YFCC100M \cite{thomee2015yfcc100m} which contains about 800,000 videos on Yahoo! Flickr with metadata such as the title, tags, the uploader, etc. There are no manual labels on this set and we automatically generate the curriculum from the metadata in a similar way. Since there are no annotations, we train the concept detectors on the most 101 frequent latent topics found in the metadata.
There are totally 47,397 webly labeled videos on the 101 concepts for training.
On FCVID, as the manual labels are available, the performance is evaluated in terms of the precision of the top 5 and 10 ranked videos (P@5 and P@10) and mean Average Precision (mAP) of 239 concepts. On YFCC100M, since there are no manual labels, for evaluation, we apply the detectors to a third public video collection called TRECVID MED which includes 32,000 Internet videos~\cite{over2014trecvid}. We apply the detectors trained on YFCC100M to the TRECVID videos and manually annotate the top 10 detected videos returned by each method for 101 concepts.
\textbf{Implementation Details}
We build our method on top of a pre-trained convolutional neural network as the low-level features (VGG network \cite{simonyan2014very}). We extract the key-frame level features and create a video feature by the average pooling. The same features are used across different methods on each dataset. The concept detectors are trained based on a hinge loss cost function. Algorithm~\ref{alg:overall} is used to train the concept models iteratively and the $\lambda$ stops increasing after 100 iterations. We automatically generate curriculum labels based on the video metadata, ASR, OCR and VGG net 1,000 classification results using latent topic modeling with word embedding matching as shown in Section 3, and derive a partial-order curriculum using the method discussed in Section 3.2.2.
\textbf{Baselines}
The proposed method is compared against the following five baseline methods which cover both the classical and the recent representative learning algorithms on webly-labeled data.
\textit{BatchTrain} trains a single SVM model using all samples in the multi-modal curriculum built as described in section 3.2.2 LT+WE.
\textit{Self-Paced Learning (SPL)} is a classical method where the curriculum is generated by the learner itself~\cite{kumar2010self}.
\textit{BabyLearning} is a recent method that simulates baby learning by starting with few training samples and fine-tuning using more weakly labeled videos crawled from the search engine \cite{liang2015towards}.
\textit{GoogleHNM} is a hard negative mining method proposed by Google \cite{varadarajan2015efficient}. It utilizes hard negative mining to train a second order mixture of experts model according to the video's YouTube topics. \textit{FastImage} \cite{han2015fast} is a video retrieval method that utilizes web images from search engine to match to the video with re-ranking. \textit{WELL} is the proposed method. The hyper-parameters of all methods including the baseline methods are tuned on the same validation set. On FCVID, the set is a standard development set with manual labels randomly selected from 10\% of the training set (No training was done using ground truth labels) whereas on YFCC100M it is also a 10\% proportion of noisy training set.
\subsection{Experiments on FCVID}
\textbf{Curriculum Comparison} As disscussed in Section 3.2.2, we compare different ways to build curriculum for noisy label learning.
Here we also compare their effectiveness by training concept detectors directly using the curriculum labels. The batch train model is used for all generated cirriculumn labels.
In Table \ref{exp-cur} we show the batch trained models' precision at 5, 10 and mean average precision on the test set of FCVID.
For LT+WE (Multi-modal), we extract curriculum from different modalities as shown in Section 3.2.2, and combine them using linear weights. The weights are hyper-parameters that are tuned on the validation set, and the optimal weights for textual metadata, ASR, image classification and OCR results are 1.0, 0.5, 0.5 and 0.05, respectively.
Results show that the curriculum generated by combining latent topic modeling and word embedding using multi-modal prior knowledge is the most accurate, which indicates our claim of exploiting multi-modal information is beneficial, and we use this method in WELL for the rest of the experiments.
\begin{table}[]
\centering
\caption{Curriculum BatchTrain Comparison}
\label{exp-cur}
\begin{tabular}{|l||c|c|c|}
\hline
Method & P@5 & P@10 & mAP \\ \hline \hline
ExactMatching & 0.730 & 0.713 & 0.419 \\
StemMatching & 0.782 & 0.763 & 0.469 \\
YouTubeTopicAPI & 0.587 & 0.563 & 0.315 \\
SearchEngine & 0.723 & 0.713 & 0.413 \\
WordEmbedding & 0.790 & 0.774 & 0.462 \\
LatentTopic & 0.731 & 0.716 & 0.409 \\
LT+WE & 0.804 & 0.795 & 0.473 \\
\textbf{LT+WE(Multi-modal)} & \textbf{0.838 } & \textbf{ 0.820 } & \textbf{ 0.486 } \\ \hline
\end{tabular}
\end{table}
\textbf{Baseline Comparison} Table~\ref{exps-basline} compares the precision and mAP of different methods where the best results are highlighted. As we see, the proposed WELL significantly outperforms all baseline methods, with statistically significant difference at $p$-level of 0.05. Comparing WELL with SPL, the effect of curriculum learning and dropout makes a significant difference in terms of performance, which suggests the importance of prior knowledge and preventing over-fitting in webly learning. The promising experimental results substantiate the efficacy of the proposed method.
\begin{table}[ht]
\centering
\footnotesize
\caption{Baseline comparison on FCVID}
\label{exps-basline}
\begin{tabular}{|l||c|c|c|c|c|c|}
\hline
Method & P@5 & P@10 & mAP \\ \hline \hline
BatchTrain & 0.838 & 0.820 & 0.486 \\
FastImage~\cite{han2015fast} &- &- & 0.284\\
SPL~\cite{kumar2011learning} & 0.793 & 0.754 & 0.414 \\
GoogleHNM~\cite{varadarajan2015efficient} & 0.781 & 0.757 & 0.472 \\
BabyLearning~\cite{liang2015towards} & 0.834 & 0.817 & 0.496 \\
\textbf{WELL} &\textbf{0.918}&\textbf{0.906} & \textbf{0.615} \\ \hline
\end{tabular}
\end{table}
\textbf{Robustness to Noise Comparison}
In this comparison we manually control the noisiness of the curriculum in order to systematically verify how our methods would perform with respect to the noisiness within the web data. The experimental results indicate the robustness of our method towards noisy labels. To this end, we randomly select video samples with ground truth labels for each concept, so that the precision of the curriculum labels are set at 20\%, 40\%, 60\%, 80\% and we fix the recall of all the labels. We then train WELL using such curriculum and test them on the FCVID testing set. We also compare WELL to three other methods with the same curriculum, among them \textit{GoogleHNM} is a recent method to train video concept detector with large-scale data. We exclude \textit{BabyLearning}, which relies on the returned results by the search engine, since in this experiment the curriculum is fixed . As shown in Table \ref{exp-noise}, as the noisiness of the curriculum grows (the precision drops), WELL maintains its performance while other methods drop significantly. Specifically, when the precision of the curriculum drops from 40\% to 20\%, other methods' mAP averagely drops 46.5\% while WELL's mAP only drops 19.1\% relatively. It shows that WELL is robust against different level of noise, which shows great potential in larger scale webly-labeled learning as the dataset gets bigger, the noisier it may become.
\begin{figure}[!ht]
\centering
\includegraphics[width=1.0\linewidth,height=55mm]{well-noise}
\caption{WELL performance with curriculum of different level of noisiness. p=k\% means the curriculum precision. The higher is k, the less noise is in the curriculum labels.}
\label{well-noise}
\end{figure}
\begin{table}[ht]
\centering
\footnotesize
\caption{WELL performance with curriculum of different level of noisiness. $p$ represents the precision of the curriculum. }
\vspace{1mm}
\label{exp-noise}
\begin{tabular}{|l||c|c|c|c|}
\hline
& p=20\% & p=40\% & p=60\% & p=80\% \\ \hline \hline
BatchTrain & 0.232 & 0.463 & 0.538 & 0.592 \\
SPL & 0.184 & 0.396 & 0.515 & 0.586 \\
GoogleHNM & 0.304 & 0.477 & 0.552 & 0.602 \\
\textbf{WELL} & \textbf{0.496} & \textbf{0.613} & \textbf{0.646} & \textbf{0.673} \\ \hline
\end{tabular}
\end{table}
\begin{table}[ht]
\centering
\footnotesize
\caption{WELL performance with curriculum of different level of noisiness. $p$ represents the precision of the curriculum. }
\vspace{1mm}
\label{exp-noise}
\begin{tabular}{|l||c|c|c|c|}
\hline
& p=80\% & p=60\% & p=40\% & p=20\% \\ \hline \hline
BatchTrain & 0.592& 0.538& 0.463& 0.232 \\
SPL & 0.586& 0.515& 0.396& 0.184 \\
GoogleHNM & 0.602& 0.552& 0.477& 0.304 \\
\textbf{WELL} & \textbf{0.673}& \textbf{0.646}& \textbf{0.613}& \textbf{0.496} \\ \hline
\end{tabular}
\end{table}
\textbf{Ground-truth Training Comparison}
In this part, we also compare our method with the state-of-the-art method trained using ground truth labels on FCVID (rDNN)~\cite{jiang2015exploiting}. We compare WELL trained using the static CNN features, the standard features provided by the authors~\cite{jiang2015exploiting}, and we also compare WELL using the late (average) fusion with CNN, motion and audio MFCC features (WELL-MM) to the method that achieves the best result on FCVID trained using the same multi-modal features. WELL-MM uses CNN, Motion and MFCC features, which is the same set of features as rDNN-F~\cite{jiang2015exploiting}. Noted that the state-of-the-art method uses the ground truth labels to train models, which includes 42,223 videos with manual labels, while our proposed method uses none of the human annotation into training but still be able to outperform one of the state-of-the-art results.
\begin{table}[ht]
\centering
\footnotesize
\caption{Ground-truth Training Comparison on FCVID. The methods with * are trained using human annotated labels. WELL-MM uses CNN, Motion and MFCC features, which is the same set of features as rDNN-F.}
\vspace{1mm}
\label{exps-sa}
\begin{tabular}{|l||c|c|c|c|c|c|}
\hline
Method & P@5 & P@10 & mAP \\ \hline \hline
WELL &\textbf{0.918}&\textbf{0.906} & \textbf{0.615} \\
Static CNN\cite{jiang2015exploiting}* &-&- & 0.638 \\
WELL-MM &\textbf{0.930}&\textbf{0.918} & \textbf{0.697} \\
rDNN-F\cite{jiang2015exploiting}* &-&- & 0.754 \\ \hline
\end{tabular}
\end{table}
\textbf{Noisy Dataset Size Comparison} To investigate the potential of concept learning on webly-labeled video data, we apply the methods on different sizes of subsets of the data. Specifically, we randomly split the FCVID training set into several subsets of 200, 500, 1,000, and 2,000 hours of videos, and train the models on each subset without using manual annotations. The models are then tested on the same test set. Table~\ref{exps-small} lists the average results of each type of subsets. As we see, the accuracy of WELL on webly-labeled data increases along with the growth of the size of noisy data while other webly learning methods' performance tend to be saturated.
Comparing to the methods trained using ground truth, In Table~\ref{exps-small}, WELL-MM trained using the whole dataset (2000h) outperforms rDNN-F(trained using manual labels) trained using around 1200h of data. And since the incremental performance increase of WELL-MM is close to linear, we conclude that with sufficient webly-labeled videos WELL-MM will be able to outperform the rDNN-F trained using 2000h of data, which is currently the largest manual labeled dataset.
\vspace{-3mm}
\begin{figure}[!ht]
\centering
\includegraphics[width=1.05\linewidth,height=70mm]{well-dataset}
\vspace{-7mm}
\caption{MAP comparison of models trained using web labels and ground-truth labels on different subsets of FCVID. The methods with * are trained using human annotated labels.}
\label{well-dataset}
\end{figure}
\begin{table}[ht]
\centering
\footnotesize
\caption{MAP comparison of models trained using web labels and ground-truth labels on different subsets of FCVID. The methods with * are trained using human annotated labels. Noted some of the numbers from \cite{jiang2015exploiting} are approximated from graphs.}
\vspace{1mm}
\label{exps-small}
\setlength{\tabcolsep}{4pt}
\renewcommand{\arraystretch}{1.1}
\begin{tabular}{|l||c|c|c|c|}
\hline
Dataset Size & 200h & 500h & 1000h & 2000h \\ \hline
BatchTrain & 0.364 & 0.422 & 0.452 & 0.486 \\
SPL~\cite{kumar2011learning} & 0.327 &0.379 &0.403 &0.414 \\
GoogleHNM~\cite{varadarajan2015efficient} & 0.361 & 0.421 &0.451 &0.472 \\
BabyLearning~\cite{liang2015towards} &0.390 & 0.447 & 0.481 & 0.496 \\
\textbf{WELL-MM} & \textbf{0.541} & \textbf{0.616} & \textbf{0.632} & \textbf{0.697} \\ \hline
Static CNN\cite{jiang2015exploiting}* & 0.485 & 0.561 & 0.604 & 0.638 \\
rDNN-F\cite{jiang2015exploiting}* & 0.550 & 0.620 & 0.650 & 0.754\\ \hline
\end{tabular}
\end{table}
\vspace{-3mm}
\subsection{Experiments on YFCC100M}
In the experiments on YFCC100M, we train 101 concept detectors on YFCC100M and test them on the TRECVID MED dataset which includes 32,000 Internet videos. Since there are no manual labels, to evaluate the performance, we manually annotate the top 10 videos in the test set and report their precisions in Table~\ref{exps-yfcc}. The MED evaluation is done by four annotators and the final results are averaged from all annotations. The Fleiss' Kappa value for these four annotators is 0.64.
A similar pattern can be observed where the comparisons substantiate the rationality of the proposed webly learning framework. Besides, the promising results on the largest multimedia set YFCC100M verify the scalability of the proposed method.
\vspace{-3mm}
\begin{table}[ht]
\centering
\footnotesize
\caption{Baseline comparison on YFCC100M}
\label{exps-yfcc}
\begin{tabular}{|l||c|c|c|c|c|c|}
\hline
Method & P@3 & P@5 & P@10\\ \hline \hline
BatchTrain & 0.535 & 0.513 & 0.487 \\
SPL~\cite{kumar2011learning} &0.485 & 0.463 & 0.454\\
GoogleHNM~\cite{varadarajan2015efficient} &0.541 & 0.525 & 0.500 \\
BabyLearning~\cite{liang2015towards} &0.548 & 0.519 & 0.466 \\
\textbf{WELL} &\textbf{0.667} & \textbf{0.663} & \textbf{0.649}\\ \hline
\end{tabular}
\end{table}
\begin{figure*}
\centering
\includegraphics[width=1.0\textwidth]{pick-all}
\vspace{-7mm}
\caption{WELL's example picks for different iterations}
\vspace{-5mm}
\label{well-pick-all}
\end{figure*}
\vspace{-4mm}
\subsection{Time Complexity Comparison}
The computation complexity of WELL is comparable to existing methods. For a single class, the complexity for our model and baseline models is $O(r \times n \times m)$, where $r$ is the number of iterations to converge, $n$ is the number of training samples and $m$ is the feature dimension. Theoretically, the complexity is comparable to the baseline models that have different $r$. In practice, on a 40 core-CPU machine, WELL and SPL takes 7 hours to converge (100 iterations) on FCVID with 239 concepts, whereas GoogleHNM and BabyLearning take around 5 hours.
In Table \ref{time}, we show the theoretical and actual run time for all methods.
\begin{table}[ht]
\centering
\caption{Runtime comparison across different methods. We report the time complexity on different method as well as their actual run time on FCVID (in hours).}
\label{time}
\begin{tabular}{|l||c|c|c|}
\hline
Method & Complexity & FCVID(h) \\ \hline \hline
BatchTrain & $O(n \times m)$ & 2.0 \\
SPL & $O(r \times n \times m)$ & 7.0 \\
GoogleHNM & $O(r \times n \times m)$ & 5.0 \\
BabyLearning & $O(r \times n \times m)$ & 5.0 \\
WELL & $O(r \times n \times m)$ & 7.0\\ \hline
\end{tabular}
\end{table}
\vspace{-3mm}
\subsection{Qualitative Analysis}
In this section we show training examples of WELL. In Figure~\ref{well-pick-all}, we demonstrate the positive samples that WELL select at different stage of training the concept "baseball", "forest", "birthday" and "cow".
For the concept "baseball", at early stage (1/93, 25/93), WELL selects easier and clearer samples such as those camera directly pointing at the playground, while at later stage (75/93, 93/93) WELL starts to train with harder samples with different lighting conditions and untypical samples for the concept.
For the concept "birthday", as we see, at later stage of the training, complex samples for birthday event like a video with two girl singing birthday song (75/84) and a video of celebrating birthday during hiking (84/84) are included in the training.
For the concept "forest", at the final iteration (95/95), a video of a man playing nunchaku is included, as the video title contains "Air Forester Nunchaku Freestyle" and it is included in the curriculum, which is reasonable as the curriculum is noisy and may contain false positive. Since WELL is able to leave outliers in later stage of the training, the affection of the false positives in the curriculum can be alleviated by early stopping. In our experiments, we stop increasing $\lambda$ after 100 iterations so in the final model, only a subset of samples are used.
\vspace{-2mm}
\section{Conclusions}
In this paper, we proposed a novel method called WELL for webly labeled video data learning. WELL extracts multi-modal informative knowledge from noisy weakly labeled video data from the web through a general framework with solid theoretical justifications.
WELL achieves the best performance only using webly-labeled data on two major video datasets. The comprehensive experimental results demonstrate that WELL outperforms state-of-the-art studies by a statically significant margin on learning concepts from noisy web video data. In addition, the results also verify that WELL is robust to the level of noisiness in the video data. The result suggests that with more webly-labeled data, which is not hard to obtain, WELL can potentially outperform models trained on any existing manually-labeled data.
\vspace{-3mm}
\bibliographystyle{abbrv}
\linespread{0.9}
\scriptsize
| 29cfa4dd9053e8603af4433c4a66439945522909 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section*{Introduction}
The minimum energy required to reset one bit of information represents one of the fundamental limits of computation arising when one bit of information is erased or destroyed. Starting with a system encoding two possible states with the same probability and finishing the procedure with only one possible state, the variation of entropy in the system is equal to the difference of entropy between the final state and the initial one, $\Delta S = k_\textnormal{B} \ln 1 - k_\textnormal{B} \ln 2 = -k_\textnormal{B} \ln 2$. On average this reduction of entropy has to be accompanied with an increase of heat in the surrounding environment in order to not violate the second law of thermodynamics, $Q_\textnormal{L} \geq k_\textnormal{B} T \ln 2$. This limit takes the name of Landauer limit and was theorized by Rolf Landauer in the 60's \cite{landauer1961irreversibility}, and remained untested for over fifty years. The development of computational methods allowed to evaluate tiny amounts of heat exchanged and recent technical advances in micro- and nano-fabrication made possible to recently test the validity of the Landauer principle. In particular, the first experimental verification of the Landauer principle has been carried out by B\'erut \mbox{\latinphrase{et al.\ }} using a colloidal particle trapped in optical tweezers \cite{berut2012experimental}. More recently Jun \mbox{\latinphrase{et al.\ }} presented similar results considering a colloidal particle in a feedback trap \cite{jun2014high}. Finally, the last experimental verification of the Landauer limit has been performed by Hong \mbox{\latinphrase{et al.\ }} in a nanomagnetic system \cite{Honge1501492}. In this case information is encoded in the magnetization state of the system while an external magnetic field is used in order to flip between two preferred magnetization states. The evaluation of the heat produced during the reset operation was demonstrated to be compatible with the Landauer limit. Even if there is no doubt nowadays on the validity of the Landauer principle, the test on micro-mechanical systems is still missing. We have recently shown that it is possible to use electro-mechanical devices to accomplish basic\cite{lopez2015operating} and complex logic operations\cite{lopez2015sub} with an arbitrarily low energy expenditure. Therefore the possibility to use this class of devices for memory storage completes the logic architecture for a complete computing device.
In the following we show the measurement on heat production when performing the reset operation in a novel memory unit based on a bistable mechanical cantilever at effective temperature. The results are in good agreement with the Landauer limit. The dependence of the dissipation with the error rate is also investigated showing a trend in accordance with the theory\cite{gammaitoni2015towards}.
\section*{Results}
\subsection*{Bistable micro-mechanical system}
The mechanical system used to perform the experiment is depicted in Figure \ref{f:schematic} (a). A triangular micro-cantilever, \SI{200}{\micro\meter} long, is used to encode one bit on information. In order to obtain two stable states two magnets with opposite magnetization are placed on the tip of the cantilever and on a movable stage facing the cantilever. In this way, depending on the distance between the magnets, $d$, and the relative lateral alignment, $\Delta x$, it is possible to induce bistability on the system. Figure \ref{f:schematic} (b) shows the potential energy as a function of $d$ reconstructed from the probability density function of the position of the cantilever at equilibrium, $\rho (x,d) = A \exp(-U(x,d)/k_\textnormal{B} T)$, which implies that $U(x,d)=-k_\textnormal{B} T \ln \rho (x,d) + U_\textnormal{0}$\cite{berut2012experimental}. When the magnets are far away the effect of the repulsive force is negligible, the system is then monostable and can be approximated to a linear system. Decreasing the distance the repulsive force between magnets tends to soften the system up to the point where two stable positions appear. The effect of reducing even more $d$ is to enlarge the separation of the rest states and to increase the potential barrier separating these two wells. Eventually, when the distance between the magnets is small enough, the system remains trapped in one well for a period of time larger than the relaxation time of the system.
Logic states are encoded in the position of the cantilever tip: logic 0 for $x<0$ and logic 1 for $x>0$.
The proposed system presents intrinsic dissipative processes that depend on the maximum displacement of the cantilever tip\cite{lopez2015sub}. The minimum heat produced when performing a physical transformation of the system is proportional to $x_\textnormal{max}^2$. In our setup it is not possible to reduce the separation between the two potential wells to a value that bounds the heat produced by intrinsic dissipation below the Landauer limit. Increasing the effective temperature increases the value for the Landauer limit making possible to have negligible intrinsic dissipation. A piezoelectric shaker is used to excite the structure with a band limited white Gaussian noise to mimic the effect of an arbitrary temperature. In the present experiment the white noise is limited to \SI{50}{\kilo\Hz}, well above the resonance frequency of the free cantilever ($f_\textnormal{0}$=\SI{5.3}{\kilo\Hz}). The dependence of the effective temperature with the root-mean-squared voltage supplied to the shaker is reported in Figure \ref{f:schematic} (c). The red dot, corresponding to an effective temperature of $T_\textnormal{eff}$ = \SI{5e7}{\kelvin}, highlights the condition considered in the present case. The solid line represent the expected trend where $T \propto V_\textnormal{rms}^2$ \cite{venstra2013stochastic}. The effective temperature has been estimated computing the power spectral density (PSD) of the system at various piezoelectric noise excitation voltages. The obtained curves have been fitted with Lorenzian curves taking as reference for the calibration the one at room temperature ($T$=\SI{300}{\kelvin}). The other curves have been used to extract the only varying parameter $T_\textnormal{eff}$, corresponding to the effective temperature of the system under external excitation.
Finally two electrostatic probes, placed one on the left and the other on the right of the cantilever, are used to apply a negative and positive forces respectively. When a voltage different to zero is applied on one probe the cantilever feels an attractive electrostatic force toward the probe due to the polarization of the cantilever itself. The voltage on the probes, the distance between the magnets and their time evolution are used to specify the protocols used in order to change the bit stored in the system as described in the following subsection.
\subsection*{Reset protocol}
Varying the magnets distance $d$ the barrier separating the two stable wells varies from the minimum value $B_\textnormal{0}$ for $d$=3.65 a.u. to $B_\textnormal{max}$ for $d$=2.8 a.u. Applying a voltage on one probe corresponds to apply a force toward the probe itself. Thus a voltage on the left probe corresponds to a negative force and a voltage on the right probe corresponds to a positive force. Details of the force calibration are presented in methods section.
In Figure \ref{f:protocol} (a) the protocol followed to reset the bit to the state 0 and 1 is presented. The procedure is similar to the ones presented in Refs. \citen{bennett1982thermodynamics} and \citen{berut2012experimental}. Initially the barrier separating the two stable states is removed moving the magnet away (red curve in the first panel) making the system monostable. Once the barrier is removed we apply a negative (positive) force to reset the bit status to 0 (1) applying a finite voltage $V_\textnormal{L}$ ($V_\textnormal{R}$) on the left (right) probe. This is represented by the magenta curve in Figure \ref{f:protocol} (a). Once the force is applied we restore the barrier to its original value. Finally, we remove the lateral force finishing in the original parameters configuration. At the end of the operation, if there are no errors the cantilever position encodes the desired bit of information. In order to be sure to perform the operation starting from both initial states, we mimic a statistical ensemble where the initial probability is 50\% to start in the left well and 50\% in the right well. A trace of the cantilever tip position, $x$, is shown in Figure \ref{f:protocol} (a) (black line), where the dashed blue lines represent the two stable positions once the barrier is restored. When the barrier is removed the system goes from the local prepared state to an undefined state with a free expansion, the entropy on the system thus increases in a irreversible manner\cite{bennett1982thermodynamics,crooks1999}. This increment is related to uncontrollable transitions from one well to the other once the barrier height is close to $k_\textnormal{B} T$. These large excursions of the cantilever can be seen in the time series of position. Figure \ref{f:protocol} (b) shows a representation of the time evolution of the potential energy during the set of the bit to the logic state 1. In a first step the barrier is removed allowing the system to oscillate in a monostable potential landscape. Then the potential is slightly tiled and when the barrier is restored the system is confined in the desired state. After these stages the bit is set to the state 1. In order to have a reliable measure of the heat produced during the considered operations, reset protocols are repeated for 800 times in order to have a large enough statistic.
\subsection*{Heat vs error probability}
In the optimal case the initial configuration is a mixed logical 0 and 1 where both states have the same probability while the final configuration is the selected state with a $100\%$ probability. This corresponds to an entropy variation of $\Delta S = -k_\textnormal{B} \ln(2)$ and a minimum heat produced of $Q\ge -T\Delta S = Q_\textnormal{L}$. It has been shown that the bound $Q_\textnormal{L} = -k_\textnormal{B} T \ln(2)$ applies only for symmetric potentials. Considering asymmetries on the system the minimum produced heat can be lowered below $Q_\textnormal{L}$ \cite{sagawa}. In our setup the system is slightly asymmetric and we have evaluated the variation of entropy from the initial to the final state from the probability density function of the tip position, $\rho(x)$, being $\Delta S_\textnormal{G} = -0.61k_\textnormal{B} $ $\Delta S_\textnormal{S} = -0.68k_\textnormal{B}$ for the Gibbs and Shannon entropy respectively, both close to $-k_\textnormal{B} \ln(2)$.
If we consider the possibility to commit errors during the reset operation the heat produced becomes a function of the probability of success\cite{gammaitoni2015towards}:
\begin{equation}
Q(P_\textnormal{s})\ge k_\textnormal{B} T \left[ \ln(2) + P_\textnormal{s} \ln (P_\textnormal{s}) + (1-P_\textnormal{s}) \ln (1-P_\textnormal{s}) \right]
\label{e:Q_ps}
\end{equation}
where $P_\textnormal{s}$ is the probability of success or success rate. When $P_\textnormal{s}$ is 0.5 no reset operation is performed and thus there is no minimum heat to be produced during the operation \cite{jun2014high,gammaitoni2015towards}.
In Figure \ref{f:Q_vs_sr} (a) we present the average heat produced for the reset operation as function of the lateral alignment of the counter magnet $\Delta x$. When the system is aligned closely to perfection (\mbox{\latinphrase{i.e.,\ }} $\Delta x \approx 0$) we estimate a heat production slightly above $k_\textnormal{B} T$ and below two times $Q_\textnormal{L}$. Asymmetrizing the potential, by means of setting $\Delta x \neq 0$, the heat produced tends to decrease reaching values this time below $Q_\textnormal{L}$. However, in this conditions the error rate in performing the reset operation have a major role, in fact in this configuration the probability of success, $P_\textnormal{s}$, decrease rapidly. This is represented by the color map of dots in Figure \ref{f:Q_vs_sr} (a), where green represents higher success rate while blue represents a higher probability of error. In Figure \ref{f:Q_vs_sr} (b) the success rate of the reset operation is reported as function of the lateral alignment. Solid violet circles represent the overall success rate while red and black symbols represent the error rate for resetting to 1 or to 0 respectively. Circles are used to report the error probability for the same initial and final state while crosses are used for 0 to 1 and 1 to 0 transitions. For instance let us consider the case where $\Delta x <0$: the counter magnet is moved towards the right and as a consequence the 0 state is more favorable respect the 1 state. From Figure \ref{f:Q_vs_sr} (b) we can see that for $\Delta x <0$ the probability of resetting toward 0 is almost 100\% while the probability of resetting toward 1 decreases rapidly reaching values below 50\%. The same behavior is present in the case $\Delta x > 0$, where the counter magnet is moved to the left, where the state 1 is more favorable.
We can now correlate the heat produced to the probability of success for resetting to 0 and 1 as presented in Figure \ref{f:Q_vs_sr} (c). Dashed lines represent the Landauer limit for a 100\% of success rate ($\approx 0.7 k_\textnormal{B} T$). While in both cases the heat produced is above the Landauer limit, in the reset to 0 case the obtained values are very close to $Q_\textnormal{L}$. As expected, decreasing the success rate the obtained values goes below the Landauer limit for both cases accordingly to Equation \ref{e:Q_ps}.\\
As it is well known the adiabatic limit in presence of dissipation mechanisms like viscous damping can be only reached if the operation is performed slowly when these mechanisms are negligible. We increased the protocol time for the reset operation from \SI{0.25}{\second} up to \SI{3.5}{\second}. The results are presented in Figure \ref{f:Q_vs_sr} (d). Increasing the protocol time decreases the heat production reaching values well below the Landauer limit. Notice that in these cases where $Q < Q_\textnormal{L}$ the $P_\textnormal{s}$ is well below 1 since the system has more time to relax and therefore tends to thermalize before the reset operation is correctly performed. In fact for protocols lasting more than \SI{1}{\second} the success rate is below 75\%.
\section*{Discussion}
We have measured the intrinsic minimum energy dissipation during the reset of one bit of information in a micro-mechanical system. We have considered a completely different physical system respect to the existing literature, \mbox{\latinphrase{i.e.,\ }} micro-electro-mechanical system. To achieve these results we have performed the experiment at an effective temperature of \SI{5e7}{\kelvin} in order to make the intrinsic dissipation of the mechanical structure negligible respect to the thermodynamic contribution. In these conditions we have reached values of heat produces consistent with the Landauer limit approaching it closely. Moreover we presented experimental data relating the minimum heat produced with the probability of success of resetting one bit of information. Nowadays where there is a lot of attention on micro-electro-mechanical systems able to perform computation at arbitrary low energy, the achieved results have a significant importance in the development of new computing paradigms based on systems different from the well established CMOS technology.
\section*{Methods}
\subsection*{Setup preparation and calibration}
The micro-cantilever used is a commercial atomic force microscopy (AFM) probe (NanoWorld PNP-TR-TL\cite{nanoworld}). It is long \SI{200}{\micro\meter}, with a nominal stiffness $k$=\SI{0.08}{\newton\per\meter} and a nominal resonace frequency of \SI{17}{\kilo\Hz}. A fragment of NdFeB (neodymium) magnet is attached to the cantilever tip with bi-component epoxy resin. To set the magnetization to a known direction the system is heated up to \SI{670}{\kelvin}, above its Curie temperature \cite{ma2002recent} in the presence of a strong external magnetic field with the desired orientation. With this additional mass the resonance frequency decreases to \SI{5.3}{\kilo\Hz}. The quality factor of the system has been estimated from the power spectral density of the displacement, $x$, fitted with a Lorenzian curve, giving a value of $Q_\textnormal{f}=320$.
The deflection of the cantilever, $x$, is determined by means of an AFM-like laser optical lever. A small bend of the cantilever provokes the deflection of a laser beam incident to the cantilever tip that can be detected with a two quadrants photo detector. The laser beam is focused on the cantilever tip with an optical lens (focal length $f=$\SI{50}{\milli\meter}). For small deflections the response of the photo detector remains linear, thus $x=r_\textnormal{x}\Delta V_\textnormal{PD}$, where $V_\textnormal{PD}$ is the voltage difference generated by the two quadrants of the photo detector. In order to determine $r_\textnormal{x}$ we look at the frequency response of the system as in \cite{lopez2015sub} under thermal excitation. The relation between the measured voltage and the expected displacement gives $r_\textnormal{x}=$\SI{1.8365e-5}{\meter\per\volt}.
The system is placed in a vacuum chamber and isolated from seismic vibrations to maximize the signal-to-noise ratio. All measurements were performed at pressure $P=$\SI{4.7e-2}{\milli\bar}.
\subsection*{Effective temperature estimation}
As the system is modeled as a harmonic oscillator with one degree of freedom, according to the Equipartition Theorem the thermal energy present in the system is simply related to the cantilever fluctuations as $ \frac{1}{2}k_\textnormal{B} T = \frac{1}{2} k \langle x^2\rangle $. According to Parseval's theorem $\langle x^2\rangle =\int_{0}^{\infty}\! \vert G(\omega) \vert ^2 \, \mathrm{d}\omega$ where $G(\omega)$ stands for the PSD of the system. This takes the form of a Lorenzian function $$ \vert G(\omega) \vert= \sqrt{\frac{4k_\textnormal{B} T}{Q_\textnormal{f} k \omega_\textnormal{0}}} \sqrt{\frac{1}{\left(1-\frac{\omega^2}{\omega_\textnormal{0}^2}\right)^2 + \frac{1}{Q_\textnormal{f}^2}\frac{\omega^2}{\omega_{0}^2}}} $$ where $\omega = 2 \pi f$.
Once the system has been calibrated at room temperature we estimated the effective temperatures fitting the only free parameter, $T$, from the measured PSDs to the expected function $\vert G(\omega) \vert$.
\subsection*{Force calibration}
A set of two electrodes (see Fig.~\ref{f:schematic}(a)) is used in order to polarize the cantilever producing a bend on the mechanical structure. Electrostatic forces depend on the voltage applied to the electrodes, i.e.\ $V_\textnormal{L}$ and $V_\textnormal{R}$ for the left and right electrode respectively. Since the restoring force of the cantilever can be expressed as $F_\textnormal{k}=-k x$, the relation between applied voltage to the probe and the force acting on the cantilever has been estimated in static conditions assuming $F_\textnormal{k}=F_\textnormal{el}$. The relation between the applied voltages, $V_\textnormal{L}$ and $V_\textnormal{R}$, and the electrostatic force $F_\textnormal{el}$ is then fitted with a $9^\textnormal{th}$ degree polynomial.
\subsection*{Heat production evaluation}
The work performed on the system along a given trajectory $x(t)$ is given by the integral\cite{seifert2012stochastic,douarche2005experimental}:
\begin{equation}
W=\int_{0}^{\tau_\textnormal{p}}{\sum_{k=1}^{M} \frac{\partial U(x,\boldsymbol{\lambda})}{\partial \lambda_k} \frac{\partial \lambda_k}{\partial t} \textnormal{d}t}
\end{equation}
where $\tau_\textnormal{p}$ is the protocol time duration, $U(x,\boldsymbol{\lambda})$ is the total potential energy of the system and $\boldsymbol{\lambda}$ is a vector containing all the $M$ control parameters. In our case we have two controls parameter, the voltage applied to the piezoelectric stage to control the energy barrier, and the electrostatic forces. To obtain the heat produced $Q$ we have to consider the variation in internal energy, $\Delta E$. Ideally the potential energy at both the bottom wells is the same, however considering asymmetries on the system $\Delta E$ can be different from zero. The evaluation of the total energy variation $\Delta E$ is obtained from the reconstructed potential energy, $U$, and the variation of the kinetic energy. The latter quantity is however negligible even for the shortest $\tau_p$, where the total kinetic energy variation is one order of magnitude lower than $Q_\textnormal{L}$ (\SI{1.4e-17}{\joule} versus \SI{5.3e-16}{\joule}). Finally the heat produced is obtained from $Q=W-\Delta E$.
| 5a145b9413ad053c9a8bd7ede060f0afe7ca4a2a | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction} \label{sec-intro}
In this paper we study sublanguages of Plotkin's functional programming language ${\mathrm{PCF}}$,
which we here take to be the simply typed $\lambda$-calculus over a single base type $\nat$,
with constants
\[ \begin{array}{rlcrl}
\num{n} & : \nat \mbox{~~for each $n \in \mathbb{N}$} \;, & &
{\mathit{suc}},{\mathit{pre}} & : \nat\rightarrow\nat \;, \\
{\mathit{ifzero}} & : \nat\rightarrow\nat\rightarrow\nat\rightarrow\nat \;, & &
Y_\sigma & : (\sigma\rightarrow\sigma)\rightarrow\sigma \mbox{~~for each type $\sigma$} \;.
\end{array} \]
As usual, we will consider this language to be endowed with a certain (call-by-name) operational semantics,
which in turn gives rise to a notion of \emph{observational equivalence} for ${\mathrm{PCF}}$ programs.
We define the \emph{level} ${\mathrm{lv}}(\sigma)$ of a type $\sigma$ inductively by
\[ {\mathrm{lv}}(\nat) ~=~ 0 \;, ~~~~~~ {\mathrm{lv}}(\sigma\rightarrow\tau) ~=~ \max\,({\mathrm{lv}}(\sigma)+1,{\mathrm{lv}}(\tau)) \;, \]
and define the \emph{pure type} $\pure{k}$ of level $k \in \mathbb{N}$ by
\[ \pure{0} ~=~ \nat \;, ~~~~~~ \pure{k+1} ~=~ \pure{k} \rightarrow \nat \;. \]
Modifying the definition of ${\mathrm{PCF}}$ so that the constants $Y_\sigma$ are admitted only
for types $\sigma$ of level $\leq k$, we obtain a sublanguage ${\mathrm{PCF}}_k$ for any $k \in \mathbb{N}$.
Our main result will be that for each $k$, the expressive power of ${\mathrm{PCF}}_{k+1}$ strictly
exceeds that of ${\mathrm{PCF}}_k$: in particular, there is no closed term of ${\mathrm{PCF}}_k$ that is observationally equivalent to $Y_{\pure{k+1}}$.
(Fortunately, `observational equivalence' has the same meaning for all the
languages in question here, as will be explained in Section~\ref{sec-background}.)
This answers a question posed explicitly by Berger in \cite{Berger-min-recursion},
but present in the folklore at least since the early 1990s.
It is worth remarking that the situation is quite different for various extensions of ${\mathrm{PCF}}$
considered in the literature, in which one may restrict to recursions at level~1 types
without loss of expressivity (see Subsection~\ref{subsec-embed}).
We can phrase our result more denotationally in terms of the type structure
${\mathsf{SF}}$ of \emph{(PCF-)sequential functionals} or its effective substructure ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$.
As will be reviewed in Section~\ref{sec-background},
the latter may be conveniently characterized up to isomorphism as the closed term model for ${\mathrm{PCF}}$
modulo observational equivalence.
Our result can therefore be understood as saying that more elements of ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$ (and hence of ${\mathsf{SF}}$)
are denotable in ${\mathrm{PCF}}_{k+1}$ than in ${\mathrm{PCF}}_k$.
From this we may easily infer that there is no finite `basis'
$B \subseteq {\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$ relative to which all elements of ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$ are $\lambda$-definable
(see Corollary~\ref{no-finite-basis-cor} below).
The models ${\mathsf{SF}}$ and ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$ are \emph{extensional}:
elements of type $\sigma \rightarrow \tau$ can be considered as mathematical functions mapping
elements of type $\sigma$ to elements of type $\tau$.
Whilst our theorem is naturally stated in terms of these extensional models,
its proof will make deep use of a more intensional model which yields ${\mathsf{SF}}$
as its extensional quotient.
This intensional model of PCF has been considered many times before in the literature,
for instance as the \emph{${\mathrm{PCF}}$ B\"ohm tree} model of Amadio and Curien \cite{Amadio-Curien},
and it is known to be isomorphic to the game models of PCF given by
Abramsky, Jagadeesan and Malacaria \cite{AJM} and by Hyland and Ong \cite{Hyland-Ong}.
In this paper, we choose to work with the \emph{nested sequential procedure} (NSP)
presentation of this model, as studied in detail in the recent book of Longley and Normann
\cite{Longley-Normann}.
(We will touch briefly on the possible use of other presentations for this purpose
in Section~\ref{sec-further-work}.)
We shall denote the NSP model by ${\mathsf{SP}}^0$;
its construction will be reviewed in Section~\ref{sec-background},
but in the meantime, let us offer a high-level overview of our proof method
without assuming detailed knowledge of this model.
As a motivating example, fix $k \in \mathbb{N}$, and consider the ${\mathrm{PCF}}$ term
\[ \Phi_{k+1} ~:~ (\nat \rightarrow \pure{k+1} \rightarrow \pure{k+1}) \rightarrow \nat \rightarrow \pure{k+1} \]
given informally by
\[ \Phi_{k+1}\;g\;n ~=~ g\;n\;(\Phi_{k+1}\;g\;(n+1)) ~=~ g\;n\;(g\;(n+1)\;(g\;(n+2)\;(g\;\cdots))) \;,
\]
or more formally by
\[ \Phi_{k+1} ~=~ \lambda g n.\; Y_{\nat \rightarrow \pure{k+1}} \,
(\lambda f m.\,g\,m\,(f(m+1)))\,n \;, \]
where $f$ has type $\nat \rightarrow \pure{k+1}$.
Clearly, $\Phi_{k+1}$ is a term of ${\mathrm{PCF}}_{k+1}$; however, we will be able to
show that the element of ${\mathsf{SF}}$ that $\Phi_{k+1}$ defines is not denotable in ${\mathrm{PCF}}_k$.
What is the essential feature of this function that puts it beyond the reach of ${\mathrm{PCF}}_k$?
To get a hint of this, we may observe from the informal definition above that
$\Phi_{k+1}$ seems implicitly to involve an infinite nested sequence of calls to its argument $g$,
and indeed the NSP model makes this idea precise.
Furthermore, each call to $g$ involves an argument of type level $k+1$ resulting from another such call.
Broadly speaking, we shall refer to such a sequence of nested calls (subject to certain other conditions)
as a \emph{$k\!+\!1$-spine} in the NSP associated with $\Phi_{k+1}$.
As a second example, we can see from the natural recursive definition of $Y_{\pure{k+1}}$ itself
that this too involves a spine of this kind:
\[ Y_{\pure{k+1}} \; h ~=~ h(Y_{\pure{k+1}}\;h) ~=~ h(h(h(\cdots))) \;. \]
where $h$ has type $\pure{k+1} \rightarrow \pure{k+1}$.
In fact, we may view $Y_{\pure{k+1}}$ as a `special case' of $\Phi_{k+1}$, since
$Y_{\pure{k+1}} \,h$ is observationally equivalent to $\Phi_{k+1} \,(\lambda n.h) \,\num{0}$.
A suitable general definition of $k\!+\!1$-spine turns out to be quite delicate to formulate;
but having done this, it will be possible to prove that no ${\mathrm{PCF}}_k$-denotable NSP contains a
$k\!+\!1$-spine (Theorem~\ref{no-gremlin-thm}).
This will be proved by induction on the generation of such NSPs: in essence, we have to show that none of
the generating operations for the interpretations of ${\mathrm{PCF}}_k$ terms are capable of manufacturing
spinal NSPs out of non-spinal ones. This already suffices to show that within the
intensional model ${\mathsf{SP}}^0$, the evident procedure for $Y_{\pure{k+1}}$ is denotable in ${\mathrm{PCF}}_{k+1}$
but not in ${\mathrm{PCF}}_k$.
However, this does not yet establish our main theorem, which concerns not ${\mathsf{SP}}^0$ but its extensional quotient ${\mathsf{SF}}$. For this purpose, we undertake a closer analysis of the function $\Phi_{k+1}$ defined above:
we show that
not only the NSP arising from the above definition, but any \emph{extensionally equivalent} NSP,
must necessarily involve a $k\!+\!1$-spine.
This shows that, within ${\mathsf{SF}}$ (or ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$), the functional given by $\Phi_{k+1}$ is
not denotable in ${\mathrm{PCF}}_k$.
This establishes Berger's conjecture that the languages ${\mathrm{PCF}}_k$ form a strict hierarchy.
Since $\Phi_{k+1}$ is definable from $Y_{\nat \rightarrow \pure{k+1}}$, the above shows that
the element $Y_{\nat \rightarrow \pure{k+1}} \in {\mathsf{SF}}$ is not ${\mathrm{PCF}}_k$-denotable.
To complete the picture, however, we would also like to know that the simpler element
$Y_{\pure{k+1}} \in {\mathsf{SF}}$ is not ${\mathrm{PCF}}_k$-denotable.
We show this via a more refined version of the above analysis which is of some interest in its own right.
Just as ${\mathrm{PCF}}$ is `stratified' into sublanguages ${\mathrm{PCF}}_k$, we show that each ${\mathrm{PCF}}_k$ may be further
stratified into sublanguages ${\mathrm{PCF}}_{k,1}, {\mathrm{PCF}}_{k,2}, \ldots$ on the basis of the `width' of the types $\sigma$
for which $Y_\sigma$ is permitted (see Definition~\ref{width-def}).
An easy adaptation of our earlier proofs then shows that, for any $l$,
there are operators $Y_\sigma$ in ${\mathrm{PCF}}_{k,l+2}$ that are not denotable in ${\mathrm{PCF}}_{k,l}$
(the appearance of $l+2$ is admittedly a curiosity here).
Since all these $Y_\sigma$ are themselves readily definable from $Y_{\pure{k+1}}$,
it follows that $Y_{\pure{k+1}} \in {\mathsf{SF}}$ itself is not denotable in ${\mathrm{PCF}}_{k,l}$ for any $l$,
and hence not in ${\mathrm{PCF}}_k$.
This finer analysis illustrates the remarkable richness of structure that ${\mathsf{SF}}$ has to offer.
The paper is organized as follows.
In Section~\ref{sec-background} we recall the necessary technical background on ${\mathrm{PCF}}$
and on the models ${\mathsf{SP}}^0$ and ${\mathsf{SF}}$, fleshing out many of the ideas outlined above.
In Section~\ref{sec-denotations} we obtain a convenient inductive characterization
of the class of procedures denotable in (the `oracle' version of) ${\mathrm{PCF}}_k$,
framed in terms of constructions on the procedures themselves.
In Section~\ref{sec-Y-k+1} we introduce the central concept of a $k\!+\!1$-spinal procedure,
and show using our inductive characterization that no ${\mathrm{PCF}}_k$-denotable procedure can be
$k\!+\!1$-spinal (this is the most demanding part of the proof).
As noted above, this already shows that ${\mathrm{PCF}}_{k+1}$ denotes more elements of ${\mathsf{SP}}^0$
than ${\mathrm{PCF}}_k$ does.
In Section~\ref{sec-extensional} we obtain the corresponding result for ${\mathsf{SF}}$, showing that the element
$\Phi_{k+1} \in {\mathsf{SF}}$, and hence $Y_{\nat \rightarrow \pure{k+1}} \in {\mathsf{SF}}$, is not ${\mathrm{PCF}}_k$-denotable.
In Section~\ref{sec-pure-type} we adapt our methods to the more fine-grained hierarchy of languages
${\mathrm{PCF}}_{k,l}$ that takes account of the widths of types; this enables us to show also that
$Y_{\pure{k+1}} \in {\mathsf{SF}}$ is not ${\mathrm{PCF}}_k$-denotable.
We conclude in Section~\ref{sec-further-work} with a discussion of related and future work.
The present paper is a revised, corrected and expanded version
of a University of Edinburgh technical report
from July 2015, the most significant changes being
the addition of the material on the pure type $\pure{k+1}$ in Section~\ref{sec-pure-type},
and a simplified approach to characterizing a suitable substructure of ${\mathsf{SP}}^0$ in
Section~\ref{sec-denotations}.
I am grateful to Ulrich Berger, Mart\'{i}n Escard\'{o}, Dag Normann and Alex Simpson
for valuable discussions and correspondence, and to Luke Ong and Colin Stirling for helping me to
navigate the existing literature on $\lambda Y$ and recursion schemes
(as discussed in Section~\ref{sec-further-work}).
Many of the participants in the Domains XII workshop in Cork and the Galop XI workshop in Eindhoven
also offered valuable comments; thanks in particular to Neil Jones for drawing his work to my attention.
Finally, I thank the anonymous referees for their careful work on the paper and their valuable suggestions;
these have resulted in significant improvements in both the overall architecture and
the formal details.
\section{Background} \label{sec-background}
We here summarize the necessary definitions and technical background from
\cite{Longley-Normann}, especially from Chapters~6 and 7.
\subsection{The language ${\mathrm{PCF}}$} \label{subsec-PCF}
In \cite{Scott-TCS}, Scott introduced the language LCF for computable functionals of simple type.
This language is traditionally called ${\mathrm{PCF}}$ when equipped with a standalone
operational semantics as in Plotkin \cite{LCF-considered}.
We will work here with the same version of ${\mathrm{PCF}}$ as in \cite{Longley-Normann},
with the natural numbers as the only base type.
Our types $\sigma$ are thus generated by
\[ \sigma ~::=~ \nat ~\mid~ \sigma \rightarrow \sigma \;, \]
and our terms will be those of the simply typed $\lambda$-calculus
constructed from the constants
\[ \begin{array}{rcll}
\num{n} & : & \nat & \mbox{for each $n \in \mathbb{N}$} \;, \\
\mathit{suc},\;\mathit{pre} & : & \nat\rightarrow\nat \;, \\
\mathit{ifzero} & : & \nat\rightarrow\nat\rightarrow\nat\rightarrow\nat \;, \\
Y_\sigma & : & (\sigma\rightarrow\sigma) \rightarrow \sigma &
\mbox{for each type $\sigma$} \;.
\end{array} \]
We often abbreviate the type $\sigma_0 \rightarrow\cdots\rightarrow \sigma_{r-1} \rightarrow \nat$
to $\sigma_0,\ldots,\sigma_{r-1} \rightarrow \mathbb{N}$ or just $\vec{\sigma} \rightarrow \nat$.
As usual, we write $\Gamma \vdash M:\sigma$ to mean that $M$ is a well-typed
term in the environment $\Gamma$ (where $\Gamma$ is a finite list of typed variables).
Throughout the paper, we shall regard the type of a variable $x$ as intrinsic to $x$, and will
often write $x^\sigma$ to indicate that $x$ carries the type $\sigma$.
For each $k \in \mathbb{N}$, the sublanguage ${\mathrm{PCF}}_k$ is obtained by admitting the constants
$Y_\sigma$ only for types $\sigma$ of level $\leq k$.
We endow the class of closed ${\mathrm{PCF}}$ terms with the following small-step reduction rules:
\[ \begin{array}{rclcrcl}
(\lambda x.M)N & \rightsquigarrow & M[x \mapsto N] \;, & &
\mathit{ifzero}\;\num{0} & \rightsquigarrow & \lambda xy.x \;, \\
\mathit{suc}\;\num{n} & \rightsquigarrow & \num{n+1} \;, & &
\mathit{ifzero}\;\num{n+1} & \rightsquigarrow & \lambda xy.y \;, \\
\mathit{pre}\;\num{n+1} & \rightsquigarrow & \num{n} \;, & &
Y_\sigma M & \rightsquigarrow & M(Y_\sigma M) \;. \\
\mathit{pre}\;\num{0} & \rightsquigarrow & \num{0} \;,
\end{array} \]
We furthermore allow these reductions to be applied in certain term contexts.
Specifically, the relation $\rightsquigarrow$ is inductively generated by the rules above along with the clause:
if $M \rightsquigarrow M'$ then $E[M] \rightsquigarrow E[M']$,
where $E[-]$ is one of the contexts
\[ [-]N \;, ~~~~~~~~ {\mathit{suc}}\,[-] \;, ~~~~~~~~
{\mathit{pre}}\,[-] \;, ~~~~~~~~ {\mathit{ifzero}}\,[-] \;. \]
We write $\rightsquigarrow^*$ for the reflexive-transitive closure of $\rightsquigarrow$.
If $Q$ is any closed ${\mathrm{PCF}}$ term of type $\nat$, it is easy to check that either
$Q \rightsquigarrow^* \num{n}$ for some $n \in \mathbb{N}$
or the (unique) reduction path starting from $Q$ is infinite.
This completes the definition of the languages ${\mathrm{PCF}}$ and ${\mathrm{PCF}}_k$.
Whilst the language ${\mathrm{PCF}}_0$ is too weak for programming purposes
(it cannot even define addition), it is not hard to show that even ${\mathrm{PCF}}_1$ is Turing-complete:
that is, any partial computable function $\mathbb{N} \rightharpoonup \mathbb{N}$ is representable
by a closed ${\mathrm{PCF}}_1$ term of type $\nat \rightarrow \nat$.
We will also refer to the non-effective language ${\mathrm{PCF}}^\Omega$
(or \emph{oracle ${\mathrm{PCF}}$}) obtained by extending the definition of ${\mathrm{PCF}}$ with a
constant $C_f: \nat\rightarrow\nat$ for every set-theoretic partial function $f : \mathbb{N} \rightharpoonup \mathbb{N}$,
along with a reduction rule $C_f \num{n} \rightsquigarrow \num{m}$ for every $n,m$ such that
$f(n)=m$. (In ${\mathrm{PCF}}^\Omega$, the evaluation of a closed term $Q:\nat$ may fail to
reach a value $\num{n}$ either because it generates an infinite computation, or because
it encounters a subterm $C_f(n)$ where $f(n)$ is undefined.)
The languages ${\mathrm{PCF}}^\Omega_k$ are defined analogously.
If $M,M'$ are closed ${\mathrm{PCF}}^\Omega$ terms of the same type $\sigma$,
and ${\mathcal{L}}$ is one of our languages ${\mathrm{PCF}}_k$, ${\mathrm{PCF}}_k^\Omega$, ${\mathrm{PCF}}$, ${\mathrm{PCF}}^\Omega$,
we say that $M,M'$ are \emph{observationally equivalent in ${\mathcal{L}}$},
and write $M \simeq_{\mathcal{L}} M'$,
if for all closed program contexts $C[-^\sigma] : \nat$ of ${\mathcal{L}}$ and all $n \in \mathbb{N}$, we have
\[ C[M] \rightsquigarrow^* \num{n} \mbox{~~iff~~} C[M'] \rightsquigarrow^* \num{n} \;. \]
(It makes no difference to the relation $\simeq_{\mathcal{L}}$ whether we take $C[-]$ to range over
single-hole or multi-hole contexts.)
Fortunately, it is easy to show that all of the above languages give rise to exactly the
same relation $\simeq_{\mathcal{L}}$.
Indeed, it is immediate from the definition that if ${\mathcal{L}},{\mathcal{L}}'$ are two of our languages
and ${\mathcal{L}} \supseteq {\mathcal{L}}'$, then $\simeq_{{\mathcal{L}}} \,\subseteq\, \simeq_{{\mathcal{L}}'}$;
it therefore only remains to verify that $M \simeq_{{\mathrm{PCF}}_0} M'$ implies
$M \simeq_{{\mathrm{PCF}}^\Omega} M'$.
We may show this by a `syntactic continuity' argument,
exploiting the idea that any of the constants $Y_\sigma$ or $C_f$ in ${\mathrm{PCF}}^\Omega$
can be `approximated' as closely as necessary by terms of ${\mathrm{PCF}}_0$.
Specifically, let us write $\bot$ for the non-terminating program $Y_0(\lambda x^0.x) : \nat$
(a term of ${\mathrm{PCF}}_0$), and for any type $\sigma$ write $\bot_\sigma$ for the term of type $\sigma$
of the form $\lambda \vec{x}.\bot$.
For any $j \in \mathbb{N}$, we may then define ${\mathrm{PCF}}_0$ terms
\begin{eqnarray*}
Y_\sigma^{(j)} & = & \lambda f^{\sigma\rightarrow\sigma}.\,f^j(\bot_\sigma) \;, \\
C_f^{(j)} & = & \lambda n.\,\mathit{case}~n~\mathit{of}~
(0 \Rightarrow \num{f(0)} \mid \cdots \mid j-1 \Rightarrow \num{f(j-1)}) \;,
\end{eqnarray*}
where we use some evident syntactic sugar in the definition of $C_f^{(j)}$.
For any ${\mathrm{PCF}}^\Omega$ term $M$, let $M^{(j)}$ denote the `approximation' obtained from
$M$ by replacing all occurrences of constants
$Y_\sigma, C_f$ by $Y_\sigma^{(j)}, C_f^{(j)}$ respectively.
It is then not hard to show that for closed $Q:\nat$, we have
\[ Q \rightsquigarrow^* \num{n} \mbox{~~~iff~~~}
\exists j.\; Q^{(j)} \rightsquigarrow^* \num{n} \;. \]
From this it follows easily that if $C[-]$ is an observing context of ${\mathrm{PCF}}^\Omega$
that distinguishes $M,M'$, then some approximation $C^{(j)}[-]$ (a context of ${\mathrm{PCF}}_0$)
also suffices to distinguish them. This establishes that
$\simeq_{{\mathrm{PCF}}_0} \,\subseteq\; \simeq_{{\mathrm{PCF}}^\Omega}$.
We may therefore write $\simeq$ for observational equivalence without ambiguity.
In fact, an even more restricted class of observing contexts suffices for ascertaining
observational equivalence of ${\mathrm{PCF}}^\Omega$ terms.
The well-known \emph{(equational) context lemma}, due to Milner \cite{Milner-PCF},
states that $M \simeq M' : \sigma_0,\ldots,\sigma_{r-1} \rightarrow \nat$
iff $M,M'$ have the same behaviour in all \emph{applicative contexts} of ${\mathrm{PCF}}$---%
that is, if for all closed ${\mathrm{PCF}}$ terms $N_0:\sigma_0,\ldots,N_{r-1}:\sigma_{r-1}$, we have
\[ M N_0 \ldots N_{r-1} \rightsquigarrow^* n \mbox{~~iff~~}
M' N_0 \ldots N_{r-1} \rightsquigarrow^* n \;. \]
Furthermore, using the above idea of approximation, it is easy to see that we obtain
exactly the same equivalence relation if we allow the $N_i$ here to range only over
closed ${\mathrm{PCF}}_0$ terms---this gives us the notion of
\emph{${\mathrm{PCF}}_0$ applicative equivalence}, which we shall denote by $\sim_{0}$.
We have concentrated so far on giving a purely operational description of ${\mathrm{PCF}}$.
We are now able to express the operational content of our main theorems as follows.
As in Section~\ref{sec-intro}, we define the type $\pure{k}$ by $\pure{0} = \nat$,
$\pure{k+1} = \pure{k} \rightarrow \nat$;
we shall write $\pure{k}$ simply as $k$ where there is no risk of confusion.
\begin{theorem} \label{operational-main-thm}
For any $k \geq 1$, there are functionals definable in ${\mathrm{PCF}}_{k+1}$ but not in ${\mathrm{PCF}}^\Omega_k$.
More specifically:
(i) There is no closed term $M$ of ${\mathrm{PCF}}^\Omega_k$ such that
$M \simeq Y_{\nat \rightarrow (k+1)}$ (or equivalently $M \sim_{0} Y_{\nat \rightarrow (k+1)}$).
(ii) There is even no closed $M$ of ${\mathrm{PCF}}^\Omega_k$ such that
$M \simeq Y_{k+1}$.
\end{theorem}
We shall obtain part~(i) of this theorem at the end of Section~\ref{sec-extensional},
then in Section~\ref{sec-pure-type} resort to a more indirect method to obtain the stronger
statement (ii).
The theorem also holds when $k=0$, but this rather uninteresting case does not require
the methods of this paper; it will be dealt with in Subsection~\ref{subsec-power-pcf1}.
Theorem~\ref{operational-main-thm} can be construed as saying that in a suitably
pure fragment of a functional language such as Haskell,
the computational strength of recursive function definitions increases strictly as the admissible
type level for such recursions is increased.
The point of the formulation in terms of $\sim_{0}$ is to present our result in
as manifestly strong a form as possible:
there is no $M \in {\mathrm{PCF}}_k$ that induces the same partial function as $Y_{k+1}$
even on closed ${\mathrm{PCF}}_0$ terms.
A more denotational formulation of our theorem can be given in terms of
the model ${\mathsf{SF}}$ of \emph{sequential functionals}, which we here define as the
type structure of closed ${\mathrm{PCF}}^\Omega$ terms modulo observational equivalence.
Specifically, for each type $\sigma$, let ${\mathsf{SF}}(\sigma)$ denote the set of closed
${\mathrm{PCF}}^\Omega$ terms $M:\sigma$ modulo $\simeq$.
It is easy to check that application of ${\mathrm{PCF}}^\Omega$ terms
induces a well-defined function
$\cdot : {\mathsf{SF}}(\sigma\rightarrow\tau) \times {\mathsf{SF}}(\sigma) \rightarrow {\mathsf{SF}}(\tau)$
for any $\sigma,\tau$;
the structure ${\mathsf{SF}}$ then consists of the sets ${\mathsf{SF}}(\sigma)$ along with these application operations.
Using the context lemma, it is easy to see that ${\mathsf{SF}}(\nat) \cong \mathbb{N}_\bot = \mathbb{N} \sqcup \{ \bot \}$,
and also that ${\mathsf{SF}}$ is \emph{extensional}:
if $f,f' \in {\mathsf{SF}}(\sigma\rightarrow\tau)$ satisfy
$f \cdot x = f' \cdot x$ for all $x \in {\mathsf{SF}}(\sigma)$, then $f = f'$.
Thus, up to isomorphism, each ${\mathsf{SF}}(\sigma\rightarrow\tau)$ may be considered simply as a
certain set of functions from ${\mathsf{SF}}(\sigma)$ to ${\mathsf{SF}}(\tau)$.
Any closed ${\mathrm{PCF}}^\Omega$ term $M:\sigma$ naturally has a denotation $\sem{M}^{\mathsf{SF}}$
in ${\mathsf{SF}}(\sigma)$, namely its own equivalence class. We may therefore restate
Theorem~\ref{operational-main-thm} as:
\begin{theorem} \label{SF-main-thm}
Suppose $k \geq 1$.
(i) The element $\sem{Y_{\nat\rightarrow(k+1)}}^{\mathsf{SF}}$ is not denotable in ${\mathrm{PCF}}^\Omega_k$.
(ii) Even $\sem{Y_{k+1}}^{\mathsf{SF}}$ is not ${\mathrm{PCF}}^\Omega_k$-denotable.
\end{theorem}
It follows immediately that in any other adequate, compositional model of ${\mathrm{PCF}}^\Omega$
(such as Scott's continuous model or Berry's stable model), the element
$\sem{Y_{k+1}}$ is not ${\mathrm{PCF}}^\Omega_k$-denotable,
since the equivalence relation on ${\mathrm{PCF}}^\Omega$ terms induced by such a model must be contained
within $\simeq$.
By taking closed terms of ${\mathrm{PCF}}$ rather than ${\mathrm{PCF}}^\Omega$ modulo observational
equivalence, we obtain the type structure ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$ of
\emph{effective sequential functionals}, which can clearly be seen as a substructure of ${\mathsf{SF}}$.
Although the above constructions of ${\mathsf{SF}}$ and ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$ are syntactic,
there are other more mathematical constructions
(for instance, involving game models \cite{AJM,Hyland-Ong})
that also give rise to these structures,
and experience suggests that these are mathematically natural classes of
higher-order functionals.
We now see that Theorem~\ref{SF-main-thm}(i)
implies an interesting absolute property of ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$,
not dependent on any choice of presentation for this structure or any selection of language primitives:
\begin{corollary}[No finite basis] \label{no-finite-basis-cor}
There is no finite set $B$ of elements of ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$ such that all
elements of ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$ are $\lambda$-definable relative to $B$.
In other words, the cartesian category of ${\mathrm{PCF}}$-computable functionals is not finitely generated.
\end{corollary}
\begin{proof}
Suppose $B = \{ b_0, \ldots, b_{n-1} \}$ were such a set.
For each $i$, take a closed ${\mathrm{PCF}}$ term $M_i$ denoting $b_i$.
Then the terms $M_0,\ldots,M_{n-1}$ between them contain only finitely many
occurrences of constants $Y_\sigma$,
so these constants are all present in ${\mathrm{PCF}}_k$ for large enough $k$.
But this means that $b_0,\ldots,b_{n-1}$, and hence all elements of ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$,
are ${\mathrm{PCF}}_k$-denotable, contradicting Theorem~\ref{SF-main-thm}(i).
$\Box$
\end{proof}
\subsection{The model ${\mathsf{SP}}^0$} \label{sec-SP0}
We turn next to an overview of the
\emph{nested sequential procedure} (or NSP) model, denoted by ${\mathsf{SP}}^0$.
Further details and motivating examples are given in \cite{Longley-Normann}.
In some respects, however, our presentation here will be more formal than that of \cite{Longley-Normann}:
in particular, our discussion of bound variables and $\alpha$-conversion issues will be somewhat
more detailed, in order to provide a solid foundation for the delicate syntactic arguments that follow.
The ideas behind this model have a complex history.
The general idea of sequential computation via nested oracle calls
was the driving force behind Kleene's later papers
(e.g.\ \cite{Kleene-revisited-IV}), although the concept did not receive a
particularly transparent or definitive formulation there.
Many of the essential ideas of NSPs can be found in early work of
Sazonov \cite{Sazonov-early}, in which a notion of
\emph{Turing machine with oracles} was used to characterize the
`sequentially computable' elements of the Scott model.
NSPs as we study them here were first explicitly introduced
in work on game semantics for ${\mathrm{PCF}}$--
both by Abramsky, Jagadeesan and Malacaria \cite{AJM}
(under the name of \emph{evaluation trees})
and by Hyland and Ong \cite{Hyland-Ong} (under the name of \emph{canonical forms}).
In these papers, NSPs played only an ancillary role;
however, it was shown by Amadio and Curien \cite{Amadio-Curien} how
(under the name of \emph{${\mathrm{PCF}}$ B\"ohm trees}) they
could be made into a model of ${\mathrm{PCF}}$ in their own right.
Similar ideas were employed again by Sazonov
\cite{Sazonov-PCF} to give a standalone characterization of
the class of sequentially computable functionals.
More recently, Normann and Sazonov \cite{Normann-Sazonov}
gave an explicit construction of the NSP model in
a somewhat more semantic spirit than \cite{Amadio-Curien},
using the name \emph{sequential procedures}.
As in \cite{Longley-Normann}, we here add the epithet `nested'
to emphasize the contrast with other flavours of sequential computation.%
\footnote{A major theme of \cite{Longley-Normann} is that NSPs serve equally well to
capture the essence of PCF computation and that of Kleene's S1--S9 computability;
this is one reason for preferring a name that is not biased towards PCF.}
As in \cite{Longley-Normann}, our NSPs are generated by means of the following
infinitary grammar, interpreted coinductively.
Here $\bot$ is a special atomic symbol and $n$ ranges over natural numbers.
\begin{eqnarray*}
\mbox{\it Procedures:}~~~~
p,q & ::= & \lambda x_0 \cdots x_{r-1}.\,e \\
\mbox{\it Expressions:}~~~~
d,e & ::= & \bot ~\mid~ n ~\mid~ {\mathtt{case}}~a~\,{\mathtt{of}}~
(i \Rightarrow e_i \mid i \in \mathbb{N}) \\
\mbox{\it Applications:}~~~~~~~
a & ::= & x\, q_0 \cdots q_{r-1}
\end{eqnarray*}
Here we write $(i \Rightarrow e_i \mid i \in \mathbb{N})$ to indicate an infinite sequence of `branches':
$(0 \Rightarrow e_0 \mid 1 \Rightarrow e_1 \mid 2 \Rightarrow e_2 \mid \cdots)$.
We will use vector notation to denote finite (possibly empty) lists of variables or procedures:
$\vec{x}$, $\vec{q}$.
Our convention will be that a list $\vec{x}$ must be non-repetitive, though a list $\vec{q}$ need not be.
We may use $t$ to range over \emph{NSP terms} of any of the above three kinds;
note that a `term' is formally a (possibly infinite) syntax tree as generated by the above grammar.
A procedure $\lambda \vec{x}.\bot$ will often be abbreviated to $\bot$.
For the most part, we will be working with terms modulo (infinitary) $\alpha$-equivalence,
and most of the concepts we introduce will be stable under renamings of bound variables.
Thus, a statement $t=t'$, appearing without qualification, will mean that $t,t'$ are $\alpha$-equivalent
(although we will sometimes write $=$ as $=_\alpha$ if we wish to emphasize this).
When we wish to work with terms on the nose rather than up to $=_\alpha$, we shall refer to them
as \emph{concrete terms}.
If each variable is assigned a simple type over $\nat$, then
we may restrict our attention to \emph{well-typed}\index{well-typed} terms.
Informally, a term will be well-typed unless a typing violation occurs
at some particular point within its syntax tree.
Specifically, within any term $t$, occurrences of procedures $\lambda \vec{x}.e$
(of any type), applications $x \vec{q}$ (of the ground type $\nat$) and expressions $e$
(of ground type) have types that are related to the types of their constituents and of variables
as usual in type $\lambda$-calculus extended by ${\mathtt{case}}$ expressions of type $\nat$.
We omit the formal definition here since everything works in the expected way;
for a more precise formulation see \cite[Section~6.1.1]{Longley-Normann}.
If $\Gamma$ is any environment (i.e.\ a finite non-repetitive list of variables),
we write $\Gamma \vdash e$ and $\Gamma \vdash a$ to mean that
$e,a$ respectively are well-typed with free variables in $\Gamma$.
We also write $\Gamma \vdash p:\tau$ when $p$ is well-typed in $\Gamma$ and
of type $\tau$.
We shall often refer to variable environments that arise from combining several lists of variables,
which may be represented by different notations, e.g.\ $\Gamma,V,\vec{x}$.
Since such environments are required to be non-repetitive, we take it to be
part of the content of a typing judgement such as
$\Gamma,V,\vec{x} \vdash t \,(:\tau)$ that the entire list $\Gamma,V,\vec{x}$ is non-repetitive.
However, the \emph{order} of variables within an environment will typically be of much less concern to us
(clearly our typing judgements $\Gamma \vdash T \,(:\tau)$ are robust under permutations of $\Gamma$),
and we will sometimes abuse notation by identifying a finite \emph{set} $Z$ of variables
with the list obtained from some arbitrary ordering of it.
It will also be convenient to place another condition on concrete
well-typed terms (not imposed in \cite{Longley-Normann}) in order to exclude \emph{variable hiding}.
Specifically, we shall insist that if $\Gamma \vdash t \,(:\tau)$
then no variable of $\Gamma$ appears as a bound variable within $t$,
nor are there any \emph{nested} bindings within $t$ of the same variable $x$.
(Clearly any concrete term not satisfying this restriction is $\alpha$-equivalent to one that does.)
This will help to avoid confusion in the course of some delicate arguments in which questions of
the identity of variables are crucial.
With these ideas in place, we may take ${\mathsf{SP}}(\sigma)$ to be the set of
well-typed procedures of type $\sigma$ modulo $=_\alpha$,
and ${\mathsf{SP}}^0(\sigma) \subseteq {\mathsf{SP}}(\sigma)$ the subset constituted by the \emph{closed} procedures
(i.e.\ those that are well-typed in the empty environment).
By inspection of the grammar for procedures, it is easy to see that ${\mathsf{SP}}^0(\nat) \cong \mathbb{N}_\bot$.
As in \cite{Longley-Normann}, we shall need to work not only with NSPs
themselves, but with a more general calculus of NSP \emph{meta-terms}
designed to accommodate the intermediate forms that arise in the course of computations:
\begin{eqnarray*}
\mbox{\it Meta-procedures:} ~~~~
P,Q & ::= & \lambda \vec{x}.\,E \\
\mbox{\it Meta-expressions:} ~~~~
D,E & ::= & \bot ~\mid~ n ~\mid~
\caseof{G}{i \Rightarrow E_i \mid i \in \mathbb{N}} \\
\mbox{\it Ground meta-terms:} ~~~~~~~~
G & ::= & E ~\mid~ x\, \vec{Q} ~\mid~ P \vec{Q}
\end{eqnarray*}
Here again, $\vec{x}$ and $\vec{Q}$ denote finite lists.
We shall use $T$ to range over meta-terms of any of the above three kinds;
once again, a meta-term is formally a syntax tree as generated by the above grammar.
(Unless otherwise stated, we use uppercase letters for general meta-terms and lowercase
ones for terms.)
Once again, we will normally work with meta-terms up to (infinitary) $\alpha$-equivalence,
but may also work with concrete meta-terms when required.
The reader is warned of an ambiguity arising from the above grammar: a surface form
$\lambda.E$ may be parsed either as a meta-procedure $\lambda \vec{x}.E$ with $\vec{x}$ empty,
or as a ground meta-term $(\lambda \vec{x}.E)\vec{Q}$ with both $\vec{x},\vec{Q}$ empty.
Formally these are two quite distinct meta-terms, bearing in mind that meta-terms are officially syntax trees.
To remedy this ambiguity, we shall therefore in practice write `$()$' to indicate the presence of
an empty argument list $\vec{Q}$ to a meta-procedure $P$,
so that the ground meta-term above will be written as $(\lambda.E)()$.
In the absence of `$()$', a surface form $\lambda.E$ should always be interpreted as a meta-procedure.
Meta-terms are subject to the expected typing discipline, leading to typing judgements
$\Gamma \vdash P:\sigma$, $\Gamma \vdash E$, $\Gamma \vdash G$
for meta-procedures, meta-expressions and ground meta-terms respectively.
Again we omit the details: see \cite[Section~6.1.1]{Longley-Normann}.
We shall furthermore require that well-typed concrete meta-terms are subject to the
no-variable-hiding condition.
We will sometimes write e.g.\ $\Gamma \vdash P$ to mean that $P$ is a well-typed meta-procedure
in environment $\Gamma$, if the type itself is of no particular concern to us.
There is an evident notion of simultaneous capture-avoiding \emph{substitution}
$T[\vec{x}\mapsto\vec{Q}]$ for well-typed concrete terms.
Specifically, given $\Gamma,\vec{x} \vdash T \,(:\tau)$ and
$\Gamma,\vec{y} \vdash Q_i : \sigma_i$ for each $i<r$,
where $\vec{x} = x_0^{\sigma_0},\ldots,x_{r-1}^{\sigma_{r-1}}$,
we will have $\Gamma,\vec{y} \vdash T[\vec{x}\mapsto\vec{Q}] \,(:\tau)$.
Note that this may entail renaming of bound variables both within $T$ (in order to avoid capture of
variables in $\vec{y}$) and in the $Q_i$ (in order to maintain the no-hiding condition for variables
bound within $T$).
The details of how this renaming is performed will not matter, provided that for each $T,\vec{x},\vec{Q}$ as above we have a determinate choice of a suitable concrete term $T[\vec{x}\mapsto\vec{Q}]$,
so that multiple appearances of the same substitution will always yield the same result.
We also note that substitution is clearly well-defined on $\alpha$-equivalence classes.
Finally, we will say a substitution $[\vec{x}\mapsto\vec{Q}]$ \emph{covers}
a set $V$ of variables if $V$ consists of precisely the variables $\vec{x}$.
As a mild extension of the concept of meta-term, we have an evident notion of a
\emph{meta-term context} $C[-]$: essentially a meta-term containing a `hole' $-$,
which may be of meta-procedure, meta-expression or ground meta-term type
(and in the case of meta-procedures, will carry some type $\sigma$).
Our convention here is that a meta-term context $C[-]$ is permitted to contain only a single
occurrence of the hole $-$.
Multi-hole contexts $C[-_0,-_1,\ldots]$ will occasionally be used, but again, each hole $-_i$ may
appear only once.
By the \emph{local variable environment} associated
with a concrete meta-term context $\Gamma \vdash C[-]$,
we shall mean the set $X$ of variables $x$ bound within $C[-]$ whose scope includes the hole,
so that the environment in force at the hole is $\Gamma,X$.
(The no-variable-hiding convention ensures that $X$ and indeed $\Gamma,X$ is non-repetitive.)
Although in principle local variable environments pertain to particular choices of concrete contexts,
most of the concepts that we define using such environments will be easily seen to be invariant
under renamings of bound variables.
Next, there is a concept of \emph{evaluation}
whereby any concrete meta-term $\Gamma \vdash T\, (:\sigma)$ evaluates to
an ordinary concrete term $\Gamma \vdash\, \dang{\!T\!} (:\sigma)$.
To define this, the first step is to introduce a \emph{basic reduction}\index{basic reduction} relation $\rightsquigarrow_b$ for concrete ground meta-terms, which we do by the following rules:
\begin{description}
\item[(b1)] \( (\lambda \vec{x}.E) \vec{Q} \rightsquigarrow_b
E [\myvec{x} \mapsto \myvec{Q}] ~~~\) (\emph{$\beta$-rule}).
\item[(b2)] \( \caseof{\bot}{i \Rightarrow E_i} \rightsquigarrow_b \bot \).
\item[(b3)] \( \caseof{n}{i \Rightarrow E_i} \rightsquigarrow_b E_n \).
\item[(b4)] \( \caseof{(\caseof{G}{i \Rightarrow E_i})}{j \Rightarrow F_j} \rightsquigarrow_b \) \\
\( \mbox{~~~~~~~~~} \caseof{G}{i \Rightarrow \caseof{E_i}{j \Rightarrow \!F_j}} \).
\end{description}
Note that the $\beta$-rule applies even when $\vec{x}$ is empty:
thus, $(\lambda.2)() \rightsquigarrow_b 2$.
From this, a \emph{head reduction} relation $\rightsquigarrow_h$
on concrete meta-terms is defined inductively:
\begin{description}
\item[(h1)] If $G \rightsquigarrow_b G'$ then $G \rightsquigarrow_h G'$.
\item[(h2)] If $G \rightsquigarrow_h G'$ and $G$ is not a ${\mathtt{case}}$ meta-term, then
\[ \caseof{G}{i \Rightarrow E_i} ~\rightsquigarrow_h~ \caseof{G'}{i \Rightarrow E_i} \;. \]
\item[(h3)] If $E \rightsquigarrow_h E'$
then $\lambda \myvec{x}.E ~\rightsquigarrow_h~ \lambda \myvec{x}.E'$.
\end{description}
Clearly, for any meta-term $T$, there is at most one $T'$ with $T \rightsquigarrow_h T'$.
We call a meta-term a \emph{head normal form} if it cannot be further
reduced using $\rightsquigarrow_h$.
The possible shapes of head normal forms are
$\bot$, $n$, $\caseof{y \vec{Q}}{i \Rightarrow E_i}$
and $y \vec{Q}$, the first three optionally prefixed by $\lambda \myvec{x}$
(where $\vec{x}$ may contain $y$).
We now define the \emph{general reduction} relation $\rightsquigarrow$ inductively as follows:
\begin{description}
\item[(g1)] If $T \rightsquigarrow_h T'$ then $T \rightsquigarrow T'$.
\item[(g2)] If $E \rightsquigarrow E'$ then
$\lambda \myvec{x}.E ~\rightsquigarrow~ \lambda \myvec{x}.E'$.
\item[(g3)] If $Q_j = Q'_j$ except at $j=k$ where $Q_k \rightsquigarrow Q'_k$, then
\begin{eqnarray*}
x \vec{Q} & \rightsquigarrow & x \vec{Q}\,' \;, \\
\caseof{x\, \vec{Q}}{i \Rightarrow E_i} & \rightsquigarrow &
\caseof{x\, \vec{Q}\,'}{i \Rightarrow E_i} \;.
\end{eqnarray*}
\item[(g4)] If $E_i = E'_i$ except at $i=k$ where $E_k \rightsquigarrow E'_k$, then
\[ \caseof{x\, \vec{Q}}{i \Rightarrow E_i} ~\rightsquigarrow~
\caseof{x\, \vec{Q}}{i \Rightarrow E'_i} \;. \]
\end{description}
It is easy to check that this reduction system is sound with respect to the typing rules.
We emphasize that the relation $\rightsquigarrow$ is defined on concrete meta-terms,
although it is clear that it also gives rise to a well-defined reduction relation on their
$\alpha$-classes.
An important point to note is that modulo the obvious inclusion of terms into meta-terms,
terms are precisely meta-terms in \emph{normal form}, i.e.\ those that cannot be reduced using $\rightsquigarrow$.
(For example, the meta-procedure $\lambda.2$ is in normal form, though the ground meta-term
$(\lambda.2)()$ is not.)
We write $\rightsquigarrow^*$ for the reflexive-transitive closure of $\rightsquigarrow$.
The above reduction system captures the finitary aspects of evaluation.
In general, however, since terms and meta-terms may be infinitely deep,
evaluation must be seen as an infinite process.
To account for this infinitary aspect, we use some familiar domain-theoretic ideas.
We write $\sqsubseteq$
for the evident syntactic orderings on concrete meta-procedures and on ground
meta-terms: thus, $T \sqsubseteq U$ iff $T$ may be obtained from $U$
by replacing zero or more ground subterms (perhaps infinitely many) by $\bot$.
It is easy to see that for each $\sigma$, the set of all concrete procedure terms of type $\sigma$
forms a directed-complete partial order under $\sqsubseteq$.
By a \emph{finite} (concrete) term $t$, we shall mean one generated by the following
grammar, this time construed inductively:
\begin{eqnarray*}
\mbox{\it Procedures:}~~~~
p,q & ::= & \lambda x_0 \ldots x_{r-1}.\,e \\
\mbox{\it Expressions:}~~~~
d,e & ::= & \bot ~\mid~ n ~\mid~ {\mathtt{case}}~a~\,{\mathtt{of}}~
(0 \Rightarrow e_0 \mid \cdots \mid r-1 \Rightarrow e_{r-1}) \\
\mbox{\it Applications:}~~~~~~~
a & ::= & x\, q_0 \ldots q_{r-1}
\end{eqnarray*}
We regard finite terms as ordinary NSP terms
by identifying the conditional branching
$(0 \Rightarrow e_0 \mid \cdots \mid r-1 \Rightarrow e_{r-1})$ with
\[ (0 \Rightarrow e_0 \mid \cdots \mid r-1 \Rightarrow e_{r-1} \mid
r \Rightarrow \bot \mid r+1 \Rightarrow \bot \mid \cdots) \;.
\]
We may now explain how a general meta-term $T$ \emph{evaluates}
to a term $\dang{\!T\!}$.
This will in general be an infinite process,
but we can capture the value of $T$ as the limit of the finite portions that
become visible at finite stages in the reduction.
To this end, for any concrete meta-term $T$ we define
\[ \Downarrow_{\mbox{\scriptsize \rm fin}} T ~=~
\{ t \mbox{~finite~} \mid \exists T'.\; T \rightsquigarrow^* T' ~\wedge~
t \sqsubseteq T' \} \;.
\]
It is not hard to check that for any meta-term $T$, the set $\Downarrow_{\mbox{\scriptsize \rm fin}} T$
is directed with respect to $\sqsubseteq$ (see \cite[Proposition~6.1.2]{Longley-Normann}).
We may therefore define
$\dang{\!T\!}$, the \emph{value} of $T$, to be the ordinary concrete term
\[ \dang{\!T\!} ~=~ \bigsqcup \,(\Downarrow_{\mbox{\scriptsize \rm fin}} T) \;. \]
Note in passing that the value $\dang{\!G\!}$ of a ground meta-term $G$
may be either an expression or an application.
In either case, it is certainly a ground meta-term.
It is also easy to see that
$\dang{\lambda \myvec{x}.E} \,=\, \lambda \myvec{x}.\dang{\!E\!}$,
and that if $T \rightsquigarrow^* T'$ then $\dang{\!T\!} \,=\, \dang{\!T'\!}$.
Whilst we have defined our evaluation operation $\dang{\!-\!}$ for concrete meta-terms,
it is clear that this induces a well-defined evaluation operation on their $\alpha$-classes,
and for the most part this is all that we shall need.
We also note that the syntactic ordering $\sqsubseteq$ on concrete terms induces a partial order
$\sqsubseteq$ on their $\alpha$-classes, and that each ${\mathsf{SP}}(\sigma)$ and ${\mathsf{SP}}^0(\sigma)$
is directed-complete with respect to this ordering.
In the present paper, an important role will be played by the tracking of variable occurrences
(and sometimes other subterms) through the course of evaluation.
By inspection of the above rules for $\rightsquigarrow$, it is easy to see that if $T \rightsquigarrow T'$,
then for any occurrence of a (free or bound) variable $x$ within $T'$, we can identify a
unique occurrence of $x$ within $T$ from which it originates (we suppress the formal
definition). The same therefore applies whenever $T \rightsquigarrow^* T'$.
In this situation, we may say that the occurrence of $x$ within $T'$ is a \emph{residual} of the
one within $T$, or that the latter is the \emph{origin} of the former.
Note, however, that these relations are relative to a particular reduction path
$T \rightsquigarrow^* T'$: there may be other paths from $T$ to $T'$ for which the origin-residual relation
is different.
Likewise, for any occurrence of $x$ within $\dang{\!T\!}$, we may pick some finite
$t \sqsubseteq \dang{\!T\!}$ containing this occurrence, and some $T' \sqsupseteq t$
with $T \rightsquigarrow^* T'$; this allows us to identify a unique occurrence of $x$ within $T$
that originates the given occurrence in $\dang{\!T\!}$.
It is routine to check that this occurrence in $T$ will be independent of the choice of
$t$ and $T'$ and of the chosen reduction path $T \rightsquigarrow^* T'$;
we therefore have a robust origin-residual relationship between variable occurrences in $T$
and those in $\dang{\!T\!}$.
A fundamental result for NSPs is the \emph{evaluation theorem}, which says that
the result of evaluating a meta-term is unaffected if we choose to evaluate certain subterms
`in advance':
\begin{theorem}[Evaluation theorem] \label{eval-thm}
If $C[-_0,-_1,\ldots]$ is any meta-term context with countably
many holes and $C[T_0,T_1,\ldots]$ is well-formed, then
\[ \dang{C[T_0,T_1,\ldots]} ~=~
\dang{C[\dang{\!T_0\!},\dang{\!T_1\!},\ldots]}. \]
\end{theorem}
The proof of this is logically elementary but administratively complex:
see \cite[Section~6.1.2]{Longley-Normann}.
One further piece of machinery will be useful:
the notion of \emph{hereditary $\eta$-expansion}, which enables us to convert
a variable $x$ into a procedure term (written $x^\eta$).
Using this, the restriction that variables may appear only at the head of applications
can be seen to be inessential: e.g.\ the `illegal term' $f x$ may be replaced by the legal term $f x^\eta$.
The definition of $x^\eta$ is by recursion on the type of $x$:
if $x:\sigma_0,\ldots,\sigma_{r-1} \rightarrow \nat$, then
\[ x^\eta ~=~ \lambda z_0^{\sigma_0} \ldots z_{r-1}^{\sigma_{r-1}} . \;
\caseof{x z_0^\eta \ldots z_{r-1}^\eta}{i \Rightarrow i} \;. \]
In particular, if $x:\nat$ then $x^\eta = \lambda.\,\caseof{x}{i \Rightarrow i}$.
The following useful properties of $\eta$-expansion are proved in
\cite[Lemma~6.1.14]{Longley-Normann}:
\begin{lemma} \label{eta-properties}
$\dang{x^\eta \vec{q}} \;=\, \caseof{x \vec{q}}{i \Rightarrow i}$, and
$\dang{\lambda \vec{y}.\,p \vec{y}\,^\eta} \;=\, p$.
\end{lemma}
The sets ${\mathsf{SP}}(\sigma)$ may now be made into a total applicative
structure ${\mathsf{SP}}$ by defining
\[ (\lambda x_0 \cdots x_r.e) \cdot q ~=~
\lambda x_1 \cdots x_r.\dang{e[x_0 \mapsto q]} \;. \]
Clearly the sets ${\mathsf{SP}}^0(\sigma)$ are closed under this application
operation, so we also obtain an applicative substructure ${\mathsf{SP}}^0$ of ${\mathsf{SP}}$.
It is easy to check that application in ${\mathsf{SP}}$ is monotone and continuous
with respect to $\sqsubseteq$.
It is also shown in \cite[Section~6.1.3]{Longley-Normann} that both ${\mathsf{SP}}$ and ${\mathsf{SP}}^0$
are typed \emph{$\lambda$-algebras}:
that is, they admit a compositional interpretation of typed $\lambda$-terms that
validates $\beta$-equality.
(The relevant interpretation of pure $\lambda$-terms is in fact given by three of the
clauses from the interpretation of ${\mathrm{PCF}}^\Omega$ as defined in Subsection~\ref{subsec-interp} below.)
\subsection{Interpretation of ${\mathrm{PCF}}$ in ${\mathsf{SP}}^0$} \label{subsec-interp}
A central role will be played by certain procedures
$Y_\sigma \in {\mathsf{SP}}^0((\sigma\rightarrow\sigma)\rightarrow\sigma)$
which we use to interpret the ${\mathrm{PCF}}$ constants $Y_\sigma$
(the overloading of notation will do no harm in practice).
If $\sigma = \sigma_0,\ldots,\sigma_{r-1} \rightarrow \nat$,
we define $Y_\sigma = \lambda g^{\sigma\rightarrow\sigma}. F_\sigma[g]$,
where $F_\sigma[g]$ is specified corecursively up to $\alpha$-equivalence by:
\[ F_\sigma[g] ~=_\alpha~ \lambda x_0^{\sigma_0} \ldots x_{r-1}^{\sigma_{r-1}}.\;
\caseof{g\;(F_\sigma[g])\; x_0^\eta \cdots x_{r-1}^\eta}{i \Rightarrow i} \;. \]
(A concrete representative of $F_\sigma[g]$ satisfying the no-hiding condition will of course feature
a different choice of bound variables $x_0,\ldots,x_{r-1}$ at each level.)
We may now give the standard interpretation of ${\mathrm{PCF}}^\Omega$ in ${\mathsf{SP}}$.
To each ${\mathrm{PCF}}^\Omega$ term $\Gamma \vdash M:\sigma$ we associate
a procedure-in-environment $\Gamma \vdash \sem{M}^{{\mathsf{SP}}}_\Gamma : \sigma$
(denoted henceforth by $\sem{M}_\Gamma$)
inductively as follows:
\begin{eqnarray*}
\sem{x^\sigma}_\Gamma & = & x^{\eta} \\
\sem{\num{n}}_\Gamma & = & \lambda.n \\
\sem{{\mathit{suc}}}_\Gamma & = & \lambda x.\,\caseof{x}{i \Rightarrow i+1} \\
\sem{{\mathit{pre}}}_\Gamma & = & \lambda x.\,\caseof{x}{0 \Rightarrow 0 \mid i+1 \Rightarrow i} \\
\sem{{\mathit{ifzero}}}_\Gamma & = &
\lambda xyz.\,{\mathtt{case}}~{x}~{\mathtt{of}}~(0 \Rightarrow \caseof{y}{j \Rightarrow j} \\
& & ~~~~~~~~~~~~~~~\mid~i+1 \Rightarrow \caseof{z}{j \Rightarrow j}) \\
\sem{Y_\sigma}_\Gamma & = & Y_\sigma \\
\sem{C_f}_\Gamma & = & \lambda x.\,\caseof{x}{i \Rightarrow f(i)} \\
\sem{\lambda x^\sigma.M}_\Gamma & = & \lambda x.\sem{M}_{\Gamma,x} \\
\sem{MN}_\Gamma & = & \sem{M}_\Gamma \cdot \sem{N}_\Gamma
\end{eqnarray*}
(In the clause for $C_f$, we interpret $f(i)$ as $\bot$ whenever $f(i)$ is undefined.)
The following key property of $\sem{-}^{\mathsf{SP}}$ is shown as Theorem~7.1.16 in \cite{Longley-Normann}:
\begin{theorem}[Adequacy] \label{SP0-adequacy}
For any closed ${\mathrm{PCF}}^\Omega$ term $M:\nat$,
we have $M \rightsquigarrow^* \num{n}$ iff $\sem{M} = \lambda.n$.
\end{theorem}
We may now clarify the relationship between ${\mathsf{SP}}^0$ and ${\mathsf{SF}}$.
First, there is a natural `observational equivalence' relation $\approx$ on each ${\mathsf{SP}}^0(\sigma)$,
defined by
\[ q \approx q' \mbox{~~~iff~~~} \forall r \in {\mathsf{SP}}^0(\sigma\rightarrow\nat).~~
r \cdot q ~=~ r \cdot q' \;. \]
It is not hard to see that if $p \approx p' \in {\mathsf{SP}}^0(\sigma\rightarrow\tau)$ and $q \approx q' \in {\mathsf{SP}}^0(\sigma)$
then $p \cdot q \approx p' \cdot q \approx p' \cdot q' \in {\mathsf{SP}}^0(\tau)$.
Explicitly, the first of these equivalences holds because for any $r \in {\mathsf{SP}}^0(\tau\rightarrow\nat)$ we have
(using Lemma~\ref{eta-properties}) that
\[ r \cdot (p \cdot q) ~=~ (\lambda x.r(\lambda \vec{z}.xq\vec{z}^{\,\eta})) \cdot p
~=~ (\lambda x.r(\lambda \vec{z}.xq\vec{z}^{\,\eta})) \cdot p' ~=~ r \cdot (p' \cdot q) \;, \]
while the second equivalence holds because for any $r \in {\mathsf{SP}}^0(\tau\rightarrow\nat)$ we have
\[ r \cdot (p' \cdot q) ~=~ (\lambda y.r(\lambda \vec{z}.p'y^\eta \vec{z}^{\,\eta})) \cdot q
~=~ (\lambda y.r(\lambda \vec{z}.p'y^\eta \vec{z}^{\,\eta})) \cdot q' ~=~ r \cdot (p' \cdot q') \;. \]
We thus obtain a well-defined applicative structure ${\mathsf{SP}}^0 /\!\approx$ as a quotient of ${\mathsf{SP}}^0$;
we write $\theta : {\mathsf{SP}}^0 \rightarrow {\mathsf{SP}}^0 /\!\approx$ for the quotient map.
It turns out that up to isomorphism, this structure ${\mathsf{SP}}^0 /\!\approx$ is none other than ${\mathsf{SF}}$.
Indeed, in \cite{Longley-Normann} this was taken as the definition of ${\mathsf{SF}}$, and the
characterization as the closed term model of ${\mathrm{PCF}}^\Omega$ modulo observational equivalence
proved as a consequence.
In order to fill out the picture a little more,
we will here exhibit the equivalence of these two definitions as a consequence of
the following non-trivial fact, given as Corollary~7.1.34 in \cite{Longley-Normann}:%
\footnote{What is actually shown in \cite{Longley-Normann} is that every element of ${\mathsf{SP}}^0$ is denotable
on the nose in a language ${\mathrm{PCF}} + \mathit{byval}$, with a certain choice of denotation for the
constant $\mathit{byval}$.
Since the latter satisfies
$\sem{\mathit{byval}} \approx \sem{\lambda f^{\nat\rightarrow\nat}x^\nat.\,{\mathit{ifzero}}\;x\,(fx)(fx)}$,
the present Theorem~\ref{SP0-completeness-thm} follows easily.}
\begin{theorem} \label{SP0-completeness-thm}
For every $p \in {\mathsf{SP}}^0(\sigma)$, there is a closed ${\mathrm{PCF}}^\Omega$ term $M: \sigma$ such that
$\sem{M} \approx p$.
\end{theorem}
\begin{proposition} \label{SF-char}
$({\mathsf{SP}}^0/\!\approx) \cong{\mathsf{SF}}$, via an isomorphism that identifies $\theta(\sem{M}^{\mathsf{SP}})$
with $\sem{M}^{\mathsf{SF}}$ for any closed ${\mathrm{PCF}}^\Omega$ term $M$.
\end{proposition}
\begin{proof}
For any element $x \in {\mathsf{SP}}^0/\!\approx$, we may take $p \in {\mathsf{SP}}^0$ with $\theta(p)=x$,
then by Theorem~\ref{SP0-completeness-thm} take $M$ closed with $\sem{M} \approx p$;
we then have that $\theta(\sem{M}) = x$.
In this way, each ${\mathsf{SP}}^0(\sigma)/\!\approx$ corresponds bijectively to the set of
closed ${\mathrm{PCF}}^\Omega$ terms $M:\sigma$ modulo some equivalence relation $\sim$.
Recall now that $\simeq$ denotes observational equivalence in ${\mathrm{PCF}}^\Omega$.
To see that $\sim \,\subseteq\, \simeq$, suppose $M \sim M'$, and let $C[-]$ be any suitable observing
context of ${\mathrm{PCF}}^\Omega$. By the compositionality of $\sem{-}^{\mathsf{SP}}$, we obtain some $\sem{C} \in {\mathsf{SP}}^0$
such that $\sem{C[M]} = \sem{C} \cdot \sem{M}$ and similarly for $M'$.
But $M \sim M'$ means that $\sem{M} \approx \sem{M'}$, whence by the definition of $\approx$
we conclude that $\sem{C[M]} = \sem{C[M']}$ at type $\nat$.
So by Theorem~\ref{SP0-adequacy} we have
$C[M] \rightsquigarrow^* \num{n}$ iff $C[M'] \rightsquigarrow^* \num{n}$.
Since $C[-]$ was arbitrary, we have shown that $M \simeq M'$.
To see that $\simeq \,\subseteq\, \sim$, suppose $M \simeq M' : \sigma$.
It will suffice to show that $\sem{M} \approx \sem{M'}$.
So suppose $r \in {\mathsf{SP}}^0(\sigma\rightarrow\nat)$, and
using Theorem~\ref{SP0-completeness-thm}, take a ${\mathrm{PCF}}^\Omega$ term $R$ such that
$\sem{R} \approx r$.
Then $r \cdot \sem{M} = \sem{R} \cdot {M} = \sem{RM}$ at type $\nat$, and similarly for $M'$.
But since $M \simeq M'$, we have $RM \rightsquigarrow^* n$ iff $RM' \rightsquigarrow^* n$,
whence $\sem{RM} = \sem{RM'}$ by Theorem~\ref{SP0-adequacy}.
Thus $r \cdot \sem{M} = r \cdot \sem{M'}$, and we have shown $\sem{M} \approx \sem{M'}$.
Since each ${\mathsf{SF}}(\sigma)$ consists of closed ${\mathrm{PCF}}^\Omega$ terms $M:\sigma$ modulo $\simeq$,
we have established a bijection $({\mathsf{SP}}^0(\sigma) /\!\approx) \cong {\mathsf{SF}}(\sigma)$ for each $\sigma$.
Moreover, both $\sem{-}^{\mathsf{SP}}$ and $\theta$ respect application, so it follows that
$({\mathsf{SP}}^0 /\!\approx) \cong {\mathsf{SF}}$, and it is immediate by construction that $\theta(\sem{M}^{\mathsf{SP}})$
is identified with $\sem{M}^{\mathsf{SF}}$.
$\Box$
\end{proof}
\vspace*{1.0ex}
As we have already seen in Subsection~\ref{subsec-PCF},
Milner's context lemma for ${\mathrm{PCF}}^\Omega$ implies that
${\mathsf{SF}}$ is extensional.
From this and Proposition~\ref{SF-char}, we may now read off the following
useful characterization of the equivalence $\approx$:
\begin{lemma}[NSP context lemma] \label{nsp-eq-context-lemma}
Suppose $p,p' \in {\mathsf{SP}}^0(\sigma_0,\ldots,\sigma_{r-1} \rightarrow \nat)$.
Then $p \approx p'$ iff
\[ \forall q_0 \in {\mathsf{SP}}^0(\sigma_0), \ldots, q_{r-1} \in {\mathsf{SP}}^0(\sigma_{r-1}). ~~
p \cdot q_0 \cdot \ldots \cdot q_{r-1} ~=~ p' \cdot q_0 \cdot \ldots \cdot q_{r-1} \;. \]
\end{lemma}
We shall also make use of the \emph{observational ordering} on ${\mathsf{SF}}$
and the associated preorder on ${\mathsf{SP}}$.
Let $\sqsubseteq$ be the usual information ordering on ${\mathsf{SF}}(\nat) \cong \mathbb{N}_\bot$,
and let us endow each ${\mathsf{SF}}(\sigma)$ with the partial order $\preceq$ defined by
\[ x \preceq x' \mbox{~~~iff~~~} \forall h \in {\mathsf{SF}}(\sigma\rightarrow\nat),\, n \in \mathbb{N}.~~
h \cdot x ~\sqsubseteq~ h \cdot x' \;. \]
It is easy to see that the application operations $\cdot$ are monotone with respect to $\preceq$.
Moreover, Milner's context lemma also exists in an \emph{inequational} form which says, in effect, that
if $f,f' \in {\mathsf{SF}}(\sigma_0,\ldots,\sigma_{r-1}\rightarrow\nat)$ then
\[ f \preceq f' \mbox{~~~iff~~~} \forall y_0 \in {\mathsf{SF}}(\sigma_0),\ldots,y_{r-1} \in {\mathsf{SF}}(\sigma_{r-1}).~~
f \cdot y_0 \cdot \ldots \cdot y_{r-1} ~\sqsubseteq f' \cdot y_0 \cdot \ldots \cdot y_{r-1} \;. \]
Thus, if elements of ${\mathsf{SF}}(\sigma\rightarrow\tau)$ are considered as functions ${\mathsf{SF}}(\sigma) \rightarrow {\mathsf{SF}}(\tau)$,
the partial order $\preceq$ coincides with the pointwise partial order on functions.
We write $\preceq$ also for the preorder on each ${\mathsf{SP}}^0(\sigma)$
induced by $\preceq$ on ${\mathsf{SF}}$: that is, $p \preceq p'$ iff $\theta(p) \preceq \theta(p')$.
Furthermore, we extend the use of the notations $\approx,\,\preceq$ in a natural way to open terms
(in the same environment), and indeed to meta-terms:
e.g.\ we may write $\vec{x} \vdash P \preceq P'$ to mean
$\dang{\lambda \vec{x}.P} \;\preceq\; \dang{\lambda \vec{x}.P'}$.
Clearly $p \approx p'$ iff $p \preceq p' \preceq p$.
The following is also now immediate:
\begin{lemma} \label{nsp-ineq-context-lemma}
Suppose $p,p' \in {\mathsf{SP}}^0(\sigma)$ where $\sigma = \sigma_0,\ldots,\sigma_{r-1} \rightarrow \nat$.
Then the following are equivalent:
(i) $p \preceq p'$.
(ii) \( \forall r \in {\mathsf{SP}}^0(\sigma\rightarrow\nat).~ r \cdot p \sqsubseteq r \cdot p' \).
(iii) \( \forall q_0 \in {\mathsf{SP}}^0(\sigma_0), \ldots, q_{r-1} \in {\mathsf{SP}}^0(\sigma_{r-1}). ~~
p \cdot q_0 \cdot \ldots \cdot q_{r-1} \,\sqsubseteq\, p' \cdot q_0 \cdot \ldots \cdot q_{r-1} \).
\end{lemma}
We conclude this subsection by reformulating some of the major milestones in our proof
using the notation now available.
Specifically, in Sections~\ref{sec-denotations} and \ref{sec-Y-k+1} we will show the following:
\begin{theorem} \label{SP0-main-thm}
For any $k \geq 1$, the element $\sem{Y_{k+1}} \in {\mathsf{SP}}^0$ is not ${\mathrm{PCF}}^\Omega_k$-denotable
(whence neither is $\sem{Y_{0 \rightarrow (k+1)}}$).
\end{theorem}
In Section~\ref{sec-extensional} we will go on
to show that no $Z \approx \sem{Y_{0 \rightarrow (k+1)}}$ can be ${\mathrm{PCF}}^\Omega_k$-denotable,
establishing Theorem~\ref{SF-main-thm}(i).
In Section~\ref{sec-pure-type} we will resort to a more refined version of our methods
to show the same for $Y_{k+1}$;
this will establish Theorem~\ref{SF-main-thm}(ii).
\subsection{The embeddability hierarchy} \label{subsec-embed}
The following result will play a crucial role in this paper:
\begin{theorem}[Strictness of embeddability hierarchy] \label{no-retraction-thm}
In ${\mathsf{SF}}$, no type $\pure{k+1}$ can be a \emph{pseudo-retract}
of any finite product $\Pi_i \sigma_i$ where each $\sigma_i$
is of level $\leq k$.
More formally, if $z$ is a variable of type $\pure{k+1}$ and each $x_i$ a variable of type $\sigma_i$,
there cannot exist procedures
\[ z \vdash t_i : \sigma_i \;, ~~~~~~ \vec{x} \vdash r : \pure{k+1} \]
such that $z \vdash {r[\vec{x} \mapsto \vec{t}\,]} \,\succeq z^\eta$.
\end{theorem}
If in the above setting we had $z \vdash {r[\vec{x} \mapsto \vec{t}\,]} \,\approx z^\eta$,
we would call $\pure{k+1}$ a \emph{retract} of $\Pi_i \sigma_i$.
In Appendix A we will show that the notions of retract and pseudo-retract actually coincide,
since $z \vdash p \succeq z^\eta$ implies $z \vdash p \approx z^\eta$.
However, this fact will not be needed for the main results of this paper.
In our statement of Theorem~\ref{no-retraction-thm},
we have referred informally to a product $\Pi_i \sigma_i$ which we have
not precisely defined (although the formal statement of the theorem gives
everything that is officially necessary).
One may readily make precise sense of this product notation within the
\emph{Karoubi envelope} ${\mathbf{K}}({\mathsf{SF}})$ as studied in \cite[Chapter~4]{Longley-Normann}:
for instance, it is not hard to show that any finite product of level $\leq k$ types can
be constructed as a retract of the pure type $\pure{k+1}$.
In the present paper, however, references to product types may be taken to be
purely informal and motivational.
The proof of Theorem~\ref{no-retraction-thm} appears in
\cite[Section~7.7]{Longley-Normann}, but because of its crucial role in the paper
we reprise it here with some minor stylistic improvements.
\vspace*{1.0ex}
\begin{proof}
By induction on $k$.
For the case $k=0$, we note that
$\nat\rightarrow\nat$ cannot be a pseudo-retract of any $\nat^r$, since (for example)
the set of maximal elements in ${\mathsf{SF}}(\nat\rightarrow\nat)$ is of larger cardinality than
the set of all elements of ${\mathsf{SF}}(\nat)^r$.
(Alternatively, one may note that $\nat\rightarrow\nat$ is not a \emph{retract} of $\nat^r$,
since the former contains strictly ascending chains of length $r+2$ while the latter does not;
then use the method of Appendix~A in the easy case $k=1$ to show that
any pseudo-retraction of the relevant type would be a retraction.)
Now assume the result for $k-1$, and suppose
for contradiction that $z \vdash t_i$ and $\vec{x} \vdash r$ exhibit $\pure{k+1}$
as a pseudo-retract of $\Pi_i \sigma_i$ where each $\sigma_i$ is of level $\leq k$.
Let $v =\, \dang{r[\vec{x} \mapsto \vec{t}\,]}$,
so that $z \vdash v \succeq z^\eta$,
whence $\dang{v[z \mapsto u]}\, \succeq u$ for any $u \in {\mathsf{SP}}^0(k+1)$.
We first check that any $v$ with this latter property must have the syntactic form
$\lambda f^k.\,\caseof{zp}{\cdots}$ for some $p$ of type $\pure{k}$.
Indeed, it is clear that $v$ does not have the form
$\lambda f.n$ or $\lambda f.\bot$, and the only other alternative form
is $\lambda f.\,\caseof{fp'}{\cdots}$.
In that case, however, we would have
\[ \dang{v[z \mapsto \lambda w^k.0]} \cdot\; (\lambda y^{k-1}.\bot) ~=~ \bot \;, \]
contradicting
$\dang{v[z \mapsto \lambda w^k.0]} \cdot\; (\lambda y^{k-1}.\bot) \,\succeq\, (\lambda w.0)(\lambda y.\bot) = 0$.
We now focus on the subterm $p$ in $v = \lambda f^k.\,\caseof{zp}{\cdots}$.
The general direction of our argument will be to show that
$\lambda f^k.p$ represents a function of type $\pure{k} \rightarrow \pure{k}$
that dominates the identity, and that moreover our construction
of $v$ as $\dang{r[\vec{x} \mapsto \vec{t}\,]}$ can be used to split this into functions
$\pure{k} \rightarrow \Pi_j \rho_j$ and $\Pi_j \rho_j \rightarrow \pure{k}$ where the
$\rho_j$ are of level $\leq k-1$, contradicting the induction hypothesis.
An apparent obstacle to this plan is that $z$ as well as $f$ may appear
free in $p$; however, it turns out that we still obtain all the properties
we need if we specialize $z$ here (somewhat arbitrarily) to $\lambda w.0$.
Specifically, we claim that $\lambda f.\!\dang{p[z \mapsto \lambda w.0]} \,\succeq {\mathit{id}}_k$.
By Lemma~\ref{nsp-ineq-context-lemma}, it will suffice to show that
$\dang{p[f \mapsto q, z \mapsto \lambda w.0]} \,\succeq q$ for any $q \in {\mathsf{SP}}^0(k)$.
The idea is that if it is not, then (ignoring the presence of $z$ in $p$ for now)
we may specialize $z$ to some $u$ that will detect the difference
between $p[f \mapsto q]$ and $q$, so that the subterm `$zp$' within $v$ will yield $\bot$,
contradicting that $z \vdash v \succeq z^\eta$.
We can even allow for the presence of $z$ in $p$ by a suitably careful choice of $u$.
Again by Lemma~\ref{nsp-ineq-context-lemma}, it suffices to show that
$\dang{p[f \mapsto q, z \mapsto \lambda w.0]} \cdot\; s \succeq q \cdot s$ for any $s$.
So suppose $q \cdot s = \lambda.n$ whereas
$\dang{p[f \mapsto q, z \mapsto \lambda w.0]} \cdot\; s \neq \lambda.n$
for some $s \in {\mathsf{SP}}^0(k-1)$ and $n \in \mathbb{N}$.
Take $u = \lambda g.\,\caseof{gs}{n \Rightarrow 0}$,
so that $u \cdot q' = \bot$ whenever $q' \cdot s \neq \lambda.n$.
Then $u \preceq \lambda w.0$ by Lemma~\ref{nsp-ineq-context-lemma},
so we have $\dang{p[f \mapsto q, z \mapsto u]} \cdot\; s \neq \lambda.n$
since $\lambda.n$ is maximal in ${\mathsf{SP}}^0(\nat)$.
By the definition of $u$, it follows that
$\dang{(zp)[f \mapsto q, z \mapsto u]}\, = \bot$, whence
$\dang{v[z \mapsto u]} \cdot\, q = \bot$, whereas $u \cdot q=0$,
contradicting $\dang{v[z \mapsto u]}\, \succeq u$.
This completes the proof that
$\lambda f.\!\dang{p[z \mapsto \lambda w.0]} \; \succeq {\mathit{id}}_{k}$.
Next, we show how to split the function represented by this procedure
so as to go through some $\Pi_j \rho_j$ as above.
Since $\dang{r[\vec{x} \mapsto \vec{t}\,]} \,= \lambda f.\,\caseof{zp}{\cdots}$, we have that
$r[\vec{x}\mapsto\vec{t}\,]$ reduces in finitely many steps to a head normal form
$\lambda f.\,\caseof{zP}{\cdots}$ where $\dang{\!P\!}\,=p$.
By working backward through this reduction sequence, we may locate
the ancestor within $r[\vec{x}\mapsto\vec{t}\,]$ of this head occurrence of $z$.
Since $z$ does not appear free in $r$, this occurs within some $t_i$, and clearly it must
appear as the head of some subterm $\caseof{zp'}{\cdots}$.
Now since $t_i$ has type $\sigma_i$ of level $\leq k$,
and $z:\pure{k+1}$ is its only free variable,
it is easy to see that all \emph{bound}
variables within $t_i$ have pure types of level $<k$.
Let $x'_0,x'_1,\ldots$ denote the finitely many bound variables that are in scope
at the relevant occurrence of $zp'$, and suppose each $x'_j$ has type
$\rho_j$ of level $<k$.
By considering the form of the head reduction sequence
$r[\vec{x}\mapsto\vec{t}\,] \rightsquigarrow^*_h \lambda f.\,\caseof{zP}{\cdots}$,
we now see that $P$ has the form $p'[\vec{x}\,' \mapsto \vec{T}]$
where each $T_j: \rho_j$ contains at most $f$ and $z$ free.%
\footnote{The reader wishing to see a more formal justification for this step may consult the
proof of Lemma~\ref{g-lemma-1}(i) below.}
Writing $^*$ for the substitution $[z \mapsto \lambda w.0]$,
define procedures
\[ f^k \vdash t'_j =\, \dang{T_j^*} \,: \rho_j \;, ~~~~~~
\vec{x}\,' \vdash r' =\, \dang{p'^*} \,: \pure{k} \;. \]
Then $\dang{r'[\vec{x}\,' \mapsto \vec{t}\,'\,]}$ coincides with the term
$\dang{\lambda f.\,P^*} \,= \lambda f.\!\dang{\!p^*\!}$,
which dominates the identity as shown above.
Thus $\pure{k}$ is a pseudo-retract of $\Pi_j \rho_j$,
which contradicts the induction hypothesis.
So $\pure{k+1}$ is not a pseudo-retract of $\Pi_i \sigma_i$ after all,
and the proof is complete.
$\Box$
\end{proof}
\vspace*{1.0ex}
As an aside, we remark that for several extensions of ${\mathrm{PCF}}$ studied in the literature,
the situation is completely different, in that the corresponding fully abstract and universal models
possess a \emph{universal} simple type $\upsilon$ of which all simple types are retracts.
It follows easily in these cases that one can indeed bound the type levels of recursion operators
without loss of expressivity. For example:
\begin{itemize}
\item In the language ${\mathrm{PCF}}+{\mathit{por}}+{\mathit{exists}}$ considered by Plotkin \cite{LCF-considered},
the type $\nat\rightarrow\nat$ is universal, and the proof of this shows that every program in this
language is observationally equivalent to one in ${\mathrm{PCF}}_1 + {\mathit{por}}+{\mathit{exists}}$.
(This latter fact was already noted in \cite{LCF-considered}.)
\item In ${\mathrm{PCF}}+{\mathit{catch}}$ (a slight strengthening of Cartwright and Felleisen's language SPCF
\cite{Cartwright-Felleisen}), the type $\nat\rightarrow\nat$ is again universal, and again
the sublanguage ${\mathrm{PCF}}_1 + {\mathit{catch}}$ has the same expressive power.
\item In the language ${\mathrm{PCF}}+H$ of Longley \cite{SRF}, the type $(\nat\rightarrow\nat)\rightarrow\nat$
is universal, but even here, all constants $Y_\sigma$ with ${\mathrm{lv}}(\sigma)>1$ are dispensable.
\end{itemize}
Further details of each the above scenarios may be found in \cite{Longley-Normann}.
These facts may offer some indication of why a `cheap' proof of our present results
in the setting of pure ${\mathrm{PCF}}$ is not to be expected.%
\footnote{That PCF manifests greater structural complexity than many stronger languages
is also a moral of Loader's undecidability theorem for finitary PCF \cite{Loader-PCF}.
However, the complexity we explore here seems quite orthogonal to that exhibited by Loader:
we are concerned purely with `infinitary' aspects of definability, the entire finitary structure
being already represented by our ${\mathrm{PCF}}_0$.}
\subsection{Other sublanguages of ${\mathrm{PCF}}$} \label{subsec-power-pcf1}
Our main theorems establish a hierarchy of languages ${\mathrm{PCF}}_1 < {\mathrm{PCF}}_2 < \cdots\,$.
Before proceeding further, however, we pause to clarify the relationship between ${\mathrm{PCF}}_0$ and ${\mathrm{PCF}}_1$,
and also to survey some of the interesting territory that lies between them,
in order to situate our theorems within a wider picture.
On the one hand, ${\mathrm{PCF}}_0$ is a rather uninteresting language.
As regards the elements of ${\mathsf{SF}}$ that it denotes, it is equivalent in expressivity to ${\mathrm{PCF}}_\bot$,
a variant of ${\mathrm{PCF}}_0$ in which we replace $Y_0$ by a constant $\bot$ (denoting $\bot \in {\mathsf{SF}}(\nat)$).
This is clear since $Y_0$ and $\bot$ are interdefinable: we have $\bot = Y_0(\lambda x.x)$,
and it is easy to see that $Y_0 = \lambda f. f\bot$ (in ${\mathsf{SF}}$).
By a syntactic analysis of the possible normal forms of type $\pure{1}$ in ${\mathrm{PCF}}_\bot$,
one can show that these are very weak languages that do not even define addition.
However, such an analysis is unnecessary for our purposes,
since there are more interesting languages that clearly subsume ${\mathrm{PCF}}_0$
but are known to be weaker than ${\mathrm{PCF}}_1$.
For instance, Berger \cite{Berger-min-recursion} considered the language ${\mathrm{T}}_0 +{\mathit{min}}$,
where ${\mathrm{T}}_0$ (a fragment of G\"odel's System~T) is the $\lambda$-calculus with
first-order primitive recursion over the natural numbers:
\[ \num{0} ~: \nat \;, ~~~~~~ {\mathit{suc}} ~: \mathbb{N} \rightarrow \mathbb{N} \;, ~~~~~~
{\mathit{rec}}_0 ~: \nat \rightarrow (\nat \rightarrow \nat \rightarrow \nat) \rightarrow (\nat \rightarrow \nat) \;, \]
and ${\mathit{min}}$ is the classical \emph{minimization} (i.e.\ unbounded search)
operator of type $(\nat\rightarrow\nat)\rightarrow\nat$.
On the one hand, it is an easy exercise to define $\bot$ in ${\mathrm{T}}_0 + {\mathit{min}}$,
and to define both ${\mathit{rec}}_0$ and ${\mathit{min}}$ in ${\mathrm{PCF}}_1$.
On the other hand, Berger showed that the ${\mathrm{PCF}}_1$-definable functional
$\Phi_0: (\nat\rightarrow\nat\rightarrow\nat) \rightarrow (\nat \rightarrow \nat)$ given by
\[ \Phi_0\;g\;n ~=~ g\;n\;(\Phi_0\;g\;(n+1)) \;, \]
is not expressible in ${\mathrm{T}}_0 + {\mathit{min}}$.%
\footnote{Berger actually considered denotability in the Scott model,
but his argument applies equally to ${\mathsf{SF}}$.}
As already indicated in Section~\ref{sec-intro},
this functional and its higher-type analogues will play a crucial role in the present paper.
This situation is revisited in \cite[Section~6.3]{Longley-Normann} from the perspective of substructures
of ${\mathsf{SP}}^0$. It is shown that ${\mathrm{T}}_0 + {\mathit{min}}$, and indeed the whole of ${\mathrm{T}} + {\mathit{min}}$,
can be modelled within
the substructure ${\mathsf{SP}}^{0,{\mbox{\scriptsize \rm lwf}}}$ of \emph{left-well-founded} procedures, whereas the above functional
$\Phi_0$ is not representable by any such procedure; thus $\Phi_0$ is not expressible in ${\mathrm{T}} + {\mathit{min}}$.
(The reader may wish to study these results and proofs before proceeding further, since they provide
simpler instances of the basic method that we will use in this paper.)
At third order, there are even `hereditarily total' functionals definable in ${\mathrm{PCF}}_1$ but not
by higher-type iterators, one example being the well-known \emph{bar recursion} operator
(see \cite{Longley-bar-recursion}).
Even weaker than ${\mathrm{T}}_0 + {\mathit{min}}$ is the language of \emph{(strict) Kleene primitive recursion
plus minimization}, denoted by ${\mathsf{Klex}}^{\mbox{\scriptsize \rm min}}$ in \cite{Longley-Normann};
this again subsumes ${\mathrm{PCF}}_0$.
It is shown in \cite{Longley-Normann} that the computational power of ${\mathsf{Klex}}^{\mbox{\scriptsize \rm min}}$
coincides with that of computable \emph{left-bounded} procedures;
this is used to show, for example, that even ${\mathit{rec}}_0$ is not computable in ${\mathsf{Klex}}^{\mbox{\scriptsize \rm min}}$.
We find it reasonable to regard left-bounded procedures
as embodying the weakest higher-order computability notion of natural interest that is still Turing complete.
\section{Sequential procedures for ${\mathrm{PCF}}_k$ terms} \label{sec-denotations}
For the remainder of the paper, we take $k$ to be some fixed natural number greater than $0$.
In this section we give a direct inductive characterization of the ${\mathrm{PCF}}^\Omega_k$-denotable
elements of ${\mathsf{SP}}$ by making explicit how our interpretation works for terms of ${\mathrm{PCF}}^\Omega_k$.
The first point to observe is that we may restrict attention to ${\mathrm{PCF}}^\Omega_k$ terms in
\emph{long $\beta\eta$-normal form}: that is, terms in $\beta$-normal form in which every
variable or constant $z$ of type $\sigma_0,\ldots,\sigma_{r-1} \rightarrow \nat$ is fully applied
(i.e.\ appears at the head of a subterm $z N_0 \ldots N_{r-1}$ of type $\nat$).
Moreover, an inductive characterization of the class of such terms is easily given.
\begin{proposition} \label{long-beta-eta-prop}
(i) A procedure $\Gamma \vdash p : \sigma$ is denotable by a ${\mathrm{PCF}}^\Omega_k$ term
$\Gamma \vdash M: \sigma$ iff it is denotable by one in long $\beta\eta$-normal form.
(ii) The class of long $\beta\eta$-normal forms of ${\mathrm{PCF}}^\Omega_k$ is inductively generated by
the following clauses:
\begin{enumerate}
\item If $\Gamma \vdash N_i : \sigma_i$ is a normal form for each $i<r$
and $x^{\sigma_0,\ldots,\sigma_{r-1} \rightarrow \nat} \in \Gamma$,
then $\Gamma \vdash x N_0 \ldots N_{r-1} : \nat$ is a normal form (note that $r$ may be $0$ here).
\item If $\Gamma, x^\sigma \vdash M : \tau$ is a normal form then so is $\Gamma \vdash \lambda x.M : \sigma \rightarrow \tau$.
\item The numeric literals $\Gamma \vdash \num{n} : \nat$ are normal forms.
\item If $\Gamma \vdash M : \nat$ is a normal form then so are $\Gamma \vdash {\mathit{suc}}\;M : \nat$,
$\Gamma \vdash {\mathit{pre}}\;M : \nat$ and $\Gamma \vdash C_f\,M : \nat$ for any $f: \mathbb{N} \rightharpoonup \mathbb{N}$.
\item If $\Gamma \vdash M : \nat$, $\Gamma \vdash N : \nat$ and $\Gamma \vdash P : \nat$ are
normal forms, then so is $\Gamma \vdash {\mathit{ifzero}}\;M\;N\;P : \nat$.
\item If $\sigma = \sigma_0,\ldots,\sigma_{r-1} \rightarrow \nat$ is of level $\leq k$
and $\Gamma \vdash M : \sigma\rightarrow\sigma$ and $\Gamma \vdash N_i : \sigma_i$ are normal forms,
then $\Gamma \vdash Y_\sigma M N_0 \ldots N_{r-1} : \nat$ is a normal form.
\end{enumerate}
\end{proposition}
\begin{proof}
(i) It is a well-known property of simply typed $\lambda$-calculi that every term $M$ is $\beta\eta$-equivalent
to one in long $\beta\eta$-normal form: indeed, we may first compute the $\beta$-normal form of $M$
and then repeatedly apply the $\eta$-rule to expand any subterms that are not already fully applied.
Moreover, it is shown in \cite[Theorem~6.1.18]{Longley-Normann} that ${\mathsf{SP}}$ is a $\lambda\eta$-algebra,
so that if $\Gamma \vdash M =_{\beta\eta} M'$ then $\sem{M}_\Gamma = \sem{M'}_\Gamma$
in ${\mathsf{SP}}$. This establishes the claim.
(ii) This is clear from the fact that no application may be headed by a $\lambda$-abstraction and that
all occurrences of variables and constants must be fully applied.
$\Box$
\end{proof}
\vspace*{1.0ex}
It follows that the class of ${\mathrm{PCF}}^\Omega_k$-denotable procedures may be generated inductively
by a set of clauses that mirror the above formation rules for long $\beta\eta$-normal ${\mathrm{PCF}}^\Omega_k$ terms.
We now consider each of these formation rules in turn in order to spell out the corresponding
operation at the level of NSPs. In Section~\ref{sec-Y-k+1} we will show that these operations cannot give rise
to \emph{$k\!+\!1$-spinal} procedures, from which it will follow that no ${\mathrm{PCF}}^\Omega_k$-denotable
procedure can be $k\!+\!1$-spinal.
For the first three formation rules, the effect on NSPs is easily described:
\begin{proposition} \label{NSP-nf-prop}
(i) If $\Gamma \vdash x N_0 \ldots N_{r-1} : \nat$ in ${\mathrm{PCF}}^\Omega$, then
$\sem{x N_0 \ldots N_{r-1}}_\Gamma = \caseof{x \sem{N_0}_\Gamma \cdots \sem{N_{r-1}}_\Gamma}{j \Rightarrow j}$.
(ii) If $\Gamma \vdash \lambda x.M : \sigma \rightarrow \tau$ in ${\mathrm{PCF}}^\Omega$, then
$\sem{\lambda x.M}_\Gamma = \lambda x.\,\sem{M}_{\Gamma,x}$.
(iii) $\sem{\num{n}}_\Gamma = \lambda.n$.
\end{proposition}
\begin{proof}
(i) Easy using the definition of $\sem{-}$ and Lemma~\ref{eta-properties}.
(ii), (iii) are part of the definition of $\sem{-}$.
$\Box$
\end{proof}
\vspace*{1.0ex}
As regards the formation rules for ${\mathit{suc}}$, ${\mathit{pre}}$, $C_f$ and ${\mathit{ifzero}}$,
the situation is again fairly straightforward, although a little more machinery is needed:
\begin{definition} \label{rightward-leaf-def}
(i) The set of \emph{rightward} (occurrences of) \emph{numeral leaves} within
a term $t$ is defined inductively by means of the following clauses:
\begin{enumerate}
\item A term $n$ is a rightward numeral leaf within itself.
\item Every rightward numeral leaf within $e$ is also one within $\lambda \vec{x}.e$.
\item Every rightward numeral leaf in each $e_i$ is also one in $\caseof{a}{i \Rightarrow e_i}$.
\end{enumerate}
(ii) If $t$ is a term and $e_i$ an expression for each $i$,
let $t[i \mapsto e_i]$ denote the result of replacing each rightward leaf occurrence $i$
in $t$ by the corresponding $e_i$.
\end{definition}
\begin{lemma} \label{rightward-leaf-prop}
$\dang{\caseof{d}{i \Rightarrow e_i}} \;= d[i \mapsto e_i]$
for any expressions $d,e_i$.
\end{lemma}
\begin{proof}
For each $c \in \mathbb{N}$, define a `truncation' operation $-^{(c)}$ on expressions as follows:
\begin{eqnarray*}
n^{(c)} ~=~ n \;, ~~~~ \bot^{(c)} & = & \bot \;, \\
\caseof{a}{i \Rightarrow e_i}^{(0)} & = & \bot \;, \\
\caseof{a}{i \Rightarrow e_i}^{(c+1)} & = & \caseof{a}{i \Rightarrow e_i^{(c)}} \;.
\end{eqnarray*}
Then clearly $d = \bigsqcup_c d^{(c)}$ and
$d[i \mapsto e_i] = \bigsqcup_c d^{(c)}[i \mapsto e_i]$.
Moreover, we may show by induction on $c$ that
\[ \dang{\caseof{d^{(c)}}{i \Rightarrow e_i}} ~=~ d^{(c)}[i \mapsto e_i] \;. \]
The case $c=0$ is trivial since $d^{(0)}$ can only have the form $n$ or $\bot$.
For the induction step, the situation for $d=n,\bot$ is trivial, so let us suppose
$d = \caseof{a}{j \Rightarrow f_j}$. Then
\begin{eqnarray*}
& & \dang{\caseof{d^{(c+1)}}{i \Rightarrow e_i}} \\
&=& \dang{\caseof{(\caseof{a}{j \Rightarrow f_j^{(c)}})}{i \Rightarrow e_i}} \\
&=& \caseof{a}{j \Rightarrow\, \dang{\caseof{f_j^{(c)}}{i \Rightarrow e_i}}} \\
&=& \caseof{a}{j \Rightarrow (f_j^{(c)}[i \mapsto e_i])} \mbox{~~by the induction hypothesis} \\
&=& (\caseof{a}{j \Rightarrow f_j^{(c)}})[i \mapsto e_i] \\
&=& d^{(c+1)}[i \mapsto e_i] \;.
\end{eqnarray*}
Since $\dang{\!-\!}$ is continuous, the proposition follows by taking the supremum over $c$.
$\Box$
\end{proof}
\vspace*{1.0ex}
From this lemma we may now read off the operations on NSPs that correspond to clauses~4 and 5
of Proposition~\ref{long-beta-eta-prop}(ii):
\begin{proposition} \label{NSP-arith-prop}
(i) If $\Gamma \vdash M : \nat$ in ${\mathrm{PCF}}^\Omega$, then
$\sem{C_f\,M}_\Gamma = \sem{M}_\Gamma [i \mapsto f(i)]$
(understanding $f(i)$ to be $\bot$ when $i \not\in {\mathrm{dom}}\;f$); similarly for ${\mathit{suc}}$ and ${\mathit{pre}}$.
(ii) If $\Gamma \vdash M : \nat$, $\Gamma \vdash N : \nat$ and $\Gamma \vdash P : \nat$, then
$\sem{{\mathit{ifzero}}\;M\;N\;P}_\Gamma = \sem{M}_\Gamma [0 \mapsto d,\, i+1 \mapsto e]$
where $\sem{N}_\Gamma = \lambda.d$ and $\sem{P}_\Gamma = \lambda.e$.
\end{proposition}
\begin{proof}
(i) The definition of $\sem{-}$ yields
\[ \sem{C_f\,M}_\Gamma ~=
\dang{\lambda.\,\caseof{\sem{M}_\Gamma}{i \Rightarrow f(i)}} \, , \]
and by Lemma~\ref{rightward-leaf-prop} this evaluates to $\sem{M}_\Gamma [i \mapsto f(i)]$.
Likewise for ${\mathit{suc}}$ and ${\mathit{pre}}$.
(ii) The definition of $\sem{-}$ yields
\[ \sem{{\mathit{ifzero}}\;M\;N\;P}_\Gamma ~=~
\dang{\lambda.\,\caseof{\sem{M}_\Gamma}{0 \Rightarrow d \mid i+1 \Rightarrow e}} \, , \]
and by Lemma~\ref{rightward-leaf-prop} this evaluates to
$\sem{M}_\Gamma [0 \mapsto d,\, i+1 \mapsto e]$.
$\Box$
\end{proof}
\vspace*{1.0ex}
It remains to consider the formation rule involving $Y_\sigma$.
It will be convenient to regard the NSP for $Y M N_0 \ldots N_{r-1}$ as a result of
plugging some simpler NSPs together, in the sense indicated by the following definition.
Here and later, we shall follow the convention that Greek capitals $\Gamma,\Delta$ denote arbitrary
environments, while Roman capitals $Z,X,V$ denote lists of variables of type level $\leq k$.
(Of course, the idea of plugging can be formulated without any restrictions on types,
but we wish to emphasize at the outset that only pluggings at level $\leq k$
will feature in our proof.)
\begin{definition}[Plugging] \label{plugging-def}
Suppose given the following data:
\begin{itemize}
\item a variable environment $\Gamma$,
\item a finite list $Z$ of `plugging variables' $z$ of level $\leq k$,
disjoint from $\Gamma$,%
\item a \emph{root expression} $\Gamma,Z \vdash e$,
\item a substitution $\xi$ assigning to each $z^\sigma \in Z$ a procedure $\Gamma,Z \vdash \xi(z) : \sigma$.
\end{itemize}
In this situation, we define the \emph{($k$-)plugging} $\Pi_{\Gamma,Z}(e,\xi)$
(often abbreviated to $\Pi(e,\xi)$)
to be the meta-term obtained from $e$ by repeatedly expanding variables $z \in Z$ to $\xi(z)$.
To formalize this, let $T^\circ$ denote the meta-term obtained from $T$ by replacing
each ground type subterm $z \vec{Q}$ (where $z \in Z$) by $\bot$.
We may now define, up to $\alpha$-equivalence,
\begin{eqnarray*}
\Pi^0(e,\xi) & = & e \;, \\
\Pi^{m+1}(e,\xi) & = & \Pi^m(e,\xi)[z \mapsto \xi(z) \mbox{~for all $z \in Z$}] \;, \\
\Pi(e,\xi) & = & \bigsqcup_m \Pi^m(e,\xi)^\circ \;,
\end{eqnarray*}
where $\bigsqcup$ denotes supremum with respect to the syntactic order on meta-terms.
\end{definition}
It is easy to see that $\Pi_{\Gamma,Z}(e,\xi)$ is well-typed in environment $\Gamma$.
Note that some renaming of bound variables will typically be necessary
in order to realize $\Pi_{\Gamma,Z}(e,\xi)$ as a concrete term conforming to the no-variable-hiding condition;
we will not need to fix on any one particular way of doing this.
The operation on NSPs corresponding to clause~6 of Proposition~\ref{long-beta-eta-prop}(ii)
may now be described as follows:
\begin{proposition} \label{NSP-Y-prop}
Suppose that $\sigma = \sigma_0,\ldots,\sigma_{r-1} \rightarrow \nat$ is of level $\leq k$
and that $\Gamma \vdash Y_\sigma M N_0 \ldots N_{r-1}$ in ${\mathrm{PCF}}^\Omega_k$,
where $\sem{M}_\Gamma = \lambda z^\sigma.p = \lambda z^{\sigma} x_0^{\sigma_0} \cdots x_{r-1}^{\sigma_{r-1}}.\,e$
and $\sem{N_i}_\Gamma = q_i$ for each $i$.
Then
\[ \sem{Y_\sigma M N_0 \ldots N_{r-1}}_\Gamma ~=~ \lambda. \dang{\Pi_{\Gamma,Z}(e,\xi)}
\]
where $Z = z,x_0,\ldots,x_{r-1}$, $\xi(z) = p$, and
$\xi(x_i) = q_i$ for each $i$.
\end{proposition}
\vspace*{1.0ex}
\begin{proof}
Note that in this instance of plugging, the repeated substitutions are needed only for the sake
of the term $\xi(z)$ which may contain $z$ free---only a single substitution step is needed for the
plugging variables $x_i$, since the $q_i$ contain no free variables from $Z$.
We may thus rewrite $\Pi_{\Gamma,Z}(e,\xi)$ as
\[ \Pi_{\Gamma,\vec{x},Z'}(e,\xi')\;[\vec{x}\mapsto{q}] \,, \]
where $Z'=\{z\}$ and $\xi'(z)=p$.
The proposition will therefore follow easily (with the help of Theorem~\ref{eval-thm}) once we know that
\[ \sem{Y_\sigma M}_\Gamma ~=~
\lambda \vec{x}.\, \dang{\Pi_{\Gamma,\vec{x},Z'}(e,\xi')} \;. \]
To see this, write $Y_\sigma = \lambda g.F_\sigma[g]$ where
$F_\sigma[g] = \lambda \vec{x}.\,\caseof{g\,(F_\sigma[g])\,\vec{x}^{\,\eta}}{i \Rightarrow i}$
as at the start of Section~\ref{subsec-interp}.
Then clearly
\[ \sem{Y_\sigma M}_\Gamma ~=~ (\lambda g. F_\sigma[g]) \cdot (\lambda z.p) ~=~
\dang{F_\sigma[\lambda z.p]} \;. \]
Here the meta-term $F_\sigma[\lambda z.p]$ is specified corecursively (up to $\alpha$-equivalence) by
\[ F_\sigma[\lambda z.p] ~=~ \lambda \vec{x}.\,\caseof{(\lambda z. p)\,(F_\sigma[\lambda z.p])\,\vec{x}^{\,\eta}}{i \Rightarrow i} \]
That is, $F_\sigma[\lambda z.p]$ coincides with the meta-term $G = \bigsqcup_m G^m$, where
\[ G^0 ~=~ \bot_\sigma \;, ~~~~~~
G^{m+1} ~=~ \lambda \vec{x}.\,\caseof{(\lambda z.p)\,G^m\,\vec{x}^{\,\eta}}{i \Rightarrow i} \;. \]
We may now compare this with the meta-term $H = \bigsqcup_m H^m$, where
\[ H^0 ~=~ \bot_\sigma \;, ~~~~~~
H^{m+1} ~=~ \lambda \vec{x}.\, e [z \mapsto H^m] \;. \]
Noting that $(\lambda z.p)\,G^m\,\vec{x}^{\,\eta} \rightsquigarrow e[z \mapsto G^m,\,\vec{x} \mapsto \vec{x}^{\,\eta}]$,
we have by Lemmas~\ref{eta-properties} and \ref{rightward-leaf-prop} that
\[ \dang{G^{m+1}} ~=~ \dang{\lambda \vec{x}.\, e[z \mapsto G^m]} \]
whence by Theorem~\ref{eval-thm} and an easy induction we have $\dang{G^m} \,=\, \dang{H^m}$
for all $m$. Hence $\dang{\!G\!} \,=\, \dang{\!H\!}$.
Moreover, it is immediate from the definition that $H$ coincides with the meta-term
$\lambda \vec{x}.\, \Pi_{\Gamma,\vec{x},Z'}(e,\xi')$ mentioned earlier.
We thus have
\[ \sem{Y_\sigma M}_\Gamma ~=~ \dang{F_\sigma[\lambda z.p]} ~=~ \dang{\!G\!} ~=~ \dang{\!H\!}
~=~ \lambda \vec{x}.\, \dang{\Pi_{\Gamma,\vec{x},Z'}(e,\xi')} \]
and the proof is complete.
(We have glossed over some fine details of variable renaming here, but these are easily attended to.)
$\Box$
\end{proof}
\vspace*{1.0ex}
Combining Propositions~\ref{NSP-nf-prop}, \ref{NSP-arith-prop} and \ref{NSP-Y-prop}
with Proposition~\ref{long-beta-eta-prop}, the results of this section may be summarized as follows.
\begin{theorem} \label{denotable-inductive-thm}
The class of ${\mathrm{PCF}}^\Omega_k$-denotable procedures-in-environment $\Gamma \vdash p$
is the class generated inductively by the following rules:
\begin{enumerate}
\item If $\Gamma \vdash q_i$ is denotable for each $i<r$ and $x \in \Gamma$, then
\[ \Gamma \vdash \lambda.\,\caseof{x q_0 \ldots q_{r-1}}{j \Rightarrow j} \] is denotable.
\item If $\Gamma,x \vdash p$ is denotable, then $\Gamma \vdash \lambda x.p$ is denotable.
\item Each $\Gamma \vdash \lambda.n$ is denotable.
\item If $\Gamma \vdash p$ is denotable and $f : \mathbb{N} \rightharpoonup \mathbb{N}$, then
$\Gamma \vdash p[i \mapsto f(i)]$ is denotable. (The constructions for ${\mathit{suc}}$ and ${\mathit{pre}}$
are special cases of this).
\item If $\Gamma \vdash p$, $\Gamma \vdash \lambda.d$ and $\Gamma \vdash \lambda.e$ are denotable,
then $\Gamma \vdash p[0 \mapsto d, i+1 \mapsto e]$ is denotable.
\item If $\Gamma \vdash \lambda z^\sigma x_0^{\sigma_0} \cdots x_{r-1}^{\sigma_{r-1}}.e$ is denotable
where $\sigma = \sigma_0,\ldots,\sigma_{r-1}\rightarrow\nat$ is of level $\leq k$,
and $\Gamma \vdash q_i:\sigma_i$ is denotable for each $i<r$, then
\[ \Gamma \vdash \lambda. \dang{\Pi_{\Gamma,Z}(e,\xi)} \]
is denotable, where $Z = z,\vec{x}$ is disjoint from $\Gamma$,
$\xi(z) = \lambda \vec{x}.e$, and $\xi(x_i)=q_i$ for each $i$.
\end{enumerate}
\end{theorem}
To conclude this section,
we introduce a useful constraint on NSPs which, although not satisfied by all ${\mathrm{PCF}}^\Omega_k$-denotable
procedures, will hold for all those that we will need to consider in the course of our main proofs.
As we shall see, this constraint will interact well with the inductive rules just presented.
Referring back to the examples in Section~\ref{sec-intro}, we see that the recursive definitions of both $Y_{k+1}$ and $\Phi_{k+1}$ involved a variable $g$ of type level $k+2$.
It is therefore natural that our analysis will involve the consideration of terms in which such a variable $g$
appears free. However, it will turn out that apart from this one designated variable, our terms need never
involve any other variables of level $>k$, and this has a pleasant simplifying effect on our arguments.
This motivates the following definition:
\begin{definition} \label{regular-def}
Suppose $g$ is a variable of type level $k+2$.
(i) An environment $\Gamma$ is \emph{($g$-)regular}
if $\Gamma$ contains $g$ but all other variables in $\Gamma$ are of type level $\leq k$.
(ii) A meta-term $T$ is regular if all free and bound variables within $T$ are of level $\leq k$,
except possibly for free occurrences of $g$.
(iii) A meta-term-in-environment $\Gamma \vdash T$ is regular if both $\Gamma, T$ are regular.
\end{definition}
There is a useful alternative characterization of regularity in the case of normal forms:
\begin{proposition} \label{regular-term-prop}
A term-in-environment $\Gamma \vdash t$ is regular iff $\Gamma$ is regular and $t$ is
not a procedure of type level $\geq k+2$.
\end{proposition}
\begin{proof}
The left-to-right implication is trivial, since a procedure of level $\geq k+2$ would have the form
$\lambda \vec{x}.\cdots$ where at least one of the $x_i$ was of level $\geq k+1$.
For the converse, suppose $\Gamma$ is regular and $t$ is not a procedure of level $\geq k+2$.
Then $t$ contains no free variables of level $\geq k$ other than $g$, so we just need to show
that all variables bound by a $\lambda$-abstraction within $t$ are of level $\leq k$.
Suppose not, and suppose that $\lambda \vec{x}.e$ is some outermost subterm of $t$
with ${\mathrm{lv}}(\vec{x}) > k$. Then $\lambda \vec{x}.e$ cannot be the whole of $t$,
since $t$ would then be a procedure of level $>k+1$.
Since $t$ is a normal form, the subterm $\lambda \vec{x}.e$ (of level $>k+1$)
must therefore occur as an argument to some variable $w$
of level $> k+2$. But this is impossible, since $\Gamma$ contains no
such variables, nor can such a $w$ be bound within $t$, since the relevant subterm
$\lambda \vec{w}.d$ would then properly contain $\lambda \vec{x}.e$, contradicting
the choice of the latter.
$\Box$
\end{proof}
\vspace*{1.0ex}
Let us now consider how the inductive clauses of Theorem~\ref{denotable-inductive-thm} may be used to
generate \emph{regular} procedures-in-environment $\Gamma \vdash t$.
The following gives a useful property of derivations involving these clauses:
\begin{proposition} \label{regular-generation-prop}
If $\Gamma \vdash t$ is regular and ${\mathrm{PCF}}^\Omega_k$-denotable,
then any inductive generation of the denotability of $\Gamma \vdash t$ via the clauses of
Theorem~\ref{denotable-inductive-thm} will consist entirely of regular procedures-in-environment.
\end{proposition}
\begin{proof}
It suffices to observe that for each of the six inductive clauses (regarded as rules), if the conclusion
is a regular procedure-in-environment then so are each of the premises.
For clause 1, this is clearly the case because the $q_i$ are subterms of the procedure in the conclusion.
For clause 2, we note that if $\lambda x.p$ is regular then so is $p$, and moreover $x$ has
level $\leq k$ so that $\Gamma,x$ is regular.
Clauses 3 and 4 are trivially handled.
For clause 5, we not that if $\Gamma \vdash p[0 \vdash d, i+1 \mapsto e]$ is regular
then $\Gamma \vdash p$, $\Gamma \vdash \lambda.d$ and $\Gamma \vdash \lambda.e$
are immediately regular by Proposition~\ref{regular-term-prop}
(regardless of whether any leaves $0$ or $i+1$ appear in $p$).
Likewise, for clause 6, we note that under the given hypotheses, both
$\lambda z \vec{x}.e$ and each $q_i$ are of level $\leq k+1$; hence if $\Gamma$ is regular
then immediately $\Gamma \vdash \lambda z \vec{x}.e$ and $\Gamma \vdash q_i$ are
regular by Proposition~\ref{regular-term-prop}.
$\Box$
\end{proof}
\vspace*{1.0ex}
In particular, let us consider again the construction of the procedure $Y_{k+1}$ as
$\lambda g.F_{k+1}[g]$, where
\[ F_{k+1}[g] ~=~ \lambda x^k.\;\caseof{g\,(F_{k+1}[g])\,x^\eta}{i \Rightarrow i} \;. \]
It is clear by inspection that $g \vdash F_{k+1}[g]$ is regular;
hence, if it were ${\mathrm{PCF}}^\Omega_k$-denotable, then Proposition~\ref{regular-generation-prop} would apply.
We shall show, however, that a purely regular derivation via the clauses of
Theorem~\ref{denotable-inductive-thm} cannot generate `spinal' terms such as $F_{k+1}[g]$;
hence $g \vdash F_{k+1}[g]$ is not ${\mathrm{PCF}}^\Omega_k$-denotable.
This will immediately imply that $Y_{k+1}$ itself is not ${\mathrm{PCF}}^\Omega_k$-denotable
(Theorem~\ref{SP0-main-thm}),
since the only means of generating non-nullary $\lambda$-abstractions is via clause 2 of Theorem~\ref{denotable-inductive-thm}.
\section{${\mathrm{PCF}}^\Omega_k$-denotable procedures are non-spinal} \label{sec-Y-k+1}
In this section, we will introduce the crucial notion of a \emph{($k\!+\!1$-)spinal term}, and
will show that the clauses of Theorem~\ref{denotable-inductive-thm} (in the regular case)
are unable to generate spinal terms from non-spinal ones.
Since the procedure $F_{k+1}[g]$ will be easily seen to be spinal, this will
establish Theorem~\ref{SP0-main-thm}.
More specifically, we will actually introduce the notion of a \emph{$g$-spinal term},
where $g$ is a free variable which we treat as fixed throughout our discussion.
We shall do this first for the case
\[ g ~:~ (k+1) \rightarrow (k+1) \]
as appropriate for the analysis of $F_{k+1}[g]$ and hence of $Y_{k+1}$.
Later we will also consider a minor variation for $g$ of type $\nat \rightarrow (k+1) \rightarrow (k+1)$,
as appropriate to the definition of $\Phi_{k+1}$ given in Section~\ref{sec-intro}.
In both cases, we shall be able to dispense with the term `$k\!+\!1$-spinal', since the type level $k+1$
may be read off from the type of $g$.
Some initial intuition for the concept of spinality was given in Section~\ref{sec-intro}.
We now attempt to provide some further motivation by examining a little more closely
the crucial difference between $Y_{k+1}$ and $Y_k$ (say) that we are trying to capture.
The most obvious difference between these procedures is that
$Y_{k+1}$ involves an infinite sequence of nested calls to a variable $g: (k+1) \rightarrow (k+1)$,
whereas $Y_k$ does not.
One's first thought might therefore be to try and show that no procedure
involving an infinite nesting of this kind can be constructed using the means at our disposal
corresponding to ${\mathrm{PCF}}^\Omega_k$ terms.
As it stands, however, this is not the case.
Suppose, for example, that ${\mathit{up}}_k: {k} \rightarrow {k+1}$ and
${\mathit{down}}_k: {k+1} \rightarrow {k}$ are ${\mathrm{PCF}}_0$ terms defining a
standard retraction ${k} \lhd {k+1}$.
More specifically, let us inductively define
\[ \begin{array}{rclrcl}
{\mathit{up}}_0 & = & \lambda x^0.\lambda z^0.x \;, &
{\mathit{down}}_0 & = & \lambda y^1.\,y\,\num{0} \;, \\
{\mathit{up}}_{k+1} & = & \lambda x^{k+1}.\lambda z^{k+1}.\,x({\mathit{down}}_k\;z) \;, &
{\mathit{down}}_{k+1} & = & \lambda y^{k+2}.\lambda w^{k}.\,y({\mathit{up}}_k\;w) \;.
\end{array} \]
Now consider the ${\mathrm{PCF}}_k$ program
\[ Z_{k+1} ~=~
\lambda g: (k+1) \rightarrow (k+1).\; {\mathit{up}}_k\,(Y_k\,({\mathit{down}}_k \circ g \circ {\mathit{up}}_k)) \;. \]
This is essentially just a representation of $Y_k$ modulo our encoding of type $k$ in type $k+1$.
A simple calculation shows that the NSPs for $Y_{k+1}$ and $Z_{k+1}$ are
superficially very similar in form, both involving an infinite sequence
of nested calls to $g: (k+1) \rightarrow (k+1)$.
(These NSPs are shown schematically in Figure~1 for the case $k=2$.)
We will therefore need to identify some more subtle property of NSPs that differentiates
between $Y_{k+1}$ and $Z_{k+1}$.
\begin{figure} \label{Y_Z_NSPs}
\begin{center}
\includegraphics[scale=.60]{Y_Z_NSPs.pdf}
\end{center}
\caption{The NSPs for $Y_3$ and $Z_3$. Here $\lambda.z$ abbreviates
$\lambda.\,\caseof{z}{i \Rightarrow i}$.}
\vspace*{1.0ex}
\end{figure}
The intuitive idea will be that in the NSP for $Z_{k+1}$, the full potency of
$g$ as a variable of type ${k+1} \rightarrow {k+1}$ is not exploited,
since both the input and output of $g$ are `funnelled' through the simpler type ${k}$.
Such funnelling will inevitably entail some loss of information,
as Theorem~\ref{no-retraction-thm} tells us that the type ${k}$ cannot
fully represent the structure of the type ${k+1}$.
A useful mental picture here is that of a $(k+1)$-dimensional space being `flattened' down to a $k$-dimensional one.
Broadly speaking, then, we shall want to define a $g$-spinal term to be one containing an infinite sequence
of nested calls to $g$ but with no essential `flattening' of the arguments.
It will then be the case that $Y_{k+1}$ is $g$-spinal, but $Z_{k+1}$ is not.
We now approach the formal definition of a $g$-spinal term,
generalizing the structure exhibited by the terms $F_{k+1}[g]$.
To get our bearings, let us examine the form of these terms one more time.
Note that $F_{k+1}[g]$ has the form $\lambda x^k. H[g,x]$, where
\[ H[g,x] ~=~ \caseof{g(\lambda x'.H[g,x']) x^\eta}{\cdots} \;. \]
Spinal expressions of this kind, in which the topmost $g$ of the spine appears at the very head of
the expression, will be referred to as \emph{head-spinal}.
In fact, we shall say that $H[g,x]$ is head-spinal with respect to the variable $x$, since as noted above,
it is significant here that $x$ is passed to $g$ with no `flattening' (in the form of the procedure $x^\eta$).
As a first attempt, then, one might hazard that we should define a concept of head-spinality relative to
a type $k$ variable coinductively as follows: an expression $e$ is $g$-head-spinal w.r.t.\ $x$ if it is of the
form
\[ \caseof{g(\lambda x'.e')x^\eta}{\cdots} \]
where $e'$ is itself $g$-head-spinal w.r.t.\ $x$.
In fact, in order for the set of non-spinal terms to have appropriate closure properties,
we shall need to relax this definition in two ways.
Firstly, we allow $\lambda x'.e'$ to be replaced by $\lambda x'.E[e']$ where $E[-]$ is any expression context:
that is, we allow the head-spinal subterm $e'$ to appear at positions other than the head of this procedure.
Secondly, we allow $x^\eta$ to be replaced by a procedure term $o$ that can be \emph{specialized} to
(something close to) $x^\eta$: the intuition is that any such $o$ will embody the whole content of $x$
with no flattening.
This leads us, at last, to the following definition. Note that this makes reference to the technical notion of
an \emph{$x,V$-closed substitution}, the explanation of which we shall defer to Definition~\ref{xV-closed-def} below. This entails that the notion of head-spinality needs to be defined relative to a certain set $V$ of
variables as well as a type $k$ variable $x$.
We shall adopt the convention that any environment denoted by $\Gamma$ will be $g$-regular
(and hence will contain $g$); recall that Roman letters such as $V,X,Z$
always denote lists of variables of level $\leq k$ (which may also contribute to the
environments we consider).
\begin{definition}[Spinal terms] \label{spinal-def}
Suppose $g$ has type $(k+1) \rightarrow (k+1)$.
Suppose $\Gamma \vdash e$ is $g$-regular,
and that $x^k \in \Gamma$ and $V \subseteq \Gamma$.
(i) In this situation, we coinductively declare $e$ to be \emph{$g$-head-spinal with respect to $x,V$}
iff $e$ has the form
\[ \caseof{g(\lambda {x'}. E[e']) o}{\cdots} \]
where $E[-]$ is an expression context, and
\begin{enumerate}
\item for some $x,\!V$-closed substitution $^\circ$ covering the free variables of $o$ other than $x$,
we have $o^\circ \succeq x^\eta$,
\item $e'$ is $g$-head-spinal with respect to $x',V'$, where $V'$ is the local variable environment for $E[-]$.
(Clearly $e'$ will automatically be $g$-regular in some $\Gamma'$ that contains both $x'$ and $V'$.)
\end{enumerate}
In other words, we take `$e$ is $g$-head-spinal w.r.t.\ $x,V$' to be the largest relation
that satisfies the above statement.
(ii) In the above setting, we may also refer to the application $g(\lambda {x'}. E[e']) o$ itself
as $g$-head-spinal w.r.t.\ $x,V$.
(iii) We say a term $t$ is \emph{$g$-spinal} if it contains a subexpression that is $g$-head-spinal
w.r.t.\ some $x,V$.
\end{definition}
Whilst this definition makes use of local variable environments which in principle pertain to concrete terms,
it is easily seen that the notion of $g$-spinal term is $\alpha$-invariant.
Since we are taking $g$ to be fixed throughout the discussion, we will usually omit mention of it
and speak simply of spinal and head-spinal terms, and of regular (meta-)terms and environments.
In condition~1, one might have expected to see
$o^\circ \approx x^\eta$, but it turns that the argument goes through most smoothly
with $\succeq$ in place of $\approx$.
In Appendix~A we will see that $o^\circ \succeq x^\eta$
is actually equivalent to $o^\circ \approx x^\eta$,
although this is somewhat non-trivial to show and is not needed for our main proof.
It remains to define the notion of an $x,V$-closed substitution.
Suppose that $^\circ = [\vec{w} \mapsto \vec{r}\,]$ is some substitution proposed for use in
condition 1 of Definition~\ref{spinal-def}(i).
Since we are wishing to compare $o^\circ$ with $x^\eta$, it is natural to require that the $\vec{r}$
contain no free variables other than $x$.
However, what we want to ensure here is intuitively that the whole unflattened content of $x$ is present
in $o$ itself rather than simply being introduced by the substitution.
This can be ensured if we allow $x$ as a free variable only in procedures $r_i$ of type level $<k$:
such procedures can only introduce `flattened' images of $x$, since the $x$ is here being funnelled through
a type of level $\leq k-1$.
For technical reasons, we furthermore need to restrict such occurrences of $x$ to those $r_i$
substituted for variables $w_i$ in a certain set $V$,
which in practice will consist of variables
bound between one spinal occurrence of $g$ and the next
(as can be seen from the specification of $V'$ above).
The necessity for the set $V$ is admittedly difficult to motivate at this point:
it is simply what the details of the proof seem to demand
(see the last page of the proof of Lemma~\ref{g-lemma-2}).
\begin{definition} \label{xV-closed-def}
If $x$ is a variable of type $k$ and $V$ a set of variables,
a substitution $^\circ = [\vec{w} \mapsto \vec{r}\,]$ is called
\emph{$x,\!V$-closed} if the $r_i$ contain no free variables, except that if $w_i \in V$
and ${\mathrm{lv}}(w_i)<k$ then $r_i$ may contain $x$ free.
\end{definition}
It is worth remarking that if we were only interested in
showing the non-definability of $Y_{k+1}$ as an element of ${\mathsf{SP}}^0$, one
could do without the notion of $x,\!V$-closedness altogether,
and more simply require in Definition~\ref{spinal-def} that $^\circ$ is closed
(and moreover that $\dang{\!o^\circ\!}\, = x^\eta$ on the nose).
The weaker definition we have given is designed with the proof of non-definability
in ${\mathsf{SF}}$ in mind: we will be able to show in Section~\ref{sec-extensional} that every (simple) procedure
representing the functional $\Phi_{k+1} \in {\mathsf{SF}}$ is spinal in this weaker sense.%
\footnote{It can be shown using Theorem~\ref{no-retraction-thm} that if $o^\circ \succeq x^\eta$
where $^\circ$ is $x,\!V$-closed, then at least one $x$ in $\dang{\!o^\circ\!}$ must originate from
$o$ rather than from $^\circ$.
We have not actually settled the question of whether there are procedures $o$ such that
$o^\circ \succeq x^\eta$ for some $x,\!V$-closed $^\circ$ but not for any closed $^\circ$;
fortunately this is not necessary for the purpose of our proof.}
We now digress briefly to explain the small modification of this machinery
that we will need in Section~\ref{sec-extensional}.
Since our purpose there will be to analyse the functional $\Phi_{k+1}$ which we defined
in Section~\ref{sec-intro}, we shall be working in a setup in which
the global variable $g$ has the slightly different type ${0}\rightarrow(k+1)\rightarrow(k+1)$.
In this setting, we may vary the above definition
by coinductively declaring $e$ to be $g$-head-spinal w.r.t.\ $x,V$ iff $e$ has the form
\[ \caseof{gb(\lambda x'. E[e']) o}{\cdots} \]
where $b$ is a procedure term of type ${0}$ and conditions~1 and 2 above are also satisfied.
Subject to this adjustment, all the results and proofs of the present section go through
in this modified setting, with the extra argument $b$ playing no active role.
For the remainder of this section, we shall work with
a global variable $g$ of the simpler type $(k+1) \rightarrow (k+1)$, on the understanding that
the extra arguments $b$ can be inserted where needed to make formal
sense of the material in the modified setting.
We do not expect that any confusion will arise from this.
Clearly $g \vdash F_{k+1}[g]$ is spinal.
The main result of this section will be that every ${\mathrm{PCF}}^\Omega_k$-denotable procedure
$\Gamma \vdash p$ is non-spinal (Theorem~\ref{no-gremlin-thm}).
We shall establish this by induction on the generation of denotable terms
as in Theorem~\ref{denotable-inductive-thm}, the only challenging case being the one for rule~6,
which involves plugging.
Here we require some technical machinery whose purpose is to show that
if the result of a plugging operation is spinal,
then a spinal structure must already have been present in one of the components of the plugging:
there is no way to `assemble' a spinal structure from material in non-spinal fragments.
The core of the proof will consist of some lemmas developing the machinery
necessary for tackling rule 6.
We start with some technical but essentially straightforward facts concerning evaluation and
the tracking of subterms and variable substitutions.
\begin{lemma} \label{g-lemma-1}
Suppose that
\[ \Gamma \;\vdash\; \dang{K[d]} \;=_\alpha\; K'[c] \]
where $K[-],K'[-]$ are concrete meta-term contexts with local environments $\vec{v},\vec{v}\,'$ respectively,
and $\Gamma,\vec{v} \vdash d = \caseof{gpq}{\cdots}$,
$\Gamma,\vec{v}\,' \vdash c = \caseof{gp'q'}{\cdots}$ are concrete expressions.
Suppose also that:
\begin{enumerate}
\item $\Gamma \vdash K[d]$ is regular;
\item in the evaluation above, the head $g$ of $c$ originates from that of $d$.
\end{enumerate}
Then:
(i) There is a substitution $^\dag = [\vec{v}\mapsto\vec{s}\,]$ of level $\leq k$
arising from the $\beta$-reductions in the above evaluation,
with $\Gamma,\vec{v}\,' \vdash \vec{s}$ regular, such that
$\Gamma,\vec{v}\,' \vdash gp'q' =_\alpha \,\dang{\!(gpq)^\dag\!}$,
whence $\dang{\!d^\dag\!}$ is of form $\caseof{gp'q'}{\cdots}$ up to $=_\alpha$.%
\footnote{Note that although both $\dang{\!d^\dag\!}$ and $c$ have the form $\caseof{gp'q'}{\cdots}$,
they will in general have different case-branches, for instance when $K[-]$ is of the form
$\caseof{-}{\cdots}$.}
(ii) If furthermore
$c$ is head-spinal w.r.t.\ some $x,V$,
then also $\dang{d^\dag}$ is head-spinal w.r.t.\ $x,V$.
(iii) If $K[-]$ contains no $\beta$-redexes $P \vec{Q}$ with $P$ of type level $k+1$,
then $^\dag$ is \emph{trivial for level $k$ variables}:
that is, there is an injection $\iota$ mapping each level $k$ variable $v_i \in \vec{v}$ to
a variable $\iota(v_i) \in \vec{v}\,'$ such that $s_i = \iota(v_i)^\eta$.
\end{lemma}
In reference to part~(iii), recall that substitutions $v \mapsto v^\eta$ have no effect on the meaning
of a term, as established by Lemma~\ref{eta-properties}.
Note that the environments $\vec{v},\vec{v}\,'$, and hence the injection $\iota$, will in general
depend on the concrete choice of $K[d]$ and $K'[c]$. However, for the purpose of proving the theorem,
it is clearly harmless to assume that $K'[c]$ is, on the nose, the concrete term obtained by evaluating $K[d]$.
In this case, we will see from the proof below that each $\iota(v_i)$ will be either $v_i$ itself or a renaming
of $v_i$ arising from the evaluation.
\vspace*{1.0ex}
\begin{proof}
(i) We first formulate a suitable property of terms that is preserved under all individual reduction steps.
Let $K[-],d,p,q$ and $\vec{v}$ be fixed as above, and suppose that
\[ K^0[\caseof{gP^0Q^0}{\cdots}] ~\rightsquigarrow~ K^1[\caseof{gP^1Q^1}{\cdots}] \]
via a single reduction step,
where the $g$ on the right originates from the one on the left,
and moreover $K^0,P^0,Q^0$ enjoy the following properties
(we write $\vec{v}\,^0$ for the local environment for $K^0[-]$):
\begin{enumerate}
\item $\Gamma \vdash K^0[\caseof{gP^0Q^0}{\cdots}]$ is regular.
\item There exists a substitution $^{\dag0} = [\vec{v}\mapsto\vec{s}\,^0]$
(with $\Gamma,\vec{v}\,^0 \vdash \vec{s}\,^0$ regular)
such that $\dang{gP^0Q^0} \,=_\alpha\, \dang{(gpq)^{\dag 0}}$.
\end{enumerate}
We claim that $K^1,P^1,Q^1$ enjoy these same properties w.r.t.\ the
local environment $\vec{v}\,^1$ for $K^1[-]$.
For property~1, clearly $K^1[\caseof{gP^1Q^1}{\cdots}]$ cannot contain
variables of level $>k$ other than $g$, because $K^0[\caseof{gP^0Q^0}{\cdots}]$ does not.
For property~2, we define the required substitution $^{\dag1} = [\vec{v}\mapsto\vec{s}\,^1]$
by cases according to the nature of the reduction step:
\begin{itemize}
\item If the subexpression $\caseof{gP^0Q^0}{\cdots}$ is unaffected by the
reduction (so that $P^0=P^1$ and $Q^0=Q^1$),
or if the reduction is internal to $P^0,Q^0$ or to the rightward portion $(\cdots)$,
or if the reduction has the form
\begin{eqnarray*}
\caseof{(\caseof{gP^0Q^0}{i \Rightarrow E^0_i})}{j \Rightarrow F_j} & \rightsquigarrow & \\
\caseof{gP^0Q^0}{i \Rightarrow \caseof{E^0_i}{j \Rightarrow F_j}}
\end{eqnarray*}
then the conclusion is immediate, noting that $\vec{v}\,^1 = \vec{v}\,^0$
and taking $^{\dag 1}=\,^{\dag 0}$.
\item If the reduction is for a $\beta$-redex $(\lambda \vec{x}.E)\vec{R}$
where the indicated subexpression $\caseof{gP^0Q^0}{\cdots}$ lies within some $R_i$,
we may again take $^{\dag 1}$ to be $^{\dag 0}$, with obvious adjustments to compensate for
any renaming of bound variables within $R_i$ or $\vec{s}\,^0$.
In this case $\vec{v}\,^1$ may contain more variables than $\vec{v}\,^0$,
but we will still have that $\Gamma,\vec{v}\,^1 \vdash \vec{s}\,^1$
once these renamings have been effected.
\item If the reduction is for a $\beta$-redex $(\lambda \vec{x}.E)\vec{R}$
where $\caseof{gP^0Q^0}{\cdots}$ lies within $E$,
then $P^1 = P^0[\vec{x} \mapsto \vec{R}\,']$ and $Q^1 = Q^0[\vec{x} \mapsto \vec{R}\,']$
for some $\vec{R}\,' =_\alpha \vec{R}$.
In this case, the local environment $\vec{v}\,^1$ for $K^1[-]$ will be $\vec{v}\,^0 - \vec{x}$
(perhaps modulo renamings of the $v^0_i$),
so that the conclusion follows if we take $^{\dag 1} = [\vec{v}\mapsto\vec{s}\,^1]$
where $s_i^1 = \dang{s_i^0 [\vec{x}\mapsto\vec{R}\,']}$ for each $i$
(modulo the same renamings).
Note here that $\Gamma,\vec{v}\,^1 \vdash \vec{s}\,^1$ is regular since $\vec{R}$ is regular
by condition~1 of the hypothesis.
\end{itemize}
Now in the situation of the lemma we will have some
finite reduction sequence
\[ K[\caseof{gpq}{\cdots}] ~\rightsquigarrow^*~ K''[\caseof{gP'Q'}{\cdots}] \;, \]
where, intuitively, $K''[-]$ is fully evaluated down as far as the hole.
More formally, there is a finite normal-form context $t[-] \sqsubseteq K''[-]$
containing the hole in $K''[-]$ such that $t[-] \sqsubseteq K'[-]$;
from this we may also see that $\dang{\!P'\!}\,=p'$, $\dang{\!Q'\!}\,=q'$
and $\dang{\!K''[-]\!}\,=K'[-]$.
Moreover, we now see that $K,p,q$ themselves trivially satisfy the above invariants
if we take $^\dag = [\vec{v} \mapsto \vec{v}\,^\eta]$ (Lemma~\ref{eta-properties} is used here).
We therefore infer by iterating the argument above that $K'',P',Q'$ also satisfy these invariants with respect to
some $^\dag = [\vec{v} \mapsto \vec{s}\,]$ with $\Gamma,\vec{v}\,' \vdash \vec{s}$ regular.
(The environment $\Gamma,\vec{v}\,'$ is correct here, as $K'[-],K''[-]$ have the same
local environment.)
We now have $gp'q' =\, \dang{gP'Q'} \,=_\alpha\, \dang{(gpq)^\dag}$.
That the $\vec{v}$ are of level $\leq k$ is automatic, because $K[d]$ is regular.
It also follows immediately that $\dang{\!d^\dag\!}$ has the stated form.
(ii) If $c$ is head-spinal w.r.t.\ $x,V$, then we see from Definition~\ref{spinal-def} that $gp'q'$
and hence $\dang{d^\dag} \,= \caseof{gp'q'}{\cdots}$ are head-spinal w.r.t.\ $x,V$.
(iii) From the proof of (i), we see that in the reduction of $K[\caseof{gpq}{\cdots}]$
to $K''[\caseof{gP'Q'}{\cdots}]$, any $v_i \in \vec{v}$
can be tracked through the local environments for the intermediate contexts $K^0[-],K^1[-],\ldots$
until (if ever) it is a substitution variable for a $\beta$-reduction.
For those $v_i$ that never serve as such a variable, it is clear from the construction that
$v_i$ gives rise to some variable $\iota(v_i) \in \vec{v}\,'$ (either $v_i$ itself or a renaming thereof),
and that $s_i = v_i^\dag = \iota(v_i)^\eta$.
We wish to show that all $v_i \in \vec{v}$ of level $k$ are in this category.
Recalling that $\vec{v}$ is the local environment for $K[-]$,
any $v_i \in \vec{v}$ of level $k$ is bound by the leading $\lambda$ of some
meta-procedure $P$ within $K[-]$ of level $k+1$.
By hypothesis, this $P$ does not occur in operator position; nor can it occur
as an argument to another $\lambda$-abstraction within $K[-]$,
since this would require a bound variable of level $\geq k+1$.
It must therefore occur as a level $k+1$ argument to $g$,
so that we have a subterm $g(\lambda v_i.E[-])\cdots$.
But this form of subterms is stable under reductions, since $g$ is a global variable;
it follows easily that this subterm has a residual $g(\lambda v_i'.E'[-])\cdots$ in each of the intermediate
reducts (where $v_i'$ is either $v_i$ or a renaming thereof),
and thus that $v_i$ and renamings thereof never serve as substitution variables for $\beta$-reductions.
$\Box$
\end{proof}
\vspace*{1.0ex}
Thus, in the setting of the above lemma,
if $c$ is head-spinal then $d$ can be specialized and evaluated to yield a head-spinal term via the substitution
$[\vec{v} \mapsto \vec{s}\,]$.
However, we wish to show more, namely that in this setting, $d$ itself is already a spinal term,
so that the $\vec{s}$ make no essential contribution to the spinal structure.
(This will give what we need in order to show that $k$-pluggings cannot manufacture spinal terms
out of non-spinal ones.)
This is shown by the next lemma, whose proof forms the
most complex and demanding part of the entire argument.
The main challenge will be to show that all the head-spinal occurrences of $g$ in
$\dang{d[\vec{v} \mapsto \vec{s}\,]}$ originate from $d$ rather than from $\vec{s}$.
The reader is advised that great care is needed as regards which variables can appear free where,
and for this reason we shall make a habit of explicitly recording the variable environment
for practically every term or meta-term that we mention.
\begin{lemma} \label{g-lemma-2}
Suppose we have regular terms
\[ \Gamma,\vec{v} \;\vdash~ d ~=~ \caseof{gpq}{\cdots} \;, ~~~~~~
\Gamma,\vec{v}\,' \vdash \vec{s} \;, ~~~~~~
{\mathrm{lv}}(\vec{v}), {\mathrm{lv}}(\vec{v}\,') \leq k \;, \]
where $\Gamma,\vec{v}\,' \vdash\, \dang{d[\vec{v}\mapsto\vec{s}\,]}$
is head-spinal with respect to some $x,V$.
Then $d$ itself is spinal.
\end{lemma}
\begin{proof}
We begin with some informal intuition.
The term $\dang{d[\vec{v}\mapsto\vec{s}\,]}$, being head-spinal, will be of the form
\[ \Gamma,\vec{v}\,' \;\vdash~ t ~=~
\caseof{g\,(\lambda x'.\,E[\caseof{gF'o'}{\cdots}])\,o}{\cdots} \;, \]
where ${o'}^\circ \succeq {x'}^\eta$ for some $^\circ$ (and likewise for $o$ and $x$).
Here the head $g$ of $t$ clearly originates from that of $d$;
likewise, the $\lambda x'$ originates from the leading $\lambda$ of $p$ within $d$,
rather than from $\vec{s}$.
Suppose, however, that the second displayed spinal occurrence of $g$ in $t$
originated from some $s_i$ rather than from $d$.
In order to form the application of this $g$ to $o'$,
the whole content of ${x'}^\eta$ would in effect need to be passed in to $s_i$
when $d$ and $\vec{s}$ are combined. But this is impossible, since the arguments to
$s_i$ are of level $<k$, so by Theorem~\ref{no-retraction-thm}
we cannot funnel the whole of ${x'}^\eta$ through them:
that is, the interface between $d$ and $\vec{s}$ is too narrow
for the necessary interaction to occur.
(The situation is made slightly more complex by the fact that
some components of $^\circ$ might also involve $x'$, but the same idea applies.)
It follows that the second spinal $g$ in $t$ originates from $d$ after all.
By iterating this argument, we can deduce that all the spinal occurrences of $g$,
and indeed the entire spinal structure, comes from $d$.
We now proceed to the formal proof.
By renaming variables if necessary,
we may assume for clarity that the same variable is never bound in two places within the entire
list of terms $d,\vec{s}$,
and that all bound variables within $d,\vec{s}$ are distinct from those of $\vec{v}$ and $\vec{v}\,'$.
Let $^\dag = [\vec{v}\mapsto\vec{s}\,]$,
and let us write the subterm $p$ appearing within $d$ as $\Gamma,\vec{v} \vdash \lambda {x'}^k.e$,
where $\Gamma,\vec{v},x' \vdash e$ is regular. Then
\[ \dang{d^\dag} ~=~ \caseof{g\,(\lambda x'.\!\dang{e^\dag})\dang{q^\dag}}{\cdots} \;, \]
and since $\dang{\!d^\dag\!}$ is head-spinal by hypothesis,
$\dang{\!{e}^\dag\!}$ will be some term
$\Gamma,x',\vec{v}\,' \vdash E[c]$, where
$\Gamma,x',\vec{v}\,',\vec{y}\,' \vdash c~=~ \caseof{gF'o'}{\cdots}$ is itself head-spinal
with respect to $x'$ and $\vec{y}\,'$.
(Here $\vec{y}\,'$ denotes the local environment for $E[-]$, so that
$\Gamma,x',\vec{v}\,',\vec{y}\,'$ contains no repetitions.)
We will first show that the head $g$ of $c$ comes from $e$ rather than from $^\dag$;
we will later show that the same argument can be repeated for
lower spinal occurrences of $g$.
\vspace*{1.0ex}
\emph{Claim 1: In the evaluation $\dang{\!e^\dag\!} \,= E[c]$, the head $g$ of $c$
originates from $e$.}
\vspace*{1.0ex}
\emph{Proof of Claim 1:} Suppose for contradiction that the head $g$ of $c$ originates from some
substituted occurrence of an $s_i$ within ${e}^\dag$,
say as indicated by ${e}^\dag = D[s_i]$ and $s_i = L[d']$,
where $\Gamma,x',\vec{v}\,' \vdash D[-]$, $\Gamma,\vec{v}\,' \vdash s_i$,
and $\Gamma,\vec{v}\,',\vec{z} \vdash d' = \caseof{gp'q'}{\cdots}$.
(Here $\vec{z}$ is the local variable environment for $L[-]$;
note that $\vec{z}$ is disjoint from $\Gamma,x',\vec{v}\,'$, but may well overlap with $\vec{y}\,'$.)
Then
\[ \Gamma,x',\vec{v}\,' \;\vdash~
\dang{\!e^\dag\!} ~=~ \dang{D[L[d']]} ~=~ E[c] \;, \]
where the head $g$ in $d'$ is the origin of the head $g$ in $c$.
We will use this to show that a head-spinal term may be obtained from $d'$
via a substitution of level $< k$;
this will provide the bottleneck through which ${x'}^\eta$ is unable to pass.
We first note that the above situation satisfies the conditions of Lemma~\ref{g-lemma-1},
where we take the $\Gamma,K,d,K',c$ of the lemma
to be respectively $(\Gamma,x',\vec{v}\,'),$ $D[L[-]],d',E,c$.
Condition~1 of the lemma holds because $\Gamma,\vec{v},x' \vdash e$ and
$\Gamma,\vec{v}\,' \vdash \vec{s}$ are clearly regular,
whence so is $\Gamma,x',\vec{v} \vdash e^\dag = D[L[d']]$;
condition~2 is immediate in the present setup.
We conclude that there is a substitution $[\vec{y}\mapsto\vec{t}\,]$
(called $[\vec{v}\mapsto\vec{s}\,]$ in the statement of Lemma~\ref{g-lemma-1})
with $\vec{y}$ the local environment for $D[L[-]]$
and $\Gamma,x',\vec{v}\,',\vec{y}\,' \vdash \vec{t}$
(recalling that $\vec{y}\,'$ are the local variables for $E[-]$),
such that $\dang{d'[\vec{y}\mapsto\vec{t}\,]}$ is head-spinal
and indeed of the form $\caseof{gF'o'}{\cdots}$ with $F',o'$ as above.
Furthermore, the only $\beta$-redexes in $e^\dag$ are those arising from the
substitution $^\dag$, with some $s_j$ of level $k$ as operator.
There are therefore no $\beta$-redexes in $e^\dag$ with a substitution variable of level $k$,
so by Lemma~\ref{g-lemma-1}(iii), the substitution $[\vec{y} \mapsto \vec{t}\,]$
is trivial for variables of level $k$.
Note also that $\vec{y}$ (the environment for $D[L[-]]$) subsumes $\vec{z}$
(the environment for $L[-]$); it is disjoint from $\Gamma,x,\vec{v}\,'$ but may well overlap with $\vec{y}\,'$.
Let us now split the substitution $[\vec{y} \mapsto \vec{t}\,]$ as
$[\vec{y}\,^+ \mapsto \vec{t}\,^+, \, \vec{y}\,^- \mapsto \vec{t}\,^-]$,
where $\vec{y}\,^+$ consists of the variables in $\vec{y}$ of level $k$,
and $\vec{y}\,^-$ consists of those of level $<k$.
As we have noted, the substitution for $\vec{y}\,^+$ is trivial:
that is, there is a mapping associating with each $y_j \in \vec{y}\,^+$ a variable $\iota(y_j) \in \vec{y}\,'$
such that $t_j = \iota(y_j)^\eta$.
Taking stock, we have that
\begin{eqnarray*}
\Gamma,x',\vec{v}\,',\vec{y}\,' & \vdash &
\dang{d'[\vec{y} \mapsto \vec{t}\,]} ~=~ \caseof{gF'o'}{\cdots} \;, \\
\Gamma,\vec{v}\,',\vec{z} & \vdash & d' = \caseof{gp'q'}{\cdots} \;, \\
\Gamma,x',\vec{v}\,',\vec{y}\,' & \vdash & \vec{t} \;,
\end{eqnarray*}
where $[\vec{y}\mapsto\vec{t}\,]$ is trivial for level $k$ variables,
and $gF'o'$ is head-spinal w.r.t.\ $x,\vec{y}\,'$.
From this we may read off that
\[ \Gamma,x',\vec{v}\,',\vec{y}\,' \;\vdash\; \dang{q'[\vec{y} \mapsto \vec{t}\,]} ~=~ o' \;. \]
Since $\vec{y}$ subsumes $\vec{z}$,
we may henceforth regard $q'$ as a term in environment $\Gamma,\vec{v}\,',\vec{y}$.
(This is compatible with the no-variable-hiding condition:
our conventions ensure that the variables of $\vec{y}-\vec{z}$ come from $d$ rather than $s_i$
and so do not appear bound in $q'$.)
We may harmlessly write $q'[\vec{y} \mapsto \vec{t}\,]$ as above,
even though there are variables of $\vec{y}$ that cannot appear in $q'$.
Since $x'$ does not occur free in $q'$, each free occurrence of $x'$ in $o'$ above must originate from some $t_j \in \vec{t}$,
which must furthermore have some type $\rho_j$ of level $<k$, since if $t_j$ had level $k$ then
we would have $t_j = \iota(y_j)^\eta$ which does not contain $x'$ free.
In fact, we may decompose the substitution $[\vec{y} \mapsto \vec{t}\,]$ as
$[\vec{y}\,^+ \mapsto \iota(\vec{y}\,^+)^\eta]$ followed by $[\vec{y}\,^- \mapsto \vec{t}\,^-]$,
since none of variables of $\vec{y}\,^-$ appear free in the $\iota(y_j)^\eta$ for $y_j \in \vec{y}\,^+$.
Setting $q'^* =\, \dang{q[\vec{y}\,^+ \mapsto \iota(\vec{y}\,^+)^\eta]}$ (so that $q'^*$ is just $q'$
with the variables in $\vec{y}\,^+$ rewritten via $\iota$),
we therefore have $\dang{q'^*[\vec{y}\,^- \mapsto \vec{t}\,^-]}\, = o'$.
Thus:
\begin{eqnarray*}
\Gamma,\vec{v}\,',\iota(\vec{y}\,^+),\vec{y}\,^- & \vdash & q'^* : \pure{k} \;, \\
\Gamma,x,\vec{v}\,',\vec{y}\,' & \vdash & t_j : \rho_j \mbox{~~~for $t_j \in \vec{t}\,^-$} \;, \\
\Gamma,x,\vec{v}\,',\vec{y}\,' & \vdash & \dang{q'^* [\vec{y}\,^- \mapsto \vec{t}\,^-]} ~=~ o' \;.
\end{eqnarray*}
Since $o'^\circ \succeq {x'}^\eta$ for a suitable $x',\vec{y}\,'$-closed substitution $^\circ$
(as part of the fact that $\dang{\!d^\dag\!}$ is head-spinal),
the above already comes close to exhibiting $\pure{k}$ as a pseudo-retract of a level $<k$ product type,
contradicting Theorem~\ref{no-retraction-thm}.
To complete the argument, we must take account of the effect of $^\circ$,
which we here write as $[\vec{w}\mapsto\vec{r}\,]$
(we may assume that $\vec{w}$ is exactly $\Gamma,\vec{v}\,',\vec{y}\,'$).
Reordering our variables, we may now write $x,\vec{w} \vdash \vec{t}\,^-$.
Next, let us split $^\circ$ into two independent parts: a substitution $[\vec{w}\,^+\mapsto\vec{r}\,^+]$
covering the variables in $\Gamma,\vec{v}\,',\vec{y}\,'$ of level $\geq k$,
and $[\vec{w}\,^-\mapsto\vec{r}\,^-]$ covering those of level $<k$.
Since $^\circ$ is $x',\vec{y}\,'$-closed, we have $\vdash \vec{r}\,^+$ and $x' \vdash \vec{r}\,^-$.
Now set ${q'}^\wr =\, \dang{{q'^*}[\vec{w}\,^+\mapsto\vec{r}\,^+]}$,
so that $\vec{u}\,^- \vdash {q'}^\wr$
where $\vec{u}\,^-$ consists of the variables of $\Gamma,\vec{v}\,',\vec{y}$ of level $<k$.
The idea is that $\vec{u}\,^- \vdash {q'}^\wr$ may now serve as one half of a suitable pseudo-retraction.
For the other half, let $[\vec{u}\,^- \mapsto \vec{a}\,^-]$ denote the effect of the substitution
$[\vec{y}\,^- \mapsto \vec{t}\,^-]$ followed by $^\circ =[\vec{w} \ \mapsto \vec{r}\,]$
(the order is important here as $\vec{y}\,^-$ and $\vec{w}$ may overlap).
Since $\vec{u}\,^- \subseteq \vec{y} \cup \vec{w}$ and $x,\vec{w} \vdash \vec{t}\,^-$
and $x \vdash \vec{r}$,
this substitution does indeed cover at least the variables of $\vec{u}\,^-$ and we have $x \vdash \vec{a}\,^-$.
We may now verify that $\vec{u}\,^- \vdash {q'}^\wr$ and $x \vdash \vec{a}\,^-$
constitute a pseudo-retraction as follows:
\begin{eqnarray*}
x' & \vdash & \dang{{q'}^\wr [\vec{u}\,^- \mapsto \vec{a}\,^-]} \\
& = & \dang{q'^* [\vec{w}\,^+ \mapsto \vec{r}\,^+] [\vec{y}\,^- \mapsto \vec{t}\,^-] [\vec{w} \ \mapsto \vec{r}\,]} \\
& = & \dang{(q'^* [\vec{y}\,^- \mapsto \vec{t}\,^-]) [\vec{w} \ \mapsto \vec{r}\,] } \\
& = & \dang{{o'}^\circ} ~\succeq~ {x'}^\eta \;.
\end{eqnarray*}
As regards the second equation here, the first substitution $[\vec{w}\,^+ \mapsto \vec{r}\,^+]$ may be safely
omitted as $\vec{w}\,^+$ and $\vec{y}\,^-$ are disjoint and the terms $\vec{r}\,^+$ do not contain
any of the $\vec{w}\,^+$ or $\vec{y}\,^-$ free.
We therefore have $\pure{k}$ as a pseudo-retract of a product of level $<k$ types.
This contradicts Theorem~\ref{no-retraction-thm}, so the proof of Claim 1 is complete.
\vspace*{1.0ex}
We may therefore suppose that in the evaluation $\dang{\!e^\dag\!} = E[c]$,
the originating occurrence of the head $g$ in $c$
is as indicated by $\Gamma,x',\vec{v} \vdash e = C[d']$,
where $\Gamma,x',\vec{v},\vec{v}\,'' \vdash d' =\caseof{gp'q'}{\cdots}$.
(Here $\vec{v}\,''$ is the local environment for $C[-]$.
The symbols $d',p',q'$ are available for recycling now that the proof of Claim~1 is complete.)
In order to continue our analysis to greater depth, note that we may write
\[ \Gamma,x',\vec{v}\,' \vdash~
\dang{\!e^\dag\!} ~=~ \dang{(\lambda \vec{v}.\,C[d'])\vec{s}} ~=~ E[c] \;, \]
where $c$ is head-spinal w.r.t.\ $x',\vec{y}\,'$, and the head $g$ of $d'$ is the origin of the head $g$ of $c$.
(Recall that $\vec{y}\,'$ is the local environment for $E[-]$ and that
$\Gamma,x',\vec{v} \vdash e = C[d']$ is regular, whence ${\mathrm{lv}}(\vec{v}\,'') \leq k$.)
We claim that once again we are in the situation of Lemma~\ref{g-lemma-1},
taking $\Gamma,K,d,K',c$ of the lemma to be respectively
$(\Gamma,x',\vec{v}\,')$, $(\lambda \vec{v}.\,C[-])\vec{s}$, $d', E, c$.
Condition~2 of the lemma is immediate in the present setup;
for condition~1, we again note that
$\Gamma,x',\vec{v}\,' \vdash C[d'] = e$ and $\Gamma,\vec{v}\,' \vdash \vec{s}$ are regular,
so by Proposition~\ref{regular-term-prop} contain no bound variables of level $>k$;
hence the same is true for $(\lambda \vec{v}.\,C[d'])\vec{s}$.
Applying Lemma~\ref{g-lemma-1}, we obtain a substitution
$^{\dag'} = [\vec{v}\,^+ \mapsto \vec{s}\,^+]$ of level $\leq k$
(with $\vec{v}\,^+ = \vec{v},\vec{v}\,''$),
where $\Gamma,x',\vec{v}\,',\vec{v}\,^+ \vdash d'$
and $\Gamma,x',\vec{v}\,',\vec{y}\,' \vdash \vec{s}\,^+$ are regular, such that
\[ \Gamma,x',\vec{v}\,',\vec{y}\,' \;\vdash\; \dang{d'[\vec{v}\,^+\!\mapsto\!\vec{s}\,^+]} \]
is head-spinal w.r.t.\ $x',\vec{y}\,'$, and indeed of the form $\caseof{gF'o'}{\cdots}$ with $F',o'$ as above.
(We may in fact write just $\Gamma,x',\vec{v}\,^+ \vdash d'$
since, by assumption, the variables of $\vec{v}\,'$ do not overlap with the free or bound variables of $d$
so do not appear in $d$.)
We may also read off that $\dang{(p')^{\dag'}} \,= F'$ and $\dang{(q')^{\dag'}} \,= o'$.
As regards the substitution $^{\dag'} = [\vec{v}\,^+ \mapsto \vec{s}\,^+]$, it is clear that this
extends $ [\vec{v} \mapsto \vec{s}\,]$ since the evaluation of $(\lambda \vec{v}.\,C[d'])\vec{s}$
starts by $\beta$-reducing this term.
Moreover, the argument of Lemma~\ref{g-lemma-1}(iii) shows that $^{\dag'}$ is trivial for
any level $k$ variables in $\vec{v}\,''$, as $C[-]$ is in normal form.
We are now back precisely where we started, in the sense that $d',\vec{v}\,^+,\vec{s}\,^+$
themselves satisfy the hypotheses of Lemma~\ref{g-lemma-2},
with $(\Gamma,x')$ now playing the role of $\Gamma$
and $(\vec{v}\,',\vec{y}\,')$ that of $\vec{v}\,'$.
Explicitly, we have regular terms
\[ \Gamma,x,\vec{v}\,^+ \vdash d' = \caseof{gp'q'}{\cdots} \;, ~~~~~~~~
\Gamma,x,\vec{v}\,',\vec{y}\,' \vdash \vec{s}\,^+ \]
(so that ${\mathrm{lv}}(\vec{v}\,^+), {\mathrm{lv}}(\vec{v}\,',\vec{y}\,') \leq k$)
where $\Gamma,x,\vec{v}\,',\vec{y}\,' \vdash \dang{d' [\vec{v}\,^+ \mapsto \vec{s}\,^+]}$
is head-spinal w.r.t.\ $x',\vec{y}\,'$.
We can therefore iterate the whole of the above argument to obtain an infinite descending chain of subterms
\[ \begin{array}{rrlrl}
\Gamma,\vec{v} \vdash &
d = & \caseof{gpq}{\cdots} \;, &
p = & \lambda x'.\,C[d'] \;, \\
\Gamma,\vec{v},x',\vec{v}\,'' \vdash &
d' = & \caseof{gp'q'}{\cdots} \;, &
p' = & \lambda x''.\,C'[d''] \;, \\
\Gamma,\vec{v},x',\vec{v}\,'',x'',\vec{v}\,'''' \vdash &
d'' = & \caseof{gp''q''}{\cdots} \;, &
p'' = & \lambda x'''.\,C''[d'''] \;, \\
& \cdots & & \cdots &
\end{array} \]
along with associated substitutions $^\dag$, $^{\dag'}$, $^{\dag''}, \ldots$
applicable to $d,d',d'',\ldots$ respectively, such that
$\dang{\!p^\dag\!}$, $\dang{(p')^{\dag'}\!}$, $\dang{(p'')^{\dag''}\!}, \ldots$
coincide with the successive procedure subterms $F,F',F'',\ldots$
from the spine of the original term $\dang{\!d^\dag\!}$,
and likewise $\dang{\!q^\dag\!}$, $\dang{(q')^{\dag'}\!}$, $\dang{(q'')^{\dag''}\!}, \ldots$
coincide with $o,o',o''\ldots$.
We cannot quite conclude that $d$ is head-spinal, because the critical $x$ in
$q^\dag$ might originate not from $q$ but from a level $k$ term in $\vec{s}$ (for example).
However, we can show that this problem does not arise for $q',q'',\ldots$, essentially because
$x',x'',\ldots$ are bound locally within $p$.
We will in fact show that $d'$ is head-spinal w.r.t.\ $x',\vec{v}\,''$,
where $\vec{v}\,''$ is the local environment for $C[-]$;
this will imply that $d$ is spinal.
In the light of Definition~\ref{spinal-def},
it will be sufficient to show that $(q')^{\circ'} \succeq {x'}^\eta$ for some $x',\vec{v}\,''$-closed
specialization $^{\circ'}$ covering the free variables of $q'$ except $x'$
(namely those of $\Gamma,\vec{v},\vec{v}\,''$);
the same argument will then obviously apply also to $q'',q''',\ldots$.
Recall that $\Gamma,\vec{v},\vec{v}\,'',x' \vdash q'$
and $\Gamma,\vec{v}\,',\vec{y}\,',x' \vdash o'$.
Since $gF'o'$ is head-spinal w.r.t.\ $x,\vec{y}\,'$,
we may as before take $^\circ = [\vec{w}\mapsto\vec{r}\,]$
$x',\vec{y}\,'$-closed such that ${o'}^\circ \succeq {x'}^\eta$,
where $\vec{w} = \Gamma,\vec{v}\,',\vec{y}\,'$ and $x' \vdash \vec{r}$.
Now define
\[ ^{\circ'} ~=~ [\vec{v} \mapsto \vec{s}\,^\circ, \;
\vec{v}\,'' \mapsto (\vec{s}\,'')^\circ, \;
\vec{w}\,^\Gamma\mapsto\vec{r}\,^\Gamma] \]
(where we write $\vec{s}\,^+$ as $\vec{s},\vec{s}\,''$,
and $\vec{w}\,^\Gamma \mapsto \vec{r}\,^\Gamma$ denotes the restriction of $^\circ$ to $\Gamma$).
This covers the free variables of $q'$ except $x'$, and we have
$x' \vdash \vec{s}\,^\circ, (\vec{s}\,'')^\circ, \vec{r}\,^\Gamma$ because
$\vec{w} \vdash \vec{s}, \vec{s}\,''$ and $x \vdash \vec{r}$.
Moreover, we have
\[ q'^{\circ'} ~=~ (q' [\vec{v}\,^+ \mapsto (\vec{s}\,^+)^\circ])[\vec{w}\,^\Gamma \mapsto \vec{r}\,^\Gamma] ~=~ (q'^{\dag'})^\circ ~\approx~ o^\circ ~\succeq~ x'^\eta \]
since $(\vec{s}\,^+)^\circ$ contains no free variables except $x$.
To check that $^{\circ'}$ is $x',\vec{v}\,''$-closed, it remains to show that
that $u^{\circ'}$ may contain $x'$ free only when $u \in \vec{v},''$ and $u$ is of level $<k$.
(Indeed, it is because of the possibility of $x'$ occurring free in these terms that
the machinery of $x,V$-closed substitutions is necessary at all.)
The remaining cases are handled as follows:
\begin{itemize}
\item The terms $\vec{s}$ exist in environment $\Gamma,\vec{v}\,'$, so do not involve $x'$
or any of the variables of $\vec{y}\,'$.
Since $^\circ$ is $x',\vec{y}\,'$-closed, it follows that the terms $\vec{s}\,^\circ$ do not involve $x'$.
\item For any variables $v \in \vec{v}\,''$ of level $k$, we have $v^{\dag'} = v^\eta$ which contains
no free variables of level $<k$, so that $(v^{\dag'})^\circ$ cannot involve $x'$.
\item The $\vec{r}\,^\Gamma$ cannot involve $x'$, since $^\circ$ is $x',\vec{y}\,'$-closed
and $\Gamma$ is disjoint from $\vec{y}\,'$.
\end{itemize}
This completes the verification that $d$ is spinal.
$\Box$
\end{proof}
\vspace*{1.0ex}
From the above lemma we may immediately conclude, for example,
that in the setting of Lemma~\ref{g-lemma-1}(ii) the term $d$ is spinal.
We are now ready for the main result of this section:
\begin{theorem} \label{no-gremlin-thm}
Every ${\mathrm{PCF}}^\Omega_k$-denotable procedure $\Gamma \vdash p$ is non-$g$-spinal
where $g: (k+1) \rightarrow (k+1)$.
\end{theorem}
\begin{proof}
In the light of Section~\ref{sec-denotations}, it will suffice to show that the
clauses of Theorem~\ref{denotable-inductive-thm}
cannot generate spinal terms from non-spinal ones.
For clauses 1--5 this is very easily seen.
For clause 6, it will be sufficient to show that non-spinal terms are closed under $k$-plugging,
and it is here that the machinery of Lemmas~\ref{g-lemma-1} and \ref{g-lemma-2} comes into play.
Suppose that $t =\, \dang{\Pi_{\Gamma,Z}(e,\xi)}$ where $\Gamma,Z \vdash e$
and $\Gamma,Z \vdash \xi(z)$ for each $z \in Z$.
For later convenience, to each $z_i \in Z$ let us associate the procedure
$\Gamma \vdash r_i =\, \dang{\Pi_{\Gamma,Z}(\xi(z_i),\xi)}$;
it is then clear from the definition of plugging and the evaluation theorem
that $t =\, \dang{e[\vec{z}\mapsto\vec{r}\,]}$
and that $r_i =\, \dang{\xi(z_i)[\vec{z}\mapsto\vec{r}\,]}$ for each $i$.
It will be natural to frame the argument contrapositively.
Suppose that $t$ is spinal, i.e.\ $t$ contains some head-spinal expression $c$
at position $K[-]$. We shall focus on the head occurrence of $g$ in $c$.
Clearly this occurrence must originate from one of the ingredients
$\Gamma,Z \vdash e$ or $\Gamma,Z \vdash \xi(z)$ of the plugging $\Pi_{\Gamma,Z}(e,\xi)$;
let us denote this ingredient by $\Gamma,Z \vdash t_0$.
We will show that $t_0$ itself is spinal.
Suppose that the relevant occurrence of $g$ in $t_0$ is as indicated by
\[ \Gamma,Z \;\vdash\; t_0 ~=~ L[d] \;, ~~~~~~
\Gamma,Z,\vec{v} \;\vdash\; d ~=~ \caseof{g p q}{\cdots} \;, \]
where $\vec{v}$ is the local environment for $L[-]$.
Writing $^\star$ for $[\vec{z}\mapsto\vec{r}\,]$, we have
$\dang{\Pi(t_0,\xi)} \,=\, \dang{t_0^\star}$;
and if $C[-]$ is the context encapsulating the remainder of the plugging $\Pi(e,\xi)$,
then we may write
\[ \Gamma \vdash~ K[c] ~=~ t ~=~ \dang{C[\Pi(t_0,\xi)]} ~=~ \dang{C[t_0^\star]}
~=~ \dang{C[L^\star[\dang{\!d^\star\!}]]} \;, \]
where
\[ \dang{\!d^\star\!} ~=~
\caseof{g \dang{\!p^\star\!} \dang{\!q^\star\!}\!}{\cdots} \;. \]
We claim that we are in the situation of Lemma~\ref{g-lemma-1},
taking $\Gamma,K,d,K',c$ of the lemma to be respectively
$\Gamma,C[L^\star[-]],\dang{\!\!d^\star\!\!},K,c$.
Condition~1 of the lemma holds because $C[t_0^\star]$ is constructed by substitution
from normal-form terms of level $\leq k$,
and condition~2 is immediate in the present setup.
By Lemma~\ref{g-lemma-1},
we may therefore conclude that
for a suitable substitution $[\vec{v} \mapsto \vec{s}\,]$
with $\Gamma, \vec{v}\,' \vdash \vec{s}$ regular,
$\Gamma \vdash\, \dang{\dang{\!d^\star\!}[\vec{v}\mapsto\vec{s}\,]}$ is head-spinal.
(Note that the local variables of $C[-]$ do not appear in $d^\star$, because $\Gamma,Z \vdash t_0$
and $\Gamma \vdash \vec{r}$.)
Equivalently, we may say that $\dang{d[\vec{v}\,^+\!\mapsto\vec{s}\,^+]}$
is head-spinal, where
\[ [\vec{v}\,^+\mapsto\vec{s}\,^+] ~=~
[\vec{z}\mapsto\vec{r}, ~\vec{v}\mapsto\vec{s}\,]
\;, \]
so that $\Gamma, \vec{v}\,' \vdash \vec{s}\,^+$
and ${\mathrm{lv}}(\vec{v}\,^+),{\mathrm{lv}}(\vec{v}\,') \leq k$.
(Note that the $\vec{z}$ do not appear free in $\vec{s}$, nor the $\vec{v}$ in $\vec{r}$.)
Since $\Gamma,Z,V \vdash d$ and $\Gamma,\vec{v}\,' \vdash \vec{s}\,^+$ are regular,
we are in the situation of Lemma~\ref{g-lemma-2}, so may conclude
that $d$ itself is spinal, and hence that $t_0$ is spinal.
We have thus shown that $k$-plugging cannot assemble spinal terms from non-spinal ones,
and this completes the proof.
$\Box$
\end{proof}
\vspace*{1.0ex}
In particular, since the procedure $g \vdash F_{k+1}[g]$ mentioned at the start of the section
is spinal, we may conclude that this procedure is not ${\mathrm{PCF}}^\Omega_k$-denotable,
and hence neither is the procedure $Y_{k+1}$. This establishes Theorem~\ref{SP0-main-thm}.
We conclude the section by mentioning some minor variations on Theorem~\ref{no-gremlin-thm}
that we will require below.
First, as already indicated, the whole of the above proof goes through for the modified notion
of spinal term appropriate to a variable $g: 0 \rightarrow (k+1) \rightarrow (k+1)$.
Secondly, the theorem also holds for an innocuous extension of
${\mathrm{PCF}}^\Omega_k$ with a constant ${\mathit{byval}} : (\nat\rightarrow\nat) \rightarrow \nat \rightarrow \nat$,
whose denotation in ${\mathsf{SP}}^0$ we take to be
\[ \lambda fx.\;\caseof{x}{i \Rightarrow \caseof{fi}{j \Rightarrow j}} \;. \]
To see that the proof of Theorem~\ref{no-gremlin-thm} goes through in the presence of ${\mathit{byval}}$,
it suffices simply to add an extra clause to the inductive proof noting that the procedure for ${\mathit{byval}}$
is non-spinal.
This mild extension will allow a significant simplification of the
forms of procedures that we need to consider in Section~\ref{sec-extensional}.%
\footnote{The operator ${\mathit{byval}}$ plays a major role in \cite[Section~7.1]{Longley-Normann},
where it is shown that every element of ${\mathsf{SP}}^0$ is denotable in ${\mathrm{PCF}}^\Omega + {\mathit{byval}}$.
The sense in which it is innocuous is that its denotation in ${\mathsf{SF}}$ coincides with that of
$\lambda fx.\,{\mathit{ifzero}}\,x\,(fx)(fx)$; thus ${\mathit{byval}}$ adds nothing to the expressivity of ${\mathrm{PCF}}^\Omega_k$
as regards ${\mathsf{SF}}$.}
\section{Non-definability in the extensional model} \label{sec-extensional}
To obtain corresponding non-definability results for ${\mathsf{SF}}$ rather than ${\mathsf{SP}}^0$,
one must show not only that the canonical procedures $Y_\tau$ considered above are not
${\mathrm{PCF}}^\Omega_k$-denotable, but also that no extensionally equivalent procedures are.
It is easy to see that there are indeed many other procedures $Z$
with the same extension as $Y_\tau$.
To give a trivial example, we may present the canonical procedure $Y_{k+1}$ as
$\lambda gx.C[g,x]$, where
\[ C[g,x] ~=~ \caseof{A[g,x]}{i \Rightarrow i} \;, ~~~~~~~~
A[g,x] ~=~ g(\lambda x'.C[g,x'])x^\eta \;. \]
However, another candidate for the fixed point operator is
\[ Z_0 ~=~ \lambda gx.\,\caseof{A[g,x]}{i \Rightarrow {C[g,x]}} \;. \]
Intuitively, this computes the desired value twice, discarding the first result.
As a slightly more subtle example, consider the procedure
\[ Z_1 ~=~ \lambda gx.\,\caseof{g(\lambda x'.\caseof{A[g,x]}{i \Rightarrow {C[g,x']}}) x^\eta}{k \Rightarrow k} \;. \]
Here, within the $\lambda x'$ subterm, we have smuggled in a repetition of the
top-level computation $A[g,x]$ before proceeding to evaluate what is really required.
The effect is that
$\lambda x'.\caseof{A[g,x]}{i \Rightarrow C[g,x']}$ may be
extensionally below $\lambda x'.C[g,x']$,
and this may indeed affect the result when $g$ is applied.
However, this can only happen when $Y_{k+1}gx$ is undefined anyway,
so it is easy to see that $Z_1$ as a whole will have the same extension as $Y_{k+1}$.
Yet another way to construct procedures extensionally equivalent to $Y_{k+1}$
is to vary the subterms of the form $x^\eta$ (where $x$ has type $k$).
For instance, in the case $k=1$, we could replace $x^\eta$ by an
extensionally equivalent term such as
\[ X_0 ~=~ \lambda y^0.\,
x(\lambda.\,\caseof{y}{0 \Rightarrow \caseof{x(\lambda.0)}{j \Rightarrow 0} \mid i+1 \Rightarrow i+1}) \;. \]
This is different in character from the previous examples:
rather than simply repeating the computation of $xy^\eta$, we are performing the specific
computation $x(\lambda.0)$ which we can see to be harmless given that
this point in the tree has been reached.
Clearly, such `time-wasting' tricks as the above may be combined and elaborated
to yield more complex examples of procedures equivalent to $Y$.
However, all of the above are rather innocuous variations
and do not really yield a fundamentally different method of computing fixed points.
For example, the bodies of both $Z_0,Z_1$ are still head-spinal terms,
and it is essentially the spines that are really computing the desired fixed point
by the canonical method.
This suggests that we should try to show that every procedure extensionally equivalent
to $Y_{k+1}$ is spinal;
from this it would follow easily by Theorem~\ref{no-gremlin-thm} that
the fixed point functional $Y_{k+1}$ in ${\mathsf{SF}}$
is not denotable in ${\mathrm{PCF}}^\Omega_k$.
Unfortunately, we are currently unable to show this in the case of $Y_{k+1}$:
indeed, the syntactic analysis of procedures $Z \approx Y_{k+1}$
appears to present considerable technical difficulties.
We shall establish the result for $Y_{0 \rightarrow (k+1)}$,
although even here, it is simplest to concentrate not on $Y_{0 \rightarrow (k+1)}$ itself,
but on a certain functional that is readily definable from it,
namely the functional $\Phi_{k+1}$ introduced in Section~\ref{sec-intro}.
Nonetheless, the above examples of `time-wasting' procedures illustrate some of the
situations that our proof will need to deal with, and they may help to motivate
some of the technical machinery that follows.
Specifically, within ${\mathrm{PCF}}_{k+1}$, let us define
\begin{eqnarray*}
\Phi_{k+1} & : & ({0} \rightarrow (k+1) \rightarrow (k+1))
\rightarrow ({0} \rightarrow (k+1)) \\
\Phi_{k+1}\,g^{0\rightarrow(k+1)\rightarrow(k+1)} & = &
Y_{{0}\rightarrow(k+1)}\, (\lambda f^{0\rightarrow(k+1)}.\lambda n.\,g\,n\,(f({\mathit{suc}}\;n))) \;,
\end{eqnarray*}
so that informally
\[ \Phi_{k+1}\,g\,n ~=~ g\,n\,(g\,(n+1)\,(g\,(n+2)\, (\cdots))) \;. \]
For the rest of this section we will write $\Phi_{k+1}$ simply as $\Phi$.
For each $n \in \mathbb{N}$, let $g \vdash \Phi^n[g] = \Phi\,g\,\num{n} : {k+1}$,
and let $p_n \in {\mathsf{SP}}(k+1)$ be the canonical NSP for $\Phi^n[g]$
(that is, the one arising from the above ${\mathrm{PCF}}$ definition
via the standard interpretation in ${\mathsf{SP}}^0$).
These procedures may be defined simultaneously by:
\[ g~ \vdash~ p_n ~=~ \lambda x^k.\,\caseof{g\,(\lambda.n)\,p_{n+1}\,x^\eta}{i \Rightarrow i}
~:~ k+1 \;. \]
By a syntactic analysis of the possible forms of (simple) procedures $g \vdash q \approx p_n$,
we will show that any such $q$ is necessarily spinal.
Here we have in mind the modified notion of spinal term that is applicable to
terms involving a global variable $g : \rho$, where $\rho = {0} \rightarrow (k+1) \rightarrow (k+1)$
(see the explanation following Definition~\ref{spinal-def}).
Using Theorem~\ref{no-gremlin-thm} (adapted to this modified setting),
it will then be easy to conclude that within ${\mathsf{SF}}$, the element
$\sem{\lambda g.\,\Phi^0[g]}$, and hence $Y_{{0}\rightarrow(k+1)}$ itself,
is not ${\mathrm{PCF}}^\Omega_k$-denotable in ${\mathsf{SF}}$.
To show that any $q \approx p_n$ is head-spinal, our approach will be as follows.
First, we show that any such $q$ must broadly resemble $p_n$ in at least its top-level
structure, in that $q$ must have the form $\lambda x.\,\caseof{garo}{\cdots}$,
where the arguments $a,r,o$ are closely related to the corresponding
arguments $(\lambda.n), p_{n+1}, x^\eta$ occurring within $p_n$.
We do this by showing that if $q$ were to deviate in any way from this prescribed form,
we would be able to cook up procedures $G \in {\mathsf{SP}}^0(\rho)$ and $X \in {\mathsf{SP}}^0(k)$
manifesting an extensional difference between $q$ and $p_n$, i.e.\ such that
$q[g \mapsto G]X \not\approx p_n[g \mapsto G]X$.
(Contrary to our usual convention, we will here use the uppercase letters $G,X$ to range over
normal-form closed procedures that may be substituted for $g,x$ respectively.)
In particular, we shall establish a sufficiently close relationship between $r$ and $p_{n+1}$
that the same analysis can then be iterated to arbitrary depth, showing that $q$ has
a spinal structure as required.
The main complication is that $r$ need not superficially resemble $p_{n+1}$,
since within $r$, the crucial application of $g$ that effectively computes the value of $p_{n+1}$
may be preceded by other `time-wasting' applications of $g$ (the idea is illustrated by the
example $Z_1$ above). However, it turns out that such time-wasting subterms
$g a^1 r^1 o^1$ must be of a certain kind if the extensional behaviour $q \approx p_n$
is not to be jeopardized: in particular, the first argument $a^1$ must evaluate to some
$i < n$. (As in the example of $Z_1$, the idea is that if the subterm $g a^1 r^1 o^1$
merely repeats some `outer' evaluation, it will make no overall difference to the extension
if the evaluation of this subterm does not terminate.)
In order to formulate the relationship between $r$ and $p_{n+1}$, we therefore need a
means of skipping past such time-wasting applications in order to reach the application of $g$
that does the real work.
We achieve this with the help of a \emph{masking} operator $\mu_{n,n'}$,
which (for any $n \leq n'$) overrides the behaviour of $g$ on numerical arguments $n \leq i < n'$
with a trivial behaviour returning the dummy value $0$.
We now proceed to our formal development.
As a brief comment on notation, we recall from Section~\ref{sec-background}
that the relations $\approx$ and $\preceq$ of observational equivalence and inequality
make sense not just for elements of ${\mathsf{SP}}^0$ but for arbitrary meta-terms (including applications), closed or otherwise.
Throughout this section, for typographical convenience,
we will tend to express the required relationships mostly at the level of meta-terms,
writing for instance $pq \approx \lambda.n$ rather than the equivalent $\dang{pq}\,=\lambda.n$
or $p \cdot q = \lambda.n$.
We shall also perpetrate other mild abuses of notation, such as writing
a procedure $\lambda.n$ simply as $n$ (except for special emphasis),
$\lambda \vec{x}.\bot$ as $\bot$, $x^\eta$ as $x$,
a meta-expression $\caseof{A}{i \Rightarrow i}$ just as $A$,
and abbreviating a substitution $[g \mapsto G,\, x \mapsto X]$ to $[G,X]$.
We shall say that $G \in {\mathsf{SP}}^0(0\rightarrow(k+1)\rightarrow(k+1))$ is \emph{strict}
if $G\bot ro \approx \bot$ for any $r,o$. Clearly, $G$ is strict iff
$G \approx \lambda izx.\,\caseof{i}{j \Rightarrow G (\lambda.j) z^\eta x^\eta}$.
In connection with meta-terms with free variable $g$, we shall write
$T \approx^\prime T'$ to mean that $T[g \mapsto G] \approx T'[g \mapsto G]$
for all \emph{strict} $G$; the notation $\preceq^\prime$ is used similarly.
We shall actually analyse the syntactic forms of procedures $g \vdash q$ based on
the assumption that $q \succeq^\prime p_n$,
where $p_n$ is the canonical procedure for $\Phi^n[g]$ as above.
We shall say a procedure $g \vdash q$ is \emph{simple} if for every application
$garo$ appearing within $q$, the first argument $a$ is just a numeral $\lambda.n$.
The following observation simplifies our analysis of terms considerably;
it uses the operator ${\mathit{byval}}$ and its NSP interpretation,
as introduced at the end of Section~\ref{sec-Y-k+1}.
\begin{proposition} \label{simple-prop}
If there is a procedure $g \vdash q \succeq^\prime p_n$-denotable in ${\mathrm{PCF}}^\Omega_k$,
then there is a simple procedure $g \vdash q' \succeq^\prime p_n$
denotable in ${\mathrm{PCF}}^\Omega_k + {\mathit{byval}}$.
\end{proposition}
\begin{proof}
Suppose $g \vdash q$ is ${\mathrm{PCF}}^\Omega_k$-denotable where $q \succeq^\prime p_n$, and write
\[ S[g] ~=~ \lambda izx.\,\caseof{i}{j \Rightarrow g(\lambda.j)z^\eta x^\eta} \;. \]
It is easy to see that
\[ S[g] ~=~ \sem{\lambda izx.\,{\mathit{byval}}\,(\lambda j. gjzx)\,i\,}_g \;. \]
Take $g \vdash q' =\, \dang{q[g \mapsto S[g]]}$,
so that $q'$ is denotable in ${\mathrm{PCF}}^\Omega_k + {\mathit{byval}}$.
Then $q \approx^\prime q'$ since $S[G] \approx G$ for all strict $G$, so
$q' \succeq^\prime p_n$. Finally, $q'$ is clearly simple: every occurrence of $g$ within
$q[g \mapsto S[g]]$ has a numeral as its first argument, so the same will be true of
$\dang{q[g \mapsto S[g]]}$.
$\Box$
\end{proof}
\vspace*{1.0ex}
For any $n \leq n'$, let us define the \emph{masking} $\mu_{n,n'}(g)$ of $g$
to be the following procedure term (here $\rho = 0 \rightarrow (k+1) \rightarrow (k+1)$):
\[ g^\rho ~\vdash~ \mu_{n,n'}(g) ~=~
\lambda izx.\,
\caseof{i}{n \Rightarrow 0 \mid \cdots \mid n'-1 \Rightarrow 0 \mid - \Rightarrow gizx} ~:~ \rho \;.
\]
(The wildcard symbol `$-$' covers all branch indices not covered by the preceding clauses.)
We may also write $\mu_{n,n'}(P)$ for $\mu_{n,n'}(g)[g \mapsto P]$ if $P$ is any
meta-procedure of type $\rho$.
We write $\mu_{n,n+1}$ simply as $\mu_n$; note also that $\mu_{n,n}(g) \approx^\prime g$.
Clearly $\mu_n(\mu_{n+1}(\cdots(\mu_{n'-1} (g)) \cdots)) \approx \mu_{n,n'}(g)$.
We shall say that a closed meta-term $\vdash P : \rho$ is \emph{trivial at $n$} if
$P(\lambda.n)\bot\bot \approx 0$.
Note that $\mu_{n,n'}(G)$ is trivial at each of $n,\ldots,n'-1$ for any closed $G$;
indeed, $G$ is trivial at $n,\ldots,n'-1$ iff $G \succeq^\prime \mu_{n,n'}(G)$.
The following lemma now implements our syntactic analysis of the top-level structure of simple procedures $q \succeq^\prime p_n$.
\begin{lemma} \label{Qn-lemma}
Suppose $g \vdash q$ is simple and $q \succeq^\prime p_n$. Then $q$ has the form
$\lambda x^k.\,\caseof{g a r o}{\cdots}$, where:
\begin{enumerate}
\item $a = \lambda.n$,
\item $o[g \mapsto G] \succeq x^\eta$ whenever $\vdash G$ is trivial at $n$,
\item $r[g \mapsto \mu_n(g),\, x \mapsto X] \succeq^\prime p_{n+1}$ for any $X$.
\end{enumerate}
\end{lemma}
\begin{proof}
Suppose $q = \lambda x^k.e$.
Clearly $e$ is not constant since $q \succeq p_n$;
and if $e$ had head variable $x$, we would have
$q[g \mapsto \lambda izx.\,\caseof{i}{- \Rightarrow 0}](\lambda w.\bot) = \bot$,
whereas
$p_n[g \mapsto \lambda izx.\,\caseof{i}{- \Rightarrow 0}](\lambda w.\bot) = 0$,
contradicting $q \succeq^\prime p_n$.
So $e$ has the form $\caseof{g a r o}{\cdots}$.
For claim~1, we have $a = \lambda.m$ for some $m \in \mathbb{N}$
because $q$ is simple. Suppose that $m \neq n$,
and consider
\[ G' ~=~ \lambda izx.\,\caseof{i}{n \Rightarrow 0 \mid - \Rightarrow \bot} \;. \]
Then for any $X$, clearly $q[G']X \approx \bot$, whereas
$p_n[G']X \approx G'(\lambda.n)(\cdots)X \approx 0$,
contradicting $q \succeq^\prime p_n$. Thus $a = \lambda.n$.
For claim~2, suppose that $G (\lambda.n) \bot \bot \approx 0$
but not $o[G] \succeq x^\eta$;
then we may take $X \in {\mathsf{SP}}^0({k})$ and $u \in {\mathsf{SP}}^0({k-1})$
such that $Xu \approx l \in \mathbb{N}$ but $o[G,X]u \not\approx l$. Now define
\[ G' ~=~ \lambda izx.\,\caseof{i}{j \Rightarrow \caseof{xu}{l \Rightarrow Gjzx \mid - \Rightarrow \bot}} \;. \]
Then $G' \preceq G$, so $o^* u \not\approx l$ where $^* = [G',X]$;
hence $G' a^* r^* o^* \approx \bot$ and so $q[G']X \approx \bot$.
On the other hand, we have
\begin{eqnarray*}
p_n[G']X & \approx & \caseof{G'(\lambda.n)p_{n+1}^*X}{i \Rightarrow i} \\
& \approx & \caseof{Xu}{l \Rightarrow G(\lambda.n)p_{n+1}^*X} ~\approx~ 0 \;,
\end{eqnarray*}
contradicting $q \succeq^\prime p_n$ (note that $G'$ is strict). Thus $o[G] \succeq x^\eta$.
For claim~3, suppose that $p_{n+1}[G]X' \approx l$ for some strict $G \in {\mathsf{SP}}^0(\rho)$
and $X' \in {\mathsf{SP}}^0(k)$.
We wish to show that $r[\mu_n(G), X]X' \approx l$ for any $X$.
Suppose not, and consider
\[ G' ~=~
\lambda izx.\,\caseof{i}{n \Rightarrow \caseof{zX'}{l \Rightarrow 0 \mid - \Rightarrow \bot} \mid - \Rightarrow Gizx} \;. \]
Then $G' \preceq \mu_n(G)$, so $r[G',X]X' \not\approx l$.
Moreover, since $a=\lambda.n$ by claim~1,
we see that $G' a^* r^* o^* \approx \bot$, where $^* = [G',X]$,
so that $q[G']X \approx \bot$.
On the other hand, we have
\begin{eqnarray*}
p_n[G']X & \approx & \caseof{G'(\lambda.n)p_{n+1}^*X}{i \Rightarrow i} \\
& \approx & \caseof{p_{n+1}^*X'}{l \Rightarrow 0} \;.
\end{eqnarray*}
Here, since $p_{n+1}$ does not contain $x$ free, we have $p_{n+1}^* = p_{n+1}[G']$.
But it is easy to see that $p_{n+1}[G'] \approx p_{n+1}[G]$, since
every occurrence of $g$ within $p_{n+1}$ is applied to $\lambda.n'$ for some $n'>n$,
and for all such $n'$ we have $G'(\lambda.n') \approx G(\lambda.n')$.
(Since $p_{n+1}$ contains infinitely many applications of $g$, an appeal to continuity is
formally required here.)
But $p_{n+1}[G]X' \approx l$ by assumption; thus $p_{n+1}^*X' \approx l$,
allowing us to complete the proof that $p_n[G']X \approx 0$.
Once again, this contradicts $q \succeq^\prime p_n$, so claim~3 is established.
$\Box$
\end{proof}
\vspace*{1.0ex}
In the light of Appendix~A, one may strengthen claim~2 of the above lemma
by writing $o[g \mapsto G] \approx x^\eta$.
This gives a fuller picture of the possible forms of terms $q \approx p_n$,
but is not needed for showing that such $q$ are spinal.
We have now almost completed a circle, in the sense that claim~3 tells us that
$\dang{r[g \mapsto \mu_n(g), x \mapsto X]}$ itself satisfies the hypothesis for $q$
(with $n+1$ in place of $n$).
However, there still remains a small mismatch between the hypothesis and the conclusion,
in that claim~3 concerns not $r$ itself but rather $\dang{r[g \mapsto \mu_n(g)]}$.
(In the light of claim~3, the variable $x$ may be safely ignored here.)
This mismatch is repaired by the following lemma, which lends itself to iteration to any depth.
Note the entry of a term context $E[-]$ here.
\begin{lemma} \label{q-structure-lemma}
Suppose $g \vdash q$ is simple and
$q[g \mapsto \mu_{n,n'}(g)] \succeq^\prime p_{n'}$.
Then $q$ has the form $\lambda x^k.\,E[\caseof{garo}{\cdots}]$, where:
\begin{enumerate}
\item $E[-]$ has empty local variable environment,
\item $a = \lambda.n'$,
\item $o[g \mapsto G] \succeq x^\eta$ whenever $\vdash G$ is trivial at $n,\cdots,n'$,
\item $r[g \mapsto \mu_{n,n'+1}(g),\,x \mapsto X] \succeq^\prime p_{n'+1}$
for any $X$.
\end{enumerate}
\end{lemma}
\begin{proof}
Let $q' =\, \dang{q[g \mapsto \mu_{n,n'}(g)]}$.
Under the above hypotheses, we have by Lemma~\ref{Qn-lemma} that $q'$ is of the form
$\lambda x.\,\caseof{ga'r'o'}{\cdots}$, where $a'=\lambda.n'$,
$o'[g \mapsto G] \succeq x^\eta$ whenever $G$ is trivial at $n'$,
and $r'[g \mapsto \mu_{n'}(g)] \succeq^\prime p_{n'+1}$.
Write $q$ as $\lambda x.\,E[\caseof{garo}{\cdots}]$ where the displayed
occurrence of $g$ originates the head $g$ of $q'$ via the substitution
$^\dag = [g \mapsto \mu_{n,n'}(g)]$.
Suppose that the hole in $E[-]$ appeared within an abstraction $\lambda y.-$;
then the hole in $E[g \mapsto \mu_{n,n'}(g)][-]$ would likewise appear within such an abstraction.
Moreover, the evaluation of $q[g \mapsto \mu_{n,n'}(g)]$ consists simply of the contraction of
certain expressions $\mu_{n,n'}(g)(\lambda.m)r''o''$ to either $0$ or $g(\lambda.m)r''o''$,
followed by some reductions $\caseof{0}{i \Rightarrow e_i} \rightsquigarrow e_0$;
thus, any residue in $q'$ of the critical $g$ identified above will likewise appear underneath $\lambda y$.
But this is impossible, because the head $g$ of $q'$ is a residue of this $g$ by assumption.
This establishes condition~1.
In the light of this, by Lemma~\ref{g-lemma-1}(i) we have
$a' \approx a^\dag$, $o' \approx o^\dag$ and $r' \approx r^\dag$.
But since $q$ is simple, $a$ is a numeral, so $a=\lambda.n'$, giving condition~2.
For condition~3, suppose $G$ is trivial at $n,\ldots,n'$.
Then $G \succeq \mu_{n,n'+1}(G)$, so that
\[ o[g \mapsto G] ~\succeq~ o[g \mapsto \mu_{n,n'}(\mu_{n'}(G))]
~\approx~ o^\dag[g \mapsto \mu_{n'}(G)] ~\approx~ o'[g \mapsto \mu_{n'}(G)]
~\succeq~ x^\eta \;, \]
since $\mu_{n'}(G)$ is trivial at $n'$.
Condition~4 also holds since
$r'[g \mapsto \mu_{n'}(g)] \approx r^\dag[g \mapsto \mu_{n'}(g)] \approx r[g \mapsto \mu_{n,n'+1}(g)]$, where
$r'[g \mapsto \mu_{n'}(g)] \succeq^\prime p_{n'+1}$.
$\Box$
\end{proof}
\begin{corollary} \label{spinal-cor}
If $g \vdash q$ is simple and $q \succeq^\prime p_n$, then $q$ is spinal
(in the modified sense).
\end{corollary}
\begin{proof}
Since condition~3 of the above lemma matches its hypotheses,
starting from the assumption that
$q \approx^\prime q[g \mapsto \mu_{n,n}(g)] \succeq^\prime p_n$,
we may apply the lemma iteratively to obtain a spinal structure as
prescribed by Definition~\ref{spinal-def}
(subject to the relevant adjustments for $g : 0 \rightarrow (k+1) \rightarrow (k+1)$).
Note that at each level, a suitable substitution $^\circ$ will be the closed one that
specializes $g$ to $\lambda izx.0$ (which is trivial at all $n$), and all variables $x^k$
other than the innermost-bound one to $\bot$.
$\Box$
\end{proof}
\vspace*{1.0ex}
Thus, if there exists a ${\mathrm{PCF}}^\Omega_k$-denotable procedure $t \approx \lambda g.p_0$, for instance,
then by Proposition~\ref{simple-prop} there is also a simple such procedure $t'$
denotable in ${\mathrm{PCF}}^\Omega_k + {\mathit{byval}}$,
and by Corollary~\ref{spinal-cor}, this $t'$ will be spinal in the modified sense.
But this contradicts Theorem~\ref{no-gremlin-thm}
(understood relative to the modified setting, and applied to ${\mathrm{PCF}}^\Omega_k + {\mathit{byval}}$ as indicated
at the end of Section~\ref{sec-Y-k+1}).
We conclude that no $t \approx \lambda g.p_0$ can be ${\mathrm{PCF}}^\Omega_k$-denotable.
Since the interpretation of ${\mathrm{PCF}}^\Omega_k$ in ${\mathsf{SF}}$ factors through ${\mathsf{SP}}^0$,
this in turn means that within the model ${\mathsf{SF}}$, the element $\sem{\lambda g.\,\Phi^0[g]}$
is not denotable in ${\mathrm{PCF}}^\Omega_k$.
On the other hand, this element is obviously denotable
relative to $Y_{{0}\rightarrow (k+1)} \in {\mathsf{SF}}$ even in ${\mathrm{PCF}}_1$,
so the proof of Theorems~\ref{operational-main-thm}(i) and \ref{SF-main-thm}(i) is complete.
This establishes Berger's conjecture, and also suffices for the proof of Corollary~\ref{no-finite-basis-cor}.
\section{Extensional non-definability of $Y_{k+1}$} \label{sec-pure-type}
We have so far shown that the element $Y_{0 \rightarrow (k+1)} \in {\mathsf{SF}}$ is not ${\mathrm{PCF}}^\Omega_k$-denotable.
We shall now refine our methods slightly to show that
even $Y_{k+1} \in {\mathsf{SF}}$ is not ${\mathrm{PCF}}^\Omega_k$-denotable.
Since $\pure{k+1}$ is clearly a ${\mathrm{PCF}}_0$-definable retract of every level $k+1$ type,
this will establish that no $Y_\sigma \in {\mathsf{SF}}$ where ${\mathrm{lv}}(\sigma)=k+1$ is denotable in
${\mathrm{PCF}}^\Omega_k$.
The idea is as follows. Within each type level $k \geq 1$, we can stratify the types into
\emph{sublevels} $(k,l)$ where $l = 1,2,\ldots$, essentially by taking account of the `width' of the type
as well as its depth.
We thus obtain a sublanguage ${\mathrm{PCF}}^\Omega_{k,l}$ of ${\mathrm{PCF}}^\Omega_k$ by admitting $Y_\sigma$
only for types $\sigma$ of sublevel $(k,l)$ or lower.
We show that, roughly speaking, all our previous methods can be adapted to show that for a given $k$,
the languages ${\mathrm{PCF}}^\Omega_{k,l}$ for $l = 1,2,\ldots$ form a strict hierarchy.
(This is true as regards definability in ${\mathsf{SP}}^0$; for ${\mathsf{SF}}$, we will actually show only that
${\mathrm{PCF}}^\Omega_{k,l}$ is strictly weaker than ${\mathrm{PCF}}^\Omega_{k,l+2}$.)
This more refined hierarchy is of some interest in its own right, and illustrates that the structure of ${\mathsf{SF}}$
is much richer than that manifested by the pure types.%
\footnote{This contrasts with the situation for extensional \emph{total} type structures over $\mathbb{N}$, for example.
There, under mild conditions, every simple type is isomorphic to a pure type:
see Theorem~4.2.9 of \cite{Longley-Normann}.}
Any term of ${\mathrm{PCF}}^\Omega_k$ will be a term of some ${\mathrm{PCF}}^\Omega_{k,l}$.
Previously we showed only that no such term could define $Y_{0 \rightarrow (k+1)} \in {\mathsf{SF}}$;
however, we now see that there is actually plenty of spare headroom between ${\mathrm{PCF}}^\Omega_{k,l}$
and the pure type $k+1$.
Specifically, there are operators $Y_\sigma$ in ${\mathrm{PCF}}^\Omega_{k,l+2}$
that are not ${\mathrm{PCF}}^\Omega_{k,l}$-denotable; and since all such $\sigma$ are of level $\leq k$
and are thus easily seen to be retracts of $\pure{k+1}$,
we may conclude that $Y_{k+1} \in {\mathsf{SF}}$ is not ${\mathrm{PCF}}^\Omega_k$-denotable.
The following definition sets out the more fine-grained stratification of types that we shall use.
\begin{definition} \label{width-def}
(i) The \emph{width} $w(\sigma)$ of a type $\sigma$ is defined inductively as follows:
\[ w(\nat) ~=~ 0 \;, ~~~~~~~~
w(\sigma_0,\ldots,\sigma_{r-1} \rightarrow \nat) ~=~ \max(r,w(\sigma_0),\ldots,w(\sigma_{r-1})) \;. \]
For $k,l \geq 1$, we say $\sigma$ has \emph{sublevel} $(k,l)$ if ${\mathrm{lv}}(\sigma)=k$ and $w(\sigma)=l$.
If $\sigma=\nat$, we simply say that $\sigma$ has \emph{sublevel} $0$.
We order sublevels in the obvious way: $0$ is the lowest sublevel,
and $(k,l) < (k',l')$ if either $k<k'$ or $k=k'$ and $l<l'$.
(ii) For each $k,l$ we define a type $\rho_{k,l}$ by:
\[ \rho_{0,l} ~=~ \nat \;, ~~~~~~~~
\rho_{k+1,l} ~=~ \rho_{k,l},\ldots,\rho_{k,l} \rightarrow \nat \mbox{~~($l$ arguments)} \;. \]
When $k,l \geq 1$, we may call $\rho_{k,l}$ the \emph{homogeneous} type of sublevel $(k,l)$.
\end{definition}
The following facts are easily established. Here we shall say that $\sigma$ is a \emph{simple retract} of $\tau$
if there is a ${\mathrm{PCF}}_0$-definable retraction $\sigma \lhd \tau$ within ${\mathsf{SP}}^0$.
\begin{proposition} \label{sublevel-facts}
(i) For $k \geq 1$, every type of sublevel $(k,l)$ or lower is a simple retract of $\rho_{k,l}$.
Hence, for all $k \geq 0$, for every finite list of types $\sigma_i$ of level $\leq k$, there is some $l$ such that
each $\sigma_i$ is a simple retract of $\rho_{k,l}$.
(ii) Every finite product of level $\leq k$ types is a simple retract of $\pure{k+1}$.
(iii) If $\sigma$ is a simple retract of $\tau$, then $Y_\sigma$ is ${\mathrm{PCF}}_0$-definable from $Y_\tau$ in ${\mathsf{SP}}_0$.
\end{proposition}
\begin{proof}
(i) The first claim (for $k \geq 1$) is easy by induction on $k$, and the second claim
(which is trivial when $k=0$) follows easily.
(ii) By induction on $k$. The case $k=0$ is easy. For $k \geq 1$,
suppose $\sigma_0,\ldots,\sigma_{m-1}$ are level $\leq k$ types,
where $\sigma_i = \tau_{i0},\ldots,\tau_{i(n_i-1)} \rightarrow \nat$ for each $i$.
Here the $\tau_{ij}$ are of level at most $k-1$, so by (i),
we may choose $l$ such that each $\tau_{ij}$ is a simple retract of $\rho_{k-1,l}$.
Taking $n = \max_i n_i$, we then have that each $\sigma_i$ is a simple retract of
$\rho_{k-1,l},\ldots,\rho_{k-1,l} \rightarrow \nat$ (with $n$ arguments).
The product $\Pi_i \sigma_i$ is therefore a simple retract of the type
$\sigma = \nat,\rho_{k-1,l},\ldots,\rho_{k-1,l} \rightarrow \nat$.
But by the induction hypothesis, the product of $\nat,\rho_{k-1,l},\ldots,\rho_{k-1,l}$
is a simple retract of $\pure{k}$ whence $\sigma$ itself is a simple retract of $\pure{k+1}$.
We leave (iii) as an exercise.
$\Box$
\end{proof}
\vspace*{1.0ex}
Next, we adapt the proof of Theorem~\ref{no-retraction-thm} to establish the crucial gap
between $\rho_{k,l}$ and $\rho_{k,l+1}$.
This gives an indication of how our methods may be used to map out the embeddability relation
between types in finer detail, although we leave an exhaustive investigation of this to future work.
\begin{theorem} \label{kl-no-retraction-thm}
Suppose $k \geq 1$. Within ${\mathsf{SF}}$, the type $\rho_{k,l+1}$ is not a pseudo-retract of any finite product
of types of sublevel $\leq (k,l)$ or lower.
\end{theorem}
\begin{proof}
In view of Proposition~\ref{sublevel-facts}(i), it will suffice to show that $\rho_{k,l+1}$ is not
a pseudo-retract of a finite power of $\rho_{k,l}$.
We argue by induction on $k$. The arguments for both the base case $k=1$ and the step case $k>1$
closely parallel the argument for the step case in Theorem~\ref{no-retraction-thm},
so we treat these two cases together as far as possible, omitting details that are easy adaptations
of those in the earlier proof.
Suppose for contradiction that there were procedures
\[ z^\rho \vdash t_i : \sigma ~~(i<m) \;, ~~~~~~~~
x_0^{\sigma},\ldots,x_{m-1}^{\sigma} \vdash r : \rho \;, \]
where $\rho = \rho_{k,l+1}$, $\sigma = \rho_{k,l}$,
such that $z \vdash r[\vec{x} \mapsto \vec{t}\,] \succeq z^\eta$.
Let $v =\, \dang{r[\vec{x} \mapsto \vec{t}\,]}$, so that $z \vdash v \succeq z^\eta$.
As in the proof of Theorem~\ref{no-retraction-thm}, one may show that
$v$ has the form $\lambda f_0 \ldots f_l.\,\caseof{zp_0 \ldots p_l}{\cdots}$
where $p_i[z \mapsto \lambda \vec{w}.0] \succeq f_i$ for each $i$.
Next, we note that $r[\vec{x} \mapsto \vec{t}\,]$ reduces in finitely many steps to a head normal form
$\lambda f_0 \ldots f_l.\, \caseof{zP_0\ldots P_l}{\cdots}$ where $\dang{P_i} \,= p_i$ for each $i$;
moreover, the ancestor of the leading $z$ here must lie within some $t_i$, say at the head of
some subterm $z\vec{P}\,'$, where $\vec{P}$ is an instance of $\vec{P}\,'$ via some substitution $^\dag$.
At this point, the arguments for the base case and step case part company.
In the base case $k=1$, we have that $z^\rho \vdash t_i : \sigma$
where $\rho = \nat^{l+1} \rightarrow \nat$ and $\sigma = \nat^l \rightarrow \nat$;
thus there are no bound variables within $t_i$ except the top-level ones---say $w_0,\ldots,w_{l-1}$,
all of type $\nat$. So in fact $^\dag$ has the form $[\vec{w} \mapsto \vec{W}]$ for certain meta-terms
$f_0^\nat,\ldots,f_l^\nat,z \vdash W_j : \nat$.
Now consider the terms $f_0,\ldots,f_l \vdash \vec{W}^*$ and $\vec{w} \vdash \vec{P}\,' {}^*$,
writing $^*$ for the substitution $[z \mapsto \lambda w.0]$.
These compose to yield $\vec{f} \, \vdash \, \dang{\vec{P}\,^*} \,=\, \dang{\vec{p}\,^*} \,\succeq f$,
so we have expressed $\nat^{l+1}$ as a pseudo-retract of $\nat^l$ within ${\mathsf{SF}}$.
As already noted in the course of the proof of Theorem~\ref{no-retraction-thm}, this is impossible.
For the step case $k>1$, we proceed much as in the proof of Theorem~\ref{no-retraction-thm},
using the substitution $^\dag$ to express $\rho_{k-1,l+1}$ as a retract of a finite product of types
of sublevel $(k-1,l)$ or lower, contrary to the induction hypothesis. We leave the remaining details
as an exercise.
$\Box$
\end{proof}
\vspace*{1.0ex}
We now outline how the ideas of Sections~\ref{sec-denotations}, \ref{sec-Y-k+1} and \ref{sec-extensional}
may be adapted to show that $Y_{\rho_{k,l+1}} \in {\mathsf{SP}}^0$ is not ${\mathrm{PCF}}^\Omega_{k,l}$-denotable.
We assume that $k \geq 2$ for the time being (the case $k=1$ will require special treatment).
The adaptations are mostly quite systematic: the type $\rho_{k,l+1}$
now plays the role of $\pure{k+1}$; types of sublevel $\leq (k,l)$ play the role of types of level $\leq k$;
$\rho_{k-1,l+1}$ plays the role of $\pure{k}$; and types of sublevel $\leq (k-1,l)$ play the role of types
of level $<k$. Since the proof we are adapting is quite lengthy, we leave many routine details to be checked
by the reader.
First, the evident adaptation of Definition~\ref{plugging-def} yields a notion of
\emph{$k,l$-plugging} where the plugging variables are required to be of sublevel $\leq (k,l)$,
and we thus obtain an inductive characterization of the ${\mathrm{PCF}}^\Omega_{k,l}$-denotable procedures
analogous to Theorem~\ref{denotable-inductive-thm}.
We also adapt the notion of \emph{regular} meta-term as follows:
\begin{definition}
Suppose $g : \rho_{k,l+1} \rightarrow \rho_{k,l+1}$ for $k,l \geq 1$.
An environment $\Gamma$ is \emph{$g$-($k,l$-)regular} if $\Gamma$ contains $g$ but no other variables
of sublevel $> (k,l)$. A meta-term $T$ is $g$-regular if it contains no variables of sublevel $> (k,l)$
except possibly for free occurrences of $g$. We say $\Gamma \vdash T$ is $g$-regular
if both $\Gamma$ and $T$ are $g$-regular.
\end{definition}
Next, we proceed to the ideas of Section~\ref{sec-Y-k+1}.
Our convention here will be that $\Gamma$ ranges over regular environments, and Roman capitals $V,Z$
range over sets of variables of sublevel $\leq (k,l)$.
The notions of $x,V$-closed substitution and spinal term carry over as follows:
\begin{definition} \label{kl-spinal-def}
Suppose $g : \rho_{k,l+1} \rightarrow \rho_{k,l+1}$ where $k \geq 2$ and $l \geq 1$.
(i) If $\vec{x}$ is a list of variables of type $\rho_{k-1,l+1}$ and $V$ a set of variables of sublevel
$\leq (k,l)$, a substitution $^\circ = [\vec{w} \mapsto \vec{r}\,]$ is called \emph{$\vec{x},V$-closed}
if the $r_i$ contain no free variables, except that if $w_i \in V$ and $w_i$ is of sublevel $\leq (k-1,l)$
then $r_i$ may contain the $\vec{x}$ free.
(ii) Suppose $\Gamma \vdash e$ is $g$-$k,l$-regular and $\vec{x},V \subseteq \Gamma$.
We coinductively declare $e$ to be \emph{$g$-head-spinal}
w.r.t.\ $\vec{x},V$ iff $e$ has the form
$\caseof{g(\lambda \vec{x}\,'.E[e'])\vec{o}}{\cdots}$, where $E[-]$ is an expression context, and
\begin{itemize}
\item for some $\vec{x},V$-closed specialization $^\circ$ covering the free variables of $\vec{o}$
other than those of $\vec{x}$, we have $\vec{o}\,^\circ \succeq \vec{x}^{\,\eta}$,
\item $e'$ is $g$-head-spinal w.r.t.\ $\vec{x}\,',V'$, where $V'$ is the local variable environment for $E[-]$.
\end{itemize}
(iii) We say a regular term $\Gamma \vdash t$ is \emph{$g$-spinal}
if it contains a $g$-head-spinal subexpression w.r.t.\ some $\vec{x},V$.
\end{definition}
Lemma~\ref{g-lemma-1} and its proof go through with the above adaptations;
here the local environments $\vec{v}, \vec{v}\,'$ are now of sublevel $\leq (k,l)$, and part~(iii)
of the lemma now states that if $K[-]$ contains no redexes with operator of sublevel $> (k,l)$, then
the substitution $^\dag$ is trivial for variables of sublevel $\geq (k-1,l+1)$.
The crucial Lemma~\ref{g-lemma-2}, which forms the heart of our proof, now translates as follows:
\begin{lemma} \label{kl-lemma-2}
Suppose $g: \rho_{k,l+1} \rightarrow \rho_{k,l+1}$ and we have $g$-regular terms
\[ \Gamma,\vec{v} \;\vdash~ d ~=~ \caseof{gpq}{\cdots} \;, ~~~~~~
\Gamma,\vec{v}\,' \vdash \vec{s} \,, \]
where $\vec{v},\vec{v}\,'$ are of sublevel $\leq (k,l)$, and
$\Gamma,\vec{v}\,' \vdash\, \dang{d[\vec{v}\mapsto\vec{s}\,]}$
is $g$-head-spinal with respect to some $\vec{x},V$.
Then $d$ itself is $g$-spinal.
\end{lemma}
The entire proof of this lemma translates systematically according to the correspondences we have indicated,
invoking Theorem~\ref{kl-no-retraction-thm} for the fact that
$\rho_{k-1,l+1}$ is not a pseudo-retract of a product of sublevel $\leq (k-1,l)$ types.
The analogue of Theorem~\ref{no-gremlin-thm} now goes through readily, so we obtain:
\begin{theorem}
If $k \geq 2$ and $l \geq 1$, every ${\mathrm{PCF}}^\Omega_{k,l}$-denotable procedure is non-$g$-spinal
where $g: \rho_{k,l+1} \rightarrow \rho_{k,l+1}$.
$\Box$
\end{theorem}
As in our original proof, we will actually use a version of this theorem for the modified notion of
spinal term that incorporates the extra argument $b$, and for the extension of ${\mathrm{PCF}}^\Omega_{k,l}$
with the operator ${\mathit{byval}}$.
To adapt the material of Section~\ref{sec-extensional}, we now take $g$ to be a variable of type
$0 \rightarrow \rho_{k,l+1} \rightarrow \rho_{k,l+1}$, and argue that the ${\mathrm{PCF}}_{k,l+2}$-denotable element
$\Phi = \lambda g.\, Y_{0 \rightarrow \rho_{k,l+1}} (\lambda fn.\,g\,n\,(f({\mathit{suc}}\,n)))$ within ${\mathsf{SF}}$
is not ${\mathrm{PCF}}^\Omega_{k,l}$-denotable. The proof is a completely routine adaptation of that in
Section~\ref{sec-extensional}. Since $Y_{0 \rightarrow \rho_{k,l+1}}$ is readily definable from $Y_{k+1}$
by Proposition~\ref{sublevel-facts}, this implies that $Y_{k+1} \in {\mathsf{SF}}$ is not ${\mathrm{PCF}}^\Omega_{k,l}$-denotable.
We have thus shown:
\begin{theorem}
(i) For $k \geq 2$ and $l \geq 1$, the element $Y_{0 \rightarrow \rho_{k,l+1}} \in {\mathsf{SF}}$ is denotable
in ${\mathrm{PCF}}_{k,l+2}$ but not in ${\mathrm{PCF}}^\Omega_{k,l}$.
(ii) For $k \geq 2$, the element $Y_{k+1} \in {\mathsf{SF}}$ is not denotable in ${\mathrm{PCF}}^\Omega_k$.
\end{theorem}
A slightly different approach is needed for the case $k=1$. This is because at level $0$ our only
type is $\nat$, so we are unable to make a distinction between sublevels $l$ and $l+1$.
To establish Lemma~\ref{kl-lemma-2} in this case, we again wish to show that we cannot pass in
the content of the relevant $\vec{x}\,'$ to the relevant $s_i$, but now the idea is to appeal to the
fact that $\vec{x}\,'$ consists of $l+1$ variables of type $\nat$, whereas $s_i$ accepts at most $l$
arguments of type $\nat$. (We have already seen that $\nat^{l+1}$ cannot be a retract of $\nat^l$.)
However, we also need to exclude the possibility that the substitution $^\circ$
is being used to import some components of $\vec{x}\,'$.
We can achieve this if we require $^\circ$ to be actually closed rather than just $\vec{x}\,',V'$-closed,
and it turns out that this is permissible if we also tighten our notion of spinal term slightly,
essentially to ensure that no intermediate applications
of $g$ appear in between those declared to constitute the spine of the term:
\begin{definition} \label{strongly-spinal-def}
Suppose $g : \rho_{1,l+1} \rightarrow \rho_{1,l+1}$ where $l \geq 1$.
Suppose $\Gamma \vdash e$ is $g$-$1,l$-regular and $\vec{x} \subseteq \Gamma$.
We coinductively declare $e$ to be \emph{strongly $g$-head-spinal} w.r.t.\ $\vec{x}$
iff $e$ has the form $\caseof{g(\lambda \vec{x}\,'.E[e'])\vec{o}}{\cdots}$,
where $E[-]$ is an expression context, and
\begin{itemize}
\item the hole within $E[-]$ does not itself occur within an application $gpq$,
\item for some closed substitution $^\circ$ covering the free variables of $\vec{o}$ other than
those of $\vec{x}$, we have $\vec{o}\,^\circ \succeq \vec{x}^{\,\eta}$,
\item $e'$ is strongly $g$-head-spinal w.r.t.\ $\vec{x}\,'$.
\end{itemize}
The notion of strongly $g$-spinal term follows suit.
\end{definition}
The counterpart of Lemma~\ref{g-lemma-1} goes through as expected, although without part~(iii):
the relevant sublevel distinction does not exist at type level $0$, and we cannot conclude that the
substitution in question is trivial for all variables of type $\nat$.
We may now indicate the required changes to Lemma~\ref{kl-lemma-2} and its proof:
\begin{lemma}
Suppose $g: \rho_{1,l+1} \rightarrow \rho_{1,l+1}$ and we have $1,l$-regular terms
\[ \Gamma,\vec{v} \;\vdash~ d ~=~ \caseof{gpq}{\cdots} \;, ~~~~~~
\Gamma,\vec{v}\,' \vdash \vec{s} \,, \]
where $\vec{v},\vec{v}\,'$ are of sublevel $\leq (1,l)$, and
$\Gamma,\vec{v}\,' \vdash\, \dang{d[\vec{v}\mapsto\vec{s}\,]}$
is strongly $g$-head-spinal with respect to some $\vec{x}$.
Then $d$ itself is strongly $g$-spinal.
\end{lemma}
\begin{proof}
The proof of Lemma~\ref{g-lemma-2} up to the end of the proof of Claim~1 adapts straightforwardly,
and is somewhat simplified by the fact that the substitution $^\circ$ is closed.
As sketched above, the crucial contradiction is provided by the fact that $\nat^{l+1}$ is not a pseudo-retract
of $\nat^l$ in ${\mathsf{SF}}$.
For the remainder of the proof, the key point to note is that $\vec{v}\,''$ (the local environment for $C[-]$)
is actually empty in this case. This is because $C[-]$ is in normal form and contains no free variables
of level $\geq 2$ except $g$, so any $\lambda$-term containing the hole would need to appear as an
argument to $g$. It would then follow that the hole within $E[-]$ lay within an argument to some occurrence of $g$, as precluded by the definition of strongly spinal term.
It follows trivially that the $\vec{x}\,',\vec{v}\,''$-closed substitution $^{\circ'}$ constructed at the very end
of the proof is actually closed. Moreover, the spinal structure of $d'$ identified by the proof cannot contain
any intermediate applications of $g$, since these would give rise under evaluation to intermediate applications
of $g$ in the spine of $\dang{d[\vec{v}\mapsto\vec{s}\,]}$ as precluded by
Definition~\ref{strongly-spinal-def}.
Thus, the identified spinal structure in $d'$ is actually a strongly spinal structure,
and the argument is complete.
$\Box$
\end{proof}
\vspace*{1.0ex}
A trivial adaptation of the proof of Theorem~\ref{no-gremlin-thm} now yields:
\begin{theorem}
No ${\mathrm{PCF}}^\Omega_{1,l}$-denotable procedure can be strongly $g$-spinal
where $g: \rho_{1,l+1} \rightarrow \rho_{1,l+1}$.
$\Box$
\end{theorem}
As before, this adapts easily to a variable $g$ of type $0 \rightarrow \rho_{1,l+1} \rightarrow \rho_{1,l+1}$.
From here on, we again follow the original proof closely.
The only additional point to note is that in place of Corollary~\ref{spinal-cor}
we now require that any simple $g \vdash q$ with $q \succeq' p_n$ must be \emph{strongly} spinal,
but this is already clear from the proof of Lemma~\ref{q-structure-lemma}.
We therefore have everything we need for:
\begin{theorem}
(i) For any $l \geq 1$, $Y_{\nat^{l+2} \rightarrow \nat} \in {\mathsf{SF}}$ is
not denotable in ${\mathrm{PCF}}^\Omega_{1,l}$.
(ii) $Y_2 \in {\mathsf{SF}}$ is not denotable in ${\mathrm{PCF}}^\Omega_1$.
$\Box$
\end{theorem}
The proof of Theorems~\ref{operational-main-thm} and \ref{SF-main-thm} is now complete.
\section{Related and future work} \label{sec-further-work}
\subsection{Other hierarchies of Y-combinators}
There have been a number of previous results from various research traditions showing that in some sense
the power of level $k$ recursions increases strictly with $k$.
Whilst many of these results look tantalizingly similar to ours,
it turns out on inspection that their mathematical substance is quite different, and
we do not expect any substantial technical connections with our own work to be forthcoming.
Nonetheless, it is interesting to see how strikingly different ideas and methods arising in other contexts
can lead to superficially similar results.
Previous results to the effect that $Y$-combinators for level $k+1$ are not definable
from those for level $k$ have been obtained by Damm \cite{Damm-IO} and
Statman \cite{Statman-lambda-Y}. It is convenient to discuss the latter of these first.
Statman works in the setting of the simply typed \emph{$\lambda Y$-calculus}, essentially the
pure $\lambda$-calculus extended with constants $Y_\sigma : (\sigma\rightarrow\sigma)\rightarrow\sigma$
and reduction rules $Y_\sigma M \rightsquigarrow M(Y_\sigma M)$.
He gives a succinct proof that $Y_{k+1}$ is not definable from $Y_k$
up to computational equality, based on the following idea.
If $Y_{k+1}$ were definable from $Y_k$, it would follow that
the recursion equation $Y_{k+1} g = g(Y_{k+1} g)$ could be derived with only finitely many
uses of the equation $Y_k M = M(Y_k M)$ (say $m$ of them).
It would then follow, roughly speaking, that $mn$ recursion unfoldings for $Y_k$
would suffice to fuel $n$ recursion unfoldings for $Y_{k+1}$.
On the other hand, it can be shown that the size of normal-form terms definable
using $n$ unfoldings of $Y_{k+1}$ (as a function of the size of the starting term) grows faster than
can be accounted for with $mn$ unfoldings of $Y_k$.
The language $\lambda Y$ is seemingly less powerful than ${\mathrm{PCF}}$,%
\footnote{One might consider translating ${\mathrm{PCF}}$ into $\lambda Y$ by representing natural numbers as Church numerals;
however, it appears that predecessor is not $\lambda Y$-definable for this representation.}
although this is perhaps not the most essential difference between Statman's work and ours.
More fundamentally, Statman's method establishes the non-definability only up to computational
equality (that is, the equivalence relation generated by the reduction rules),
whereas we have been concerned with non-definability modulo observational
(or extensional) equivalence.
Even for non-denotability in ${\mathsf{SP}}^0$, an approach along Statman's lines
would be unlikely to yield much information,
since there is no reason why the number of unfoldings of $Y_k$ required to generate
the NSP for $Y_{k+1}$ to depth $n$ should not grow dramatically as a function of $n$.
A result very similar to Statman's was obtained earlier in Damm \cite{Damm-IO},
but by a much more indirect route as part of a far-reaching investigation of the
theory of tree languages.
In Damm's setting, programs are \emph{recursion schemes}---%
essentially, families of simultaneous (and possibly mutually recursive) defining
equations in typed $\lambda$-calculus---but in essence these can be considered as terms of $\lambda Y$
relative to some signature consisting of typed constants.
(Actually, Damm's $\lambda$-terms involve a restriction to \emph{derived types},
which has the effect of limiting attention to what are elsewhere called \emph{safe} recursion schemes.)
Any such program can be expanded to an
infinite tree (essentially a kind of B\"ohm tree), and Damm's result (Theorem~9.8 of \cite{Damm-IO})
is that if programs are considered up to \emph{tree equality},
then safe level $k+1$ recursions give more expressive power than safe level $k$ ones.
Damm's result is thus distinguished from Statman's in two ways: by the restriction to safe recursion schemes,
and by the use of tree equality in place of the stricter computational equality.
This latter point brings Damm's work somewhat closer in spirit to our work:
indeed, in the case of pure $\lambda Y$, tree equality agrees with equality of innocent strategies
if the base type is interpreted as a certain trivial game---or equivalently with equality in a variant of
our ${\mathsf{SP}}^0$ with no ground type values.
However, in the case of a signature for ${\mathrm{PCF}}$, tree equality will still be considerably more fine-grained than
equality in ${\mathsf{SP}}^0$, let alone in ${\mathsf{SF}}$, since in effect the ${\mathrm{PCF}}$ constants are left uninterpreted.
It therefore seems unlikely that an approach to our theorems via Damm's methods is viable.
The strictness of the recursion scheme hierarchy was further investigated by Ong
\cite{Ong-model-checking}, who used innocent game semantics to show that the complexity of
certain model checking problems for trees generated by level $k$ recursions increases strictly with $k$.
It was also shown by Hague, Murawski, Ong and Serre \cite{Collapsible-PDA}
that the trees generated by level $k$ recursion schemes (with no safety restriction)
were precisely those that could be generated by \emph{collapsible pushdown automata} of order $k$.
Later, Kartzow and Parys \cite{Kartzow-Parys} used pumping lemma techniques to show that
the collapsible pushdown hierarchy was strict, hence so was the recursion scheme hierarchy
with no safety restriction.
Again, despite the intriguing parallels to our work, these results appear to be manifesting something
quite different:
in our setting, the power of level $k$ recursion has nothing to do with the `difficulty of computing' the relevant
NSPs in the sense of automata theory, since the full power of Turing machines is required even at level~1.
Also of interest is the work of Jones \cite{Jones-cons-free} from the functional programming community.
Jones's motivation is close to ours in that he seeks to give mathematical substance to the programming
intuition that some (combinations of) language features yield
`more expressive power' than others.
As Jones notes, an obvious obstacle to obtaining such results is that all programming languages
of practical interest are Turing complete, so that no such expressivity distinctions are visible at the level of the
(first-order) computable functions.
Whereas we have responded to this by considering the situation at higher types, where genuine
expressivity differences do manifest themselves, Jones investigates the power of (for example)
recursion at different type levels in the context of a restricted language of `read-only' or `cons-free' programs.
Amongst other results, he shows that in such a language, if data values and general recursion of type level
$k \geq 1$ are admitted, then the computable functions from lists of booleans to booleans are exactly the
{\sc{exp}$^k$\sc{time}} computable ones.
These results inhabit a mathematical territory very different from ours, although yet again,
the basic moral that the power of recursion increases strictly with its type level shines through.
\subsection{Relationship to game semantics}
Next, we comment briefly how our work relates
to the known \emph{game models} of PCF \cite{AJM,Hyland-Ong}.
It is known that these models are in fact isomorphic to our ${\mathsf{SP}}^0$,
although the equivalence between the game-theoretic definition of application and our own
is mathematically non-trivial (see \cite[Section~7.4]{Longley-Normann}).
This raises the obvious question of whether our proofs could be conducted equally well, or better, in a game-semantical setting.
Whilst a direct translation is presumably possible, our present impression
is that the sequential procedure presentation, and our calculus of meta-terms in particular,
allows one to see the wood for the trees more clearly,
and also to draw more easily on familiar intuitions from $\lambda$-calculus.
However, a closer look at game-semantical approaches would be needed in order to judge
whether either approach really offers some genuine mathematical or conceptual advantage over the other.
\subsection{Sublanguages of ${\mathrm{PCF}}$: further questions} \label{sublang-questions}
We now turn our attention to some potential extensions and generalizations of our work.
So far, we have worked mainly with a coarse stratification of types in terms of their levels,
although we have illustrated in Section~\ref{sec-pure-type} how finer stratifications are also possible.
Naturally, there is scope for a still more fine-grained analysis of types and the relative strength of their
$Y$-combinators; this is of course closely related to the task of mapping out the embeddability relation
between types (as in Subsection~\ref{subsec-embed}) in finer detail.
Even at level~1, there is some interest here.
Our analysis in Section~\ref{sec-pure-type} has shown that, for $l \geq 1$,
\begin{itemize}
\item the element $Y_{\nat^{l+1} \rightarrow \nat} \in {\mathsf{SP}}^0$ is not ${\mathrm{PCF}}^\Omega_0$-definable
from $Y_{\nat^l \rightarrow \nat}$,
\item the element $Y_{\nat^{l+2} \rightarrow \nat} \in {\mathsf{SF}}$ is not ${\mathrm{PCF}}^\Omega_0$-definable
from $Y_{\nat^l \rightarrow \nat}$.
\end{itemize}
However, this leaves us with a small gap for ${\mathsf{SF}}$: e.g.\ we have not shown either that
$Y_{\nat\rightarrow\nat}$ is strictly weaker than $Y_{\nat^2\rightarrow\nat}$ or that
$Y_{\nat^2\rightarrow\nat}$ is strictly weaker than $Y_{\nat^3\rightarrow\nat}$,
although according to classical logic, at least one of these must be the case.
(This is reminiscent of some well-known situations from complexity theory.)
We expect that a more delicate analysis will allow us to fill this gap.
One can also envisage an even more fine-grained hierarchy obtained by admitting other base types
such as the booleans or unit type.
A closely related task is to obtain analogous results for the \emph{call-by-value}
interpretation of the simple types (as embodied in Standard~ML, for example).
As is shown in \cite[Section~4.3]{Longley-Normann}, a call-by-value (partial) type structure
${\mathsf{SF}}_v$ can be constructed by fairly general means from ${\mathsf{SF}}$:
here, for example, ${\mathsf{SF}}_v(\pure{1})$ consists of all partial functions $\mathbb{N} \rightharpoonup \mathbb{N}$
rather than (monotone) total functions $\mathbb{N}_\bot \rightarrow \mathbb{N}_\bot$,
and ${\mathsf{SF}}_v(\pure{2})$ consists of partial functions $\mathbb{N}_\bot^\mathbb{N} \rightharpoonup \mathbb{N}$.
From known results on the interencodability of call-by-name and call-by-value models
(see \cite[Section~7.2]{Longley-Normann}), it is easy to read off the analogue of
Corollary~\ref{no-finite-basis-cor} for ${\mathsf{SF}}_v^{\mbox{\scriptsize \rm eff}}$;
however, more specific results on the relative strengths of various $Y$-combinators
within ${\mathsf{SF}}_v$ would require some further reworking of our arguments.%
Of course, one can also pose relative definability questions for other elements besides
$Y$-combinators.
For instance, it is natural to consider the \emph{higher-order primitive recursors} ${\mathit{rec}}_\sigma$
of System~T, as well as the closely related \emph{iterators} ${\mathit{iter}}_{\sigma\tau}$:
\begin{eqnarray*}
{\mathit{rec}}_\sigma & : &
\sigma \rightarrow (\sigma \rightarrow \nat \rightarrow \sigma) \rightarrow \nat \rightarrow \sigma \;, \\
{\mathit{iter}}_{\sigma\tau} & : & (\sigma\rightarrow (\sigma + \tau)) \rightarrow \sigma \rightarrow \tau \;.
\end{eqnarray*}
The idea behind the latter is to embody the behaviour of a \emph{while} construct for
imperative-style loops with state $\sigma$ and exit type $\tau$.%
\footnote{The sum type $\sigma + \tau$ is not officially part of our system, but can
(for any given $\sigma,\tau$) be represented as a retract of some existing simple type.}
It is shown in \cite[Section~6.3]{Longley-Normann} that each ${\mathit{rec}}_\sigma$ may be
interpreted by a \emph{left-well-founded} procedure (cf.\ Subsection~\ref{subsec-power-pcf1}),
and it is not hard to check that the same is true for each ${\mathit{iter}}_{\sigma\tau}$.
Furthermore, it is clear that all left-well-founded procedures are non-spinal,
so it requires only the addition of an easy base case in the proof of Theorem~\ref{no-gremlin-thm}
to show that $Y_{k+1}$ cannot be defined
even in ${\mathrm{PCF}}_k^\Omega$ extended with all primitive recursors and iterators.
(Actually, we can dispense with the primitive recursors here,
as it is straightforward to define them from suitable iterators.)
The dual question, roughly speaking, is whether any or all of the recursors ${\mathit{rec}}_\sigma$
or iterators ${\mathit{iter}}_{\sigma\nat}$ for types $\sigma$ of level $k+1$ are definable in ${\mathrm{PCF}}_k$.
We conjecture that they are not, and that this could be shown by suitably
choosing a substructure of ${\mathsf{SP}}^0$ so as to exclude such ${\mathit{iter}}_{\sigma\nat}$.
(This would incidentally answer Question~2 of \cite[Section~5]{Berger-min-recursion}.)
One could also look for substructures that more precisely determine the strength of
various \emph{bar recursion} operators or the \emph{fan functional}.
All in all, our experience leads us to expect that many further substructures of ${\mathsf{SP}}^0$
should be forthcoming, leading to a harvest of non-definability results exhibiting a rich
`degree structure' among ${\mathrm{PCF}}$-computable functionals.
Another very natural kind of question is the following: given a particular sublanguage ${\mathcal{L}}$
of ${\mathrm{PCF}}$, what is the \emph{simplest} possible type for an element of ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$ that is
not denotable in ${\mathcal{L}}$? Or to look at it another way: given a type $\sigma$, what is the
smallest natural sublanguage of ${\mathrm{PCF}}$ that suffices for defining all elements of
${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}(\sigma)$?
Here the analysis of \cite[Section~7.1]{Longley-Normann} yields several
positive definability results, whilst the analysis of the present paper provides ammunition on the
negative side. The current state of our knowledge is broadly as follows.
As in \cite{Longley-Normann}, we write ${\Klex^\smallmin}$ of the language of \emph{Kleene $\mu$-recursion}:
this is equivalent (in its power to denote elements of ${\mathsf{SF}}$)
to ${\mathrm{PCF}}_0$ extended with a strict primitive recursor for ground type and a strict iterator for ground type,
but with no form of general recursion.
\begin{itemize}
\item For first-order types $\sigma$, even ${\Klex^\smallmin}$ suffices for defining
all elements of ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}(\sigma)$; likewise, the oracle version ${\Klex^\smallmin}^\Omega$
suffices for ${\mathsf{SF}}(\sigma)$.
\item For second-order types $\sigma$ of the special form $(\nat\rightarrow\nat)^r \rightarrow \nat$,
${\Klex^\smallmin}^\Omega$ still suffices for ${\mathsf{SF}}(\sigma)$; however, this result is non-constructive,
and ${\Klex^\smallmin}$ does not suffice for ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}(\sigma)$.
(We draw here on some recent work of Dag Normann \cite{Normann-sequential-dcpo}.)
\item For general second-order types, ${\Klex^\smallmin}^\Omega$ no longer suffices, but
the languages ${\mathrm{PCF}}_1$, ${\mathrm{PCF}}_1^\Omega$ suffice for ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$, ${\mathsf{SF}}$ respectively---%
indeed, even the single recursion operator $Y_{\nat\rightarrow\nat}$ is enough here.
\item For third-order types, we do not know whether ${\mathrm{PCF}}_1$ suffices (for ${\mathsf{SF}}^{\mbox{\scriptsize \rm eff}}$).
We do know that ${\mathrm{PCF}}_2$ suffices, and that $Y_{\nat\rightarrow\nat}$ alone is not enough.
\item For types of level $k \geq 4$, ${\mathrm{PCF}}_{k-3}$ does not suffice,
but ${\mathrm{PCF}}_{k-2}$ does.
\end{itemize}
Again, there is scope for a more fine-grained view of the hierarchy of types.
\subsection{Other languages and models}
We have so far concentrated almost entirely on ${\mathrm{PCF}}$-style sequential computation.
To conclude, we briefly consider which other notions of higher-order computation
are likely to present us with an analogous situation.
As already noted at the end of Subsection~\ref{subsec-embed},
several extensions of ${\mathrm{PCF}}$ studied in the literature present a strikingly different picture:
in these settings, universal types exist quite low down, and as a consequence,
only $Y$-combinators of low type (along with the
other constants of the language) are required for full definability.
There is, however, one important language which appears to be more analogous
to pure ${\mathrm{PCF}}$ in these respects, namely an extension with \emph{local state}
(essentially Reynolds's \emph{Idealized Algol}).
This language was studied in \cite{Game-semantics}, where a fully abstract game model
was provided (consisting of well-bracketed but possibly non-innocent strategies).
Unpublished work by Jim Laird has shown that there is no universal type
in this setting.
We would conjecture also that the recursion hierarchy for this language is strict,
where we consider expressibility modulo observational equivalence in Idealized Algol.
This would constitute an interesting variation on our present results.
Related questions also arise in connection with \emph{total} functionals.
Consider, for example, the type structure ${\mathsf{Ct}}$ of total continuous functionals,
standardly constructed as the extensional collapse (relative to $\mathbb{N}$) of
the Scott model ${\mathsf{PC}}$ of partial continuous functionals.
It is shown by Normann \cite{Normann2000} that every effective element of ${\mathsf{Ct}}$
is represented by a \emph{${\mathrm{PCF}}$-denotable} element of ${\mathsf{PC}}$, and the proof actually
shows that ${\mathrm{PCF}}_1$ suffices here. (The further generalization of these ideas by
Longley \cite{Ubiquity} makes some use of second-order recursions as in ${\mathrm{PCF}}_2$;
we do not know whether these can be eliminated.) Thus, in this setting,
only recursions of low type are needed to obtain all computable functionals.
Similar remarks apply to the total type structure ${\mathsf{HEO}}$, obtained as the extensional collapse
of ${\mathsf{PC}}^{\mbox{\scriptsize \rm eff}}$.
On the other hand, one may consider the \emph{Kleene computable} functionals over ${\mathsf{Ct}}$,
or over the full set-theoretic type structure ${\mathsf{S}}$, as defined by the schemes S1--S9.
As explained in \cite[Chapter~6]{Longley-Normann}, sequential procedures can be
seen as abstracting out the algorithmic content common to both ${\mathrm{PCF}}$-style and
Kleene-style computation (note that Kleene's S9 in some sense does duty for the
$Y$-combinators of ${\mathrm{PCF}}$).
This naturally suggests that our strict hierarchy for ${\mathrm{PCF}}$ may have an analogue for
the Kleene computable functionals (say over ${\mathsf{S}}$ or ${\mathsf{Ct}}$), where at level $k$ we consider
the evident restriction of S9 to types of level $\leq k$.
We conjecture that this is indeed the case, although the required counterexamples may be
more difficult to find given that we are limited to working with total functionals.
\section*{Appendix A: Super-identity procedures}
In the course of our proof, we have frequently encountered assertions of the form
$p \succeq x^\eta$ for various procedures $x^k \vdash p$.
Although not necessary for our main argument, it is natural to ask whether there are
any such procedures other than those for which $p \approx x^\eta$.
The following theorem shows that the answer is no: in other words,
no procedure $\lambda x.p : \pure{k} \rightarrow \pure{k}$
can extensionally `improve on' the identity function.
We present this as a result of some interest in its own right,
whose proof is perhaps less trivial than one might expect.
Recall that $\preceq$ denotes the extensional order on ${\mathsf{SF}}$,
as well as the associated preorder on ${\mathsf{SP}}^0$.
Within ${\mathsf{SF}}$, we will write $f \prec f'$ to mean $f \preceq f'$ but $f \neq f'$;
we also write $f \sharp f'$ to mean that $f,f'$ have no upper bound
with respect to $\preceq$.
We call an element of ${\mathsf{SP}}^0$ \emph{finite} if it is a finite tree once branches
of the form $i \Rightarrow \bot$ have been deleted.
We say an element of ${\mathsf{SF}}$ is \emph{finite} if it is represented by some finite
element of ${\mathsf{SP}}^0$.
We write ${\mathsf{SP}}^{0,{\mbox{\scriptsize \rm fin}}}(\sigma), {\mathsf{SF}}^{\mbox{\scriptsize \rm fin}}(\sigma)$ for the set of finite elements
in ${\mathsf{SP}}^0(\sigma), {\mathsf{SF}}(\sigma)$ respectively.
\begin{theorem} \label{super-identity-thm}
(i) If $f \in {\mathsf{SF}}^{\mbox{\scriptsize \rm fin}}(k)$ and $f \prec f'$, then there exists $f'' \sharp f'$
with $f \prec f''$.
(ii) If $\Phi \in {\mathsf{SF}}(k \rightarrow k)$ and $\Phi \succeq {\mathit{id}}$,
there can be no $f \in {\mathsf{SF}}(k)$ with $\Phi(f) \succ f$; hence $\Phi = {\mathit{id}}$.
\end{theorem}
\begin{proof}
(i) The cases $k=0,1$ are easy, so let us assume $k \geq 2$.
Suppose $f \prec f'$ where $f$ is finite.
Then for some $g \in {\mathsf{SF}}(k-1)$ we have $f(g)=\bot$ but $f'(g)=n \in \mathbb{N}$, say,
and by continuity in ${\mathsf{SP}}^0$ we may take $g$ here to be finite.
Take $p,q \in {\mathsf{SP}}^{0,{\mbox{\scriptsize \rm fin}}}$ representing $f,g$ respectively;
we may assume that $p,q$ are `pruned' so that every subtree containing no
leaves $m \in \mathbb{N}$ must itself be $\bot$.
\textit{Case 1:} $g(\bot^{k-2}) = a \in \mathbb{N}$.
In this case, we may suppose that $q = \lambda x.a$.
Consider the computation of $p \cdot q$.
Since all calls to $q$ trivially evaluate to $a$, this computation follows the
rightward path through $p$ consisting of branches $a \Rightarrow \cdots$.
But since $p$ is finite and $p \cdot q = \bot$ (because $f(g)=\bot$),
this path must end in a leaf occurrence of $\bot$ within $p$.
Now extend $p$ to a procedure $p'$ by replacing this leaf occurrence
by some $n' \neq n$; then clearly $p' \cdot q = n'$.
Taking $f''$ to be the function represented by $p'$, we then have
$f \preceq f''$ and $f''(g) = n' \,\sharp\, n = f'(g)$, so $f'' \sharp f'$
(whence also $f'' \neq f$ so $f \prec f''$).
\textit{Case 2:} $g(\bot^{k-2}) = \bot$.
Take $N$ larger than $n$ and all numbers appearing in $p,q$.
Define $p' \sqsupseteq p$ as follows: if $p = \lambda x.\bot$, take $p' = \lambda x.N$,
otherwise obtain $p'$ from $p$ by replacing each case branch $j \Rightarrow \bot$
anywhere within $p$ by $j \Rightarrow N$ whenever $j \leq N$.
Extend $q$ to $q'$ in the same way.
Note in particular that every ${\mathtt{case}}$ subterm within $p',q'$ will now be equipped
with a branch $N \Rightarrow N$.
Now consider the computation of $p \cdot q$.
Since $p,q$ are finite and $f(g)=\bot$, this evaluates to an occurrence of $\bot$
which originates from $p$ or $q$. Since no numbers $>N$ ever arise in the computation,
this occurrence of $\bot$ cannot be part of a branch $j \Rightarrow \bot$ for $j>N$,
so will have been replaced by $N$ in $p'$ or $q'$.
Now suppose that we head-reduce $pq$ until $\bot$ first appears in head position,
and let $U$ be the resulting meta-term.
Then it is easy to see that $p'q'$ correspondingly reduces to a meta-term $U'$
with $N$ in head position. (Formally, we reason here by induction on the length
of head-reduction sequences not involving the rule for $\caseof{\bot}{\cdots}$.)
We now claim that $p' \cdot q' = N$. Informally, this is because the head
occurrence of $N$ in $U'$ will be propagated to the top level by the case branches
$N \Rightarrow N$ within both $p'$ and $q'$.
More formally, let us define the set of meta-expressions \emph{led by $N$}
inductively as follows:
\begin{itemize}
\item $N$ is led by $N$.
\item If $E$ is led by $N$, then so is $\caseof{E}{i \Rightarrow D_i}$.
\end{itemize}
We say that an NSP meta-term $T$ is \emph{saturated at $N$}
if every ${\mathtt{case}}$ subterm within $T$ has a branch $N \Rightarrow E$
where $E$ is led by $N$.
Clearly $p'q'$ is saturated at $N$, and it is
easy to check that the terms saturated at $N$ are closed under head reductions;
thus $U'$ is saturated at $N$.
But we have also seen that $U'$ has $N$ in head position, so is led by $N$.
Finally, an easy induction on term size shows that
every finite meta-term that is led by $N$ and saturated at $N$ evaluates to $N$ itself.
This shows that $p' \cdot q' = N$.
To conclude, let $f'',g'$ be the functions represented by $p',q'$ respectively,
so that $f \preceq f''$ and $g \preceq g'$.
Then $f'(g')=n$, but $p'' \cdot q = N$ so $f''(g')=N \neq n$, whence $f'' \sharp f'$
(and also $f'' \neq f$ so $f \prec f''$).
(ii) Suppose $\Phi \succeq {\mathit{id}}$ and $\Phi(f) \succ f$ for some $f$.
Again by continuity, we may take $f$ to be finite.
Then by (i), we may take $f'' \succ f$ such that $f'' \,\sharp\, \Phi(f)$.
But this is impossible because $\Phi(f'') \succeq f''$ and $\Phi(f'') \succeq \Phi(f)$.
Thus $\Phi={\mathit{id}}$.
$\Box$
\end{proof}
\vspace*{1.0ex}
It is easy to see that the above theorem holds with any finite type over $\nat$
in place of $\pure{k}$.
However, it will trivially fail if the unit type $\unit$ is admitted as an additional base type:
e.g.\ the function $(\lambda x.\top) \in {\mathsf{SF}}(\unit\rightarrow\unit)$ strictly dominates the identity.
An interesting question is whether the theorem holds for all finite types over
the type $\bool$ of booleans: note that the above proof fails here
since it requires the base type to be infinite.
For comparison, we mention that in other models of computation,
improvements on the identity are sometimes possible for such types.
For example, if $\sigma=\bool\rightarrow\bool$, then a functional
of type $\sigma\rightarrow\sigma$ strictly dominating the identity exists in the Scott model.
Indeed, such a function $J$ can be defined in
${\mathrm{PCF}}$ augmented with the parallel conditional ${\mathit{pif}}$, e.g.\ as
\[ J ~=~ \lambda f^\sigma.\lambda x^\bool.\,\mathit{vote}(f(x),f({\mathit{tt}}),f({\mathit{ff}})) \;. \]
Here $\mathit{vote}$ is Sazonov's voting function, definable by
\[ \mathit{vote}(x,y,z) ~=~ {\mathit{pif}}(x,{\mathit{pif}}(y,{\mathit{tt}},z),{\mathit{pif}}(y,z,{\mathit{ff}})) \;. \]
The point is that $J$ will `improve' the function $\lambda x.\,{\mathit{if}}(x,{\mathit{tt}},{\mathit{tt}})$
to $\lambda x.{\mathit{tt}}$.
We do not know whether phenomena of this kind can arise within the model ${\mathsf{SF}}$.
| 0dfa0ce7332e04960bc644a671086f5ea5e0bbd1 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
In the conference, I talked about
the Hodge theoretic aspect of injectivity and
vanishing theorems (see \cite{fujino1}, \cite{fujino2},
and \cite{fujino3}).
Here, I will explain some analytic generalizations.
In \cite{fujino-matsumura}, Shin-ichi Matsumura and I
established the
following theorems.
\begin{thm}[{\cite[Theorem A]{fujino-matsumura}}]\label{f-thm1}
Let $F$ be a holomorphic line bundle
on a compact K\"ahler manifold $X$ and let $h$ be a singular
hermitian metric on $F$.
Let $M$ be a holomorphic line bundle on $X$
equipped with a smooth hermitian metric $h_M$.
We assume that
$$
\sqrt{-1}\Theta_{h_M}(M)\geq 0 \quad
\text{and}\quad \sqrt{-1}\Theta_h(F)-a\sqrt{-1}\Theta_{h_M}(M)\geq 0
$$ for some $a>0$.
Let $s$ be a nonzero global section of $M$.
Then the map
$$
\times s: H^i(X, \omega_X\otimes F\otimes
\mathcal J(h))\to H^i(X, \omega_X\otimes F\otimes \mathcal J(h)\otimes M)
$$
induced by $\otimes s$ is injective for every $i$, where
$\omega_X$ is the canonical bundle of $X$ and
$\mathcal J(h)$ is the multiplier ideal sheaf of $h$.
\end{thm}
Theorem \ref{f-thm1} is a generalization of Enoki's injectivity theorem
(see \cite[Theorem 0.2]{enoki}).
Although the formulation of Theorem
\ref{f-thm1} may look artificial, it
has many interesting applications (see \cite{fujino-matsumura}).
Theorem \ref{f-thm2} below is a Bertini-type theorem for multiplier ideal
sheaves.
\begin{thm}[{\cite[Theorem 1.10]{fujino-matsumura}}]\label{f-thm2}
Let $X$ be a compact complex manifold, let $\Lambda$ be a free
linear system on $X$ with $\dim \Lambda\geq 1$, and
let $\varphi$ be a quasi-plurisubharmonic
function on $X$.
We put
$$
\mathcal G=\{ H\in \Lambda \, |\, {\text{$H$ is smooth
and $\mathcal J(\varphi|_H)=\mathcal J(\varphi)|_H$}}\}.
$$ Then $\mathcal G$ is dense in $\Lambda$ in the classical topology.
Note that $\mathcal J (\varphi)$ is the multiplier ideal
sheaf of $\varphi$.
\end{thm}
The main purpose of this
paper is to prove the following theorem, which is a
slight generalization of \cite[Theorem D]{fujino-matsumura},
as an application of Theorem \ref{f-thm1} and Theorem \ref{f-thm2}.
\begin{thm}[Vanishing theorem of Koll\'ar--Nadel type]\label{f-thm3}
Let $f:X\to Y$ be a holomorphic map from
a compact K\"ahler manifold $X$ to
a projective variety $Y$.
Let $F$ be a holomorphic
line bundle on $X$ equipped with
a singular hermitian metric $h$.
Let $H$ be an ample line bundle on $Y$.
Assume that there exists a smooth hermitian metric $g$ on
$f^*H$ such that
$$
\sqrt{-1}\Theta_g(f^*H)\geq 0 \quad \text{and}\quad
\sqrt{-1}\Theta_h(F)-\varepsilon
\sqrt{-1}\Theta_g(f^*H)\geq 0
$$
for some $\varepsilon >0$.
Then we have
$$
H^i(Y, R^jf_*(\omega_X\otimes F\otimes \mathcal J(h)))=0
$$
for every $i>0$ and $j$,
where $\omega_X$ is the canonical bundle of $X$
and $\mathcal J(h)$ is the multiplier ideal
sheaf associated to the singular hermitian metric $h$.
\end{thm}
We can easily see that Theorem \ref{f-thm3}
contains Demailly's original formulation of the Nadel vanishing
theorem (see \cite[Theorem 1.4]{fujino-matsumura})
and Koll\'ar's vanishing theorem (see \cite[Theorem 2.1 (iii)]{kollar})
as special cases.
Therefore, we call Theorem \ref{f-thm3} the vanishing theorem
of Koll\'ar--Nadel type.
For a related vanishing theorem,
see \cite[Theorem 1.3]{matsumura}.
In this paper, we will freely use the same notation as in \cite{fujino-matsumura}.
\section{Proof of Theorem \ref{f-thm3}}
In this section, we prove Theorem \ref{f-thm3} as an application of
Theorem \ref{f-thm1} and Theorem \ref{f-thm2}.
I hope that the following proof will show the reader how to use
Theorem \ref{f-thm1} and Theorem \ref{f-thm2}.
\begin{proof}[Proof of Theorem \ref{f-thm3}]
We use the induction on $\dim Y$.
If $\dim Y=0$, then the statement is obvious.
We take a sufficiently large positive integer $m$ and a general
member $B\in |H^{\otimes m}|$ such that
$D=f^{-1}(B)$ is smooth,
contains no associated primes of $\mathcal O_X/\mathcal J(h)$,
and satisfies $\mathcal J(h|_D)=\mathcal
J(h)|_D$ by Theorem \ref{f-thm2}.
By the Serre vanishing theorem, we may further assume that
\begin{equation}\label{f-eq-1}
H^i(Y, R^jf_*(\omega_X\otimes F\otimes \mathcal J(h))
\otimes H^{\otimes m})=0
\end{equation}
for every $i>0$ and $j$.
We have the following big commutative diagram.
$$
\xymatrix{
& 0\ar[d] & 0\ar[d]&\\
0 \ar[r]& \mathcal J(h) \otimes \mathcal O_X(-D)
\ar[d]
\ar[r]^{\quad \quad \alpha}
&\mathcal J(h)\ar[d]\ar[r]& \Coker \alpha
\ar[r]\ar[d]^{\beta}& 0 \\
0 \ar[r]& \mathcal O_X(-D)\ar[d]
\ar[r]
&\mathcal O_X \ar[d]\ar[r]& \mathcal O_{D}\ar[r]& 0 \\
&\left(\mathcal O_X/\mathcal J(h)\right)\otimes
\mathcal O_X(-D) \ar[r]^{\quad \quad \quad \gamma}
\ar[d]& \mathcal O_X/\mathcal J(h)\ar[d]&\\
& 0& 0&
}
$$
Since $D$ contains no associated primes of $\mathcal O_X/\mathcal J(h)$,
$\gamma$ is injective.
This implies that $\beta$ is injective by the
snake lemma and that $\Coker \alpha=\mathcal J(h)|_D
=\mathcal J(h|_D)$.
Thus we obtain the following short exact sequence:
$$
0\to \mathcal J(h)\otimes \mathcal O_X(-D)\to
\mathcal J(h)\to \mathcal J(h|_D)\to 0.
$$
By taking $\otimes \omega_X\otimes F\otimes \mathcal O_X(D)$
and using adjunction,
we obtain the short exact sequence:
$$
0\to \omega_X\otimes F\otimes \mathcal J(h)
\to \omega_X\otimes F\otimes \mathcal J(h) \otimes f^*H^{\otimes m}
\to \omega_D\otimes F|_D\otimes \mathcal J(h|_D)\to 0.
$$
Therefore, we see that
\begin{equation}\label{f-eq-2}
\begin{split}
0 &\to R^jf_*(\omega_X\otimes F\otimes \mathcal J(h))
\to R^jf_*(\omega_X\otimes F\otimes \mathcal J(h))
\otimes H^{\otimes m}\\&
\to R^jf_*(\omega_D\otimes F|_D\otimes \mathcal J(h|_D))\to 0
\end{split}
\end{equation}
is exact
for every $j$ since $B$ is a general member of $|H^{\otimes m}|$.
By induction on $\dim Y$, we have
\begin{equation}\label{f-eq-3}
H^i(B, R^jf_*(\omega_D\otimes F|_D\otimes
\mathcal J(h|_D)))=0
\end{equation} for
every $i>0$ and $j$.
By taking the long exact sequence associated to \eqref{f-eq-2},
we obtain
\begin{equation*}
\begin{split}
H^i(Y, R^jf_*(\omega_X\otimes F\otimes \mathcal J(h)))
=H^i(Y, R^jf_*(\omega_X\otimes F\otimes \mathcal J(h))
\otimes H^{\otimes m})
\end{split}
\end{equation*}
for every $i\geq 2$ and $j$ by \eqref{f-eq-3}.
Thus we have
\begin{equation}\label{f-eq-4}
H^i(Y, R^jf_*(\omega_X\otimes F\otimes \mathcal J(h)))=0
\end{equation}
for every $i\geq 2$ and $j$ by \eqref{f-eq-1}.
By Leray's spectral sequence and \eqref{f-eq-1} and \eqref{f-eq-4},
we have the
following commutative diagram:
$$
\xymatrix{
H^1(Y, \mathcal F^j)\ar[d]_\alpha
\ar@{^{(}->}[r]& H^{j+1}(X,
\omega_X\otimes F\otimes \mathcal J(h))
\ar@{^{(}->}[d]^{\beta}
\\
H^1(Y, \mathcal F^j \otimes H^{\otimes m}) \ar@{^{(}->}[r]& H^{j+1}(X,
\omega_X\otimes F\otimes \mathcal J(h)\otimes
f^*H^{\otimes m})
}
$$
for every $j$,
where $\mathcal F^j=R^jf_*(\omega_X\otimes F\otimes \mathcal
J(h))$. Note that the horizontal arrows are
injective.
Since $\beta$ is injective by Theorem \ref{f-thm1},
we obtain that $\alpha$ is also injective.
By \eqref{f-eq-1}, we have
\begin{equation*}
H^1(Y, R^jf_*(\omega_X\otimes F\otimes \mathcal J(h))
\otimes H^{\otimes m})=0
\end{equation*}
for every $j$.
Therefore, we see that $H^1(Y,
R^jf_*(\omega_X
\otimes F\otimes \mathcal J(h)))=0$ for every $j$.
Thus we obtain the desired vanishing theorem:~Theorem \ref{f-thm3}.
\end{proof}
We close this section with a remark on Nakano semipositive
vector bundles.
\begin{rem}\label{f-rem2.1}
Let $E$ be a Nakano semipositive
vector bundle on $X$.
We can easily see that Theorem \ref{f-thm3} holds
even when $\omega_X$ is replaced by $\omega_X\otimes E$.
We leave the details as an exercise for the reader
(see \cite[Section 6]{fujino-matsumura}).
\end{rem}
\begin{ack}
I was partially supported by Grant-in-Aid for Young Scientists (A) 24684002,
Grant-in-Aid for Scientific Research (S) 16H06337,
and Grant-in-Aid for Scientific Research (B) 16H03925 from JSPS.
I thank Shin-ichi Matsumura very much whose
comments made Theorem \ref{f-thm3}
better than my original formulation.
\end{ack}
| fc48fcc883b22ea89b990b13ee7a6268b795a8e3 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
In this note, we study the area of a minimal submanifold in the unit ball in $\mathbb{R}^n$. For a minimal submanifold $\Sigma$ that passes through the center of the ball, it is well-known that the area of $\Sigma$ is bounded from below by the area of a flat $k$-dimensional disk:
\begin{theorem}
\label{consequence.of.monotonicity}
Let $\Sigma$ be a $k$-dimensional minimal submanifold in the unit ball $B^n$ which passes through the origin and satisfies $\partial \Sigma \subset \partial B^n$. Then $|\Sigma| \geq |B^k|$.
\end{theorem}
Theorem \ref{consequence.of.monotonicity} is a direct consequence of the well-known monotonicity formula for minimal submanifolds. This technique is discussed, for example, in \cite{Ekholm-White-Wienholtz} and \cite{Simon}.
In 1973, Alexander and Osserman \cite{Alexander-Osserman} studied a closely related problem. More precisely, they considered a minimal surface in the unit ball in $\mathbb{R}^3$ which passes through a prescribed point in the interior of the ball (not necessarily the center of the ball). In the special case of disk-type minimal surfaces, they were able to show that the area of the surface is bounded from below by the area of a flat disk. However, their argument does not work for minimal surfaces of other topological types, nor does it generalize to higher dimensions. In 1974, Alexander, Hoffman, and Osserman \cite{Alexander-Hoffman-Osserman} proved an analogous inequality in higher dimensions, but only in the special case of area-minimizing surfaces.
In this note, we completely settle this question for minimal submanifolds of arbitrary dimension and codimension:
\begin{theorem}
\label{main.theorem}
Let $\Sigma$ be a $k$-dimensional minimal submanifold in the unit ball $B^n$ which passes through a point $y \in B^n$ and satisfies $\partial \Sigma \subset \partial B^n$. Then $|\Sigma| \geq |B^k| \, (1-|y|^2)^{\frac{k}{2}}$. Moreover, the inequality is strict unless $\Sigma$ is a flat $k$-dimensional disk which is orthogonal to $y$.
\end{theorem}
The proof of Theorem \ref{main.theorem} relies on an application of the first variation formula for minimal submanifolds (cf. \cite{Allard}, \cite{Simon}) to a carefully chosen vector field in ambient space. In particular, our argument generalizes immediately to the varifold setting. A similar technique was used in \cite{Brendle} to prove a sharp bound for the area of a free-boundary minimal surface in a ball. The main difficulty in this approach is to find the correct vector field. The vector field used in \cite{Brendle} was obtained as the gradient of the Green's function for the Neumann problem on the unit ball. By contrast, the vector field used in the proof of Theorem \ref{main.theorem} is not a gradient field, and does not have any obvious geometric interpretation.
\section{Proof of Theorem \ref{main.theorem}}
Let us fix a point $y \in B^n$. We define a vector field $W$ on $B^n \setminus \{y\}$ in the following way: For $k>2$, we define
\begin{align*}
W(x)
&= -\frac{1}{k} \, \bigg ( \Big ( \frac{1-2 \langle x,y \rangle+|y|^2}{|x-y|^2} \Big )^{\frac{k}{2}} - 1 \bigg ) \, (x-y) \\
&+ \frac{1}{k-2} \, \bigg ( \Big ( \frac{1-2 \langle x,y \rangle+|y|^2}{|x-y|^2} \Big )^{\frac{k-2}{2}} - 1 \bigg ) \, y.
\end{align*}
For $k=2$, we define
\begin{align*}
W(x)
&= -\frac{1}{2} \, \Big ( \frac{1-2 \langle x,y \rangle+|y|^2}{|x-y|^2} - 1 \Big ) \, (x-y) \\
&+ \frac{1}{2} \, \log \Big ( \frac{1-2 \langle x,y \rangle+|y|^2}{|x-y|^2} \Big ) \, y.
\end{align*}
Note that $1-2 \langle x,y \rangle+|y|^2 \geq |x-y|^2 > 0$ for all points $x \in B^n \setminus \{y\}$. This shows that $W$ is indeed a smooth vector field on $B^n \setminus \{y\}$.
\begin{lemma}
\label{a}
For every point $x \in B^n$ and every orthonormal $k$-frame $\{e_1,\hdots,e_k\} \subset \mathbb{R}^n$, we have
\[\sum_{i=1}^k \langle D_{e_i} W,e_i \rangle \leq 1.\]
\end{lemma}
\textbf{Proof.}
We compute
\begin{align*}
\sum_{i=1}^k \langle D_{e_i} W,e_i \rangle
&= 1 - \Big ( \frac{1-2 \langle x,y \rangle+|y|^2}{|x-y|^2} \Big )^{\frac{k}{2}} \\
&+ \frac{(1-2 \langle x,y \rangle+|y|^2)^{\frac{k-2}{2}}}{|x-y|^k} \, \sum_{i=1}^k \langle y,e_i \rangle \, \langle x-y,e_i \rangle \\
&+ \frac{(1-2 \langle x,y \rangle+|y|^2)^{\frac{k}{2}}}{|x-y|^{k+2}} \, \sum_{i=1}^k \langle x-y,e_i \rangle^2 \\
&- \frac{(1-2 \langle x,y \rangle+|y|^2)^{\frac{k-4}{2}}}{|x-y|^{k-2}} \, \sum_{i=1}^k \langle y,e_i \rangle^2 \\
&- \frac{(1-2 \langle x,y \rangle+|y|^2)^{\frac{k-2}{2}}}{|x-y|^k} \, \sum_{i=1}^k \langle x-y,e_i \rangle \, \langle y,e_i \rangle \\
&= 1 - \frac{(1-2 \langle x,y \rangle+|y|^2)^{\frac{k}{2}}}{|x-y|^{k+2}} \, \Big ( |x-y|^2 - \sum_{i=1}^k \langle x-y,e_i \rangle^2 \Big ) \\
&- \frac{(1-2 \langle x,y \rangle+|y|^2)^{\frac{k-4}{2}}}{|x-y|^{k-2}} \, \sum_{i=1}^k \langle y,e_i \rangle^2 \\
&\leq 1.
\end{align*}
Note that the preceding calculation is valid both for $k > 2$ and for $k=2$. This proves the assertion. \\
\begin{lemma}
\label{b}
The vector field $W$ vanishes along the boundary $\partial B^n$.
\end{lemma}
\textbf{Proof.}
Suppose that $x \in \partial B^n$. Then $1-2 \langle x,y \rangle+|y|^2 = |x-y|^2$. This directly implies $W(x)=0$. Again, this conclusion holds both for $k>2$ and for $k=2$. \\
\begin{lemma}
\label{c}
We have
\[W(x) = -(1-|y|^2)^{\frac{k}{2}} \, \frac{x-y}{k \, |x-y|^k} + o \Big ( \frac{1}{|x-y|^{k-1}} \Big )\]
as $x \to y$.
\end{lemma}
\textbf{Proof.}
By definition of $W(x)$, we have
\begin{align*}
W(x)
&= -(1-2 \langle x,y \rangle+|y|^2)^{\frac{k}{2}} \, \frac{x-y}{k \, |x-y|^k} + o \Big ( \frac{1}{|x-y|^{k-1}} \Big ) \\
&= -(1-|y|^2)^{\frac{k}{2}} \, \frac{x-y}{k \, |x-y|^k} + o \Big ( \frac{1}{|x-y|^{k-1}} \Big )
\end{align*}
as $x \to y$. This proves the assertion. \\
We now describe the proof of Theorem \ref{main.theorem}. To that end, we assume that $\Sigma$ is a minimal surface in $B^n$ passing through the point $y$. Since the vector field $W$ vanishes along the boundary $\partial \Sigma \subset \partial B^n$, we obtain
\begin{equation}
\label{first.variation}
\int_{\Sigma \setminus B_r(y)} (1-\text{\rm div}_\Sigma W) = |\Sigma \setminus B_r(y)| - \int_{\Sigma \cap \partial B_r(y)} \langle W,\nu \rangle
\end{equation}
by the divergence theorem. Here, $\nu$ denotes the inward pointing unit normal to the region $\Sigma \cap B_r(y)$ within the surface $\Sigma$. In other words, the vector $\nu$ is tangential to $\Sigma$, but normal to $\Sigma \cap \partial B_r(y)$. It is easy to see that
\[\nu = -\frac{x-y}{|x-y|} + o(1)\]
for $x \in \Sigma \cap \partial B_r(y)$. Using Lemma \ref{c}, we obtain
\[\langle W,\nu \rangle = (1-|y|^2)^{\frac{k}{2}} \, \frac{1}{k \, r^{k-1}} + o \Big ( \frac{1}{r^{k-1}} \Big )\]
for $x \in \Sigma \cap \partial B_r(y)$. Since
\[|\Sigma \cap \partial B_r(y)| = |\partial B^k| \, r^{k-1} + o(r^{k-1}),\]
we conclude that
\begin{equation}
\label{term.1}
\lim_{r \to 0} \int_{\Sigma \cap \partial B_r(y)} \langle W,\nu \rangle = \frac{1}{k} \, |\partial B^k| \, (1-|y|^2)^{\frac{k}{2}} = |B^k| \, (1-|y|^2)^{\frac{k}{2}}.
\end{equation}
Combining (\ref{first.variation}) and (\ref{term.1}) gives
\[\lim_{r \to 0} \int_{\Sigma \setminus B_r(y)} (1 - \text{\rm div}_\Sigma W) = |\Sigma| - |B^k| \, (1-|y|^2)^{\frac{k}{2}}.\]
On the other hand, by Lemma \ref{a} we have the pointwise inequality
\[1 - \text{\rm div}_\Sigma W \geq 0.\]
Putting these facts together, we obtain $|\Sigma| - |B^k| \, (1-|y|^2)^{\frac{k}{2}} \geq 0$, as claimed.
Finally, we study the case of equality. Suppose that $|\Sigma| - |B^k| \, (1-|y|^2)^{\frac{k}{2}} = 0$. In this case, we have
\[1 - \text{\rm div}_\Sigma W = 0\]
for each point $x \in \Sigma \setminus \{y\}$. Hence, if $x$ is an arbitrary point on $\Sigma \setminus \{y\}$ and $\{e_1,\hdots,e_k\}$ is an orthonormal basis of $T_x \Sigma$, then we have
\[|x-y|^2 - \sum_{i=1}^k \langle x-y,e_i \rangle^2 = \sum_{i=1}^k \langle y,e_i \rangle^2 = 0.\]
This implies that $\Sigma$ is a flat $k$-dimensional disk which is orthogonal to $y$. This completes the proof of Theorem \ref{main.theorem}.
| 4bdd4dc78d8ad2b18f09cce337723b7fedfac883 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
In this paper we study bounds for the density of a diffusion process at a small time under a local strong H\"ormander condition. To be more precise, let $X$ denote the process in ${\mathbb R}^n$ solution to
\be{equation-intro}
dX_{t}=\sum_{j=1}^{d}\sigma
_{j}(t,X_{t})\circ dW_{t}^{j}+b(t,X_{t})dt,\quad \quad X_0=x_0.
\end{equation}
where $W=(W^{1},...,W^{d})$ is a standard Brownian motion and $\circ
dW_{t}^{j} $ denotes the Stratonovich integral.
We assume nice differentiability and boundedness or sublinearity properties for the diffusion coefficients $b$ and $\sigma_j$, $j=1,\ldots,d$, and we consider a degenerate case:
\begin{equation}\label{dim}
\mathrm{dim}\,\sigma(0,x_0)=\mathrm{dim}\,\mathrm{Span}\{\sigma_1(0,x_0), \ldots,\sigma_d(0,x_0)\}<n,
\end{equation}
$\mathrm{dim}\,S$ denoting the dimension of the vector space $S$.
Our aim is to study lower and upper bounds for the density of the solution to
\eqref{equation-intro} at a small (but not asymptotic) time, say $\delta$, under the following local strong H\"ormander condition:
\begin{equation}\label{hor-intro}
\mathrm{Span}\{\sigma_i(0,x_0),[\sigma_p,\sigma_j](0,x_0),\ i,p,j=1,\ldots,d\}={\mathbb R}^n
\end{equation}
in which $[\cdot,\cdot]$ denotes the standard Lie bracket vector field. Notice that we ask for a H\"orman\-der condition at time $0$. Our estimates are written in terms of a norm which reflects the
non-isotropic structure of the problem: roughly speaking, in a small time interval of length $\delta$, the diffusion process moves with speed $\sqrt{\delta}$ in the direction of the diffusion vector fields $\sigma _{j}$ and with speed $\delta=\sqrt{\delta}\times \sqrt{\delta}$ in the direction of $[\sigma _{i},\sigma _{j}]$.
In order to catch this behavior we introduce
the following norms. Let $A_\delta(0,x_0)$ denote the $n\times d^2$ matrix
$$
A_\delta(0,x_0)=[A_{1,\delta}(0,x_0),\ldots,A_{d^2,\delta}(0,x_0)]
$$
where the general column $A_{l,\delta}(0,x_0)$, $l=1,\ldots,d^2$, is defined as follows:
\begin{itemize}
\item
for $l=(p-1)d+i$ with $p,i\in\{1,\dots,d\}$ and $p\neq i$ then
$$
A_{l,\delta}(0,x_0) =[\sqrt\delta\sigma_{i},\sqrt\delta\sigma_{p}](0,x_0)
=\delta[\sigma_{i},\sigma_{p}](0,x_0);
$$
\item
for $l=(p-1)d+i$ with $p,i\in\{1,\dots,d\}$ and $p= i$ then
$$
A_{l,\delta}(0,x_0) =\sqrt\delta\,\sigma_{i}(0,x_0).
$$
\end{itemize}
Under \eqref{hor-intro}, the rank of $A_\delta(0,x_0)$ is equal to $n$, hence the following norm is well defined:
$$
|\xi|_{A_\delta(0,x_0)}=\<(A_\delta(0,x_0)A_\delta(0,x_0)^T)^{-1}\xi,\xi\>^{1/2},\quad \xi\in{\mathbb R}^n,
$$
where the supscript $T$ denotes the transpose and $\<\cdot,\cdot\>$ stands for the standard scalar product. We prove in \cite{BCP2} that the metric
given by this norm is locally equivalent with the control distance $d_{c}$
(the Carath\'eodory distance) which is usually used in this framework. We denote by $p_{\delta}(x_0,\cdot)$ the density of the solution to \eqref{equation-intro} at time $\delta$. Under \eqref{hor-intro} and assuming suitable hypotheses on the boundedness and sublinearity of the coefficients $b$ and $\sigma_j$, $j=1,\ldots, d$ (see Assumption \ref{assumption1} for details), we prove the following result (recall $\mathrm{dim}\,\sigma(0,x_0)$ given in \eqref{dim}):
\begin{description}
\item[]
\textsc{[lower bound]}
there exist positive constants $r,\delta_*,C$ such that for every $\delta\leq \delta_*$ and for every $y$ with $|y- x_0-b(0,x_0)\delta|_{A_\delta(0,x_0)}\leq r$ one has
\[
p_\delta(x_0,y)\geq \frac{1}{C \delta^{n-\frac{\mathrm{dim}\,\sigma(0,x_0)}{2} }};
\]
\item []
\textsc{[upper bound]}
for any $p>1$, there exists a positive constant $C$ such that for every $\delta\leq 1$ and for every $y\in {\mathbb R}^n$ one has
\[
p_\delta(x_0,y)\leq \frac{1}{\delta^{n-\frac{\mathrm{dim}\,\sigma(0,x_0)}{2}}
} \frac{C}{1+|y-x_0|_{A_\delta(0,x_0)}^p}.
\]
\end{description}
This is stated in Theorem \ref{main-th}, where an exponential upper bound is achieved as well, provided that stronger boundedness assumptions on the diffusion coefficients hold (see Assumption \ref{assumption1bis}).
In the context of a degenerate diffusion coefficient which fulfills a strong H\"ormander condition, the main result in this direction is due to Kusuoka and Stroock. In the celebrated paper \cite{KusuokaStroock:87}, they prove the following two-sided Gaussian bounds: there exists a constant $M\geq 1$ such that
\be{ksintro}
\begin{split}
&\frac{1}{M |B_{d_c}(x_0,\delta^{1/2})|}\exp\left( -\frac{M d_c(x_0,y)^2}{\delta}\right)\\
&\quad \quad \quad \leq p_\delta(x_0,y)
\leq \frac{M}{|B_{d_c}(x_0,\delta^{1/2})|}\exp\left( -\frac{d_c(x_0,y)^2}{M \delta}\right)
\end{split}
\end{equation}
where $\delta\in (0,1]$, $x_0,y\in {\mathbb R}^n$, $B_{d_c}(x,r)=\{y\in {\mathbb R}^n: d_c(x,y)<r\}$, $d_c$ denoting the control (Carath\'eodory) distance, and $|B_{d_c}(x,r)|$ stands for the Lebesgue measure of $B_{d_c}(x,r)$. It is worth to be said that \eqref{ksintro} holds under special hypotheses: in \cite{KusuokaStroock:87} it is assumed that the coefficients do not depend on the time variable and that $b(x)=\sum_{j=1}^d \alpha_i \sigma_i(x)$, with $\alpha_i\in C_b^\infty({\mathbb R}^n)$ (i.e. the drift is generated by the vector fields of the diffusive part, which is a quite restrictive hypothesis).
Other
celebrated estimates for the heat kernel under strong H\"{o}rmander
condition are provided in \cite{BenArousLeandre:91, BenArousLeandreII:91}.
The subject has also been widely studied by analytical methods - see
for example \cite{Jerison1987} and \cite{Sanchez-Calle1984}. We stress that these are asymptotic results, whereas we prove estimate for a finite, positive and fixed time. In \cite{pigato:14}, \cite{DelarueMenozzi:10}, non-isotropic norms similar to $|\cdot|_{A_\delta(0,x_0)}$ are used to
provide density estimates for SDEs under H\"ormander conditions of weak type. We also refer to \cite{CattiauxMesnager:02}, which considers the existence of the density for SDEs with time dependent coefficients, under very weak regularity assumption.
The paper \cite{BCP2} will follow the present work, considering related results from a ``control'' point of view, discussing in particular tube estimates and the connection with the control-Carath\'eodory distance.
Tube estimates are estimates on the probability that an It\^o process remains around a deterministic path up to a given time. These will be obtained from a concatenation of the short-time density estimates presented here.
Then we will consider, as in \cite{KusuokaStroock:87}, $b(t,x)=b(x)$ and $\sigma(t,x)=\sigma(x)$. Defining the semi distance $d$ via: $d(x,y) < \sqrt{\delta}$ if and only if $|x-y|_{A_\delta(x)}<1$, we will prove in \cite{BCP2} the local equivalence of $d$ and $d_c$. This will give a rewriting of the upper/lower estimates of the density in terms of the control distance as well.
The paper is organized as follows. In Section \ref{settings} we set-up the framework and give the precise statement of our main result (Theorem \ref{main-th}). The proof is split in two sections: Section \ref{lower}, which is devoted to the lower bound, and Section \ref{upper}, in which we deal with the upper bound. The main tool we are going to use is given by the estimates of localized densities which have been developed in \cite{BC14}. These need to use techniques from Malliavin calculus, so we briefly report in Appendix \ref{app-mall} all these arguments. But in order to set-up our program we also require some other facts, which have been collected in other appendices. First, we use a key decomposition of the solution $X_\delta$ to \eqref{equation-intro} at a small (but not asymptotic) time $\delta$ (see Section \ref{lower-key}), and we postpone the proof in Appendix \ref{app-dec}. This decomposition allows us to work with a random variable whose law, conditional to a suitable $\sigma$-algebra, is Gaussian, and in Appendix \ref{app-supp} we study some useful support properties that are applied to our case. Moreover, since the key-decomposition brings to handle a perturbed Gaussian random variable, in Appendix \ref{sectioninversefunction} we prove density estimates via local inversion for such kind of random variables.
\section{Notations and main results}\label{settings}
We need to recall some notations. For $f,g:{\mathbb R}^{+}\times {\mathbb R}^{n}\rightarrow {\mathbb R}^{n}$ we define the directional derivative (w.r.t. the space variable $x$) $\partial_{g}f(t,x)=\sum_{i=1}^{n}g^{i}(t,x)\partial _{x_{i}}f(t,x)$, and we recall that the Lie bracket (again w.r.t. the space variable)
is defined as $[g,f](t,x)=\partial _{g}f(t,x)-\partial _{f}g(t,x)$.
Let $M\in \mathcal{M}_{n\times m}$ be a matrix with full row rank. We write $M^T$ for the transposed matrix, and $MM^T$ is invertible. We denote by $\lambda_*(M)$ (respectively $\lambda^*(M))$ the smallest (respectively the largest) singular value of $M$.
We recall that singular values are the square roots of the eigenvalues of $MM^T$, and that, when $M$ is symmetric, singular values coincide with the absolute values of the eigenvalues of $M$. In particular, when $M$ is a covariance matrix, $\lambda_*(M)$ and $\lambda^*(M)$ are the smallest and the largest eigenvalues of $M$.
We consider the following norm on ${\mathbb R}^{n}$:
\be{Not4}
\left\vert y\right\vert _{M}=\sqrt{\left\langle (MM^T)^{-1}y,y\right\rangle}.
\end{equation}
Hereafter, $\alpha =(\alpha _{1},...,\alpha _{k})\in \{1,...,n\}^{k}$ represents a multi-index with length $\left\vert \alpha \right\vert =k$ and $%
\partial _{x}^{\alpha }=\partial _{x_{\alpha _{1}}}...\partial _{x_{\alpha
_{k}}}$. We allow the case $k=0$, giving $\alpha=\emptyset$, and $\partial_x^\alpha f= f$.
Finally, for given vectors $v_1,\ldots,v_n\in {\mathbb R}^m$, we define $\langle v_1,\ldots,v_n\rangle \subset {\mathbb R}^{m}$ the vector space spanned by $v_1,\ldots,v_n$.
Let $X$ denote the process in ${\mathbb R}^n$ already introduced in \eqref{equation-intro}, that is
\be{equation}
dX_{t}=\sum_{j=1}^{d}\sigma
_{j}(t,X_{t})\circ dW_{t}^{j}+b(t,X_{t})dt,\quad \quad X_0=x_0,
\end{equation}
$W$ being a standard Brownian motion in ${\mathbb R}^d$. We suppose the diffusion coefficients fulfill the following requests:
\bas{assumption1}
There exists a constant $\kappa>0$ such that, $\forall t\in [0,1],\,\forall x \in {\mathbb R}^n$:
\begin{align*}
&\sum_{j=1}^d |\sigma_j(t,x)| + |b(t,x)| + \sum_{j=1}^d \sum_{0\leq |\alpha| \leq 2}
|\partial_x^{\alpha}\partial_t\sigma_j(t,x)|
\leq \kappa (1+|x|)\\
&\sum_{j=1}^d \sum_{1\leq |\alpha| \leq 4}
|\partial_x^{\alpha} \sigma_j(t,x)|+\sum_{1\leq |\alpha| \leq 3}
|\partial_x^{\alpha} b(t,x)|
\leq \kappa
\end{align*}
\end{assumption}
Remark that Assumption \ref{assumption1} ensures the strong existence and uniqueness of the solution to \eqref{equation}. We do not assume here ellipticity but a non degeneracy of H\"ormander type. In order to do this, we need to introduce the $n\times d^2$ matrix $A(t,x)$ defined as follows. We set $m=d^2$ and define the function
\be{lip}
l(i,p)=(p-1)d+i\in\{1,\dots,m\},\quad p,i\in\{1,\dots,d\}.
\end{equation}
Notice that $l(i,p)$ is invertible. For $l=1,\ldots,m$, we set the (column) vector field $A_{l}(t,x)$ in ${\mathbb R}^n$ as follows:
\be{Al}
\begin{split}
A_{l}(t,x) &=[\sigma_{i},\sigma_{p}](t,x) \quad \mbox{if}\quad l=l(i,p) \quad \mbox{with}\quad i\neq p, \\
&= \sigma_{i}(t,x)\quad \mbox{if}\quad l=l(i,p) \quad \mbox{with}\quad i= p
\end{split}
\end{equation}
and we set $A(t,x)$ to be the $n\times m$ matrix whose columns are given by $A_1(t,x),\ldots,A_m(t,x)$:
\be{Alucia}
A(t,x)=[A_1(t,x),\ldots,A_m(t,x)].
\end{equation}
$A(t,x)$ can be interpreted as a directional matrix. We denote by $\lambda (t,x)$ the smallest singular value of $A(t,x)$, i.e.
\be{Not9'}
\lambda (t,x)^2=
\lambda_*(A(t,x))^2= \inf_{\left\vert \xi \right\vert =1}\sum_{i=1}^{m}\left\langle
A_{i}(t,x),\xi \right\rangle ^{2}.
\end{equation}
In this paper, we assume the following non degeneracy condition. We write it in a ``time dependent way'' because this is useful in \cite{BCP2}, which represents the second part of the present article. In fact, we use here just $A(0,x_0)$ and $\lambda(0,x_0)$, whereas in \cite{BCP2} we consider $A(t,x_t)$ and $\lambda(t,x_t)$, $x_t$ denoting a skeleton path.
\bas{assumption2}
Let $x_0$ denote the starting point of the diffusion $X$ solving \eqref{equation}. We suppose that
\[
\lambda (0,x_0)>0.
\]
\end{assumption}
Notice that Assumption \ref{assumption2} is actually equivalent to require that the first order H\"ormander condition holds in the starting point $x_0$, i.e. the vector fields $\sigma _{i}(0,x_0)$, $[\sigma_{j},\sigma _{p}](0,x_0)$, as $i,j,p=1,...,d$, span the whole ${\mathbb R}^{n}$.
\smallskip
We define now the $m\times m$ diagonal scaling matrix $\mathcal{D}_\delta$ as
\[
\begin{split}
(\mathcal{D}_\delta)_{l,l}&=\delta \quad \mbox{if}\quad l=l(i,p) \quad \mbox{with}\quad i\neq p, \\
&= \sqrt{\delta} \quad \mbox{if}\quad l=l(i,p) \quad \mbox{with}\quad i= p
\end{split}
\]
and the scaled directional matrix
\be{Adlucia}
A_\delta(t,x)=A(t,x)\mathcal{D}_\delta.
\end{equation}
Notice that the $l$th column of the matrix $A_\delta(t,x)$ is given by $\sqrt{\delta}\sigma_{i}(t,x)$ if $l=l(i,p)$ with $ i= p$ and by $\delta[\sigma_{i},\sigma_{p}](t,x)=[\sqrt{\delta}\sigma_{i},\sqrt{\delta}\sigma_{p}](t,x)$ if $i\neq p$. Therefore, $A_\delta(t,x)$ is the matrix given in \eqref{Alucia} when the original diffusion coefficients $\sigma_j(t,x)$, $j=1,\ldots,d$, are replaced by $\sqrt {\delta}\sigma_j(t,x)$, $j=1,\ldots,d$.
This matrix and the associated norm $|\cdot|_{A_\delta(0,x_0)}$ are the tools that allow us to account of the different speeds of propagation of the diffusion: $\sqrt{\delta}$ (diffusive scaling) in the direction of $\sigma$ and $\delta$ in the direction of the first order Lie Brackets.
In particuar, straightforward computations easily give that
\be{Norm3}
\frac{1}{\sqrt{\delta}\lambda^*(A(t,x))}\left\vert y\right\vert \leq
\left\vert y\right\vert _{A_{\delta}(t,x)}\leq \frac{1}{\delta\lambda_*(A(t,x))}%
\left\vert y\right\vert ,
\end{equation}
We also consider the following assumption, as a stronger version of Assumption \ref{assumption1} (morally we ask for boundedness instead of sublinearity of the coefficients, in the spirit of Kusuoka-Stroock estimates in \cite{KusuokaStroock:87}).
\bas{assumption1bis}
There exists a constant $\kappa>0$ such that for every $t\in [0,1]$ and $x \in {\mathbb R}^n$ one has
\begin{align*}
\sum_{0\leq |\alpha| \leq 4} \Big[
\sum_{j=1}^d |\partial_x^{\alpha} \sigma_j(t,x)|
+ |\partial_x^{\alpha} b(t,x)|
+ |\partial_x^{\alpha}\partial_t\sigma_j(t,x)| \Big]
\leq \kappa.
\end{align*}
\end{assumption}
The aim of this paper is to prove the following result:
\bt{main-th}
Let Assumption \ref{assumption1} and \ref{assumption2} hold. Let $p_{X_t}$ denote the density of $X_t$, $t>0$, with the starting condition $X_0=x_0$. Then the following holds.
\begin{enumerate}
\item
There exist positive constants $r,\delta_*,C$ such that for every $\delta\leq \delta_*$ and for every $y$ such that $|y- x_0-b(0,x_0)\delta|_{A_\delta(0,x_0)}\leq r$,
\[
\frac{1}{C \delta^{n-\frac{\mathrm{dim}\langle \sigma(0,x_0) \rangle}{2} }} \leq
p_{X_\delta}(y).
\]
\item For any $p>1$, there exists a positive constant $C$ such that for every $\delta\leq 1$ and for every $y\in {\mathbb R}^n$
\[
p_{X_\delta}(y)\leq \frac{1}{\delta^{n-\frac{\mathrm{dim}\langle \sigma(0,x_0) \rangle}{2}}
} \frac{C}{1+|y-x_0|_{A_\delta(0,x_0)}^p}.
\]
\item If also Assumption \ref{assumption1bis} holds (boundedness of coefficients) there exists a constant $C$ such that for every $\delta\leq 1$ and for every $y\in {\mathbb R}^n$.
\[
p_{X_\delta}(y)\leq \frac{C}{\delta^{n-\frac{dim \langle \sigma(0,x_0) \rangle}{2}}} \exp\big(- \frac{1}{C}|y-x_0|_{A_\delta(0,x_0)}\big).
\]
\end{enumerate}
Here $\mathrm{dim}\langle \sigma(0,x_0) \rangle$ denotes the dimension of the vector space spanned by $\sigma_1(0,x_0),\ldots,$ $\sigma_d(0,x_0)$.
\end{theorem}
\br{1}
It might appear contradictory that the lower estimate (1) in Theorem \ref{main-th} is centered in $x_0+\delta b(x_0)$,
whereas the upper estimates are centered in $x_0$. In fact, this is important only for the lower bound, the upper bounds (2) and (3) holding true either if we write
$|y-x_0-\delta b(x_0)|_{A_\delta(0,x_0)}$ or $|y-x_0|_{A_\delta(0,x_0)}$ (see next Remark \ref{3}).
\end{remark}
\begin{remark}
As already mentioned, the two sided bound \eqref{ksintro} by Kusuoka and Stroock \cite{KusuokaStroock:87} is proved under a strong H\"ormander condition of any order, but the drift coefficient must be generated by the vector fields of the diffusive part, and the diffusion coefficients $b$ and $\sigma_j$, $j=1,\ldots,d$, must not depend on time. Here, on the contrary, we allow for a general drift and time dependence in the coefficients, but we consider only first order Lie Brackets. Moreover, in assumption \ref{assumption1}, we also relax the hypothesis of bounded coefficients. Anyways, the two estimates are strictly related, since our matrix norm is locally equivalent to the Carath\'eodory distance -- this is proved \cite{BCP2} (see Section 4 therein).
\end{remark}
\begin{remark}
Our main application is developed in \cite{BCP2}, which is the second part of this paper and concerns tube estimates. To this aim, we are mostly interested in the diagonal estimates, that is, around $x_0+\delta b(0,x_0)$. In particular, what we need is the precise exponent
$n-\mathrm{dim} \langle \sigma(0,x_0) \rangle/2$, which accounts for the time-scale of the heat kernel when $\delta$ goes to zero. However, our results are not asymptotic, but hold uniformly for $\delta$ small enough. This is crucial for our application to tube estimates, and this is also a main difference with the estimates in \cite{BenArousLeandre:91,BenArousLeandreII:91}.
\end{remark}
\begin{remark}
The upper bounds in (2) and (3) of Theorem \ref{main-th} give also the tail estimates, which are exponential if we assume the boundedness of the coefficients, polynomial otherwise.
\end{remark}
The proof of Theorem \ref{main-th} is long, also different according to the lower or upper estimate, and we proceed by organizing two sections where such results will be separately proved. So, the lower estimate will be discussed in Section \ref{lower} and proved in Theorem \ref{lower-th}, whereas Section \ref{upper} and Theorem \ref{upper-th} will be devoted to the upper estimate.
\section{Lower bound}\label{lower}
We study here the lower bound for the density of $X_\delta$.
\subsection{The key-decomposition}\label{lower-key}
We start with the decomposition of the process that will allow us to produce the lower bound in short (but not asymptotic) time.
We first use a development in stochastic Taylor series of order two of the diffusion process $X$ defined through \eqref{equation}. This gives
\begin{equation}
X_{t}=x_0+Z_{t}+b(0,x_0)t+R_{t} \label{D1bis}
\end{equation}%
where
\begin{equation}\label{Decomp0}
\begin{array}{l}
\displaystyle
Z_t=\sum_{i=1}^{d}a_{i}W_{t}^{i}+\sum_{i,j=1}^{d}a_{i,j}%
\int_{0}^{t}W_{s}^{i}\circ dW_{s}^{j} \smallskip\\
\mbox{with $a_{i}=\sigma _{i}(0,x_{0})$, $a_{i,j}=\partial _{\sigma _{i}}\sigma _{j}(0,x_{0})$}
\end{array}
\end{equation}%
and
\begin{align}
R_{t}=& \sum_{j,i=1}^{d}\int_{0}^{t}\int_{0}^{s}(\partial _{\sigma
_{i}}\sigma _{j}(u,X_{u})-\partial _{\sigma _{i}}\sigma _{j}(0,x_0))\circ
dW_{u}^{i}\circ dW_{s}^{j} \label{D2bis} \\
& +\sum_{i=1}^{d}\int_{0}^{t}\int_{0}^{s}\partial _{b}\sigma
_{i}(u,X_{u})du\circ
dW_{s}^{i}+\sum_{i=1}^{d}\int_{0}^{t}\int_{0}^{s}\partial _{u}\sigma
_{j}(u,X_{u})du\circ dW_{s}^{i} \notag \\
& +\sum_{i=1}^{d}\int_{0}^{t}\int_{0}^{s}\partial _{\sigma
_{i}}b(u,X_{u})\circ dW_{u}^{i}ds+\int_{0}^{t}\int_{0}^{s}\partial
_{b}b(u,X_{u})duds. \notag
\end{align}%
Since $R_t=\mathcal{O}(t^{3/2})$, we expect the behavior of $X_t$ and $Z_t$ to be somehow close for small values of $t$. Our first goal is to give a decomposition for $Z_t$ in \eqref{Decomp0}. We start introducing some notation. We fix $\delta >0$ and set
$$
s_{k}(\delta )=\frac{k}{d}\delta,\quad k=1,\ldots,d.
$$
We now consider the following random variables: for $i,k=1,\ldots,d$,
\be{delta1}
\Delta _{k}^{i}(\delta ,W)=W_{s_{k}(\delta )}^{i}-W_{s_{k-1}(\delta
)}^{i},\quad \Delta _{k}^{i,j}(\delta ,W)=\int_{s_{k-1}(\delta
)}^{s_{k}(\delta )}(W_{s}^{i}-W_{s_{k-1}}^{i})\circ dW_{s}^{j}.
\end{equation}%
Notice that $\Delta _{k}^{i,j}(\delta ,W)$ is the Stratonovich integral, but
for $i\neq j$ it coincides with the It\^o integral. When no confusion is
possible we use the short notation $s_{k}=s_{k}(\delta ),\Delta
_{k}^{i}=\Delta _{k}^{i}(\delta ,W),\Delta _{k}^{i,j}=\Delta
_{k}^{i,j}(\delta ,W).$ We also denote the random vector $\Delta(\delta,W)$ in ${\mathbb R}^m$
\be{deltal}
\begin{split}
\Delta_{l}(\delta,W) &=\Delta_p^{i,p}(\delta,W) \quad \mbox{if $l=l(i,p)$ with $i\neq p$}, \\
&= \Delta_p^p(\delta,W) \quad \mbox{if $l=l(i,p)$ with $i=p$}.
\end{split}
\end{equation}
(recall $l(i,p)$ in \eqref{lip}). Moreover, with $\sum_{l>p}^{d}=\sum_{p=1}^{d}\sum_{l=p+1}^{d}$, we define
\begin{equation} \label{Decomp2}
\begin{array}{l}
V(\delta,W) =
\displaystyle\sum_{p=1}^d\Big[
\sum_{i\neq p}\Delta^{i}_{p} +
\sum_{i\neq j,i\neq p,j\neq
p}a_{i,j}\Delta _{p}^{i,j}+\sum_{l=p+1}^{d}\sum_{i\neq p}\sum_{j\neq
l} a_{i,j}\Delta _{l}^{j}\Delta _{p}^{i}+\frac{1}{2}\sum_{i\neq
p} a_{i,i}\left\vert \Delta _{p}^{i}\right\vert ^{2}\Big]; \\
\varepsilon _{p}(\delta ,W) = \displaystyle \sum_{l>p}^{d} \sum_{j\neq
l} a_{p,j}\Delta _{l}^{j}+\sum_{p>l}^{d}\sum_{j\neq l} a_{j,p}\Delta
_{l}^{j}+\sum_{j\neq p} a_{p,j}\Delta _{p}^{j},\quad p=1,...,d; \\
\eta _{p}(\delta ,W) = \displaystyle\frac{1}{2}a_{p,p}\left\vert \Delta
_{p}^{p}\right\vert ^{2}+ \sum_{l>p}^{d} a_{p,l}\Delta _{l}^{l}\Delta
_{p}^{p}+\Delta _{p}^{p}\varepsilon _{p}(\delta,W),\quad p=1,...,d.%
\end{array}%
\end{equation}
We have the following decomposition:
\bl{decZ-l}
Let $\Delta(\delta,W)$ and $A(0,x_0)$ be given in \eqref{deltal} and \eqref{Alucia} respectively. One has
\be{decZ}
Z_\delta=
V(\delta,W)+A(0,x_0) \Delta(\delta,W)+\eta(\delta ,W),
\end{equation}
where $V(\delta,W)$ is given in \eqref{Decomp2} and $\eta (\delta ,W)=\sum_{p=1}^{d}\eta _{p}(\delta ,W)$, $\eta_p(\delta ,W)$ being given in \eqref{Decomp2}.
\end{lemma}
The proof of Lemma \ref{decZ-l} is quite long, so it is postponed to Appendix \ref{app-dec}.
\br{111}
The reason of this decomposition is the following. We split the
time interval $(0,\delta )$ in $d$ sub intervals of length $\delta /d.$ We also split the Brownian motion in corresponding increments: $%
(W_{s}^{p}-W_{s_{k-1}}^{p})_{s_{k-1}\leq s\leq s_{k}},p=1,...,d.$ Let us fix
$p.$ For $s\in (s_{p-1},s_{p})$ we have the processes $%
(W_{s}^{i}-W_{s_{p-1}}^{i})_{s_{p-1}\leq s\leq s_{p}},i=1,...,d.$ Our idea
is to settle a calculus which is based on $W^{p}$ and to take conditional
expectation with respect to $W^{i},i\neq p.$ So $%
(W_{s}^{i}-W_{s_{p-1}}^{i})_{s_{p-1}\leq s\leq s_{p}},i\neq p$ will appear
as parameters (or controls) which we may choose in an appropriate way. The random variables on which the calculus is based are $\Delta
_{p}^{p}=W_{s_{p}}^{p}-W_{s_{p-1}}^{p}$ and $\Delta
_{p}^{i,p}=\int_{s_{p-1}}^{s_{p}}(W_{s}^{i}-W_{s_{p-1}}^{i})dW_{s}^{p},j\neq
p.$ These are the r.v. that we have emphasized in the
decomposition of $Z_\delta.$ Notice that, conditionally to the controls $%
(W_{s}^{i}-W_{s_{p-1}}^{i})_{s_{p-1}\leq s\leq s_{p}},i\neq p,$ this is a
centered Gaussian vector and, under appropriate hypothesis on the controls
this Gaussian vector is non degenerate (we treat in section \ref{app-supp} the
problem of the choice of the controls).
In order to handle the term $\Delta_{p}^{p,i}=\int_{s_{p-1}}^{s_{p}}(W_{s}^{p}-W_{s_{p-1}}^{p})dW_{s}^{i}.$
we use the identity $\Delta _{p}^{p,i}=\Delta _{p}^{i}\Delta _{p}^{p}-\Delta
_{p}^{i,p}$.
\end{remark}
We now emphasize the scaling in $\delta$ in the random vector $\Delta(\delta,W)$.
We define
$B_{t}=\delta ^{-1/2}W_{t\delta }$
and denote
\be{New3}
\begin{split}
\Theta_l &=\frac{1}{\delta }\Delta
_{p}^{i,p}=\int_{\frac{p-1}{d}}^{\frac{p}{d}}(B_{s}^{i}-B_{\frac{p-1}{d}}^{i})dB_{s}^{p}\quad \mbox{if $l=l(i,p)$ with $i\neq p$}, \\
&= \frac{1}{\sqrt{\delta }}\Delta_{p}^{p}=B_{\frac{p}{d}}^{p}-B_{\frac{p-1}{d}}^{p} \quad \mbox{if $l=l(i,p)$ with $i=p$},
\end{split}
\end{equation}
$l(i,p)$ being given in \eqref{lip}. For $p=1,...,d$ we denote with $\Theta_{(p)}$ the $p$th block of $\Theta$ with length $d$, that is
$$
\Theta _{(p)}=(\Theta _{(p-1)d+1},...,\Theta
_{pd}),
$$
so that $\Theta=(\Theta _{(1)},\ldots,\Theta _{(d)})$.
We will also denote
\be{New3bis}
l(p)=l(p,p)=(p-1)d+p\quad\mbox{and}\quad\Theta_{l(p)}=\frac{1}{\sqrt{\delta }
}\Delta_{p}^{p}.
\end{equation}
Consider now the $\sigma$ field
\begin{equation}
\mathcal{G}:=\sigma (W_{s}^{j}-W_{s_{p-1}(\delta )}^{j},s_{p-1}(\delta )\leq
s\leq s_{p}(\delta ),p=1,...d,j\neq p). \label{New4}
\end{equation}%
Then conditionally to $\mathcal{G}$ the random variables $%
\Theta _{(p)},p=1,...,d$ are independent centered Gaussian $d$
dimensional vectors and the covariance matrix $Q_{p}$ of $\Theta _{(p)}$
is given by
\begin{equation}\label{Qp}
\begin{array}{rl}
Q_{p}^{p,j}&
\displaystyle
=Q_{p}^{j,p}=\int_{\frac{p-1}{d}}^{\frac{p}{d}}\left(B_{s}^{j}-B_{\frac{p-1}{d}}^{j}\right)ds,\quad j\neq p,\smallskip\\
Q_{p}^{i,j}&
\displaystyle
=\int_{\frac{p-1}{d}}^{\frac{p}{d}}\left(B_{s}^{j}-B_{\frac{p-1}{d}}^{j}\right)\left(B_{s}^{i}-B_{\frac{p-1}{d}}^{i}\right)ds,\quad j\neq
p,i\neq p, \smallskip\\
Q_{p}^{p,p}& =\frac{1}{d}.
\end{array}
\end{equation}
It is easy to see that $\det Q_{p}\neq 0$ almost surely.
It follows that conditionally to $\mathcal{G}$ the random variable $\Theta
=(\Theta _{(1)},...,\Theta _{(d)})$ is a centered $m=d^{2}$ dimensional
Gaussian vector. Its covariance matrix $Q$ is a block-diagonal matrix built with $Q_{p},\,p=1,\dots,d$:
\begin{equation}\label{Q}
Q=
\left(
\begin{array}{ccc}
Q_1 & & \\
&\ddots & \\
& & Q_d
\end{array}
\right)
\end{equation}
In particular $\det Q=\prod_{p=1}^{d}\det Q_{p}\neq 0$
almost surely, and
$\lambda _{\ast }(Q)=\min_{p=1,...,d}\lambda _{\ast }(Q_p)$. We also have $\lambda ^{\ast }(Q)=\max_{p=1,...,d}\lambda^{\ast }(Q_p)$. We will need to work on subsets where we have a quantitative control of this quantities, so we will come back soon on these covariance matrices. But let us show now how one can rewrite decomposition \eqref{decZ} in terms of the random vector $\Theta$. As a consequence, the scaled matrix $A_\delta=A_\delta(0,x_0)$ in \eqref{Adlucia} will appear.
We denote by $A_{\delta }^{i}\in {\mathbb R}^{m},i=1,...,n$ the rows of the matrix $%
A_{\delta }$. We also denote $S=\langle A_{\delta }^{1},...,A_{\delta
}^{n}\rangle \subset {\mathbb R}^{m}$ and $S^{\bot }$ its orthogonal. Under Assumption \ref{assumption2} the columns of $A_{\delta }$ span ${\mathbb R}^{n}$
so the rows $A_{\delta }^{1},...,A_{\delta
}^{n}$ are linearly independent and $S^{\bot }$ has
dimension $m-n$. We take $\Gamma _{\delta }^{i},i=n+1,...,m$ to be an
orthonormal basis in $S^{\bot}$ and we denote $\Gamma _{\delta
}^{i}=A_{\delta }^{i}(0,x_0)$ for $i=1,...,n$.
We also denote $\underline{\Gamma}_{\delta}$ the $(m-n) \times m $ matrix with rows $\Gamma_\delta^i,i=n+1, \dots, m$.
Finally we denote by $\Gamma
_{\delta }$ the $m\times m$ dimensional matrix with rows $\Gamma _{\delta
}^{i},i=1,...,m$.
Notice that
\begin{equation}
\Gamma _{\delta }\Gamma _{\delta }^T=\left(
\begin{tabular}{ll}
$A_{\delta }A_{\delta }^T(0,x_0)$ & $0$ \\
$0$ & $\mathrm{Id}_{m-n}$%
\end{tabular}%
\right) \label{New7}
\end{equation}%
where $\mathrm{Id}_{m-n}$ is the identity matrix in ${\mathbb R}^{m-n}.$ It follows that for a
point $y=(y_{(1)},y_{(2)})\in {\mathbb R}^{m}$ with $y_{(1)}\in {\mathbb R}^{n},y_{(2)}\in
{\mathbb R}^{m-n}$ we have%
\be{GammaA}
\left\vert y\right\vert _{\Gamma _{\delta }}^{2}=\left\vert
y_{(1)}\right\vert _{A_{\delta }(0,x_0)}^{2}+\left\vert y_{(2)}\right\vert ^{2}
\end{equation}
where we recall that $\left\vert y\right\vert _{\Gamma _{\delta
}}^{2}=\left\langle (\Gamma _{\delta }\Gamma _{\delta }^T)^{-1}y,y\right\rangle$.
For $a\in {\mathbb R}^{m}$ we define the immersion
\begin{equation}
J_{a}:{\mathbb R}^{n}\rightarrow {\mathbb R}^{m},\quad
(J_{a}(z))_{i}=z_{i},i=1,...,n\quad\mbox{and}\quad (J_{a}(z))_{i}=\left\langle \Gamma _{\delta
}^{i},a\right\rangle ,i=n+1,...,m. \label{New9}
\end{equation}%
In particular $J_{0}(z)=(z,0,...,0)$ and
\begin{equation}
\left\vert J_{0}z\right\vert _{\Gamma _{\delta }}=\left\vert z\right\vert
_{A_{\delta }(0,x_0)}. \label{New9a}
\end{equation}%
Finally we denote%
\begin{equation}\label{New10}
\begin{array}{rcl}
V_{\omega } &=&V(\delta,W) \smallskip\\
\eta _{\omega }(\Theta ) &=&
\displaystyle
\sum_{p=1}^{d}\left(\frac{a_{p,p}}{2}\delta \Theta
_{l(p)}^{2}+\delta ^{1/2}\Theta _{l(p)}\varepsilon
_{p}(\delta,W)+\sum_{q>p}^{d}a_{p,q}\delta \Theta _{l(q)}\Theta _{l(p)}\right)
\end{array}
\end{equation}%
where $V(\delta,W)$ and $\varepsilon_{p}(\delta ,W)$ are
defined in (\ref{Decomp2}) and $\Theta _{l(p)}$ is given in \eqref{New3bis}. We notice that $\eta _{\omega }(\Theta )=\sum_{p=1}^{d}\eta _{p}(\delta
,W)$, $\eta_{p}(\delta ,W)$ being defined in (\ref{Decomp2}).
We also remark that both $V(\delta,W)$ and $\varepsilon_{p}(\delta ,W)$ are $\mathcal{G}$-measurable, so
\eqref{New10} stresses a dependence on $\omega$ which is $\mathcal{G}$-measurable and a dependence on the random vector $\Theta$ whose conditional law w.r.t. $\mathcal{G}$ is Gaussian.
Now the decomposition \eqref{decZ} may be written as
\[
Z_\delta=V_{\omega }+A_{\delta }(0,x_0)\Theta +\eta _{\omega }(\Theta ).
\]
We embed this relation in ${\mathbb R}^{m}$ and obtain%
\[
J_{\Theta }(Z_\delta)=J_{0}(V_{\omega })+\Gamma _{\delta }\Theta
+J_{0}(\eta _{\omega }(\Theta )).
\]
We now multiply with $\Gamma _{\delta }^{-1}$: setting
\be{deftilde}
\widetilde{Z}=\Gamma _{\delta }^{-1}J_{\Theta }(Z_\delta),\quad \widetilde{V}_{\omega }=\Gamma _{\delta }^{-1}J_{0}(V_{\omega }),\quad
\widetilde{\eta }_{\omega }(\Theta )=\Gamma _{\delta
}^{-1}J_{0}(\eta _{\omega }(\Theta ))
\end{equation}
and
\be{defG}
G=\Theta +\widetilde{\eta }_{\omega}(\Theta),
\end{equation}
we get
\be{New11}
\widetilde{Z}=\widetilde{V}_{\omega }+G.
\end{equation}
Notice that, conditionally to $\mathcal{G}$, $\widetilde Z$ is a translation of the random variable $G=\Theta+\widetilde{\eta }_{\omega}(\Theta)$ which is a perturbation of a centred Gaussian random variable. Thanks to this fact, we can to apply the results in Appendix \ref{sectioninversefunction}: we use a local inversion argument in order to give bounds for the conditional density of $\widetilde Z$, which will be used in order to get bounds for the non conditional density. As a consequence, we will get density estimates for $Z_\delta$.
\subsection{Localized density for the principal term $\widetilde Z$}
We study here the density of $\widetilde Z$ in \eqref{New11}, ``around'' (that is, localized on) a suitable set of Brownian trajectories, where we have a quantitative control on the ``non-degeneracy'' (conditionally to $\mathcal{G}$) of the main Gaussian $\Theta$.
We denote
\begin{equation}\label{qp}
q_{p}(B)=\sum_{j\neq p}\left\vert B_{\frac{p}{d}}^{j}-B_{\frac{p-1}{d}}^{j}\right\vert
+\sum_{j\neq p,i\neq p,i\neq j}\left\vert
\int_{\frac{p-1}{d}}^{\frac{p}{d}}(B_{s}^{j}-B_{\frac{i-1}{d}}^{j})dB_{s}^{i}\right\vert.
\end{equation}
For fixed $\varepsilon,\rho>0$, we define
\begin{equation}\label{def-Lambda}
\begin{array}{l}
\Lambda _{\rho,\varepsilon ,p}=\Big\{\det Q_{p}\geq \varepsilon^\rho,\sup_{\frac{p-1}{d}\leq t\leq \frac{p}{d}}\sum_{j\neq p}\vert
B_{t}^{j}-B_{\frac{p-1}{d}}^{j}\vert \leq \varepsilon^{-\rho},q_{p}(B)\leq
\varepsilon \Big\},\ p=1,\ldots,d\smallskip\\
\Lambda _{\rho,\varepsilon}=\cap_{p=1}^d\Lambda _{\rho,\varepsilon ,p}.
\end{array}
\end{equation}%
Notice that $\Lambda _{\rho,\varepsilon ,p}\in\mathcal{G}$ for every $p=1,\ldots,d$.
By using some results in Appendix \ref{app-supp}, we get the following.
\begin{lemma}\label{lemma-Lambda}
Let $\Lambda _{\rho,\varepsilon }$ be as in \eqref{def-Lambda}.
There exist $c$ and $\varepsilon _{\ast }$ such that for every $\varepsilon\leq \varepsilon_\ast$ one has
\begin{equation}
{\mathbb P}\big(\Lambda _{\rho,\varepsilon }\big)\geq c\times \varepsilon ^{\frac{1}{2%
}m(d+1)}. \label{N9}
\end{equation}%
\end{lemma}
\begin{proof}
We apply here Proposition \ref{SUPORT3}. Let $p\in\{1,\ldots d\}$ be fixed and consider the Brownian motion $\widehat B_t=\sqrt d(B_{\frac{p-1+t}d}-B_{\frac{p-1}d})$. Let $Q(\widehat B)$ be the matrix in \eqref{Q-app}. Up to a permutation of the components of $\widehat B$, we easily get $Q^{p,p}(\widehat B)=d\times Q_p^{p,p}$, $Q^{p,j}(\widehat B)=d^{3/2}\times Q_p^{p,j}$ for $j\neq p$, $Q^{i,j}(\widehat B)=d^2\times Q_p^{i,j}$ for $i\neq p$ and $j\neq p$. Therefore,
$$
\det Q_p=d^{2d-1}\det Q(\widehat B)\geq \det Q(\widehat B).
$$
Let now $q(\widehat B)$ be the quantity defined in \eqref{Sup4}. With $q_p(B)$ as in \eqref{qp}, it easily follows that
$$
q_p(B)\leq q(\widehat B).
$$
Moreover, $\sup_{t\leq 1}|\widehat B_t|=\sqrt d\, \sup_{\frac{p-1}d\leq s\leq \frac pd}|B_s-B_{\frac {p-1}d}|\geq \sup_{\frac{p-1}d\leq s\leq \frac pd}|B_s-B_{\frac {p-1}d}|$.
As a consequence, with $\Upsilon_{\rho ,\varepsilon }$ the set defined in \eqref{Sup3}, we get
$$
\Upsilon_{\rho ,\varepsilon }(\widehat B)
\subset \Lambda_{\rho, \varepsilon, p}
$$
and by using \eqref{Sup2}, we may find some constants $c$ and $%
\varepsilon _{\ast }$ such that
${\mathbb P}(\Lambda _{\rho,\varepsilon ,p})\geq c\varepsilon ^{\frac{1}{2}%
d(d+1)}$, for $\varepsilon \leq \varepsilon _{\ast }$.
This holds for every $p$. Since $\Lambda _{\rho,\varepsilon }=\cap _{p=1}^{d}\Lambda _{\rho,\varepsilon ,p}$, by using the independence property we get \eqref{N9}.
\end{proof}
Let $Q$ be the matrix in \eqref{Q}. On the set $\Lambda _{\rho,\varepsilon }\in \mathcal{G}$ we have $\det Q=\prod_{p=1}^{d}\det Q_{p}\geq \varepsilon ^{d\rho}$. Remark that
\be{complambdanorm}
\frac{\lambda^{\ast }(Q)}{\sqrt{m}}
\leq |Q|_l := \left(\frac{1}{m}\sum_{1\leq i,j\leq m} Q_{i,j}^2 \right)^{1/2}
\leq \lambda^{\ast }(Q).
\end{equation}
For $a>0$ we introduce the following function,
\[
\psi_a(x)=1_{|x|\leq a}+\exp\left( 1-\frac{a^2}{a^2-(x-a)^2} \right)1_{a<|x|<2a},
\]
which is a mollified version of $1_{[0,a]}$.
We can now define our \emph{localization variables}.
\begin{equation}\label{Utilde}
\widetilde{U}_\varepsilon=(\psi_{a_1}(1/\det Q))
\psi_{a_2}(|Q|_l)
\psi_{a_3}(q(B)),\quad\mbox{with}\quad a_1=\varepsilon^{-d\rho},\ a_2=\varepsilon^{-2\rho}, \ a_3=d\varepsilon
\end{equation}
in which we have set
$$
q(B)=\sum_{p=1}^dq_p(B).
$$
Remark that
$\widetilde{U}_\varepsilon$ is measurable w.r.t. $\mathcal{G}$.
The following inclusions hold: for every $\varepsilon$ small enough,
\be{N10}
\Lambda _{\rho,\varepsilon }
\subset
\Big\{\det Q \geq \varepsilon^{d\rho},|Q|_l\leq \varepsilon^{-2\rho},q(B)\leq
d \varepsilon \Big\}= \{\widetilde{U}_\varepsilon= 1\}\subset \{\widetilde{U}_\varepsilon\neq 0\}.
\end{equation}
We can consider $\widetilde{U}_\varepsilon$ as a smooth version of the indicator function of $\Lambda_{\rho,\varepsilon}$.
We also define, for fixed $r>0$,
\begin{equation}\label{Ubar}
\bar{U}_r=\prod_{i=1}^n \psi_r(\Theta_i).
\end{equation}
In order to state a lower estimate for the (localized) density of $\widetilde Z$ in \eqref{New11}, we define the following set of constants:
\be{CC}
\mathcal{C}=\Big\{C>0\,:\,C=\exp\Big({c \Big(\frac{\kappa}{\lambda(0,x_0)}\Big)^{q}}\Big),\ \exists\ c,q>0\Big\}
\end{equation}
and we set
$$
1/\mathcal{C}=\{C>0\,:\,1/C\in\mathcal{C}\}.
$$
\bl{lemmaZ}
Suppose Assumption \ref{assumption1} and \ref{assumption2} both hold. Let $U_{\varepsilon,r}=\widetilde U_\varepsilon\bar U_r$, $\widetilde U_\varepsilon$ and $\bar U_r$ being defined in \eqref{Utilde} and \eqref{Ubar} respectively, with $\rho=\frac 1{8m}$. Set $d{\mathbb P}_{U_{\varepsilon,r}}=U_{\varepsilon,r}d{\mathbb P}$ and let $p_{\widetilde{Z},U_{\varepsilon,r}}$ denote the density of $\widetilde Z$ in \eqref{New11} when we endow $\Omega$ with the measure ${\mathbb P}_{U_{\varepsilon,r}}$.
Then there exist $C\in \mathcal{C}$, $\varepsilon,r\in 1/\mathcal{C}$ such that for $|z|\leq r/2$,
\be{estZ}
p_{\widetilde{Z},U_{\varepsilon,r}}(z)\geq \frac{1}{C}.
\end{equation}
\end{lemma}
\begin{proof}
STEP 1: lower bound for the localized conditional density given $\mathcal{G}$.
\smallskip
Let $p_{\widetilde{Z},\bar{U}_r|\mathcal{G}}$ denote the localized density w.r.t. the localization $\bar U_r$ of $\widetilde{Z}$ conditioned to $\mathcal{G}$, i.e.
\be{defcond}
E_N^{\eta_N}[f(\widetilde{Z})\bar{U}_r|\mathcal{G}] = \int f(z) p_{\widetilde{Z},\bar{U}_r|\mathcal{G}}(z)dz,
\end{equation}
for $f$ positive, measurable, with support included in $B(0,r/2)$.
We start proving that there exist $C\in \mathcal{C}$, $\varepsilon,r \in 1/\mathcal{C}$ such that, on $\widetilde{U}_\varepsilon\neq 0$, for $|z|\leq r/2$
\[
p_{\widetilde{Z},\bar{U}_r|\mathcal{G}}(z)\geq \frac{1}{C}.
\]
We recall \eqref{New11}: $\widetilde Z=\widetilde V_\omega+\Theta+\widetilde\eta_\omega(\Theta)$, where $\omega\mapsto\widetilde V_\omega$ and $\omega\mapsto\widetilde\eta_\omega(\cdot)$ are both $\mathcal{G}$-measurable and the conditional law of $\Theta$ given $\mathcal{G}$ is Gaussian. This allows us to use the results in Appendix \ref{sectioninversefunction}. In particular, we are interested in working on the set $\{\widetilde{U}_\varepsilon\neq 0\}\in\mathcal{G}$, so one has to keep in mind that $\omega\in\{\widetilde{U}_\varepsilon\neq 0\}$.
\smallskip
On $\widetilde{U}_\varepsilon\neq 0$, by \eqref{Utilde} and \eqref{complambdanorm} one has $\lambda^*(Q) \leq 2 \sqrt{m} \varepsilon^{-2\rho}$, and
\begin{equation*}
\frac{\varepsilon ^{d\rho } }{2} \leq \det Q\leq \lambda _{\ast }(Q)\lambda ^{\ast}(Q)^{m-1} \leq \lambda _{\ast }(Q) (2 \sqrt{m})^{m-1}\varepsilon ^{-2\rho (m-1)},
\end{equation*}%
and this gives $\lambda _{\ast }(Q)\geq \frac{ \varepsilon ^{3 m \rho } }{(2\sqrt{m})^m} $. So, fixing $\rho=1/(8m)$, for $\varepsilon \leq \varepsilon^*$,
\be{rlh}
\frac{1}{16 m^2}\frac{\lambda_*(Q)}{\lambda^*(Q)} \geq
C_m \varepsilon ^{3 m \rho+2\rho } \geq
\varepsilon.
\end{equation}
To apply \eqref{invest} to $G=\Theta +\widetilde{\eta }_{\omega}(\Theta)$ we need to check the hypothesis of Lemma \ref{invfun2}. We are going to use the notation of Appendix \ref{sectioninversefunction}, in particular for $c_*(\widetilde{\eta}_\omega,r)$ in \eqref{c*} and $c_i(\widetilde{\eta}_\omega)$, $i=2,3$, in \eqref{ci}. Recall that $\widetilde\eta_\omega$ is defined in \eqref{deftilde} through $\eta_\omega$ given in \eqref{New10}. Since the third order derivatives of $\eta_\omega$ are null, we have $c_{3}(\widetilde{\eta }_{\omega})=0$. Also, for $i=l(p)$ and $j=l(q)$ we have $\partial _{i,j}\eta _{\omega }(\Theta )=\delta a_{ij}$, otherwise we get $\partial _{i,j}\eta _{\omega }(\Theta )=0$. So $\left|\partial _{i,j}\eta _{\omega}(\Theta)\right| \leq \delta \sum_{i,j} |a_{i,j}|$. Using \eqref{Norm3} we obtain
\begin{equation*}
\left| \partial _{i,j}\widetilde{\eta }_{\omega}(\Theta)\right|
=\left| J_{0}(\partial _{i,j}\eta _{\omega
}(\Theta))\right|_{\Gamma_{\delta}}
=\left| \partial_{i,j}\eta_{\omega}(\Theta)\right|_{A_{\delta}}
\leq \frac{\sum_{i,j} |a_{i,j}|}{\lambda(0,x_0)}\leq C\in \mathcal{C}.
\end{equation*}%
So, with $h_{\eta_\omega}$ as in \eqref{defheta}, we get
\be{c2}
h_{\eta_\omega}=\frac{1}{16 m^2 (c_2(\widetilde{\eta }_{\omega})+\sqrt{c_3(\widetilde{\eta }_{\omega})})}\geq \frac{1}{C_1}, \quad \exists C_1\in \mathcal{C}
\end{equation}
We compute now the first order derivatives. For $j\notin \{l(p):p=1,...,d\}$
we have $\partial _{j}\eta _{\omega }=0$ and for $j=l(p)$ we have
\[
\partial_j\eta _{\omega }(\Theta )=\delta \sum_{q=p}^{d}a_{p\wedge
q,p\vee q}\Theta _{l(q)}+\sqrt{\delta }\varepsilon_{j}(\delta,W).
\]
So, as above, we obtain $\left| \partial _{j}\widetilde{\eta }_{\omega}(\Theta)\right|
\leq C (|\Theta|+|\varepsilon_j(\delta,W)|/\sqrt{\delta})$. Remark now that on $\{\bar{U}_r\neq 0\}$ we have $|\Theta|\leq C r$, and on $\{\widetilde{U}_\varepsilon\neq 0\}$ we have $q(B)\leq 2d\varepsilon$, so
\be{ept}
\sum_{j=1}^d |\varepsilon_j(\delta,W)|\leq C\sqrt{\delta} q(B) \leq C\sqrt{\delta} \varepsilon.
\end{equation}
Therefore
\be{cstar}
c_*(\widetilde{\eta }_{\omega},16r)\leq C_2 (r +\varepsilon), \quad \exists C_2\in \mathcal{C}.
\end{equation}
We also consider the following estimate of $|\widetilde{V}_{\omega }| =|V_{\omega }|_{A_{\delta }}$. First, we rewrite $V_\omega$ as follows:
$$
\begin{array}{c}
\displaystyle
V_{\omega }
=\sum_{p}a_p\mu_p(\delta,W)+\sum_p\psi_p(\delta, W),\quad\mbox{with}\\
\displaystyle
\mu _{p}(\delta ,W)
= \displaystyle\sum_{i\neq p}\Delta _{i}^{p}
\mbox{ and }
\psi _{p}(\delta ,W) =
\!\!\!\!\!\!\!\sum_{i\neq j,i\neq p,j\neq
p}\!\!\!\!\!\!a_{i,j}\Delta _{p}^{i,j}+\!\!\sum_{l=p+1}^{d}\sum_{i\neq p}\sum_{j\neq
l} a_{i,j}\Delta _{l}^{j}\Delta _{p}^{i}+\frac{1}{2}\sum_{i\neq
p} a_{i,i}\left\vert \Delta _{p}^{i}\right\vert ^{2}
\end{array}
$$
Using again \eqref{Norm3} we have
\[
\Big|\sum_{p=1}^{d}a_{p}\mu _{p}(\delta ,W)\Big| _{A_{\delta }}=%
\frac{1}{\sqrt{\delta }}\Big| A_{\delta }J_{0}\Big(\sum_{p=1}^{d}\mu _{p}(\delta
,W)\Big)\Big|_{A_{\delta }}\leq \sum_{p=1}^{d}\frac{1}{\sqrt{\delta }}
|\mu_{p}(\delta ,W))|\leq C q(B)
\]
and
\[
\left\vert \psi (\delta ,W)\right\vert _{A_{\delta }}\leq \frac{\left\vert
\psi (\delta ,W)\right\vert}{\delta \sqrt{\lambda (0,x_0)}}
\leq C q(B).
\]
Since $\omega\in\{\widetilde{U}_\varepsilon\neq 0\}$ we get
\be{V}
\begin{split}
|\widetilde{V}_{\omega }| \leq C q(B) \leq C_3 \varepsilon,\quad \exists C_3\in \mathcal{C}.
\end{split}
\end{equation}
We consider \eqref{V}, and fix $\frac{r}{\varepsilon}=2 C_3 \in \mathcal{C}$, so $|\widetilde{V}_{\omega}|\leq r/2$. Then we consider \eqref{cstar} and we obtain
\[
c_*(\widetilde{\eta }_{\omega},4r)\leq C_2 (2 C_3 +1)\varepsilon \leq \varepsilon^{1/2}, \quad \mbox{for }\, \varepsilon\leq \frac{1}{(4 C_2 C_3)^2} \in \frac{1}{\mathcal{C}}.
\]
Moreover, looking at \eqref{c2}
\[
r=2 C_3\varepsilon \leq \frac{1}{C_1}\quad \mbox{for }\, \varepsilon \leq \frac{1}{2 C_1 C_3}\in \frac{1}{\mathcal{C}}.
\]
So, with
\[
\varepsilon=\varepsilon^*\wedge \frac{1}{(4 C_2 C_3)^2} \wedge \frac{1}{2 C_1 C_3}\in \frac{1}{\mathcal{C}},
\]
and $r=2 C_3 \varepsilon$ we have
\[
|\widetilde{V}_{\omega}|\leq \frac{r}{2},
\qquad c_*(\widetilde{\eta}_\omega,4 r)\leq \varepsilon^{1/2},
\qquad r\leq \frac{1}{C_1}.
\]
Now, by using also \eqref{rlh} and \eqref{c2}, it follows that \eqref{hpimpl} holds, and we can apply Lemma \ref{invfun2}. We obtain
\[
\begin{split}
\frac{1}{K \det Q^{1/2}
}\exp\left(-\frac{K}{\lambda_*(Q)}|z|^2 \right) \leq
p_{G,\bar{U}_r|\mathcal{G}}(z)
\end{split}
\]
for $|z|\leq r$, where $K$ does not depend on $\sigma,b$. Remark that, using $\lambda_*(Q)\geq \frac{\varepsilon^{3m\rho}}{(2\sqrt{m})^m}$, $\rho=\frac{1}{8m}$, $r/\varepsilon=2 C_1$ and $\varepsilon\leq 1/(4 C_2 C_1)^2$,
\be{noexp}
\frac{|z|^2}{\lambda_*(Q)}\leq \frac{(2\sqrt{m})^m r^2}{\varepsilon^{3m\rho}}\leq
(2\sqrt{m})^m \frac{ r^2}{\varepsilon}
\leq
(2\sqrt{m})^m \frac{ r^2}{\varepsilon^2}\varepsilon
\leq
(2\sqrt{m})^m (2C_1)^2 \varepsilon \leq \bar{K}
\end{equation}
where $\bar{K}$ does not depend on $\sigma$, $b$. Therefore $ p_{G,\bar{U}_r|\mathcal{G}}(z)\geq \frac{1}{C} $, for $|z|\leq r$, for some $C\in\mathcal{C}$, on $\widetilde{U}_\varepsilon\neq 0$. Now, by recalling that $|\widetilde{V}_\omega|\leq r/2$ and \eqref{New11}, we have
\be{Zc}
p_{\widetilde{Z},\bar{U}_r|\mathcal{G}}(z)\geq \frac{1}{C},\quad\mbox{for $|z|\leq r/2$ on the set $\{\widetilde U_\varepsilon\neq 0\}$}.
\end{equation}
STEP 2: we get rid of the conditioning on $\mathcal{G}$, to have non-conditional bound for $p_{\widetilde{Z},U_{\varepsilon,r}}$.
Since $\widetilde{U}_\varepsilon$ is $\mathcal{G}$ measurable, for every non-negative and measurable function $f$ with support included in $B(0,r/2)$ we have
\[
E_N^{\eta_N}(f(\widetilde{Z})U_{\varepsilon,r})=E_N^{\eta_N}\big(\widetilde{U}_\varepsilon
E_N^{\eta_N}(f(\widetilde{Z})\bar{U}_r|\mathcal{G})\big).
\]
By \eqref{defcond} and \eqref{Zc}, we obtain
\[
E_N^{\eta_N}(f(\widetilde{Z})U_{\varepsilon,r})\geq \frac{1}{C}E_N^{\eta_N}(\widetilde{U}_\varepsilon) \int f(z) dz
\]
Since $\Lambda_{\rho, \varepsilon} \subset \{\widetilde{U}_\varepsilon= 1\}$, $E_N^{\eta_N}(\widetilde{U}_\varepsilon)\geq {\mathbb P}(\Lambda_{\rho, \varepsilon})$, so by using \eqref{N9} and $\varepsilon\in 1/\mathcal{C}$ we finally get that $E_N^{\eta_N}(\widetilde{U}_\varepsilon)\geq \frac{1}{C}$, so \eqref{estZ} is proved.
\end{proof}
\subsection{Lower bound for the density of $X_\delta$}
We study here a lower bound for the density of $X_\delta$, $X$ being the solution to \eqref{equation}. Recall decomposition \eqref{D1bis}:
$$
X_\delta-x_0-b(0,x_0)\delta=Z_\delta+R_\delta.
$$
Our aim is to ``transfer'' the lower bound for $\widetilde Z=\Gamma^{-1}_\delta J_\Theta(Z_\delta)$ already studied in Lemma \ref{lemmaZ} to a lower bound for $X_\delta$. In order to set up this program, we use results on the distance between probability densities which have been developed in \cite{BallyCaramellino:12}. In particular, we are going to use now Malliavin calculus. Appendix \ref{app-mall} is devoted to a recall of all the results and notation the present section is based on. In particular, we denote with $D$ the Malliavin derivative with respect to $W$, the Brownian motion driving the original equation \eqref{equation}.
But first of all, we need some properties of the matrix $\Gamma_\delta$, which can be resumed as follows. We set $SO(d)$ the set of the $d\times d$ orthogonal matrices and we denote with $\mathrm{Id}_d$ the $d\times d$ identity matrix.
\begin{lemma}\label{Gamma-prop}
Set for simplicity $A_\delta=A_\delta(0,x_0)$ and let $\Gamma_\delta$ be as in \eqref{New7}. There exist $\mathcal{U}\in SO(n)$, $\underline{\mathcal{U}}\in SO(m-n)$ and $\mathcal{V}\in SO(m)$ such that
\[
\Gamma_\delta = \left(
\begin{array}{cc}
\mathcal{U} & 0 \\
0^T & \underline{\mathcal{U}}
\end{array}
\right) \left(
\begin{array}{cc}
\bar{\Sigma} & 0 \\
0^T & \mathrm{Id}_{m-n}
\end{array}
\right) \mathcal{V}^T
\]
where $0$ denotes a null $n\times (m-n)$ matrix and $\bar{\Sigma}=\bar{\Sigma}=\mathrm{Diag}(\lambda_1(A_\delta),\ldots,\lambda_n(A_\delta))$, $\lambda_i(A_\delta)$, $i=1,\ldots,n$, being the singular values of $A_\delta$ (which are strictly positive because $A_\delta$ has full rank).
\end{lemma}
\begin{proof}
We recall that
\[
\Gamma_\delta=\left( \begin{array}{c}A_\delta\\
\underline{\Gamma}_\delta
\end{array}\right),
\]
where $\underline{\Gamma}_\delta$ is a $(m-n)\times n$ matrix whose rows are vectors of ${\mathbb R}^m$ which are orthonormal and orthogonal with the rows of $A_\delta$. We take a singular value decomposition for $A_\delta$ and for $\underline{\Gamma}_\delta$. So, there exist $\mathcal{U}\in SO(n)$ and $\bar{\mathcal{V}}\in SO(m)$ such that
$$
A_\delta= \mathcal{U} \big(\bar{\Sigma}\ 0\big) \bar{\mathcal{V}}^T,
$$
$0$ denoting the $n\times (m-n)$ null matrix. Similarly, there exist
$\underline{\mathcal{U}}\in SO(m-n)$ and $\underline{\mathcal{V}}\in SO(m)$ such that
$$
\underline{\Gamma}_\delta= \underline{\mathcal{U}} \big(0^T\ \mathrm{Id}_{m-n}\big) \underline{\mathcal{V}}^T,
$$
the diagonal matrix being $\mathrm{Id}_{m-n}$ because the rows of $\underline{\Gamma}_\delta$ are orthonormal. Therefore, we get
\[
\Gamma_\delta = \left(
\begin{array}{cc}
\mathcal{U} & 0 \\
0^T & \underline{\mathcal{U}}
\end{array}
\right) \left(
\begin{array}{cc}
\bar{\Sigma} & 0 \\
0^T & \mathrm{Id}_{m-n}
\end{array}
\right) \mathcal{V}^T
\]
where $\mathcal{V}$ is a $m\times m$ matrix whose first $n$ columns are given by the first $n$ columns of $\overline{\mathcal{V}}$ and the remaining $m-n$ columns are given by the last $m-n$ columns of $\underline{\mathcal{V}}$. Moreover, since each row of $A_\delta$ is orthogonal to any row of $\underline{\Gamma_\delta}$, it immediately follows that all columns of $\mathcal{V}$ are orthogonal. This proves that $\mathcal{V}\in SO(m)$, and the statement follows.
\end{proof}
Then we have
\bl{estF}
Suppose Assumption \ref{assumption1} and \ref{assumption2} both hold. Let $U_{\varepsilon, r}$ denote the localization in Lemma \ref{lemmaZ} and let $\mathcal{U}$ and $\bar \Sigma$ be the matrices in Lemma \ref{Gamma-prop}. Set
$$
\alpha= \mathcal{U}\bar{\Sigma} \quad\mbox{and}\quad \widehat X_\delta=\alpha^{-1} (X_\delta-x_0-b(0,x_0)\delta).
$$
Then there exist $C\in \mathcal{C}$, $\delta_*,r\in 1/\mathcal{C}$ such that for $\delta\leq \delta_*, \, |z|\leq r/2$,
\be{estlocF}
p_{\widehat{X}_\delta,U_{\varepsilon,r}}(z)\geq \frac 1C,
\end{equation}
$p_{\widehat{X}_\delta,U_{\varepsilon,r}}$ denoting the density of $\widehat{X}_\delta$ with respect to the measure ${\mathbb P}_{U_{\varepsilon,r}}$.
\end{lemma}
\begin{proof}
We set $\widehat{Z}_\delta=\alpha^{-1}Z_\delta$ and we use Proposition \ref{generalbounds}, with the localization $U=U_{\varepsilon,r}$, applied to $F=\widehat{X}_\delta$ and $G=\widehat Z_\delta$. Recall that the requests in (1) of Proposition \ref{generalbounds} involve several quantities: the lowest singular value (that in this case coincides with the lowest eigenvalue) $\lambda_*(\gamma_{\widehat{X}_\delta})$ and $\lambda_*(\gamma_{\widehat{Z}_\delta})$ of the Malliavin covariance matrix of $\widehat{X}_\delta$ and $\widehat{Z}_\delta$ respectively, as well as $m_{U_{\varepsilon,r}}(p)$ in \eqref{defmU}, the Sobolev-Malliavin norms $\|\widehat{X}_\delta\|_{2,p,U_{\varepsilon,r}}$, $\|\widehat{Z}_\delta\|_{2,p,U_{\varepsilon,r}}$, and $\|\widehat{X}_\delta-\widehat{Z}_\delta\|_{2,p,U_{\varepsilon,r}}=\| \alpha^{-1} R_\delta\|_{2,p,U_{\varepsilon,r}}$. First of all, by using Assumption \ref{assumption1}, one easily gets
\begin{equation}\label{est1}
\|\alpha^{-1} R_\delta\|_{2,p}\leq C\delta^{-1} \delta^{3/2}=C\sqrt \delta
\quad\mbox{and}\quad
\|\widehat{X}_\delta\|_{2,p}+\|\widehat Z_\delta\|_{2,p}
\leq C,\quad \exists C\in\mathcal{C}.
\end{equation}
We now check that $m_{U_{\varepsilon,r}}(p)<\infty$ for every $p$. Standard computations and \eqref{Sup1} give, for every $p$,
\[
\| 1/\det Q \|_{2,p}+
\| \,|Q|_l \|_{2,p}+
\|q(B)\|_{2,p}+
\|\Theta \|_{2,p}\leq C,
\]
so we can apply \eqref{normlocalization} and conclude
\be{estimateU}
m_{U_{\varepsilon,r}}(p) \leq C \in \mathcal{C}.
\end{equation}
We now study the lower eigenvalue of the Malliavin covariance matrix of $\widehat{Z}_\delta$. From the definition of $\widehat{Z}_\delta$, we have
\be{ZG}
\widetilde{Z}=
\mathcal{V} \left(
\begin{array}{c}
\alpha^{-1} Z_\delta \\
\underline{\mathcal{U}}^T\underline{\Gamma}_\delta\Theta
\end{array}
\right)
=
\mathcal{V} \left(
\begin{array}{c}
\widehat{Z}_\delta \\
\underline{\mathcal{U}}^T\underline{\Gamma}_\delta\Theta
\end{array}
\right),
\end{equation}
(see the proof of Lemma \ref{Gamma-prop} for the definition of $\underline{\Gamma}_\delta$).
As an immediate consequence, one has $ \lambda_*(\gamma_{\widehat{Z}_\delta}) \geq
\lambda_*(\gamma_{\widetilde{Z}})$, and it suffices to study the lower eigenvalue of $\widetilde{Z}$. By using \eqref{New11}, we have
\[
\begin{split}
\langle\gamma_{\widetilde{Z}} \xi, \xi \rangle
& = \sum_{i=1}^d \int_0^\delta \langle D^i_s \widetilde{Z},\xi \rangle^2
= \sum_{i=1}^d \int_{s_{i-1}(\delta)}^{s_i(\delta)} \langle D^i_s \widetilde{Z},\xi \rangle^2
= \sum_{i=1}^d \int_{s_{i-1}(\delta)}^{s_i(\delta)} \langle D^i_s (\Theta+\widetilde \eta(\Theta)),\xi \rangle^2\\
& \geq \sum_{i=1}^d \int_{s_{i-1}(\delta)}^{s_i(\delta)} \Big(\frac{1}{2}\langle D^i_s \Theta,\xi \rangle^2 - \langle D^i_s \eta(\Theta),\xi \rangle^2\Big) ds\\
&=S_1+S_2.
\end{split}
\]
We write
\begin{align*}
&S_1=\sum_{i=1}^d \int_{s_{i-1}(\delta)}^{s_i(\delta)} \frac{1}{2}\langle D^i_s \Theta,\xi \rangle^2\geq \frac{\lambda_*(Q)}{2}\\
&S_2=\sum_{i=1}^d \int_{s_{i-1}(\delta)}^{s_i(\delta)} \langle \nabla\eta(\Theta) D^i_s \Theta,\xi \rangle^2 ds
= \sum_{i=1}^d \int_{s_{i-1}(\delta)}^{s_i(\delta)} \langle D^i_s \Theta, \nabla\eta(\Theta)^T \xi \rangle^2 ds
\leq \lambda^*(Q) |\nabla\eta(\Theta)|^2 |\xi|^2,
\end{align*}
so that
\[
\begin{split}
\lambda_*(\gamma_{\widehat{Z}_\delta}) \geq
\lambda_*(\gamma_{\widetilde{Z}}) \geq \lambda_*(Q) \left(\frac{1}{2} - \frac{\lambda^*(Q)}{\lambda_*(Q)} |\nabla\eta(\Theta)|^2\right).
\end{split}
\]
On $\{\widetilde{U}_\varepsilon\neq 0\}$, we have already proved in Lemma \ref{lemmaZ} that $c_*(\eta,\Theta)\leq \frac{\sqrt{\lambda_*(Q)/\lambda^*(Q)}}{2m}$. Since $|\nabla\eta(\Theta)| \leq m c_*(\eta,\Theta)$, we obtain
\[
|\nabla\eta(\Theta)| \leq \frac{1}{2}\sqrt{\frac{\lambda_*(Q)}{\lambda^*(Q)}},
\]
and therefore $\lambda_*(\gamma_{\widehat{Z}_\delta})\geq 4 \lambda_*(\gamma_{\widetilde{Z}})\geq \lambda_*(Q)\geq \varepsilon^{3m\rho}$, which implies that $E_N^{\eta_N}_{U_{\varepsilon,r}}(\lambda_*(\widehat{Z}_\delta)^{-p})<\infty$ for every $p$.
Let us study the lowest eigenvalue of $\gamma_{\widehat{X}_\delta}$. We use here some results from next Section \ref{upper}, namely Lemma \ref{gammaF}. There, we actually prove the desired bound for the Malliavin covariance matrix of $\alpha^{-1}(X_\delta-x_0)$. Here we are considering $\widehat{X}_\delta=\alpha^{-1}(X_\delta-x_0-b(0,x_0)\delta)$, but their Malliavin covariance matrix is the clearly the same. Then, Lemma \ref{gammaF} gives that
$E_N^{\eta_N}(\lambda_*(\gamma_{\widehat{X}_\delta})^{-p})<\infty$ for every $p$.
So, we have proved that all the requests in Proposition \ref {generalbounds} hold. Then, we can apply \eqref{generallowerbound} and we get
$$
p_{{\widehat{X}_\delta},U_{\varepsilon,r}}(z)
\geq p_{{\widehat{Z}_\delta},U_{\varepsilon,r}}(z)
-C'\sqrt\delta
$$
with $C'\in \mathcal{C}$. Now, from \eqref{ZG} and \eqref{estZ}, with a simple change of variables, we get
\be{boundG}
p_{{\widehat{Z}_\delta},U_{\varepsilon,r}}(z) \geq
\frac{1}{C}, \quad \mbox{for}\quad |z|\leq \frac{r}{2}.
\end{equation}
We can assert the existence of $\delta_*\in 1/\mathcal{C}$ and $C\in \mathcal{C}$ such that for all $\delta\leq \delta_*$
$$
p_{{\widehat{X}_\delta},U_{\varepsilon,r}}(z)
\geq \frac 1{2C},
$$
and the statement follows.
\end{proof}
We are now ready for the proof of the lower bound:
\bt{lower-th}
Let Assumption \ref{assumption1} and \ref{assumption2} hold. Let $p_{X_t}$ denote the density of $X_t$, $t>0$. Then there exist positive constants $r,\delta_*,C$ such that for every $\delta\leq \delta_*$ and for every $y$ such that $|y- x_0-b(0,x_0)\delta|_{A_\delta(0,x_0)}\leq r$,
\[
p_{X_\delta}(y) \geq
\frac{1}{C \delta^{n-\frac{\mathrm{dim}\langle \sigma(0,x_0) \rangle}{2} }},
\]
$\mathrm{dim}\langle \sigma(0,x_0) \rangle$ denoting the dimension of the vector space spanned by $\sigma_1(0,x_0),\ldots,\sigma_d(0,x_0)$.
Here, $C\in\mathcal{C}$ and $r,\delta_\ast\in 1/\mathcal{C}$.
\end{theorem}
\begin{proof}
We take the same $\delta_*,r$ as in Lemma \ref{estF} and let ${\widehat{X}_\delta}$ denotes the r.v. handled in Lemma \ref{estF}. By construction, we have $X_\delta=x_0+b(0,x_0)+\alpha {\widehat{X}_\delta}$, so by applying Lemma \ref{estF} we get
\[
\begin{split}
E_N^{\eta_N}(f(X_\delta)) & \geq E_N^{\eta_N}_{U_{\varepsilon,r}}(f(X_\delta)) = E_N^{\eta_N}_{U_{\varepsilon,r}}(f(x_0 + b(0,x_0)\delta+ \alpha \widehat{X}_\delta))\\
&= \int f(x_0 +b(0,x_0)\delta+ \alpha z) p_{\widehat{X}_\delta,U_{\varepsilon,r}}(z) dz\\
&\geq \frac{1}{C} \int_{\{|z|\leq r/2\}} f(x_0 +b(0,x_0)\delta+ \alpha z) dz\\
&\geq \frac{1}{C|\det \alpha |} \int_{|y|_{\alpha}\leq r/2}
f(x_0 +b(0,x_0)\delta+ \,y) dy\\
\end{split}
\]
From \eqref{Adlucia} and the Cauchy-Binet formula we obtain
\be{CB}
\frac{1}{C} \delta^{n-\frac{\mathrm{dim}\langle \sigma \rangle}{2}} \leq
\sqrt{|\det A_\delta A_\delta^T |} =\det(\alpha)\leq
C \delta^{n-\frac{\mathrm{dim}\langle \sigma \rangle}{2}}
\end{equation}
and the statement follows.
\end{proof}
\begin{remark}\label{classC}
We observe that if the diffusion coefficients are bounded, that is Assumption \ref{assumption1bis} holds, then the class $\mathcal{C}$ in \eqref{CC} of the constants can be replaced by
$$
\mathcal{D}_0=\Big\{C>0\,:\,C=c \Big(\frac{\kappa}{\lambda(0,x_0)}\Big)^{q},\ \exists\ c,q>0\Big\}
$$
and, as before, $1/\mathcal{D}_0=\{C>0\,:\,1/C\in\mathcal{D}_0\}$. This is because in the estimates for $\|\widehat{X}_\delta-\widehat Z_\delta\|_{2,p}$ and $\|\widehat{X}_\delta\|_{2,p}$ one does not need any more to use the Gronwall's Lemma but it suffices to use the boundedness of the coefficients and the Burkholder inequality.
\end{remark}
\section{Upper bound}\label{upper}
We study here the upper bound for the density of $X_\delta$.
\subsection{Rescaling of the diffusion}
As for the lower bound, we again scale $X_\delta$. We recall the results and the notation in Lemma \ref{Gamma-prop} and we define the change of variable
\be{Ta}
T_\alpha:{\mathbb R}^n \rightarrow {\mathbb R}^n , \quad\quad T_\alpha(y)=\alpha^{-1} y, \quad \mbox{where}\quad \alpha=\mathcal{U} \bar{\Sigma}
\end{equation}
and its adjoint $T_\alpha^*(v)=\alpha^{-1,T} v$. Note that $\alpha$ is a $n\times n$ matrix. We write $A_{\delta,j}$, for $j=1,\dots, m$, for the columns of $A_\delta$ (which can be $\sqrt{\delta}\sigma_i$ or $\delta[\sigma_i,\sigma_p]$). The following properties hold:
\begin{lemma}
Let $T_\alpha$ be defined in (\ref{Ta}). Then one has:
\begin{align}
& |y|_{A_\delta}=|T_\alpha y|=|y|_\alpha,\quad \forall y\in {\mathbb R}^n, \quad\mbox{ and }\det \alpha= \sqrt{\det A_\delta A_\delta^T} \label{p1}\\
& \forall v\in {\mathbb R}^n \mbox{ with }|v|=1,\quad \exists j=1,\dots, m : \quad |T_\alpha^*v \cdot A_{\delta,j}|\geq \frac{1}{m} \label{p2}\\
& \forall j=1,\dots,d, \quad \sqrt{\delta}|T_\alpha\sigma_j| \leq 1 \label{p3}
\end{align}
\end{lemma}
\begin{proof}{}
\eqref{p1} follows easily from $\alpha=\mathcal{U} \bar{\Sigma}$ and the definition \eqref{Not4} of $|\cdot|_{M}$. Now, $(T_\alpha^*v)^T A_\delta = v^T \alpha^{-1} A_\delta = [v^T 0] \mathcal{V}^T$. So $|(T_\alpha^*v)^T A_\delta|=|[v^T 0] \mathcal{V}^T|=1$. Recall that $A_{\delta,j}$ are the columns of $A_\delta$, therefore $\exists j=1,\dots, m : \quad | (T_\alpha^*v)^T A_{\delta,j}|\geq \frac{1}{m}$, which is equivalent to \eqref{p2}. Moreover, $T_\alpha A_\delta=[\mathrm{Id}_n\, 0]\mathcal{V}^T$. This easily implies that $\forall i=1,\dots, m,\quad | T_\alpha A_{\delta,i}|\leq 1$. For $A_{\delta,i}=\sigma_j(0,x_0)\sqrt{\delta}$ we have \eqref{p3}.
\end{proof}
We define now
\begin{equation}\label{F}
F=\alpha^{-1} (X_\delta-x_0)=T_\alpha (X_\delta-x_0).
\end{equation}
As for the lower bound, we first estimate the density of $F$, using the results in Appendix \ref{app-mall} (specifically, \eqref{generalupperbound} in Proposition \ref{generalbounds}), and then recover the estimates for the density of $X_\delta$ via a change of variable.
\subsection{Malliavin Covariance Matrix}\label{sectionnorris}
Let $F$ be as in \eqref{F}. To prove the upper bound for its density $p_F$ we need a quantitative control on the Malliavin covariance matrix $\gamma_F$ of $F$. We start with some preliminary results.
The following lemma is a slight modification of Lemma 2.3.1. in \cite{Nualart:06}.
\bl{lemma:matrixmoment}
Let $\gamma$ be a symmetric nonnegative definite $n\times n$ random matrix. Denoting $|\gamma|=\sum_{1\leq i,j \leq n} |\gamma^{i,j}|^2)^{1/2}$, we assume that, for $p\geq 2$, $E_N^{\eta_N}|\gamma|^{p+1} < \infty$, and that there exists $\varepsilon_0>0$ such that for $\varepsilon \leq \varepsilon_0$,
\[
\sup_{|\xi|=1} {\mathbb P} [\langle\gamma \xi,\xi \rangle<\varepsilon]\leq \varepsilon^{p+2n}
\]
Then there exists a constant $C$ depending only on the dimension $n$ such that
\[
E_N^{\eta_N} \lambda_*(\gamma)^{-p} \leq C E_N^{\eta_N} |\gamma|^{1+p} \varepsilon_0^{-p}
\]
\end{lemma}
We also need the following technical result.
\bl{covT}
Let $\delta\in(0,1]$ and let $a_t, b_t$, $t\in [0,\delta]$ be stochastic processes which are a.s. increasing. Assume that $b_0=0$. Suppose for fixed $p\geq 1$ and for all $ t\in[0,\delta]$ one has
\[
E_N^{\eta_N}[b_t^p]\leq C_p t^{2p}\quad\mbox{ and }\quad a_t\geq \frac{t-b_t}{\delta}.
\]
Then for all $\varepsilon>0$
\[
{\mathbb P}( a_\delta \leq \varepsilon)\leq
4^{p}C_p \varepsilon^{p}.
\]
\end{lemma}
\begin{proof}
Set
\[
S_\varepsilon = \inf \left\{ s \geq 0: b_s \geq \frac{\delta\varepsilon}{2} \right\} \wedge \delta,
\]
Remark that for any $p>0$
\[
{\mathbb P}(S_\varepsilon < \delta\varepsilon)
=
{\mathbb P}\left(
b_{\delta\varepsilon}^p
\geq \left(\frac{\delta\varepsilon}{2}\right)^p
\right)
\leq 2^p \frac{E_N^{\eta_N} b_{\delta\varepsilon}^p}
{(\delta\varepsilon)^{p}}
\leq 2^p C_p (\delta \varepsilon)^p .
\]
On the other hand, on $S_\varepsilon\geq \delta\varepsilon$,
\[
a_{S_\varepsilon} \geq a_{\delta\varepsilon} \geq \frac{\delta\varepsilon-\delta\varepsilon/2 }{\delta} \geq
\varepsilon/2.
\]
Therefore
\[
{\mathbb P}(a_\delta < \varepsilon/2) \leq
{\mathbb P}(a_\delta < \varepsilon/2, S_\varepsilon<\delta\varepsilon)+{\mathbb P}(a_\delta < \varepsilon/2, S_\varepsilon\geq \delta\varepsilon)
\leq
{\mathbb P}(S_\varepsilon < \delta\varepsilon)\leq 2^p C_p\varepsilon^p.
\]
This implies that ${\mathbb P}(a_\delta < \varepsilon) \leq 4^{p}C_p \varepsilon^{p}$.
\end{proof}
The following Lemma \ref{norris} is a refinement of what was proved by Norris in \cite{Norris}, in the sense that we take care of the same quantities, but handling more carefully the dependence on the final time $t_0$. This is a key estimate for the proof of next Theorem \ref{upper-th}.
\bl{norris}
Suppose $u(t)=(u_1(t),\dots, u_d(t))$ and $a(t)$ are a.s. continuous and adapted processes such that for some $p\geq 1$, $C>0$ and for every $t_0\leq 1$ one has
\be{int_norr}
E_N^{\eta_N} \left[\sup_{0\leq s \leq t_0} |u_s|^p \right]\leq \frac{C}{t_0^p} ,\quad\quad E_N^{\eta_N} \left[\sup_{0\leq s \leq t_0} |a_s|^p \right]\leq \frac{C}{t_0^p}.
\end{equation}
Set
\[
Y(t)=y+\int_0^t a(s) ds + \sum_{k=1}^d \int_0^t u_k(s) dW^k_s.
\]
Then, for any $q>4$ and $r>0$ such that $6r+4<q$, there exists $\varepsilon_0(q,r,p)$ such that for every $t_0 \leq 1$ and $\varepsilon\leq \varepsilon_0(q,r,p)$ one has
\[
{\mathbb P}\left\{\int_0^{t_0} Y_t^2 dt <\varepsilon^q,\, \int_0^{t_0} |u(t)|^2 dt \geq \frac{6\varepsilon}{t_0}
\right\}\leq (2^pC+1)\varepsilon^{rp}.
\]
\end{lemma}
\begin{proof}{}
Set $\theta_t=|a_t|+|u_t|$, and
\[
\tau=\inf\left\{ s\geq 0: \sup_{0\leq u\leq s} \theta_u >\frac{\varepsilon^{-r}}{t_0} \right\} \wedge t_0.
\]
We have
\[
{\mathbb P}\left\{\int_0^{t_0} Y_t^2 dt <\varepsilon^q,\, \int_0^{t_0} |u(t)|^2 dt \geq \frac{\varepsilon}{t_0}
\right\}
\leq A_1+A_2.
\]
where $A_1={\mathbb P}[\tau<t_0]$ and
\[
A_2= {\mathbb P}\left\{
\int_0^{t_0} Y_t^2 dt<\varepsilon^q,\int_0^{t_0} |u_t|^2dt\geq \frac{\varepsilon}{t_0}, \tau=t_0
\right\}
\]
An upper bound for $A_1$ easily follows from \eqref{int_norr}. Indeed
\[
{\mathbb P}[\tau<t_0]\leq {\mathbb P}\left[\sup_{0\leq s\leq t_0} \theta_u >\frac{\varepsilon^{-r}}{t_0}\right]
\leq t_0^p \varepsilon^{rp}E_N^{\eta_N}\left[\sup_{0\leq s\leq t_0} \theta_s^p\right]
\leq 2^p C\varepsilon^{rp}.
\]
for $\varepsilon\leq \varepsilon_0$. To estimate $A_2$ we introduce
\begin{align*}
&N_t=\int_0^t Y_s \sum_{k=1}^d u^k_s dW^k_s\quad\mbox{and}\\
&B= \left\{\langle N\rangle_\tau <\rho,\sup_{0\leq s\leq \tau }|N_s|\geq \delta \right\},\quad\mbox{with}
\quad \delta=\frac{\varepsilon^{2r+2}}{t_0} \quad\mbox{and}\quad \rho=\frac{\varepsilon^{q-2r}}{t_0^2}.
\end{align*}
By the exponential martingale inequality,
\[
{\mathbb P}(B)\leq \exp (\frac{-\delta^2}{2\rho} )\leq \exp\left( -\varepsilon^{6r+4-q} \right).
\]
So, in order to conclude the proof, it suffices to show that
\be{incl}
\left\{
\int_0^{t_0} Y_t^2 dt<\varepsilon^q,\int_0^{t_0} |u_t|^2 dt\geq \frac{6\varepsilon}{t_0}, \tau =t_0
\right\}\subset B,
\end{equation}
We suppose $\omega \notin B$, $\int_0^{t_0} Y_t^2 dt<\varepsilon^q$ and $\tau=t_0$ and show $\int_0^{t_0} |u_t|^2dt < 6\varepsilon/t_0$. With these assumptions,
\[
\langle N \rangle_\tau =\int_0^{\tau } Y_t^2 |u_t|^2 dt
\leq \int_0^{t_0} Y_t^2 dt\,\sup_{0\leq t\leq \tau }|u_t|^2 <\frac{\varepsilon^{q-2r}}{t_0^2}=\rho.
\]
So, for $\omega\notin B$ then one necessarily has $\sup_{0\leq t\leq \tau } |\int_0^{t} Y_s \sum_{k=1}^d u^k_s dW^k_s|<\delta=\varepsilon^{2r+2}/t_0$. From $6r+4\leq q$, if $\tau =t_0$ then
\begin{align*}
\sup_{0\leq t\leq \tau }\left|\int_0^t Y_s a_s ds\right|
&\leq \left( t_0 \int_0^\tau Y_s^2 a_s^2 ds\right)^{1/2}
\leq t_0\Big(\int_0^{t_0}Y_s^2ds\sup_{0\leq s\leq \tau }|a_s|^2\Big)^{1/2}\\
&\leq t_0\Big(\varepsilon^q\,\frac {\varepsilon^{-2r}}{t_0^2}\Big)^{1/2}
\leq
\frac{\varepsilon^{2r+2}}{t_0}.
\end{align*}
Thus
\[
\sup_{0\leq t\leq \tau } \left|\int_0^t Y_s dY_s \right|\leq
\sup_{0\leq t\leq \tau } \left|\int_0^t Y_s a_s ds+\int_0^t Y_s u_s dW_s \right|
\leq \frac{2 \varepsilon^{2r+2}}{t_0}.
\]
By It\^o's formula, $Y_t^2=y^2+2\int_0^t Y_s dY_s+\langle M\rangle_t$ with $\langle M\rangle_t=\int_0^t |u_s|^2ds$. So, recalling that $q>2r+2$,
\[
\int_0^\tau \langle M\rangle_tdt =
\int_0^\tau Y_t^2 dt - \tau y^2 - 2\int_0^\tau \int_0^t Y_s dY_s dt \\
< \varepsilon^q +4 \frac{\varepsilon^{2r+2}}{t_0}<5\frac{\varepsilon^{2r+2}}{t_0}.
\]
Since $t\mapsto \langle M\rangle_t$ is non negative and increasing, for $0<\gamma <\tau $ we have
\[
\gamma \langle M \rangle_{\tau -\gamma}\leq \int_{\tau -\gamma}^\tau \langle M\rangle_tdt
\leq 5\frac{\varepsilon^{2r+2}}{t_0}.
\]
Using also the fact that
\[
\langle M\rangle_\tau -\langle M\rangle_{\tau -\gamma}=\int_{\tau -\gamma}^\tau |u_s|^2 ds \leq \gamma \frac{\varepsilon^{-2r}}{t_0^2},
\]
we have
\[
\langle M \rangle_\tau <\frac{5\varepsilon^{2r+2}}{\gamma}+ \gamma \frac{\varepsilon^{-2r}}{t_0^2}.
\]
With $\gamma=t_0 \varepsilon^{2r+1}$, this gives $\int_0^{t_0} |u_s|^2 ds =\langle M \rangle_\tau <\frac{6\varepsilon}{t_0}$.
\end{proof}
We are now ready to prove the non degeneracy of the Malliavin covariance matrix. More precisely, we prove a quantitative version of this property: the $L^p$ norm of the inverse of the Malliavin covariance matrix of $F$ is upper bounded by a constant in $\mathcal{C}$, $\mathcal{C}$ being defined in \eqref{CC}.
\bl{gammaF}
Let $\alpha$, $T_\alpha$ and $F=T_\alpha (X_\delta-x_0)$ be defined as in \eqref{Ta} and \eqref{F}. Let $\gamma_F$ denote the Malliavin covariance matrix of $F$. Then for any $p>1$ there exists $C\in\mathcal{C}$ such that, for $\delta\leq 1$, $E_N^{\eta_N} |\lambda_*(\gamma_F)|^{-p}\leq C$.
\end{lemma}
\begin{proof}
We need a bound for the moments of the inverse of
\[
\gamma_F= \sum_{k=1}^d \int_0^\delta D^k_s F D^k_s F^T ds.
\]
Following \cite{Nualart:06} we define the tangent flow $Y$ of $X$ as the derivative with respect to the initial condition of $X$: $Y_t:=\partial_x X_t$. We also denote its inverse $Z_t=Y_t^{-1}$. Then one has (remark that the equations we consider for $X$, $Y$ and $Z$ are all in Stratonovich form):
\be{defZY}
\begin{split}
Y_t&=\mathrm{Id} + \sum_{k=1}^d\int_0^t \nabla_x \sigma_k(s,X_s) Y_s \circ dW^k_s+\int_0^t \nabla_x b(s,X_s) Y_s ds\\
Z_t&=\mathrm{Id} - \sum_{k=1}^d\int_0^t Z_s \nabla_x \sigma_k(s,X_s) \circ dW^k_s-\int_0^t Z_s \nabla_x b(s,X_s)ds,
\end{split}
\end{equation}
where $\nabla_x \sigma_k$ and $\nabla_x b$ are the Jacobian matrix with respect to the space variable. It holds
\[
D_s X_\delta=Y_\delta Z_s \sigma(s,X_s),\quad s<\delta.
\]
By applying It\^o's formula we have the following representation, for $\phi\in C^{1,2}$:
\be{devZphim}
\begin{split}
Z_t \phi(t,X_t)&=
\phi(0,x_0)+\int_0^t Z_s \sum_{k=1}^d [\sigma_k,\phi](s,X_s) dW_s^k\\
&+
\int_0^t Z_s \left\{ [b,\phi]+ \frac{1}{2} \sum_{k=1}^d [\sigma_k,[\sigma_k,\phi]]+ \frac{\partial \phi}{\partial s} \right\} (s,X_s)\,ds
\end{split}
\end{equation}
(details are given in \cite{Nualart:06}, remark that in the r.h.s. above we are taking into account an It\^o integral). We now compute
\[
D_s F = \alpha^{-1} D_s X_\delta
=\alpha^{-1} Y_\delta Z_s \sigma(s,X_s)
=\alpha^{-1} Y_\delta \alpha \alpha^{-1} Z_s \sigma(s,X_s)
\]
so
\[
\gamma_F = \alpha^{-1} Y_\delta \alpha \, \bar{\gamma}_F \, (\alpha^{-1} Y_\delta \alpha)^T \quad \mbox{ where }\quad
\bar{\gamma}_F =
\alpha^{-1}
\int_0^\delta
Z_s \sigma(s,X_s)
\sigma(s,X_s)^T Z_s^T ds \,
\alpha^{-1,T},
\]
and
\[
\gamma^{-1}_F = (\alpha^{-1} Y_\delta \alpha)^{-1,T} \, \bar{\gamma}_F^{-1} \, (\alpha^{-1} Y_\delta \alpha)^{-1}.
\]
Now,
\[
(\alpha^{-1} Y_\delta \alpha)^{-1} =
\alpha^{-1} Z_\delta \alpha =
\mathrm{Id}_n
+ \alpha^{-1} (Z_\delta-\mathrm{Id}_n) \alpha
\]
Using the fact that $\lambda^*(\cdot)$ is a norm on the set of matrices, and that for two $n\times n$ matrices $A,B$, $\lambda^*(AB)\leq n \lambda^*(A) \lambda^*(B) $, we have
\[
\lambda_*\big(\gamma_F\big)^{-1}=
\lambda^*\big(\gamma_F^{-1}\big) \leq n^2 \lambda^*\big(\bar{\gamma}_F^{-1}\big) \,\lambda^*\big( (\alpha^{-1} Y_\delta \alpha)^{-1}\big)^2
\]
and
\[
\lambda^*\big((\alpha^{-1} Y_\delta \alpha)^{-1}\big)\leq 1+ n^2\lambda^*(\alpha^{-1}) \lambda^*(Z_\delta-\mathrm{Id}_n) \lambda^*(\alpha).
\]
Standard estimates (see also \eqref{defZY}) give $\lambda^*(Z_\delta-\mathrm{Id}_n) \leq C_1\sqrt{\delta}$ for some $C_1\in \mathcal{C}$. Moreover
\[
\begin{split}
& \lambda^*(\alpha)=\lambda^*(A_\delta)=\lambda^*(A \mathcal{D}_\delta) \leq n \lambda^*(A) \lambda^*(\mathcal{D}_\delta)\leq C_2 \sqrt{\delta}, \quad C_2\in \mathcal{C}\\
&\lambda^*(\alpha^{-1})\leq \frac{1}{\lambda_*(A \mathcal{D}_\delta)} \leq \frac{C_3}{\delta} , \quad C_3\in \mathcal{C}
\end{split}
\]
and so for all $q>1$ exists $C \in \mathcal{C}$ such that
\[
E_N^{\eta_N} \lambda^*\left((\alpha^{-1} Y_\delta \alpha)^{-1}\right)^{q} \leq C
\]
We now need to estimate the reduced matrix, i.e. prove
that for all $q>1$ exists $C \in \mathcal{C}$ such that
\be{estred}
E_N^{\eta_N} \lambda^*(\bar{\gamma}_F^{-1})^q=
E_N^{\eta_N} \lambda_*(\bar{\gamma}_F)^{-q} \leq C
\end{equation}
We show now that for any $p>0$, $\sup_{|v|=1} {\mathbb P}\left( \langle \bar{\gamma}_F v,v \rangle\leq \varepsilon\right) \leq \varepsilon^p$, for $\delta\leq 1$ for $\varepsilon\leq \varepsilon_0\in 1/\mathcal{C}$ not depending on $\delta$. Together with lemma \ref{lemma:matrixmoment} this implies \eqref{estred}.
Denote $\xi = T_\alpha^* v=\alpha^{-1,T}v$. From \eqref{p2} and the definition \eqref{Adlucia} of $A_\delta$ we have two possible cases:
A) $|\xi \cdot \sigma_j(0,x_0)| \geq \frac{1}{m\delta^{1/2}}$ for some $j=1,\dots, d$, or B)
$|\xi \cdot [\sigma_j,\sigma_l](0,x_0)|\geq \frac{1}{m\delta}$ for some $j,l=1,\dots, d,\, j\neq l$. Moreover
\be{changevariablematrix}
\alpha \bar{\gamma}_F \alpha^T
=\int_0^\delta
Z_s \sigma(s,X_s)
\sigma(s,X_s)^T Z_s^T ds.
\end{equation}
Therefore, with $\xi = T_\alpha^* v$, we have for any $q>1$
\[
\begin{split}
{\mathbb P}(\langle \bar{\gamma}_F v,v \rangle\leq \varepsilon^q) &=
{\mathbb P}\left(\xi^T \int_0^\delta Z_s \sigma(s,X_s)\sigma(s,X_s)^T Z_s^T ds\, \xi \leq \varepsilon^q \right) \\
&=
{\mathbb P}\left( \sum_{i=1}^d \int_0^\delta |\xi^T Z_s \sigma_i(s,X_s) |^2 ds \leq \varepsilon^q\right)
\end{split}
\]
We decompose this probability:
\[
\begin{split}
{\mathbb P}(\langle &\bar{\gamma}_F v,v \rangle\leq \varepsilon^q)
=
{\mathbb P}\left( \sum_{i=1}^d \int_0^\delta |\xi^T Z_t \sigma_i(t,X_t) |^2 dt \leq \varepsilon^q\right)\\
&\leq
{\mathbb P}\left( \sum_{i=1}^d \int_0^\delta |\xi^T Z_t \sigma_i(t,X_t) |^2 dt \leq \varepsilon^q,\sum_{i,k=1}^d \int_0^\delta |\xi^T Z_t [\sigma_i,\sigma_k](t,X_t) |^2 dt \leq \frac{\varepsilon}{\delta}\right)\\
&+
{\mathbb P}\left( \sum_{i=1}^d \int_0^\delta |\xi^T Z_t \sigma_i(t,X_t) |^2 dt \leq \varepsilon^q, \sum_{i,k=1}^d \int_0^\delta |\xi^T Z_t [\sigma_i,\sigma_k](t,X_t) |^2 dt > \frac{\varepsilon}{\delta} \right)\\
&=:I_1+I_2
\end{split}
\]
To estimate $I_1$ we distinguish the two cases A) and B) above.
Case A): $|\xi \cdot \sigma_j(0,x_0)| \geq \frac{1}{m\delta^{1/2}}$ for some $j=1,\dots, d$. We fix this $j$. Then,
\begin{align*}
&I_1
\leq
{\mathbb P}\Big( \int_0^\delta |\xi^T Z_t \sigma_j(t,X_t) |^2 dt \leq \varepsilon^q,
\int_0^{\delta} \Big|\xi^T \sum_{k=1}^d Z_t [\sigma_k,\sigma_j](t,X_t)\Big|^2 dt < \frac{ \varepsilon}{\delta} \Big) \\
&\leq
{\mathbb P}\Big( \int_0^\delta |\xi^T Z_t \sigma_j(t,X_t) |^2 dt \leq \varepsilon^q,
\sup_{0\leq t\leq \delta} \Big| \int_0^t \xi^T \sum_{k=1}^d Z_s [\sigma_k,\sigma_j](s,X_s) dW^k_s \Big|^2 < \frac{1}{12 m^2 \delta} \Big)\\
&+{\mathbb P}\Big(\sup_{0\leq t\leq \delta} |\int_0^t \xi^T \sum_{k=1}^d Z_s [\sigma_k,\sigma_j](s,X_s) dW_s^k |^2 \geq \frac{1}{12 m^2 \delta},
\int_0^\delta |\xi^T \sum_{k=1}^d Z_t [\sigma_k,\sigma_j](t,X_t)|^2 dt <\frac{ \varepsilon}{\delta} \Big)
\end{align*}
Set $u_s=(\xi^T Z_s [\sigma_k,\sigma_j](s,X_s))_{k=1,\dots, d}$. From the exponential martingale inequality we have
\be{mart}
\begin{split}
&{\mathbb P}\Big(\sup_{0\leq t\leq\delta }| \sum_{k=1}^d \int_0^t u^k_s dW^k_s |^2 \geq \frac{1}{12 m^2 \delta},\quad \int_0^\delta |u_t|^2 dt < \frac{\varepsilon}{\delta} \Big) \\
&\quad \leq
2\exp\left( -\frac{1}{ 12 m^2\delta} \times \frac{\delta}{ 2\varepsilon} \right)
=2\exp\left( -\frac{1}{ 24 m^2 \varepsilon} \right)<\varepsilon^p,
\end{split}
\end{equation}
the latter inequality holding for every $p>1$ and $\varepsilon\leq \varepsilon_0$.
We now define
$$
D:=\Big\{
\sup_{0\leq t\leq \delta} |\int_0^t \xi^T \sum_{k=1}^d Z_s [\sigma_k,\sigma_j](s,X_s) dW_s^k |^2 < \frac{1}{12 m^2 \delta}\Big\}
$$
and prove
\[
{\mathbb P}\left( \left\{\int_0^\delta |\xi^T Z_t \sigma_j(t,X_t) |^2 dt \leq \frac{\varepsilon^q}{4m^2}\right\}
\cap D \right)\leq \varepsilon^p
\]
which is equivalent to the desired estimate ${\mathbb P}\left( \left\{\int_0^\delta |\xi^T Z_t \sigma_j(t,X_t) |^2 dt \leq \varepsilon^q\right\} \cap D \right)\leq \varepsilon^p$.
From representation \eqref{devZphim}, for $\phi=\sigma_j$ we find
\[
Z_t \sigma_j(t,X_t)= \sigma_j(0,x_0)+ \int_0^t \sum_{k=1}^d Z_s [\sigma_k,\sigma_j](s,X_s)dW^k_s
+R_t,
\]
with
\[
R_t= \int_0^t Z_s \left\{ [b,\sigma_j]+ \frac{1}{2} \sum_{k=1}^d [\sigma_k,[\sigma_k,\sigma_j ]]+\frac{\partial\sigma_j}{\partial s}\right\}(s,X_s)\,ds.
\]
From $(a+b+c)^2\geq a^2/3 - b^2-c^2$ and $|\xi\cdot \sigma_j(0,x_0)|\geq \frac{1}{m\delta^{1/2}}$, for $\bar{t}\leq \delta$ we can write
\[
\begin{split}
& \int_0^{\bar{t}} |\xi^T Z_t \sigma_j(t,X_t) |^2 dt \\
&\quad\geq
\frac{\bar{t} |\xi^T \sigma_j(0,x_0) |^2}{3} -
\int_0^{\bar{t}} | \sum_{k=1}^d \int_0^t \xi^T Z_s [\sigma_k,\sigma_j](s,X_s)dW^k_s|^2 dt
-\int_0^{\bar{t}} | \xi^T R_t|^2 dt \\
&\quad \geq
\frac{\bar{t}}{3 \delta m^2}
- \int_0^{\bar{t}} | \sum_{k=1}^d \int_0^t \xi^T Z_s [\sigma_k,\sigma_j](s,X_s)dW^k_s|^2 dt
-\int_0^{\bar{t}} |\xi^T R_t|^2 dt.
\end{split}
\]
On the set $D$ one has
\[
\int_0^{\bar{t}} | \sum_{k=1}^d \int_0^t \xi^T Z_s [\sigma_k,\sigma_j](s,X_s)dW^k_s|^2 dt
\leq {\bar{t}}\frac{1}{12 m^2\delta},
\]
so
\[
\begin{split}
\int_0^{\bar{t}} |\xi^T Z_t \sigma_j(t,X_t) |^2 dt \geq
\frac{\bar{t}}{4 m^2 \delta} -
\int_0^{\bar{t}} | \xi^T R_t|^2 dt,
\end{split}
\]
that we rewrite as
\be{hplab}
\begin{split}
4 m^2 \int_0^{\bar{t}} |\xi^T Z_t \sigma_j(t,X_t) |^2 dt \geq
\frac{\bar{t} -
4 m^2 \delta \int_0^{\bar{t}} | \xi^T R_t|^2 dt }{\delta}.
\end{split}
\end{equation}
We now set
\[
\mbox{$\displaystyle a_{\bar{t}}= 4 m^2 \int_0^{\bar{t}} |\xi^T Z_t \sigma_j(t,X_t) |^2 dt$ on the set $D$ and $a_{\bar{t}}=\bar{t}/\delta$ on the set $D^c$,}
\]
$D^c$ denoting the complement of $D$.
Standard computations, considering also $|\xi|=|T_\alpha^* v|\leq |v| C/\delta = C/\delta$, give
$E_N^{\eta_N} (\int_0^{\bar{t}} | \xi^T R_t|^2 dt)^q \leq C {\bar{t}}^{3 q}/\delta^{2 q},$
so
$E_N^{\eta_N} ( 4 \delta m^2 \int_0^{\bar{t}} | \xi^T R_t|^2 dt)^q \leq C {\bar{t}}^{2 q} $, for $C\in \mathcal{C}$ (recall also $\bar{t} \leq \delta$).
This estimate and \eqref{hplab} allow us to apply lemma \ref{covT} with $a_{\bar{t}}$ defined above and
\[
b_{\bar{t}}= 4 \delta m^2 \int_0^{\bar{t}} | \xi^T R_t|^2 dt.
\]
We find
\[
\left\{ \int_0^\delta |\xi^T Z_t \sigma_j(t,X_t) |^2 dt \leq \frac{\varepsilon^q}{4m^2}\right\}
\cap D =\{ a_\delta \leq \varepsilon^q \} \cap D
\]
and we have
\[
{\mathbb P}\Big( \Big\{ \int_0^\delta |\xi^T Z_t \sigma_j(t,X_t) |^2 dt \leq \frac{\varepsilon^q}{4m^2}\Big\}
\cap D \Big)
={\mathbb P}( \{a_\delta \leq \varepsilon^q\} \cap D )
\leq
{\mathbb P}( a_\delta \leq \varepsilon^q )\leq \varepsilon^p.
\]
We the obtain $I_1 <\varepsilon^p $ for any $p>1$, for $\delta\leq 1$, $\varepsilon\leq \varepsilon_0$.
\smallskip
Case B) $|\xi \cdot [\sigma_j,\sigma_l](t,x_0)|\geq \frac{1}{m\delta}$ for some $j,l=1\dots d,\, j\neq l$. In this case we write
\[
I_1\leq {\mathbb P}\left(\int_0^\delta | \xi^T Z_t [\sigma_j,\sigma_l](t,X_t) |^2 dt \leq \frac{\varepsilon}{\delta} \right)
\]
From representation \eqref{devZphim} with $\phi=[\sigma_j,\sigma_l]$ we find
\[
Z_t [\sigma_j,\sigma_l](t,X_t)=[\sigma_j,\sigma_l](0,x_0)+R_t,
\]
with
\be{repR2}
\begin{split}
R_t & = \int_0^t Z_s \sum_{k=1}^d [\sigma_k,[\sigma_j,\sigma_l] ](s,X_s) dW_s^k\\
&+
\int_0^t Z_s \left\{ [b,[\sigma_j,\sigma_l]]+ \frac{1}{2} \sum_{k=1}^d [\sigma_k,[\sigma_k,[\sigma_j,\sigma_l] ]]+\frac{\partial [\sigma_j,\sigma_l]}{\partial s} \right\}(s,X_s)\, ds.
\end{split}
\end{equation}
From $(a+b)^2\geq a^2/2-b^2$ and $|\xi\cdot[\sigma_j,\sigma_l](0,x_0)|\geq \frac{1}{m \delta}$, for $\bar{t}\leq \delta$ we have
\be{inR}
\int_0^{\bar{t}} |\xi^T Z_t [\sigma_j,\sigma_l](t,X_t)|^2 dt
\geq \frac{ \bar{t} |\xi^T [\sigma_j,\sigma_l](0,x_0)|^2 }{2} - \int_0^{\bar{t}} |\xi^T R_t|^2 dt
\geq \frac{ \bar{t} }{2\delta^2 m^2} - \int_0^{\bar{t}} |\xi^T R_t|^2 dt.
\end{equation}
We apply lemma \ref{covT} with
\[
a_{\bar{t}}=2m^2 \delta \int_0^{\bar{t}} | \xi^T Z_s [\sigma_j,\sigma_l](s,X_s) |^2 ds\quad\mbox{and}\quad
b_{\bar{t}}=2m^2 \delta^2 \int_0^{\bar{t}} |\xi^T R_t|^2 dt
\]
Indeed from \eqref{repR2} and $|\xi|\leq C/\delta$,
\[
E_N^{\eta_N} |b_{\bar{t}}|^q \leq C \bar{t}^2
\]
and from \eqref{inR} we have $a_{\bar{t}}\geq \frac{\bar t-b_{\bar t}}\delta$. So, we find $I_1<\varepsilon^p$, for $\delta\leq 1$, $\varepsilon\leq \varepsilon_0$.
We estimate now
\[
I_2={\mathbb P}\left( \sum_{i=1}^d \int_0^\delta |\xi^T Z_s \sigma_i(s,X_s) |^2 ds \leq \varepsilon^q,
\sum_{i,j=1}^d \int_0^\delta |\xi^T Z_s [\sigma_i,\sigma_j](s,X_s) |^2 ds > \frac{\varepsilon}{\delta} \right).
\]
By using again \eqref{devZphim}, we find
\[
\begin{split}
\xi^T Z_t \sigma_i(t,X_t)
& = \sigma_i(0,x_0) +\sum_{j=1}^d \int_0^t \xi^T Z_s [\sigma_j,\sigma_i](s,X_s) dW^j_s \\
& + \int_0^t
\xi^T Z_s \left\{ [b,\sigma_i]+ \frac{1}{2} \sum_{j=1}^n [\sigma_j,[\sigma_j,\sigma_i]]+\frac{\partial \sigma_i}{\partial s} \right\}(s,X_s) ds.
\end{split}
\]
For $t_0= \delta$ and from the fact that $|\xi|\leq \frac{C}{\delta}$, we have
\[
\begin{split}
&E_N^{\eta_N} [\sup_{0\leq s \leq \delta} |\xi^T Z_s [\sigma_j,\sigma_i](s,X_s)|^p ] \leq \frac{C}{\delta^p}, \quad C\in {\mathcal{C}}, \quad \mbox{and}\\
&E_N^{\eta_N} \left[\sup_{0\leq s \leq \delta} |\xi^T Z_s \left\{[b,\sigma_i]+ \frac{1}{2} \sum_{j=1}^d [\sigma_j,[\sigma_j,\sigma_i]]\right\}(s,X_s)|^p \right] \leq \frac{C}{\delta^p}, \quad C\in {\mathcal{C}}.
\end{split}
\]
Thus we can apply Lemma \ref{norris} and we get
\[
{\mathbb P}\left( \sum_{i=1}^d \int_0^\delta |\xi^T Z_s \sigma_i(s,X_s) |^2 ds \leq \varepsilon^q \mbox{ and }
\sum_{i,j=1}^d \int_0^\delta |\xi^T Z_s [\sigma_i,\sigma_j](s,X_s) |^2 ds > \frac{\varepsilon}{\delta} \right)\leq \varepsilon^p
\]
for any $p>1$, $\delta\leq 1$ for $\varepsilon\leq \varepsilon_0$. We have now both the estimates of $I_1$ and $I_2$, so we have
$\sup_{|v|=1} {\mathbb P}(\langle \bar{\gamma}_F v,v \rangle\leq \varepsilon^q)\leq \varepsilon^p$ for
$p>1$, $\delta\leq 1$ for $\varepsilon\leq \varepsilon_0$, and the statement holds.
\end{proof}
\subsection{Upper bound for the density of $X_\delta$}
\bt{upper-th}
Let Assumption \ref{assumption1} and \ref{assumption2} hold. Let $p_{X_t}$ denote the density of $X_t$, $t>0$. Then, for any $p>1$, there exists a positive constant $C\in \mathcal{C}$ such that for every $\delta\leq 1$ and for every $y\in {\mathbb R}^n$
\[
p_{X_\delta}(y)\leq \frac{1}{\delta^{n-\frac{\mathrm{dim}\langle \sigma(0,x_0) \rangle}{2}}
} \frac{C}{1+|y-x_0|_{A_\delta(0,x_0)}^p}.
\]
Again, $\mathrm{dim}\langle \sigma(0,x_0) \rangle$ denotes the dimension of the vector space spanned by $\sigma_1(0,x_0),\ldots,$ $\sigma_d(0,x_0)$.
\end{theorem}
\begin{proof}
Set $F=T_\alpha(X_\delta-x_0)$. We apply estimate \eqref{generalupperbound}: there exist constants $p$ and $a$ depending only on the dimension $n$, such that
\[
p_F(z) \leq C \max\{1,E_N^{\eta_N} |\lambda_* (\gamma_F)|^{-p} \|F\|_{2,p} \} {\mathbb P}(|F-z|<2)^{a}.
\]
We first show that $\|F\|_{2,p}\leq C\in \mathcal{C}$, as a consequence of Assumption \ref{assumption1}. We prove just that $\|F\|_p\leq C$ for every $p$, for the Malliavin derivatives the proof is heavier but analogous. We write
$$
F=T_\alpha \Big( \sum_{j=1}^{d}
\int_0^\delta \sigma_{j}(t,X_{t})\circ dW_{t}^{j}+ \int_0^\delta b(t,X_{t})dt
\Big)
=T_\alpha \Big(
\sum_{j=1}^{d} \sigma_{j}(0,x_0) W_\delta^{j} +
B_\delta \Big),
$$
where
\[
B_\delta=
\sum_{j=1}^{d} \int_0^\delta \big(\sigma_{j}(t,X_{t})-\sigma_{j}(0,x_0)\big) \circ dW_{t}^{j}+ \int_0^\delta b(t,X_{t})dt.
\]
Therefore
\be{boundF}
|F|
\leq
\sum_{j=1}^{d} |T_\alpha \sigma_{j}(0,x_0) W_\delta^{j} | +
| T_\alpha B_\delta |.
\end{equation}
\eqref{p3} implies $ |T_\alpha \sigma_{j}(0,x_0) W_\delta^{j} | \leq C W_\delta^{j}/\sqrt{\delta}$, for $j=1,\dots,d$.
Moreover $| T_\alpha B_\delta |\leq |B_\delta|_{A_\delta}\leq C |B_\delta|/\delta$. If assumption \ref{assumption1} holds we conclude that $E_N^{\eta_N} |F|^p \leq C\in \mathcal{C}$.
As in \cite{BC14}, Remark 2.4, it is easy to reduce the estimate of ${\mathbb P}(|F-z|<2)$ to the tail estimate of $F$, and then to use Markov inequality to relate the estimate of the tails to the moments of $F$:
\be{tails}
{\mathbb P}(|F-z|<2)\leq {\mathbb P}(|F|>|z|/2) \leq C \frac{1\vee E_N^{\eta_N} |F|^p}{1+|z|^p},\quad \forall z\in {\mathbb R}^n
\end{equation}
Since, from Assumption \ref{assumption1}, all the moments of $F$ are bounded by constants in $\mathcal{C}$, we have that for any exponent $p>1$ this term decays faster than $|z|^{-p}$ for $|z|\rightarrow \infty$.
In Lemma \ref{gammaF} we have already proved that $E_N^{\eta_N} |\lambda_*(\gamma_F)|^{-q}\leq C\in \mathcal{C}$, for $\delta \leq 1$. We conclude that $p_F(z)\leq \frac{C}{1+|z|^p}$. The upper bound for the density of $X_\delta$ comes from the simple change of variable
$y=x_0+\alpha z$. For a positive and bounded measurable function $f:{\mathbb R}^n\rightarrow {\mathbb R}$, we write
\[
E_N^{\eta_N} f(X_\delta)=E_N^{\eta_N} f(x_0 + \alpha F)
= \int f(x_0 +\alpha z) p_F(z) dz
\]
and we apply our density estimate, so that
\[
E_N^{\eta_N} f(X_\delta)\leq \int \frac{C f(x_0 +\alpha z)}{1+|z|^p} dz
\leq \frac{C}{|\det\alpha|} \int \frac{f(y)}{1+|x_0-y|_{A_\delta(0,x_0)}^p} dy,
\]
in which we have used \eqref{p1}. Concerning $|\det \alpha|$, we recall \eqref{CB} and we obtain
\[
p_{X_\delta}(y)\leq \frac{1}{\delta^{n-\frac{dim \langle \sigma(0,x_0) \rangle}{2}}} \frac{C}{1+|x_0-y|_{A_\delta(0,x_0)}^p}.
\]
\end{proof}
\br{2}
If Assumption \ref{assumption1bis} holds then the upper estimate in Theorem \ref{upper-th} is of exponential type:
there exists a constant $C\in \mathcal{C}$ such that for every $\delta\leq 1$ and for every $y\in {\mathbb R}^n$
\[
p_{X_\delta}(y)\leq \frac{C}{\delta^{n-\frac{dim \langle \sigma(0,x_0) \rangle}{2}}} \exp(- \frac{1}{C}|y-x_0|_{A_\delta(0,x_0)}) .
\]
The proof is identical to the previous one except for the last part. In fact, looking at \eqref{boundF}, in this case the boundedness of the coefficients allows one to apply the exponential martingale inequality, so instead of \eqref{tails} we obtain the exponential bound ${\mathbb P}(|F|>|y|/2)\leq C\exp(-|y|/C)$. This actually gives the proof of (3) in Theorem \ref{main-th}.
\end{remark}
\br{3}
In Theorem \ref{lower-th} the lower bound is centered at $x_0+\delta b(x_0)$ but for the upper estimate in Theorem \ref{upper-th}, one can choose to center at $x_0$ or at $x_0+\delta b(x_0)$. In fact, in this case we notice that
\[
|\delta b(x_0)|_{A_\delta(0,x_0)}\leq \frac{C'}{\delta} |\delta b(x_0)|\leq C'',
\]
so
\[
\frac{C_1}{1+|x_0-y|_{A_\delta(0,x_0)}}\leq
\frac{C_2}{1+|x_0+\delta b(x_0)-y|_{A_\delta(0,x_0)}}\leq
\frac{C_3}{1+|x_0-y|_{A_\delta(0,x_0)}},
\]
and the estimate of Theorem \ref{upper-th} can be equivalently written as
\[
p_{X_\delta}(y)\leq \frac{1}{\delta^{n-\frac{\mathrm{dim}\langle \sigma(0,x_0) \rangle}{2}}
} \frac{C}{1+|y-x_0-\delta b(x_0)|_{A_\delta(0,x_0)}^p}.
\]
\end{remark}
\begin{remark}
Theorem \ref{upper-th} can be seen as an improvement of the upper bound in \cite{KusuokaStroock:85} in the sense that it precisely identifies the exponent
$n-\frac{\mathrm{dim} \langle \sigma(0,x_0) \rangle}{2}$, which accounts of the time-scale of the heat kernel when $\delta$ goes to zero. This is evident when we consider the diagonal estimate $y=x_0$, and the same consideration holds when $y$ is close to $x_0$. When looking at the tails ($y$ far from $x_0$), it is not clear which of the two upper bounds is more accurate, unless we further specify the model.
\end{remark}
| cbab8484e5834e65127a784252bf5cb1d12fa1a2 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Models of the nonlinear Schr{\"o}dinger (NLS)
type~\cite{sulem,ablowitz,siambook}
have proven to be rather universal in describing envelope
nonlinear wave structures in dispersive media. Such
structures emerge in fields ranging from water waves
and
nonlinear optics~\cite{kivshar} to plasmas~\cite{infeld}
and atomic Bose-Einstein
condensates (BECs)~\cite{emergent}.
A particularly interesting setting, recognized early
on (i.e., since the 1970's) in nonlinear optics in the context of
interaction of waves of different frequency
is that of multi-component NLS models~\cite{manakov}.
Among these, arguably, the most prototypical one
is the integrable~\cite{intman2} so-called Manakov model, which is
characterized by equal nonlinear
interactions within and across components.
Two decades after these initial developments within nonlinear
optics, a renewed interest has emerged for such multi-component
systems through the advent of ultra-cold atomic Bose-Einstein
condensates (BECs)~\cite{stringari,roman}. Numerous experiments
since then have focused on realizing such multi-component
BECs as mixtures of, e.g.,
different spin states of the same atom species
(so-called pseudo-spinor condensates)~\cite{Hall1998a,chap01:stamp}, or
different Zeeman sub-levels of the same hyperfine level
(spinor condensates)~\cite{Stenger1998a,kawueda,stampueda}.
A remarkable feature of these atomic systems when they pertain
to the same atomic species is that the so-called scattering lengths
controlling the inter-atomic interactions and hence effectively
the nonlinear prefactors are nearly equal within and across
components both in settings of, e.g., $^{87}$Rb and of $^{23}$Na.
This, in turn, translates in models well approximated by
the Manakov nonlinearity, enabling the experimental
realization not only of ground states, but also of numerous
solitonic excitations, most notably
of dark-bright solitons and their
variants~\cite{hamburg,pe1,pe2,pe3,azu,pe4,pe5}. These
developments have been recently summarized in a number of
reviews and books~\cite{siambook,emergent,revip}.
Our aim in the present work is to consider the harmonically trapped setting
of atomic Bose-Einstein condensates in the context of the multi-component
models discussed above. In earlier work, both a subset of the present
authors~\cite{todd1,todd2}, as well as other researchers~\cite{feder}
explored the use of analytical techniques in order to examine the existence
and stability of solutions in the vicinity of the well-understood linear
(quantum harmonic oscillator) limit of single-component models featuring one
atomic species. The relevant methods included, e.g., among others the use of
Lyapunov-Schmidt conditions for persistence of solutions near this limit, as
well as the use of the Krein signature and related topological index
tools~\cite{toddbook} to characterize the stability of the resulting
excitations. The topological tools are used to determine the potential number
of unstable directions associated with an excitation, while the analysis
allows us to determine which of the potential instabilities are realized.
Here, we extend such considerations to the more involved
setting of two-component systems. While we will not do so in this paper, the
ideas presented herein can be used to consider the existence and spectral
stability of solutions to systems with three or more components. As in the
one-component setting, nonlinear states emanate (bifurcate) from a
corresponding linear state. In the one-component setting the linear state
corresponds to an eigenfunction with a specified number of nodes which is
associated with a simple eigenvalue. In the two-component case the
eigenvalues are semi-simple, and in the cases considered herein will be of
multiplicity two. At the linear level the first component will have $m$
nodes (i.e., we will denote the number of nodes
of that component by $m$),
while the second component will have $n$ nodes. The value of such
topological and analytical tools in uncovering the potential number of
unstable eigendirections of each such pair $(m,n)$ can be considerable in
shaping the expectation of the potential experimental observability of
different states.
It should be highlighted here that it does not escape us that
low atom numbers are more prone to effects of quantum fluctuations
potentially detrimental to the existence of the states
(although it is our understanding that the study of such effects
in multi-component systems is fairly limited). Nevertheless, our
argument is that the value of considerations such as the topological
ones presented herein is that they are of broader value in
uncovering potential eigendirections beyond the vicinity of
the linear limit and hence of relevance to regimes where the
states could be observable as described by the mean-field Manakov-like
limit discussed herein and as has been revealed
experimentally e.g. in~\cite{hamburg,pe1,pe2,pe3}. For instance, an intriguing example of a finding
that we present herein is that even the very robust (and experimentally
identified) dark-bright soliton not only possesses a potentially unstable
eigendirection, but this instability is realized provided that the
inter-component interaction is increased sufficiently (an experimentally
feasible scenario via the tuning of the inter-component
scattering length by means of so-called
Feshbach resonances~\cite{siambook}).
Our presentation will be structured as follows. In section II, we will
briefly present the theoretical setup and the analysis of the
existence of the different solutions $(m,n)$. In section III, we will
present a general framework for considering the stability of these
states. In section IV, we catalogue the different possible
states $(m,n)$ with $0 \leq m,n \leq 2$. Finally, in section V
we summarize our findings and present our conclusions, as well
as a number of directions for future study.
\section{Theoretical setup and Existence Results}
We consider the following two-component system, bearing in mind
the setting of two hyperfine states of, e.g., $^{87}$Rb~\cite{revip}
\begin{eqnarray}
\label{eqn00_1}
i\partial_t {\phi}_1(x,t) &=& (-\frac{1}{2}\partial_{xx}+\frac{1}{2}\Omega^2 x^2-\mu_1){\phi}_1+ ( g_{11}|{\phi}_1|^2 + g_{12}|\phi_2|^2 )\phi_1, \\
\label{eqn00_2}
i\partial_t \phi_2(x,t) &=& (-\frac{1}{2}\partial_{xx}+\frac{1}{2}\Omega^2 x^2-\mu_2)\phi_2+ ( g_{21}|\phi_1|^2 + g_{22}|\phi_2|^2 )\phi_2
\end{eqnarray}
where $\phi_j\in \mathbb{C}$ is the mean-field wave-function of species
$j,\,g_{jk}\in \mathbb{R}^+$ with $g_{12} = g_{21}$, $\mu_j\in \mathbb{R}$
represents the chemical potential for species $j$, and parabolic trapping
potentials are considered here with the same trapping frequency $\Omega$ for
both species; $\Omega$ effectively represents the ratio of the trapping
strengths along the longitudinal (elongated) and transverse (strongly
trapped) directions.
Focusing on the wave functions such that $\int_{\mathbb{R}}|\phi_j(x)|^2 dx = O(\epsilon)$ where $\epsilon\ll 1$ (i.e., the small amplitude, near-linear
limit discussed in the previous section), we introduce the scaling $\phi_j = \epsilon^{1/2} \psi_j$ and obtain the following equations:
\begin{eqnarray}
\label{eqn0_1}
i\partial_t {\psi}_1(x,t) &=& f_1= (-\frac{1}{2}\partial_{xx}+\frac{1}{2}\Omega^2 x^2-\mu_1) {\psi}_1+\epsilon( g_{11}| {\psi}_1|^2 + g_{12}| {\psi}_2|^2 ) {\psi}_1, \\
\label{eqn0_2}
i\partial_t {\psi}_2(x,t) &=& f_2= (-\frac{1}{2}\partial_{xx}+\frac{1}{2}\Omega^2 x^2-\mu_2) {\psi}_2+\epsilon( g_{12}| {\psi}_1|^2 + g_{22}| {\psi}_2|^2 ) {\psi}_2
\end{eqnarray}
where $\int_{\mathbb{R}}| {\psi}_j(x)|^2 dx = O(1)$ now. Due to the gauge
invariance of the system, if $\{\psi_1, \psi_2\}$ are a solution, then
$\{\psi_1 e^{i\theta_1}, \psi_2 e^{i\theta_2} \}$ will also be a solution for
any real $\theta_1$ and $\theta_2$. In this paper we will focus on the existence
and spectral stability of real-valued steady-state solutions for $\psi_1$ and
$\psi_2$.
Set
\[
\boldsymbol{\psi}=\left(\begin{array}{c}\psi_1\\\psi_2\end{array}\right),\quad
\boldsymbol{f}=\left(\begin{array}{c}f_1\\f_2\end{array}\right),\quad
\boldsymbol{\mu}=\left(\begin{array}{c}\mu_1\\\mu_2\end{array}\right).
\]
We seek the stationary solutions
$\boldsymbol{\psi}(x,t)=\boldsymbol{\psi}(x)$ through the continuation of a
nontrivial solution for $\epsilon=0$. For the moment assume the asymptotic
expansions,
\begin{equation}\label{e:expand}
\boldsymbol{\psi}=\boldsymbol{\psi}^{(0)}+\epsilon\boldsymbol{\psi}^{(1)}+O(\epsilon^2),\,\,
\boldsymbol{\mu}=\boldsymbol{\mu}^{(0)}+\epsilon\boldsymbol{\mu}^{(1)}+O(\epsilon^2).
\end{equation}
These expansions will be verified through a Lyapunov-Schmidt reduction, which
requires a detailed understanding of the linearized problem associated with
(\ref{eqn0_1})-(\ref{eqn0_2})~\cite{nirenberg}. We linearize about the steady
state solution by taking the Fr\'{e}chet derivative of $\boldsymbol{f}(
{\boldsymbol{\psi}}, {\boldsymbol{\psi}}^*, \epsilon)$ and $\boldsymbol{f^*}(
{\boldsymbol{\psi}}, {\boldsymbol{\psi}}^*, \epsilon)$ with respect to
${\boldsymbol{\psi}}$ and ${\boldsymbol{\psi}}^*$; star here stands for
complex conjugation. Let $\mathcal{L}$ denote the operator associated with
the linearization
having the asymptotic expansion
$\mathcal{L}=\mathcal{L}^{0)}+\epsilon\mathcal{L}^{(1)}+O(\epsilon^2)$, where
\begin{align}
\label{eqn0_L0}
\mathcal{L}^{(0)}=
\left(
\begin{array}{cccc}
-\frac{1}{2}\partial_{xx} + \frac{1}{2}\Omega^2 x^2 - \mu_1^{(0)} & 0 & 0 & 0\\
0 & -\frac{1}{2}\partial_{xx} + \frac{1}{2}\Omega^2 x^2 - \mu_1^{(0)} & 0 & 0\\
0 & 0 & -\frac{1}{2}\partial_{xx} + \frac{1}{2}\Omega^2 x^2 - \mu_2^{(0)} & 0\\
0 & 0 & 0 & -\frac{1}{2}\partial_{xx} + \frac{1}{2}\Omega^2 x^2 - \mu_2^{(0)}
\end{array}
\right),
\end{align}
and
\begin{align}
\label{eqn0_L}
\begin{split}
\mathcal{L}^{(1)}=
\left(
\begin{array}{cccc}
2g_{11}|{\psi}_1^{(0)}|^2+g_{12}|{\psi}_2^{(0)}|^2 & g_{11}({\psi}_1^{(0)})^2 & g_{12}{\psi}_1^{(0)} {\psi}_2^{(0)} & g_{12}{\psi}_1^{(0)} {\psi}_2^{(0)}\\
g_{11}({\psi}_1^{(0)})^2 & 2g_{11}|{\psi}_1^{(0)}|^2+g_{12}|{\psi}_2^{(0)}|^2 & g_{12}{\psi}_1^{(0)}{\psi}_2^{(0)} & g_{12}{\psi}_1^{(0)}{\psi}_2^{(0)}\\
g_{12}{\psi}_1^{(0)}{\psi}_2^{(0)} & g_{12}{\psi}_1^{(0)}{\psi}_2^{(0)} & 2g_{22}|\psi_2^{(0)}|^2+g_{21}|\psi_1^{(0)}|^2 & g_{22}({\psi}_2^{(0)})^2\\
g_{12}{\psi}_1^{(0)}{\psi}_2^{(0)} & g_{12}{\psi}_1^{(0)}{\psi}_2^{(0)} & g_{22}({\psi}_2^{(0)})^2 & 2g_{22}|{\psi}_2^{(0)}|^2+g_{21}|{\psi}_1^{(0)}|^2
\end{array}
\right).
\end{split}
\end{align}
Focusing on real solutions, we can directly see a symmetry of $\mathcal{L}$:
\begin{equation}\label{e:L-property}
J_1^T \mathcal{L} J_1 = \mathcal{L},\quad
J_1 =
\left(
\begin{array}{cccc}
0 & 1 & 0 & 0\\
1 & 0 & 0 & 0\\
0 & 0 & 0 & 1\\
0 & 0 & 1 & 0
\end{array}
\right).
\end{equation}
Using the linear eigenvalues of the quantum harmonic oscillator,
\[
\mu_1^{(0)}=\Omega(m+\frac{1}{2}),\quad
\mu_2^{(0)}=\Omega(n+\frac{1}{2}),
\]
it can be directly observed that
$\mathcal{L}^{(0)}$ has a non-empty kernel spanned by $\{
(u_{m},0,0,0)^T , (0,u_{m},0,0)^T , (0,0,u_{n},0)^T, (0,0,0,u_{n})^T \}$,
where
\[
u_{k}(x) = \sqrt{\frac{1}{2^k k!
}}\left(\frac{\Omega}{\pi}\right)^{1/4} H_k(\sqrt{\Omega} x)e^{-\frac{\Omega
x^2}{2}},
\]
and $H_k$ are the Hermite polynomials. The first three states are
\[
u_0(x)=\left(\frac{\Omega}{\pi}\right)^{1/4} e^{-\Omega x^2/2},\,\,
u_1(x)=\left(\frac{\Omega}{\pi}\right)^{1/4}\sqrt{2\Omega} x e^{-\Omega x^2/2},\,\,
u_2(x)=\left(\frac{\Omega}{\pi}\right)^{1/4}\sqrt{\frac{1}{2}}(2\Omega x^2-1) e^{-\Omega x^2/2}.
\]
The collection of states $\{u_0,u_1,u_2,\dots\}$ has the properties that:
\begin{enumerate}
\item $\displaystyle{\left(-\frac{1}{2}\partial_{xx}+\frac{1}{2}\Omega^2
x^2\right)u_{j}=\Omega(j+\frac{1}{2})u_j}$ for each $j=0,1,2,\dots$
\item $u_j(x)$ has $j$ simple zeros for each $j=0,1,2,\dots$
\item the set is orthonormal under the inner product $\displaystyle{\langle
g,h\rangle=\int_\mathbb{R}g(x)h(x)^*\,\mathrm{d}x}$
\item the set is a basis for $L^2(\mathbb{R})$.
\end{enumerate}
We now consider the existence problem. We apply the Lyapunov-Schmidt
Reduction Method to Eqs.~(\ref{eqn0_1})-(\ref{eqn0_2}) with
\[
\boldsymbol{\mu}\approx\left(\begin{array}{c}\Omega(m+1/2)\\\Omega(n+1/2)\end{array}\right).
\]
The state will hereafter be denoted as $(m,n)$. Since the vector field is
smooth, and the eigenvalues are semi-simple, the reduction guarantees that
both $\boldsymbol{\mu}$ and $\boldsymbol{\psi}$ will have an asymptotic
expansion in $\epsilon$ of~(\ref{e:expand}).
Equations~(\ref{eqn0_1})--(\ref{eqn0_2}) at order $O(1)$ are
\begin{eqnarray}
\label{eqn1_1}
0 &=& (-\frac{1}{2}\partial_{xx}+\frac{1}{2}\Omega^2 x^2-\mu_1^{(0)}){\psi}_1^{(0)}, \\
\label{eqn1_2}
0 &=& (-\frac{1}{2}\partial_{xx}+\frac{1}{2}\Omega^2 x^2-\mu_2^{(0)}){\psi}_2^{(0)}.
\end{eqnarray}
The nontrivial solution is the expected one,
\[
\boldsymbol{\psi}^{(0)}=\left(\begin{array}{c}au_m\\bu_n\end{array}\right),\quad
\boldsymbol{\mu}^{(0)}=\left(\begin{array}{c}\Omega(m+1/2)\\\Omega(n+1/2)\end{array}\right),
\]
where $a,b\in\mathbb{R}$.
The next set of equations at $O(\epsilon)$ will provide the definitive values
that $a$ and $b$ must assume. Equations~(\ref{eqn0_1})--(\ref{eqn0_2}) at order
$O(\epsilon)$ are
\begin{eqnarray}
\label{eqn2_1}
0 &=& (-\frac{1}{2}\partial_{xx}+\frac{1}{2}\Omega^2 x^2-\mu_1^{(0)}){\psi}_1^{(1)} + ( g_{11}|{\psi}_1^{(0)}|^2 + g_{12}|{\psi}_2^{(0)}|^2 - \mu_1^{(1)} ){\psi}_1^{(0)} , \\
\label{eqn2_2}
0 &=& (-\frac{1}{2}\partial_{xx}+\frac{1}{2}\Omega^2 x^2-\mu_2^{(0)}){\psi}_2^{(1)} + ( g_{12}|{\psi}_1^{(0)}|^2 + g_{22}|{\psi}_2^{(0)}|^2 - \mu_2^{(1)} ){\psi}_2^{(0)}.
\end{eqnarray}
Solvability requires
\begin{eqnarray}
\label{eqn1_bifur}
0 &=& a( \mu_1^{(1)} - A g_{11}a^2 - B g_{12}b^2 ) , \\
\label{eqn2_bifur}
0 &=& b( \mu_2^{(1)} - B g_{12}a^2 - C g_{22}b^2 )
\end{eqnarray}
where
\[
A=A_m=\langle u_m^2,u_m^2\rangle,\quad B=B_{m,n}=\langle u_m^2,u_n^2\rangle,\quad C=C_n=\langle u_n^2,u_n^2\rangle.
\]
Note that $A,B,C>0$ are positive real numbers; in particular, for a few
values of the indices we have
\[
A_0=C_0=B_{0,0}=\sqrt{\frac{\Omega}{2\pi}},\,\,
A_1=C_1=B_{1,1}=\frac{3}{4}\sqrt{\frac{\Omega}{2\pi}},\,\,
A_2=C_2=B_{2,2}=\frac{41}{64}\sqrt{\frac{\Omega}{2\pi}},
\]
and
\[
B_{1,0}=B_{0,1}=\frac{1}{2}\sqrt{\frac{\Omega}{2\pi}},\,\,
B_{2,0}=B_{0,2}=\frac{3}{8}\sqrt{\frac{\Omega}{2\pi}},\,\,
B_{2,1}=B_{1,2}=\frac{7}{16}\sqrt{\frac{\Omega}{2\pi}}.
\]
Solving Eqs.~(\ref{eqn1_bifur})--(\ref{eqn2_bifur}) requires that for
nontrivial solutions the pair $(a, b)$ should satisfy one of the following:
\begin{enumerate}
\item $a=0$, and $\displaystyle{b^2=\frac{\mu_2^{(1)}}{C g_{22}}}$ for
$\mu_2^{(1)}>0$
\item $b=0$, and $\displaystyle{a^2=\frac{\mu_1^{(1)}}{A g_{11}}}$ for
$\mu_1^{(1)}> 0$
\item $a,b\neq 0$ and $\displaystyle{\left(
\begin{array}{cc}
A g_{11} & B g_{12}\\
B g_{12} & C g_{22}
\end{array}
\right)
\left(
\begin{array}{cc}
a^2\\
b^2
\end{array}
\right)=
\left(
\begin{array}{cc}
\mu_1^{(1)}\\
\mu_2^{(1)}
\end{array}
\right)}$.
\end{enumerate}
Both the first and second case correspond to effectively single-component
solutions, and will not be further considered in what follows except in a
parenthetical manner. Regarding the third case,
\[
AC g_{11}g_{22} - B^2 g_{12}^2\neq 0\,\,\leadsto\,\,
a^2=\frac{Cg_{22}\mu_1^{(1)}-Bg_{12}\mu_2^{(1)}}{AC
g_{11}g_{22} - B^2 g_{12}^2}>0 ,\quad
b^2=\frac{Ag_{11}\mu_2^{(1)}-Bg_{12}\mu_1^{(1)}}{AC g_{11}g_{22} - B^2
g_{12}^2}>0.
\]
On the other hand, if the coefficients $g_{11},g_{12},g_{22}$ are special
enough such that $A_m C_n g_{11}g_{22} - B_{m,n}^2 g_{12}^2=0$ for some
$(m,n)$, then the nontrivial two-component solutions for the state $(m,n)$
will exist only if
\[
\frac{\mu_1^{(1)}}{\mu_2^{(1)}}=\frac{A g_{11}}{B g_{12}}.
\]
Moreover, when those two-component solutions exist, $a$ and $b$ are not
uniquely determined by $\mu_1^{(1)}$ and $\mu_2^{(1)}$ as in the above but
there exists a family of available values for them. It is worthwhile to note
that this condition is reminiscent of the phase separation criterion between
two components and, in fact, coincides with the latter when
$m=n$~\cite{siambook,emergent}.
Nevertheless, we will not focus on this singular case here and in that
light,
in all that follows we
focus on the cases where $ab\neq 0$ and $AC g_{11}g_{22} - B^2 g_{12}^2\neq 0$, and without loss of generality assume $m\leq n$.
\section{Spectral stability}
If $\boldsymbol{{\psi}}$ is a steady-state solution to the system
(\ref{eqn0_1})--(\ref{eqn0_2}), then we consider the perturbation ansatz
${\tilde{\psi}}_j(x,t)={\psi}_j+\delta(e^{\lambda t}v_j(x)+e^{ \lambda^* t}
w_j^*(x))$ of such a solution. After substituting
$\boldsymbol{{\tilde{\psi}}}$ back into the system and linearizing around the
solution $\boldsymbol{{\psi}}$, we obtain the eigenvalue problem,
\begin{eqnarray}
\label{eqn_ev}
J \mathcal{L} \boldsymbol{\xi} = i\lambda\boldsymbol{\xi}\quad\leadsto\quad
(-iJ)\mathcal{L} \boldsymbol{\xi} = \lambda\boldsymbol{\xi},
\end{eqnarray}
where $J={\rm diag}(1,-1,1,-1)$ and $\boldsymbol{\xi}=(v_1,w_1,v_2,w_2)^T$.
The operator $-iJ$ is skew-symmetric, and the operator $\mathcal{L}$ is
self-adjoint. Consequently, this is a Hamiltonian eigenvalue problem. Because
the solutions are purely real, an important consequence is that the
eigenvalues satisfy the four-fold symmetry, $\{\pm\lambda,\pm\lambda^*\}$, which can also be explicitly seen from (\ref{e:L-property}).
Moreover, because of the unbounded potential term $\Omega^2x^2/2$ in the
operator $\mathcal{L}$, the spectrum is purely discrete, and each eigenvalue
has finite geometric and algebraic multiplicity.
\subsection{The unperturbed spectrum}
The spectrum for small $\epsilon$ will be determined via a perturbation
expansion from the $\epsilon=0$ spectrum. Consequently, it is important to
first have a detailed description of the unperturbed spectrum. For a given
eigenvalue, $\lambda$, let $E_\lambda$ denote the corresponding eigenspace.
It is straightforward to infer that given the quantum harmonic oscillator
nature of its constituents,
the eigenvalues of $(-iJ)\mathcal{L}^{(0)}$ are
$-i\ell\Omega,\,\ell\in\mathbb{Z}$. Due to the four-fold spectral symmetry we
can focus on the lower-half complex plane in the following. For each
nonnegative $\ell$ there are three possibilities:
\begin{enumerate}
\item if $\ell\leq m$, then for $\lambda^{(0)}=-i\ell\Omega$,
\[
E_{-i\ell\Omega}=\mathrm{Span}\left\{\left(\begin{array}{c}u_{m+\ell}\\0\\0\\0\end{array}\right),
\left(\begin{array}{c}0\\u_{m-\ell}\\0\\0\end{array}\right),
\left(\begin{array}{c}0\\0\\u_{n+\ell}\\0\end{array}\right),
\left(\begin{array}{c}0\\0\\0\\u_{n-\ell}\end{array}\right)\right\}
\]
\item if $m<\ell\leq n$, then for $\lambda^{(0)}=-i\ell\Omega$,
\[
E_{-i\ell\Omega}=\mathrm{Span}\left\{\left(\begin{array}{c}u_{m+\ell}\\0\\0\\0\end{array}\right),
\left(\begin{array}{c}0\\0\\u_{n+\ell}\\0\end{array}\right),
\left(\begin{array}{c}0\\0\\0\\u_{n-\ell}\end{array}\right)\right\}
\]
\item If $\ell> n$, then for $\lambda^{(0)}=-i\ell\Omega$,
\[
E_{-i\ell\Omega}=\mathrm{Span}\left\{\left(\begin{array}{c}u_{m+\ell}\\0\\0\\0\end{array}\right),
\left(\begin{array}{c}0\\0\\u_{n+\ell}\\0\end{array}\right)\right\}.
\]
\end{enumerate}
The kernel has dimension four.
\subsection{Krein Signature and Hamiltonian-Krein Index}
The spectrum of $(-iJ)\mathcal{L}$ is completely known for the unperturbed
problem. In particular, it is purely imaginary, so that the unperturbed wave
is spectrally stable. Because of the four-fold symmetry, eigenvalues which
are simple will remain purely imaginary for small $\epsilon$. However, as we
see above the unperturbed eigenvalues are semi-simple, which implies that
some could gain a nontrivial real part upon perturbation. Our first goal is
to show via the Hamiltonian-Krein index (HKI) that all but a finite number of
the eigenvalues will remain purely imaginary under small perturbation.
Moreover, the index will precisely locate which among the infinitely many
eigenvalues can gain nonzero real part under perturbation. See
~\cite{toddbook} for a more detailed exposition of what follows.
For the operator $(-iJ)\mathcal{L}$ let $k_\mathrm{r}$ denote the total number of
real positive eigenvalues (counting multiplicity), and $k_\mathrm{c}$ the total
number of eigenvalues with positive real part and nonzero imaginary part
(counting multiplicity). Regarding the purely imaginary eigenvalues, let
$\lambda$ be a purely imaginary eigenvalue with finite multiplicity, and let
$E_\lambda$ denote the associated eigenspace. The negative Krein index
associated with $E_\lambda$ is
$k_\mathrm{i}^-(\lambda)=\mathrm{n}(\mathcal{L}|_{E_\lambda})$. Here $\mathrm{n}(S)$ denotes
the number of negative eigenvalues (counting multiplicity) associated with a
Hermitian matrix $S$, and $\mathcal{L}|_{E_\lambda}$ denotes the Hermitian
matrix induced by restricting $\mathcal{L}$ to operate on $E_\lambda$. If
$\lambda$ is a simple eigenvalue with associated eigenvector
$\boldsymbol{\xi}$, then $\mathrm{n}(\mathcal{L}|_{E_\lambda})=\mathrm{n}(\langle
\boldsymbol{\xi}, \mathcal{L}\boldsymbol{\xi} \rangle)$. The eigenvalue is
said to have positive Krein signature if $\mathrm{n}(\mathcal{L}|_{E_\lambda})=0$;
otherwise, it is said to have negative Krein signature. Let $k_\mathrm{i}^-$ denote
the total negative Krein index,
\[
k_\mathrm{i}^-=\sum_{\lambda\in\sigma(-(iJ)\mathcal{L})\cap i\mathbb{R}}k_\mathrm{i}^-(\lambda).
\]
The HKI is the sum of all three indices,
\[
K_{\mathrm{Ham}}=k_\mathrm{r}+k_\mathrm{c}+k_\mathrm{i}^-.
\]
Because of the four-fold eigenvalue symmetry, $k_\mathrm{c}$ and $k_\mathrm{i}^-$ will be
even integers. In particular, there will be precisely $k_\mathrm{c}/2$ eigenvalues
with positive real part and negative imaginary part, and $k_\mathrm{i}^-/2$ purely
imaginary eigenvalues with negative imaginary part and negative Krein index.
The negative Krein index can be easily computed for the unperturbed problem.
Again, we focus only on those eigenvalues with negative imaginary part. Using
the diagonal form of $\mathcal{L}^{(0)}$, and the bases for the spectral
subspaces given in the previous subsection, we find
\begin{enumerate}
\item if $\lambda=-i\ell\Omega$ with $0<\ell\le m$, then
$k_\mathrm{i}^-(-i\ell\Omega)=2$
\item if $\lambda=-i\ell\Omega$ with $m<\ell\le n$, then
$k_\mathrm{i}^-(-i\ell\Omega)=1$
\item if $\lambda=-i\ell\Omega$ with $n<\ell$, then
$k_\mathrm{i}^-(-i\ell\Omega)=0$.
\end{enumerate}
The four-fold symmetry implies that the eigenvalues with positive imaginary
part satisfy $k_i^-(i \ell \Omega)=k_i^-(-i \ell \Omega)$ for any $l\in \mathbb{N}$.
Consequently, the total negative Krein index is
\[
k_\mathrm{i}^-=4m+2(n-m)=2(m+n),
\]
so the HKI for the unperturbed problem is
\[
K_{\mathrm{Ham}}=2(m+n).
\]
Half of these eigenvalues have negative imaginary part, and half have
positive imaginary part.
Since the index is integer-valued, for operators which depend continuously on
parameters it remains unchanged for small perturbations. This statement,
however, requires that no additional eigenvalues can be added into the mix
via a bifurcation from the origin. Recall that we consider only those waves
which are nontrivial in both components. The gauge symmetry implies that the
geometric multiplicity of the origin will always be minimally two, and the
Hamiltonian structure of the spectral problem means the algebraic
multiplicity will always then be minimally four. For the unperturbed problem
the algebraic multiplicity of the origin is precisely four. Since the origin
is isolated, this then implies that for small perturbations the multiplicity
will remain four. Consequently, we know that for small $\epsilon$,
\[
K_{\mathrm{Ham}}=2(m+n),
\]
and for those eigenvalues associated with the HKI having nonzero imaginary
part, half will have positive imaginary part, and half will have negative
imaginary part.
The HKI provides for an upper bound of the number of eigenvalues with
positive real part. In order to locate those eigenvalues with small positive
real part for the perturbed problem, we do a perturbation expansion. However,
it is not necessary for us to perform an expansion for each eigenvalue.
Purely imaginary eigenvalues can leave the imaginary axis only via the
collision of eigenvalues of opposite Krein signature. This implies that for
the perturbation expansion we only need to consider those eigenvalues for which
the induced matrix $\mathcal{L}^{(0)}|_{E_{-i\ell\Omega}}$ is indefinite.
Restricting to those eigenvalues with negative imaginary part, this means we
only have $\ell>0$. If $0<\ell\le m$, the facts that
$\mathrm{dim}[E_{-i\ell\Omega}]=4$ and $k_\mathrm{i}^-(-i\ell\Omega)=2$ imply
that at most two eigenvalues can be created with positive real part
(collision of a pair of eigenvalues with negative Krein signature with a pair
with positive Krein signature). If $m<\ell\le n$, the facts that
$\mathrm{dim}[E_{-i\ell\Omega}]=3$ and $k_\mathrm{i}^-(-i\ell\Omega)=1$ imply
that at most one eigenvalue can be created with positive real part (collision
of one eigenvalue with negative Krein signature with one with positive
Krein signature). Finally, if $\ell>n$ then the unperturbed eigenvalue has
positive Krein signature, and will consequently remain purely imaginary under
small perturbation. In conclusion, when performing the perturbation expansion
we need only start with those unperturbed eigenvalues with $0<\ell\le n$.
\subsection{Reduced Eigenvalue Problem}
Knowing the Hamiltonian-Krein index, we will examine the exact number of
eigenvalue pairs with nonzero growth rates (i.e., associated with
instabilities) by finding the leading-order correction to each eigenvalue.
Since the eigenvalues are semi-simple, and the underlying solution is smooth
in $\epsilon$, the eigenvalues and associated eigenfunctions have the
expansions,
\[
\lambda=\lambda^{(0)}+\epsilon \lambda^{(1)}+O(\epsilon^2),\quad
\boldsymbol{\xi}=\boldsymbol{\xi}^{(0)}+\epsilon \boldsymbol{\xi}^{(1)}+O(\epsilon^2).
\]
The $O(1)$ and $O(\epsilon)$ reductions of Eqn.~(\ref{eqn_ev}) as
\begin{eqnarray}
\label{eqn_ev0}
\left(\mathcal{L}^{(0)}-i\lambda^{(0)}J\right)\boldsymbol{\xi}^{(0)} &=& \boldsymbol{0}\\
\label{eqn_ev1}
\left(\mathcal{L}^{(0)}-i\lambda^{(0)}J\right)\boldsymbol{\xi}^{(1)}&=&
\left(i\lambda^{(1)}J-\mathcal{L}^{(1)}\right)\boldsymbol{\xi}^{(0)}.
\end{eqnarray}
Letting $\{\boldsymbol{\eta}_i\}$ be an orthonormal basis for
$E_{-i\ell\Omega}$, upon writing $\boldsymbol{\xi}^{(0)}=\sum
c_i\boldsymbol{\eta}_i$ the solvability condition for (\ref{eqn_ev1}) is the
reduced spectral problem,
\begin{eqnarray}
\label{eqn_ev_reduced}
M \boldsymbol{c} = i\lambda^{(1)}\boldsymbol{c}.
\end{eqnarray}
Here
\[
M_{j,k}=\langle \boldsymbol{\eta}_j, J\mathcal{L}^{(1)} \boldsymbol{\eta}_k \rangle,
\]
where the inner-product on each component is the standard one for
$L^2(\mathbb{R})$.
Similar calculations have been presented in different examples (chiefly for
single component systems); see for one such example, e.g.,~\cite{todd3}. Note
that if the spectrum of $M$ is purely real, then to leading order the
eigenvalues will be purely imaginary. On the other hand, eigenvalues of $M$
which have nonzero imaginary part lead to an oscillatory instability for the
underlying wave.
For the expansion we need only consider those eigenvalues with
$\lambda^{(0)}=-i\ell\Omega$ for $0<\ell\le n$. The perturbed eigenvalues for
$\ell>n$ will remain purely imaginary. The size of the matrix $M$ will depend
upon the value of $\ell$; in particular, if $0<\ell\le m$, then
$M\in\mathcal{M}_{4\times 4}(\mathbb{R})$, while if $m<\ell\le n$, then
$M\in\mathcal{M}_{3\times 3}(\mathbb{R})$. Defining
\[
D_{p,q,r,s}:=\langle u_pu_q,u_ru_s\rangle,
\]
the explicit expression for $M$ is:
\begin{itemize}
\item [(a)] $\lambda^{(0)}=-i\ell\Omega$ with $0<\ell\leq m$, then $M=M_a$,
where
\[
M_a=\left(
\begin{array}{cc}
\tilde{M}_{11} & \tilde{M}_{12}\\
\tilde{M}_{21} & \tilde{M}_{22}
\end{array}\right),
\]
and the individual blocks are defined via
\[
\tilde{M}_{11}=\left(
\begin{array}{cc}
2 g_{11} a^2 B_{m, m+\ell}+g_{12}b^2 B_{n,m+\ell}-\mu_1^{(1)} & g_{11}a^2 D_{m,m,m+\ell,m-\ell} \\
-g_{11}a^2 D_{m,m,m+\ell,m-\ell} & -(2 g_{11} a^2 B_{m, m-\ell}+g_{12}b^2 B_{n,m-\ell})+\mu_1^{(1)}
\end{array}
\right),
\]
and
\[
\tilde{M}_{12}=g_{12}a b\left(
\begin{array}{cc}
D_{m,n,m+\ell,n+\ell} & D_{m,n,m+\ell,n-\ell} \\
-D_{m,n,m-\ell,n+\ell} & -D_{m,n,m-\ell,n-\ell}
\end{array}
\right),\quad\tilde{M}_{21}=g_{12}a b\left(
\begin{array}{cc}
D_{m,n,m+\ell,n+\ell} & D_{m,n,m-\ell,n+\ell} \\
-D_{m,n,m+\ell,n-\ell} & -D_{m,n,m-\ell,n-\ell}
\end{array}
\right),
\]
and
\[
\tilde{M}_{22}=\left(
\begin{array}{cc}
2 g_{22} b^2 B_{n, n+\ell}+g_{12}a^2 B_{m,n+\ell}-\mu_2^{(1)} & g_{22}b^2 D_{n,n,n+\ell,n-\ell} \\
-g_{22}b^2 D_{n,n,n+\ell,n-\ell} & -(2 g_{22} b^2 B_{n, n-\ell}+g_{12}a^2 B_{m,n-\ell})+\mu_2^{(1)}
\end{array}
\right)
\]
\item [(b)] If $\lambda^{(0)}=-i\ell\Omega$ where $m<\ell\leq n$, then
$M=M_b$, and $M_b$ is simply the submatrix obtained from $M_a$ after
removing the second row and second column.
\end{itemize}
Before continuing, we briefly comment on what the above perturbation
calculation says about the spectral stability of one-component solutions. If
$a=0$, then
\[
\tilde{M}_{11}= \left(
\begin{array}{cc}
g_{12}b^2 B_{m+\ell,n}-\mu_1^{(1)} & 0 \\
0 & -g_{12}b^2 B_{m-\ell,n}+\mu_1^{(1)}
\end{array}
\right),\quad
\tilde{M}_{22}=g_{22} b^2 \left(
\begin{array}{cc}
2 B_{n,n+\ell}-C & D_{n,n,n-\ell,n+\ell} \\
-D_{n,n,n-\ell,n+\ell} & C-2 B_{n-\ell,n}
\end{array}
\right),
\]
and $\tilde{M}_{12}=\tilde{M}_{21}=\boldsymbol{0}$. Thus, $M_a$ can have
complex eigenvalues only if $|B_{n-\ell,n}+B_{n,n+\ell}-C| < |D_{n,n,n-\ell,n+\ell}| $.
Similarly, when $b=0,\,M_a$ will have a complex spectrum only if
$|B_{m-\ell,m}+B_{m,m+\ell}-A| < |D_{m,m,m-\ell,m+\ell}|$.
When $a=0$, the condition for $M_b$ to have complex eigenvalues will be the same as that for $M_a$. If $b=0$, $M_b$ will simply become a diagonal matrix and always have a real spectrum.
Therefore, for a one-component solution where $(\psi_1^{(0)},\psi_2^{(0)})=(a
u_m, 0)$, its spectral stability can be examined by checking conditions
$|B_{m-\ell,m}+B_{m,m+\ell}-A_m| < |D_{m,m,m-\ell,m+\ell}|$ for $0<\ell\leq m$. Here we note
that this is the same stability result if we consider a single one-component
equation. For $m\leq 2$ we can directly check the stability conditions for
the one-component solutions to get:
\begin{enumerate}
\item if $m=0$, the solutions continued from $a u_0$ are spectrally stable
for small $\epsilon$;
\item if $m=1$, it can be checked that
$|B_{0,1}+B_{1,2}-A_1|=|1/2+7/16-3/4|
> |\sqrt{2}/8|=|D_{0,1,1,2}|$, so the solutions continued from $a
u_1$ are spectrally stable for small $\epsilon$;
\item if $m=2$, it can be checked that
$|B_{1,2}+B_{2,3}-A_2|=|7/16+51/128-41/64|
> |5\sqrt{6}/64|=|D_{1,2,2,3}|$ but
$|B_{0,2}+B_{2,4}-A_2|=|3/8+329/1024-41/64| <
|3\sqrt{6}/128|=|D_{0,2,2,4}|$, so the splitting of eigenvalues at
$2\Omega$ will enter the complex plane and the solutions continued from
$a u_2$ are spectrally unstable (these stability features are well
known, e.g., from the work of~\cite{coles}).
\end{enumerate}
In fact, in Section~\ref{section:summary-stability} we show that, in general,
\[
\frac{|B_{m-1,m}+B_{m,m+1}-A_m|}{|D_{m,m,m-1,m+1}|} = \sqrt{\frac{m}{m+1}}
+\sqrt{\frac{m+1}{m}} >2,\quad m\geq 1.
\]
That is to say, the eigenvalues for one-component solutions near $-i\Omega$ will always stay on the imaginary axis, although this perturbation calculation itself doesn't rule out the possibility for other eigenvalues to enter the complex plane.
\section{Catalogue of Different $(m,n)$ Cases}
\label{section:stability-catologue}
We now use the theory of the previous section to compute the spectral
stability of various two-component solutions. In particular, we will assume
$0\le m\le n\le2$.
{In what follows, the different branches are presented for $\Omega=0.1$, although similar results have been obtained for other values of $\Omega$. In fact, the value of $\Omega$ does not have a significant bearing
on the agreement between analytical predictions and computational results (including in the more physically realistic case of $\Omega\ll 1$).}
\subsection{$(m,n)=(0,0)$}
For $\mu_1^{(0)}=\mu_2^{(0)}=\Omega/2$, we consider the branches of
solutions continued from $({\psi}_1^{(0)}, {\psi}_2^{(0)})=(a u_{0}, b
u_{0})$, where
\[
a^2=\sqrt{\frac{2\pi}{\Omega}}\,
\frac{g_{22}\mu_1^{(1)}-g_{12}\mu_2^{(1)}}{g_{11}g_{22}-g_{12}^2},\quad
b^2=\sqrt{\frac{2\pi}{\Omega}}\,
\frac{g_{11}\mu_2^{(1)}-g_{12}\mu_1^{(1)}}{g_{11}g_{22}-g_{12}^2}.
\]
This state features the fundamental (ground state) waveform in both
components of the system. Since $K_{\mathrm{Ham}}=0$ when $\epsilon=0$, the
wave is spectrally (indeed, orbitally) stable for small $\epsilon$, and it
is not necessary to perform the perturbation calculation.
\subsection{$(m,n)=(0,1)$ (interchange all subscripts to obtain case $(1,0)$)}
If $\mu_1^{(0)}=\Omega/2$ and $\mu_2^{(0)}=3\Omega/2$, we consider the
continuation of $({\psi}_1^{(0)}, {\psi}_2^{(0)})=(a u_{0}, b u_{1})$, where
\[
a^2=\sqrt{\frac{2\pi}{\Omega}}\,\frac{3g_{22}\mu_1^{(1)}-2g_{12}\mu_2^{(1)}}{3g_{11}g_{22}-g_{12}^2},\quad
b^2=2\sqrt{\frac{2\pi}{\Omega}}\,\frac{2g_{11}\mu_2^{(1)}-g_{12}\mu_1^{(1)}}{3 g_{11}g_{22} - g_{12}^2},
\]
which corresponds to a ``dark-bright" configuration. This configuration has been extensively studied in experiments over the past decade, as has been
recently summarized e.g.
in~\cite{revip}
Regarding the spectral stability we have $K_{\mathrm{Ham}}=2$, with the
dangerous eigenvalues at $\lambda^{(0)}=\pm i\Omega$. At most one eigenvalue
with positive real part will emerge from $-i\Omega$. For the perturbation
calculation we only need consider case (b), where
$M_b\in\mathcal{M}_{3\times3}(\mathbb{R})$ is
\begin{equation}\label{e:matrix-b-0-1}
M_b=\frac{1}{8}\sqrt{\frac{\Omega}{2\pi}}\left(
\begin{array}{ccc}
2 g_{12}b^2 & \sqrt{2}g_{12}a b & 4 g_{12}a b \\
\sqrt{2} g_{12}a b & g_{22}b^2 - g_{12}a^2 & \sqrt{2} g_{22}b^2 \\
-4 g_{12}a b & -\sqrt{2} g_{22}b^2 & -2 g_{22}b^2 - 4 g_{12}a^2
\end{array}
\right).
\end{equation}
Regarding the spectrum of $M_b$, we have the following proposition:
\begin{proposition}\label{proposition-0-1}
$M_b$ for $(m,n)=(0,1)$ (i.e. the matrix in (\ref{e:matrix-b-0-1})) has an eigenvalue zero with associated
eigenvector $(-a/b,-\sqrt{2},1)^\mathrm{T}$, and two other eigenvalues
\[
-\frac{1}{16}\sqrt{\frac{\Omega}{2\pi}}\left(5 a^2 g_{12} - 2 b^2 g_{12} +
b^2 g_{22}\pm \sqrt{9 a^4 g_{12}^2 + 18 a^2 b^2 g_{12}(g_{22}-2
g_{12})+b^4(2 g_{12}+g_{22})^2}\right).
\]
The eigenvalues of $M_b$ will have nonzero imaginary parts if and only if
$g_{12}>g_{22}$.
\end{proposition}
The presence of the zero eigenvalue is well-known to be associated with the
invariance of the condensate to dipolar oscillations with the frequency of
the trap $\Omega$, yielding the so-called Kohn mode in the spectrum with the
trap frequency (and hence vanishing perturbations off of the linear
limit)~\cite{stringari}.
This proposition can be verified via direct calculation and in
Section~\ref{section:summary-stability} we will state more general results.
It is intriguing that the expression for the nonzero eigenvalues for $M_b$
here does not include $g_{11}$. This is due to the fact that
$A_0-2B_{0,1}=0$. As stated in Proposition~\ref{proposition-0-1}, $M_b$ will
have eigenvalues with nonzero imaginary parts --leading to an instability--
if $g_{12}>g_{22}$, i.e., the inter-component interactions have to be
stronger than the interactions within the ``dark'' species.
As an example, when $a=b=1$ the growth rate is
\[
|\mathrm{Re}(\lambda^{(1)})|=
\frac{1}{16}\sqrt{\frac{\Omega}{2\pi}(23g_{12}+g_{22})(g_{12}-g_{22})}\,I_{g_{12}>g_{22}}(g_{12}),
\]
where
\[
I_{g_{12}>g_{22}}(g_{12})=\begin{cases}
0,\quad&g_{12}\le g_{22}\\
1,\quad&g_{12}> g_{22}.
\end{cases}
\]
In Figure \ref{fig1_2}, a case associated with this potential instability
scenario of the $(0,1)$ branch is shown. In particular, the maximal real part
of numerically computed eigenvalues from Eqn.~(\ref{eqn_ev}) is plotted with
respected to $g_{12}$ and $\epsilon$. We see that the numerical result is in
good agreement with with our prediction $\epsilon|\mathrm{Re}(\lambda^{(1)})|$. It is
relevant to indicate that in the integrable limit of $g_{ij}=1$, this
instability does not manifest itself, but it should be observable in systems
away from this limit provided that the first excited (dark) state is
initialized in the ``wrong'' component i.e., the one with intra-component
interactions $g_{22}<g_{12}$, while the fundamental state is initialized in
the component with $g_{11}$.
\begin{figure}[!htbp]
\centering
\begin{tabular}{ccc}
\includegraphics[width=5cm]{g12_eps_lambda_re_numerical_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{g12_eps01_lambda_re_numerical_predict_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{g12_15_eps_lambda_re_numerical_predict_new-eps-converted-to.pdf}
\end{tabular}
\caption{Stability features of the $(0,1)$ branch:
in the left panel, we set $g_{22}=1.06, a=b=1$ and plot the maximal real part of numerically computed eigenvalues as a function of $g_{12}$ and $\epsilon$. In the middle (right) panel, we fix $\epsilon=0.1$ ($g_{12}=1.5$) for the setups in the left panel and compare the numerical result with analytical prediction $\epsilon|\mathrm{Re}(\lambda^{(1)})|$, as a function of
$g_{12}$ ($\epsilon$). We find very good agreement in the dependence
of the relevant eigenvalue.}
\label{fig1_2}
\end{figure}
In Fig.~\ref{fig1_2_1}, we present the example with $a=b=1, g_{11}=1.03,
g_{12}=1.04, g_{22}=1.06, \Omega=0.1$, comparing eigenvalue predictions with
corrections up to $O(\epsilon)$ with corresponding numerical results.
We find that all of the
eigenvalues in the numerical computation are on the imaginary axis, which
matches our analytical prediction. According to our numerical computation, the spectrum will remain purely imaginary even when $\epsilon$ is large, which is shown in the left panel of Fig.~\ref{fig1_2_1b}. In addition, we find that $\phi_2$ becomes $0$ at
$\epsilon\approx 2.6$ where the branch of solutions meets the branch of
single-component solutions on $\phi_1$.
\begin{figure}[!htbp]
\begin{tabular}{ccc}
\includegraphics[width=5cm]{compare_01_1_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_01_2_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_01_u12_density_new-eps-converted-to.pdf}
\end{tabular}
\caption{The left (middle) panel shows the imaginary parts of the eigenvalues
around $-i\Omega$ ($-i2\Omega$) as functions of $\epsilon$ with $O(\epsilon)$ corrections
(solid lines) and corresponding numerical results (dashed lines with circles)
in the case of $(m,n)=(0,1)$, i.e., a prototypical example of a dark state with $n=1$ coupled to
a fundamental state of $m=0$. In the right panel, we show the densities of $\phi_1$ and $\phi_2$ at $\epsilon=0.1$.}
\label{fig1_2_1}
\end{figure}
\begin{figure}[!htbp]
\begin{tabular}{cc}
\includegraphics[width=7cm]{ev_01_1_re_new-eps-converted-to.pdf}
\includegraphics[width=7cm]{compare_01_u12_square_long_new-eps-converted-to.pdf}
\end{tabular}
\caption{In the left panel, we plot the imaginary parts of the eigenvalues over a wide interval of parametric variation of $\epsilon$. The right panel shows the $L^2$-norm of the corresponding solution of $\phi_1$ and $\phi_2$ for such interval of $\epsilon$.}
\label{fig1_2_1b}
\end{figure}
As another example, we consider a case that is ``immediately unstable''
in the vicinity of the linear limit. In particular,
if $a=b=1, g_{11}=1.03, g_{12}=1.2, g_{22}=1.06, \Omega=0.1$,
the numerical computation shows that all of the eigenvalues except a quartet
(near $\pm i\Omega$) are purely imaginary, as shown in Fig.~\ref{fig1_2_2}.
As $\epsilon$ increases, we notice that the quartet will finally come back to
the real axis at $\epsilon\approx 0.7$ and split along it, which is shown in
Figure \ref{fig1_2_3}.
\begin{figure}[!htbp]
\begin{tabular}{ccc}
\includegraphics[width=5cm]{compare_01_b_1_re_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_01_b_2_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_01_b_1_im_new-eps-converted-to.pdf}
\end{tabular}
\caption{ The left (middle) panel shows the imaginary parts of the eigenvalues around $-i\Omega$ ($-i2\Omega$) as functions of $\epsilon$ with $O(\epsilon)$ corrections (solid lines) and corresponding numerical results (dashed lines with circles). In the left panel, the red lines and green lines (solid and dashed) are
identical since a pair of eigenvalues of $M_b$ for $\lambda^{(0)}=-i\Omega$ are complex conjugates. Moreover, the nonzero imaginary parts
of $i \lambda$ for this pair imply the instability (this is the only source of the instability) of the solution, as shown in the right panel
(solid line for the $O(\epsilon)$ correction using this pair of complex conjugates and the dashed line for the numerical computation of the real parts of the
relevant eigenvalues).}
\label{fig1_2_2}
\end{figure}
\begin{figure}[!htbp]
\begin{tabular}{ccc}
\includegraphics[width=5cm]{ev_01_2_re_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{ev_01_2_im_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_01_u12_square_long_2_new-eps-converted-to.pdf}
\end{tabular}
\caption{ The left (middle) panel shows the variation of the
imaginary (real) parts of the eigenvalues for the branch of solutions with $a=b=1$, $g_{11}=1.03$, $g_{12}=1.2$, $g_{22}=1.06$, $\Omega=0.5$. In the right panel, we plot the $L^2$-norm of the solution of $\phi_j$ over the same interval of $\epsilon$.}
\label{fig1_2_3}
\end{figure}
Figure~\ref{fig1_2_4} illustrates the numerical evolution of the unstable configuration shown in Figure \ref{fig1_2_2} with $\epsilon=0.1$. If a small initial perturbation is added to the solution, the development of the instability can be observed over intermediate time scales in Figure~\ref{fig1_2_4}.
To determine the fate of the solution under the action of this
instability, we have performed considerably longer simulations focusing
on the dynamics of the unstable waveform (see the middle panels
of Figure~\ref{fig1_2_4}).
{There we find an oscillatory pattern of the long-term dynamics of the solution,
featuring breathing (yet not genuinely periodic) recurrences over time. To be more specific, the bottom panels in Figure~\ref{fig1_2_4}
reveal, through a dynamical
decomposition to the lowest order harmonic oscillator
modes, that the system does not stay at a certain state but quantitatively alternates between the states $(0,1)$ and $(1,0)$. We also observe that the instability of the state $(0,1)$ (similar for the state $(1,0)$) is essentially caused by the eigenmodes that are related to the unstable eigenvalues near $\pm i\Omega$, which is verified by the fact that both the time evolution of $|c_1|$ for $\phi_1$ and that of $|c_0|$ for $\phi_2$ bear small oscillations whose frequency is close to $\frac{2\pi}{\Omega}$.}
\begin{figure}[!htbp]
\begin{tabular}{cc}
\includegraphics[width=7cm]{dynamics_01_phi1_g12_shorter_new-eps-converted-to.pdf}
\includegraphics[width=7cm]{dynamics_01_phi2_g12_shorter_new-eps-converted-to.pdf}\\
\includegraphics[width=7cm]{dynamics_01_phi1_g12_longer_2_reduced-eps-converted-to.pdf}
\includegraphics[width=7cm]{dynamics_01_phi2_g12_longer_2_reduced-eps-converted-to.pdf}\\
\includegraphics[width=7cm]{dynamics_01_phi1_g12_longer_2_eigenmodes-eps-converted-to.pdf}
\includegraphics[width=7cm]{dynamics_01_phi2_g12_longer_2_eigenmodes-eps-converted-to.pdf}
\end{tabular}
\caption{ The top left (right) panel illustrates an example of the dynamics of $|\phi_1|^2$ ($|\phi_2|^2$) with $a=b=1$, $g_{11}=1.03$, $g_{12}=1.2$, $g_{22}=1.06$, $\Omega=0.1$ and $\epsilon=0.1$ for the $(0,1)$ state. The middle panels show the dynamics for the same setup for a longer time, revealing its oscillatory
nature, while the bottom panels (left for $\phi_1$ and right for $\phi_2$) show the dynamics upon decomposition to the orthonormal basis $\{u_k\}$, where
the diagnostic $c_k$ used
stands for the coefficient (i.e., prefactor) of $u_k$ in the decomposition.}
\label{fig1_2_4}
\end{figure}
\subsection{$(m,n)=(1,1)$}
When $\mu_1^{(0)}=\mu_2^{(0)}=3\Omega/2$, we consider $({\psi}_1^{(0)},
{\psi}_2^{(0)})=(a u_{1}, b u_{1})$, where
\[
a^2=\frac{4}{3}\sqrt{\frac{2\pi}{\Omega}}\,
\frac{g_{22}\mu_1^{(1)}-g_{12}\mu_2^{(1)}}{g_{11}g_{22}-g_{12}^2},\quad
b^2=\frac{4}{3}\sqrt{\frac{2\pi}{\Omega}}\,
\frac{g_{11}\mu_2^{(1)}-g_{12}\mu_1^{(1)}}{g_{11}g_{22}-g_{12}^2}.
\]
This state corresponds to a (co-located) dark-dark type configuration
featuring a first excited state in both components.
Regarding the spectral stability we have $K_{\mathrm{Ham}}=4$, with the
dangerous eigenvalues again at $\lambda^{(0)}=\pm i\Omega$. At most two
eigenvalues with positive real part will emerge from $-i\Omega$. For the
perturbation calculation we need to consider case (a), where
$M_a\in\mathcal{M}_{4\times4}(\mathbb{R})$ is
\[
M_a=\frac{1}{16}\sqrt{\frac{\Omega}{2\pi}}\left(
\begin{array}{cccc}
2 g_{11}a^2 -5 g_{12}b^2 & 2\sqrt{2} g_{11} a^2 & 7 g_{12}a b & 2\sqrt{2}g_{12}a b \\
-2\sqrt{2} g_{11} a^2 & -4 g_{11}a^2 +4 g_{12}b^2 & -2\sqrt{2}g_{12}a b & -8 g_{12}a b\\
7 g_{12}a b & 2\sqrt{2}g_{12}a b & 2 g_{22}b^2 -5 g_{12}a^2 & 2\sqrt{2} g_{22} b^2\\
-2\sqrt{2}g_{12}a b & -8 g_{12}a b & -2\sqrt{2} g_{22} b^2 & -4 g_{22}b^2 +4 g_{12}a^2
\end{array}
\right).
\]
Examining the spectrum of $M_a$, we find that one eigenvalue is zero with
associated eigenvector $(-\sqrt{2}\,a/b,a/b,-\sqrt{2},1)^\mathrm{T}$; once
again, this is associated with the invariance to dipolar oscillations with
the frap frequency. Since the matrix is real-valued, this then implies that
there is at most one pair of eigenvalues with nonzero imaginary part. This is
an important conclusion that is particular to the case of the parabolic trap:
the presence of the well-known symmetry associated with the dipolar
oscillations~\cite{stringari} does not allow in this case the broader
spectrum of two potentially unstable eigendirections to lead to
instabilities; instead, only such instability direction may be realized in
practice.
For a particular example, if we fix $a=b=1$ and $g_{11}=g_{22}=1$, then the
remaining three eigenvalues of $M_a$ are
\[
\sqrt{\frac{\Omega}{2\pi}}\,\frac{1+g_{12}}{8},\quad
\sqrt{\frac{\Omega}{2\pi}}\,\frac{-1\pm\sqrt{1-56
g_{12}+136 g_{12}^2}}{16}
\]
Under these specific parameter values, there is one pair of eigenvalues that can
enter the complex plane for
\[
g_{12}\in\left(\frac{14-9\sqrt{2}}{68},\frac{14+9\sqrt{2}}{68}\right) \approx (0.0187,
0.3931).
\]
Again, setting $a=b=1, g_{11}=1.03, g_{12}=1.04, g_{22}=1.06, \Omega=0.1$ we
compare some predicted eigenvalues up to $O(\epsilon)$ with corresponding
numerical eigenvalues in Fig.~\ref{fig1_3}. All of the numerically computed
eigenvalues from Eqn.~(\ref{eqn_ev}) for this example are imaginary, which
matches the analytical result of the reduced spectral eigenvalue problem from
Eqn.~(\ref{eqn_ev_reduced}). As $\epsilon$ becomes large, we see a pair of
eigenvalues enter the complex plane near $\pm i\Omega$ at $\epsilon\approx
4.9$ when the eigenvalue from $i \Omega$ collides with the eigenvalue from
$3i\Omega$ (see Figure {\ref{fig1_3_2}}). We also note that this complex pair
will come back to the imaginary axis at $\epsilon\approx 5.2$, i.e., the
parametric interval of instability is fairly narrow in this case.
\begin{figure}[!htbp]
\begin{tabular}{ccc}
\includegraphics[width=5cm]{compare_11_1_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_11_2_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_11_u12_density_new-eps-converted-to.pdf}
\end{tabular}
\caption{ Case of $a=b=1, g_{11}=1.03, g_{12}=1.04, g_{22}=1.06, \Omega=0.1$
for the $(1,1)$ branch:
The left (middle) panel shows the imaginary parts of the eigenvalues
around $-i\Omega$ ($- 2i\Omega$) as functions of $\epsilon$ with $O(\epsilon)$ corrections (solid lines) and corresponding numerical results
(dashed lines with circles). The right panel shows the densities of $\phi_1$ and $\phi_2$ at $\epsilon=0.1$.}
\label{fig1_3}
\end{figure}
\begin{figure}[!htbp]
\begin{tabular}{ccc}
\includegraphics[width=5cm]{ev_11_1_re_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{ev_11_1_im_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_11_u12_square_long_new-eps-converted-to.pdf}
\end{tabular}
\caption{ The left (middle) panel shows the change of imaginary (real) parts of the eigenvalues for the branch of solutions with $a=b=1, g_{11}=1.03, g_{12}=1.04, g_{22}=1.06, \Omega=0.1$ for the $(1,1)$ branch. In the right panel, we plot the $L^2$-norm of the solution of $\phi_j$ as a function of $\epsilon$.}
\label{fig1_3_2}
\end{figure}
If $a=b=1, g_{11}=1, g_{12}=0.25, g_{22}=1, \Omega=0.1$, the numerical
computation shows that all of the eigenvalues except a quartet (near $\pm
i\Omega$) are on the imaginary axis, as shown in Fig.~\ref{fig1_3_3}. As
$\epsilon$ grows, the complex pair of eigenvalues from $-i\Omega$ will return
to the real axis and split into two, as seen in Figure {\ref{fig1_3_4}}. The
split eigenvalue going upward will meet with the eigenvalue coming down from
$-i3\Omega$ and produce another pair (quartet) of complex eigenvalues, which
will go back to the imaginary axis and split again. One of the split
eigenvalue will move upward and collide with the eigenvalue from $-i5\Omega$,
which will again lead to complex eigenvalues, and so on. In
Figure~\ref{fig1_3_5}, we illustrate the numerical evolution of this unstable
configuration for $\epsilon=0.1$. It can be seen that the weak (and clearly
discerned to be oscillatory) nature of the instability only allows it to
manifest over fairly long time scales, resulting in breathing dynamics.
\begin{figure}[!htbp]
\begin{tabular}{ccc}
\includegraphics[width=5cm]{compare_11_b_1_re_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_11_b_2_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_11_b_1_im_new-eps-converted-to.pdf}
\end{tabular}
\caption{ Case of the $(1,1)$ branch with
$a=b=1, g_{11}=1, g_{12}=0.25, g_{22}=1, \Omega=0.1$: The left (middle) panel shows the imaginary parts of the eigenvalues
around $-i\Omega$ ($-i2\Omega$) as functions of $\epsilon$
with $O(\epsilon)$ corrections (solid lines) and corresponding numerical
results (dashed lines with circles). In the left panel, the red lines and
blue lines (solid and dashed) are almost identical since a pair of
eigenvalues of $M_a$ for $\lambda^{(0)}=-i\Omega$ are complex conjugates. Moreover, the nonzero
imaginary parts of this pair imply the instability (this is the only source of the instability) of the solution, as shown in the right panel
(solid line for the $O(\epsilon)$ correction using this pair of complex
conjugates and the dashed line for the numerical computation of the real
parts of the eigenvalues). }
\label{fig1_3_3}
\end{figure}
\begin{figure}[!htbp]
\begin{tabular}{cc}
\includegraphics[width=5cm]{ev_11_2_re_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{ev_11_2_im_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_11_u12_square_long_2_new-eps-converted-to.pdf}
\end{tabular}
\caption{ The left (middle) panel shows the change of imaginary (real) parts of the eigenvalues for the $(1,1)$
branch of solutions with $a=b=1, g_{11}=1, g_{12}=0.25, g_{22}=1, \Omega=0.1$. This is for a lengthy parametric continuation over
$\epsilon$, featuring not only the original instability near
the linear limit but subsequent splits (stabilizations) and further
collisions (destabilizations) of the relevant solution. Again the right panel shows the $L^2$-norm of the solution of $\phi_j$ as a function of $\epsilon$.
}
\label{fig1_3_4}
\end{figure}
\begin{figure}[!htbp]
\begin{tabular}{cc}
\includegraphics[width=7cm]{dynamics_11_phi1_g12_shorter_new-eps-converted-to.pdf}
\includegraphics[width=7cm]{dynamics_11_phi2_g12_shorter_new-eps-converted-to.pdf}
\end{tabular}
\caption{ The left (right) panel shows an example of the dynamics of $|\phi_1|^2$ ($|\phi_2|^2$) of the $(1,1)$ branch with $a=b=1, g_{11}=1, g_{12}=0.25, g_{22}=1$, $\Omega=0.1$ and $\epsilon=0.1$. }
\label{fig1_3_5}
\end{figure}
\subsection{$(m,n)=(0,2)$}
For $\mu_1^{(0)}=\Omega/2$ and $\mu_2^{(0)}=5\Omega/2$, we consider the
continuation of $({\psi}_1^{(0)}, {\psi}_2^{(0)})=(a u_{0}, b u_{2})$, where
\[
a^2=64\sqrt{\frac{2\pi}{\Omega}}\,
\frac{41g_{22}\mu_1^{(1)}-24g_{12}\mu_2^{(1)}}{41 g_{11}g_{22}- 9g_{12}^2},\quad
b^2=512\sqrt{\frac{2\pi}{\Omega}}\,
\frac{8 g_{11}\mu_2^{(1)}-3g_{12}\mu_1^{(1)}}{41 g_{11}g_{22}-9g_{12}^2}.
\]
Regarding spectral stability we have $K_\mathrm{Ham}=4$, except that now the
dangerous eigenvalues are at $\lambda^{(0)}=\pm i\Omega,\pm i2\Omega$. At
most one eigenvalue with positive real part will emerge from each of these
dangerous eigenvalues.
First consider the perturbation calculation associated with
$\lambda^{(0)}=-i\Omega$. We consider case (b), and the matrix
$M_b\in\mathcal{M}_{3\times3}(\mathbb{R})$ is
\begin{equation}\label{E-b-0-2}
M_b=\frac{1}{64}\sqrt{\frac{\Omega}{2\pi}}\left(
\begin{array}{ccc}
4 g_{12}b^2 & 4\sqrt{3}\,g_{12}a b & 8\sqrt{2}\, g_{12}a b \\
4\sqrt{3}\, g_{12}a b & 10 g_{22}b^2 - 4 g_{12}a^2 & 5\sqrt{6}\, g_{22}b^2 \\
-8\sqrt{2}\, g_{12}a b & -5\sqrt{6}\, g_{22}b^2 & -15 g_{22}b^2 - 8 g_{12}a^2
\end{array}
\right).
\end{equation}
\begin{proposition}\label{proposition-0-2}
For $(m,n)=(0,2)$, the eigenvalues of $M_b$ in (\ref{E-b-0-2}) are $0$ and
\[
-\frac{1}{128}\sqrt{\frac{\Omega}{2\pi}}\left(12 a^2 g_{12} - 4 b^2 g_{12}
+ 5 b^2 g_{22}\pm \sqrt{16 a^4 g_{12}^2-8 a^2 b^2 g_{12}(28 g_{12}-25
g_{22})+b^4(4 g_{12}+5 g_{22})^2 }\right).
\]
The imaginary parts of the eigenvalues for (\ref{E-b-0-2}) will be nonzero if
$g_{12}>5g_{22}/4$.
\end{proposition}
\noindent As is the case for the continuation of $(0,1)$, the parameter
$g_{11}$ does not appear in the expressions of the eigenvalues.
Now consider the perturbation calculation associated with
$\lambda^{(0)}=-i2\Omega$.
In this case, the matrix
$M_b\in\mathcal{M}_{3\times3}(\mathbb{R})$ is
\[
M_b=\frac{1}{512}\sqrt{\frac{\Omega}{2\pi}}\left(
\begin{array}{ccc}
-128 g_{11}a^2 + 136 g_{12}b^2 & 12\sqrt{6}g_{12}a b & 192 g_{12}a b \\
12\sqrt{6}\, g_{12}a b & g_{22}b^2 - 52 g_{12}a^2 & 12\sqrt{6}\, g_{22}b^2 \\
-192 g_{12}a b & -12\sqrt{6}\, g_{22}b^2 & -56 g_{22}b^2 - 320 g_{12}a^2
\end{array}
\right).
\]
Unfortunately, the expressions of eigenvalues are not as
straightforward/enlightening in an analytical form (although available). As
the numerical computations below show, it is possible for this matrix to have
a pair of eigenvalues with nonzero imaginary part.
For the numerical computations, we again let $a=b=1, g_{11}=1.03,
g_{12}=1.04, g_{22}=1.06, \Omega=0.1$, and compute the continuation of
two-component solutions. We compare
our analytical predictions providing the eigenvalues up to $O(\epsilon)$ with
the corresponding numerical eigenvalues in Fig.~\ref{fig1_4}. As $\epsilon$
grows, all of the numerically computed eigenvalues are on the imaginary axis
and their change with respect to $\epsilon$ is illustrated in the left
panel of Fig.~\ref{fig1_4b}. Additionally, we find that $\phi_2$ becomes zero
at $\epsilon\approx 3.2$ where this branch of solutions meets the branch of
one-component solutions on $\phi_1$. We note that this resembles the first example of case (0,1) very much.
\begin{figure}[!htbp]
\begin{tabular}{ccc}
\includegraphics[width=5cm]{compare_02_1_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_02_2_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_02_u12_density_new-eps-converted-to.pdf}
\end{tabular}
\caption{Case of the $(0,2)$ branch with $a=b=1, g_{11}=1.03,
g_{12}=1.04, g_{22}=1.06, \Omega=0.1$: The left (middle) panel shows the imaginary parts of the eigenvalues
around $-i\Omega$ ($-i2\Omega$) as functions of $\epsilon$ with $O(\epsilon)$ corrections
(solid lines) and corresponding numerical results (dashed lines with circles). The right panel shows the densities of $\phi_1$ and $\phi_2$ at $\epsilon=0.1$ for this $(0,2)$ waveform.}
\label{fig1_4}
\end{figure}
\begin{figure}[!htbp]
\begin{tabular}{cc}
\includegraphics[width=7cm]{ev_02_1_re_new-eps-converted-to.pdf}
\includegraphics[width=7cm]{compare_02_u12_square_long_new-eps-converted-to.pdf}\\
\end{tabular}
\caption{In the left panel, we keep track of the imaginary parts of the eigenvalues for large $\epsilon$ where the imaginary parts remain zero. The right panel shows the $L^2$-norm of the solution of $\phi_j$ for such variation of $\epsilon$.}
\label{fig1_4b}
\end{figure}
However, in this case too, we can explore realistic scenarios where the
instability manifests itself immediately in the vicinity of the linear limit.
In particular, if $a=b=1, g_{11}=1, g_{12}=1, g_{22}=0.5, \Omega=0.1$, the
numerical computation shows that there exist two quartets of eigenvalues
(near $\pm i\Omega$ and $\pm i2\Omega$) that do not lie on the imaginary
axis, as shown in Fig.~\ref{fig1_4_3}. I.e., in this case, {\it both}
unstable eigendirections of the system are realized and, in fact, potentially
concurrently (contrary, e.g., to the case of $(1,1)$ waves). As $\epsilon$
grows, we observe that the complex pairs near $-i\Omega$ and $-i2\Omega$ tend
to come back to the imaginary axis and split along the axis, as shown in
Fig.~\ref{fig1_4_4}. We observe that $\phi_2$ vanishes at $\epsilon\approx
0.8$ where the branch of solutions meets the one-component branch of
solutions on $\phi_1$ there.
In Figure~\ref{fig1_4_5}, the
numerically-monitored dynamics of the steady-state solution with a small
initial perturbation verifies its instability. Here too, the instability
manifests its oscillatory character and weak growth rate over longer time
scales.
{In particular, the middle and bottom panels of Figure~\ref{fig1_4_5} show that the system first quantitatively alternates between unstable states $(0,2)$ and $(2,0)$ and then transits to the states that are close to $(0,1)$ and $(1,0)$. It can be checked that the dynamics of $|c_2|$ and $|c_0|$ comes with small oscillations with frequency close to $\frac{2\pi}{2\Omega}$, which implies that the instability in the first phase is related to the unstable eigenvalues near $\pm 2i\Omega$. Similarly, the time evolution of $|c_1|$ oscillates at the frequency of approximately $\frac{2\pi}{\Omega}$, which is connected to the unstable eigenvalues near $\pm i\Omega$.
We note that the two-phase time evolution shown in Figure~\ref{fig1_4_5} is typical for our setup.}
\begin{figure}[!htbp]
\begin{tabular}{ccc}
\includegraphics[width=5cm]{compare_02_b_1_re_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_02_b_2_re_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_02_b_im_new-eps-converted-to.pdf}
\end{tabular}
\caption{Case of $a=b=1, g_{11}=1, g_{12}=1, g_{22}=0.5, \Omega=0.1$
for the $(0,2)$ branch: The left (middle) panel shows the imaginary parts of $\lambda$'s around $\Omega$ ($2\Omega$) as functions of $\epsilon$ with
$O(\epsilon)$ corrections (solid lines) and corresponding numerical
results (dashed lines with circles). In these two panels, the red lines and green lines
(solid and dashed) are almost identical since both $M_b$ for $\lambda^{(0)}=-i\Omega$ and
$M_b$ for $\lambda^{(0)}=-i2\Omega$ have a pair of eigenvalues that are complex conjugates. Moreover, the
nonzero imaginary parts of these pairs imply the instability of the solution, as shown in the right panel (solid lines for the $O(\epsilon)$ corrections using these pairs of complex conjugates and the dashed lines for the numerical computation of the real parts of the eigenvalues, blue for the ones
near $-i\Omega$ and green for the ones near $-i2\Omega$). }
\label{fig1_4_3}
\end{figure}
\begin{figure}[!htbp]
\begin{tabular}{ccc}
\includegraphics[width=5cm]{ev_02_2_re_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{ev_02_2_im_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_02_u12_square_long_2_new-eps-converted-to.pdf}
\end{tabular}
\caption{ The left (middle) panel shows the change of
imaginary (real) parts of $i \lambda$ for the branch of solutions with
$a=b=1, g_{11}=1, g_{12}=1, g_{22}=0.5, \Omega=0.1$. The presence of the two associated instabilities is
evident in the middle panel. In the right panel, we monitor the change of the $L^2$-norm of the solution of $\phi_j$ over $\epsilon$.
}
\label{fig1_4_4}
\end{figure}
\begin{figure}[!htbp]
\begin{tabular}{cc}
\includegraphics[width=7cm]{dynamics_02_phi1_g12_shorter_new-eps-converted-to.pdf}
\includegraphics[width=7cm]{dynamics_02_phi2_g12_shorter_new-eps-converted-to.pdf}\\
\includegraphics[width=7cm]{dynamics_02_phi1_g12_longer_2_reduced-eps-converted-to.pdf}
\includegraphics[width=7cm]{dynamics_02_phi2_g12_longer_2_reduced-eps-converted-to.pdf}\\
\includegraphics[width=7cm]{dynamics_02_phi1_g12_longer_2_eigenmodes-eps-converted-to.pdf}
\includegraphics[width=7cm]{dynamics_02_phi2_g12_longer_2_eigenmodes-eps-converted-to.pdf}
\end{tabular}
\caption{ The top left (right) panel shows an example of the dynamics
of $|\phi_1|^2$ ($|\phi_2|^2$) for the $(0,2)$ branch
with $a=b=1, g_{11}=1, g_{12}=1, g_{22}=0.5$,
$\Omega=0.1$ and $\epsilon=0.1$. In the middle panels, we monitor the dynamics of the same solution for a longer time, revealing the two stage
nature of the evolution of the instability (see the relevant
discussion in the text). The bottom panels show the same dynamics (left for $\phi_1$ and right for $\phi_2$) using the dynamical decomposition to the
orthonormal basis $\{u_k\}$, where $c_k$ stands for the coefficient for $u_k$
in the decomposition. The transition between modes $(0,2)$ and $(2,0)$
originally to
$(0,1)$ and $(1,0)$ eventually is evident in the mode dynamics.}
\label{fig1_4_5}
\end{figure}
\subsection{$(m,n)=(1,2)$ (interchange all subscripts to obtain $(m,n)=(2,1)$)}
For $\mu_1^{(0)}=3\Omega/2$ and $\mu_2^{(0)}=5\Omega/2$ we consider the
continuation of $({\psi}_1^{(0)}, {\psi}_2^{(0)})=(a u_{1}, b u_{2})$, where
\[
a^2=4\sqrt{\frac{2\pi}{\Omega}}\,
\frac{41g_{22}\mu_1^{(1)}-28g_{12}\mu_2^{(1)}}{123 g_{11}g_{22}- 49g_{12}^2},\quad
b^2=16\sqrt{\frac{2\pi}{\Omega}}\,
\frac{12 g_{11}\mu_2^{(1)}-7g_{12}\mu_1^{(1)}}{123 g_{11}g_{22}-49g_{12}^2}.
\]
Regarding spectral stability we have $K_\mathrm{Ham}=6$, and the dangerous
eigenvalues are $\lambda^{(0)}=\pm i\Omega,\pm i2\Omega$. It is possible for
a pair of eigenvalues with nonzero real part to emerge from $\pm i\Omega$,
while at most one eigenvalue with positive real part can emerge from $\pm
i2\Omega$.
First consider the perturbation calculation with $\lambda^{(0)}=-i\Omega$. We
consider case (a), and $M_a\in\mathcal{M}_{4\times4}(\mathbb{R})$ is
\[
M_a=\frac{1}{64}\sqrt{\frac{\Omega}{2\pi}}\left(
\begin{array}{cccc}
8 g_{11} a^2 + 13 g_{12}b^2 & 8\sqrt{2}\,g_{11}a^2 & 5\sqrt{6}\,g_{12}a b & 28 g_{12}a b \\
-8\sqrt{2}\,g_{11}a^2 & -16 g_{11} a^2+4 g_{12}b^2 & -4\sqrt{3}\,g_{12}a b & -8\sqrt{2}\, g_{12}a b \\
5\sqrt{6}\, g_{12}a b & 4\sqrt{3}\,g_{12}a b & 10 g_{22}b^2 - 6 g_{12}a^2 & 5\sqrt{6}\, g_{22} b^2 \\
-28 g_{12}a b & -8\sqrt{2}\, g_{12}a b & -5\sqrt{6}\, g_{22} b^2 & -15 g_{22}b^2 - 20 g_{12}a^2
\end{array}
\right).
\]
As per the dipolar mode that we discussed previously,
one eigenvalue of $M_a$ is zero, with associated eigenvector
$(-a/b,-a/(\sqrt{2}\,b,-\sqrt{6}/2,1)^\mathrm{T}$. Consequently, $M_a$ can
have at most one pair of eigenvalues with nonzero imaginary part, which
implies that at most one pair of eigenvalues with nonzero real part can
emerge under the perturbation. If we particularly set $g_{11}=g_{22}=1$ and
$a=b=1$, numerical results suggest a pair of complex conjugate eigenvalues
will arise for $g_{12}>1$, as illustrated in the example below
Now we consider the perturbation calculation with $\lambda^{(0)}=-i2\Omega$.
We consider case (b), and $M_b\in\mathcal{M}_{3\times3}(\mathbb{R})$ is
\[
\frac{1}{512}\sqrt{\frac{\Omega}{2\pi}}\left(
\begin{array}{ccc}
-32 g_{11} a^2-20 g_{12}b^2 & 46\sqrt{2}\,g_{12}a b & 32\sqrt{3}\, g_{12}a b \\
46\sqrt{2}\, g_{12}a b & g_{22}b^2 - 74 g_{12}a^2 & 12\sqrt{6}\, g_{22}b^2 \\
-32\sqrt{3}\, g_{12}a b & -12\sqrt{6}\, g_{22}b^2 & -56 g_{22}b^2 - 32 g_{12}a^2
\end{array}
\right).
\]
This matrix can have at most one pair of eigenvalues with nonzero imaginary
part, an example of which can be obtained for $g_{11}=g_{22}=1$, $a=b=1$ and
$g_{12}>0$ (see Fig.~\ref{fig1_5}).
For the numerical calculations we let $a=b=1, g_{11}=1, g_{12}=1.2, g_{22}=1,
\Omega=0.1$ and we compare our analytical predictions for the leading order
corrections to the eigenvalues up to $O(\epsilon)$ against the corresponding
numerical eigenvalues in Fig.~\ref{fig1_5}. We find two eigenvalue pairs (one
pair each near $\pm i\Omega$ and $\pm i2\Omega$) introducing respective
instability eigendirections.
In Fig.~\ref{fig1_5_1}, we see that the pairs near $-i\Omega$ and $-i2\Omega$
will eventually return to the imaginary axis, over considerably wider
parametric continuations in $\epsilon$, splitting along the axis.
Among these returned imaginary eigenvalues, the one that stems from
$-i2\Omega$ and goes upward will meet the eigenvalue coming from $-i4\Omega$
to generate another pair of complex eigenvalues at $\epsilon\approx 2.7$.
Shortly after (parametrically), these complex eigenvalues will come back to
the axis and split into two eigenvalue pairs, with one of them going up to
further repeat this process at $\epsilon\approx 3.6$ and $\epsilon\approx 4.2$.
\begin{figure}[!htbp]
\begin{tabular}{cc}
\includegraphics[width=8cm]{compare_12_1_re_new-eps-converted-to.pdf}
\includegraphics[width=8cm]{compare_12_2_re_new-eps-converted-to.pdf}\\
\includegraphics[width=8cm]{compare_12_u12_density_new-eps-converted-to.pdf}
\includegraphics[width=8cm]{compare_12_im_new-eps-converted-to.pdf}
\end{tabular}
\caption{ Case of $a=b=1, g_{11}=1, g_{12}=1.2, g_{22}=1,
\Omega=0.1$ for the $(1,2)$ branch: The top left panel shows the real parts of $i\lambda$ around $\Omega$ as functions
of $\epsilon$ with $O(\epsilon)$ corrections (solid lines) and
corresponding numerical results (dashed lines with circles). The top right panel shows the real parts of $i \lambda$ around $2\Omega$. In both panels,
it should be noticed that the red lines and green lines (solid and dashed)
are almost identical since a pair of eigenvalues of $M_a$ and $M_b$ are complex conjugates. Moreover,
the nonzero imaginary parts
of these two pairs imply the instability of the solution, as shown in the bottom right panel
(the growth rates are shown as
solid lines for the predicted $O(\epsilon)$ corrections and as dashed lines for the
numerical computation of the real parts of the eigenvalues; green for $M_b$ with $\lambda^{(0)}=-i2\Omega$
and blue for $M_a$ with $\lambda^{(0)}=-i\Omega$). The densities of $\phi_1$ and $\phi_2$ at $\epsilon=0.1$
are given in the bottom left panel.}
\label{fig1_5}
\end{figure}
\begin{figure}[!htbp]
\begin{tabular}{ccc}
\includegraphics[width=5cm]{ev_12_re_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{ev_12_im_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_12_u12_square_long_new-eps-converted-to.pdf}
\end{tabular}
\caption{ The left (middle) panel shows the change of imaginary (real) parts of the eigenvalues
for the $(1,2)$ branch of solutions with $a=b=1, g_{11}=1, g_{12}=1.2, g_{22}=1, \Omega=0.1$. In this
extended parametric continuation, the
signature of the instabilities is evident in the middle panel of
the figure. The right panel shows the change of the $L^2$-norm of the solution of $\phi_j$ over $\epsilon$.}
\label{fig1_5_1}
\end{figure}
In Figure~\ref{fig1_5_2}, we illustrate the numerical evolution of the
unstable configuration shown in Figure \ref{fig1_5} with $\epsilon=0.1$. With
a small initial perturbation, the oscillation around the stationary solution
gradually grows and the instability becomes apparent in the dynamics.
\begin{figure}[!htbp]
\begin{tabular}{cc}
\includegraphics[width=7cm]{dynamics_12_phi1_g12_shorter_new-eps-converted-to.pdf}
\includegraphics[width=7cm]{dynamics_12_phi2_g12_shorter_new-eps-converted-to.pdf}
\end{tabular}
\caption{ The left (right) panel shows an example of the dynamics of $|\phi_1|^2$ ($|\phi_2|^2$) for the $(1,2)$ state
with $a=b=1, g_{11}=1, g_{12}=1.2, g_{22}=1$, $\Omega=0.1$ and $\epsilon=0.1$. Again, the oscillatory
nature of the associated instability
eventually kicks in over longer time scales.}
\label{fig1_5_2}
\end{figure}
\subsection{$(m,n)=(2,2)$}
When $\mu_1^{(0)}=\mu_2^{(0)}=5\Omega/2$ we consider $({\psi}_1^{(0)},
{\psi}_2^{(0)})=(a u_{2}, b u_{2})$, where
\[
a^2=\frac{64}{\sqrt{41}}\sqrt{\frac{2\pi}{\Omega}}\,
\frac{g_{22}\mu_1^{(1)}-g_{12}\mu_2^{(1)}}{g_{11}g_{22}-g_{12}^2},\quad
b^2=\frac{64}{\sqrt{41}}\sqrt{\frac{2\pi}{\Omega}}\,
\frac{g_{11}\mu_2^{(1)}-g_{12}\mu_1^{(1)}}{g_{11}g_{22}-g_{12}^2}.
\]
Regarding spectral stability we have $K_\mathrm{Ham}=8$, and the dangerous
eigenvalues are $\lambda^{(0)}=\pm i\Omega,\pm i2\Omega$. It is possible for
two pairs of eigenvalues with nonzero real part to emerge from each of the
dangerous eigenvalues.
First consider the perturbation calculation associated with
$\lambda^{(0)}=-i\Omega$. We have case (a), and the matrix $M_a$ is
\[
M_a=\frac{1}{128}\sqrt{\frac{\Omega}{2\pi}}\left(
\begin{array}{cccc}
20 g_{11} a^2 - 31 g_{12}b^2 & 10\sqrt{6}\,g_{11}a^2 & 51 g_{12}a b & 10\sqrt{6}\, g_{12}a b \\
-10\sqrt{6}\,g_{11}a^2 & -30 g_{11} a^2 +26 g_{12}b^2 & -10\sqrt{6}\,g_{12}a b & -56 g_{12}a b \\
51 g_{12}a b & 10\sqrt{6}\,g_{12}a b & 20 g_{22}b^2 - 31 g_{12}a^2 & 10\sqrt{6}\, g_{22} b^2 \\
-10\sqrt{6}\, g_{12}a b & -56 g_{12}a b & -10\sqrt{6}\, g_{22} b^2 & -30 g_{22}b^2 + 26 g_{12}a^2
\end{array}
\right).
\]
One of the eigenvalues is zero, with associated eigenvector
$(-\sqrt{6}a/(2b),a/b,-\sqrt{6}/2,1)^\mathrm{T}$, for the same
(dipolar) symmetry reasons as before. Consequently, $M_a$ can
have at most one pair of eigenvalues with nonzero imaginary part, so at most
one pair of eigenvalues with nonzero real part can emerge from $\pm i\Omega$.
As an example, if we assume $a=b=1$ and $g_{11}=g_{22}=1$, then the other
eigenvalues of $M_a$ are
\[
-\frac5{64}\sqrt{\frac{\Omega}{2\pi}}\,(1+g_{12}),\quad
\sqrt{\frac{\Omega}{2\pi}}\,
\frac{-5\pm\sqrt{25-2900 g_{12}+6124g_{12}^2}}{128}
\]
where the instability (nonzero imaginary parts) will emerge for
\[
g_{12}\in (\frac{725-285\sqrt{6}}{3062}, \frac{725+285\sqrt{6}}{3062} )\approx (0.0088, 0.4648).
\]
Now consider the perturbation calculation associated with
$\lambda^{(0)}=-i2\Omega$. We again have case (a), and the matrix $M_a$ is
now
\[
M_a=\frac{1}{1024}\sqrt{\frac{\Omega}{2\pi}}\left(
\begin{array}{cccc}
2 g_{11} a^2 - 327 g_{12}b^2 & 24\sqrt{6}\,g_{11}a^2 & 329 g_{12}a b & 24\sqrt{6}\, g_{12}a b \\
-24\sqrt{6}\,g_{11}a^2 & -112 g_{11} a^2 +272 g_{12}b^2 & -24\sqrt{6}\,g_{12}a b & -384 g_{12}a b \\
329 g_{12}a b & 24\sqrt{6}\,g_{12}a b & 2 g_{22}b^2 - 327 g_{12}a^2 & 24\sqrt{6}\, g_{22} b^2 \\
-24\sqrt{6}\, g_{12}a b & -384 g_{12}a b & -24\sqrt{6}\, g_{22} b^2 & -112 g_{22}b^2 + 272 g_{12}a^2
\end{array}
\right).
\]
If we set $a=b=1$ and $g_{11}=g_{22}=1$, then the eigenvalues of $M_a$ are
\[
\sqrt{\frac{\Omega}{2\pi}}\,
\frac{-55(1+g_{12}) \pm i3\sqrt{23}\,|1+g_{12}|}{1024},\quad
\sqrt{\frac{\Omega}{2\pi}}\,
\frac{-55\pm \sqrt{-207-67872 g_{12}+426880g_{12}^2}}{1024}.
\]
Minimally one pair of eigenvalues will gain nonzero imaginary part, and if
\[
g_{12}\in (\frac{3(1414-599\sqrt{6})}{53360}, \frac{3(1414+599\sqrt{6})}{53360}) \approx (-0.0030, 0.1620)
\]
two pairs of eigenvalues with nonzero imaginary part will emerge.
If $a=b=1, g_{11}=1, g_{12}=0.1, g_{22}=1, \Omega=0.1$ we
provide the relevant comparison of analytical predictions
and numerically computed eigenvalues in Fig.~\ref{fig1_6}.
In this case, we identify three quartets of unstable eigenvalues (one near
$\pm i\Omega$ and two near $\pm i2\Omega$). As $\epsilon$ increases, we see
that the complex eigenvalues near $-i\Omega$ (at $\epsilon\approx 5.2$) and
the ones near $-i2\Omega$ (at $\epsilon\approx 1.2$ and $\epsilon\approx 4.9$)
will return to the imaginary axis and split along it as shown in
Fig.~\ref{fig1_6_1}. Additionally, the split eigenvalues from $-i\Omega$ and
$-i2\Omega$ going upward will collide with the eigenvalues from $-i3\Omega$
and $-i4\Omega$, respectively, to generate new eigenvalues with nonzero real
part, a feature illustrated in the extended parametric continuation of
Fig.~\ref{fig1_6_1}. In Figure~\ref{fig1_6_2}, we illustrate the numerical
evolution of this unstable configuration $\epsilon=0.1$. {The instability
settles in an oscillatory manner after a long time evolution, redistributing
the atoms within the condensate and resulting in the recurrence of different states. We note that this is a more complicated case than the breathing case in Figure~\ref{fig1_2_4} since there are more possible unstable eigendirections in this case.}
\begin{figure}[!htbp]
\begin{tabular}{cc}
\includegraphics[width=8cm]{compare_22_1_re_new-eps-converted-to.pdf}
\includegraphics[width=8cm]{compare_22_2_re_new-eps-converted-to.pdf}\\
\includegraphics[width=8cm]{compare_22_u12_density_new-eps-converted-to.pdf}
\includegraphics[width=8cm]{compare_22_im_new-eps-converted-to.pdf}
\end{tabular}
\caption{ Case of
$a=b=1, g_{11}=1, g_{12}=0.1, g_{22}=1, \Omega=0.1$ for the
$(2,2)$ branch: The top left panel shows the imaginary parts of the eigenvalues around $-i\Omega$ as functions of
$\epsilon$ with $O(\epsilon)$ corrections (solid lines) and
corresponding numerical results for comparison
(dashed lines with circles). The red and green lines are almost identical
since one pair of the eigenvalues of $M_a$ for $\lambda^{(0)}=-i\Omega$ corresponds
to complex conjugates. The top right panel shows the imaginary parts of the eigenvalues around $-i2\Omega$.
It should be noticed again that the red lines and green lines (solid and dashed) are essentially
identical and the blue lines and black lines (solid and dashed) are almost the same since two
quartets of eigenvalues of $M_a$ for $\lambda^{(0)}=-i2\Omega$ arise in this case. Moreover,
the nonzero real parts
of $\lambda$'s also imply the instability of the solution, as shown in the bottom
right panel (solid lines for the $O(\epsilon)$ correction using this pair of complex
conjugates and the dashed lines for the numerical computation of the real parts of the
eigenvalues; green is used for $M_a$ for $\lambda^{(0)}=-i\Omega$ while red and blue
denote the imaginary parts for the $M_a$ with $\lambda^{(0)}=-i2\Omega$ eigenvalues).
The densities of $\phi_1$ and $\phi_2$ at $\epsilon=0.1$ are given in the
bottom left panel, showcasing the second excited state nature of both
fields. }
\label{fig1_6}
\end{figure}
\begin{figure}[!htbp]
\begin{tabular}{cc}
\includegraphics[width=5cm]{ev_22_re_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{ev_22_im_new-eps-converted-to.pdf}
\includegraphics[width=5cm]{compare_22_u12_square_long_new-eps-converted-to.pdf}
\end{tabular}
\caption{ The left (middle) panel shows the change of imaginary (real) parts of the
eigenvalues for the $(2,2)$ branch of solutions with $a=b=1, g_{11}=1, g_{12}=0.1, g_{22}=1, \Omega=0.1$.
In this case, the middle panel illustrates the potential
of the configuration for 3 instabilities all of which are manifested
(possibly even concurrently for small $\epsilon$). Interestingly though,
for sufficiently large $\epsilon$, there exists a potential
parametric interval of spectral stability. In the right panel, we plot the $L^2$-norm of the solution of $\phi_j$ as a function of $\epsilon$.}
\label{fig1_6_1}
\end{figure}
\begin{figure}[!htbp]
\begin{tabular}{cc}
\includegraphics[width=7cm]{dynamics_22_phi1_g12_shorter_new-eps-converted-to.pdf}
\includegraphics[width=7cm]{dynamics_22_phi2_g12_shorter_new-eps-converted-to.pdf}
\end{tabular}
\caption{ The top left (right) panel shows an example of the dynamics of $|\phi_1|^2$
($|\phi_2|^2$) with $a=b=1, g_{11}=1, g_{12}=0.1, g_{22}=1$, $\Omega=0.1$ and $\epsilon=0.1$. Over
longer time scales, the oscillatory instability
sets in re-arranging the atomic distribution in both components.}
\label{fig1_6_2}
\end{figure}
\subsection{Summary of the spectral stability results}
\label{section:summary-stability}
In Section~\ref{section:stability-catologue}, we have examined solutions with different $(m, n)$ pairs for $0\leq m, n \leq
2$. The spectral stability of each solution has been studied both
analytically and topologically via perturbation theory for small $\epsilon$.
The topological results are robust, and valid for all pairs of $(m,n)$.
In order to upgrade the stability results for general $(m,n)$, we first introduce several identities for $m,n\ge 1$ as follows:
\begin{eqnarray}
\label{eqn_claim_1_1}
\alpha_m:=
\frac{ A_m - 2B_{m,m+1} }{ D_{m,m,m-1,m+1} }=\frac{D_{m,m,m-1,m+1}}{ A_m - 2B_{m,m-1} }=-\sqrt{\frac{m}{m+1}}
\end{eqnarray}
\begin{eqnarray}
\label{eqn_claim_1_2}
\frac{D_{m,n,m+1,n+1}+\alpha_n\cdot D_{m,n,m+1,n-1}}{B_{m,n}-B_{m+1,n}} = \frac{D_{m,n,m-1,n+1}+\alpha_n\cdot D_{m,n,m-1,n-1}}{\alpha_m (B_{m,n}-B_{m-1,n})}
=\sqrt{\frac{m+1}{n+1}}
\end{eqnarray}
\begin{eqnarray}
\label{eqn_claim_1_3}
\frac{B_{0,n-1}-B_{0,n}}{D_{0,1,n-1,n}}=-\frac{1}{\alpha_n} \frac{B_{0,n}-B_{0,n+1}}{D_{0,1,n,n+1}} =
\frac{1}{\sqrt{n}},\quad
B_{0,n}=\frac{(2n-1)!}{(n-1)!\, n!\, 2^{2n-1}} \sqrt{\frac{\Omega}{2 \pi}}.
\end{eqnarray}
Though no analytical proofs for these identities are provided here, we have verified them for general $(m,n)$ pairs through extensive numerical experiments.
Then for the stability results:
\begin{itemize}
\item When $m>0$, there are four eigenvalues of $-iJ \mathcal{L}$ near
$-i\ell\Omega$ for $1 \leq\ell \leq m$. According to the
Hamiltonian-Krein index, at most two eigenvalues among four can have
positive real parts (leading to instability). However, the perturbation
calculation suggests that not all of the four eigenvalues can enter the
complex plane for $m\ge 1$.
\begin{remark}
\label{remark_4_1} For $1\leq m\leq n$, it can be directly checked
using (\ref{eqn_claim_1_1})--(\ref{eqn_claim_1_2}) that $M_a$ for
$\ell=1$ has an eigenvalue $0$ with eigenvector $\displaystyle{\left(1,
\sqrt{\frac{m}{m+1}}, \frac{b}{a}\sqrt{\frac{n+1}{m+1}},
\frac{b}{a}\sqrt{\frac{n}{m+1}}\right)^\mathrm{T}}$.
\end{remark}
Thus, near $-i\ell \Omega$, two eigenvalues will always stay on the imaginary axis (one of them is $0$) and there are at most one pair of complex eigenvalues. We discussed previously the physical origin of the corresponding
(dipolar) symmetry removing the potential for one among the
pertinent instability eigendirections.
\item When $m<n$, there are three eigenvalues near $-i\ell\Omega$ for
$m<\ell\leq n$. At most one pair of these eigenvalues will have nonzero
real part. Similar to Remark~\ref{remark_4_1}, we particularly notice that one of eigenvalues near $-i\Omega$ will always be zero.
\begin{remark}
\label{remark_4_1a} For $m=0$ and $1\leq n$, it can be directly checked
using (\ref{eqn_claim_1_1})--(\ref{eqn_claim_1_3}) that $M_b$ for
$\ell=1$ has an eigenvalue $0$ with eigenvector
$\displaystyle{\left(\frac{a}{b}\frac{1}{\sqrt{n}},
\sqrt{\frac{n+1}{n}}, 1\right)^\mathrm{T}}$.
\end{remark}
\item The Hamiltonian-Krein index, $K_\mathrm{Ham}=2(m+n)$, gives an upper
bound for the number of pairs of eigenvalues that can leave the
imaginary axis and bring about an instability. In the examined
examples, this upper bound can be reached only when $m=0$. For $m>0$,
the exact upper bound will be $2(m+n-1)$, given the presence of the
symmetry/invariance associated with dipolar motion of the condensate
removing one of the potentially unstable associated eigendirections
\item When $(m,n)=(0,1)$, an instability will arise if and only if
$g_{12}>g_{22}>0$, i.e., the inter-component nonlinear interactions are
stronger than the nonlinear interactions within the ``dark" species.
\end{itemize}
As $\epsilon$ grows away from $0$, we notice that the eigenvalue starting
from $-i\Omega$ can collide with the eigenvalues from $-i3\Omega$,
$-i5\Omega,\dots$ on the imaginary axis to generate eigenvalues with nonzero
real part. Similarly, the eigenvalue from $-i2\Omega$ can meet with the
eigenvalues from $-i4\Omega,\,-i6\Omega,\dots$ on the imaginary axis to
produce new pairs of eigenvalues with nonzero real part.
Our numerical results suggest that (given their respective
parities) eigenmodes at odd multiples of $\Omega$ interact with
other ones such and similarly even ones interact with even.
While our analysis does not lend itself to the
consideration of this wide parametric regime in $\epsilon$, numerical
computations reveal the corresponding potential (oscillatory) instabilities
and their customary restabilization for some interval of wider
parametric variations of
$\epsilon$.
\section{Conclusions \& Future Challenges}
In the present work, we illustrated the usefulness of Lyapunov-Schmidt
reductions, as well as of Hamiltonian-Krein index theory, in acquiring a
systematic understanding of bifurcations from the linear limit of the
multi-component system of atomic gases. Here, we have focused on the
two-component case, yet it should be evident from the analysis how general
multi-component cases will modify the specifics yet not the overall
formulation of the present setting. Once again, this mean-field limit may be
of somewhat limited applicability to the atomic case for very small atom
numbers (mathematically, squared $L^2$ norms), as there additional (quantum)
effects may skew the picture. Nevertheless, optical settings (with suitably
tailored refractive index profiles) can lend themselves to the analysis
presented herein. Moreover, and arguably more importantly, the
topological nature of the tools developed provides insights on the number of
potentially unstable eigendirections even far from the linear limit, where
the mean field model has been successfully used to monitor different
multi-component excited states, such as most notably e.g. dark-bright
solitons and their close relatives (such as dark-dark ones). We have found a
number of surprising results in the process, such as the fact that $(0,1)$
states (involving one fundamental and one excited state) may be unstable
provided that inter- to intra-component interaction ratios are suitably
chosen. Another intriguing feature is that the presence of additional
symmetry (embedded in the dipolar motion inside the trap) may prevent
particular instability eigendirections from manifesting themselves.
It would be interesting to extend the present considerations
to spinor systems that are intensely studied over the past few
years in atomic experiments~\cite{kawueda,stampueda}. Additionally,
higher dimensional settings, both two-dimensional ones where
vortex-bright and related states have been devised~\cite{VB},
but also three-dimensional ones involving vortex-rings~\cite{rings} and
skyrmions~\cite{ruost} or related patterns would be especially interesting
to attempt to explore through this methodology, as traditionally
the complexity of such states limits the potential for
analytical results. Lastly, it does not escape us that an
equally interesting and analytically tractable (at least to some
degree) limit is that of large chemical potentials where the solitary
waves can be treated as particles. Developing a general theory
of that limit and connecting that with the low amplitude
limit presented herein, would be of particular interest.
This would also allow to showcase the connection between
the two tractable limits via numerical computations and to confirm the
robustness of the topological tools in revealing the potential
for instability while traversing the continuum from one to
the other limit. Such studies are currently in progress and
will be reported in future publications.
{\it Acknowledgements.}\\
P.G.K. gratefully acknowledges support from NSF-DMS-1312856. T.K. gratefully acknowledges support from Calvin College through a Calvin Research Fellowship.
| 080937ba6a4fa6c5c31aef58456d6dcab4655fee | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
\subsection{Bloch functions} Let $\D$ be the unit disk in the complex plane and $\vp$ an analytic function defined in $\D$. We say that $\vp$ is a \emph{Bloch function} if
\be \label{Bloch}
\beta(\vp) = \sup_{z\in\D}(1-|z|^2)|\varphi'(z)|<\infty.
\ee
This defines a seminorm, and the Banach space $\cB$ of all Bloch functions equipped with the norm $\|\varphi\|_\mathcal B = |\varphi(0)|+\beta(\vp) $ is called \emph{Bloch space}. We refer to \cite{ACP}, \cite{Da}, \cite{Po70}, \cite{Pom}, \cite{SW} and \cite{Zhu} for information on the Bloch space.
For $\vp$ analytic in $\D$ the \emph{schlicht radius} $d_\vp(z)$ is defined as the radius of the largest disk lying on the Riemann image of $\vp$ and centered at the point $\vp(z)$, whenever $z$ is not a branch point, \emph{i.e.}$\!$ if $\vp'(z)\neq 0$. At a branch point of $\vp$ the schlicht radius is defined as zero. It was shown in \cite{SW} that every analytic function satisfies
$$
d_\vp(z) \, \leq \, (1-|z|^2)|\vp'(z)|, \qquad z\in\D.
$$
A similar inequality in the reverse direction was also shown in \cite[\S31]{SW} for the case when the schlicht radius is uniformly bounded. Thus, $\vp\in\cB$ if and only if $\sup_{z\in\D}d_\vp(z)<\infty$.
For univalent functions these inequalities take the simpler form of
\be \label{rad}
\frac{1}{4}(1-|z|^2)|\vp'(z)| \, \leq \, d_\vp(z) \, \leq \, (1-|z|^2)|\vp'(z)|, \qquad z\in\D
\ee
due to Koebe's $1/4$-Theorem. Note that in this case $d_\vp(z)$ is simply the distance between $\vp(z)$ and the boundary of $\vp(\D)$ and, therefore, $\vp\in\cB$ if and only if $\vp(\D)$ does not contain arbitrarily large disks.
Yet another close connection between Bloch functions and univalent functions was found in \cite{Po70}. Namely, if $\psi$ is univalent then $\beta(\log\psi') \leq 6$ and, conversely, if $\beta(\vp)\leq 1$ then $\vp=\log \psi'$ for some univalent function $\psi$.
\subsection{Harmonic mappings} A planar harmonic mapping is a complex-valued harmonic function $f$ defined on a domain $\Omega\subset\mathbb{C}$. When $\Omega$ is simply connected, the mapping has a \emph{canonical decomposition} $f=h+\overline{g}$, where $h$ and $g$ are analytic in $\Omega$. Since the Jacobian of $f$ is given by $J_f=|h'|^2 - |g'|^2$, it is locally univalent and \emph{orientation-preserving} if and only if $|g'|< |h'|$, or equivalently, if $h'(z)\neq0$ and the dilatation $\omega=g'/h'$ has the property $|\omega(z)|<1$ in $\Omega$. We say that $f$ is \emph{orientation-reversing} if $\overline{f}$ is orientation-preserving.
Since the mid-80s and especially after the work of J. Clunie and T. Sheil-Small \cite {CS} in 1984, there has been a great interest in trying to extend the classic results of the analytic world to their harmonic analogues. Some work in this direction for the Bloch space was done by F. Colonna \cite{Co}, whose point of departure was the \emph{metric} characterization of $\cB$, namely, $f\in\cB$ if and only if $f$ is Lipschitz between $\D$ endowed with the hyperbolic metric and $\SC$ endowed with the euclidean metric. For a harmonic mapping $f=h+\overline g\,$ this Lipschitz condition was proved in \cite{Co} to be equivalent to both $h$ and $g$ belonging to $\cB$.
The \emph{schlicht radius} $d_f(z)$ of a harmonic mapping $f=h+\overline{g}$ is defined as the radius of the largest disk which is the injective image of some subdomain of $\D$ and is centered at $f(z)$. We set $d_f(z)=0$ if no such disk exists. A generalization of the \emph{geometric} definition of Bloch functions would be to ask that $f$ satisfy $\sup_{z\in\D}d_f(z)<\infty$. However, we shall prove in Lemma \ref{radius} that if $f$ is univalent and normalized then
$$
\frac{1}{16}(1-|z|^2) (|h'(z)|-|g'(z)|) \; \leq \; d_f(z) \; \leq \; \frac{\pi}{2} (1-|z|^2) |h'(z)|, \qquad z\in\D.
$$
It can also be shown that no two of the above three quantities are comparable. Therefore, an analytic characterization of the geometric definition for harmonic mappings is, as far as we know, yet to be found.
\subsection{Harmonic Bloch-type functions} Our starting point will be the \emph{analytic} definition \eqref{Bloch}. Noting that the Jacobian of an analytic function $\vp$ is given by $J_\vp = |\vp'|^2$, we feel justified in introducing the following definition.
\begin{definition}
Let $f=h+\overline{g}$ be harmonic in $\D$. We say that $f$ is a \emph{Bloch-type function} if
$$
\beta(f) \; = \; \sup_{z\in\D}(1-|z|^2)\sqrt{|J_f(z)|}<\infty.
$$
We denote this class of functions by $\cB_H$.
\end{definition}
Indeed, we shall see in Section 2 that this definition gives rise to a class rather than a linear space. However, $\cB_H$ contains the Bloch space defined in \cite{Co}. We shall prove that $\cB_H$ is both affine and linearly invariant. In Section 3 we show a connection between $\cB_H$ and univalent harmonic mappings that resembles Pommerenke's theorem \cite{Po70}. We also study the schlicht radius in $\cB_H$. In Section 4 we give growth and coefficients estimates for sense-preserving functions in $\cB_H$.
\section{The class of harmonic Bloch-type mappings}
Our first task will be to show the affine and linear invariance of $\cB_H$. Throughout the paper we will denote by $\vp_\al \, (\al\in\D)$ the disk automorphism given by $\vp_\al(z)=(\al+z)/(1+\overline{\al}z), z\in\D$.
\begin{proposition} If $f\in\mathcal{B}_H$ then
\begin{enumerate}
\item[(i)] $a f+b\overline{f} \in\cB_H$ for any $a,b\in\mathbb C$. (affine invariance) \\
\item[(ii)] $f\circ \vp_\al\in\cB_H$ for any $\al\in\D$. (linear invariance)
\end{enumerate}
\end{proposition}
\begin{proof} Let $f=h+\overline{g}$. To prove (i) we write
$$
F = a f+b\, \overline{f} = a h +b g +\overline{\overline{a}g+\overline{b}h}
$$
and compute
$$
J_F = |a h' +b g'|^2 - |\overline{a}g'+\overline{b}h'|^2 = (|a|^2-|b|^2) \, J_f.
$$
The assertion now easily follows.
For claim (ii) we write $F=f\circ \vp_\al = H+\overline{G}$ and compute
$$
H'(z) = \frac{h'\big(\vp_\al(z)\big) (1-|\al|^2) }{(1+\overline{\al}z)^2}, \qquad G'(z) = \frac{g'\big(\vp_\al(z)\big) (1-|\al|^2) }{(1+\overline{\al}z)^2}.
$$
Hence
\begin{align*}
(1-|z|^2)\sqrt{|J_F(z)|} &= \frac{(1-|z|^2) (1-|\al|^2) }{|1+\overline{\al}z|^2}\sqrt{|J_f\big(\vp_\al(z)\big)|} \\
& = \big(1-|\vp_\al(z)|^2\big) \sqrt{|J_f\big(\vp_\al(z)\big)|}.
\end{align*}
Taking the supremum over $z\in\D$ we get that $\beta(F) = \beta(f)$.
\end{proof}
In what follows, Example \ref{ex-no-space} shows that $\cB_H$ is not a linear space. It also shows that functions in $\cB_H$ may grow arbitrarily fast. Hence, in order to get growth and coefficient estimates in Section 4 we shall restrict ourselves to sense-preserving functions in $\cB_H$.
\begin{example} \label{ex-no-space}
Consider an analytic function $h$ for which $h'(z) = (1-z)^{-p}$, for some $p>2$. Set $f = h +\overline{h} = 2 \,{\rm Re}\, \{h\}$ and see that, since $J_f \equiv 0$, $f$ belongs to $\cB_H$. Obviously, the identity $\text{id}(z)= z$ belongs to $\cB_H$, but we will see that $f+\text{id}$ does not. Indeed,
$$
J_{f+\text{id}} = |h'+1|^2 - |h'|^2 = 1 +2{\rm Re}\, \{h'\}
$$
and therefore, for $0<x<1$ we have
$$
(1-x^2)^2 |J_{f+\text{id}}(x)| = (1+x)^2 \frac{2+(1-x)^p}{(1-x)^{p-2}} \longrightarrow \infty
$$
as $x\to1^-$.
\end{example}
Example \ref{ex-extr} shows that the harmonic Bloch space considered in \cite{Co} is strictly contained in $\cB_H$. Recall that in \cite{Co} the definition of a Bloch function $f= h + \overline{g}$ is equivalent to both $h$ and $g$ belonging to $\cB$.
\begin{example} \label{ex-extr}
Let $f = h + \overline{g}$ be given by $h(z)=\frac{2}{\sqrt{1-z}}$ and $\omega(z)=(g'/h')(z)=z$. Then $f\in\cB_H$ since $h'(z) = (1-z)^{-3/2}$ and
$$
(1-|z|^2) \sqrt{J_f(z)} = \left( \frac{1-|z|^{2}}{|1-z|} \right)^{3/2} \leq 2 \sqrt{2}.
$$
Note that $h\notin\cB$ since, for $0<x<1$, we have
$$
(1-x^2)|h'(x)| = \frac{1+x}{\sqrt{1-x}} \longrightarrow \infty
$$
as $x\to1^-$. Therefore $f$ is not a Bloch function for \cite{Co}.
\end{example}
\section{Univalent functions}
Let $f=h+\overline{g}$ be a harmonic, univalent and sense-preserving mapping in $\D$. Let $\omega = g'/h' : \D\to\D$ be its dilatation and write
$$
h(z)=\sum_{n=0}^\infty a_n z^n \qquad \text{and} \qquad g(z)=\sum_{n=1}^\infty b_n z^n.
$$
We say that $f\in S_H$ if it satisfies $a_0=1-a_1=0$ and that $f\in S_H^0$ if in addition $b_1=0$.
A simple use of the Schwarz Lemma \cite[\S 5.4]{Du} yields the sharp inequality \mbox{$|b_2|\leq 1/2$} for functions in $S_H^0$. It takes more ef\mbox{}fort to prove that $|a_2|<49$ in $S_H^0$ \cite[\S 6.3]{Du}, and still, the best known constant $49$ is quite distant from the conjectured $5/2$.
For the larger class $S_H$, we have that $|b_1|<1$ simply because $f$ is sense-preserving. Also, it is possible to translate the preceding inequalities by means of an affine transformation. Given $f\in S_H$, the function
\be \label{aff}
f_0 = \frac{f-\overline{b_1}\overline{f}}{1-|b_1|^2}
\ee
belongs to $S_H^0$. This transformation is invertible, so that $f =f_0 +\overline{b_1}\overline{f_0}$. Hence, it is not difficult to see that
\be \label{a_2}
|a_2| < 49 + \frac{|b_1|}{2}
\ee
for functions in $S_H$.
In the recent work \cite{HM} a new \emph{Schwarzian} derivative for harmonic locally univalent functions was defined and studied. Also, a \emph{pre-Schwarzian} derivative was defined as
$$
P_f = \frac{h''}{h'} - \frac{\overline{\omega}\omega'}{1-|\omega|^2}
$$
and with it the following \emph{Becker-type} criterion for univalence was proved.
\begin{theoremO}[\cite{HM}] \label{Becker}
Let $f=h+\overline{g}$ be a sense-preserving harmonic function in the unit disk with dilatation $\omega$. If for all $z\in\D$
$$
|zP_f(z)| + \frac{|z\omega'(z)|}{1-|\omega(z)|^2} \leq \frac{1}{1-|z|^2},
$$
then $f$ is univalent.
\end{theoremO}
For any $\omega:\D\to\D$ analytic we define its \emph{hyperbolic derivative} by
$$
\omega^*(z) = \frac{\omega'(z) (1-|z|^2)}{1-|\omega(z)|^2}.
$$
We set $\|\omega\|_h = sup_{z\in\D}|\omega^*(z)|$ for its \emph{hyperbolic norm}. See \cite[\S 5]{BM}.
We recall that a sense-preserving homeomorphism $f$ is called \emph{quasiconformal} if it maps infinitesimal circles onto infinitesimal ellipses having ratio of the major over the minor axis bounded by some constant. This is equivalent to saying that its (second complex) dilatation $\om=\overline{f_{\overline{z}}}/f_z$ is bounded away from one, that is, $|\om(z)| \leq k<1$. See \cite[\S 1.2]{Du}.
\subsection{Connection between univalent and Bloch-type functions.} A well-known theorem of Pommerenke \cite{Po70} gives yet another characterization of the analytic Bloch space $\cB$. It states that a function $f$ is Bloch if and only if there exists a constant $c>0$ and a univalent function $g$ such that $f=c\log g'$. The following theorems show a similar connection between harmonic univalent mappings and the class $\cB_H$.
\begin{theorem} Let $F=H+\overline{G}$ be univalent and sense-preserving in $\D$. Let \mbox{$h=\log(H')$} and consider any $\omega:\D\to\D$ analytic. Then $f=h+\overline{g}$, having dilatation $\omega_f = \omega$, belongs to $\cB_H$.
\end{theorem}
\begin{proof}
Let $\alpha\in\D$ and compose $F$ with a disk automorphism to obtain
$$
T(z) = \frac{F\left(\frac{\alpha +z}{1+\overline{\alpha}z}\right) - F(\alpha) }{(1-|\alpha|^2)H'(\alpha)}.
$$
It can easily be seen that $T\in S_H$ and that the second coefficient of the analytic part of $T$ is given by
$$
a_2(\alpha) = (1-|\alpha|^2)\frac{H''(\alpha)}{2H'(\alpha)} - \overline{\alpha}.
$$
We turn to $f=h+\overline{g}$ and compute
\begin{align*}
(1-|\alpha|^2)\sqrt{J_f(\alpha)} & \, \leq \, (1-|\alpha|^2) |h'(\alpha)| \\
&= \, (1-|\alpha|^2) \left| \frac{H''(\alpha)}{H'(\alpha)} \right| \\
&= \, 2 |a_2(\alpha)+\overline{\alpha}| \\
&< \, 101,
\end{align*}
in view of \eqref{a_2}. The proof is complete.
\end{proof}
In the opposite direction we have the following theorem.
\begin{theorem}
Let $f=h+\overline g \in \cB_H$ be sense-preserving and suppose that $g\in\cB$. Let $0<\ve<1$. Set
$$
H(z) = \int_0^z \exp\left(\frac{\ve}{c} \,h(\ze)\right) d\ze,
$$
where $c = \sqrt{\beta(g)^2 +\beta(f)^2}$, and consider any analytic $\omega:\D\to\D$ satisfying $\|\omega\|_h\leq (1-\ve)/2$. Then $F=H+\overline{G}$, having dilatation $\omega_F = \omega$, is univalent.
\end{theorem}
\begin{proof}
We apply Theorem \ref{Becker} to the function $F$. Since $f\in\cB_H$ and $g\in\cB$, we have that
$$
(1-|z|^2)^2 |h'(z)|^2 \, \leq \, \beta(f)^2 +(1-|z|^2)^2 |g'(z)|^2 \, \leq \, c^2.
$$
Hence
$$
\left| \frac{H''(z)}{H'(z)} \right| \, = \, \frac{\ve}{c} |h'(z)| \, \leq \, \frac{\ve}{1-|z|^2}.
$$
Also, the definition of the hyperbolic norm and our hypothesis lead to
$$
\frac{|\omega'(z)|}{1-|\omega(z)|^2} \, \leq \, \frac{\|\omega\|_h}{1-|z|^2} \, \leq \, \frac{1-\ve}{2(1-|z|^2)}.
$$
We may now compute
\begin{align*}
|zP_F(z)| + \frac{|z\omega'_F(z)|}{1-|\omega_F(z)|^2} \, & \leq \, \left| \frac{H''(z)}{H'(z)} \right| + \frac{2|\omega'(z)|}{1-|\omega(z)|^2} \\
& \leq \, \frac{1}{1-|z|^2}
\end{align*}
and conclude that $F$ is univalent by Theorem \ref{Becker}.
\end{proof}
\subsection{Schlicht radius.} A well-known covering theorem \cite[\S 6.2]{Du} states that all functions in $S_H^0$ contain in their image a disk centered at the origin, having radius $1/16$. (The conjectured constant is $1/6$.) Applying as before the affine transformation \eqref{aff} it is easy to see that
\be \label{cover}
\left\{ w\in\SC \; : \; |w|<\frac{1-|b_1|}{16}\right\} \subset f(\D)
\ee
for every $f\in S_H$ \cite[Corollary 4.5]{CS}.
A result in the opposite direction states that each function in $S_H$ omits some point on the circle $|w|=\frac{\pi}{2}$. In other words
\be \label{Hall}
\big(\SC\backslash f(\D)\big) \bigcap \left\{|w|=\frac{\pi}{2}\right\} \neq \emptyset.
\ee
The constant $\frac{\pi}{2}$ was given by Hall \cite{Ha} and is best possible. See also \cite[\S 6.2]{Du}.
As mentioned in Section 1, the \emph{schlicht radius} $d_f(z)$ of a harmonic mapping $f=h+\overline{g}$ at a point $z\in\D$ is defined as the radius of the largest disk which is the injective image of some subdomain of $\D$ and is centered at $f(z)$. If there is no such disk then we set $d_f(z)=0$. The existence of a universal lower bound for $\sup_{z\in\D}d_f(z)$ is commonly refered to as a \emph{Bloch theorem}. It was shown in \cite{CGH} that openness (\emph{i.e.}$\!$ the property of mapping open sets to open sets) and the normalization $g'(0)=1-h'(0)=0$ are sufficient conditions for a Bloch theorem to hold. Moreover, it was shown that the normalization alone is not a sufficient condition.
Since here we will be concerned only with univalent functions, the schlicht radius coincides with the distance between $f(z)$ and the boundary of $f(\D)$. The following lemma provides us with some estimates.
\begin{lemma} \label{radius}
If $f\in S_H$ then
$$
\frac{1}{16}(1-|z|^2) (|h'(z)|-|g'(z)|) \; \leq \; d_f(z) \; \leq \; \frac{\pi}{2} (1-|z|^2) |h'(z)|,
$$
for all $z\in\D$.
\end{lemma}
\begin{proof}
Let $\al\in\D$ and compose with a disk automorphism to obtain
$$
F(z) = \frac{f\left(\frac{\al +z}{1+\overline{\al}z}\right) - f(\al) }{(1-|\al|^2)h'(\al)} = H(z) + \overline{G(z)}.
$$
Since $F\in S_H$, the covering theorem \eqref{cover} and Hall's result \eqref{Hall} imply that the radius $d_F(0)$ of the largest disk centered at the origin and contained in the image of $F$ satisfies
$$
\frac{1-|B_1|}{16} \; \leq \; d_{F}(0) \; \leq \; \frac{\pi}{2}.
$$
We compute
$$
d_{F}(0) = \frac{d_{f}(\al)}{(1-|\al|^2)|h'(\al)|}
$$
and $B_1=g'(\al)/\overline{h'(\al)}$, the first coefficient of $G$. The inequality follows upon substitution.
\end{proof}
\begin{theorem} \label{rad-B}
Let $f\in S_H$.
\begin{enumerate}
\item[(i)] If $f\in\cB_H$ then $\displaystyle d_f(z) = O\left(\frac{1}{\sqrt{1-|z|}}\right), |z|\to1^-$.
\item[(ii)] If $d_f(z) = O\left(\sqrt{1-|z|}\right), |z|\to1^-$ then $f\in\cB_H$.
\end{enumerate}
If in addition $f$ is quasiconformal then $f\in\cB_H$ if and only if $\,\sup_{z\in\D}d_f(z)<\infty$.
\end{theorem}
We shall need the following lemma. See \cite{Ga}, page 3.
\begin{lemmO} \label{L1}
If $\om:\D\to\D$ is analytic then
$$
|\om(z)| \leq \frac{|\om(0)| + |z|}{ 1 +|\om(0)||z|}.
$$
\end{lemmO}
\begin{proof}[Proof of Theorem \ref{rad-B}]
Note that $f\in\cB_H$ is equivalent to
$$
(1-|z|^2) |h'(z)| \sqrt{1-|\om(z)|^2} \leq \beta(f), \quad z\in\D.
$$
Also note that $\om(0)=b_1$. An application of lemmas \ref{radius} and \ref{L1} yields
$$
d_f(z) \; \leq \; \frac{\pi}{2} (1-|z|^2) |h'(z)| \; \leq \; \frac{\pi}{2} \frac{\beta(f)}{\sqrt{1-|\om(z)|^2}} \; \leq \; \frac{\pi}{2} \sqrt{\frac{1+|b_1|}{1-|b_1|}} \frac{\beta(f)}{\sqrt{1-|z|}},
$$
so that claim (i) is proved. For assertion (ii) we use again lemmas \ref{radius} and \ref{L1} to get
$$
(1-|z|^2)\sqrt{J_f(z)} \; \leq \; 16 \, d_f(z) \sqrt{\frac{1+|\om(z)|}{1-|\om(z)|}} \; \leq \; 16 \sqrt{2} \sqrt{\frac{1+|b_1|}{1-|b_1|}} \frac{d_f(z)}{\sqrt{1-|z|}},
$$
thus $f\in\cB_H$.
Suppose now that $f$ is quasiconformal and see that its dilatation $\om=g'/h':\D\to\D$ satisfies
$$
\|\om\|_\infty \; = \; \sup_{z\in\D} |\om(z)| \; < \; 1.
$$
Arguing as before but using only Lemma \ref{radius} we get
$$
d_f(z) \; \leq \; \frac{\pi}{2} \frac{\beta(f)}{\sqrt{1-\|\om\|_\infty}}
$$
and in the opposite direction
$$
(1-|z|^2)\sqrt{J_f(z)} \; \leq \; 16 \, d_f(z) \sqrt{\frac{1+\|\om\|_\infty}{1-\|\om\|_\infty}}.
$$
The proof is complete.
\end{proof}
\section{Growth and coefficients estimates}
For a harmonic sense-preserving function $f = h + \overline{g}$ with dilatation $\om=g'/h':\D\to\D$, we write the power series
$$
h(z)=\sum_{n=0}^\infty a_n z^n, \quad g(z)=\sum_{n=1}^\infty b_n z^n \quad \text{and} \quad \om(z)=\sum_{n=0}^\infty c_n z^n.
$$
Of course $c_0=b_1/a_1$. We will also make use of the notation
$$
M_\infty(r,f) = \max_{|z|=r} |f(z)|.
$$
We now present some growth and coefficients estimates for the class $\cB_H$. Note, however, that these bounds are not uniform throughout $\cB_H$, but rather, to each of its subclasses having prescribed $|c_0|$.
\begin{theorem}\label{T1}
If $f = h + \overline{g} \in \cB_H$ is sense-preserving then
$$
\max\{|h(z)-a_0|, |g(z)| \} \leq \beta(f) \sqrt{\frac{1+|c_0|}{1-|c_0|}} \frac{r}{\sqrt{1-r^2}}, \quad |z|=r.
$$
This estimate is sharp in order of magnitude.
\end{theorem}
\begin{proof}
Let $|z|=r<1$ and write
$$
h(z)-a_0 \, = \, \int_0^z h'(\ze) d\ze \, = \, z \int_0^1 h'(tz) dt.
$$
We have
$$
|h(z)-a_0| \leq r \int_0^1 |h'(tz)| dt \leq r \int_0^1 \frac{\beta(f)}{(1-r^2t^2) \sqrt{1-|\om(tz)|^2}} dt,
$$
since $f\in\cB_H$. We use Lemma \ref{L1} to get
$$
|h(z)-a_0| \, \leq \, \beta(f) \sqrt{\frac{1+|c_0|}{1-|c_0|}}\, r \int_0^1 \frac{dt }{(1-r^2t^2)^{3/2}}.
$$
We compute the integral
$$
\int_0^1 \frac{dt }{(1-r^2t^2)^{3/2}} \, = \, \frac{1}{\sqrt{1-r^2}}
$$
and thus complete the proof of the desired inequality for the function $h$.
We easily get the same bound for $g$ by computing
$$
|g(z)| \, \leq \, r \int_0^1 |g'(tz)| dt
$$
and using the fact that $|g'|\leq|h'|$.
We now prove the sharpness of the order of magnitude. When $c_0=0$, both inequalities (for functions $h$ and $g$) are optimal in view of example \ref{ex-extr}. Our considerations here will contain this as a special case. We take $f = h +\overline{g}$, for which $h'(z) = (1-z)^{-3/2}$, as in example \ref{ex-extr}, but here we take the dilatation to be a self-map of $\D$ whose image is a horodisk centered at some $t\in[0,1)$, that is, $\omega(z) = (g'/h')(z)= t +(1-t)z$. We see that $f\in\cB_H$ since
\begin{align*}
(1-|z|^2) \sqrt{J_f(z)} & \, = \, \frac{1-|z|^{2}}{|1-z|^{3/2}}\sqrt{1-|\omega(z)|^2} \\
& \, = \, \frac{1-|z|^{2}}{|1-z|} \sqrt{ \frac{1-|z|^{2} -2t {\rm Re}\,\big(\overline{z}(1-z)\big) -t^2|1-z|^2}{|1-z|} }\\
& \, \leq \, 2 \sqrt{2}\sqrt{1+t}.
\end{align*}
The sharpness of the inequality for $h$ is now obvious since $h(z)=\frac{2}{\sqrt{1-z}}$ in our example.
For the function $g$ of this example we compute
$$
g'(z) = \frac{1}{(1-z)^{-3/2}} -\frac{1-t}{\sqrt{1-z}}.
$$
Integrating we get
$$
g(z)=\frac{2}{\sqrt{1-z}}+2(1-t)\sqrt{1-z},
$$
hence, for every $\ve>0$ we have that
$$
(1-x)^{1/2-\ve} |g(x)| \longrightarrow \infty,
$$
when $x\to1^-$. The proof is complete.
\end{proof}
\pagebreak
\begin{theorem}\label{T2}
If $f = h + \overline{g} \in \cB_H$ is sense-preserving then
$$
|a_1| \leq \frac{\beta(f)}{\sqrt{1-|c_0|^2}}
$$
and
$$
\max\{|a_n|, |b_n| \} \leq \beta(f) \left(\frac{e}{3}\right)^{3/2} \sqrt{\frac{1+|c_0|}{1-|c_0|}} \sqrt{n+2}, \quad n\geq2.
$$
\end{theorem}
\begin{proof}
For the first inequality we put $z=0$ in the definition of $\cB_H$ and get
$$
\sqrt{|a_1|^2-|b_1|^2} \, \leq \, \beta(f).
$$
Let $n\geq2$. By Cauchy's formula we have that
$$
|a_n| \, = \, \frac{|h^{(n)}(0)|}{n!} \, = \, \left| \frac{1}{n\,2\pi i} \int_{|\ze|=r} \frac{h'(\ze)}{\ze^n} d\ze\right| \, \leq \, \frac{M_\infty(r,h')}{n \, r^{n-1}},
$$
for any $r\in(0,1)$. Similarly, and also due to the fact that $f$ is sense-preserving, we have that
$$
|b_n| \, \leq \, \frac{M_\infty(r,g')}{n \, r^{n-1}} \, \leq \, \frac{M_\infty(r,h')}{n \, r^{n-1}}.
$$
The definition of $\cB_H$ implies that
$$
\max\{|a_n|, |b_n| \} \, \leq \, \frac{\beta(f)}{n \, r^{n-1}(1-r^2)\sqrt{1-M_\infty^2(r,\om)}}.
$$
Using Lemma \ref{L1} we get that
\begin{align*}
\max\{|a_n|, |b_n| \} \, & \leq \, \frac{\beta(f) (1+|c_0|r)}{n \, r^{n-1}(1-r^2)^{3/2} \sqrt{1-|c_0|^2}} \\
& \leq \, \frac{\beta(f)}{n} \sqrt{\frac{1+|c_0|}{1-|c_0|}} \frac{1}{r^{n-1}(1-r^2)^{3/2}}.
\end{align*}
This inequality is true for all $r$ in $(0,1)$. Therefore, in order to minimize the expression on the right hand side we see that $r^{n-1}(1-r^2)^{3/2}$ is maximized for $r=\sqrt{\frac{n-1}{n+2}}$. Making this choice we get
\begin{align*}
\max\{|a_n|, |b_n| \} \leq & \frac{\beta(f)}{n} \sqrt{\frac{1+|c_0|}{1-|c_0|}} \left(\frac{n+2}{n-1}\right)^{\frac{n-1}{2}} \left(\frac{n+2}{3}\right)^{3/2} \\
= & \frac{\beta(f)}{3\sqrt{3}} \sqrt{\frac{1+|c_0|}{1-|c_0|}} \vp(n) \sqrt{n+2},
\end{align*}
where
$$
\vp(x) = \left[\left( 1+ \frac{3}{x-1}\right)^{\frac{x-1}{3}}\right]^{3/2} \left(1+\frac{2}{x}\right).
$$
Note that $\vp(x)\to e^{3/2}$ when $x\to+\infty$. We will now show that $\vp$ increases to its limit. First note that $\vp(x)>0$ for $x\geq2$. We compute
$$
\log\vp(x) = \frac{x-1}{2} \log\left(\frac{x+2}{x-1}\right) +\log\left(\frac{x+2}{x}\right).
$$
Differentiating we get
$$
\psi(x) := \frac{\vp'(x)}{\vp(x)} = \frac{1}{2} \log\left(\frac{x+2}{x-1}\right) - \frac{3x+4}{2x(x+2)}.
$$
One more differentiation yields
$$
\psi'(x) = -\frac{x^2+8}{2x^2(x+2)^2(x-1)},
$$
which for $x\geq2$ obviously satisfies $\psi' <0$. Therefore $\psi$ decreases, so that
$$
\psi(x) > \lim_{x\to\infty} \psi(x) =0,
$$
hence $\vp' >0 $ and the proof is complete.
\end{proof}
| 182090120b768ae1b737b18dd680539ecf27ca0c | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
\label{sec:intro}
As a key part of the human visual system, visual attention mechanisms filter irrelevant visual stimuli in order to focus more on the important parts. Computational models of attention try to mimic this process through the use of machines and algorithms. These models have gained increasing attention lately. The reason behind this growing interest lies in their use in different computer vision and multimedia problems including but not limited to image retrieval~\cite{5730497}, visual quality assessment~\cite{7303954,7444164} video resizing/summarization~\cite{6529173,6527322}, action recognition~\cite{Wang2017}, event detection~\cite{Gan2015} either as a tool for visual feature extraction or as a mechanism for selecting features. These models are also important for generic applications such as advertisement and web design~\cite{7294708} as attention plays a key role in both user interfaces and human-machine interaction.
In the literature, the computational attention models developed so far generally aim to predict where humans fixate their eyes in images~\cite{borji2013state}. Specifically, they produce the so-called saliency maps from the visual data where a high saliency score at an image location indicates that the point is more likely to be fixated. These models are largely inspired by the hierarchical processing of human visual system~\cite{Hubel1962} and the theoretical studies like Feature Integration Theory~\cite{Treisman:CogPsych1980} or Guided Search Model~\cite{Wolfe:PsychBul1994}. They consider low-level image features (color, orientation, contrast, motion, etc.) and/or high-level features (pedestrians, faces, text) while predicting saliency maps. In this process, while low-level features are employed to examine how different an image point from its surroundings, high-level features are employed as it is experimentally shown that humans have a tendency to fixate on certain object classes more than others.
Another current trend in the existing literature is to detect salient objects~\cite{6331539,borji2015salient,7347445,7514760} from the images. These models specifically aim at identifying the most prominent objects in an image that attract attention under free-viewing conditions and then segmenting them out from the background. These computational models for visual attention can be further grouped into two according to their inputs as static and dynamic saliency models. While static models work on images, dynamic models take video sequences as input.
Saliency prediction from videos leads to great challenges when it is compared to carrying out the same task in still images. The reason is that the dynamic saliency frameworks require taking into account both spatial and temporal characteristics of the video sequences. Static saliency models use features like color, intensity and orientation, however dynamic models need to focus more on the moving objects or image parts as it is shown that humans there is a tendency for humans to look at them while viewing. Hence, the preliminary models proposed for dynamic saliency prediction extend the existing saliency models suggested for still images so that they consider extra motion features~\cite{harel2006graph,guo2008spatio,cui2009temporal,seo2009static}. However, more recent works approach the same task from a different point of view and propose novel solutions~\cite{hou2007saliency,mathe2012dynamic,rudoy2013learning}.
\subsection{Overview of our approach}
Deep learning has been successfully applied to saliency prediction in still images in the last few years, providing state-of-the-art results~\cite{Kümmerer2014b,pan2016shallow,kruthiventi2015deepfix,vig2014large,zhao2015saliency,Bruce_2016_CVPR,Jetley_2016_CVPR}. The early models utilize pre-trained deep convolutional neural networks (CNNs) proposed to classify images as generic feature extractors and build classifiers on top of those features to classify fixated image regions~\cite{vig2014large,Kümmerer2014b}. Later models, however, approach the problem from an end-to-end perspective and either train networks from scratch or most of the time fine-tune the weights of a pre-trained model~\cite{Kruthiventi_2016_CVPR, Jetley_2016_CVPR,pan2016shallow}. The modifications in the network architectures are usually about integrating multi-scale processing or using different loss functions~\cite{HuangSALICON2015,Jetley_2016_CVPR}. It has been investigated that the power of these deep models mainly comes from the property that the features learned by these networks are semantically very rich~\cite{Bruce_2016_CVPR}, capturing high-level factors important for saliency detection. Motivated by the success of these works, in this study, we explore the use of two-stream CNNs for saliency prediction from videos. To the best of our knowledge, our work is the first deep model for dynamic saliency, which is trained in an end-to-end manner, that learns to combine spatial and temporal information in an optimal manner within a two-stream network architecture.
\subsection{Our contributions}
The contributions of our work can be summarized as follows:
\begin{enumerate}
\item We study two-stream convolutional neural networks which mimic the visual pathways in the brain and combine networks trained on temporal and spatial information to predict saliency map of a given video frame. Although these network architectures have been previously investigated for some computer vision problems such as video classification~\cite{karpathy2014large} and action recognition~\cite{simonyan14b}, to our knowledge, we are the first to apply two-stream deep models for saliency prediction from videos in the literature. In particular, in our study, we investigate two different fusion strategies, namely element-wise and convolutional fusion strategies, to integrate spatial and temporal streams.
\item We carry out extensive experiments on DIEM ~\cite{mital2011clustering} and UCF-Sports~\cite{MatheSminchisescuPAMI2015} datasets and compare our deep spatio-temporal saliency networks against several state-of-the-art dynamic saliency models. Our evaluation demonstrates that the proposed STSConvNet model outperforms these models in nearly all of the evaluation metrics on these datasets.
\item On a number of challenging still images, we also show that our spatio-temporal saliency network can predict the human fixations better than the state-of-the-art deep static saliency models. The key idea that we follow is to extract optical flow from these static images by using a recently proposed method~\cite{walker2015dense} and feed them to our network along with the appearance image.
\end{enumerate}
\section{Related Work}
In this study, we focus on bottom-up modeling of dynamic saliency. Below, we first summarize the existing dynamic saliency models from the literature and then provide a brief overview of the proposed deep-learning based static saliency models which are related to ours.
\subsection{Dynamic Saliency}
Early examples of saliency models for dynamic scenes extend the previously proposed static saliency models which process images in a hierarchical manner by additionally considering features related to motion such as optical flow. For instance, in~\cite{harel2006graph}, Harel \emph{et al.} propose a graph-theoretic solution to dynamic saliency by representing the extracted feature maps in terms of fully connected graphs and by predicting the final saliency map. In~\cite{cui2009temporal}, Cui \emph{et al.} extract salient parts of video frames by performing spectral residual analysis on the Fourier spectrum of these frames over the spatial and the temporal domains. In~\cite{guo2008spatio}, Guo \emph{et al.} propose a similar spectral analysis based formulation. In~\cite{sultani2014human}, Sultani and Saleemi extend Harel \emph{et al.}~\cite{harel2006graph}'s model by using additional features such as color and motion gradients and by post-processing the predicted maps via a graphical model based on Markov Random Fields. In~\cite{seo2009static}, Seo and Milanfar employ self similarities of spatio-temporal volumes to predict saliency. ~\cite{mauthner2015encoding}, Mauthner \emph{et al.} also present a video saliency detection method for using as a prior information for activity recognition algorithms. Instead of using a data driven approach they propose an unsupervised algorithm for estimating salient regions of video sequences.
Following these early works, other researchers rather take different perspectives and devise novel solutions for dynamic saliency. For example, in~\cite{hou2007saliency}, Hou and Zhang consider rarity of visual features while extracting saliency maps from videos and propose an entropy maximization-based model. In~\cite{rahtu2010segmenting}, Rahtu \emph{et al.} extract saliency by estimating local contrast between feature distributions. In~\cite{Ren2012}, Ren \emph{et al.} propose a unified model with the temporal saliency being estimated by sparse and low-rank decomposition and the spatial saliency map being extracted by considering local-global contrast information. In~\cite{mathe2012dynamic}, Mathe \emph{et al.} devise saliency prediction from videos as a classification task where they integrate several visual cues through learning-based fusion strategies. In another study, Rudoy \emph{et al.}~\cite{rudoy2013learning} propose another learning based model for dynamic saliency. It differs from Mathe \emph{et al.}'s model~\cite{mathe2012dynamic} in that they take into account a sparse set of gaze locations thorough which they predict conditional gaze transitions over subsequent video frames. Zhou \emph{et al.}~\cite{zhou2014learning} oversegment video frames and use low-level features from the extracted regions to estimate regional contrast values. Zhao \emph{et al.}~\cite{zhao2015fixation} learn a bank of filters for fixations and use it to model saliency in a location-dependent manner. Khatoonabadi \emph{et al.}~\cite{hossein2015many} propose a saliency model that depends on compressibility principle. In a very recent study, Leboran \emph{et al.}~\cite{awsd}, propose another dynamic saliency model by using the idea that perceptual relevant information is carried by high-order statistical structures.
\subsection{Deep Static Saliency}
In recent years, deep neural networks based models provide state-of-the-art results in many computer vision problems such as image classification~\cite{he2015deep}, object detection~\cite{girshick2015fast}, activity recognition~\cite{zhou2014learning}, semantic segmentation~\cite{long2015fully} and video classification~\cite{karpathy2014large}. These approaches perform hierarchical feature learning specific to a task, and thus gives results better than the engineered features. Motivated by the success of these models, a number of researchers have recently proposed deep learning based models for saliency prediction from images~\cite{Kümmerer2014b,pan2016shallow,kruthiventi2015deepfix,vig2014large,Bruce_2016_CVPR,Jetley_2016_CVPR,Li2016}.
Vig \emph{et al.}~\cite{vig2014large} use an ensemble of CNNs which learns biologically inspired hierarchical features for saliency prediction. K\"{u}mmerer \emph{et al.}~\cite{Kümmerer2014b} employ deep features learned through different layers of the AlexNet~\cite{krizhevsky2012imagenet} and learn how to integrate them for predicting saliency maps. Kruthiventi et al.~\cite{kruthiventi2015deepfix} adopt VGGNet~\cite{simonyan15b} for saliency estimation where they introduce a location-biased convolutional layer to model the center-bias, and train the model on SALICON dataset using Euclidean loss. Jetley \emph{et al.}~\cite{Jetley_2016_CVPR} also use the VGGNet architecture but they especially concentrate on investigating different kinds of probability distance measures to define the loss function. Pan \emph{et al.}~\cite{pan2016shallow} very recently propose two CNN models having different layer sizes by approaching saliency prediction as a regression task. Li \emph{et al.}~\cite{Li2016} employ a fully convolutional neural network within a multi-task learning framework to jointly detect saliency and perform object class segmentation. It is important to note that all these models are proposed for predicting saliency in still images not videos. Bruce \emph{et al.}~\cite{Bruce_2016_CVPR} propose yet another fully convolutional network to predict saliency and they try to understand factors and learned representations when training these type of networks for saliency estimation.
Motivated by the deep static saliency models, in our paper we investigate the use of two-stream CNNs for saliency prediction from videos. In fact, investigating layered formulations is not new for saliency prediction. As discussed earlier, most of the traditional dynamic saliency models are inspired from the hierarchical processing in the low-level human vision~\cite{Hubel1962}. These models, however, employ hand-crafted features to encode appearance and motion contrast to predict where humans look at in dynamic scenes. Since they depend on low-level cues, they often fail to capture semantics of scenes at its full extent, which is evidently important for gaze prediction. More recent models, on the other hand, employ learning-based formulations to integrate these low-level features with the detection maps for faces, persons, and other objects. This additional supervision boosts the prediction accuracies, however, the performance is limited by the discrimination capability of the considered features and the robustness of the employed detectors.
As compared to the previous dynamic saliency models, our deep spatio-temporal saliency networks are trained to predict saliency in an end-to-end manner. This allows us to learn hierarchical features, both low-, mid- and high-level,~\cite{Cichy2016,bylinskii2016should} that are specialized for the gaze prediction task. For instance, while the early layers learn filters that are sensitive to edges or feature contrasts, the filters in the top layers are responsible from capturing more complex, semantic patterns that are important for the task. In our case, our deep two-stream saliency networks learn multiple layers of spatial and temporal representations and ways to combine them to predict saliency.
In particular, in our study we extract temporal information via optical flow between consecutive video frames and investigate different ways to use this additional information in saliency prediction within a deep two-stream spatio-temporal network architecture~\cite{simonyan14b}. These two-stream networks are simple to implement and train, and to our interest, are in line with the hierarchical organization of the human visual system. Specifically, the biological motivation behind these architectures is the so-called two-streams hypothesis~\cite{two-stream-hypothesis} which speculate that human visual cortex is comprised of two distinct streams, namely ventral and the dorsal streams, which are respectively specialized to process appearance and motion information.
Here, an alternative deep architecture could be to stack two or more frames together and feeding this input to a deep single-stream CNN, which was investigated in several action recognition networks~\cite{7410867,6165309,Taylor2010,7410879}. In this work, we do not pursue this direction because of two reasons. Firstly, this approach requires learning 3D convolutional filters~\cite{7410867,6165309,Taylor2010} in order to capture spatio-temporal regularities among input video frames but using 3D filters highly increases the complexity of the models and these 3D convolutional networks are harder to train with limited training data~\cite{7410879} (which is the case for the existing dynamic saliency datasets). Secondly, 3D convolutional filters are mainly used for expressing long-range motion patterns which could be important for recognizing an action since they cannot easily be captured by optical-flow based two-stream models. For dynamic saliency prediction though, we believe that such long-range dependencies are minimally important as human attention shifts continuously, and optical flow information is sufficient to establish the link between motion and saliency.
\begin{figure*}[!t]
\centering
\includegraphics[width=\textwidth]{sources/saliency-networks.jpg} \\
\begin{tabular}{cp{1cm}p{1cm}c}
\small{(a) Single stream saliency networks} & & & \small{(b) Two-stream saliency networks}\\
\end{tabular}
\caption{(a) The baseline single stream saliency networks. While SSNet utilizes only spatial (appearance) information and accepts still video frames, TSNet exploits only temporal information whose input is given in the form of optical flow images. (b) The proposed two-stream spatio-temporal saliency networks. STSMaxNet performs fusion by using element-wise max fusion, whereas STSConvNet employs convolutional fusion after the fifth convolution layers.}
\label{fig:saliency-networks}
\end{figure*}
\section{Our Models}
The aim of our study is to investigate the use of deep architectures for predicting saliency from dynamic scenes. Recently, CNNs provided drastically superior performance in many classification and regression tasks in computer vision. While the lower layers of these networks respond to primitive image features such as edges, corners and shared common patterns, the higher layers extract semantic information like object parts, faces or text~\cite{Bruce_2016_CVPR,bylinskii2016should}. As mentioned before, such low and high-level features are shown to be both important and complementary in estimating visual saliency. Towards this end, we examine two baseline single stream networks (spatial and temporal) given in Figure~\ref{fig:saliency-networks}(a) and two two-stream networks~\cite{simonyan14b} shown in Figure~\ref{fig:saliency-networks}(b), which combine spatial and temporal cues via two different integration mechanisms: element-wise max fusion and convolutional fusion, respectively. We describe these models in detail below.
\subsection{Spatial Saliency Network}
For the basic single stream baseline model, we retrain the recently proposed static saliency model in~\cite{pan2016shallow} for dynamic saliency prediction by simply ignoring temporal information and using the input video frame alone. Hence, this model does not consider the inter-frame relationships while predicting saliency for a given video. As shown in the top row of Figure~\ref{fig:saliency-networks}(a), this CNN resembles the VGG-M model~\cite{simonyan15b} -- the main difference being that the final layer is a deconvolution (fractionally strided convolution) layer to up sample to the original image size. Note that it does not use any temporal information and exploits only appearance information to predict saliency in still video frames. We refer to this network architecture as SSNet.
\subsection{Temporal Saliency Network}
Saliency prediction from videos is inherently different than estimating saliency from still images in that our attention is highly affected by the local motion contrast of the foreground objects. To understand the contribution of temporal information to the saliency prediction, we develop a second single stream baseline. As given in the bottom row of Figure~\ref{fig:saliency-networks}(a), this model is just a replica of the spatial stream net but the input is provided in the form of optical flow images, as in~\cite{simonyan14b}, computed from two subsequent frames. We refer to this single stream network architecture as TSNet.
\subsection{Spatio-Temporal Saliency Network with Direct Averaging}
As a baseline model, we define a network model which integrates the responses of the final layers of the spatial and the temporal saliency networks by using direct averaging. Note that this model does not consider a learning strategy on how to combine these two-stream network and consider each one of the single-stream networks equally reliable. We refer to this two-stream network architecture as STSAvgNet.
\subsection{Spatio-Temporal Saliency Network with Max Fusion}
This network model accepts both a video frame and the corresponding optical flow image as inputs and merges together the spatial and temporal single stream networks via element-wise max fusion. That is, given two feature maps $\mathbf{x}^{s},\mathbf{x}^{t}\in \mathbb{R}^{H\times W \times D}$ from the spatial and temporal streams, with $W,H,D$ denoting the width, height and the number of channels (filters), max fusion takes the maximum of these two feature maps at every spatial location $i$ and $j$, and channel $d$, as:
\begin{equation}
y^{max}_{i,j,d} = \max\left(x^{s}_{i,j,d} , x^{t}_{i,j,d}\right)\;.
\end{equation}
The use of the $\max$ operation makes the ordering of the channels in a convolutional layer arbitrary. Hence, this fusion strategy assumes arbitrary correspondences between the spatial and temporal streams. That said, this spatio-temporal model seeks filters so that these arbitrary correspondences between feature maps become as meaningful as possible according to the joint loss. After this fusion step, it also uses a deconvolution layer to produce an up-sampled saliency map as the final result as illustrated in the top row of Figure~\ref{fig:saliency-networks}(b). We refer to this two-stream network architecture as STSMaxNet.
\subsection{Spatio-Temporal Saliency with Convolutional Fusion}
This network model integrates spatial and temporal streams by applying convolutional fusion. That is, the corresponding feature maps $\mathbf{x}^{s}$ and $\mathbf{x}^{t}$ respectively from the spatial and temporal streams are stacked together and then combined as follows:
\begin{equation}
\mathbf{y}^{conv} = \left[ \mathbf{x}^{s} \quad\mathbf{x}^{t}\right]*\mathbf{f}+\mathbf{b}\;,
\end{equation}
where $\mathbf{f}\in\mathbb{R}^{1\times 1 \times 2D \times D}$ denotes a bank of $1\times 1$ filters, and $\mathbf{b}\in \mathbb{R}^D$ represents the bias term.
The main advantage of the convolutional fusion over the element-wise max fusion is that the filterbank $\mathbf{f}$ learns the optimal correspondences between the spatial and temporal streams based on the loss function, and reduces the number of channels by a factor of two through the weighted combinations of $\mathbf{x}^{s}$ and $\mathbf{x}^{t}$ with weights given by $\mathbf{f}$ at each spatial location. As demonstrated in the bottom row of Figure~\ref{fig:saliency-networks}(b), this is followed by a number of convolutions and a final deconvolution layer to produce the saliency map. We refer to this network architecture as STSConvNet.
\subsection{Spatio-Temporal Saliency Network with Direct Fusion}
Finally, as another baseline model, we design a single stream network in which the appearance and optical flow images are stacked together and fed to the network as input. This model implements an early fusion strategy at the very beginning of the network architecture and can be seen as a special case of STSConvNet. Here, each layer of the network learns a set of filters that directly acts on the given appearance and motion frames. We refer to this model as STSDirectNet.
\section{Implementation Details}
\label{ssec:implementation_details}
\subsection{Network Architectures}
The architecture of our single stream models is the same with that of the deep convolution network proposed in~\cite{pan2016shallow}. They take $320\times 240\times 3$ pixels images and processes them by the following operations: $C(96,7,3)$ $\rightarrow$ $LRN$ $\rightarrow$ $P$ $\rightarrow$ $C(256,5,2)$ $\rightarrow$ $P$ $\rightarrow$ $C(512,3,1)$ $\rightarrow$ $C(512,5,2)$ $\rightarrow$ $C(512,5,2)$ $\rightarrow$ $C(256,7,3)$ $\rightarrow$ $C(128,11,5)$ $\rightarrow$ $C(32,11,5)$ $\rightarrow$ $C(1,13,6)$ $\rightarrow$ $D$. Here, $C(d,f,p)$ represents a convolutional layer with $d$ filters of size $f\times f$ applied to the input with padding $p$ and stride 1. $LRN$ denotes a local response normalization layer that carries out a kind of lateral inhibition, and $P$ indicates a max pooling layer over $3\times 3$ regions with stride 2. Finally, $D$ is a deconvolution layer with filters of size $8\times 8\times 1$ with stride 4 and padding 2 which upscales the final convolution results to the original size. All convolutional layers except the last one are followed by a ReLU layer. Our spatial and temporal stream models in particular differ from each other in their inputs. While the first one processes still images, the next one accepts optical flow images as input.
For the proposed spatio-temporal saliency networks shown in Figure~\ref{fig:saliency-networks}(b), we employ element-wise max and convolutional fusion strategies to integrate the spatial and temporal streams. Performing fusion after the fifth convolutional layer gives the best results for both of these fusion strategies. In STSMaxNet, the single stream networks are combined by applying element-wise max operation, which is followed by the same deconvolution layer in the single stream models. On the other hand, STSConvNet performs fusion by stacking the feature maps together and integrating them by a convolution layer $C(512,1,0)$ whose weights are initialized with identity matrices. The remaining layers are the same with those of the single stream models.
\subsection{Data Preprocessing}
We employ three publicly available datasets, 1.DIEM (Dynamic Images and Eye Movements)~\cite{mital2011clustering}, 2. UCF-Sports~\cite{MatheSminchisescuPAMI2015} datasets and 3. MIT 300 dataset~\cite{mit-saliency-benchmark}, which are described in detail in Section~\ref{sec:experiments}, in our experiments. Since our networks accept inputs of size $320\times 240\times 3$ pixels and outputs saliency maps of the same size, all videos and ground truth fixation density maps are rescaled to this size prior to training. We use the publicly available implementation of DeepFlow~\cite{weinzaepfel2013deepflow} and we additionally extract optical flow information from the rescaled versions of subsequent video frames. Optical flow images are then generated by stacking horizontal and vertical flow components and the magnitude of the flow together. Some example optical flow images are shown in Figure~\ref{fig:opticalflow}.
\begin{figure}[!t]
\centering
\begin{tabular}{c@{\;}c@{\;}c}
\includegraphics[width=0.15\textwidth,height=1.8cm]{sources/ucfd1.jpg} &
\includegraphics[width=0.15\textwidth,height=1.8cm]{sources/ucfg1.jpg} &
\includegraphics[width=0.15\textwidth,height=1.8cm]{sources/ucfc1.jpg}\\
\includegraphics[width=0.15\textwidth,height=1.8cm]{sources/ucfd_flow.jpg} &
\includegraphics[width=0.15\textwidth,height=1.8cm]{sources/ucfg_flow.jpg} &
\includegraphics[width=0.15\textwidth,height=1.8cm]{sources/ucfc_flow.jpg}
\end{tabular}
\caption{Sample optical flow images generated for some frames of a video sequence from UCF-Sports dataset.\vspace{-6pt}}
\label{fig:opticalflow}
\end{figure}
\subsection{Data Augmentation}
Data augmentation is a widely used approach to reduce the effect of over-fitting and improve generalization of neural networks. For saliency prediction, however, classical techniques such as cropping, horizontal flipping, or RGB jittering are not very suitable since they alter the visual stimuli used in the eye tracking experiments in collecting the fixation data. Having said that, horizontal flipping is used in~\cite{pan2016shallow} as a data augmentation strategy although there is no theoretical basis for why this helps to obtain better performance.
In our study, we propose to employ a new and empirically grounded data augmentation strategy for specifically training saliency networks. In~\cite{judd2011fixations}, the authors performed a thorough analysis on how image resolution affects the exploratory behavior of humans through an eye-tracking experiment. Their experiments revealed that humans are quite consistent about where they look on high and low-resolution versions of the same images. Motivated with this observation, we process all video sequences and produce their low-resolution versions by down-sampling them by a factor of 2 and 4, and use these additional images with the fixations obtained from original high-resolution images in training. We note that in reducing the resolution of optical flow images the magnitude should also be rescaled to match with the down-sampling rate. It is worth-mentioning that this new data augmentation strategy can also be used for boosting performances of deep models for static saliency estimation.
\subsection{Training}
We employ the weights of the pretrained CNN model in~\cite{pan2016shallow} to set the initial weights of our spatial and temporal stream networks. In training the models, we use Caffe framework~\cite{jia14} and employed Stochastic Gradient Descent with Euclidean distance between the predicted saliency map and the ground truth. The networks were trained over 200K iterations where we used a batch size of 2 images, momentum of 0.9 and weight decay of 0.0005, which is reduced by a factor of 0.1 at every 10K iterations. Depending on the network architectures, it takes between 1 day to 3 days to train our models on the DIEM dataset by using a single 2GB GDDR5 NVIDIA GeForce GTX 775M GPU on a desktop PC equipped with 4-core Intel i5 (3.4 GHz) Processor and 16 GB memory. At test time, it takes approximately 2 secs to extract the saliency map of a single video frame.
\section{Experimental Results}
\label{sec:experiments}
In the following, we first review the datasets on which we perform our experiments and provide the list of state-of-the-art computational saliency models that we compared against our spatio-temporal saliency networks. We then provide the details of the quantitative evaluation metrics that are used to assess the model performances. Next, we discuss our experimental results.
\subsection{Datasets}
To validate the effectiveness of the proposed deep dynamic saliency networks, our first set of experiments is carried out on the DIEM dataset~\cite{mital2011clustering}. This dataset is collected for the purpose of investigating where people look at dynamic scenes. It includes 84 high-definition natural videos including movie trailers, advertisements, etc. Each video sequence has eye fixation data collected from approximately 50 different human subjects. In our experiments, we only used the right-eye positions of the human subjects as done in~\cite{borji2013quantitative}.
We perform our second set of experiments on the UCF-Sports ~\cite{MatheSminchisescuPAMI2015}. This dataset is collected from broadcast television channels such as the BBC and ESPN which consists of a set of sport actions~\cite{MatheSminchisescuPAMI2015}. The video sequences are collected from wide range of websites. This dataset contains 150 video sequences with $720\times 480$ resolution and cover a range of scene and viewpoints. The dataset includes several actions such as diving, golf swing, kicking and lifting, and is used for action recognition. However, recently, additional human gaze annotations were collected in~\cite{MatheSminchisescuPAMI2015}. These fixations are collected over 16 human subjects under task specific and task-independent free viewing conditions.
Lastly, for the experiments on predicting eye fixations on still images, we choose a number of images from the MIT 300 dataset~\cite{mit-saliency-benchmark}. Selected images are the ones especially depicting an action and including objects that are interpreted as in motion. This dataset has eye fixation data collected from 39 subjects under free-viewing conditions for 3 secs for a total of 300 natural images with longest dimension 1024 pixels and the other dimension varied from 457 to 1024 pixels.
\subsection{The compared computational saliency models}
Through our experiments on DIEM and UCF-Sports datasets, we compare our deep network models with eight state-of-the-art dynamic saliency models: GVBS~\cite{harel2006graph}, PQFT~\cite{guo2008spatio}, SR~\cite{hou2007saliency}, Seo and Milanfar~\cite{seo2009static}, Rudoy \emph{et al.}~\cite{rudoy2013learning}, Fang \emph{et al.}~\cite{fang2014}, Zhou \emph{et al.}~\cite{zhou2014learning}, and DWS~\cite{awsd} models. Moreover, we compare our two-stream deep models STSMaxNet and STSConvNet to a certain extent with deep static saliency model DeepSal~\cite{pan2016shallow} as the architectures of our TSNet and SSNet models are adapted from this model.
\subsection{Evaluation Measures}
We employ Area Under Curve (AUC), shuffled AUC (sAUC)~\cite{zhang2008sAUC}, Pearson's Correlation Coefficient (CC), Normalized Scanpath Saliency (NSS)~\cite{nss}, Normalized Cross Correlation (NCC) and ${\chi}^2$ distance throughout our experiments for performance evaluation. We note that NCC and ${\chi}^2$ distance are not widely-used measures but we report them as some recent studies~\cite{rudoy2013learning,zhao2015fixation,mauthner2015encoding} employ them in their analysis.
AUC measure considers the saliency maps as a classification map and uses the receiver operator characteristics curve to estimate the effectiveness of the predicted saliency maps in capturing the ground truth eye fixations. An AUC value of~1 indicates a perfect match with the ground truth while the performance of chance is indicated by a value close to 0.5. The AUC measure cannot account for the tendency of human subjects to look at the image center of the screen, i.e. the so-called center bias. Hence, we also report the results of the shuffled version of AUC (sAUC) where the center bias is compensated by selecting the set of fixations used as the false positives from another randomly selected video frame from the dataset instead of the processed frame.
CC treats the predicted saliency and the ground truth human fixation density maps as random variables and measures the strength of a linear relationship between two using a Gaussian kernel density estimator. While a value close to 0 indicates no correlation, a CC value close to +1/-1 demonstrates a good linear relationship. NSS estimates the average normalized saliency score value by examining the responses at the human fixated locations on the predicted saliency map which has been normalized to have zero mean and unit standard deviation. While a NSS value of 0 indicates chance in predicting eye fixtions, a non-negative NSS value, especially that of greater than 1, denotes correspondence between maps above chance.
NCC is a general measure used for assessing image similarity. It treats the ground truth fixation map and the predicted saliency map as images and estimates a score with values close to 1 implying high similarity and negative values indicating low similarity. ${\chi}^2$ distance considers the saliency maps as a probability distribution map and compares the predicted map with the ground truth human fixation map accordingly. A perfect prediction model needs to provide a distance close to~0 for the ${\chi}^2$ distance.
\begin{figure*}[!t]
\centering
\begin{tabular}{c@{\;}c@{\;}c@{\;}c@{\;}c@{\;}c@{\;}c}
\includegraphics[width=0.135\textwidth]{sources/159_motion.jpg} &
\includegraphics[width=0.135\textwidth]{sources/159_gt.jpg} &
\includegraphics[width=0.135\textwidth]{sources/159_ssnet.jpg} &
\includegraphics[width=0.135\textwidth]{sources/159_tsnet.jpg} &
\includegraphics[width=0.135\textwidth]{sources/159_avgnet2.jpg} &
\includegraphics[width=0.135\textwidth]{sources/159_stsmaxnet.jpg} &
\includegraphics[width=0.135\textwidth]{sources/159_stsconvnet.jpg} \\
\includegraphics[width=0.135\textwidth]{sources/321_motion.jpg} &
\includegraphics[width=0.135\textwidth]{sources/321_gt.jpg} &
\includegraphics[width=0.135\textwidth]{sources/321_ssnet.jpg} &
\includegraphics[width=0.135\textwidth]{sources/321_tsnet.jpg} &
\includegraphics[width=0.135\textwidth]{sources/321_avgnet2.jpg} &
\includegraphics[width=0.135\textwidth]{sources/321_stsmaxnet.jpg} &
\includegraphics[width=0.135\textwidth]{sources/321_stsconvnet.jpg} \\
\includegraphics[width=0.135\textwidth]{sources/191_motion.jpg} &
\includegraphics[width=0.135\textwidth]{sources/191_gt.jpg} &
\includegraphics[width=0.135\textwidth]{sources/191_ssnet.jpg} &
\includegraphics[width=0.135\textwidth]{sources/191_tsnet.jpg} &
\includegraphics[width=0.135\textwidth]{sources/191_avgnet2.jpg} &
\includegraphics[width=0.135\textwidth]{sources/191_stsmaxnet.jpg} &
\includegraphics[width=0.135\textwidth]{sources/191_stsconvnet.jpg} \\
\includegraphics[width=0.135\textwidth]{sources/240_motion.jpg} &
\includegraphics[width=0.135\textwidth]{sources/240_gt.jpg} &
\includegraphics[width=0.135\textwidth]{sources/240_ssnet.jpg} &
\includegraphics[width=0.135\textwidth]{sources/240_tsnet.jpg} &
\includegraphics[width=0.135\textwidth]{sources/240_avgnet2.jpg} &
\includegraphics[width=0.135\textwidth]{sources/240_stsmaxnet.jpg} &
\includegraphics[width=0.135\textwidth]{sources/240_stsconvnet.jpg} \\
\small{Overlayed images} & \small{Ground Truth} & \small SSNet & \small TSNet & \small STSAvgNet & \small STSMaxNet & \small STSConvNet
\end{tabular}
\caption{Qualitative evaluation of the proposed saliency network architectures. For these sample frames from the DIEM dataset, our STSConvNet provides the most accurate prediction as compared to the other network models.}
\label{fig:network-comparisons}
\end{figure*}
\subsection{Experiments on DIEM}
In our analysis, we first both qualitatively and quantitatively compare our proposed deep dynamic saliency networks, SSNet, TSNet, STSDirectNet, STSAvgNet, STSMaxNet and STSConvNet, with each other. Following the experimental setup of~\cite{borji2013quantitative}, we split the dataset into a training set containing 64 video sequences and a testing set including the remaining 20 representative videos covering different concepts. Specifically, we use the same set of splits used in~\cite{rudoy2013learning}.
As our STSMaxNet and STSConvNet models integrate spatial and temporal streams by respectively using element-wise and convolutional fusion strategies, we perform an extensive set of initial experiments to determine the optimum layers for the fusion process to take place in these two-stream networks. In particular, we train STSMaxNet and STSConvNet models by considering different fusion layers, and test each one of them on the test set by considering sAUC measure. In Table~\ref{tab:invest}, we provide these performance comparisons at different fusion layers. Interestingly, as can be seen from the table, fusing the spatial and temporal streams after the fifth convolution layer achieves the best results for both spatio-temporal networks. In the remainder, we use these settings for our STSMaxNet and STSConvNet models.
In Figure~\ref{fig:network-comparisons}, for some sample video frames we provide the outputs of the proposed networks along with the corresponding human fixation density maps (the ground truth). The input frames are given as overlayed images by compositing them with their consecutive frames to show the motion in the scenes. Saliency maps are shown as heatmaps superimposed over the original image for visualization purposes. As can be seen from these results, SSNet, which does employ appearance but not motion information, in general provides less accurate saliency maps and misses the foreground objects or their parts that are in motion. TSNet gives relatively better results, but as shown in the second and the third row, it does not identify all of the salient regions as it focuses more on the moving parts of the images. Directly averaging the saliency maps of these two single stream models, referred to as STSAvgNet, does not produce very satisfactory results either. As expected, STSMaxNet is slightly better since max fusion enforces to learn more effective filters in order to combine the spatial and temporal streams. Overall, STSConvNet is the best performing model. This can be rooted in the application of $1 \times 1$ convolutional filters that learn the optimal weights to combine appearance and motion feature maps.
\begingroup
\begin{table}[!t]
\centering
\caption{Performance comparison of our spatio-temporal saliency networks at different fusion layers on the DIEM dataset. The reported values are sAUC scores. Best performance is achieved after the fifth convolution layer.}
\begin{tabular}{|ccc|}
\hline
Fusion Layers & STSMaxNet & STSConvNet\\
\hline \hline
Conv2 & 0.52 & 0.71\\
Conv3 & 0.67 & 0.70\\
Conv4 & 0.76 & 0.83\\
Conv5 & 0.81 & 0.84\\
Conv6 & 0.80 & 0.79\\
Conv7 & 0.81 & 0.79\\
\hline
\end{tabular}
\label{tab:invest}
\end{table}
\endgroup
\renewcommand*{\arraystretch}{1.2}
\begin{table}[!t]
\centering
\caption{Performance comparisons on the DIEM dataset.}
\begin{tabular}{|p{2.25cm}cccccc|}
\hline
& AUC & sAUC & CC & NSS & ${\chi}^2$ & NCC\\
\hline \hline
SSNet & 0.72 & 0.69 & 0.35 & 1.85 & 0.48 & 0.41\\
TSNet & 0.79 & 0.77 & 0.41 & 1.98 & 0.40 & 0.43\\
\hline
STSDirectNet & 0.71 & 0.60 & 0.37 & 1.53 & 0.49 & 0.28\\
STSAvgNet & 0.68 & 0.62 & 0.37 & 1.67 & 0.49 & 0.37\\
STSMaxNet & 0.83 & 0.81 & 0.46 & 2.01 & 0.31 & 0.45\\
STSConvNet & 0.87 & 0.84 & 0.47 & 2.15 & 0.29 & 0.46\\
STSConvNet* & \textbf{0.88} & \textbf{0.86} & \textbf{0.48} & 2.18 & \textbf{0.28} & \textbf{0.47}\\
\hline
GBVS~\cite{harel2006graph} & 0.74 & 0.70 & 0.47 & 2.04 & 0.47 & 0.38\\
SR~\cite{hou2007saliency} & 0.69 & 0.64 & 0.30 & 2.22 & 0.57 & 0.40\\
PQFT~\cite{guo2008spatio} & 0.71 & 0.67 & 0.33 & 1.77 & 0.52 & 0.33\\
Seo-Milanfar~\cite{seo2009static} & 0.59 & 0.51 & 0.10 & 0.12 & 0.75 & 0.28\\
Rudoy \emph{et al.}~\cite{rudoy2013learning} & -- & 0.74 & -- & -- & 0.31 & --\\
Fang \emph{et al.}~\cite{fang2014} & 0.71 & 0.48 & 0.21 & 0.55 & 0.87 & 0.40\\
Zhou \emph{et al.}~\cite{zhou2014learning} & 0.60 & 0.52 & 0.13 & 0.24 & 0.71 & 0.22\\
DWS~\cite{awsd} & 0.81 & 0.79 & 0.32 & \textbf{2.97} & 0.40 & 0.39\\
\hline
\end{tabular}
\label{tab:DIEMresults}
\end{table}
\begin{figure*}[!t]
\centering
\begin{tabular}{c@{\;}c@{\;}c@{\;}c@{\;}c@{\;}c@{\;}c}
\includegraphics[width=0.135\textwidth]{sources/279_280.jpg} &
\includegraphics[width=0.135\textwidth]{sources/2295_qt.jpg} &
\includegraphics[width=0.135\textwidth]{sources/2295_our.jpg} &
\includegraphics[width=0.135\textwidth]{sources/2295_gvbs.jpg} &
\includegraphics[width=0.135\textwidth]{sources/2295_quaf.jpg} &
\includegraphics[width=0.135\textwidth]{sources/d2.jpg} &
\includegraphics[width=0.135\textwidth]{sources/2295_nips.jpg} \\
\includegraphics[width=0.135\textwidth]{sources/111_112.jpg} &
\includegraphics[width=0.135\textwidth]{sources/dsogt_1.jpg} &
\includegraphics[width=0.135\textwidth]{sources/deor_1.jpg} &
\includegraphics[width=0.135\textwidth]{sources/ro1.jpg} &
\includegraphics[width=0.135\textwidth]{sources/roo1.jpg} &
\includegraphics[width=0.135\textwidth]{sources/d5.jpg} &
\includegraphics[width=0.135\textwidth]{sources/rooo1.jpg} \\
\includegraphics[width=0.135\textwidth]{sources/20_21.jpg} &
\includegraphics[width=0.135\textwidth]{sources/dsogt_3.jpg} &
\includegraphics[width=0.135\textwidth]{sources/deor_3.jpg} &
\includegraphics[width=0.135\textwidth]{sources/ro3.jpg} &
\includegraphics[width=0.135\textwidth]{sources/roo3.jpg} &
\includegraphics[width=0.135\textwidth]{sources/d6.jpg}&
\includegraphics[width=0.135\textwidth]{sources/rooo3.jpg} \\
\includegraphics[width=0.135\textwidth]{sources/14_15.jpg} &
\includegraphics[width=0.135\textwidth]{sources/dsogt_5.jpg} &
\includegraphics[width=0.135\textwidth]{sources/deor_5.jpg} &
\includegraphics[width=0.135\textwidth]{sources/ro5.jpg} &
\includegraphics[width=0.135\textwidth]{sources/roo5.jpg} &
\includegraphics[width=0.135\textwidth]{sources/d7.jpg} &
\includegraphics[width=0.135\textwidth]{sources/rooo5.jpg} \\
\includegraphics[width=0.135\textwidth]{sources/224_225.jpg} &
\includegraphics[width=0.135\textwidth]{sources/dsogt_13.jpg} &
\includegraphics[width=0.135\textwidth]{sources/deor_13.jpg} &
\includegraphics[width=0.135\textwidth]{sources/ro13.jpg} &
\includegraphics[width=0.135\textwidth]{sources/roo13.jpg} &
\includegraphics[width=0.135\textwidth]{sources/d8.jpg} &
\includegraphics[width=0.135\textwidth]{sources/rooo13.jpg} \\
\includegraphics[width=0.135\textwidth]{sources/171_172.jpg} &
\includegraphics[width=0.135\textwidth]{sources/2772_qt.jpg} &
\includegraphics[width=0.135\textwidth]{sources/2772_our.jpg} &
\includegraphics[width=0.135\textwidth]{sources/2772_gvbs.jpg} &
\includegraphics[width=0.135\textwidth]{sources/2772_quaf.jpg} &
\includegraphics[width=0.135\textwidth]{sources/d3.jpg} &
\includegraphics[width=0.135\textwidth]{sources/2772_nips.jpg} \\
\small Overlayed frames &
\small Ground Truth & \small STSConvNet* & \small GBVS~\cite{harel2006graph} & \small PQFT~\cite{guo2008spatio} & \small SR~\cite{hou2007saliency} & \small DWS~\cite{awsd}
\end{tabular}
\caption{Qualitative comparison of our STSConvNet* model against some dynamic saliency models on DIEM dataset. Our model clearly produces better results.}
\label{fig:tab:model-comparisons}
\end{figure*}
In Table~\ref{tab:DIEMresults}, we present the quantitative results averaged over all video sequences and frames on the test split of the DIEM dataset. Here, we compare and contrast our single- and two-stream saliency networks with eight existing dynamic saliency methods, namely GVBS~\cite{harel2006graph}, SR~\cite{hou2007saliency}, PQFT~\cite{guo2008spatio}, Seo and Milanfar~\cite{seo2009static}, Rudoy \emph{et al.}~\cite{rudoy2013learning}\footnote{Since the code provided by the authors is not working correctly, sAUC and $\chi^2$ scores are directly taken from~\cite{rudoy2013learning}.}, Fang \emph{et al.}~\cite{fang2014}, Zhou \emph{et al.}~\cite{zhou2014learning} and DWS~\cite{awsd} models.
Among our deep saliency networks, we empirically find that STSDirectNet provides the worst quantitative results. This is in line with our observation in Table~\ref{tab:invest} that delaying the integration of appearance and motion streams to a certain extent leads to more effective learning of mid and low level features. Secondly, we see that SSNet performs considerably lower than Temporal stream network, which demonstrates that motion is more vital for dynamic saliency. STSMaxNet gives results better than those of the single stream models but our STSConvNet model performs even better. It can be argued that STSConvNet learns more effective filters that combine spatial and temporal streams in an optimal manner. In addition, when we employ the data augmentation strategy proposed in the previous section, it further improves the overall performance of STSConvNet. In the remainder, we refer to this model with data augmentation as STSConvNet*. When we compare our proposed STSMaxNet, STSConvNet, and STSConvNet* models to the previous dynamic saliency methods, our results demonstrate the advantages of two-stream deep CNNs that they consistently outperform all those approaches, including the very recently proposed DWS model, according to five out of six evaluation measures.
We present some qualitative results in Figure~\ref{fig:tab:model-comparisons} where we again provide the input frames as transparent overlayed images showing the inherent motion. We observe that the proposed STSConvNet* model localizes the salient regions more accurately than the existing models. For example, for the frame given in the first row, none of the compared models correctly capture the fixations over the painting brush. Similarly, for the second and the third frames, only our spatio-temporal saliency network fixates to the text and the cellular phone in the frames, respectively.
\subsection{Experiments on UCF-Sports}
Learning-based models might sometimes fail to provide satisfactory results for a test sample due to a shift from the training data domain. To validate generalization ability of our best-performing STSConvNet* model, we perform additional experiments on UCF-Sports dataset. In particular, we do not carry out any training for our model from scratch or fine-tune it on UCF-Sports but rather use the predictions of the model trained only on DIEM dataset.
\begingroup
\renewcommand*{\arraystretch}{1.2}
\begin{table}[!t]
\centering
\caption{Performance comparisons on the UCF-SPORTS dataset.}
\begin{tabular}{|p{2.25cm}cccccc|}
\hline
& AUC & sAUC & CC & NSS & ${\chi}^2$ & NCC\\
\hline \hline
GBVS~\cite{harel2006graph} & 0.83 & 0.52 & 0.46 & 1.82 & 0.54 & \textbf{0.59}\\
SR~\cite{hou2007saliency} & 0.78 & 0.69 & 0.26 & 1.20 & 0.42 & 0.52\\
PQFT~\cite{guo2008spatio} & 0.69 & 0.51 & 0.29 & 1.15 & 0.64 & 0.48\\
Seo-Milanfar~\cite{seo2009static} & 0.80 & 0.72 & 0.31 & 1.37 & 0.56 & 0.36\\
Fang \emph{et al.}~\cite{fang2014} & \textbf{0.85} & 0.70 & 0.44 & 1.95 & 0.52 & 0.33\\
Zhou \emph{et al.}~\cite{zhou2014learning} & 0.81 & 0.72 & 0.36 & 1.71 & 0.56 & 0.37\\
DWS~\cite{awsd} & 0.76 & 0.70 & 0.28 & 2.01 & 0.40 & 0.49\\
STSConvNet* & 0.82 & \textbf{0.75} & \textbf{0.48} & \textbf{2.13} & \textbf{0.39} & 0.54\\
\hline
\end{tabular}
\label{tab:ucf-results}
\end{table}
\endgroup
\begin{figure*}[!t]
\centering
\resizebox{\textwidth}{!} {
\begin{tabular}{c@{\;}c@{\;}c@{\;}c@{\;}c@{\;}c}
\includegraphics[width=0.18\textwidth]{sources/306_307.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usir_3.jpg} &
\includegraphics[width=0.18\textwidth]{sources/ueor_3.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usirr_3.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usirrr_3.jpg} \\
\includegraphics[width=0.18\textwidth]{sources/269_270.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usir_2.jpg} &
\includegraphics[width=0.18\textwidth]{sources/ueor_2.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usirr_2.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usirrr_2.jpg} \\
\includegraphics[width=0.18\textwidth]{sources/288_289.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usir_6.jpg} &
\includegraphics[width=0.18\textwidth]{sources/ueor_6.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usirr_6.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usirrr_6.jpg} \\
\includegraphics[width=0.18\textwidth]{sources/78_79.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usir_8.jpg} &
\includegraphics[width=0.18\textwidth]{sources/ueor_8.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usirr_8.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usirrr_8.jpg} \\
\includegraphics[width=0.18\textwidth]{sources/58_59.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usir_1.jpg} &
\includegraphics[width=0.18\textwidth]{sources/ueor_1.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usirr_1.jpg} &
\includegraphics[width=0.18\textwidth]{sources/usirrr_1.jpg} \\
\footnotesize Overlayed input frames & \footnotesize Ground Truth & \footnotesize STSConvNet* & \footnotesize Fang \emph{et al.}~\cite{fang2014} & \footnotesize DWS~\cite{awsd}
\end{tabular}}
\caption{Qualitative comparison of our STSConvNet* model against some previous dynamic saliency models on UCF-Sports dataset. Our spatio-temporal saliency network outperforms the others.}
\label{fig:tab:early-ucf}
\end{figure*}
\begin{figure}[!t]
\centering
\resizebox{\linewidth}{!} {
\begin{tabular}{c@{\;}c@{\;}c@{\;}c@{\;}c}
\includegraphics[width=2.4cm,height=1.8cm]{sources/m1.jpg} &
\includegraphics[width=2.4cm,height=1.8cm]{sources/mg1.jpg} &
\includegraphics[width=2.4cm,height=1.8cm]{sources/mf1.jpg} &
\includegraphics[width=2.4cm,height=1.8cm]{sources/ms1.jpg} &
\includegraphics[width=2.4cm,height=1.8cm]{sources/mr1.jpg} \\
\includegraphics[width=2.4cm,height=3.4cm]{sources/m2.jpg} &
\includegraphics[width=2.4cm,height=3.4cm]{sources/mg2.jpg} &
\includegraphics[width=2.4cm,height=3.4cm]{sources/mf2.jpg} &
\includegraphics[width=2.4cm,height=3.4cm]{sources/ms2.jpg} &
\includegraphics[width=2.4cm,height=3.4cm]{sources/mr2.jpg} \\
\includegraphics[width=2.4cm,height=2cm]{sources/m6.jpg} &
\includegraphics[width=2.4cm,height=2cm]{sources/mg6.jpg} &
\includegraphics[width=2.4cm,height=2cm]{sources/mf6.jpg} &
\includegraphics[width=2.4cm,height=2cm]{sources/ms6.jpg} &
\includegraphics[width=2.4cm,height=2cm]{sources/mr6.jpg} \\
\includegraphics[width=2.4cm,height=3.5cm]{sources/m5.jpg} &
\includegraphics[width=2.4cm,height=3.5cm]{sources/mg5.jpg} &
\includegraphics[width=2.4cm,height=3.5cm]{sources/mf5.jpg} &
\includegraphics[width=2.4cm,height=3.5cm]{sources/ms5.jpg} &
\includegraphics[width=2.4cm,height=3.5cm]{sources/mr5.jpg} \\
\includegraphics[width=2.4cm,height=3.2cm]{sources/m4.jpg} &
\includegraphics[width=2.4cm,height=3.2cm]{sources/mg4.jpg} &
\includegraphics[width=2.4cm,height=3.2cm]{sources/mf4.jpg} &
\includegraphics[width=2.4cm,height=3.2cm]{sources/ms4.jpg} &
\includegraphics[width=2.4cm,height=3.2cm]{sources/mr4.jpg} \\
\normalsize Input Image &\normalsize Ground Truth & \normalsize DeepFix~\cite{kruthiventi2015deepfix} & \normalsize SALICON~\cite{jiang2015salicon} & \normalsize STSConvNet
\end{tabular}}
\caption{Some experiments on still images. While the top performing deep static saliency models fail to compute satisfactory results in these images (results taken from~\cite{bylinskii2016should}), our spatio-temporal saliency network (STSConvNet) can produce better saliency maps using predicted optical flow maps.}
\label{fig:motioneffect}
\end{figure}
In Table~\ref{tab:ucf-results}, we provide the performance of our model compared to the previous dynamic saliency models which are publicly available on the web. As can be seen, our STSConvNet* model performs better than the state-of-the-art models according to majority of the evaluation measures. It especially outperforms the recently proposed DWS model in terms of all measures. These results suggest that our two-stream network generalizes well beyond the DIEM dataset.
In Figure~\ref{fig:tab:early-ucf}, we present sample qualitative results of Fang \emph{et al.}~\cite{fang2014} and DWS model~\cite{awsd} (two recently proposed models) and our STSConvNet model on some video frames. For instance, we observe that for the sample frame given in the first row, our model fixates to both the runner and the crowd. For the second and the third sample frames, the compared models do not accurately localize the weight lifter and the cowboys as salient, respectively. Similarly, the proposed STSConvNet* model predicts the eye fixations better than the competing models for the fourth image containing a guardian walking in a corridor. For the last diving image, STSConvNet* and Fang~\emph{et~al.} give results fairly close to the ground truth, while DWS output some spurious regions as salient.
\subsection{Experiments on Still Images from MIT300}
Deep static saliency networks achieve excellent performances on existing benchmark datasets for static saliency estimation. These models, however, only exploit spatial information captured in still images, but sometimes an image, despite being taken in an instant, might carry plenty of information regarding the inherent motion exist in it. In a recent study by Bylinski~\emph{et~al.}~\cite{bylinskii2016should}, the authors demonstrate that the areas showing these kind of activities are indeed evidently important for saliency prediction since humans have tendency to look at the objects that they think in motion or that are in interaction with humans or some other objects. Motivated by these observations, in this section, we present the failures or the shortcomings of the current deep static saliency models through some examples, and show how motion information exist in still images can be utilized to fill in the semantic gap exist in the current static saliency models.
Figure~\ref{fig:motioneffect} presents sample images taken from~\cite{bylinskii2016should} and which are from the MIT 300 dataset~\cite{mit-saliency-benchmark} where highly fixated regions (which cover the 95th percentile of the human fixation maps) are highlighted with yellow curves. As can be clearly seen from these examples, the state-of-the-art deep static models generally fail to give high saliency values to regions where an action occurs or which contains objects that are interpreted as in motion. To capture those regions, we employ the deep optical flow prediction model~\cite{walker2015dense} which extracts optical flow from static images. Once we estimate the motion map of a still image, we can exploit this information together with the RGB image as inputs to our spatio-temporal saliency network (STSConvNet) to extract a saliency map. We observe that using these (possibly noisy) motion maps within our framework provides more accurate predictions than the existing deep static saliency models, and even captures the objects of gaze as illustrated in the first two sample images. These experiments reveal that the performances of static saliency networks can be improved by additionally considering motion information inherent in still images.
\section{Conclusion}
In this work, we have investigated several deep architectures for predicting saliency from dynamic scenes. Two of these deep models are single-stream convolutional networks respectively trained for processing spatial and temporal information. Our proposed spatio-temporal saliency networks, on the other hand, are built based on two-stream architecture and employ different fusion strategies, namely direct averaging, max fusion and convolutional fusion, to integrate appearance and motion features, and they are all trainable in an end-to-end manner. While training these saliency networks, we additionally employ an effective and well-founded data augmentation method that utilizes low-resolution versions of the video frames and the ground truth saliency maps, giving a significant boost in performance. Our experimental results demonstrate that the proposed STSConvNet model achieves superior performance over the state-of-the-art methods on DIEM and UCF-Sports datasets. Lastly, we provide some illustrative example results on a number of challenging still images, which show that static saliency estimation can also benefit from motion information. This is left as an interesting topic for future research.
\section*{Acknowledgment}
This research was supported in part by TUBITAK Career Development Award 113E497 and Hacettepe BAP FDS-2016-10202.
\ifCLASSOPTIONcaptionsoff
\newpage
\fi
\bibliographystyle{IEEEtran}
| 4456daf6a3728f0301623af6572edef7f7c27b6c | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{\label{sec:level1}First-level heading:\protect\\ The line
Tunnel field-effect transistor (TFET), a promising replacement of classical metal-oxide-semiconductor field-effect transistor (MOSFET) for future low-power integrated circuits, has been intensively studied over a decade. The advantages of TFET come from its steep subthreshold swing (SS) that overcomes the 60mV/dec limit of a conventional MOSFET, allowing substantial supply voltage ($V_{\rm{DD}}$) scaling \cite{ionescu2011tunnel}. However, because of low tunnel probability the steep SS usually occurs at very low current level \cite{seabaugh2010low,lu2014tunnel}. This leads to insufficient ON-state current ($I_{\rm{ON}}$) and thus large switching delay ($CV_{\rm{DD}}/I_{\rm{ON}}$). Various approaches have been proposed to improve the low $I_{\rm{ON}}$. In particular, GaSb/InAs heterojunction based TFETs can considerably boost $I_{\rm{ON}}$ due to their broken/staggered-gap band alignment \cite{mohata2011demonstration,lind2015iii}.
However, as the channel length scales to sub-20nm as projected by International Technology Roadmap for Semiconductors (ITRS) for the next technology nodes \cite{itrs_website}, the GaSb/InAs n-type TFETs suffer from large ambipolar and source-to-drain tunneling leakage due to the small band gap and the small effective masses of the InAs channel. These leakage can be reduced by reducing the body thickness \cite{avci2013heterojunction}, because the band gap and the effective masses of the InAs channel increase as the body thickness decreases. Meanwhile, the large band gap and large effective masses also reduce the tunneling probability across the tunnel junction. The resonant TFET with a reversed InAs/GaSb heterojunction can have a steep SS at short gate length \cite{avci2013heterojunction} but the $I_{\rm{ON}}$ is limited by the narrow resonant transmission peak \cite{long2016design}. The channel heterojunction design with a large band-gap AlInAsSb alloy as the channel material has also been proposed \cite{long2016high,long2016extremely} to mitigate the short-channel effects. However, a good-quality dielectric on top of AlInAsSb has not been experimentally demonstrated yet. Therefore, InP, a material with a high-quality dielectric already demonstated, has been investigated as the alternative channel material \cite{long2016iprm}. It is found that the lattice-mismatched InP channel imposes biaxial compressive strain on the GaSb/InAs tunnel junction, compromising the improvement.
\begin{figure}
\includegraphics[width=4.25cm]{geo_flat_thick.eps}
\includegraphics[width=4.25cm]{geo_non_uniform.eps}
\caption{\label{fig:geo} GaSb/InAs n-type TFETs with conventional uniform (a) and proposed non-uniform (b) body thickness. The oxide and the gate in (b) are conformal to the channel so that the oxide thickness is equal to that of (a). Here, Tch/Tox is the channel/oxide thickness, $\epsilon_{ox}$ is the oxide dielectric constant, Ls/Lg/Ld is the source/channel/drain length, and Ns/Nd is the source/drain doping density. The extra parameters in (b) are the source thickness Ts and the wide channel length Lw.}
\end{figure}
In this study, we show that by reducing the InAs channel body thickness meanwhile retaining a relatively large body thickness at the source tunnel junction, the leakage can be significantly reduced without compromising the large source tunnel probability. The devices are simulated and optimized by solving Poisson equation and open-boundary Schr\"{o}dinger equation \cite{luisier2006atomistic} self-consistently within NEMO5 tool \cite{steiger2011nemo5}. The Hamiltonian employed is in the atomistic $sp^3d^5s^*$ tight binding (TB) basis including spin-orbit coupling, with the room temperature TB parameters fitted to the band structures as well as the wave functions of density functional theory calculations for better transferability \cite{tan2015tight, Tan2016Online}. Due to the high semiconductor-to-oxide barrier height \cite{robertson2006}, the oxide is treated as an impenetrable potential barrier and only modeled in the Poisson equation. Various scattering mechanisms, such as the electron-phonon scattering \cite{Knoch2010, Luisier2010} and the discrete dopant scattering \cite{Khayer2011}, are not included, which may lead to underestimated SS. Non-ideality effects, such as the interface trap assisted tunneling and the Shockley-Read-Hall generation-recombination processes \cite{Mookerjea2010}, are not considered in this study.
\begin{figure}
\includegraphics[width=8.5cm]{EcEv_mass.eps}
\caption{\label{fig:edge_mass} (a) Conduction and valence band edges (Ec and Ev), (b) electron and hole effective masses (m*e and m*h), of the GaSb and InAs UTBs, as functions of the UTB thickness. The confinement is in the [001] orientation and the effective masses are in the [100] orientation.}
\end{figure}
\begin{figure}
\includegraphics[width=8.5cm]{IV_IoffIon_uniform.eps}
\caption{\label{fig:IV_uni} (a) $I_{\rm{DS}}$-$V_{\rm{GS}}$ characteristics ($V_{\rm{DS}}$=0.3V) of the uniform device (Fig.~\ref{fig:geo}~(a)) as a function of the body thickness Tch. Confinement/transport is in the [001]/[100] orientation. Lg=15nm, Tox=1.8nm, $\epsilon_{ox}$=9.0, Ns=$-5\times10^{19}/\rm{cm}^{3}$, and Nd=$+2\times10^{19}/\rm{cm}^{3}$. (b) $I_{\rm{ON}}$-$I_{\rm{OFF}}$ curves with $V_{\rm{DD}}=0.3$V.}
\end{figure}
First, we analyze the conventional GaSb/InAs UTB n-type TFETs with uniform body thickness, as illustrated in Fig.~\ref{fig:geo}~(a). The band edges and the effective masses of the ultra-thin-body (UTB) structures for different body thicknesses are plotted in Fig.~\ref{fig:edge_mass}. As can be seen, the InAs UTB band gap (InAs Ec$-$InAs Ev) and the tunnel barrier height (InAs Ec$-$GaSb Ev) both increase as the UTB thickness becomes smaller. The InAs Ec calculated is in agreement with the experiment data \cite{brar1993photoluminescence} with upto 0.2eV mismatch \footnote{The experimental data is shifted by 0.15eV to account for the valence band offset (VBO) between bulk InAs and confined AlSb (with 5nm thickness), assuming the VBO between bulk InAs and bulk AlSb is 0.18eV \cite{vurgaftman2001band} and the confined heavy hole energy of AlSb is 0.03eV. Note that the experiment data were obtained at low temperature.}. The effective masses (electron and hole) of the InAs UTB also increase significantly as the UTB thickness decreases. Therefore, the I-V characteristic of the device is a strong function of the UTB thickness. Indeed, as shown in Fig.~\ref{fig:IV_uni}~(a), a large body thickness (Tch=3.6nm or 3.0nm) leads to a large turn-on current, but also a large SS and a high $I_{\rm{OFF}}$. A small body thickness (Tch=1.8nm) gives rise to a small SS and a low $I_{\rm{OFF}}$, but a small turn-on current. With $I_{\rm{OFF}}=1\times10^{-3}$A/m and $V_{\rm{DD}}=0.3$V, the optimal body thickness for Lg=15nm is around 2.4nm, providing $I_{\rm{ON}}=25$A/m (Fig.~\ref{fig:IV_uni}~(b)). This is too low for any practical logic application.
\begin{figure}
\includegraphics[width=8.5cm]{Ion_Tch_Lw_full_IV.eps}
\caption{\label{fig:IV_uni_non_uni} (a) $I_{\rm{ON}}$ (at $V_{\rm{DD}}$=0.3V and $I_{\rm{OFF}}$=$1\times10^{-3}$A/m) of the nonuniform design (Fig.~\ref{fig:geo}~(b)) for different values of Lw and Tch, with fixed Ts=3.6nm. (b) Full $I_{\rm{DS}}$-$V_{\rm{GS}}$ characteristics of the Ts/Tch/Lw=3.6nm/1.8nm/4.0nm case in (a), in comparison with the uniform 3.6nm and uniform 1.8nm cases. All other device parameters are the same as those in Fig.~\ref{fig:IV_uni}.}
\end{figure}
\begin{figure}
\includegraphics[width=8.5cm]{band_trans.eps}
\caption{\label{fig:band_trans} Comparison of the band diagrams (a) and the transmission functions (b) of the three cases in Fig.~\ref{fig:IV_uni_non_uni}~(b): uniform 3.6nm body thickness, uniform 1.8nm thickness, and non-uniform 3.6nm/1.8nm body thickness.}
\end{figure}
The proposed design that can overcome this dilemma is illustrated in Fig.~\ref{fig:geo}~(b). Compared with Fig.~\ref{fig:geo}~(a), this design has reduced body thickness {\it only} in part of the channel region (and in the drain).
Three parameters need to be optimized, {\it i.e.}, Ts, Tch, and Lw. In this study, we fix Ts to be 3.6nm and then optimize Tch and Lw to maximize $I_{\rm{ON}}$. As shown in Fig.~\ref{fig:IV_uni_non_uni}~(a), there is an optimal Lw for each Tch and the optimal Lw is smaller for smaller Tch, which will be explained in a moment. A tradeoff of Tch is also clearly observed, since a large Tch does not have sufficiently large band gap and effective masses needed for the leakage suppression, while a small Tch would affect the ON-state transmission due to the large reflection at the waveguide discontinuity. With Tch=1.8nm and Lw=4.0nm, we obtain the largest $I_{\rm{ON}}$ (284A/m), which is more than an order of magnitude larger than that of the uniform case (25A/m). The full $I_{\rm{DS}}$-$V_{\rm{GS}}$ curve is further displayed in Fig.~\ref{fig:IV_uni_non_uni}~(b) along with two uniform thickness cases, showing that both steep SS and large turn-on current are simultaneously obtained in the nonuniform case.
The improvements can be better understood from the band diagrams and the transmissions plotted in Fig.~\ref{fig:band_trans}, where the three cases in Fig.~\ref{fig:IV_uni_non_uni}~(b) are compared. Above the channel Ec, the nonuniform 3.6/1.8nm case has larger transmission than the uniform 1.8nm case, giving rise to its larger turn-on current. This is due to its smaller tunneling barrier height and smaller effective masses at the tunnel junction. Its transmission over channel Ec is even larger than the uniform 3.6nm case, benefiting from its larger tunneling electric field and resonance-enhanced tunneling, both resulting from the confined band offset. Below the channel Ec, the nonuniform 3.6/1.8nm case has steeper transmission slope than the uniform 3.6nm case, implying steeper SS. This is partly due to the larger channel band gap and larger channel effective masses, partly due to the better electrostatics at the channel-drain junction, and partly due to the smaller drain Fermi degeneracy (the energy distance between the drain Fermi level and the drain Ec).
\begin{figure}
\includegraphics[width=8.5cm]{LDOS_.eps}
\caption{\label{fig:ldos} LDOS (in logarithmic scale) of the optimized design in Fig.~\ref{fig:IV_uni_non_uni}, at OFF state (a) and ON state (b). Band diagrams (dashed lines) and contact Fermi levels (solid lines) are superimposed. The quasi-bound states are highlighted (circles).}
\end{figure}
The local denstiy of states (LDOS) is further depicted in Fig.~\ref{fig:ldos}. Similar to the channel heterojunction design \cite{long2016high}, the confined conduction band edges in the channel form a quantum well which creates a quasi-bound state. The energy level of this state needs to be aligned with the channel conduction band edge at the ON state so that it enhances the $I_{\rm{ON}}$. At the OFF state, in order to reduce phonon-assisted tunneling leakage, the energy of this state has to be higher than the valence band edge at the source by at least the optical phonon energy \cite{koswatta2010possibility} ($\hbar\omega_{op}\approx$30meV for bulk InAs and here it is 75meV higher). To satisfy these requirements, the parameter Lw needs to be adjusted for a given Tch, as shown in Fig.~\ref{fig:IV_uni_non_uni}~(a). In fact, a smaller Tch leads to a larger confined band offset and thus the Lw needs to be reduced properly to shift the quasi-bound state upward.
\begin{figure}
\includegraphics[width=8.5cm]{IV_variations.eps}
\caption{\label{fig:variations} Sensitivities of the $I_{\rm{DS}}$-$V_{\rm{GS}}$ curve ($V_{\rm{DS}}$=0.3V) to the variations of Ts (a), Tch (b), and Lw (d), for the optimized design in Fig.~\ref{fig:IV_uni_non_uni}. The amount of Ts and Tch variations is one monolayer (about $\pm0.3$nm) and the Lw variation is $\pm0.5$nm.}
\end{figure}
Fig.~\ref{fig:variations} shows the sensitivities of the I-V curve to the geometry variations.
We find that the I-V curve can tolerate certain amount of Ts and Lw variations.
It is however very sensitive to the channel thickness (Tch) variations resulting in unacceptable $I_{\rm{OFF}}$ or $I_{\rm{ON}}$ level.
This suggests that precise fabrication control of Tch (at atomic level) is necessary.
\begin{figure}
\includegraphics[width=8.5cm]{IV_scale_vs_mosfet.eps}
\caption{\label{fig:scale} $I_{\rm{DS}}$-$V_{\rm{GS}}$ characteristics ($V_{\rm{DS}}$=0.3V) of the proposed TFETs, in comparison with Si MOSFETs, for three channel lengths (Lg=15nm, 12nm, and 9nm). (a) Ts/Tch/Lw=3.6/1.8/4.0nm for TFETs and Tch=1.8nm for MOSFETs, and (b) Ts/Tch/Lw=3.6/1.2/2.5nm for TFETs and Tch=1.2nm for MOSFETs.}
\end{figure}
ITRS 2020 and 2023 technology nodes require channel length Lg to be scaled to about 12nm and 9nm \cite{itrs_website}. At such short channel lengths, the source-to-drain tunneling leakage becomes more prominent. As compared in Fig.~\ref{fig:scale}~(a), when Lg is reduced from 15nm to 12nm, the $I_{\rm{ON}}$ of TFET drops from 284A/m to 106A/m, while the $I_{\rm{ON}}$ of Si MOSFET only drops from 49A/m to 41A/m. When Lg is further scaled to 9nm, the TFET cannot provide a decent ON/OFF ratio and does not possess an advantage over Si MOSFET. To improve the scalability, the channel thickness (Tch) can be reduced from 1.8nm to 1.2nm to enlarge the channel band gap and channel effective masses. As seen in Fig.~\ref{fig:scale}~(b), the SS of TFET degrades less, with $I_{\rm{ON}}$=209A/m (50A/m) obtained at Lg=12nm (9nm), which is still considerably larger than $I_{\rm{ON}}$=47A/m (31A/m) of Si MOSFET.
In summary, a novel transistor design with a nonuniform body thickness is proposed to improve the scalability of GaSb/InAs n-type TFETs. At 15nm channel length, it achieves ballistic $I_{\rm{ON}}$ that is over $10\times$ greater than that is achievable with uniform body thickness. Moreover, it enables the channel to be scaled towards sub-10nm lengths. This design can also be employed to improve InAs/GaSb p-type TFETs.
This work uses nanoHUB.org computational resources operated by the Network for Computational Nanotechnology funded by the U.S. National Science Foundation under Grant EEC-0228390, Grant EEC-1227110, Grant EEC-0634750, Grant OCI-0438246, Grant OCI-0832623, and Grant OCI-0721680. This material is based upon work supported by the National Science Foundation under Grant 1509394. NEMO5 developments were critically supported by an NSF Peta-Apps award OCI-0749140 and by Intel Corp.
\nocite{*}
\bibliographystyle{apl}
| d62cd034a758076c0c8422dd586fbea236aeebed | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{INTRODUCTION}
In recent years deep learning methods have demonstrated significant improvements in various tasks. These methods outperform human-level object detection in some tasks \cite{resnet}, and achieve state-of-the-art results in machine translation \cite{nmt} and speech processing \cite{graves2013speech}. Additionally, deep learning combined with reinforcement learning techniques was able to beat human champions in challenging games such as Go \cite{d_silver}. There are three different reasons for the outstanding results of Deep Learning models:
\begin{enumerate}
\item Powerful computing resources such as fast GPUs.
\item Utilizing efficiently large collections of datasets, e.g. ImageNet \cite{imagenet} for image processing.
\item Advanced academic research on training methods and network architectures \cite{batch_norm}, \cite{alexnet}, \cite{adadelta}, \cite{dropout}.
\end{enumerate}
Error correcting codes for channel coding are used in order to enable reliable communications at rates close to the Shannon capacity. A well-known family of linear error correcting codes are the low-density parity-check (LDPC) codes \cite{galmono}. LDPC codes achieve near Shannon channel capacity with the belief propagation (BP) decoding algorithm, but can typically do so for relatively large block lengths. For high density parity check (HDPC) codes \cite{jiang2006iterative}, \cite{dimnik2009improved}, \cite{yufit2011efficient}, \cite{zhang2012adaptive}, such as common powerful algebraic codes, the BP algorithm obtains poor results compared to the maximum likelihood decoder \cite{helmling2014efficient}.
In this work we focus on HDPC codes and demonstrate how the BP algorithm can be improved. The naive approach to the problem is to assume a neural network type decoder without restrictions, and train its weights using a dataset that contains a large amount of codewords. The training goal is to reconstruct the transmitted codeword from a noisy version after transmitting over the communication channel. Unfortunately, when using this approach our decoder is not given any side information regarding the structure of the code. In fact it is even not aware of the fact that the code is linear. Hence we are required to train the decoder using a huge collection of codewords from the code, and due to the exponential nature of the problem, this is infeasible, e.g., for a BCH(63,45) code we need a dataset of $2^{45}$ codewords. On top of that, the database needs to reflect the variability due to the noisy channel.
In order to overcome this issue, our proposed approach is to assign weights to the edges of the Tanner graph that represent the given linear code, thus yielding a ``soft'' Tanner graph. These edges are trained using deep learning techniques. A well-known property of the BP algorithm is the independence of the performance on the transmitted codeword. A major ingredient in our new method is that this property is preserved by our decoder. Thus it is sufficient to use a single codeword for training the parameters of our decoder. We demonstrate improvements over BP for various high density parity check codes, including BCH(63,36), BCH(63,45), and BCH(127,106) .
\section{THE BELIEF PROPAGATION ALGORITHM}
The renowned BP decoder \cite{galmono}, \cite{ru_book} can be constructed from the Tanner graph, which is a graphical representation of some parity check matrix that describes the code. In this algorithm, messages are transmitted over edges. Each edge calculates its outgoing message based on all incoming messages it receives over all its edges, except for the message received on the transmitting edge. We start by providing an alternative graphical representation to the BP algorithm with $L$ full iterations when using parallel (flooding) scheduling. Our alternative representation is a trellis in which the nodes in the hidden layers correspond to edges in the Tanner graph.
Denote by $N$, the code block length (i.e., the number of variable nodes in the Tanner graph), and by $E$, the number of edges in the Tanner graph. Then the input layer of our trellis representation of the BP decoder is a vector of size $N$, that consists of the log-likelihood ratios (LLRs) of the channel outputs. The LLR value of variable node $v$, $v=1,2,\ldots,N$, is given by
$$
l_v = \log\frac{\Pr\left(C_v=1 | y_v\right)}{\Pr\left(C_v=0 | y_v\right)}
$$
where $y_v$ is the channel output corresponding to the $v$th codebit, $C_v$.
All the following layers in the trellis, except for the last one (i.e., all the hidden layers), have size $E$. For each hidden layer, each processing element in that layer is associated with the message transmitted over some edge in the Tanner graph. The last (output) layer of the trellis consists of $N$ processing elements that output the final decoded codeword. Consider the $i$th hidden layer, $i=1,2,\ldots,2L$. For odd (even, respectively) values of $i$, each processing element in this layer outputs the message transmitted by the BP decoder over the corresponding edge in the graph, from the associated variable (check) node to the associated check (variable) node. A processing element in the first hidden layer ($i=1$), corresponding to the edge $e=(v,c)$, is connected to a single input node in the input layer: It is the variable node, $v$, associated with that edge.
Now consider the $i$th ($i>1$) hidden layer. For odd (even, respectively) values of $i$, the processing node corresponding to the edge $e=(v,c)$ is connected to all processing elements in layer $i-1$ associated with the edges $e'=(v,c')$ for $c'\ne c$ ($e'=(v',c)$ for $v'\ne v$, respectively). For odd $i$, a processing node in layer $i$, corresponding to the edge $e=(v,c)$, is also connected to the $v$th input node.
The BP messages transmitted over the trellis graph are the following. Consider hidden layer $i$, $i=1,2,\ldots,2L$, and let $e=(v,c)$ be the index of some processing element in that layer. We denote by $x_{i,e}$, the output message of this processing element. For odd (even, respectively), $i$, this is the message produced by the BP algorithm after $\lfloor (i-1)/2 \rfloor$ iterations, from variable to check (check to variable) node.
For odd $i$ and $e=(v,c)$ we have (recall that the self LLR message of $v$ is $l_v$),
\begin{equation}
x_{i,e=(v,c)} = l_v + \sum_{e'=(v,c'),\: c'\ne c} x_{i-1,e'}
\label{eq:x_ie_RB}
\end{equation}
under the initialization, $x_{0,e'}=0$ for all edges $e'$ (in the beginning there is no information at the parity check nodes). The summation in~\eqref{eq:x_ie_RB} is over all edges $e'=(v,c')$ with variable node $v$ except for the target edge $e=(v,c)$. Recall that this is a fundamental property of message passing algorithms~\cite{ru_book}.
Similarly, for even $i$ and $e=(v,c)$ we have,
\begin{equation}
x_{i,e=(v,c)} = 2\tanh^{-1} \left( \prod_{e'=(v',c),\: v'\ne v} \tanh \left( \frac{x_{i-1,e'}}{2} \right) \right)
\label{eq:x_ie_LB}
\end{equation}
The final $v$th output of the network is given by
\begin{equation}
o_v = l_v + \sum_{e'=(v,c')} x_{2L,e'}
\label{eq:ov}
\end{equation}
which is the final marginalization of the BP algorithm.
\section{THE PROPOSED DEEP NEURAL NETWORK DECODER}
We suggest the following parameterized deep neural network decoder that generalizes the BP decoder of the previous section. We use the same trellis representation for the decoder as in the previous section. The difference is that now we assign weights to the edges in the Tanner graph. These weights will be trained using stochastic gradient descent which is the standard method for training neural networks. More precisely, our decoder has the same trellis architecture as the one defined in the previous section. However, Equations~\eqref{eq:x_ie_RB}, \eqref{eq:x_ie_LB} and \eqref{eq:ov} are replaced by
\begin{equation}
x_{i,e=(v,c)} = \tanh \left(\frac{1}{2}\left(w_{i,v} l_v + \sum_{e'=(v,c'),\: c'\ne c} w_{i,e,e'} x_{i-1,e'}\right)\right)
\label{eq:x_ie_RB_NN}
\end{equation}
for odd $i$,
\begin{equation}
x_{i,e=(v,c)} = 2\tanh^{-1} \left( \prod_{e'=(v',c),\: v'\ne v}{x_{i-1,e'}}\right)
\label{eq:x_ie_LB_NN}
\end{equation}
for even $i$, and
\begin{equation}
o_v = \sigma \left( w_{2L+1,v} l_v + \sum_{e'=(v,c')} w_{2L+1,v,e'} x_{2L,e'} \right)
\label{eq:ov_NN}
\end{equation}
where $\sigma(x) \equiv \left( 1+e^{-x} \right)^{-1}$ is a sigmoid function. The sigmoid is added so that the final network output is in the range $[0,1]$. This makes it possible to train the network using a cross entropy loss function, as described in the next section. Apart of the addition of the sigmoid function at the outputs of the network, one can see that by setting all weights to one, Equations \eqref{eq:x_ie_RB_NN}-\eqref{eq:ov_NN} degenerate to \eqref{eq:x_ie_RB}-\eqref{eq:ov}. Hence by optimal setting (training) of the parameters of the neural network, its performance can not be inferior to plain BP.
It is easy to verify that the proposed message passing decoding algorithm \eqref{eq:x_ie_RB_NN}-\eqref{eq:ov_NN} satisfies the message passing symmetry conditions \cite[Definition 4.81]{ru_book}. Hence, by \cite[Lemma 4.90]{ru_book}, when transmitting over a binary memoryless symmetric (BMS) channel, the error rate is independent of the transmitted codeword. Therefore, to train the network, it is sufficient to use a database which is constructed by using noisy versions of a single codeword. For convenience we use the zero codeword, which must belong to any linear code. The database reflects various channel output realizations when the zero codeword has been transmitted. The goal is to train the parameters $\left \{ w_{i,v},w_{i,e,e'},w_{i,v,e'} \right \}$ to achieve an $N$ dimensional output word which is as close as possible to the zero codeword. The network architecture is a non-fully connected neural network. We use stochastic gradient descent to train the parameters.
The motivation behind the new proposed parameterized decoder is that by setting the weights properly, one can compensate for small cycles in the Tanner graph that represents the code. That is, messages sent by parity check nodes to variable nodes can be weighted, such that if a message is less reliable since it is produced by a parity check node with a large number of small cycles in its local neighborhood, then this message will be attenuated properly.
The time complexity of the deep neural network is similar to plain BP algorithm. Both have the same number of layers and the same number of non-zero weights in the Tanner graph. The deep neural network architecture is illustrated in Figure~\ref{fig:BCH_15_11_arch} for a BCH(15,11) code.
\begin{figure}[thpb]
\centering
\includegraphics[width=0.983101925\linewidth]{nn_15_11_highres.pdf}
\caption{Deep Neural Network Architecture For BCH(15,11) with 5 hidden layers which correspond to 3 full BP iterations. Note that the self LLR messages $l_v$ are plotted as small bold lines. The first hidden layer and the second hidden layer that were described above are merged together. Also note that this figure shows 3 full iterations and the final marginalization.}
\label{fig:BCH_15_11_arch}
\end{figure}
\section{EXPERIMENTS}
\subsection{Neural Network Training}
We built our neural network on top of the TensorFlow framework~\cite{abadi2015tensorflow} and used an NVIDIA Tesla K40c GPU for accelerated training. We applied cross entropy as our loss function,
\begin{equation}
L{(o,y)}=-\frac{1}{N}\sum_{v=1}^{N}y_{v}\log(o_{v})+(1-y_{v})\log(1-o_{v})
\label{eq:cross_entropy}
\end{equation}
where $o_{v}$, $y_{v}$ are the deep neural network output and the actual $v$th component of the transmitted codeword (if the all-zero codeword is transmitted then $y_{v}=0$ for all $v$).
Training was conducted using stochastic gradient descent with mini-batches. The mini-batch size was $120$ examples. We applied the RMSPROP~\cite{rmsprop} rule with a learning rate equal to $0.001$. The neural network has $10$ hidden layers, which correspond to $5$ full iterations of the BP algorithm. Each processing element in an odd (even, respectively) indexed hidden layer is described by Equation~\eqref{eq:x_ie_RB_NN} (Equation~\eqref{eq:x_ie_LB_NN}, respectively).
At test time, we inject noisy codewords after transmitting through an AWGN channel and measure the bit error rate (BER) in the decoded codeword at the network output.
When computing~\eqref{eq:x_ie_RB_NN}, we also clip the input to the tanh function such that the absolute value of the input is always smaller than some positive constant $A < 10$. This is also required for practical (finite block length) implementations of the BP algorithm, in order to stabilize the operation of the decoder. We trained our decoding network on few different linear codes, including BCH(15,11), BCH(63,36), BCH(63,45) and BCH(127,106).
\subsection{Neural Network Training With Multiloss}
The proposed neural network architecture has the property that after every odd $i$ layer we can add final marginalization. We can use that property to add additional terms in the loss. The additional terms can increase the gradient update at the backpropagation algorithm and allow learning the lower layers. At each odd $i$ layer we add the final marginalization to loss:
\begin{equation}
L{(o,y)}=-\frac{1}{N}\sum_{i=1,3}^{2L-1}\sum_{v=1}^{N}y_{v}\log(o_{v,i})+(1-y_{v})\log(1-o_{v,i})
\label{eq:multiloss_cross_entropy}
\end{equation}
where $o_{v,i}$, $y_{v}$ are the deep neural network output at the odd $i$ layer and the actual $v$th component of the transmitted codeword. This network architecture is illustrated in Figure 2.
\begin{figure}[thpb]
\centering
\hspace*{-3.5cm}\includegraphics[width=1.4\linewidth, left]{nn_15_11_multiloss_highres.pdf}
\caption{Deep Neural Network Architecture For BCH(15,11) with training
multiloss. Note that the self LLR messages $l_v$ are plotted as small bold lines.}
\label{fig:BCH_15_11_arch_multiloss}
\end{figure}
\subsection{Dataset}
The training data is created by transmitting the zero codeword through an AWGN channel with varying SNRs ranging from $1{\rm dB}$ to $6{\rm dB}$. Each mini batch has $20$ codewords for each SNR (a total of $120$ examples in the mini batch). For the test data we use codewords with the same SNR range as in the training dataset. Parity check matrices were taken from \cite{ParityCheckMatrix}.
\subsection{Results}
In this section we present the results of the deep neural decoding networks for various BCH block codes. In each code we observed an improvement compared to the BP algorithm. Note that when we applied our algorithm to the BCH(15,11) code, we obtained close to maximum likelihood results with the deep neural network. For larger BCH codes, the BP algorithm and the deep neural network still have a significant gap from maximum likelihood. The BER figures~\ref{fig:BCH_63_36_ber}, \ref{fig:BCH_63_45_ber} and \ref{fig:BCH_127_106_ber} show an improvement of up to $0.75{\rm dB}$ in the high SNR region. Furthermore, the deep neural network BER is consistently smaller or equal to the BER of the BP algorithm. This result is in agreement with the observation that our network cannot perform worse than the BP algorithm.
Figure \ref{fig:BCH_63_45_multiloss_ber} shows the results of training the deep neural network with multiloss. It shows an improvement of up to $0.9{\rm dB}$ compared to the plain BP algorithm. Moreover, we can observe that we can achieve the same BER performance of 50 iteration BP with 5 iteration of the deep neural decoder, This is equal to complexity reduction of factor 10.
We compared the weights of the BP algorithm and the weights of the trained deep neural network for a BCH(63,45) code. We observed that the deep neural network produces weights in the range from $-0.8$ to $2.2$, in contrast to the BP algorithm which has binary $“1”$ or $“0”$ weights. Figure~\ref{fig:weight_hist} shows the weights histogram for the last layer. Interestingly, the distribution of the weights is close to a normal distribution. In a similar way, every hidden layer in the trained deep neural network has a close to normal distribution. Note that Hinton \cite{hinton2010practical} recommends to initialize the weights with normal distribution.
In Figures~\ref{fig:layer4_bp} and~\ref{fig:layer4_dl} we plot the weights of the last hidden layer. Each column in the figure corresponds to a neuron described by Equation ~\eqref{eq:x_ie_RB_NN}. It can be observed that most of the weights are zeros except the Tanner graph weights which have value of 1 in Figure~\ref{fig:layer4_bp} (BP algorithm) and some real number in Figure~\ref{fig:layer4_dl} for the neural network. In Figure~\ref{fig:layer4_bp} and ~\ref{fig:layer4_dl} we plot a quarter of the weights matrix for better illustration.
\begin{figure}[thpb]
\centering
\includegraphics[width=0.983101925\linewidth]{bch_63_36_ber.png}
\caption{BER results for BCH(63,36) code.}
\label{fig:BCH_63_36_ber}
\end{figure}
\begin{figure}[thpb]
\centering
\includegraphics[width=0.983101925\linewidth]{bch_63_45_ber.png}
\caption{BER results for BCH(63,45) code.}
\label{fig:BCH_63_45_ber}
\end{figure}
\begin{figure}[thpb]
\centering
\includegraphics[width=0.983101925\linewidth]{bch_127_106_ber.png}
\caption{BER results for BCH(127,106) code.}
\label{fig:BCH_127_106_ber}
\end{figure}
\begin{figure}[thpb]
\centering
\includegraphics[width=0.983101925\linewidth]{bch_63_45_ber_multiloss.png}
\caption{BER results for BCH(63,45) code trained with multiloss.}
\label{fig:BCH_63_45_multiloss_ber}
\end{figure}
\begin{figure}[thpb]
\centering
\includegraphics[width=0.983101925\linewidth]{weight_hist.png}
\caption{Weights histogram of last hidden layer of the deep neural network for BCH(63,45) code.}
\label{fig:weight_hist}
\end{figure}
\begin{figure}[thpb]
\centering
\includegraphics[width=0.983101925\linewidth]{weight_mat_0_2D.png}
\caption{Weights of the last hidden layer in the BP algorithm for BCH(63,45) code.}
\label{fig:layer4_bp}
\end{figure}
\begin{figure}[thpb]
\centering
\includegraphics[width=0.983101925\linewidth]{weight_mat_3_2D.png}
\caption{Weights of the last hidden layer in deep neural network for BCH(63,45) code.}
\label{fig:layer4_dl}
\end{figure}
\section{CONCLUSIONS}
In this work we applied deep learning techniques to improve the performance of the BP algorithm. We showed that a ``soft'' Tanner graph can produce improvements when used in the BP algorithm instead of the standard Tanner graph. We believe that the BER improvement was achieved by properly weighting the messages, such that the effect of small cycles in the Tanner graph was partially compensated. It should be emphasized that the parity check matrices that we worked with were obtained from \cite{ParityCheckMatrix}. We have not evaluated our method on parity check matrices for which an attempt was made to reduce the number of small cycles. A notable property of our neural network decoder is that once we have trained its parameters, we can improve performance compared to plain BP without increasing the required computational complexity. Another notable property of the neural network decoder is that we learn the channel and the linear code simultaneously.
We regard this work as a first step in the implementation of deep learning techniques for the design of improved decoders. Our future work include possible improvements in the error rate results by exploring new neural network architectures and combining other decoding methods. Furthermore, we plan to investigate the connection between the parity check matrix and the deep neural network decoding capabilities.
\addtolength{\textheight}{-12cm}
\section*{ACKNOWLEDGMENT}
This research was supported by the Israel Science Foundation, grant no. 1082/13. The Tesla K40c used for this research was donated by the NVIDIA Corporation.
\bibliographystyle{IEEEtran}
| fcaa99d3f5d11897ce84b0d678159208a2dab04e | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}\label{sec:intro}
Microquasars are X-ray binaries that exhibit spatially resolved, extended radio emission. The nomenclature is motivated by a structural similarity with the quasar family of active galactic nuclei (AGN). Both object classes are believed to comprise a compact central object embedded in a flow of accreting material, and both exhibit relativistic, collimated jets. In the current paradigm, both microquasars and AGN derive their power from the gravitational potential energy that is liberated as ambient matter falls onto the compact object. Notwithstanding their morphological resemblance, microquasars and radio-loud AGN represent complementary examples of astrophysical jet production on dramatically disparate spatial and temporal scales. Indeed, conditions of accretion and mass provision that pertain to the supermassive ($10^{6}M_{\odot}\lesssim M_{\rm BH} \lesssim 10^{9}M_{\odot}$) black holes that power AGN and of the stellar-mass compact primaries of microquasars are markedly different.
In the latter, a companion star (or donor) provides the reservoir of matter for accretion onto a compact stellar remnant (or primary), which can be either a neutron star or a black hole. Partial dissipation of the resultant power output occurs in a disk of material surrounding the primary, producing the thermal and non-thermal X-ray emission, which is characteristic of all X-ray binary systems. Microquasars are segregated on the basis of associated non-thermal radio emission, indicative of synchrotron radiation in a collimated outflow, which carries away a sizeable fraction of the accretion luminosity \citep{2004MNRAS.355.1105F}.
In AGN, superficially similar jet structures are known to be regions of particle acceleration and non-thermal photon emission. The resulting radiation spectrum can extend from radio wavelengths into the very high-energy (VHE; $E_{\gamma}>100$ GeV) $\gamma$-ray\ regime.
Very high-energy $\gamma$-ray\ emission has been observed from many AGN in the blazar sub-class\footnote{ \url{http://tevcat.uchicago.edu/}},
where the jet axis is aligned close to the observer line-of-sight, as well as from a few radio galaxies (e.g. M87, \citealp{2003A&A...403L...1A}; Cen A, \citealp{2009ApJ...695L..40A}; NGC 1275, \citealp{Aleksic2012}) and starburst galaxies (e.g. M82, \citealp{Acciari2009}; NGC 253, \citealp{Abramowski2012}).
If similar jet production and efficient particle acceleration mechanisms operate in microquasars and AGNs, this might imply that the former object class are plausible sources of detectable VHE $\gamma$-ray\ emission as well, assuming that appropriate environmental conditions prevail.
The primarily relevant environmental conditions include the density of nearby hadronic material, which provides scattering targets for inelastic proton scattering interactions; these interactions produce pions that produce $\gamma$-rays when they subsequently decay. The ambient magnetic field strength is also important and influences the rate at which electrons lose energy via synchrotron radiation. Synchrotron photons contribute to the reservoir of soft photons that are available for inverse Compton (IC) up-scattering into the VHE $\gamma$-ray regime.
The argument for phenomenological parity between AGN and microquasars, possibly related to their structural resemblance, has been strengthened in recent years as the spectral properties of both radio and X-ray emission are remarkably similar for both stellar mass and supermassive black holes. In recent years these similarities led to the postulation of a so-called fundamental plane, which describes a three-dimensional, phenomenological correlation between the radio (5 GHz) and X-ray (2--10 keV) luminosities and the black hole mass \citep{uniplane, Falcke2004}.
However, the fundamental plane does not appear to extend into the TeV band. To date, only one well-established microquasar has been observed to emit in the VHE $\gamma$-ray\ regime. This is the Galactic black hole Cygnus X-1, which was marginally detected (at the $\sim4\sigma$ level) by the MAGIC telescope immediately prior to a $2-50\,$keV X-ray flare observed by the INTEGRAL satellite, the \textit{Swift} Burst Alert Telescope (BAT), and the \textit{RXTE} All-Sky Monitor (ASM) \citep{2007ApJ...665L..51A,2008A&A...492..527M}. \cite{Laurent2011} recently identified linear polarized soft $\gamma$-ray emission from Cygnus X-1 (see also \citealp{Jourdain2012}), thereby locating the emitter within the jets and identifying their capacity to accelerate particles to high energies (see however \citealp{Romero2014}).
Further motivation for observing microquasars in the VHE band arises from the recent identification of the high-mass microquasar Cygnus X-3 as a transient high-energy (HE; 100MeV $<E_{\gamma}<$ 100 GeV) $\gamma$-ray\ source by the \textit{Fermi} \citep{2009Sci...326.1512F} and AGILE \citep{2009Natur.462..620T} satellites. The identification of a periodic modulation of the HE signal is consistent with the orbital frequency of Cygnus X-3 and provides compelling evidence for effective acceleration of charged particles to GeV energies within the binary system \citep{2009Sci...326.1512F}. Based on evidence from subsequent reobservations, the HE $\gamma$-ray\ flux from Cygnus X-3 appears to be correlated with transitions observed in X-rays in and out of the so-called ultra-soft state, which exhibits bright soft X-ray emission and low fluxes in hard X-rays and is typically associated with contemporaneous radio flaring activity \cite[e.g.][]{2012MNRAS.421.2947C}.
Unfortunately, repeated observations of Cygnus X-3 using the MAGIC telescope did not yield a significant detection \citep{2010ApJ...721..843A}, despite the inclusion of data that were obtained simultaneously with the periods of enhanced HE emission detected using \textit{Fermi}. However, the intense optical and ultraviolet radiation fields produced by the Wolf-Rayet companion star in Cygnus X-3 imply a large optical depth for VHE $\gamma$-ray s due to absorption via $e^{+}e^{-}$ pair production \citep[e.g.][]{2010MNRAS.406..689B, 2012MNRAS.421.2956Z}. Accordingly, particle acceleration mechanisms akin to those operating in Cygnus X-3 may yield detectable VHE $\gamma$-ray\ fluxes in systems with fainter or cooler donors.
Mechanisms for $\gamma$-ray production in microquasars have been widely investigated, resulting in numerous hadronic \citep[see e.g.][]{2003A&A...410L...1R} and leptonic \citep[see e.g.][]{1999MNRAS.302..253A, 2002A&A...388L..25G, 2006A&A...447..263B, 2006ApJ...643.1081D, 2010MNRAS.404L..55D} models, describing the expected fluxes and spectra of microquasars in the GeV-TeV band. In both scenarios, a highly energetic population of the relevant particles is required and, consequently, emission scenarios generally localize the radiating region within the jet structures of the microquasar.
Leptonic models rely upon IC scattering of photons from the primary star in the binary system or photons produced through synchrotron emission along the jet to produce VHE $\gamma$-ray\ emission. In this latter scenario, they closely resemble models of extragalactic jets \citep{1981ApJ...243..700K,1989ApJ...340..181G}, but typically invoke internal magnetic fields that are stronger by factors $\sim1000$.
Consideration of hadronic models is motivated by the detection of Doppler-shifted emission lines associated with the jets of the microquasar SS 433 \citep[e.g.][]{1984ARA&A..22..507M}, indicating that at least some microquasar jets comprise a significant hadronic component.
Models of VHE $\gamma$-ray\ production by hadronic particles generally invoke electromagnetic cascades initiated by both neutral and charged pion decays \citep{2003A&A...410L...1R,1996A&A...309..917A,2005ApJ...632.1093R}.
Electron-positron pair production, $\gamma\gamma\rightarrow e^{+}e^{-}$ can absorb VHE $\gamma$-rays . In the case of 1 TeV $\gamma$-rays, the cross section for this process is maximised for ultraviolet target photons ($E_{\rm ph}\sim10$ eV), where its value may be approximated in terms of the Thomson cross section as $\sigma_{\gamma\gamma}\approx\sigma_{T}/5$ \citep[e.g.][]{1967PhRv..155.1404G}. In high-mass systems, the companion star is expected to produce a dense field of these target photons to interact with the $\gamma$-rays \citep[e.g.][]{1987ApJ...322..838P,1995SSRv...72..593M,2005ApJ...634L..81B,2006A&A...451....9D}. This process can be very significant and probably contributes to the observed orbital modulation in the VHE $\gamma$-ray\ flux from LS 5039 \citep{2006A&A...460..743A}. In contrast, the ultraviolet spectrum of low-mass microquasars is likely dominated by the reprocessing of X-ray emission in the cool outer accretion flow \citep{1994A&A...290..133V,2009MNRAS.392.1106G}, although jet emission might also be significant \citep{2006MNRAS.371.1334R}. Regardless of their origin, the observed optical and ultraviolet luminosities of low-mass X-ray binaries (LMXBs) are generally orders of magnitude lower than those of high-mass systems \citep{2006MNRAS.371.1334R}, and the likelihood of strong $\gamma$-ray\ absorption is correspondingly reduced.
However, microquasars may only become visible in the TeV band during powerful flaring events. These transient outbursts, characterised by the ejection of discrete superluminal plasma clouds, are usually observed at the transition between low- and high-luminosity X-ray states \citep{2004MNRAS.355.1105F}. Monitoring black-hole X-ray binaries with radio telescopes and X-ray satellites operating in the last decade enabled a classification scheme of such events to be established \citep{Homan2005}. Hardness-intensity diagrams (HIDs) plot the source X-ray intensity against X-ray colour (or hardness) and have subsequently been extensively used to study the spectral evolution of black-hole outbursts. At the transition from the so-called low-hard state to the high-soft states through the hard-to-soft intermediate states, the steady jet associated with the low-hard state is disrupted. These transient ejections, produced once the accretion disk collapses inwards, are more relativistic than the steady low-hard jets \citep{2004MNRAS.355.1105F}. Internal shocks can develop in the outflow, possibly accelerating particles that subsequently give rise to radio optically thin flares observed from black-hole systems; this phenomenological description is also extensible to neutron stars, although in that case jet radio power is lower by a factor 5--30 (\citealp{2006MNRAS.366...79M}).
Outburst episodes have also been observed in cases in which the source remained in the hard state without transition to the soft state \citep{Homan2005}. The detection (at the $\sim4\sigma$ level) by the MAGIC telescope of the high-mass, black-hole binary Cygnus X-1 took place during an enhanced $2-50\,$keV flux low-hard state as observed with the INTEGRAL satellite, the \textit{Swift} BAT, and the \textit{RXTE} ASM \citep{2008A&A...492..527M}. However, although the source X-ray spectrum remained unchanged throughout the TeV flare, such a bright hard state was unusually long when compared with previous observations of the source.
Here we report on contemporaneous observations with H.E.S.S. and \textit{RXTE} of the three microquasars V4641~Sgr, GRS~1915+105, and Circinus X-1. Information on the targets, the H.E.S.S. and \textit{RXTE} observations, and the corresponding trigger conditions are detailed in Sect.~\ref{sec:targets}. Analysis results are reported in Sect.~\ref{sec:mqs:results} and discussed in Sect.~\ref{sec:mqs:context}. In the appendix, detailed information on the X-ray analysis is reported, which in particular includes HIDs corresponding to the time of observations for the three studied sources.
\section{Targets and observations}
\label{sec:targets}
\subsection{Observations}\label{sec:mqs:observations}
The H.E.S.S.\ Imaging Atmospheric Cherenkov Telescope (IACT) array is situated on the Khomas Highland plateau of Namibia (23$^{\circ}16'18\arcsec$ south, $16^{\circ}30'00\arcsec$ east), at an elevation of 1800 m above sea level, and is capable of detecting a Crab-like source close to the zenith at the 5$\sigma$ level within $<5$ minutes under good observational conditions \citep{2006A&A...457..899A}. The point source sensitivity of H.E.S.S.\ enables it to detect a $2.0\times10^{-13}~\rm{cm}^{-2}\rm{s}^{-1}$ $\gamma$-ray\ flux above 1 TeV, at the 5$\sigma$ level within 25 hours, which, together with a low-energy threshold ($\sim$100 GeV), makes H.E.S.S.\ an invaluable instrument for studying the VHE $\gamma$-ray emission from microquasars. A fifth and larger telescope (commissioned in 2013) will allow the energy threshold to be lowered and will further increase the sensitivity of the instrument. For the analysis presented here, H.E.S.S. observations were carried out using the full, original four-telescope array. Owing to the diverse morphologies of the three binary systems, unique observational trigger criteria were established for each target employing various combinations of the observed X-ray state and radio flaring activity. Details are provided in subsequent paragraphs.
The \textit{Rossi X-ray Timing Explorer} (\textit{RXTE}) was a space-based X-ray observatory launched on 30 December 30 1995 and decommissioned on 5 January 2012. The primary mission of \textit{RXTE} was to provide astrophysical X-ray data with high timing resolution. This observatory occupied a circular low-earth orbit with an orbital period of $\sim90$ minutes and carried three separate X-ray telescopes. The \textit{Proportional Counter Array} (PCA) on board \textit{RXTE} comprised five copointing xenon and propane \textit{Proportional Counter Units} (PCUs), which were nominally sensitive in the energy range $\sim2-60$ keV with an energy resolution of $<18\%$ at 6 keV \citep{1993SPIE.2006..324Z}. For studies of rapidly varying sources like X-ray binaries, the PCA timing resolution of $\sim1\;\mu$s can prove invaluable. However, rapid timing measurements also require a bright source to provide sufficient counting statistics within such short time bins. The \textit{High Energy X-ray-Timing Experiment} (HEXTE) comprised two independent clusters of four phoswich scintillation detectors, which were sensitive to photons in the $\sim12-250$ keV energy range and had an energy resolution of $\sim9$ keV at 60 keV. The maximum timing resolution of HEXTE was $\sim8\;\mu$s. The \textit{All-Sky Monitor} (ASM) was a wide field-of-view instrument that monitored $\sim80\%$ of the sky over the course of each $\sim90$ minute orbit. This instrument consisted of three identical scanning shadow cameras and was designed to provide near-continuous monitoring of bright X-ray sources. Nominally, the ASM was sensitive in the energy range from $2-10\,$keV and had a rectangular field of view spanning $110^{\circ}\times12^{\circ}$.
Contemporaneous X-ray (\textit{RXTE}) and VHE $\gamma$-ray\ (H.E.S.S.) observations were performed at the epochs listed in Table \ref{table:targets}. In the following, we briefly review the observational characteristics of the target microquasars, GRS 1915+105, Circinus X-1, and V4641 Sgr. Established system parameters that characterise the three target microquasars are collated in Table \ref{tab:target_properties}.
\begin{table*}
\caption{Observationally established parameters of the target microquasars. $P_{\rm orb}$ is the binary orbital period, $M_{\star}$ is the mass of the companion star, $M_{\rm CO}$ is the compact object mass, $\theta_{\rm Jet}$ is the inclination of the observed jet with respect to the line of sight, and $d$ is the estimated distance to the microquasar.}\label{tab:target_properties}
\begin{center}
\begin{tabular}{lllll}
\hline\hline
& GRS 1915+105 & Circinus X-1 & V4641 Sgr \\
\hline
$P_{\rm orb} $ [d] & $33.85\pm 0.16$ (1) & 16.6 (2) & 2.82 (3) \\
$M_{\star} [M_{\odot}] $ & $0.47\pm 0.27$ (1) & $3-10$ (5) & $2.9\pm 0.4$ (10)\\
& $0.28\pm 0.02$ (4) & & \\
$M_{\rm CO} [M_{\odot}] $ & $12.4^{+2.0}_{-1.8}$ (6) & $\lesssim1.4$ (8) & $6.4\pm 0.6$ (10)\\
$\theta_{\rm Jet} [^{\circ}]$ & $60\phantom{.0}\pm 5$ (6) & $\lesssim3$ (9) & $\lesssim 12$ (3)\\
$d$ [kpc] & $8.6^{+2.0}_{-1.6}$ (6,7) & $9.4^{+0.8}_{-1.0}$ (9) & $6.2\pm 0.7$ (10)\\
\hline
\end{tabular}
\end{center}
\tablebib{(1) \citet{0004-637X-768-2-185}; (2) \citet{2007ATel..985....1N}; (3) \citet{2001ApJ...555..489O}; (4) \citet{ziolkowski2015}; (5) \citet{1999MNRAS.308..415J, 2007MNRAS.374..999J}; (6) \citet{0004-637X-796-1-2};(7) \citet{Zdziarski2014};(8) \citet{tennant1, 2010ApJ...719L..84L}; (9) \citet{0004-637X-806-2-265}; (10)\citet{0004-637X-784-1-2}; }
\end{table*}
\begin{table*}[th!]
\caption{\label{table:targets} Observational epochs for each target microquasar.}
\centering
\begin{tabular}{ l c c c } \hline \hline
~Target~ & \textit{RXTE} ObsId & ~\textit{RXTE} Observations (MJD)~ & ~H.E.S.S. Observations (MJD)~\\
\hline
GRS 1915+105
&90108-01-01-00&53123.091 $\rightarrow$ 53123.109 & 53123.067 $\rightarrow$ 53123.150\\
&90108-01-02-00&53124.074 $\rightarrow$ 53124.094 & 53124.079 $\rightarrow$ 53124.162\\
&90108-01-03-00&53125.130 $\rightarrow$ 53125.149 & 53125.083 $\rightarrow$ 53125.148\\
&90108-01-04-00&53126.114 $\rightarrow$ 53126.129 & 53126.109 $\rightarrow$ 53126.132\\
&90108-01-05-00&53127.097 $\rightarrow$ 53127.114 & 53127.106 $\rightarrow$ 53127.165\\
&90108-01-06-00&53128.150 $\rightarrow$ 53128.165 & 53128.149 $\rightarrow$ 53128.165\\
\hline
Cir X-1
&90124-02-01-00&53174.749 $\rightarrow$ 53174.761 & 53174.748 $\rightarrow$ 53174.832\\
&90124-02-02-00&53175.768 $\rightarrow$ 53175.780 & 53175.735 $\rightarrow$ 53175.822\\
&90124-02-03-00&53176.781 $\rightarrow$ 53176.793 & 53176.772 $\rightarrow$ 53176.858\\
\hline
V4641 Sgr
&90108-03-01-00&53193.904 $\rightarrow$ 53193.924 & Not Observed\\
&90108-03-02-00&53194.887 $\rightarrow$ 53194.908 & 53194.883 $\rightarrow$ 53194.926\\
&90108-03-03-00&53195.871 $\rightarrow$ 53195.892 & 53195.890 $\rightarrow$ 53195.931\\
\hline
\end{tabular}
\end{table*}
\subsection{GRS 1915+105}\label{sec:upperlims:grs1915}
GRS 1915+105 is a dynamically established black-hole binary first identified by the WATCH all-sky monitor on board the \textit{GRANAT} satellite \citep{castro}. Observations in the optical and near-infrared using the Very Large Telescope succeeded in identifying the stellar companion as a low-mass KM III giant \citep[][]{greiner2}. GRS 1915+105 gained a measure of celebrity as the prototype Galactic superluminal source \citep{1994Natur.371...46M}.
In a detailed study of the X-ray light curves of GRS 1915+105, \citet[][]{2000A&A...355..271B} succeeded in identifying 12 distinct variability classes, internally characterised by the duration and juxtaposition of three separate spectral states.
Episodes of class $\chi$ behaviour, belonging to state C and lasting several days, are known as plateaux and are invariably terminated by flaring activity in the radio, infrared, and X-ray bands \citep[][]{2004ARA&A..42..317F}. In contrast with the evidence for self-absorbed synchrotron radiation seen in the spectrally hard, low-luminosity state C, and often associated with continuous relativistic jets \citep[][]{2002MNRAS.331..745K}, radio spectra obtained during the end-plateau flaring episodes indicate optically thin synchrotron emission \citep[][]{1997MNRAS.290L..65F,1998ApJ...494L..61E}.
Occasionally, these flaring episodes are linked to powerful discrete plasma ejections with instantaneous power output reaching $\gtrsim10^{39}$ erg s$^{-1}$ \citep[][]{1994Natur.371...46M, Zdziarski2014}.
Modelling the emission from these discrete relativistic ejecta, \citet[][]{1999MNRAS.302..253A} showed that inverse Comptonisation of emitted synchrotron photons into the GeV-TeV regime could produce significant and persistent $\gamma$-ray\ fluxes that remain detectable for several days.
\citet{2009A&A...508.1135H} and \citet{2009arXiv0907.1017S} reported VHE $\gamma$-ray\ observations of GRS 1915+105; these authors derived integral flux upper limits of $6.1\times10^{-13}$ cm$^{-2}$s$^{-1}$ above 410 GeV and $1.17 \times10^{-12}$ cm$^{-2}$s$^{-1}$ above 250 GeV, respectively.
For the analysis presented here, GRS 1915+105 was observed by H.E.S.S.\ between 28 April and 3 May 2004 in response to an apparent decrease in the 15 GHz radio flux, which was monitored by the Ryle Telescope during a $\sim50$ day plateau state \citep{Pooley2006}, as shown in Figure \ref{evol_GRS1915}, in which coloured markers indicate the H.E.S.S.\ observation epochs. On the basis of previously observed behaviour, it was thought likely that the observed radio evolution signalled the end of the plateau state and, therefore, that flaring activity would begin within the subsequent 24 hours.
The \textit{RXTE} observations of GRS 1915+105 comprised six individual pointings, contributing to accumulated PCA and HEXTE livetimes of $7.6\,$ksec and $5176\,$s, respectively. Fifteen contemporaneous H.E.S.S.\ observations were obtained, constituting an overall livetime of 6.9 hours.
\subsection{Circinus X-1}\label{sec:upperlims:cirx1}
Circinus X-1 (hereafter Cir X-1) has been extensively studied since its initial identification \citep{1971ApJ...169L..23M}, revealing a somewhat confusing collection of complex observational characteristics.
Repeated observation of type I X-ray bursts \citep[][]{tennant2,tennant1,2010ApJ...719L..84L} definitively identifies the compact primary in Cir X-1 as a low magnetic field ($B\lesssim10^{11}$G) neutron star. Further sub-classification as a Z or atoll source \citep[see, for example,][for an explanation of the distinction between these two classes]{2007A&ARv..15....1D} is not possible since Cir X-1 exhibits several confusing spectral and timing properties, subsets of which are characteristic of both source types \citep[see e.g.][]{1999ApJ...517..472S, ooster}. Accordingly, established paradigms for disk-jet coupling in X-ray binaries with neutron star primaries \citep[e.g.][]{2006MNRAS.366...79M} cannot be reliably employed.
At radio wavelengths, the jets of Cir X-1 display notable structure on arcsecond scales, appearing as a bright core with significant extension along the axial direction of the arcminute jets \citep[][]{1998ApJ...506L.121F}. In fact, the observed extension is rather asymmetric with a ratio of at least two between the observed fluxes of the two opposing jets. Interpreted as pure relativistic aberration, this asymmetry implies a jet velocity $\gtrsim0.1c$. Cir X-1 has also been observed to eject condensations of matter with apparently superluminal velocities $\gtrsim15c$ \citep[][]{fenderradio}. These observations imply a physical velocity for the ejecta $v>0.998c$ with a maximum angle between the velocity vector and the line of sight $\theta<5^{\circ}$.These results identify Cir X-1 as a microblazar, a Galactic, small-scale analog of the blazar class of AGN, several of which are known sources of VHE $\gamma$-rays.
Definitive classification of the donor star in Cir X-1 is somewhat problematic. The low apparent magnitude of the detected optical counterpart implies a dereddened luminosity consistent with a low-mass or sub-giant companion, implying that Cir X-1 is a LMXB with a high orbital eccentricity $e\sim0.7-0.9$ \citep[e.g.][]{1999MNRAS.308..415J}. Nonetheless, recent near-infrared \citep[][]{2003A&A...400..655C} and I-band optical \citep[][]{2007MNRAS.374..999J} observations reveal emission features that are consistent with a mid-B supergiant, suggesting a more moderate eccentricity $e\sim0.45$.
Observations of Cir X-1 in the X-ray band reveal a long-term evolution of the average source brightness. Fluxes rose monotonically from near-undetectable in the early 1970s to a peak value of $\sim 1.5-2$ Crab ($1.5-10\,$keV) at the turn of the millennium, before returning over a period of $\sim4$ years to their pre-rise levels \citep[][]{2003ApJ...595..333P}. Various X-ray spectra, obtained during epochs of both high and low flux, display evidence of complex and variable emission and absorption processes.
A previous analysis of H.E.S.S.\ observations of Cir X-1 was presented by \cite{2008AIPC.1085..245N}, who derived a preliminary upper limit to the $\gamma$-ray\ flux above 1 TeV of $1.9\times10^{-13}$ cm$^{-2}$s$^{-1}$ corresponding to a detector livetime of 28 hours.
The H.E.S.S.\ observations of Cir X-1 reported here began on 18 June 2004 and were scheduled to coincide with the periastron passage of the binary components. The previous observation of regular radio flares during this orbital interval were thought to provide a good chance of observing during a period of outburst with the associated possibility that superluminal ejections might occur.
The \textit{RXTE} observations of Cir X-1 comprised three individual pointings, corresponding to orbital phase intervals $0.0486\le\phi\le0.0498$, $0.1104\le\phi\le0.1112,$ and $0.1718\le\phi\le0.1725$ (using the radio flare ephemeris of \citet{2007ATel..985....1N}), and contributing to an accumulated PCA livetime of $2576\,$s. A data set comprising 12 contemporaneous H.E.S.S.\ observations yielded a combined livetime of 5.4 hours.
\subsection{V4641 Sgr}\label{sec:upperlims:v4641sgr}
V4641 Sgr is the optical designation of the habitually weak X-ray source SAX J1819.3-2525 (XTE J1819-254), which was independently identified using the \textit{BeppoSAX} \citep[][]{IAUC7119zand} and \textit{RXTE} \citep[][]{IAUCmarkwardt} satellites. Optical spectroscopic measurements \citep[][]{2001ApJ...555..489O, 2005MNRAS.363..882L} strongly suggest a late B- or early A-type companion with an effective temperature $T_{\rm{eff}}\approx10500$ K.
The mass of the compact primary, $6.4\pm0.6\, M_{\odot}$ \citep[][]{0004-637X-784-1-2}, categorises V4641 Sgr as a firm black hole candidate.
V4641 Sgr is probably best known for its exhibition of rapid and violent outbursts. Perhaps the most spectacular of these events was the super-Eddington flare detected by the \textit{RXTE ASM} in September 1999. The observed X-ray fluxes ($2-12\,$keV) increased sharply, reaching $\approx12.2$ Crab within eight hours before fading again to below 0.1 Crab in under two hours \citep[][]{2002A&A...391.1013R}. Powerful contemporaneous flares were also observed at hard X-ray \citep[][]{IAUCmccollough}, optical \citep[][]{IAUC7253stubbings}, and radio \citep[][]{hjellming} wavelengths. In fact, Very Large Array (VLA) radio observations obtained within a day of the X-ray flare resolved a bright jet-like radio structure $\approx0.25$ arcsec in length \citep[][]{hjellming}. Assuming the most likely hypothesis, i.e., that the ejection is coincident with some phase of the X-ray flare, proper motions in the range $0.22\lesssim\mu_{\rm{jet}}\lesssim1.1$ arcsec day$^{-1}$ are derived. At the minimum distance $d=5.5$ kpc, the implied lower limit to the apparent velocity of the ejecta is $7c\lesssim v_{\rm{min}}\lesssim35c$, which is comparable with the extragalactic jets seen in blazars. Indeed, the remarkably high apparent velocities imply that V4641 Sgr may be a microblazar with a relativistic jet moving close to the line of sight ($\theta_{\rm{jet}}\lesssim12^{\circ}$; from \cite{2001ApJ...555..489O}). Subsequent, weaker broadband outbursts
have also been observed, suggesting recurrent activity on a timescale $\sim1-2$ years \citep[e.g.][]{hjellming61, 2002IAUC.7928....2R, rupen172, swank295}.
Observations of V4641 Sgr with H.E.S.S.\ were initiated on 7 July 7 2004 (MJD 53193) in response to the source brightening rapidly in the radio \citep{2004ATel..296....1R}, optical \citep{2004ATel..297....1R}, and X-ray \citep{swank295} bands. The resultant \textit{RXTE} exposure comprised three observations, each contributing to an accumulated PCA livetime of $5\,$ksec.
Two pairs of $\sim$30 minute H.E.S.S.\ observations were obtained contemporaneously with the final two \textit{RXTE} pointings. In total, the four separate exposures constitute an overall livetime of 1.76 hours.
\section{Analysis and results}\label{sec:mqs:results}
X-ray data reduction with the \texttt{FTOOLS 5.3.1} software suite employed the data selection criteria regarding elevation, offset, electron contamination, and proximity to the South Atlantic Anomaly recommended by the \textit{RXTE} Guest Observer Facility website\footnote{http://heasarc.nasa.gov/docs/xte/xhp\_proc\_analysis.html}. For each observation, the PCA STANDARD2 data were extracted from all available PCUs.\ For all observations, HEXTE Archive mode data for both clusters were extracted following the recommended procedures for time filtering and background estimation. Spectral analysis was carried out using the \texttt{XSPEC 12.6.0} package \citep{1996ASPC..101...17A}. Spectral fits for GRS 1915+105 use both PCA and HEXTE data, including an energy range of $3-200\,$keV. For bright X-ray sources, such as GRS 1915+105, statistical errors on the number of counts per spectral bin become insignificant relative to dominant uncertainties in the instrument response. Accordingly, a 1\% systematic error was added to all PCA channels. The remaining sources, Cir X-1 and V4641 Sgr, were not significantly detected by HEXTE and therefore only PCA data in the $3-20\,$keV range were considered to ensure good data quality.
These targets were sufficiently faint that the spectral bin uncertainties were statistically dominated and the addition of a systematic error component was not required.
In the case of GRS 1915+105, power density spectra (PDS) were derived using the ftool \texttt{powspec}. For each \textit{RXTE} pointing of GRS 1915+105, individual PDS were extracted from $128\,$s intervals comprising $2^{14}$ bins. The resulting spectra were then averaged to produce a PDS for the total light curve with errors estimated using the standard deviation of the average of the power in each frequency bin. The overall PDS were logarithmically rebinned and normalised to represent the squared fractional RMS in each frequency bin \citep[see e.g.][]{1988SSRv...46..273L}. Corrections for instrument deadtime \citep[see, for example,][]{2000A&A...363.1013R} were applied (although this was found to have a negligible effect in the frequency range under consideration) and the expected white noise level was subtracted \citep{1983ApJ...266..160L}. Similar temporal analyses for the remaining targets proved unfeasible because of insufficient count statistics at all but the lowest frequencies.
The $\gamma$-ray\ analysis followed the standard point-source procedure described in \cite{crabpaper}. The reflected background model \citep[see, for example,][]{2007A&A...466.1219B} was used to derive overall results in conjunction with both the \textit{hard} and \textit{standard} event selection cuts described by \cite{crabpaper}. Hard cuts (image size $\ge200$ photoelectrons) tend to enhance the signal of sources with power-law spectral slopes that are harder in comparison to the dominant cosmic ray background. Standard cuts (image size $\ge80$ photoelectrons) provide less sensitivity in such cases but allow a lower energy threshold. No significant detection was obtained for any of the three targets. Upper limits to the VHE $\gamma$-ray flux above the instrumental threshold energy were therefore derived at the 99\% confidence level using the profile likelihood method \citep{2005NIMPA.551..493R}.
\subsection{GRS 1915+105}\label{subsec:grs_res}
As illustrated by the PCA and ASM light curves shown in Figure \ref{fig:grs_lightcurves}, the X-ray count rate was stable to within $\sim10\%$ during each observation and varied by no more than $\sim20\%$ between observations. Indeed, the long-term \textit{RXTE} ASM light curve in Figure \ref{fig:grs_lightcurves} (top panel) clearly indicates that the H.E.S.S.\ observation epochs occurred during an extended and relatively faint plateau in the $2-10\,$keV flux.
\begin{figure*}[thp!]
\includegraphics[width=\textwidth]{grs_lc_plot2-eps-converted-to.pdf}
\caption{\textit{RXTE} ASM, and PCA light curves for GRS 1915+105 together with H.E.S.S. upper limits derived from individual $\sim28$ minute runs using standard event selection cuts. The blue shaded bands on the ASM light curve indicate the extent of the H.E.S.S. observations, while on the H.E.S.S. upper limit plots similar bands illustrate the duration of the contemporaneous PCA observations. The plotted H.E.S.S.\ upper limits correspond to different threshold energies and the vertical scale of each light curve has been optimised for the plotted data.}\label{fig:grs_lightcurves}
\end{figure*}
The $3-200\,$keV X-ray spectra shown in Figure \ref{fig:grs1915_spec} also exhibit remarkable stability between observations. The individual spectra are dominated by a hard non-thermal component and strongly suggest class $\chi$ (in state C) behaviour \citep[e.g.][]{2001ApJ...554L..45Z,2001ApJ...558..276T}, which is confirmed by the location of the observations in the HID of Figure \ref{HID_GRS1915}, according to the classification of \citet[][]{2000A&A...355..271B}.
Figure \ref{evol_GRS1915} shows the contextual X-ray and 15 GHz radio light curves of GRS 1915+105 during a two-month period that brackets the H.E.S.S.\ observation epochs. It is evident from the figure that H.E.S.S.\ observed the target during and extended radio-loud plateau ($\sim$80 mJy; for historical flux comparison, a three-year monitoring campaign is presented in \citet{Pooley2006}). The plateau ended approximately ten days later with a combined radio and X-ray flaring episode. The assertion of radio-loud behaviour at the H.E.S.S.\ observation epochs is supported by the quasi-periodic oscillation (QPO) analysis presented in Figure \ref{fig:grs_powspecs}.
For a detailed discussion see Appendix~\ref{app:GRS1915state}.
In summary, the combined spectral and temporal analyses indicate a robust association of the contemporaneous H.E.S.S.\ observation with the radio-loud $\chi$ state, and the presence of steady, mildly relativistic jets at the time of observation may be confidently inferred.
The contemporaneous H.E.S.S.\ observations did not yield a significant VHE $\gamma$-ray\ detection. The significances corresponding to the total H.E.S.S.\ exposure are computed using Eq. 17 from \citet{1983ApJ...272..317L} and are listed in Table \ref{tab:mqs:hess_sigmas}. Figure \ref{fig:grs_lightcurves} plots runwise 99\% confidence level upper limits to the integral VHE $\gamma$-ray\ flux above the instrumental threshold energy and illustrates the overlap between the \textit{RXTE} and H.E.S.S.\ observations. Integral flux upper limits, which correspond to the overall H.E.S.S.\ exposure, are listed in Table \ref{tab:mqs:hess_uls}.
\begin{table*}[th!]
\caption{H.E.S.S. VHE $\gamma$-ray\ significances corresponding to hard and standard event selection regimes.}\label{tab:mqs:hess_sigmas}
\centering
\begin{tabular}{lllllll}
\hline\hline
Target & Image Cuts &$N_{\rm ON}$ [events]& $N_{\rm OFF}$ [events]& $\alpha$ & Excess [events]& Significance [$\sigma$] \\
\hline
\multirow{2}{*}{GRS 1915+105} & Standard & 471 & 7127 & 0.073 & -51.6 & -2.2 \\
& Hard & 36& 783 &0.060 & -10.9 & -1.6 \\
\hline
\multirow{2}{*}{Cir X-1} & Standard & 385& 5959& 0.068& -20.1& -1.0 \\
& Hard & 45 &648 &0.056 &\phantom{-}9.1 &\phantom{-}1.4 \\
\hline
\multirow{2}{*}{V4641 Sgr} & Standard & 161 &2373& 0.067& \phantom{-}1.2 &\phantom{-}0.1\\
& Hard & 11 &275 &0.055 &-4.2 &-1.11\\
\hline
\end{tabular}
\end{table*}
\begin{table*}[th!]
\caption{H.E.S.S.\ VHE $\gamma$-ray\ integral flux upper limits above the telescope energy threshold corresponding to both event selection regimes. The upper limits are derived at the 99\% confidence level, assuming a power-law spectrum ($dN/dE\propto E^{-\Gamma}$) with the photon index $\Gamma_{\rm std} = 2.6$ for standard cuts and $\Gamma_{\rm hard} = 2.0$ for hard cuts. The rather high threshold energies derived for GRS 1915+105 and Cir X-1 are the result of large maximum observational zenith angles.}\label{tab:mqs:hess_uls}
\centering
\begin{tabular}{llllll}
\hline\hline
Target & Cuts & $T_{\rm Live}$ [s] & $\bar{Z}_{\max}$ [$^{\circ}$] & $E_{\rm thresh}$ [GeV]& $I(>E_{\rm thresh})$ [ph cm$^{-2}$s$^{-1}$] \\
\hline
\multirow{2}{*}{GRS 1915+105} & Standard & 24681 & 40.6 & 562 &$<7.338\times10^{-13}$\\
& Hard & 24681 & 40.6 & 1101 & $<1.059\times10^{-13}$\\
\hline
\multirow{2}{*}{Cir X-1} & Standard & 19433 & 43.6& 562 &$<1.172\times10^{-12}$ \\
& Hard & 19433 & 43.6 & 1101 & $<4.155\times10^{-13}$\\
\hline
\multirow{2}{*}{V4641 Sgr} & Standard & 6335 & 8.4 & 237 &$<4.477\times10^{-12}$\\
& Hard & 6335 & 8.4 & 422 & $<4.795\times10^{-13}$ \\
\hline
\end{tabular}
\end{table*}
\subsection{Cir X-1}\label{subsec:cir_res}
\begin{figure*}[thp!]
\includegraphics[width=\textwidth,angle=0]{cir_lc_plot2-eps-converted-to.pdf}
\caption{\textit{RXTE} ASM and PCA light curves for Cir X-1 together with H.E.S.S. upper limits derived from individual $\sim28$ minute runs using standard event selection cuts. The blue shaded bands on the ASM light curve indicate the extent of the H.E.S.S. observations, while on the H.E.S.S. upper limit plots similar bands illustrate the duration of the contemporaneous PCA observations (OBS $1-3$). The partitioning of OBS 3 into sub-intervals A$-$D based on $2-20\,$keV X-ray flux is illustrated in the bottom right panel. The plotted H.E.S.S.\ upper limits correspond to different threshold energies, and the vertical scale of each light curve has been optimised for the plotted data.}\label{fig:cir_lightcurves}
\end{figure*}
The ASM light curve shown in Figure \ref{fig:cir_lightcurves} reveals that the H.E.S.S.\ observation epochs occurred during an extended $\sim4$ day dip in the $2-10\,$keV X-ray flux. Additionally, it should be noted that the observations reported here were obtained during an extremely faint episode in the secular X-ray flux evolution of Cir X-1 \citep{2003ApJ...595..333P}, which is also evident from the HID presented in Figure \ref{HID_CirX1}. As a consequence, the measured X-ray fluxes are significantly lower than most others reported for this source. As illustrated in Figure \ref{fig:cir_lightcurves}, the individual PCA light curves obtained during the first two pointings are characterised by a relatively low count rate, which remains approximately constant throughout each observation. In marked contrast, the third observation exhibits clear variability with count rates doubling on timescales of $\sim50$s.
A detailed analysis of the obtained spectra (see Appendix~\ref{app:CirState}) reveals that the observed flux variability is accompanied by marked variations in spectral shape. These can be interpreted as hinting towards a strong mass transfer during the periastron passage and subsequent dramatic evolution of the local radiative environment.
H.E.S.S.\ observations obtained contemporaneously with the \textit{RXTE} pointings yield a non-detection that is evident from the significances listed in Table \ref{tab:mqs:hess_sigmas}. Figure \ref{fig:cir_lightcurves} plots runwise 99\% confidence level upper limits to the integral VHE $\gamma$-ray\ flux above the instrumental threshold energy and illustrates complete overlap between the \textit{RXTE} and H.E.S.S.\ observations. Integral flux upper limits, which correspond to the overall H.E.S.S.\ exposure, are listed in Table \ref{tab:mqs:hess_uls}.
\subsection{V4641 Sgr}\label{subsec:v46_res}
\begin{figure*}[thp!]
\includegraphics[width=\textwidth]{v4s_lc_plot2-eps-converted-to.pdf}
\caption{\textit{RXTE} ASM and PCA light curves for V4641 Sgr together with H.E.S.S. upper limits derived from individual $\sim28$ minute runs using standard event selection cuts. The blue shaded bands on the ASM light curve indicate the extent of the H.E.S.S. observations, while on the H.E.S.S. upper limit plots they illustrate the duration of the contemporaneous PCA observations. The plotted H.E.S.S.\ upper limits correspond to different threshold energies and the vertical scale of each light curve has been optimised for the plotted data.}\label{fig:v4s_lightcurves}
\end{figure*}
Figure \ref{fig:v4s_lightcurves} shows \textit{RXTE} PCA light curves derived from three pointed observations. The individual light curves indicate various degrees of X-ray variability with the clearest evidence for flaring visible as a sharp $\sim5$-fold count rate fluctuation during the first observation. In marked contrast, the second observation is uniformly faint with the $\chi^{2}$ probability of constant count rate $P_{\rm const}=0.97$ and, hence, this observation is consistent with a period of steady, low-level emission. Subsequently, the third observation reveals a reemergence of mild variability ($P_{\rm const}=0.07$) with $\sim2$-fold count rate fluctuations occurring on timescales of $\sim500$s.
Radio data shown on Figure \ref{fig:V4641hardness_radio} right were obtained using the VLA and Australia Telescope Compact Array (ATCA) between MJD 53190 and MJD 53208. They indicate rapid variability with peak flux densities of $\sim30$ mJy observed on MJD 53193 \citep{2004ATel..296....1R, 2004ATel..302....1S,2004ATel..303....1R}. An optically thin radio spectrum ($S_{\nu}\propto \nu^{-0.7}$) observed on MJD 53191 was interpreted by \citet{2004ATel..296....1R} as the signature of a decaying radio flare.
Radio observations were triggered by an optical alert from VSNET (MJD 53190) in combination with a \textit{RXTE} PCA measurement during a Galactic bulge scan (MJD 53189) that
revealed a 2-10 keV X-ray flux equivalent to 8.2 mCrab. For comparison, the August 2003 flare of V4641 Sgr reached 66 mCrab, while quiescent fluxes are typically < 0.5 mCrab \citep{swank295}.
As shown in Figure \ref{fig:V4641hardness_radio} right, the dedicated \textit{RXTE} PCA observation and H.E.S.S.\ observations took place between two radio flares, which is consistent with the X-ray variability evolution illustrated in Figure \ref{fig:v4s_lightcurves}.
While V4641 Sgr is evidently the most X-ray-faint binary in the studied sample, it simultaneously exhibits the hardest spectrum, as shown by the hardness values in Figure \ref{fig:V4641hardness_radio} (left-hand panel). Furthermore, the evolution of the hardness is consistent with contemporaneous observations of rapid flux evolution in the radio band \citep{2004ATel..302....1S}\footnote{ \url{http://www.ph.unimelb.edu.au/~rsault/astro/v4641/}} (Figure \ref{fig:V4641hardness_radio}).
To help place the H.E.S.S.\ and \textit{RXTE} observations in a historical context, the HID for V4641 Sgr in Figure \ref{HID_V4641Sgr} displays the entire archival \textit{RXTE} PCA data set for this target, and compares the X-ray characteristics corresponding to the H.E.S.S.\ observation periods (different symbols are used to indicate observations obtained on each day in the range MJD 53193-5) with three flaring episodes observed with \textit{RXTE}.
On 15 September 1999 (orange markers in Figure \ref{HID_V4641Sgr}), a 1500s \textit{RXTE} observation revealed a source flux evolution that is characterised by rapid, large-amplitude variability before reverting to a soft, low intensity state after $\sim1000$s. An optical flare that preceded the \textit{RXTE} observations likely corresponds with the onset of the short 10-hour outburst, which \citet{Wijnands2000} associated with a low $\dot{M}$ accretion event. Historically, flaring episodes exhibited by V4641 Sgr are often short, unpredictable, and relatively faint, which implies that many may go unnoticed.
Data corresponding to two longer outbursts, spanning the periods 24-26 May 2002 and 5-7 August 2003, are also illustrated in Figure \ref{HID_V4641Sgr}. In the coordinates of the HID, both episodes are topologically similar to the 1999 outburst, but shifted towards fainter harder regions.
Evidently, the X-ray fluxes that correspond with H.E.S.S.\ observation epochs indicated in Figure \ref{HID_V4641Sgr} are substantially fainter than any of these historically remarkable outbursts.
In summary, in view of the various multiwavelength data, it seems likely that V4641 Sgr underwent a period of mild activity that spanned the H.E.S.S.\ observation epochs.
The contemporaneous H.E.S.S.\ data are consistent with non-detection with the corresponding $\gamma$-ray\ significances listed in Table \ref{tab:mqs:hess_sigmas}. Technical issues prevented $\gamma$-ray\ data corresponding to the first \textit{RXTE} observation from being obtained.
Simultaneous $\gamma$-ray\ observations were obtained corresponding to the second \textit{RXTE} exposure, which showed no indications of X-ray variability. Although the source began to show increased X-ray activity during the third \textit{RXTE} observation, the degree of overlap with the corresponding H.E.S.S.\ observations was minimal. At radio, optical, and X-ray energies, V4641 Sgr exhibits rapid variability on timescales $\sim10$ minutes or less \citep[e.g.][]{2005IBVS.5626....1U,2006ApJ...637..992M}. Optimistically, the compelling evidence for mild broadband flaring admits the possibility that the H.E.S.S.\ observations monitor a transient outburst event.
Integral flux upper limits above the instrumental threshold energy, which correspond to the overall H.E.S.S.\ exposure at the position of V4641 Sgr, are listed in Table \ref{tab:mqs:hess_uls}.
\section{Discussion}\label{sec:mqs:context}
The principal aim of this investigation was to obtain contemporaneous X-ray and VHE $\gamma$-ray\ observations of three known superluminal microquasars during major flaring events. However, the results presented in $\S$\ref{sec:mqs:results} indicate that the interpretation of the VHE $\gamma$-ray\ non-detections cannot proceed under the assumption of energetic flaring or bulk superluminal ejections at the time of observation. Nonetheless, upper limits to the VHE $\gamma$-ray\ flux were derived and an analysis of the contemporaneous \textit{RXTE} observations has helped to reveal the X-ray behaviour corresponding to the H.E.S.S.\ observation epochs. These data facilitate the straightforwards derivation of constraints on the $\gamma$-ray\ luminosity of the target binary systems. In Table \ref {tab:mqs:luminosties} the calculated flux upper limits were used to infer the maximum $\gamma$-ray\ luminosities above the target-specific, instrumental threshold energy for each target binary system by assuming the maximum source distance estimate presented in Table \ref{tab:target_properties}.
\begin{table*}[thp!]
\caption{Estimated maximum VHE $\gamma$-ray\ luminosities of the target microquasars, which would still be consistent with a non-detection given the flux upper limits presented in Table \ref{tab:mqs:hess_uls}. Source distances correspond to the largest estimate that was found in the literature (see $\S$ \ref{sec:targets}). The energy threshold of Cherenkov telescope arrays increases with observational zenith angle.}\label{tab:mqs:luminosties}
\centering
\begin{tabular}{llll}
\hline\hline
Target & Maximum Distance Estimate& $E_{\rm thresh}$& Luminosity above $E_{\rm thresh}$\\
& [kpc] & [GeV] & [erg s$^{-1}$] \\
\hline
GRS 1915+105 & 10.6 & 562 & $<2.3\times10^{34}$\\
Cir X-1 & 10.2 & 562 & $<3.4\times10^{34}$\\
V4641 Sgr & 6.9 & 237 & $<2.5\times10^{34}$\\
\hline
\end{tabular}
\end{table*}
Analysis of the contemporaneous X-ray and radio observations conclusively places GRS 1915+105 in a radio-loud plateau state at the time of observation. In contrast with the superluminal flaring episodes, this state is characterised by the production of continuous, mildly relativistic radio jets with an estimated power of $\sim3\times10^{38}$ erg s$^{-1}$ (\cite{2002MNRAS.331..745K}, assuming a distance of 11 kpc). Theoretically, it seems unlikely that bright VHE $\gamma$-ray\ emission would be expected from the compact self-absorbed jets, which are typical of the plateau state of GRS 1915+105. For example, a leptonic emission model developed by \cite{2006A&A...447..263B} to simulate the broadband emission of microquasar jets in the low-hard state predicts VHE $\gamma$-ray\ luminosities $\lesssim10^{33}$ erg s$^{-1}$ that are consistent with the H.E.S.S.\ non-detection.
Notwithstanding the plausibility of VHE $\gamma$-ray\ emission in the plateau state, a comparison of the estimated jet power with the maximum $\gamma$-ray\ luminosity listed in Table \ref{tab:mqs:luminosties} reveals that the jet power conversion efficiency is constrained to be $\lesssim0.008\%$ for $\gamma$-ray\ production above 562 GeV. For comparison, corresponding efficiency estimates for $\gamma$-ray\ production were derived for the steady, compact jets of other microquasars that were observed in appropriate states. The published MAGIC upper limit on the VHE $\gamma$-ray\ luminosity of Cygnus X-3 during its hard state implies a somewhat larger maximum conversion efficiency of 0.07\% \citep{2010ApJ...721..843A} and a similar value is obtained from MAGIC upper limits on the steady VHE emission from Cygnus X-1 \citep{2007ApJ...665L..51A}.
These efficiencies are inferred from the directly observed jet power, and should be distinguished from the higher jet powers that were indirectly derived from the observation of radio-emitting bubbles inflated by microquasar jets (see e.g. \citealp{Gallo2005} for Cyg X-1, and \citealp{Pakull2010}, \citealp{Soria2010} for S26 in NGC 7793).
We presented an analysis of the entire H.E.S.S.\ data set for GRS 1915+105 \citep{2009A&A...508.1135H} and we derived an upper limit to the $\gamma$-ray\ flux above 0.41 TeV of $6.1\times10^{-13}$ cm$^{-2}$s$^{-1}$, corresponding to a detector livetime of 24.1 hours. The somewhat higher upper limits presented in $\S$\ref{subsec:grs_res} utilise a more limited data set and are therefore consistent with the previously published value. None of the H.E.S.S.\ observations of GRS 1915+105 coincide with bright flaring episodes at longer wavelengths.
Observations of Cir X-1 were obtained during an extended dip in the X-ray flux, at phase intervals close to the periastron passage of the binary components. Spectral analysis of the \textit{RXTE} data showed some evidence for a recent increase in mass transfer, producing strong signatures of X-ray absorption. It was hoped that H.E.S.S.\ observations would coincide with one of the quasi-regular radio flares, which often accompany periastron passage in Cir X-1.
The ephemeris of \citet{2007ATel..985....1N} predicts the onset of a radio flare $\sim19-20$h before the first \textit{RXTE} observation. Unfortunately, despite the undoubted occurrence of quasi-periodic radio flares from Cir X-1 near periastron, a robust correlation between the observed X-ray and radio behaviour is yet to be identified. Although rapid brightening of the X-ray continuum might indicate accompanying radio flares, evidence for a definitive association is far from clear \citep{2007wmdr.confE..37S, 2008MNRAS.390..447T}. Recent radio observations of Cir X-1 \citep[e.g.][]{fenderradio, 2008MNRAS.390..447T} focus primarily on the ultrarelativistic ejection events that manifest as $\gtrsim3$ day episodes of flaring on timescales of a few hours. In principle, the lack of contemporaneous radio data admits the possibility of such persistent outbursts at the time of observation. By analogy with canonical black hole binaries, it is possible that the inferred variation in the mass accretion rate between the first and second \textit{RXTE} observations also implies an evolution of the jet properties \citep{2006MNRAS.366...79M}, but this is far from clear in such an unusual system. Moreover, \citet{2008MNRAS.390..447T} report compelling evidence that prior to 2006, Cir X-1 underwent a $\sim6$ year episode of unusual radio quiescence, suggesting that jet formation was somewhat suppressed during the epochs of H.E.S.S.\ observation.
Accordingly, without strictly simultaneous radio data indicating otherwise, the most likely scenario is that no outflows were present. In this context the absence of a detectable $\gamma$-ray\ signal is not surprising.
As a confirmed high-mass black hole binary, V4641 Sgr is the studied target that most closely resembles the Cygnus X-1 and Cygnus X-3 systems. Moreover, the H.E.S.S.\ observations were obtained during a period of sporadic broadband flaring, and comparing these observations with the results of \cite{2007ApJ...665L..51A}, VHE $\gamma$-ray\ emission might have been expected. The detection of Cyg X-1 using the MAGIC telescopes appeared to coincide with the rising part of a strong X-ray flare. In contrast, radio spectra obtained close to the H.E.S.S.\ observational epochs are indicative of the decay following a flaring episode \citep{2004ATel..302....1S}.
Assuming that the $\gamma$-ray\ emission mechanisms operating in Cyg X-1 also occur in V4641 Sgr, the absence of a significant H.E.S.S.\ detection might be viewed as evidence that production of GeV and TeV photons is a highly transient process. This would further suggest that $\gamma$-ray\ emission originates in a spatially compact region that is at most a few light hours in size.
Absorption of $\gamma$-rays by pair production is expected to be negligible in GRS 1915+105, since the donor star is too cool and faint to produce a strong ultraviolet photon field. The same is true of Cir X-1 if the conventional assumption of a low-mass companion is adopted.
For completeness, Figure \ref{fig:mqs:abs} plots the level of $\gamma$-ray\ absorption predicted by a numerical implementation of the model presented by \cite{2006A&A...451....9D}, assuming that the companion star in Cir X-1 is in fact a mid-B supergiant as proposed by \cite{2007MNRAS.374..999J}. The separate curves are representative of the three orbital phase intervals corresponding to the H.E.S.S.\ observation epochs, and were derived using the system parameters derived by \cite{2007MNRAS.374..999J} in conjunction with typical values for the temperature ($T_{\rm eff}\approx20000$ K) and radius ($R\approx9$ R$_{\odot}$) of a mid-B supergiant. It is evident that some non-negligible absorption is expected, particularly during the first observation interval. Nonetheless, it seems unlikely that the expected levels of attenuation ($\lesssim20\%$) would suppress an otherwise detectable $\gamma$-ray\ flux sufficiently to yield the low significances listed in Table \ref{tab:mqs:hess_sigmas}.
The situation with regard to $\gamma$-ray\ absorption is clearer in the case of V4641 Sgr, since the companion has been spectroscopically identified as a late B- or early A-type star. Using the system parameters derived by \cite{2001ApJ...555..489O} and assuming a circularised orbit, the model presented by \cite{2006A&A...451....9D} was used to predict the expected levels of $\gamma$-ray\ absorption as a function of orbital phase. As illustrated in Figure \ref{fig:mqs:abs} (bottom panel), absorption might have an important effect during the first H.E.S.S.\ observation interval, although as with Cir X-1 the predicted levels of absorption ($\lesssim25\%$) would not attenuate a bright $\gamma$-ray\ signal so far below the detection threshold. During the second H.E.S.S.\ observation interval, when X-ray data show marginal indications of source activity, the predicted absorption due to pair production on the stellar radiation field is negligible.
We note however that, as in the case of Cir X-1, the relative inclination of the jets from V4641 Sgr with respect to the accretion disk may be low \citep{2006ESASP.604..201S} and, therefore, further absorption of $\sim100$~GeV -- TeV $\gamma$-ray photons could occur via interaction with the disk thermal photon field (see e.g. \citealp{1992A&A...264..127C}).
It should also be noted that all the confirmed VHE $\gamma$-ray\ binaries lie at distances of $2-4$ kpc. In contrast, the targets reported here have maximum distances in the range $7-11$ kpc, resulting in flux dilution factors that are greater by one order of mangnitude. Obviously, this has strong implications for the detectability of any emitted $\gamma$-ray\ signal.
\begin{figure}[h!]
\begin{center}
\includegraphics[width=0.53\textwidth]{cir_abs_spec-eps-converted-to.pdf}
\includegraphics[width=0.53\textwidth]{v4s_abs_lc-eps-converted-to.pdf}
\caption{Levels of $\gamma$-ray\ absorption due to pair production with stellar photons as predicted by the numerical model outlined by \cite{2006A&A...451....9D}. \textit{Top panel:} expected $\gamma$-ray\ transmission as a function of photon energy for Cir X-1 assuming an inclination $i=66^{\circ}$ and using the best-fit ephemeris derived by \cite{2007MNRAS.374..999J}, which is appropriate for a mid-B supergiant companion. The individual curves correspond to the orbital phases of the first ({blue dashed}), second (red dot-dashed), and third (black solid) H.E.S.S.\ observation intervals. \textit{Bottom panel:} expected $\gamma$-ray\ transmission for V4641 Sgr as a function of orbital phase derived using the orbital solution of \citep{2001ApJ...555..489O} and assuming a circularised orbit. The individual curves represent photon energies of 10 GeV ({black solid}), 1 TeV ({blue dashed}), and 10 TeV ({red dot-dashed}). Vertical lines indicate the first ({dot-dashed}) and second ({dashed}) H.E.S.S.\ observation epochs.}
\label{fig:mqs:abs}
\end{center}
\end{figure}
\section{Conclusions}\label{sec:conclusion}
Contemporaneous VHE $\gamma$-ray\ and X-ray observations of GRS 1915+105, Cir X-1, and V4641 Sgr were obtained using H.E.S.S.\ and \textit{RXTE}. Analysis of the resultant H.E.S.S.\ data did not yield a significant detection for any of the target microquasars. However, X-ray binaries are dynamic systems and as such are likely to exhibit evolution of their radiative properties, both as a function of orbital phase and also in response to non-deterministic properties. It follows that the non-detections presented in this work do not indicate that the target binary systems do not emit detectable VHE $\gamma$-ray\ emission at phases other than those corresponding to the H.E.S.S.\ observations.
GRS 1915+105 appears to have been observed during an extended plateau state, the archival multiwavelength data suggesting the presence of continuous, mildly relativistic radio jets at the time of observation. The \textit{RXTE} observations of Cir X-1 yield data that are consistent with strongly varying obscuration of the X-ray source shortly after periastron passage, but these data are not indicative of bright flaring during the H.E.S.S.\ observation epochs. Conversely, V4641 Sgr appears to have been observed during an episode of mild, transient flaring, although rapid source variability, combined with the limited duration of the strictly simultaneous H.E.S.S.\ and \textit{RXTE} exposure, complicates interpretation.
Microquasars continue to be classified as targets of opportunity for IACTs, requiring a rapid response to any external trigger to maximise the likelihood of obtaining a significant detection. These conditions are realised with the commissioning of the H.E.S.S. 28 m telescope, which aims to lower the energy threshold from 100 GeV to about 30 GeV (\cite{Parsons2015}; \cite{Holler2015}; \cite{HollerCrab2015}) while simultaneously enabling very rapid follow-up observations \citep{Hofverberg2013a}. To exploit these new opportunities and an increasing understanding of the behaviour of microquasars, the triggering strategies for TeV follow-up observations have evolved significantly in recent years. In the future, alternative observational strategies, including continuous monitoring of candidate microquasars in the VHE $\gamma$-ray\ band, may become possible using dedicated sub-arrays of the forthcoming Cherenkov Telescope Array \citep{2010arXiv1008.3703C}.
Irrespective of the non-detections presented herein, the tantalising observations of Cygnus X-3 at GeV energies and Cygnus X-1 by the MAGIC telescope ensures that the motivations for observing microquasars using IACTs remain compelling. Indeed, by further constraining the $\gamma$-ray\ emission properties of microquasars, subsequent observations will inevitably yield an enhanced understanding of astrophysical jet production on all physical scales. More optimistically, the detection of additional $\gamma$-ray -bright microquasars would greatly facilitate a comprehensive characterisation of the particle acceleration and radiative emission mechanisms that operate in such systems.
\begin{acknowledgements}
The support of the Namibian authorities and of the University of Namibia
in facilitating the construction and operation of H.E.S.S. is gratefully
acknowledged, as is the support by the German Ministry for Education and
Research (BMBF), the Max Planck Society, the German Research Foundation (DFG),
the French Ministry for Research,
the CNRS-IN2P3 and the Astroparticle Interdisciplinary Programme of the
CNRS, the U.K. Science and Technology Facilities Council (STFC),
the IPNP of the Charles University, the Czech Science Foundation, the Polish
Ministry of Science and Higher Education, the South African Department of
Science and Technology and National Research Foundation, and by the
University of Namibia. We appreciate the excellent work of the technical
support staff in Berlin, Durham, Hamburg, Heidelberg, Palaiseau, Paris,
Saclay, and in Namibia in the construction and operation of the
equipment.\end{acknowledgements}
\bibliographystyle{aa}
| 6766c44b5014386b2d928064431606c5b530db62 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
Human Activity Recognition (HAR) is the understanding of human behaviour from data captured by pervasive sensors, such as cameras or wearable devices. It is a powerful tool in medical application areas, where consistent and continuous patient monitoring can be insightful. Wearable devices provide an unobtrusive platform for such monitoring, and due to their increasing market penetration, feel intrinsic to the user. This daily integration into a user's life is crucial for increasing the understanding of overall human health and wellbeing. This is referred to as the ``quantified self" movement.
Wearables, such as actigraph accelerometers, generate a continuous time series of a person's daily physical exertion and rest. This ubiquitous monitoring presents substantial amounts of data, which can ({\em i})~ provide new insights by enriching the feature set in health studies, and ({\em ii})~ enhance the personalisation and effectiveness of health, wellness, and fitness applications. By decomposing an accelerometer's time series into distinctive activity modes or actions, a comprehensive understanding of an individual's daily physical activity can be inferred. The advantages of longitudinal data are however complemented by the potential of noise in data collection from an uncontrolled environment. Therefore, the data sensitivity calls for robust automated evaluation procedures.
In this paper, we present a robust automated human activity recognition (RAHAR) algorithm. We test our algorithm in the application area of sleep science by providing a novel framework for evaluating sleep quality and examining the correlation between the aforementioned and an individual's physical activity. Even though we evaluate the performance of the proposed HAR algorithm on sleep analysis, RAHAR can be employed in other research areas such as obesity, diabetes, and cardiac diseases.
\section{Related Work}
Human activity recognition (HAR) has been an active research area in computer vision and machine learning for many years. A variety of approaches have been investigated to accomplish HAR ranging from analysis of still images and videos to motion capture and inertial sensor data.
Video has been the most widely studied data source in HAR literature. Hence, there exists a wealth of papers in this particular domain. The most recent literature on HAR from videos include trajectory-based descriptors~\cite{ HWang:IJCV15, BFernando:TPAMI16, IAtmosukarto:WACV15}, spatio-temporal feature representations~\cite{ SMa:CVPR15, ZZhou:TMM15, DTran:ICCV15}, feature encoding~\cite{ VKantorov:CVPR14, XPeng:ECCV14, HKuehne:WACV16}, and deep learning~\cite{ JDonahue:CVPR15, LWang:CVPR15, LSun:ICCV15}. Reviewing the extensive list of video-based HAR studies, however, goes beyond the scope of this study and we refer the reader to~\cite{ SKe:Computers13, PBorges:TCSVT13} for a collection of more comprehensive surveys on the topic.
Unlike HAR from video, existing approaches for HAR from still images are somewhat limited, and range from histogram-based representations~\cite{NIkizler:ICPR08, CThurau:CVPR08} and color descriptors~\cite{FKhan:IJCV13} to pose-, appearance- and parts-based representations~\cite{WYang:CVPR10, SMaji:CVPR11, BYao:ICCV11, GSharma:TPAMI16}. Guo and Lai recently provided a comprehensive survey of the studies on still image-based HAR in~\cite{GGuo:PatRec14}.
Several techniques have been proposed, on the other hand, for HAR from 3D data, encompassing representations based on bag-of-words~\cite{ WLi:CVPRW10, LXia:CVPRW12}, eigen-joints~\cite{ XYang:CVPRW12}, sequence of most informative joints~\cite{ FOfli:JVCI14}, linear dynamical systems~\cite{ RChaudhry:CVPRW13}, actionlets~\cite{ JWang:CVPR12}, Lie algebra embedding~\cite{ RVemulapalli:CVPR14}, covariance descriptors~\cite{ MHussein:IJCAI13}, hidden Markov models~\cite{ FLv:ECCV06}, subspace view-invariant metrics~\cite{ YSheikh:ICCV05} and occupancy patterns~\cite{ JWang:ECCV12, AVieira:CIARP12}. Aggarwal and Xia presented a recent survey summarizing state-of-the-art techniques in HAR from 3D data~\cite{JAggarwal:PatRec14}.
Unlike vision-based HAR systems, sensor-based HAR technologies commonly deal with time series of state changes and/or various parameter values collected from a wide range of sensors such as contact sensors, accelerometers, audio and motion detectors, etc. Chen et al.~\cite{LChen:TCMCC12} and Bulling et al.~\cite{ABulling:CSUR14} present comprehensive reviews of sensor-based activity recognition literature. The most recent work in this domain includes knowledge-based inference~\cite{ACalzada:EMBC14, DBiswas:HUMOV15}, ensemble methods~\cite{ATripathi:EAIS15, CCatal:ASOC15}, data-driven approaches~\cite{RAkhavian:WSC15, LLiu:KNOSYS15}, and ontology-based techniques~\cite{GOkeyo:PMCJ14}.
All of the aforementioned studies investigate recognition/classification of fully observed action or activity, e.g., jumping, walking, running, drinking, etc. (i.e., activities of daily living), using well-curated datasets. However, thanks to the ``quantified self" movement, myriad of consumer-grade wearable devices have become available for individuals who have started monitoring their physical activity on a continuous basis, generating tremendous amount of data. Therefore, there is an urgent need for automatic analysis of data coming from fitness trackers to assess the physical activity levels and patterns of individuals for the ultimate goal of quantifying their overall wellbeing. This task requires understanding of longitudinal, noisy physical activity data at a rather higher (coarser) level than specific action/activity recognition level. Main challenges as well as opportunities of HAR from personalized data and lifelogs have been discussed in several dimensions in~\cite{BDobkin:CurrOpinNeurol13, OLara:SURV13, JBort-Roig:Sports14, MRehman:Sensors15, SMukhopadhyay:JSEN15}.
There has been a number of initiatives to overcome the challenge of collecting annotated personalized data to further research on HAR from continuous measurement of real-world physical activities~\cite{MZhou:ICMR13, CMeurisch:UbiComp15}. Even though such systems exhibit a crucial attempt in furthering research in mining personalized data, they have limited practical importance as they rely on manual annotation of the acquired data. There has also been recent attempts to automatically recognize human activities from continuous personalized data~\cite{JHamm:MobiCASE13, CDobbins:CIT15, MUddin:WearSys15, OBanos:EMBC15}. However, most of these studies are designed to recognize only a predefined set of activities, and hence, not comprehensive and robust enough to quantify the physical activity levels for the overall assessment of individuals' wellbeing.
\section{Background}
Sleep pattern evaluation is a paragon of cumbersome testing and requires extensive manual evaluation and interpretation by clinical experts. Unhealthy sleep habits can impede physical, mental and emotional wellbeing, and lead to exacerbated health consequences~\cite{strine2005associations}. Since patient referral to sleep specialists is often based on self-reported abnormalities, exacerbation often precedes diagnosis.
Clinical diagnosis of complex sleep disorders involves a variety of tests, including an overnight lab stay with oxygen and brain wave monitoring (polysomnography and electroencephalogram, respectively), and a daily sleep history log with a subjective questionnaire. The daily sleep logs and questionnaires are often found to be unreliable and inconsistent with actual observed activity. This is especially true in adolescents~\cite{arora2013investigation}. The overnight stay allows specialists to manually monitor the patient's sleep period. This requires the active involvement of a clinical sleep specialist. Furthermore, the monitoring is only for one night and in a clinical setting, rather than the patient's own home. Using wearable devices provides both a context-aware and longitudinal monitoring.
The inconvenience and inaccuracy of daily logs, coupled with the invasiveness of an overnight lab stay, substantiate the need and adoption of wearable devices for first pass diagnostic screening. More generally, using our HAR approach with a wearable device empowers users to self-monitor their sleep patterns, and reform their activity habits for optimised sleep and an improved quality of life.
\section{Preliminaries}
In this section we present a description of the dataset and the context-aware definitions used for our application area.
\subsection{Data}
Data was collected as part of a research study to examine the impact of sleep on health and performance in adolescents by Weil COrnell Medical COllege - Qatar. Two international high schools were selected for cohort development. Student volunteers were provided with an actigraph accelerometer, ActiGraph GT3X+\footnote{http://actigraphcorp.com/support/activity-monitors/gt3xplus/}, to wear on their non-dominant wrist, continuously throughout the study (i.e. even when sleeping). Deidentified data collected in the study were used in the current analysis.
The ActiGraph GT3X+ is a clinical-grade wearable device that has been previously validated against clinical polysomnography~\cite{PFreedson:MedSciSports98}. The device samples the user's sleep-wake activity at 30-100 Hertz. Currently sleep experts use this device in conjunction with the accompanying software, ActiLife\footnote{http://actigraphcorp.com/products-showcase/software/actilife/}, to evaluate an individual's sleep period. We evaluate our results side-by-side with ActiLife's results.
\subsection{Definitions}
\begin{figure*}
\caption{Sleep science definitions on an example accelerometer data extract}
\centering
\includegraphics[scale=0.25]{ActivityExample.jpg}
\label{fig:sleep_example}
\end{figure*}
\begin{table*}
\centering
\caption{Relevant sleep science equations~\cite{CIber:AASM07}}
\centering
\begin{tabular}{c|c}
\hline
Sleep Period & $ \big[ \text{Sleep Onset Time}, \text{Sleep Awakening Time} \big] $ \T\B \\
\hline
Sleep Period Duration & $ \left \| \text{Sleep Awakening Time} - \text{Sleep Onset Time} \right \| $ \T\B \\
\hline
Wake After Sleep Onset (WASO) & $ \sum_{\text{n}=\text{onset}}^{\text{awake}} \left \| \text{Wakefulness} \right \| $ \T\B \\
\hline
Latency & $ \big[ \text{Preceding Sedentary Time}, \text{Sleep Onset Time} \big] $ \T\B \\
\hline
Total Minutes in Bed & $ \left \| \text{Sleep Awakening Time} - \text{Preceding Sedentary Time} \right \| $ \T\B \\
\hline
Total Sleep Time & $ \left \| \text{Sleep Period Duration} - \text{WASO} - \text{Latency} \right \| $ \T\B \\
\hline
Sleep Efficiency & $ \text{Total Sleep Time} / \text{Total Minutes in Bed} $ \T\B \\
\hline
\end{tabular}
\label{tab:sleep_eqs}
\end{table*}
To apply our methodology to the area of sleep science, it is important to note the definitions mentioned in this section. In traditional sleep study literature, a sleep period is bounded between the sleep-onset-time and sleep-awakening-time~\cite{CIber:AASM07}. Experts characterise the sleep-onset-time as the first minute after a self-reported bedtime, that is followed by 15 minutes of continuous sleep~\cite{sadeh2000sleep}. We propose a modified definition, that allows for automatic evaluation and deems sleep diaries unnecessary. As a result, we can infer the ``bedtime" of an individual in reverse, based on their sedentary activity before the onset of sleep. Epoch records that contain no triaxial movement, 0 steps taken, and an inclinometer output of not lying down, are candidate sleep records, and are further tested for whether they are a component of the sleep period. We define sleep-onset-time as the first candidate epoch record in a series of 15 continuous candidate sleep minutes. Likewise, the sleep-awakening-time is defined as the last epoch record in a series of 15 continuous candidate sleep minutes, that is followed by 30 continuous non-candidate sleep minutes, (i.e. 30 minutes of active awake time). The sleep period duration can be computed as the time passed between sleep onset and sleep awakening.
Within the sleep period, there are periods of unrest or wakefulness. For example, when a user re-adjusts positions, or uses the bathroom. If the duration of movement exceeds 5 consecutive minutes of activity, it is marked as a time of ``wakefulness." The total sum of all moments of wakefulness is referred to as wake-after-sleep-onset, also known as WASO.
Immediately preceding the start of the sleep onset, is the time-in-bed, which quantifies the sedentary time an individual spends before they have fallen asleep. This sedentary time can be observed in the actigraph accelerometer data. The time that the preceding sedentary activity begins until the time of the sleep onset is called the sleep latency.
From the aforementioned values, total sleep time and an overall sleep efficiency score can be deduced. Total sleep time covers the defined sleep period, less the wake after sleep onset time and less the latency. Lastly, sleep efficiency is the ratio of total sleep time to total minutes in bed. All of the above definitions are summarised in Table~\ref{tab:sleep_eqs}, and visualised in Fig.~\ref{fig:sleep_example}. In this study, we use sleep efficiency as the metric to measure sleep quality~\cite{JCacioppo:PsycSci02} among other metrics such as latency, wake after sleep onset, awakening index, total sleep time, etc.~\cite{SScholle:SleepMedicine11}.
\section{Methodology}
Our methodology for RAHAR is shown algorithmically in Fig.~\ref{algo:rahar}. We elaborate on the details of our algorithm in the sequel.
\begin{figure}
\begin{algorithmic}[1]
\STATE{\textbf{input:} Raw accelerometer data}
\STATE{\textbf{output:} Time-series segments with activity intensity level annotations}
\FORALL{segment (daily or otherwise)}
\FORALL{epoch (minutes, hour, etc.)}
\STATE{implement activity cut points}
\ENDFOR
\STATE{change points $\gets$ implement hierarchical divisive estimation}
\STATE{change point intervals $\gets$ divide time series by change points}
\ENDFOR
\FORALL{change point interval}
\STATE{activity mode $\gets$ statistical mode of cut points}
\ENDFOR
\end{algorithmic}
\caption{Algorithm for Robust Automated Human Activity Recognition (RAHAR).}
\label{algo:rahar}
\end{figure}
\begin{figure*}
\centering
\caption{Classification labelling of each change point interval during an example awake time}
\includegraphics[width=1.0\linewidth]{RAHAR.png}
\label{fig:change_point_labeling}
\end{figure*}
\subsection{Pre-Processing}
The accelerometer of choice, Actigraph GT3X+, sampled each person's activity at 30-100 Hertz. The stored data included the triaxial accelerometer coordinates as well as a computed epoch step count based on the vertical axis, and post-processed inclinometer orientation. This raw data was downloaded and aggregated to a minute-by-minute granularity. An epoch of one minute was selected in order to optimise the interpretability of the physical activity~\cite{KGabriel:IJBNPA10}, as well as for implementing the state-of-the-art cut point methodology~\cite{troiano2008physical}. In other contexts, a different granularity may be sufficient.
\subsection{Automated Annotation and Segmentation}
Due to the context of sleep disorders, sleep periods needed to be annotated within the raw ActiGraph output. Candidate sleep records, epochs with no triaxial movement, 0 steps taken, and an inclinometer output of not lying down, were identified in the time series and tested to find the sleep onset time, and sleep awakening time. The details of this terminology is elaborated in the preliminaries section. All test instances that fell within these two boundary times, were annotated as ``Sleep," and constituted the sleep period.
Whilst analysing the data, we found that several participants had multiple sleep periods in a day, implying that they took daily naps or followed a polyphasic, or biphasic, sleep pattern. Upon closer analysis of the length and time of the sleep period, no discernible patterns were visible. Thus we opted to segment the time series by the end of a sleep period rather than the traditional approach of segmenting by day. Each sleep period was linked to its preceding activity, extending until the previous sleep period. We refer to these segments as sleep-wake segments. The result of this decision is that the activity immediately before each sleep period is used for the correlation analysis for its subsequent sleep period, rather than the total for that day.
\begin{figure*}
\centering
\caption{ROC curves for sleep efficiency}
\begin{subfigure}[b]{0.49\linewidth}
\centering
\caption{RAHAR}
\includegraphics[width=\linewidth]{Ours_ROC.png}
\label{fig:roc_RAHAR}
\end{subfigure}
\begin{subfigure}[b]{0.49\linewidth}
\centering
\caption{Sleep Expert + ActiLife}
\includegraphics[width=\linewidth]{SleepExpert_ROC.png}
\label{fig:roc_SEAL}
\end{subfigure}
\label{fig:roc_curve}
\end{figure*}
\begin{table*}
\centering
\caption{Sleep efficiency results}
\begin{tabular}{c|cc|cc|cc|cc|cc}
\hline
&\multicolumn{2}{c}{AU-ROC} & \multicolumn{2}{c}{F1 Score}& \multicolumn{2}{c}{Recall}& \multicolumn{2}{c}{Precision}& \multicolumn{2}{c}{Accuracy}
\\
\hline
& SE+AL & RAHAR & SE+AL & RAHAR & SE+AL & RAHAR & SE+AL & RAHAR & SE+AL & RAHAR
\\
\hline
\hline
Ada & 0.7489 & 0.8132 & 0.5574 & 0.6885 & 0.5484 & 0.5526 & 0.5667 & 0.9130 & 0.6966 & 0.7206\\
RF & 0.8115 & 0.8746 & 0.6885 & 0.7500 & 0.6774 & 0.6316 & 0.7000 & 0.9231 & 0.7865 & 0.7647\\
SVM & 0.7497 & 0.7895 & 0.3721 & 0.7077 & 0.2581 & 0.6053 & 0.6667 & 0.8519 & 0.6966 & 0.7206\\
LogR & 0.5884 & 0.8649 & - & 0.6875 & - & 0.5789 & - & 0.8462 & - & 0.7059\\
\hline
Average & 0.7246 & 0.8355 & 0.5393* & 0.7154* & 0.4946* & 0.5965* & 0.6445* & 0.8960* & 0.7266* & 0.7353*\\
\hline
\multicolumn{11}{@{}l}{{\scriptsize * logistic regression (LogR) score is not included in averaging.}} \\
\end{tabular}
\label{tab:slp_eff_res}
\end{table*}
\subsection{Activity Mode Detection}
The actigraph accelerometer data contains post-filtered ``counts" for each of the axes. These counts quantify the frequency and intensity of the user's activity\footnote{http://actigraphcorp.com/wp-content/uploads/2015/06/ActiGraph-White-Paper\_What-is-a-Count\_.pdf}. Using Troiano's cut point scale~\cite{troiano2008physical}, the age of a user, and their accelerometer triaxial count, each epoch is labeled with an intensity level: Sedentary, Light, Moderate, or Vigorous. Since each epoch is 1 minute in length, this provides an unnecessary granularity to an individual's activity levels and is highly subject to noise. We ``smooth" the activity intensity levels over activity modes using change point detection.
Once the time series is segmented into sleep-wake segments, we identify the distinctive activity modes using the multiple change point detection algorithm, hierarchical divisive estimation~\cite{NJames:JSS15}. We tested the change points to a statistical significance level of 0.01 and used a maximum number of random permutations of 99. Each change point result is treated as the interval boundaries for distinctive activity modes.
Each sleep-wake segment now consists of a series of change point intervals. The activity intensity classification label for each change point interval is computed by taking the statistical mode of the minute-by-minute labels over every epoch existing within the interval. Fig.~\ref{fig:change_point_labeling} illustrates the classification labelling of an individual's awake time.
\subsection{Modeling}
In sleep science, sleep quality is defined by a number of metrics, including total sleep time, wake after sleep onset, awakening index, and sleep efficiency~\cite{SScholle:SleepMedicine11}. In our analysis, we focus on sleep efficiency metric for our experiments~\cite{JCacioppo:PsycSci02}. Sleep efficiency is computed as a numerical value ranging from 0 to 1. According to specialists, a sleep efficiency below 0.85 (i.e., 85\%) indicates poor sleep quality. Thus, each sleep period can be classified as having ``good sleep efficiency" or ``poor sleep efficiency"~\cite{williams1974electroencephalography}.
To model the effect of daily physical activity on sleep, the duration of each intensity level of activity was aggregated over the sleep-wake segment. The percentage of awake time in each mode was used as the model input.
\section{Experiments and Results}
\begin{figure*}
\centering
\caption{Comparison of the performance of random forest model for each approach}
\begin{subfigure}[b]{0.49\linewidth}
\centering
\caption{ROC}
\includegraphics[width=\linewidth]{RandomForest_comp.png}
\label{fig:rf_roc_comp}
\end{subfigure}
\begin{subfigure}[b]{0.49\linewidth}
\centering
\caption{Sensitivity-Specificity}
\includegraphics[width=\linewidth]{SensSpec_comp.png}
\label{fig:rf_ss_comp}
\end{subfigure}
\label{fig:rf_comp}
\end{figure*}
\begin{table*}
\centering
\caption{Sleep efficiency -- sensitivity and specificity}
\begin{tabular}{c|cc|cc|cc|cc}
\hline
&\multicolumn{2}{c}{AU-ROC} & \multicolumn{2}{c}{F1 Score}& \multicolumn{2}{c}{Sensitivity}& \multicolumn{2}{c}{Specificity}
\\
\hline
& SE+AL & RAHAR & SE+AL & RAHAR & SE+AL & RAHAR & SE+AL & RAHAR
\\
\hline
\hline
Ada & 0.7489 & 0.8132 & 0.5574 & 0.6885 & 0.5484 & 0.5526 & 0.7759 & 0.9333\\
RF & 0.8115 & 0.8746 & 0.6885 & 0.7500 & 0.6774 & 0.6316 & 0.8448 & 0.9333\\
SVM & 0.7497 & 0.7895 & 0.3721 & 0.7077 & 0.2581 & 0.6053 & 0.9310 & 0.8667\\
LogR & 0.5884 & 0.8649 & - & 0.6875 & - & 0.5789 & - & 0.8667\\
\hline
Average & 0.7246 & 0.8355 & 0.5393* & 0.7154* & 0.4946* & 0.5965* & 0.8505* & 0.9111*\\
\hline
\multicolumn{9}{@{}l}{{\scriptsize * logistic regression (LogR) score is not included in averaging.}} \\
\end{tabular}
\label{tab:sens_spec}
\end{table*}
RAHAR is fundamentally a feature extraction algorithm for HAR in the context of quantifying daily physical activity levels of individuals. We therefore test the quality of activity recognition by RAHAR as compared to an expert-based HAR using a tool on continuous physical activity data from a wearable sensor. Since there is \textit{no} ground truth on human activity in this context, our objective is to evaluate which HAR approach leads to better quality models for sleep research, i.e., models for predicting sleep quality, specifically, sleep efficiency.
We selected four models for evaluating the performance of RAHAR against the performance of an expert-based HAR using a tool on the described actigraphy dataset: logistic regression, support vector machines with radial basis function kernel, random forest, and adaboost.
\begin{itemize}
\item Logistic Regression (LogR): We chose this model because it is an easily interpretable binary classifier. It is also relatively robust to noise, which as explained earlier is a complication on data collected in an uncontrolled environment.\footnote{Even though we included logistic regression (LogR) in our experiments, it is important to note that LogR model failed to stratify the dataset successfully for the state-of-the-art baseline approach, and predicted all cases to be in a single class. Therefore, we excluded LogR score of RAHAR from analysis whenever corresponding LogR score of the state-of-the-art baseline approach was not available.}
\item Support Vector Machine (SVM): This model was selected because it, also, is a binary classifier. We chose a radial basis function kernel, and so it differs from logistic regression in that it does not linearly divide the data.
\item Random Forest (RF): This model was tested as an alternative because of its easy straightforward interpretation, which is particularly relevant in the healthcare or consumer domains. It also is not restricted to linearly dividing the data.
\item Adaboost (Ada): Lastly, Adaboost was tested because it is less prone to overfitting than random forest.
\end{itemize}
For comparison purposes, we use the results from a sleep specialist using Actigraph's ActiLife software as a baseline. The sleep specialist segmentation of the ActiLife results uses the preceding day's activity for each sleep period, and aggregates the activity to an epoch of an hour. ActiLife requires the sleep specialist to manually adjust the sleep period boundaries, and then automatically computes the efficiency and other sleep metrics.
Figs.~\ref{fig:roc_RAHAR} and~\ref{fig:roc_SEAL} show the ROC curves for RAHAR and the sleep expert using ActiLife software (denoted as ``SE+AL"), respectively, while Table~\ref{tab:slp_eff_res} summarises the results for both RAHAR and SE+AL. One of the most important performance measures for HAR is the area under ROC (AU-ROC). Based on AU-ROC scores, both RAHAR and SE+AL performed best with random forest model. Furthermore, SE+AL achieved an average AU-ROC of 0.7246 whereas RAHAR achieved 0.8355, a 15\% improvement of AU-ROC score on average by our algorithm as opposed to the sleep expert using ActiLife. With an AU-ROC score of 0.5884 for SE+AL approach, the logistic regression model was, however, unable to stratify the dataset, and so predicted all cases to be in a single class. We considered this to be a failure of the logistic regression model for this problem, and thus, did not include its results in our discussion whenever it was appropriate to do so. For this reason, the misleading results have also been removed from Table~\ref{tab:slp_eff_res}.
Another important performance measure for HAR is the F1 score, which is computed as the harmonic mean of precision and recall. According to Table~\ref{tab:slp_eff_res}, RAHAR performed better than SE+AL in terms of precision and recall for all models, and hence, yielded significantly higher F1 scores. Specifically, F1 score for RAHAR, on average, was 0.7154 whereas it was 0.5393 for SE+AL (excluding logistic regression in both cases), yielding a solid margin of about 0.18 points (i.e., more than 30\% improvement). On the other hand, the accuracy scores, on average, were 0.7353 for RAHAR and 0.7266 for SE+AL (again excluding logistic regression), and exhibited a relatively less significant difference still in favor of RAHAR.
\section{Discussion of Results in Medical Context}
In this section we discuss the results of the best performing model and its broader impact to the area of sleep science. As seen in Fig.~\ref{fig:roc_curve} random forest and logistic regression were the two best performing models with the RAHAR algorithm. Based on the desired threshold value of true positive rate, TPR, (i.e., sensitivity), either model could be preferred to minimize false positive rate, FPR, (i.e., 1-specificity), which is equivalent to maximizing specificity. Random forest was also the best performing model for the SE+AL approach as mentioned earlier. If we compare the ROC as well as the sensitivity-specificity plots of the best model of each approach (i.e., random forest), we see that RAHAR outperforms SE+AL almost always as illustrated in Fig.~\ref{fig:rf_comp}.
Table~\ref{tab:sens_spec}, on the other hand, summarises sensitivity and specificity scores for RAHAR and SE+AL. Average sensitivity score for SE+AL and RAHAR across all models except logistic regression were 0.4946 and 0.5965, respectively. In other words, average sensitivity score for RAHAR is 20\% higher than that of SE+AL. As for specificity, RAHAR with an average score of 0.9111 outperforms SE+AL with an average score of 0.8505, which corresponds to a 7\% improvement.
As we seek to determine in our study whether a person had a ``good quality sleep" based on his physical activity levels during awake period prior to sleep, a false positive occurs when the model predicts ``good quality sleep" when the person actually had a ``poor quality sleep." Therefore, the number of false positives needs to be kept at a minimum for a desired number of true positives. In other words, a high specificity score is sought after while keeping the sensitivity score at the desired level. As can be seen from Fig.~\ref{fig:rf_ss_comp} with this perspective in mind, for a large range of sensitivity scores, RAHAR achieved higher specificity scores almost all the time than SE+AL did. For example, RAHAR achieved a sensitivity score of 0.9 with a specificity score of 0.8 whereas SE+AL remained at a specificity score of 0.6 for the same sensitivity threshold.
In summary, RAHAR outperforms state-of-the-art procedure in sleep research in many aspects. However, its application is not limited to sleep and it can be used for understanding and treatment of other health issues such as obesity, diabetes, or cardiac diseases. Moreover, RAHAR allows for fully automated interpretation without the necessity of manual input or subjective self-reporting.
Given the current interest in deep learning, a natural question that may arise is why an approach based on feature extraction and model building has been used instead of using deep learning models directly on the raw sensor data for HAR. In medical community, the explainability of a model is of utmost important as the medical professionals are interested in learning cause-and-effect relationships and using this knowledge to support their decision making processes. In this particular case, for example, sleep experts are interested in understanding how and when certain physical activity levels effect sleep in order to make decisions to improve sleep quality of individuals accordingly. However, it is an interesting idea to explore deep learning to see what is the best model from a model accuracy perspective to understand the limits of the value of continuous monitoring of individuals' physical activity, not only from a medical perspective in particular but also from a ``quantified self" perspective in general.
\section{Conclusion}
In this paper, we presented a robust automated human activity recognition (RAHAR) algorithm for multi-modal phenomena, and evaluated its performance in the application area of sleep science. We tested the results of RAHAR against the results of a sleep expert using ActiLife for predicting sleep quality, specifically, sleep efficiency. Our model a) automated the activity recognition, and b) improved the current state-of-the-art results, on average, by ~15\% in terms of AU-ROC and ~30\% in terms of F1 scores across different models. Automating the human activity recognition puts sleep science evaluation in the hands of wearable device users. This empowers users to self-monitor their sleep-wake habits, and take action to improve the quality of their life. The improved results demonstrate the robustness of RAHAR as well as the capabilities of implementing the algorithm within clinical software such as ActiLife.
The application of RAHAR is, however, not limited to sleep science. It can be used to monitor physical activity levels and patterns of individuals with other health issues such as obesity, diabetes, and cardiac diseases. Besides, RAHAR can also be used in the general context of the ``quantified self" movement, and provide individuals actionable information about their overall fitness levels.
\bibliographystyle{IEEEtran}
| 8cff7666dfdc658a4cb29acb696f0418307f95d9 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |
\section{Introduction}
\setcounter{section}{1}\setcounter{equation}{0}
Lie conformal (super)algebras, originally introduced by Kac in \cite{K1,K2}, encode an axiomatic
description for the singular part of the operator product expansion of chiral fields in
two-dimensional conformal field theory. They are very closely related to vertex algebras (cf. \cite{B,R}) by the same way as Lie algebras correspond to
their universal enveloping algebras. On the other hand, the theory of Lie conformal (super)algebras give us powerful tools for the
study of infinite-dimensional Lie (super)algebras and associative algebras satisfying the locality property described in \cite{K}.
The conformal (super)algebras have drawn much attention in branches of physics and mathematics since the introduction.
The structure theory, representation theory and cohomology theory of finite (i.e., finitely generated as $\C[\pa]$-modules)
Lie conformal algebras have been well developed (cf. \cite{BK, CK,CKW,DK}), and finite simple Lie conformal
superalgebras were classified in \cite{FK} and their representation theories were developed in \cite{BKL1,BKL2,KO}.
The object investigated in this paper is a Lie conformal superalgebra closed related to the loop super-Virasoro algebra $\mathfrak{sl}$ whose structures were studied in \cite{DHS}. It is defined as
a infinite-dimensional Lie superalgebra with a basis $\{L_{\al,i}, G_{\mu,j}\mid \al,i,j\in\Z, \mu\in \frac12+\Z\}$
satisfying the following commutation relations:
\begin{eqnarray}\label{p2.1}
[L_{\al,i},L_{\beta,j}]=(\al-\beta)L_{\al+\beta,i+j},\
[L_{\al,i},G_{\mu,j}]=(\frac{\al}{2}-\mu)G_{\al+\mu,i+j},\
[G_{\mu,i},G_{\nu,j}]=2L_{\mu+\nu,i+j},
\end{eqnarray} the even and odd parts of which are $\mathfrak{sl}^{\bar{0}}={\rm span}\{L_{\al,i}\mid\al,i\in\Z\}$ and
$\mathfrak{sl}^{\bar{1}}={\rm span}\{G_{\mu,j}\mid j\in\Z,\mu\in \frac12+\Z\}$, respectively.
Clearly, $\mathfrak{sl}^{\bar{0}}$ is just the loop Virasoro algebra (cf. \cite{GLZ}) and $ \mathfrak{sl}^{\bar{1}}$ is its module. Hence, the loop super-Virasoro algebra can be seen as a super extension of the loop Virasoro algebra.
Motivated by the idea from \cite{K2} we associate a Lie conformal superalgebra with the loop super-Virasoro algebra.
It is called the \textit{loop super-Virasoro conformal superalgebra}, denoted by $\mathfrak{cls}$, which is a $\C[\pa]$-module
$\mathfrak{cls}=(\bigoplus_{i\in\Z}\C[\pa]L_i)\oplus(\bigoplus_{i\in\Z}\C[\pa]G_i)$ with a $\C[\partial]$-basis $\{L_i,G_i\mid i\in\Z\}$ satisfying the following $\lambda$-brackets:
\begin{eqnarray}
&&[L_{i\ \lambda} L_j]=(\partial+2\lambda) L_{i+j},\
[L_{i\ \lambda} G_j]=(\partial+\frac32\lambda) G_{i+j},\label{d1.2}\\ \
&&[G_{i\ \lambda} L_j]=(\frac12\partial+\frac32\lambda) G_{i+j}, \
[G_{i\ \lambda} G_j]=2L_{i+j}, \forall i,j\in\Z. \label{d1.3}
\end{eqnarray} Note that this is an infinite simple Lie conformal superalgebra, containing the loop Virasoro Lie conformal algebra
$\mathfrak{clv}=\bigoplus_{i\in\Z}\C[\pa]L_i$ (see \cite{WCY}) as its subalgebra.
As pointed out previously, the theory of finite simple Lie conformal (super)algebras were well developed, but so far there is no systematic theory for the infinite case. So it is interesting and necessary to develop the theory for infinite simple Lie conformal superalgebras. This is one of our motivations for studying the loop super-Virasoro conformal superalgebra. We shall study the superderivation algebra of $\mathfrak{cls}$ and free ($\Z$-graded) $\mathfrak{cls}$-modules of rank $\leq2$. One interesting aspect is that free ($\Z$-graded) $\mathfrak{cls}$-modules of rank 1 are trivial, which is totally different from the loop Virasoro Lie conformal algebra case (all $V_{a,b}$ and $V_{A,b}$ are its nontrivial conformal modules of rank one); and the other interesting aspect is that the even or odd part of a $\Z$-grade free $\mathfrak{cls}$-module of rank two has the form $V_{A,b}$ if and only if the other part is $A_{\frac12,b}$. We remark that an important strategy frequently used in the present paper is to pass modules over $\mathfrak{cls}$ to modules over $\mathfrak{clv}$.
This paper is organized as follows. In Sect. 2, we collect some facts and notions related
to Lie conformal superalgebras. In Sect. 3, we determinate conformal superderivations of $\mathfrak{cls}$.
The Sect. 4 is devoted to giving the classification of all nontrivial free $\mathfrak{cls}$-modules of rank less than two. We also determine the irreducibility of these modules and therefore classify all inequivalent irreducible free $\mathfrak{cls}$-modules of rank two.
And free $\Z$-graded $\mathfrak{cls}$-modules of rank less than or equal to two are classified in Sect. 5. Moreover, all irreducible submodules of free $\Z$-graded $\mathfrak{cls}$-modules of rank two are completely determined.
Throughout this paper, all vector spaces are assumed to be over the complex field $\C$ and all linear maps are $\C$-linear. The main results of this paper are summarized in Theorems \ref{main3}, \ref{freeofranktwo}, \ref{main4-2} and \ref{main5}.
\section{Preliminaries}
\setcounter{section}{2}\setcounter{equation}{0}
In this section, we recall some notions related to Lie conformal superalgebras and conformal modules from\cite{DK,K1,K2}.
We say that a vector space $U$ is $\Z_2$-graded if $U=U^{\bar0}\oplus U^{\bar1}$, and $x\in U^{\bar i}$ is called $\Z_2$-homogenous and write $|x|= i$. For any two $\Z_2$-graded vector spaces $U$ and $V$, a linear map $f: U\rightarrow V$ is called homogenous of degree $\bar i\in\Z_2$ if $f(U^{\bar j})\subseteq V^{\overline{i+j}}$ for all $\bar j\in\Z_2$.
\begin{defi}
A Lie conformal superalgebra is a $\Z_2$-graded $\C[\partial]$-module $\mathcal{A}$ endowed with a linear map
$\mathcal{A}\otimes\mathcal{A}\rightarrow \C[\lambda]\otimes \mathcal{A}, a\otimes b\mapsto [a_\lambda b]$, called the $\lambda$-bracket, and
satisfying the following axioms $(a, b, c\in \mathcal{A}):$
\begin{eqnarray}
&&[(\partial a)_\lambda b]=-\lambda[a_\lambda b],\quad [a_\lambda b]=-(-1)^{|a||b|}[b_{-\lambda-\partial} a],\label{m2.2}\\
&&[a_\lambda [b_\mu c]]=[[a_\lambda b]_{\lambda+\mu}c]+(-1)^{|a||b|}[b_\mu [a_\lambda c]].
\end{eqnarray}
\end{defi}
It follows from the axioms in (\ref{m2.2}) that
\begin{eqnarray*}
[(f(\partial)a)_\lambda b]=f(-\lambda)[a_\lambda b]\ \mbox{and}\
[a_\lambda (f(\partial)b)]=f(\partial+\lambda)[a_\lambda b],\quad \forall f(\pa)\in\C[\pa].
\end{eqnarray*}
\begin{defi}
A conformal module over a Lie conformal superalgebra $\mathcal{A}$ or an $\mathcal A$-module is a $\Z_2$-graded $\C[\partial]$-module $V$ endowed with a $\lambda$-action
$\mathcal{A}\otimes V\rightarrow \C[\lambda]\otimes V, a\otimes v\mapsto a_\lambda v$, and satisfying the following axioms $(a, b \in \mathcal{A}, v\in V):$
\begin{eqnarray*}&&(\partial a)_\lambda v=-\lambda a_\lambda v,\quad a_\lambda (\partial v)=(\partial +\lambda)a_\lambda v,\\
&&[a_\lambda b]_{\lambda+\mu}v=a_\lambda (b_\mu v)-(-1)^{|a||b|}b_\mu (a_\lambda v).
\end{eqnarray*}
\end{defi}
The rank of an $\mathcal A$-module $V$ is defined to be the rank of $V$ as $\C[\partial]$-module.
\begin{defi}\label{intermod}Let $\mathcal A$ be a Lie conformal superalgebra.
\begin{itemize}\parskip-3pt\item[{\rm (1)}] $\mathcal{A}$ is called {\it $\Z$-graded} if $\mathcal{A}=\oplus_{i\in \Z}{\mathcal{A}}^i$, each ${\mathcal{A}}^i$ is a $\C[\partial]$-submodule
and
$[{\mathcal{A}}^i\,{}_\lambda\, {\mathcal{A}}^j]\subseteq \mathcal{A}^{i+j}[\lambda]$ for any $i,j\in\Z$.
\item[{\rm (2)}] A conformal module $V$ over $\mathcal A$ is {\it $\Z$-graded} if $V=\oplus_{i\in\Z}V_i$, each $V_i$ is a $\C[\partial]$-submodule and $\mathcal (\mathcal A^i)_\lambda V_j\subseteq V_{i+j}[\lambda]$ for any $i,j\in\Z$.
If each $V_i$ is a free $\C[\partial]$-module of rank $n$, then $V$ is called a free $\Z$-graded $\mathcal A$-module of rank $n$.
\end{itemize}
\end{defi}
Note that the loop super-Virasoro conformal superalgebra $\mathfrak{cls}=(\mathfrak{cls})^{\bar0}\oplus(\mathfrak{cls})^{\bar1}$ is $\Z_2$-graded with $(\mathfrak{cls})^{\bar 0}=\oplus_{i\in\Z}\C[\partial]L_i$ and $(\mathfrak{cls})^{\bar1}=\oplus_{i\in\Z}\C[\partial]G_i$ such that $[{(\mathfrak{cls})^{\alpha}} \ _\lambda (\mathfrak{cls})^{\beta}]\subseteq (\mathfrak{cls})^{\alpha+\beta}[\lambda]$ for any $\alpha,\beta\in\Z_2$, and on the other hand $\mathfrak{cls}=\oplus_{i\in\Z}(\mathfrak{cls})_i$ is also $\Z$-graded with $(\mathfrak{cls})_i=\C[\partial]L_i\oplus\C[\partial]G_i$ for each $i$ in the sense of Definition \ref{intermod}.
\section{Conformal superderivations }
\setcounter{theo}{0}
A homogenous linear map $D_\lambda:\mathcal{A}\rightarrow \mathcal{A}[\lambda]$ of degree $\bar i\in\Z_2$
is called a\textit{ homogeneous conformal superderivation of degree} $\bar{i}$ if the following conditions hold:
\begin{eqnarray*}
D_\lambda(\partial a)=(\partial+\lambda)D_\lambda a,\
D_\lambda[a_\mu b]=[[D_\lambda a]_{\lambda+\mu} b]+(-1)^{i|a|}[a_\mu [D_\lambda b]],\quad \forall a,b\in \mathcal A.
\end{eqnarray*} And we write $D$ instead of $D_\lambda$ for simplicity. Denote the set of all conformal superderivations of degree $\alpha\in\Z_2$ by ${\rm CDer}^\alpha(\mathcal{A})$. Then we call ${\rm CDer(\mathcal{A})}={\rm CDer^{\bar0}(\mathcal{A})}\oplus {\rm CDer^{\bar1}(\mathcal{A})}$ the conformal superdirivation algebra of $\mathcal A$ and each element of ${\rm CDer(\mathcal A)}$ a superderivation of $\mathcal A$.
For any $a\in \mathcal{A}$, one can see easily that the linear map $({\rm ad}_a)_\lambda: \mathcal{A}\rightarrow \mathcal{A}[\lambda]$ given by
$({\rm ad}_a)_\lambda b=[a_\lambda b]$ for $b\in \mathcal{A}$ is a conformal superderivation, which is called an \textit{inner conformal superderivation} of $\mathcal A$. Denote the set of all inner conformal superdirivations of $\mathcal A$ by ${\rm CInn(\mathcal A)}$.
Now we are ready to give the main result of this section, which establishes the equality between the two sets ${\rm CDer}(\mathfrak{cls})$ and ${\rm CInn}(\mathfrak{cls})$.
\begin{theo}\label{main3} Every conformal superderivation of $\mathfrak{cls}$ is inner, i.e.,
{\rm CDer}$(\mathfrak{cls})={\rm CInn}(\mathfrak{cls})$.
\end{theo}
\begin{proof}
Take any $D\in {\rm CDer}(\mathfrak{cls})$. For this fixed superderivation $D$ and any $c\in\Z$, define $D^c(x_j)=\pi_{c+j}D(x_j)$ for any $j\in\Z$ and $x_j\in\mathcal {(CL)}_j$, where $\pi_c$ is the natural projection from
$\C[\lambda]\otimes \mathfrak{cls}$ to $\C[\lambda]\otimes \mathfrak {(cls)}_c$. Then it is easy to check that $D^c$ is a conformal superderivation of $\mathfrak{cls}$.
We assert that each $D^c$ is inner. For this, we only need to consider the case that $D^c$ is $\Z_2$-homogenous.
\begin{case} $D^c\in {\rm CDer}^{\bar{0}}(\mathfrak{cls})$.
\end{case}
In this case, assume that $$D_\lambda^c(L_i)=f_i(\partial,\lambda)L_{i+c}\quad {\rm and}\quad D_\lambda^c(G_i)=g_i(\partial,\lambda)G_{i+c}$$for some $f(\partial, \lambda), g_i(\partial, \lambda)\in\C[\partial,\lambda]$.
Applying $D^c_{\lambda}$ to $[L_{0\ \mu} L_i]=(\partial+2\mu) L_{i}$ and comparing the coefficients of $L_{i+c}$ give
\begin{equation*}
(\partial+\lambda+2\mu)f_i(\partial,\lambda)=(\partial +2\lambda+2\mu)f_0(-\lambda-\mu,\lambda)+(\partial+2\mu)f_i(\partial+\mu,\lambda)
\end{equation*}
Setting $\mu=0$ in the formula above, we get
\begin{equation*}
f_i(\partial,i)=(\partial+2\lambda)\frac{f_0(-\lambda,\lambda)}\lambda.
\end{equation*}
Similarly, it follows from $[L_{0\ \mu} G_j]=(\partial+\frac32\mu) G_{j}$ that
\begin{equation}\label{m4.1}
(\partial+\lambda+\frac32\mu)g_{j}(\partial,\lambda)=(\partial+\frac{3(\lambda+\mu)}2)f_0(-\lambda-\mu,\lambda)
+(\partial+\frac{3\mu}2)g_j(\partial+\mu,\lambda).
\end{equation}
Setting $\mu=0$ in (\ref{m4.1}), one has
\begin{equation*}
g_j(\partial,\lambda)=(\partial+\frac{3\lambda}2)\frac{f_0(-\lambda,\lambda)}{\lambda}.
\end{equation*}
Thus, $D^c_\lambda={\rm ad}_{\frac{f_0(\partial,-\partial)}{-\partial}L_c}$.
\begin{case} $D^c\in {\rm CDer}^{\bar{1}}(\mathfrak{cls})$.
\end{case}
Assume that $D^c_\lambda(L_i)=g_i(\partial,\lambda)G_{i+c}$, $D^c_\lambda(G_i)=f_i(\partial,\lambda)L_{i+c}$.
It follows from applying $D^c_{\lambda}$ to $[L_{0\ \mu} L_i]=(\partial+2\mu) L_{i}$ that
\begin{eqnarray*}
(\partial+\lambda+2\mu)g_i(\partial,\lambda)
=g_0(-\lambda-\mu,\lambda)(\frac12\partial+\frac32(\lambda+\mu))G_{i+c}+g_i(\partial+\mu,\lambda)(\partial+\frac{3\mu}2),
\end{eqnarray*}
from which by setting $\mu=0$ one has
\begin{equation*}
g_i(\partial,\lambda)=(\frac12\partial+\frac32\lambda)\frac{g_0(-\lambda,\lambda)}{\lambda}.
\end{equation*}
Using $[{L_0}\ _\mu G_i]=(\partial+\frac32\mu)G_i$, one has
\begin{eqnarray*}
(\partial+\lambda+\frac{3\mu}2)f_i(\partial,\lambda)
=\big(2g_0(-\lambda-\mu,\lambda)-f_i(\partial+\mu,\lambda)(\partial+2\mu)\big).
\end{eqnarray*}
from which by choosing $\mu=0$ gives \begin{equation*}
f_i(\partial,\lambda)=\frac{2g_0(-\lambda,\lambda)}{\lambda}.
\end{equation*}
Whence one can see that
\begin{equation*}
D^c_\lambda={\rm ad}_{\frac{g_0(\partial,-\partial)}{-\partial}G_c}.
\end{equation*}
So in either case, we see that $D^c={\rm ad}_{x_c}$ for some $x_c\in (\mathfrak{cls})_c$, is inner, completing the assertion. Note from the definition of $D^c$ we see that $D=\sum_{c\in\Z}D^c$. In particular, $$D(L_0)=\sum_{c\in\Z}{\rm ad}_{x_c}(L_0)=\sum_{c\in\Z, x_c\neq 0}{\rm ad}_{x_c}(L_0)=\sum_{c\in\Z, x_c\neq 0}{\rm ad}_{x_c}(L_0),$$ which must be a finite sum by the fact that $D$ is a linear map from $\oplus_{i\in\Z}(\mathfrak{cls})_i$ to $\oplus_{i\in\Z}(\mathfrak{cls})_i[\lambda]$. Now it follows from the fact $0\neq {\rm ad}_{y_c}(L_0)\in (\mathfrak{cls})_c$ for any $0\neq y_c\in (\mathfrak{cls})_c$ that all but finitely many $x_c$ are zero, and therefore $\sum_{c\in\Z}x_c\in\mathfrak{cls}$. This implies $D=\sum_{c\in\Z}{\rm ad}_{x_c}={\rm ad}_{\sum_{c\in\Z}x_c}$ is an inner conformal superderivation.
\end{proof}
\section{Free modules of rank $\leq 2$}
\setcounter{case}{0}
Let $V=\C[\pa]x\oplus \C[\pa]y$ be a free $\C[\pa]$-module of rank two with $V^{\bar{0}}=\C[\pa]x$ and $V^{\bar{1}}=\C[\pa]y$. For any $a,b\in\C$ and $c\in\C^*=\C\setminus\{0\}$, on the one hand, define the actions of $L_i$ and $G_i$ on $V$ as follows:
\begin{eqnarray}\label{m4.9}
&&L_i\ _\lambda x=c^i(\pa+a\lambda+b)x,\ L_i\ _\lambda y=c^i(\pa+(a+\frac12)\lambda+b)y,\nonumber\\
&&G_i\ _\lambda x=c^iy,\ G_i\ _\lambda y=c^i(\pa+2a\lambda+b)x;
\end{eqnarray}
on the other hand, actions are given by in another way:
\begin{eqnarray}\label{m4.10}
&&L_i\ _\lambda x=c^i(\pa+a\lambda+b)x,\ L_i\ _\lambda y=c^i(\pa+(a-\frac12)\lambda+b)y, \nonumber\\
&&G_i\ _\lambda x=c^i(\pa+(2a-1)\lambda+b)y,\ G_i\ _\lambda y=c^ix.
\end{eqnarray}It is not hard to see that these two different actions can be extended to the whole $\mathfrak{cls}$ such that in both cases $V$ is a $\mathfrak{cls}$-module. Let us denote the former $\mathfrak{cls}$-module by $M_{a,b,c}$ and the latter by $M_{a,b,c}^\prime.$
\begin{prop}\label{prop-5.1}
(i)\ The $\mathfrak{cls}$-module $M_{a,b,c}$ is irreducible if and only if $a\neq 0$, and $M^\prime_{a,b,c}$ is irreducible if and only if $a\neq \frac12$. Moreover, $\C[\partial](\partial+b)x\oplus\C[\partial]y$ and $\C[\partial]x\oplus\C[\partial](\partial+b)y$ are the unique nontrival $\mathfrak{cls}$-submodules of $M_{0,b,c}$ and $M^\prime_{\frac12,b,c}$, respectively.
(ii)\ For any $R, T\in\{M, M^\prime\}$, then $R_{a,b,c}\cong T_{\alpha,\beta,\gamma}$ if and only if $(a,b,c)=(\alpha,\beta,\gamma)$ and $R=T$.
\end{prop}
\begin{proof}(i)\quad
We only restrict ourself to the irreducibility of $M_{a,b,c}$, the other one can be treated similarly. It is clear that $\C[\partial](\partial+b)x\oplus\C[\partial]y$ is the unique maximal submodule of $M_{0,b,c}$. So the irreducibility of $M_{a,b,c}$ implies that $a\neq 0$.
Conversely, we show that $M_{a,b,c}$ is irreducible if $a\neq 0$. This is equivalent to showing that any submodule $I_{f(\partial), g(\partial)}$ generated by the single nonzero element $f(\partial)x+g(\partial)y$ is the entire $M_{a,b,c}.$ Note by applying the action of $G_i$ on this element if necessary we may assume that $f(\partial)\neq 0$. Due to the irreducibility of $\C[\partial]x$ as an $\mathfrak{clv}$-module (see \cite{DK,WCY}), we can obtain that $$x+h(\partial)y\in I_{f(\partial), g(\partial)}$$ for some $h(\partial)\in\C[\partial]$ by applying the actions of $\mathfrak{clv}$. So $c_i^{-1}G_i\ _\lambda \big(x+h(\partial)y\big)=y+h(\partial+\lambda)(\partial+2a\lambda+b)x\in I_{f(\partial), g(\partial)}[\lambda],$ and then $$y+h(\partial)(\partial+b)x\in I_{f(\partial),g(\partial)}.$$ Now these two elements give $0\neq x-\big({h}(\partial)\big)^2(\partial+b)x\in I_{f(\partial),g(\partial)}.$ It follows from the irreducibility of $\C[\partial]x$ as an $\mathfrak{clv}$-module again that $x\in I_{f(\partial),g(\partial)}$ and therefore $y\in I_{f(\partial),g(\partial)}$ by \eqref{m4.9}. Hence, $I_{f(\partial),g(\partial)}=M_{a,b,c}$, completing the proof.
(ii)\quad Assume that $R_{a,b,c}=\C[\partial]x\oplus \C[\partial]y$ and $T_{\alpha,\beta,\gamma}=\C[\partial]x^\prime\oplus\C[\partial]y^\prime$, and let $$\phi: R_{a,b,c}\rightarrow T_{\alpha,\beta,\gamma}$$ be an isomorphism. Then there exist some $f(\partial), g(\partial)\in\C[\partial]$ such that $$\phi(x)=f(\partial)x^\prime\quad {\rm and}\quad \phi(y)=g(\partial)y^\prime.$$ It follows immediately from $\phi(L_i\ _\lambda x)=L_i\ _\lambda\phi(x)$ that $$c^i(\partial+a\lambda+b)f(\partial)x^\prime=\gamma^i f(\partial+\lambda)(\partial+\alpha\lambda+\beta)x^\prime, $$ from which we can see that $(a,b,c)=(\alpha,\beta,\gamma)$ and that $f(\partial)$ is a constant term, which may be assumed to be $1$ by means of replacing $x^\prime$ by $f(\partial)x^\prime$. It remains to show that $R=T$. Suppose on the contrary that $R\neq T$. Without loss of generality, we assume that $R=M$ and $T=M^\prime$. Then from $\phi(G_i\ _\lambda x)=G_i\ _\lambda\phi(x)$ we obtain that $a=\frac12$. By (i), $M_{\frac12,b,c}$ is irreducible but $M^\prime_{\frac12,b,c}$ is reducible, so $\phi$ can not be isomorphic, a contradiction.
\end{proof}
The aim of this section is to classify all free $\mathfrak{cls}$-modules of rank two. In fact, the two classes of $\mathfrak{cls}$-modules constructed as above exhaust all free $\mathfrak{cls}$-modules of rank two.
\begin{theo}\label{freeofranktwo}
Suppose that $V=V^{\bar{0}}\oplus V^{\bar{1}}$ is a nontrivial free $\mathfrak{cls}$-module of rank two. Then $V$
is either isomorphism to $M_{a,b,c}$ defined by {\rm (\ref{m4.9})} or $M^{\pr}_{a,b,c}$ defined by {\rm(\ref{m4.10})}.
\end{theo}
Let $V=V^{\bar{0}}\oplus V^{\bar{1}}$ be a nontrivial free $\mathfrak{cls}$-module of rank two. Then both $V^{\bar 0}$ and $V^{\bar 1}$ are nontrivial. To see this, suppose on the contrary that $V^{\bar i}=0$, then $G_j\ _\lambda V^{\overline {i+1}}=0$ and hence $L_j\ _\lambda V^{\overline {i+1}}=0$ for any $j\in\Z$. So in this case $V$ is a trivial $\mathfrak{cls}$-module, a contradiction.
Hence both $V^{\bar{0}}=\C[\partial]x\ {\rm and}\ V^{\bar{1}}=\C[\partial]y$ are $\C[\partial]$ and hence $\mathfrak{clv}$-modules of rank one.
Observe that $V$ cannot be a trivial $\mathfrak{clv}$-module, since the relations $[L_{i\ \lambda} G_j]=(\partial+\frac32\lambda) G_{i+j}$ for any $i,j\in\Z$ would imply the actions of $G_i$ on $V$ are all trivial.
Now it follows from \cite[Proposition 4.3]{WCY} that there exist $a, a^\prime, b, b^\prime, c,d\in\C$ such that $$L_i\ _\lambda x=c^i(\pa+a\lambda+b)x,\
L_i \ _\lambda y={d}^i(\pa+a^\prime\lambda+b^\prime)y, \ \forall i\in\Z.$$ Note that at least one of $c$ and $d$ is nonzero, without loss of generality, we assume that $c\neq 0$.
In order to determine the module structure of $\mathfrak{cls}$ on $V$ we only need to give the explicit actions of $G_i$ on $x$ and $y$. Assume \begin{eqnarray*}
G_i\ _\lambda x=c^ig_i(\partial,\lambda)y\quad {\rm and}
\quad G_i\ _\lambda y=c^ig^\prime_i(\partial,\lambda)x\quad {\rm for\ some}\ g_i(\partial,\lambda), g^\prime_i(\partial,\lambda)\in\C[\pa,\lambda].
\end{eqnarray*}
It follows from $[G_i\ _\lambda G_j]_{\lambda+\mu}x=2L_{i+j}\ _{\lambda+\mu}x$ that
\begin{eqnarray}
&&g_i(\pa+\lambda, \mu)g^{\pr}_j(\pa,\lambda)+g_i(\pa+\mu,\lambda)g^{\pr}_j(\pa,\mu)=\pa+a(\lambda+\mu)+b,\quad \forall i,j\in\Z.\label{d4.5}
\end{eqnarray}
In particular, setting $\lambda=\mu$ and $i=j$ in above formula one has
\begin{equation}\label{m4.7}
g_i(\pa+\lambda,\lambda)g^{\pr}_i(\pa,\lambda)=\pa+2a\lambda+b,
\end{equation}which immediately implies
\begin{equation}
g_i(\pa,\lambda)\neq 0,\ g^{\pr}_i(\pa,\lambda)\neq 0,\ \forall i\in\Z.
\end{equation}
This together with $[G_i\ _\lambda G_j]_{\lambda+\mu}y=2L_{i+j}\ _{\lambda+\mu}y$ forces $d\neq 0$. So we have arrived at the following lemma.
\begin{lemm}
The polynomials $g_i(\partial,\lambda), g^\prime_i(\partial,\lambda)$ for all $i\in\Z$ and complex numbers $c, d$ are all nonzero.
\end{lemm}
Using this lemma, we now can give the proof of the theorem above.
\bigskip
\noindent{\em Proof of Theorem \ref{freeofranktwo}}\quad Note that for any $i\in\Z$ by \eqref{m4.7} either ${\rm deg}_{\partial}g_{i}(\partial,\lambda)=1$ and ${\rm deg}_{\partial}g_{i}^\prime(\partial,\lambda)=0$ or ${\rm deg}_{\partial}g_{i}(\partial,\lambda)=0$ and ${\rm deg}_{\partial}g_{i}^\prime(\partial,\lambda)=1$.
\begin{case}
There exists some $i\in\Z$ such that ${\rm deg}_{\partial}g_{i}(\partial,\lambda)=1$ and ${\rm deg}_{\partial}g_{i}^\prime(\partial,\lambda)=0$. \end{case}
It follows from $[L_j\ _\lambda G_i]_{\lambda+\mu}x=(\pa+\frac32\lambda)G_{j+i}\ _{\lambda+\mu} x$ for any $i,j\in\Z$ that
\begin{eqnarray}
g_i(\pa+\lambda,\mu)(d/c)^j(\pa+a^\prime\lambda+b^\prime)-g_i(\pa,\mu)(\pa+\mu+a\lambda+b)
=(\frac12\lambda-\mu)g_{j+i}(\pa,\lambda+\mu).\label{d4.4}
\end{eqnarray} Now view terms on both sides of \eqref{d4.4} as polynomials in the variable $\partial$ with coefficients in $\C[\lambda,\mu]$, then the coefficient of $\partial^2$ in the left hand side must be zero, since ${\rm deg}_{\partial} g_{i+j}(\partial, \lambda+\mu)\leq 1$. This forces $d=c$ and hence \eqref{d4.4} turns out to be \begin{eqnarray}
g_i(\pa+\lambda,\mu)(\pa+a^\prime\lambda+b^\prime)-g_i(\pa,\mu)(\pa+\mu+a\lambda+b)=(\frac12\lambda-\mu)g_{j+i}(\pa,\lambda+\mu).\label{d4.41}\end{eqnarray} Furthermore, by choosing $\lambda=0$ in \eqref{d4.41} and comparing the coefficients of $\partial\mu$ one can see that ${\rm deg}_{\partial}g_j(\partial,\lambda)=1$ for all $j\in\Z$. This allows us for each $i\in\Z$ to write \begin{eqnarray*}g_i(\partial,\lambda)=\partial s_i(\lambda)+t_i(\lambda)\quad{\rm for\ some}\ s_i(\lambda),t_i(\lambda)\in\C[\lambda].\end{eqnarray*} Using these expressions, \eqref{d4.41} is equivalent to
\begin{eqnarray}
& s_i(\mu)\big((a^\prime-a)\lambda+b^\prime-b-\mu\big)+\lambda s_i(\mu)=(\frac12\lambda-\mu)s_{i+j}(\lambda+\mu)\label{eq-4.42}\\
&{\rm and} \quad t_i(\mu)\big((a^\prime-a)\lambda+b^\prime-b-\mu\big)+\lambda(a^\prime\lambda+b^\prime) s_i(\mu)=(\frac12\lambda-\mu)t_{i+j}(\lambda+\mu)\label{eq-4.43}.
\end{eqnarray}
It is not hard to observe from \eqref{eq-4.42} that all $s_i(\lambda)$ are equal to a nonzero constant $\gamma$, and also that $a^\prime=a-\frac12$, $b^\prime=b$. Moreover, replacing $y$ by $\gamma y$ we may assume that $\gamma=1$. At this very moment, \eqref{eq-4.43} can be rewritten as
\begin{equation}\label{eq-4.43-1}
-(\frac12\lambda+\mu)t_i(\mu)+(a-\frac12)\lambda^2+b\lambda =(\frac12\lambda-\mu)t_{i+j}(\lambda+\mu),\end{equation} from which one can see that ${\rm deg}t_i(\lambda)\leq 1$ and thus has the form $t_i(\lambda)=\alpha_i\lambda+\beta_i$ for any $i\in\Z$ and some $\alpha_i,\beta_i\in\C.$ Substituting these explicit expressions into \eqref{eq-4.43-1} and carrying a direct computation show that $t_i(\lambda)=(2a-1)\lambda+b$ for any $i\in\Z$.
To sum up, so far under the assumption $$L_i\ _\lambda x=c^i(\pa+a\lambda+b)x$$ we have obtained the following actions\begin{eqnarray*}
L_i\ _\lambda y=c^i(\pa+(a-\frac12)\lambda+b)y\quad {\rm and}\quad G_i\ _\lambda x=c^i(\pa+(2a-1)\lambda+b)y.
\end{eqnarray*}
Then by the remark at the very beginning of this proof, $g_i^\prime(\lambda):=g_i^\prime(\partial,\lambda)\in\C[\lambda]$. It follows from $[G_i\ _\lambda G_j] \ _{\lambda+\mu}y=2L_{i+j}\ _{\lambda+\mu}y$ that $$(\partial+(2a-1)\lambda+b)g_j^\prime(\mu)+(\partial+(2a-1)\mu+b)g_i^\prime(\lambda)=2\big(\partial+(a-\frac12)(\lambda+\mu)+b\big),$$ which gives immediately rise to $g^\prime_i(\lambda)=1$ for all $i\in\Z$. Whence $V\cong M^\prime_{a,b,c}.$
\begin{case}
There exists some $i\in\Z$ such that ${\rm deg}_{\partial}g_{i}(\partial,\lambda)=0$ and ${\rm deg}_{\partial}g_{i}^\prime(\partial,\lambda)=1$. \end{case}
For this case, similar arguments as for the previous one can be adopted to show that $V\cong M_{a,b,c}.$
\QED
\bigskip
We conclude this section with the following classification result, which is a consequence of Proposition \ref{prop-5.1} and Theorem \ref{freeofranktwo}.
\begin{theo}\label{main4-2}(i)
The set $\{M_{a,b,c}, M^\prime_{\alpha,\beta,\gamma}\mid a,b,\alpha,\beta,\in\C\ {\rm and}\ c,\gamma\in\C^* \ {\rm with}\ a\neq 0, \alpha\neq \frac12\}$ is a complete list of inequivalent irreducible $\mathfrak{cls}$-modules of rank two.
(ii) Any $\mathfrak{cls}$-module $V$ generated by two elements is either trivial or one of the two $\mathfrak{cls}$-modules $M_{a,b,c}$ and $M^\prime_{a,b,c}$ for some $a,b,c\in\C$.
\end{theo}
\begin{proof} (i) is obvious. (ii) Since any $\C[\partial]$-module generated by two elements is a quotient of some free $\C[\partial]$-module of rank two, $V$ is a quotient of $M_{a,b,c}$ or $M^\prime_{a,b,c}$ for some $a,b,c\in\C$ by Theorem \ref{freeofranktwo}. Then (ii) follows from the fact that the nontrivial quotient of $M_{0,b,c}$ or $M_{\frac12,b,c}^\prime$ is a trivial $\mathfrak{cls}$-module by Proposition \ref{prop-5.1} (i).
\end{proof}
\section{Free $\Z$-graded modules of rank $\leq 2$}
\setcounter{section}{5}\setcounter{theo}{0}\setcounter{case}{0}
Let us first collect some results concerning nontrivial free $\Z$-graded $\mathfrak{clv}$-modules of rank one. Two classes of such modules were introduced on the free $\C[\partial]$-module $\oplus_{i\in\Z}\C[\partial]v_i$ in \cite{WCY}:
\begin{itemize}\parskip-3pt
\item[(i)] Denote by $V_{a,b}$ when $\oplus_{i\in\Z}\C[\partial]v_i$ is equipped with the $\mathfrak{clv}$-module structure given by
$$L_i\ _\lambda v_j=(\pa+a\lambda+b)v_{i+j},\quad \forall i\in\Z, $$ where $a,b\in\C$;
\item[(ii)] Denote by $V_{A,b}$ when $\oplus_{i\in\Z}\C[\partial]v_i$ is equipped with the $\mathfrak{clv}$-module structure given by
\begin{equation*}
L_i\ _\lambda v_j=\begin{cases}(\pa+b)v_{i+j},&\mbox{ if}\ (a_j,a_{i+j})=(0,0),\\
(\pa+\lambda+b)v_{i+j},&\mbox{ if}\ (a_j,a_{i+j})=(1,1),\\
v_{i+j},&\mbox{ if}\ (a_j,a_{i+j})=(0,1),\\
(\pa+b)(\pa+\lambda+b)v_{i+j},&\mbox{ if}\ (a_j,a_{i+j})=(1,0),
\end{cases}\end{equation*}
\end{itemize}where $A=(a_i)_{i\in\Z}$ is an element of the product $\prod_{i\in\Z}\Z_2$ of $\{\Z_2\}_{i\in\Z}$. Note that $V_{\{0\}_{i\in\Z},b}$ coincides with $V_{0,b}$ and $V_{\{1\}_{i\in\Z},b}$ coincides with $V_{1,b}.$
We cite the classification result of nontrivial free $\Z$-graded $\mathfrak{clv}$-modules of rank one from \cite{WCY} as a lemma here.
\begin{lemm}\label{lemm-6.1}
Suppose that $V$ is a free $\Z$-graded $\mathfrak{clv}$-module of rank one. Then $V=V_{a,b}$ or $V=V_{A,b}$ for some $a,b\in\C$ and $A\in\prod_{i\in\Z}\Z_2$.
\end{lemm}
Let $\{x_i\mid i\in\Z\}$ and $\{y_i\mid i\in\Z\}$ be two $\C[\partial]$-linearly independent sets, and form the $\C[\partial]$-module $V=\bigoplus_{i\in\Z}(\C[\partial]x_i\oplus\C[\partial]y_i)$. For any $a,b\in\C$ and $A\in\prod_{i\in\Z}\Z_2$, we next give four different actions of $\mathfrak{cls}$ on $V$ such that $\oplus_{i\in\Z}\C[\partial]x_i$ and $\oplus_{i\in\Z}\C[\partial]y_i$ have the form $V_{a,b}$ or $V_{A,b}$ as $\mathfrak{clv}$-modules:
\begin{eqnarray*}\label{m6.3}
\noindent{\rm (i)}&&\ L_i\ _\lambda x_j=(\pa+a\lambda+b)x_{i+j},\quad\quad\quad\quad L_i\ _\lambda y_j=(\pa+(a-\frac12)\lambda+b)y_{i+j},\nonumber\\
&&\ G_i\ _\lambda x_j=(\pa+(2a-1)\lambda+b)y_{i+j},\quad G_i\ _\lambda y_j=x_{i+j}; \\
{\rm (ii)} &&L_i\ _\lambda x_j=(\pa+a\lambda+b)x_{i+j},\quad\quad \quad \quad ~ L_i\ _\lambda y_j=(\pa+(a+\frac12)\lambda+b)y_{i+j},\nonumber\\
&& G_i\ _\lambda x_j=y_{i+j}, \quad\quad\quad \quad\quad\quad \quad\quad\quad~ ~ G_i\ _\lambda y_j=(\pa+2a\lambda+b)x_{i+j};\\
{\rm (iii)}&& L_i\ _\lambda x_j=(\pa+\frac12\lambda+b)x_{i+j}\quad({\rm the\ case}\ a=\frac12);\nonumber\\
&&L_i\ _\lambda y_j=\begin{cases}(\pa+b)y_{i+j},&\mbox{ if}\ (a_j,a_{i+j})=(0,0),\\
(\pa+\lambda+b)y_{i+j},&\mbox{ if}\ (a_j,a_{i+j})=(1,1),\\
y_{i+j},&\mbox{ if}\ (a_j,a_{i+j})=(0,1),\\
(\pa+b)(\pa+\lambda+b)y_{i+j},&\mbox{ if}\ (a_j,a_{i+j})=(1,0);\nonumber\\
\end{cases}\\
&& G_i\ _\lambda x_j=\begin{cases}(\pa+b)y_{i+j},&\mbox{ if}\ (a_j,a_{i+j})\in\{(0,0),(1,0)\},\\
y_{i+j},&\mbox{ if}\ (a_j,a_{i+j})\in\{(1,1),(0,1)\};\\
\end{cases}\\
&& G_i\ _\lambda y_j=\begin{cases}(\pa+\lambda+b)x_{i+j},&\mbox{ if}\ (a_j,a_{i+j})\in\{(1,1),(1,0)\},\\
x_{i+j},&\mbox{ if}\ (a_j,a_{i+j})\in\{(0,0),(0,1)\}.\nonumber
\end{cases}
\end{eqnarray*}
\begin{eqnarray*}
{\rm (iv)}
&&L_i\ _\lambda x_j=\begin{cases}(\pa+b)x_{i+j},&\mbox{ if}\ (a_j,a_{i+j})=(0,0),\\
(\pa+\lambda+b)x_{i+j},&\mbox{ if}\ (a_j,a_{i+j})=(1,1),\\
x_{i+j},&\mbox{ if}\ (a_j,a_{i+j})=(0,1),\\
(\pa+b)(\pa+\lambda+b)x_{i+j},&\mbox{ if}\ (a_j,a_{i+j})=(1,0);\nonumber
\end{cases}\\
&& L_i\ _\lambda y_j=(\pa+\frac12\lambda+b)y_{i+j}\quad({\rm the\ case}\ a=\frac12);\nonumber\\
&& G_i\ _\lambda x_j=\begin{cases}(\pa+\lambda+b)y_{i+j},&\mbox{ if}\ (a_j,a_{i+j})\in\{(1,1),(1,0)\},\\
y_{i+j},&\mbox{ if}\ (a_j,a_{i+j})\in\{(0,0),(0,1)\};\nonumber
\end{cases}\\
&& G_i\ _\lambda y_j=\begin{cases}(\pa+b)x_{i+j},&\mbox{ if}\ (a_j,a_{i+j})\in\{(0,0),(1,0)\},\\
x_{i+j},&\mbox{ if}\ (a_j,a_{i+j})\in\{(1,1),(0,1)\};\\
\end{cases}\\
\end{eqnarray*}
Denote $V$ by $M_{a,b}$, $M^\prime_{a,b}$, $M_{A,b}$, $M^\prime_{A,b}$ in the cases (i)-(iv), respectively. One can verify under the given actions, $M_{a,b}$, $M^\prime_{a,b}$, $M_{A,b}$ and $M^\prime_{A,b}$ becomes $\mathfrak{cls}$-modules. Note that they are all $\Z_2$-graded with the even part $\oplus_{i\in\Z}\C[\partial]x_i$ and the odd part $\oplus_{i\in\Z}\C[\partial]y_i$, and also $\Z$-graded with $\C[\partial]x_i\oplus\C[\partial]y_i$ as their $i$-th gradation. So the four classes of modules above are all free $\Z$-graded $\mathfrak{cls}$-modules of rank two.
\begin{theo}\label{main5}
Suppose that $V=V_{\bar{0}}\oplus V_{\bar{1}}$ is a nontrivial free $\Z$-graded $\mathfrak{cls}$-module of rank two. Then $V$ is isomorphic to one of the following modules: $$M_{a,b}, M^{\pr}_{a,b}, M_{A,b}, M^\prime_{A,b}, \quad {\rm where}\ a,b\in\C\ {\rm and}\ A\in \prod_{i\in\Z}\Z_2.$$
\end{theo}
\begin{proof}Note that both $V^{\bar0}$ and $V^{\bar1}$ are nontrivial by the remark after Theorem \ref{freeofranktwo}, hence are nontrivial free $\Z$-graded $\mathfrak{clv}$-modules. Let $\{x_i\mid i\in\Z\}$ and $\{y_i\mid i\in\Z\}$ be $\C[\pa]$-bases of $V_{\bar{0}}$
and $V_{\bar{1}}$. Then by Lemma \ref{lemm-6.1}, $V^{\bar0}=V_{a,b}$ or $V^{\bar0}=V_{A,b}$ for some $a,b\in\C$ and $A\in\prod_{i\in\Z}\Z_2$. Assume that
\begin{eqnarray*}
&&L_i\ _\lambda x_j=f_{i,j}(\partial,\lambda)x_{i+j}, \quad G_i\ _\lambda x_j=g_{i,j}(\partial,\lambda)y_{i+j},\\
&&L_i\ _\lambda y_j=f^\prime_{i,j}(\partial,\lambda)y_{i+j}, \quad G_i\ _\lambda y_j=g^\prime_{i,j}(\partial,\lambda)x_{i+j}
\end{eqnarray*} for any $i,j,k\in\Z$ and some $f(\partial,\lambda), f^\prime(\pa,\lambda), g(\pa,\lambda), g^\prime(\pa,\lambda)\in\C[\pa,\lambda]$. It follows from $[L_i\ _\lambda G_j]_{\lambda+\mu}x_k=(\pa+\frac32\lambda)G_{i+j}\ _{\lambda+\mu}x_k$ that
\begin{eqnarray}
&&g_{j,k}(\pa+\lambda,\mu)f^{\pr}_{i,j+k}(\pa,\lambda)-g_{j,i+k}(\pa,\mu)f_{i,k}(\pa+\mu,\lambda)\nonumber\\
&=&(\frac12\lambda-\mu)g_{i+j,k}(\pa,\lambda+\mu),\forall i,j,k\in\Z.\label{m5.6}
\end{eqnarray}
\begin{clai} If
$V^{\bar0}=V_{a,b}$ as $\mathfrak{clv}$-modules, then $V^{\bar1}=V_{a-\frac12,b}$, $V_{a+\frac12,b}$ or $V_{A,b}\ (a=\frac12)$. In this case $V=M_{a,b}$, $M^\prime_{a,b}$ or $M_{A,b}$.
\end{clai} Assume that $V^{\bar0}=V_{a,b}$.
This entails us to write explicitly the expressions of $f_{i,j}(\partial,\lambda)$ as: $$f_{i,j}(\partial,\lambda)=\pa+a\lambda+b.$$ It follows from $[G_i\ _\lambda G_j]_{\lambda+\mu}x_k=2L_{i+j}\ _{\lambda+\mu}x_k$ that
\begin{eqnarray}
g_{j,k}(\pa+\lambda, \mu)g^{\pr}_{i,j+k}(\pa,\lambda)+g_{i,k}(\pa+\mu,\lambda)g^{\pr}_{j,i+k}(\pa,\mu)=2\big(\pa+a(\lambda+\mu)+b\big),\ \forall\ i,j\in\Z.\label{m5.5}
\end{eqnarray}
Setting $j=i$ and $\mu=\lambda$ in (\ref{m5.5}) gives
\begin{equation}\label{m5.4-1}
g_{i,k}(\pa+\lambda,\lambda)g^{\pr}_{i,i+k}(\pa,\lambda)=\pa+2a\lambda+b,\ \forall i, k\in\Z,
\end{equation}
from which it is not hard to see that for any $i,k\in\Z,$\begin{eqnarray} &{\rm either}\ g_{i,k}(\pa,\lambda):=\mathfrak g_{i,k} \ {\rm is\ a\ constant} \ {\rm and}\ g^{\pr}_{i,i+k}(\pa,\lambda)=\mathfrak g_{i,k}^{-1}(\pa+2a\lambda+b)\nonumber\\
& {\rm or}\ g^\prime_{i,i+k}(\pa,\lambda):= \mathfrak g_{i,i+k}\ {\rm is\ a\ constant}\ {\rm and}\ g_{i,k}(\pa,\lambda)={\mathfrak g^\prime}^{-1}_{i,i+k}(\pa+(2a-1)\lambda+b).\label{analysis}\end{eqnarray}
\begin{case}\label{c-1}
As $\mathfrak{clv}$-modules, $V^{\bar1}=V_{A,\beta}$ for some $A\in\prod_{i\in\Z}\Z_2\setminus\{\{0\}_{i\in\Z},\{1\}_{i\in\Z}\}$ and $\beta\in\C$. Then $a=\frac12, \beta=b$ and $V=M_{A,b}$.
\end{case}Note that there are four different expressions for $f^\prime_{i,j}(\pa,\lambda).$ We will check it case by case.
\noindent{(a)}\ There exist some $p,q\in\Z$ such that $f^\prime_{p,q}(\partial,\lambda)=(\partial+\beta)(\partial+\lambda+\beta)$.
Choosing $i=p$, $k=q-j$ and using $f_{i,k}(\partial,\lambda)=\pa+a\lambda+b$ in \eqref{m5.6} we have
\begin{eqnarray}
&&g_{j,q-j}(\pa+\lambda,\mu)(\partial+\beta)(\partial+\lambda+\beta)-g_{j,p+q-j}(\pa,\mu)(\pa+\mu+a\lambda+b)\nonumber\\
&=&(\frac12\lambda-\mu)g_{p+j,q-j}(\pa,\lambda+\mu), \quad\forall j\in\Z.\label{m5.6-1}
\end{eqnarray}Combining this with \eqref{analysis}, then we must have\begin{eqnarray*}& \beta=b, a=\frac12, \mathfrak g_{j,q-j}=g_{j, q-j}(\pa+\lambda,\mu)\in\C^*,\\ & g_{j,p+q-j}(\pa,\mu)=\mathfrak g_{j,q-j}(\pa+b) \ {\rm and}\ g_{p+j,q-j}(\pa,\lambda+\mu)=\mathfrak g_{j,q-j}(\partial+b).\end{eqnarray*}
It can be also observed from \eqref{m5.6-1} that all the leading coefficients of $g_{k,l}(\pa,\lambda)$ are equal (here we regard $g_{k,l}(\pa,\lambda)$ as polynomials in $\partial$ with coefficients in $\C[\lambda]$). By rescaling the basis $\{y_i\mid i\in\Z\}$ we may assume that all the leading coefficients of $g_{k,l}(\pa,\lambda)$ are equal 1. Hence, \begin{eqnarray*} g_{j, q-j}(\pa,\lambda)=1\quad {\rm and}\quad g_{j,p+q-j}(\pa,\lambda)=\pa+b,\quad \forall j\in\Z, \end{eqnarray*} which in turn by using \eqref{analysis} forces $$g_{j, q}^\prime(\pa,\lambda)=\pa+\lambda+b\quad {\rm and}\quad g^\prime_{j, p+q}(\pa,\lambda)=1,\quad\forall j\in\Z.$$ In particular, $$g_{p,q}(\pa,\lambda)=\pa+b\quad{\rm and}\quad g^\prime_{p,q}(\pa,\lambda)=\pa+\lambda+b.$$
Similarly, for any $j\in\Z$ we have:
\noindent{(b)}\ $g_{p,q}(\pa,\lambda)=\pa+b, g^\prime_{p,q}(\pa,\lambda)=1$ if $f^\prime_{p,q}(\partial,\lambda)=\pa+b$;\\
\noindent{(c)}\ $g_{p,q}(\pa,\lambda)=1, g^\prime_{p,q}(\pa,\lambda)=\pa+\lambda+b$ if $f^\prime_{p,q}(\partial,\lambda)=\pa+\lambda+b$;\\
\noindent{(d)}\ $g_{p,q}(\pa,\lambda)= g^\prime_{p,q}(\pa,\lambda)=1$ if $f^\prime_{p,q}(\partial,\lambda)=1$.\\ This is exactly the module $M_{A,b}$, completing Case \ref{c-1}.
\begin{case}
As $\mathfrak{clv}$-modules $V^{\bar1}=V_{\alpha,\beta}$ for some $\alpha,\beta\in\C$. Then $\alpha=a\pm \frac12$, $\beta=b$ and $V=M_{a,b}$ or $M^\prime_{a,b}$.
\end{case}
In this case $f_{i,j}^\prime(\pa,\lambda)=\pa+\alpha\lambda+\beta$. So \eqref{m5.6} now becomes \begin{eqnarray*}
&&g_{j,k}(\pa+\lambda,\mu)(\pa+\alpha\lambda+\beta)-g_{j,i+k}(\pa,\mu)(\pa+\mu+a\lambda+b)\nonumber\\
&=&(\frac12\lambda-\mu)g_{i+j,k}(\pa,\lambda+\mu),\forall i,j,k\in\Z.
\end{eqnarray*}Observe from this formula that the coefficients of $\pa$ in $g_{j,i+k}(\pa+\lambda,\lambda)$, $g_{j,k}(\pa,\mu)$ and $g_{i+j,k}(\pa,$ $\lambda+\mu)$ are equal, which may be assumed to be 1 for convenience.
Moreover, the degree of $g_{j,k}(\pa,\lambda)$ for any $j,k\in\Z$ is a constant, i.e, deg$_\pa g_{j,k}(\pa,\lambda)=1$ for all $j,k\in\Z$ or deg$_\pa g_{j,k}(\pa,\lambda)=0$ for all $j,k\in\Z.$ In the former case, $$g_{j,k}(\pa,\lambda)=\pa+(2a-1)\lambda+b, g^\prime_{j,k}(\pa,\lambda)=1, \beta=b\quad {\rm and}\quad \alpha=a-\frac12,$$ that is, $V=M_{a,b}$; while in the later case, $$g_{j,k}(\pa,\lambda)=1, g^\prime_{j,k}(\pa,\lambda)=\pa+2a\lambda+b, \beta=b\quad{\rm and}\quad \alpha=a+\frac12,$$ that is, $V=M_{a,b}^\prime$.
\begin{clai}\label{clai--2} The case $V^{\bar{0}}=V_{A,b}$ and $V^{\bar{1}}= V_{B,\beta}$ for some $A,B\in\prod_{i\in\Z}\Z_2\setminus\{\{0\}_{i\in\Z},\{1\}_{i\in\Z}\}$ and $b, \beta\in\C$ can not occur.\end{clai}
Suppose on the contrary that $V^{\bar{0}}= V_{A,b}$ and $V^{\bar{1}}= V_{B,\beta}$. Choosing $i=\mu=0$ in (\ref{m5.6}), one has
\begin{eqnarray*}
g_{j,k}(\pa+\lambda,0)f^{\pr}_{0,k}(\pa,\lambda)-g_{j,k}(\pa,0)f_{0,k}(\pa,\lambda)=\frac12\lambda g_{j,k}(\pa,\lambda).
\end{eqnarray*} While by definition of the actions of $\mathfrak{clv},$
\begin{eqnarray*}\label{m5.07}f_{0,k}(\pa,\lambda)=\pa+r_k\lambda+b,\ f^{\prime}_{0,k}(\pa,\lambda)=\pa+t_k\lambda+\beta,\quad{\rm where}\ r_k,t_k\in\Z_2.\end{eqnarray*}
Combining these two formulae gives
\begin{eqnarray}\label{m5.8}
g_{j,k}(\pa+\lambda,0)(\pa+t_k\lambda+b^{\prime})-g_{j,k}(\pa,0)(\pa+r_k\lambda+b)=\frac12\lambda g_{j,k}(\pa,\lambda).
\end{eqnarray}
Then $b^{\prime}=b$, by choosing $\lambda=0$. Now we put (\ref{m5.8}) in the following form:
\begin{eqnarray*}
g_{j,k}(\pa,\lambda)=2(\pa+b)\frac{g_{j,k}(\pa+\lambda,0)-g_{j,k}(\pa,0)}{\lambda}+2rg_{j,k}(\pa+\lambda,0)-2tg_{j,k}(\pa,0).
\end{eqnarray*}Taking the operation $\lim_{\lambda\rightarrow 0}$ on both sides of the above formula we see that $g_{j,k}(\partial,0)$ is solutions of the ordinary differential equation
$$2(\pa+b)\frac{d\phi}{d\pa}=(1-2(r_k-t_k))\phi .$$
It is well-known that the general solution of this differential equation is of the form $c(\pa+b)^{\frac{1-2(r_k-t_k)}{2}}$ with $c\in\C$.
Thus $g_{j,k}(\pa,0)\in\C[\pa]\cap\C(\pa+b)^{\frac{1-2(r_k-t_k)}{2}}=0$, since $\frac{1-2(r_k-t_k)}{2}$ is not an integer.
Hence, $g_{j,k}(\pa,0)=0$ and thereby $g_{j,k}(\pa,\lambda)=0$ by (\ref{m5.8}), a contradiction.
\begin{clai}
If $V^{\bar0}=V_{A,b}$ for some $A\in\prod_{i\in\Z}\Z_2\setminus\{\{0\}_{i\in\Z},\{1\}_{i\in\Z}\}$ and $b\in \C$, then $V^{\bar1}=V_{\frac12,b}$ and $V=M^\prime_{A,b}$.
\end{clai}
By Claim \ref{clai--2}, $V^{\bar1}=V_{\alpha,\beta}$ as $\mathfrak{clv}$-modules for some $\alpha,\beta\in\C$. Now it follows from proof of Case \ref{c-1} that $V^{\bar1}=V_{\frac12,b}$ and thereby $V=M^\prime_{A,b}$.
\end{proof}
The irreducible submodules of nontrivial free $\Z$-graded $\mathfrak{cls}$-modules of rank two can be characterized in the proposition.
\begin{prop}We have the following results:
\begin{itemize}
\item[{\rm (1)}]Any nontrivial irreducible $\mathfrak{cls}$-submodule of $M_{a,b}$ has the form \begin{eqnarray*}
&\bigoplus_{k\in\Z}\C[\pa](\pa+b)\sum_{i\in I}c_ix_{i+k}\oplus \bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I}c_iy_{i+k}\quad {\rm if}\ a=0,\\
&\bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I}c_ix_{i+k}\oplus \bigoplus_{k\in\Z}\C[\pa](\pa+b)\sum_{i\in I}c_iy_{i+k}\quad {\rm if}\ a=\frac12,\\
&\bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I}c_ix_{i+k}\oplus \bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I}c_iy_{i+k}\quad {\rm if}\ a\neq 0\ {\rm and}\ a\neq \frac12,
\end{eqnarray*}where $I$ is a finite subset of $\Z$ and $(c_i)_{i\in I}$ is a sequence of complex numbers;
\item[{\rm (2)}]Any nontrivial irreducible $\mathfrak{cls}$-submodule of $M^\prime_{a,b}$ has the form \begin{eqnarray*}
&\bigoplus_{k\in\Z}\C[\pa](\pa+b)\sum_{i\in I}c_ix_{i+k}\oplus \bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I}c_iy_{i+k}\quad {\rm if}\ a=0,\\
&\bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I}c_ix_{i+k}\oplus \bigoplus_{k\in\Z}\C[\pa](\pa+b)\sum_{i\in I}c_iy_{i+k}\quad {\rm if}\ a=-\frac12,\\
&\bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I}c_ix_{i+k}\oplus \bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I}c_iy_{i+k}\quad {\rm if}\ a\neq 0\ {\rm and}\ a\neq -\frac12,
\end{eqnarray*}where $I$ is a finite subset of $\Z$ and $(c_i)_{i\in I}$ is a sequence of complex numbers;
\item[{\rm (3)}]Any nontrivial irreducible $\mathfrak{cls}$-submodule of $M_{A,b}$ has the form
$$\bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I}c_i\delta_{i+k}(\pa)y_{i+k}\oplus\bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I}c_iy_{i+k} ,$$
where $I$ is a finite subset of $\Z$, $(c_i)_{i\in I}$ is a sequence of complex numbers and $$
\delta_i(\pa)=\begin{cases}(\pa+b),&\mbox{ if}\ a_i=0,\\
1,&\mbox{ if}\ a_i=1;\\
\end{cases}$$
\item[{\rm (4)}]Any nontrivial irreducible $\mathfrak{cls}$-submodule of $M^\prime_{A,b}$ has the form
$$\bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I}c_i\delta_{i+k}(\pa)x_{i+k}\oplus \bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I}c_iy_{i+k},$$
where $I$ is a finite subset of $\Z$, $(c_i)_{i\in I}$ is a sequence of complex numbers and $\delta_i(\pa)$ is defined as in (3).
\end{itemize}
\end{prop}
\begin{proof}
Here we only give the proofs of (1) and (3). Denote by $X$ and $Y$ the sets $\{x_i\mid i\in\Z\}$ and $\{y_i\mid i\in\Z\},$ respectively. Let $S_{a,b}$ be a nontrivial irreducible $\mathfrak{cls}$-submodule.
For any $0\neq u\in S_{a,b}$, there exists a finite subset ${\rm Supp}(u)$ of $X\cup Y$ such that $u\in \bigoplus_{z\in {\rm Supp}(u)}\C[\pa]z$. Take $0\neq u_0\in S_{a,b}$ with $\sharp {\rm Supp}(u_0)$ minimal. Then we can write $u_0=\sum_{z\in I_{u_0} }f_z(\pa)z$ for some $f_z(\pa)\in\C[\pa]\setminus\{0\}$.
(1)\ Consider the case $a\neq0$. Without loss of generality, we assume that the set ${\rm Supp}(u_0)\cap\{x_i\mid i\in\Z\}$ is nonempty. Fix $z^\prime\in X\cap {\rm Supp}(u_0).$ By the irreducibility of $\C[\pa]z^\prime$ as $\C[\pa] L_0$-module \cite{DK}, we may assume that $f_{z^\prime}(\partial)=1.$ Consider the elements \begin{eqnarray*}&&u_0^\prime(c_1,c_2)=L_0\ _\lambda(u_0)|_{\lambda=c_1}-L_0\ _\lambda(u_0)|_{\lambda=c_2}-a(c_1-c_2)u_0\\
&=&\sum_{z^\prime\neq z\in X\cap {\rm Supp}(u_0)}\Big(f_z(\pa+c_1)(\pa+ac_1+b)-f_z(\pa+c_2)(\pa+ac_2+b)-a(c_1-c_2)f_z(\pa)\Big)z\\
&&+\sum_{ z\in Y\cap {\rm Supp}(u_0)}\Big(f_z(\pa+c_1)\big(\pa+(a-\frac12)c_1+b\big)-f_z(\pa+c_2)\big(\pa+(a-\frac12)c_2+b\big)\\
&&~~~~~~~~~~~~~~~~~~~~~~-a(c_1-c_2)f_z(\pa)\Big)z\in S_{a,b}\end{eqnarray*} where $c_i\in\C.$ Now the minimality of $ {\rm Supp}(u_0)$ implies that $u_0^\prime(c_1,c_2)=0,$ namely, for any $c_i\in\C$ we have
\begin{eqnarray*}0&=&f_z(\pa+c_1)(\pa+ac_1+b)-f_z(\pa+c_2)(\pa+ac_2+b)\\ &&-a(c_1-c_2)f_z(\pa),\quad \forall z\in X\cap {\rm Supp}(u_0)
\end{eqnarray*} and \begin{eqnarray*}0&=&f_z(\pa+c_1)(\pa+(a-\frac12)c_1+b)-f_z(\pa+c_2)(\pa+(a-\frac12)c_2+b)\\ &&-a(c_1-c_2)f_z(\pa), \quad\forall z\in Y\cap {\rm Supp}(u_0).\end{eqnarray*} Hence, $0\neq c_z:=f_z(\pa)\in\C\ (c_{z^\prime}=1)$ for each $ z\in X\cap {\rm Supp}(u_0)$ and $f_z(\pa)=0$ for each $z\in Y\cap {\rm Supp}(u_0)$. By the minimality of ${\rm Supp}(u_0)$ again, $Y\cap {\rm Supp}(u_0)$ is empty. That is, $u_0=\sum_{z\in X\cap {\rm Supp}(u_0)}c_zz\in S_{a,b}.$ Set $I_{u_0}=\{i\mid x_i\in X\cap{\rm Supp}(u_0)\}.$ Then $u_0$ can be written as $u_0=\sum_{i\in I_{u_0}}c_ix_i$ with $c_i=c_{x_i}$. Now one can see that \begin{eqnarray*} S_{a,b}=\bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I_{u_0}}c_ix_{i+k}\oplus \bigoplus_{k\in\Z}\C[\pa](\pa+b)\sum_{i\in I_{u_0}}c_iy_{i+k}\quad &{\rm if}&\ a=\frac12\\
{\rm and}\ \ \ \ \ S_{a,b}=\bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I_{u_0}}c_ix_{i+k}\oplus \bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I_{u_0}}c_iy_{i+k}\quad &{\rm if}&\ a\neq \frac12.\end{eqnarray*}
Let us assume that $a=0$ and in this case we start with the nonempty set $Y\cap {\rm Supp}(u_0)$. Similarly, one can show that $$S_{a,b}=\bigoplus_{k\in\Z}\C[\pa](\pa+b)\sum_{i\in I_{u_0}}c_ix_{i+k}\oplus \bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I_{u_0}}c_iy_{i+k}.$$
(3)\ As for the case $a\neq 0$ in the proof of (1), $u_0=\sum_{i\in I_{u_0}}c_ix_i\in S_{a,b}.$ Then it is not hard to see that $$S_{a,b}=\bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I_{u_0}}c_ix_{i+k}\oplus \bigoplus_{k\in\Z}\C[\pa]\sum_{i\in I_{u_0}}c_i\delta_{i+k}(\pa)y_{i+k}.$$
\end{proof}
\section*{Acknowledgments}
This paper was supported by NSF grants 11371278,~11431010,~11101056,~11501417,~11161010,
the Fundamental Research Funds for the Central Universities of China,
Innovation Program of Shanghai Municipal Education Commission and Program for
Young Excellent Talents in Tongji University.
| 79b5c8593aef413b63341c6ceb650204000db555 | {
"file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz"
} |